Split (1)

train · 1.56M rows

text stringlengths 1.23k 293k	tokens float64 290 66.5k	created stringdate 1-01-01 00:00:00 2024-12-01 00:00:00	fields listlengths 1 6
Data Driven Natural Gas Spot Price Prediction Models Using Machine Learning Methods : Natural gas has been proposed as a solution to increase the security of energy supply and reduce environmental pollution around the world. Being able to forecast natural gas price beneﬁts various stakeholders and has become a very valuable tool for all market participants in competitive natural gas markets. Machine learning algorithms have gradually become popular tools for natural gas price forecasting. In this paper, we investigate data-driven predictive models for natural gas price forecasting based on common machine learning tools, i Introduction The world total primary energy supply (TPES) by fuel in 2016 was as follows: oil (31.9%), coal (27.1%), natural gas (22.1%), biofuels and waste (9.8%), nuclear (4.9%), hydro (1.8%), and other (0.1%) [1]. Natural gas thus has the third-largest share among the TPES. Furthermore, natural gas production continues to grow at a higher pace, most notably with a 3.6% increase in 2017 compared to 2016 that constitutes the largest increase since 2010. In today's world, concerns about air quality and climate change are growing, but renewable energy is expanding at a limited rate and low-carbon energy sources are hard to find in some areas. Natural gas offers many potential benefits as a solution to environmental problems. Natural gas generates heat, power, and mobility with fewer emissions, including both carbon-dioxide (CO 2 ) emissions and air pollutants, than the other fossil fuels, helping to address widespread concerns over air quality. Because the natural gas energy causes less pollution to the environment than other kinds of energy resource, it has received much more recognition recently. Natural gas exploitation has significantly helped many countries to reduce CO 2 emissions nationally and globally since 2014 [2]. Natural gas, which is one of the most important energy resources, is going to play an expanded role in the future of global energy due to its significant environmental benefits. Forecasting natural gas prices is a powerful and essential tool which has become more important for different stakeholders in the natural gas market, allowing them to make better decisions for managing the potential risk, reducing the gap between the demand and supply, and optimizing the model validation techniques, model parameter selection, and empirical study results using prepared models. Finally, Section 5 presents the conclusions. ANN The neural network is a computational model created in 1943 by McCulloch and Pitts based on mathematics and threshold logic algorithms [18]. The neural network is a framework for absorbing various machine learning algorithms for cooperative work, and thus it is not an algorithm. ANN has risen in importance since the 1980s and has been a research hotspot in the field of artificial intelligence. ANN has widespread applications in terms of data processing, classification, regression, function approximation, and numerical control. ANN process the information like the neural network of the human brain, i.e., by connecting different simple units/nodes, named as artificial neurons, to form different complex networks [19]. Each node includes an activation function to produce an output value based on one or multiple inputs. The output signal of a node can be passed to another node using a weighted connection. Thus, the output an ANN depends on the connection structure, the weight value and activation function [20]. A simple signal layer ANN is illustrated in Figure 1. Given one unit (an artificial neuron) j, let its input signals connected from other units be xi (i = 1, 2,…, n) with corresponding weights wij (i = 1, 2,…, n). There are two basic operations in the processing unit, namely summation and activation of the input signals [18]. The unit's output of yj is defined as: where tj is a bias for the unit j, and f is the activation function which can be commonly defined as the sigmoid function [21]: The output yj will pass to the connected units from the next layer as an input signal. Given a designed architecture, all units in ANNs are interconnected in different layers. A simple example of a three-layer ANN is shown in Figure 2: the information flows through the input, hidden and output layers. The input layer or the first layer contains the same number of units as the input vector size, followed by the hidden layer with an arbitrary number of units. The output layer aggregates the weighted outputs from the hidden layer units. Given one unit (an artificial neuron) j, let its input signals connected from other units be x i (i = 1, 2, . . . , n) with corresponding weights w ij (i = 1, 2, . . . , n). There are two basic operations in the processing unit, namely summation and activation of the input signals [18]. The unit's output of y j is defined as: where t j is a bias for the unit j, and f is the activation function which can be commonly defined as the sigmoid function [21]: The output y j will pass to the connected units from the next layer as an input signal. Given a designed architecture, all units in ANNs are interconnected in different layers. A simple example of a three-layer ANN is shown in Figure 2: the information flows through the input, hidden and output layers. The input layer or the first layer contains the same number of units as the input vector size, followed by the hidden layer with an arbitrary number of units. The output layer aggregates the weighted outputs from the hidden layer units. SVM SVM is mainly based on the theory of statistical learning theory. It a machine learning tool for solving a convex quadratic problem containing linear constraints, in which local minimum does not exist. Linear support vector machines can be extended into nonlinear space by introducing the kernel method, which barely increases the computational cost for high-dimensional data. Vapnik and Chervonenkis proposed the VC theory of exponential function sets, which motivated the progress of support vector machines in the historical stage [22]. With the development and maturity of theoretical research, Boser et al. proposed SVM by harnessing the theory of statistical learning knowledge and realized the nonlinear support vector machine according to the kernel function theory [23]. With respect to the risk minimization principle, ANN employs the empirical mode while SVM relies on the structural mode. So far, it has been widely used in various classification tasks, such as pattern recognition, images and text classification. In addition, SVM has been used for the regression task with some loss functions [24]. SVM can realize the regression version, which is called support vector regression (SVR). SVR solves regression problems on the basis of support vector machines [25,26]. SVR has powerful learning ability and generalization ability suitable for a small set of samples and has got great success in forecasting work with promising results [27,28]. This results from that it can adaptively process various tasks through generalization and meanwhile gets gain access to satisfactory answers. In particular, a nonlinear relation can be transferred into a linear relation with the utilization of SVR and thus the task can be easily solved. A drawback is that this becomes time-consuming when the task scale becomes large [29,30]. For the decision function, it can be represented by: where w, x, and b represent the weight vector, the input vector, and the bias, respectively. For the training input vector and target output, we assume xm and ym be the m-th (m = 1, 2,…, M) input and output, respectively. Let C be a positive parameter for penalizing the samples which does not satisfy inequality constraints. The error function can be defined as: The error function has two terms, which are used for penalizing model complexity and representing the ε-insensitive loss function, respectively. The loss function is mathematically given as: SVM SVM is mainly based on the theory of statistical learning theory. It a machine learning tool for solving a convex quadratic problem containing linear constraints, in which local minimum does not exist. Linear support vector machines can be extended into nonlinear space by introducing the kernel method, which barely increases the computational cost for high-dimensional data. Vapnik and Chervonenkis proposed the VC theory of exponential function sets, which motivated the progress of support vector machines in the historical stage [22]. With the development and maturity of theoretical research, Boser et al. proposed SVM by harnessing the theory of statistical learning knowledge and realized the nonlinear support vector machine according to the kernel function theory [23]. With respect to the risk minimization principle, ANN employs the empirical mode while SVM relies on the structural mode. So far, it has been widely used in various classification tasks, such as pattern recognition, images and text classification. In addition, SVM has been used for the regression task with some loss functions [24]. SVM can realize the regression version, which is called support vector regression (SVR). SVR solves regression problems on the basis of support vector machines [25,26]. SVR has powerful learning ability and generalization ability suitable for a small set of samples and has got great success in forecasting work with promising results [27,28]. This results from that it can adaptively process various tasks through generalization and meanwhile gets gain access to satisfactory answers. In particular, a nonlinear relation can be transferred into a linear relation with the utilization of SVR and thus the task can be easily solved. A drawback is that this becomes time-consuming when the task scale becomes large [29,30]. For the decision function, it can be represented by: where w, x, and b represent the weight vector, the input vector, and the bias, respectively. For the training input vector and target output, we assume x m and y m be the m-th (m = 1, 2, . . . , M) input and output, respectively. Let C be a positive parameter for penalizing the samples which does not satisfy inequality constraints. The error function can be defined as: The error function has two terms, which are used for penalizing model complexity and representing the ε-insensitive loss function, respectively. The loss function is mathematically given as: This function cannot penalize errors less than ε in order to reduce model complexity. The solution of minimizing the error function can be defined as: in which the parameters α m and α * m are Lagrange multipliers. Support vectors from the training vectors give non-zero Lagrange multipliers and contribute directly to the solution, while the non-support vectors do not. The relationship between support vectors and the model complexity is that its number can measure the latter [31,32]. The key in SVR applications is the utilization of kernel functions, which can map non-linear data into spaces in which they are essentially linear for the optimization process. One can utilize the following kernel function: Then, the model is changed into: For kernel functions, Gaussian, polynomial, and hyperbolic tangent type can be used. The Gaussian type is one of the most popular kernel functions used in empirical analysis, mathematically described as follows: GBM Friedman proposed the concept of gradient boosting and introduced the gradient boosting machine (GBM) technique for the extension of the boosting to the regression [33][34][35]. GBM has many applications in regression and classification by generating forecasting models in the form of the set of weak forecasting models (usually decision tree). GBM is a kind of boosting method. The basic idea of boosting is to utilize some way such that base learners of each round are more concerned about the misclassified learning samples of last round during the process of training, but GBM focuses on each modelling using the gradient descent direction in the previous model loss function. The negative gradient is used as the measurable indicator that base learners of last round and are fitted in the next round learning to correct mistakes appeared in the last round. The difference with traditional boosting is that the purpose of each computation is to reduce the residual of last computation. In order to eliminate the residual, one can refer to the gradient direction of residual decrease for model build-up. Thus, the building-up of each new model is to make the residual of the previous model cut down along the gradient direction in the gradient boosting, which greatly differs from traditional boosting in weighting correct and wrong samples. GBM has been shown significant success deployed in various kinds of practical applications in different fields. It can be considered as a methodological framework, i.e., it is very flexible and customizable which can achieve high accuracy in many data mining tasks. However, in practice, the most issue of GBM is it memory-consumption. With respect to a GBM algorithm, given an output y, an input x, a training set x i , y j n i=1 , and number of iterations M, we seek an approximationF(x) for a function F(x) and the approximation requires minimizing the expected value of some specified loss function L(y, F(x)): The gradient boosting method assumes a real-valued y and seeks an approximationF(x) and a base- can be approximated in a greedy way: The negative gradient is calculated by: h m (x) is utilized for fitting y i through minimizing the squared error: One can employ line search for ρ m so as to minimize the loss function: Finally, it is updated: GPR GPR builds upon the Gaussian process (GP). GP is a kind of random processes that generalizes multivariate normal distribution in the infinite dimensional space and therefore is defined in the distribution over functional values. It is a powerful nonlinear multivariate interpolation tool. The prior information of GP is used for the inference of continuous value, which is referred to as GPR. GPR is an interpolation method, in which the interpolation is controlled by prior covariances, and can give the best linear unbiased prediction for median under the appropriate prior assumptions. It can also be illustrated by the theory of Bayesian inference [36]. GPR is proposed by Rasmussen et al. at first, which is an efficient machine learning algorithm based on kernel function and provides a principle, practicable, and probabilistic method for kernel machine learning [37,38]. In comparison with other machine learning methods, the advantages of GPR lie in being able to seamlessly integrate multiple machine learning tasks such as parameter estimation; therefore, the regression can be obviously simplified and the objective influence for the result is slight, more convenient for an explanation. GPR possesses excellent performance with relatively small training dataset and offers confidence interval for prediction by adopting flexible kernel function. However, a known bottleneck is that the computation complexity in prediction is cubic about \|x\|, thus it is infeasible for large datasets. Let X be the matrix of input vectors and x i be the i-th training input vector. Let K(X, X) be the covariance matrix for pairwise vectors, such that: The variable f follows the following distribution: which is a multivariate Gaussian density distribution. m f (µ, Σ) represents the normal density function. Additionally, some noise can be added for the estimation of target output y: where the noise ε yields the independent normal distribution with mean zero and standard deviation σ n . Then, given an input vector x * from the test data, the outputf * becomes: Data Preparation and Description There exist a great number of factors that affect the natural gas price, e.g., crude oil, heating oil, drilling activity, temperature, natural gas supply, demand, storage, imports, etc. To better forecast the gas price, we should consider as many relevant factors as possible. To begin with, the relationship between crude oil price and natural gas price is a continuous research focus [39]. They often behave a strong connection and crude oil price fluctuations usually affect the natural gas price. Secondly, the natural gas can be seen as an alternative energy resource to heating oil in US industry and power plants, thus, their demands can be shifted between each other based on the market prices [40]. Thirdly, the prosperity level of the natural gas market can be reflected based on the gas drilling activities [41]. Fourthly, the price of natural gas often suffers from the temperature and weather changes are also a key factor [42]. The demand for natural gas varies with season. In addition, obviously, natural gas supply, demand, storage, and imports are important indicators related to natural gas price. Thus, the abovementioned variables include crude oil price (WTI), heating oil price (HO), natural gas rotary rigs (NGRR), heating degree-days (HDD), cooling degree-days (CDD), natural gas marketed production (NGMP), natural gas total consumption (NGTC), natural gas underground storage volume (NGUSV), and natural gas imports (NGI) are used as inputs in this study. Natural gas price is the output variable and Henry Hub Natural Gas Spot Price is adopted, which is from the US EIA [43] and quoted in dollars per million btu. Henry Hub is the earliest representative American natural gas spot trading centre and is also the largest one until now. It has greatly benefited the development of the natural gas market and energy economic prosperity since its establishment. For the abovementioned variables, our dataset comprises monthly data within the period from January 2001 to October 2018 and all data in the months in this period are complete, which guarantees the availability of data. Historical trends of natural gas and relevant variables are described in Figure 3, in which Figure 3a depicts the natural Henry Hub spot price from January 2001 to October 2018, containing a total of 214 records. It can be known that during this period, the average natural price was $4.71, the median was $4.05, and the standard deviation was $2.22. As shown in Figure 3a One can easily find that there are a few sharp rises for natural gas before 2009. The first sharp rise appeared in the winter of 2003, which is mainly caused by extremely cold weather, consistent with the case of storage reserves being substantially reduced, as shown in Figure 3i. The second upsurge appeared in October to December of 2005 when hurricanes Katrina and Rita happened in succession. Hurricanes make natural gas supply drop markedly. The lowest marketed production is in September 2005, only 1,400,941 million cubic feet, while in October, it increased by 74,708 million cubic feet, which can be observed in Figure 3g. The third rapid rise is before the global financial crisis in 2008, which, to a great extent, was caused by an increase of the crude oil prices that led a strong demand for natural gas, as shown in Figure 3b-d. Since 2009, natural gas in the US has remained at a low level and this is mainly due to two aspects: one is that the demand for natural gas is not urgent. The other aspect is the oversupply due to the rapid development of shale gas. The fluctuation range of HDD doubles that of CDD or so and thereby the uncertainty of temperature changes in winter is greater than that in autumn in the US, affecting the expectation of natural gas demand in winter. It can be seen from Figure 3g,i that the supply of natural gas basically went up and the storage volume increased continually during the observation period, which results from the shale gas revolution in the US. Total consumption of natural gas has a rising trend while the imports show the opposite. Table 2 lists the descriptive statistics for natural gas spot price (NGSP)and relevant variables including max, min, median, mean, standard deviation (SD), and relative standard deviation (RSD). It can be seen from Table 2 that we can know that the standard deviation of natural gas price is far smaller than that of other variables, except for heating oil price, while the relative standard deviation of natural gas is at the mid-level among all the variables. The fluctuation range of HDD doubles that of CDD or so and thereby the uncertainty of temperature changes in winter is greater than that in autumn in the US, affecting the expectation of natural gas demand in winter. It can be seen from Figure 3g,i that the supply of natural gas basically went up and the storage volume increased continually during the observation period, which results from the shale gas revolution in the US. Total consumption of natural gas has a rising trend while the imports show the opposite. Table 2 lists the descriptive statistics for natural gas spot price (NGSP)and relevant variables including max, min, median, mean, standard deviation (SD), and relative standard deviation (RSD). It can be seen from Table 2 that we can know that the standard deviation of natural gas price is far smaller than that of other variables, except for heating oil price, while the relative standard deviation of natural gas is at the mid-level among all the variables. Table 3 shows the correlations among the variables. It can be known from Table 3 that with respect to the correlation between the natural gas price and other variables [45], the strongest is with the number of rotary rigs and second are marketed production and imports. The weakest correlations are with heating degree-days and cooling degree-days. Among relevant variables, the crude oil price and the heating oil price have the strongest correlation. The variables with strong correlations also occur between the number of rotary rigs and marketed production, the number of rotary rigs and imports, heating degree-days and cooling degree-days, heating degree-days and total consumption, and marketed production and imports. The variables with weak correlations occur between crude oil price and total consumption, the number of rotary rigs and heating degree-days, and the number of rotary rigs and cooling degree-days, in which the weakest correlation is between the number of rotary rigs and heating oil price. Forecasting Performance Evaluation Criteria There existed many model evaluation criteria and we choose some classic statistical criteria for testing model forecasting performance. R-square (R 2 ) is often used for measuring the goodness-of-fit [46] where y t and y t represent the actual value and the prediction value at the time t, respectively. N means the total number of samples in the dataset. R 2 can manifest the goodness and badness of fitting by using changes in the data. The value of R 2 ranges from is from zero to one. The regression fitting effect becomes better when the value of R 2 is closer to 1 [47,48]. In general, the value greater than 0.8 can indicate that the model has a high enough goodness-of-fit. Moreover, there are four criteria used for measuring the forecasting errors. Mean absolute error (MAE) is also a common forecasting performance evaluation tool obtained by averaging the absolute errors [49,50]: Mean square error (MSE) is an average of the quadratic sum of the deviation of forecasting value and actual value and is good at measuring the degree of change [51,52]. The prediction model gets better accuracy with the decrease of the MSE value. In contrast to MAE, MSE enlarges the value of great prediction deviation and compares the stability of different prediction models. It is represented as: Root-mean-square error (RMSE) can be directly obtained from MSE by calculating the square root [53,54]. It is very sensitive to the very large or very small error value and thus has a good reflection for precision, formulated as: Mean absolute percentage error (MAPE) is often used as a loss function, since it can intuitively explain the relative errors [55,56]. It considers not only the deviation of prediction value and actual value, but also the ratio between them. It is given by: In the above-mentioned evaluation indices including MAE, MSE, RMSE, and MAPE, the smaller the value, the better the forecasting model accuracy. Model Validation Technique Cross-validation is an important method for model selection to obtain feasible and stable models in machine learning. In cross-validation, K-fold validation is a common method of preventing over-fitting when the model is too complex. It has been demonstrated that compared with Holdout method and Leave-one-out cross-validation, K-fold validation has more advantages in precision estimation and model selection [57]. With the help of K-fold validation, we partition training dataset into K equal folds, of which K − 1 folds are employed for model training while the remaining fold is used for validation. This operation of training and validation is repeatedly done by rotating K folds. After carrying out this operation K times, we gather all the errors to calculate out the final error. Note that, K is often taken as 10 in reality [57]. Model Parameter Selection Machine learning methods involve many parameters, some of which are key parameters and thus are carefully selected. The complexity of the model is often related to these key parameters, which is also called model selection parameters. With regard to ANN, we choose a nonlinear autoregressive model with external input. The number of hidden neurons and the number of delays are selected as 10 and 2, respectively. In the train network, the training algorithm chosen is Levenberg-Marquardt, which typically needs more memory but less time. According to the MSE value of the validation samples, the training will stop automatically once the generalization cannot improve. The kernel function should be carefully selected in the SVM model. In order to pick off the appropriate kernel function, our strategy is to test different functions and then the one with minimum error is what we need. Accordingly, six common kernel functions including linear type, quadratic type, cubic type, fine Gaussian, medium Gaussian, and coarse Gaussian are investigated, as shown in Table 4. Over one hundred experiments are deduced to the cubic type, which has better results than others overall and thereby serves as the kernel function of SVM in this study. GBM requires a reasonable loss function. The gradient boosting strategy harnesses several popular loss criteria: least-squares, least absolute deviation, Huber, logistic binomial log-likelihood, etc. We choose least-squares as loss function, since only the case of least-squares is suitable for regression model while others are usually applied in classification. In addition, the covariance function or kernel function is needed in GPR, such as Gaussian kernel and squared exponential kernel used in many machine learning algorithms. In this paper, squared exponential is used as the kernel function, since it behaves better results. Empirical Study Results Based on the before-mentioned machine learning methods, prepared data, forecasting performance evaluation criteria, model validation technique, and selected model parameters, the empirical study is carried out. Table 5 lists the data that describe the forecasting performance of four machine learning methods, in which each value is the average of one hundred tests for the purpose of more accurate results. Observing data of four criteria can easily find that the forecasting performance of ANN and SVM is better than that of GBM and GPR. In particular, ANN is obviously superior to other methods while GBM has the worst behaviour. Overall, the performance ranking is ANN, SVM, GPR, and GBM from strong to weak. Figure 5 shows the distribution suitcases between predictive values and actual values. It can be seen that the actual normal price is centered around 2 to 8 and these four methods give a good prediction for normal values. The fluctuation of ANN is less than that of others. However, they are unsatisfactory when the price goes up greater than 8, e.g., from September to December and April to July 2008. The prediction of abnormal values is still a difficult task so far. Figure 5 shows the distribution suitcases between predictive values and actual values. It can be seen that the actual normal price is centered around 2 to 8 and these four methods give a good prediction for normal values. The fluctuation of ANN is less than that of others. However, they are unsatisfactory when the price goes up greater than 8, e.g., from September to December and April to July 2008. The prediction of abnormal values is still a difficult task so far. Figure 5 shows the distribution suitcases between predictive values and actual values. It can be seen that the actual normal price is centered around 2 to 8 and these four methods give a good prediction for normal values. The fluctuation of ANN is less than that of others. However, they are unsatisfactory when the price goes up greater than 8, e.g., from September to December and April to July 2008. The prediction of abnormal values is still a difficult task so far. From the above results, we can understand that there are obviously distinct differences among the four machine learning models, and there is a general ranking reference. Overall, ANN and SVM have better prediction performance than GBM and GPR, where ANN wins the best and GBM is the worst. ANN has a good ability of self-learning, self-adapting, and self-organizing [58], which can analyze the patterns and rules of observed data and form complex non-linear functions through training, and adapt to large-scale, multi-factor, incomplete, and inaccurate data processing. The number of delays in ANN is 10 in this study, which is greater than traditional ANN's (the number of delays is 3 at most). That may be one reason that ANN is superior to SVM. SVM has an abundant and theoretical foundation and has strong approximation ability and generalization ability based on structural risk minimization. SVM aims at small sample, of which the optimal solution is based on limited sample. GPR has a simplified regression process and easily explains the consequence. It has favourable performance in number and stability, but it is suitable for research on relatively small training dataset. GBM has a relatively weaker performance than the above methods in forecasting natural gas prices, however, we can obtain the degree of importance of explanatory variables by using GBM. Conclusions The aim of this study is to investigate natural gas price forecasting based on four machine learning methods (ANN, SVM, GBM, and GPR). Monthly Henry Hub natural gas spot price data from January 2001 to October 2018 (there are 215 observations) were used in four prediction methods. Nine variables were investigated as inputs, which are NGSP, WTI, HO, NGRR, HDD, CDD, NGMP, NGTC, NGUSV, and NGI. The method cross-validation was used in model training. Four forecasting performance evaluation criteria including R 2 , MSE, RMSE, and MAPE are employed in prediction methods. Finally, the empirical results demonstrate that four prediction methods have decent performance in forecasting natural gas price. Overall, ANN and SVM have better forecasting performance than GBM and GPR. In particular, ANN obviously outperforms the other methods while GBM is the worst. This study could be improved by more thorough research, e.g., by comparing more different aspects of the prediction performance such as computation efficiency and using more diverse machine learning methods such as random forest [59,60]. For future work, we will evaluate the effect of emerging machine learning algorithms, such as deep learning and reinforcement learning, on energy price and correlation prediction.	7,186.2	2019-05-03T00:00:00.000	[ "Environmental Science", "Computer Science" ]
Low-Frequency Magnetic Field Exposure System for Cells Electromagnetic Biocompatibility Studies : The advancement in science and technology has resulted in the invention and widespread usage of many electrical devices in the daily lives of humans. The exponential use of modern electronic facilities has increased electromagnetic ﬁeld exposure in the current population. Therefore, the presented article deals with designing, constructing, and testing a new applicator system developed for cells electromagnetic biocompatibility studies. The applicator system is intended for studying the non-thermal impacts of low-frequency magnetic ﬁeld on cell cultures growth. Main attention is focused on increasing the capacity of the applicator and effectivity of the experiments. The key idea is to reach high level of the magnetic ﬁeld homogeneity in an area of interest and the temperature stability during the biocompatibility studies. The applicator system is designed based on numerical simulations and its construction, measurements, and properties evaluation are also reported for proving the applicator’s functionality. The new applicator allows performing ﬁve parallel experiments at the same time under the same conditions. The simulation together with the experimental results conﬁrm that the magnetic ﬁeld homogeneity reaches 99% in the area of interest and the maximum temperature instability is lower than 2% during the experiments. The effectiveness of new applicator is tested and proved during preliminary experiments with Saccharomyces Cerevisiae cells. The observed effects of MF exposure represent maximal stimulation of 74% and maximal inhibition of 49%. The reason why MF with the same parameters induces inhibition in one sample and stimulation in the other will be the subject of further research. Introduction This article aims to design, develop, construct, and verify a new exposure system to study eukaryotic cells' low-frequency electromagnetic biocompatibility. The research on the electromagnetic field (EMF) effects on living cells in the low frequency (LF) and extremely low-frequency (ELF) range is a worldwide longtime discussed topic. Even though the thermal effects are relatively well understood nowadays, there are still more questions than answers around nonthermal impacts. An impressive amount of research works conducted within this area during the last few decades [1][2][3][4][5][6] succeeded at least in terms of the necessity for nonthermal effects consideration when elucidating the response of the biological system to external EMF. In the LF range, the biological effects are usually investigated at weak levels, thus not likely to cause the thermal ones. One of the significant challenges in investigating the LF EMF nonthermal biological impacts is the lack of precise results in epidemiological and laboratory studies, pointing to a reliable mechanism of action between the LF EMF and living cells. The problem, whether electric field (EF) or the magnetic one (MF) is responsible for the biological effects of LF EMF in the nonthermal area, is not yet clarified. Still, some researchers point to the threshold for Design of Exposure System The exposure system design assumes the use of two identical applicators of MF. Using two applicators will ensure the same cultivation conditions for exposed and control samples. One applicator serves as a source of MF; the second one represents a state without intervention/negative control. Furthermore, the design assumes the use of an incubator maintaining a constant temperature, Erlenmeyer flasks with liquid yeast-cell culture, shielding, and a system for additional temperature monitoring. The applicator of the exposure system suitable for LF MF irradiation should meet the following demands: • The area of interest (AOI) with uniform exposure-the homogeneity of the MF represented by magnetic flux density field (B-field) at a minimum level of 98%; • Identical cultivation conditions for exposed and control samples; • The possibility to perform at least five experiments in parallel-five exposed and five control samples, to enhance the number of results for statistical evaluation and thus reduce the time required for experiments; • The applicator geometries optimized for an inner dimension of the incubator Q-Cell 240; • Sufficient airflow in the incubator for both control and exposed samples; • Safe handling with samples. Due to the above-defined requirements, a simple long cylindrical coil seems to be the most suitable choice for the applicator. Other groups reported the employment of Helmholtz coil for in vitro [25] or in vivo experiments [26], two flattened orthogonal coils [27], exposure system sXc-ELF composed of a set of square Helmholtz coils [28], also solenoid coil for in vitro experiments [29,30], a system of coupling coils with adjustable transfer distance [31], tetra-coil [32], and many more setups presented in [33][34][35][36][37]. Most of the mentioned systems are designed for irradiation of only one sample or multiple samples with a minimal volume using a microplate and do not consider the inhomogeneity of the B-field. To enhance the identical cultivation conditions for exposed and control samples, the laboratory incubator Q-cell 240 is chosen. The incubator maintains a stable temperature with an accuracy of 0.5 • C via forced air circulation in the chamber. The inner dimensions of the incubator 1315 × 504 × 320 mm 3 (height × width × depth), determine the maximal size of the coils. Therefore, to ensure sufficient air circulation in the coil cavity, the maximal possible length of the coil is 1000 mm. The inner diameter of the coil is determined by the pursuit to reach a high level of the magnetic field homogeneity-the smaller the diameter of the coil, the higher the homogeneity-and concerning the sample size (Erlenmeyer bank) and safe handling of them. The experiment used flasks with a volume of 100 mL, a height of 105 mm, and a diameter of 64 mm. Therefore, the inner diameter of the coil is set to 135 mm. The conceptual drawing of the cylindrical coil design is shown in Figure 1. The coil should cover the entire space of the samples to achieve the necessary uniformity of B-field within the cell culture samples; therefore, the edge samples should be located 100 mm from the respective edge of the coil. The level height of the culture medium placed in the bank is also considered to determine the position from the edge of the coil. The exposure coil is proposed to be driven by sinusoidal electric current to produce a high enough B-field within the irradiated volume but not affecting thermal conditions. For this purpose, tightly wound wire with a thickness of 1 mm and sufficient current load capacity is chosen, resulting in coils with 2000 turns each. Numerical Simulations of Exposure System Simulations are performed using two commercially available simulation software-CST Studio Suite and COMSOL Multiphysics to compare the results of numerical simulations of the identical model concerning the B-field distribution and its homogeneity in the AOI [38]. Each AOI is in real experiments Erlenmeyer flask filled with yeast extractpeptone-dextrose (YPD) medium with beer yeast cells Saccharomyces Cerevisiae. Within the simulations, the AOI is represented by a cylinder with dimensions of 68 mm in width, 40 mm in height, and a volume of 145 cm 3 and is filled with distilled water, Figure 2. Considering the percentage of distilled water in the YPD medium-95%, its substitution for distilled water, in terms of dielectric material properties, in the simulations, is permissible. Primary numerical simulations were realized to adjust the appropriate parameters of the system, where it was investigated how the input parameters influence the magnetic flux density distribution and its homogeneity [39]. The model includes two identical applicators-the exposed one and the control one, shielding between the coils, a metal incubator and ten samples representing the AOI, all depicted in Figure 2. The shielding with dimensions 1000 × 160 mm 2 (height × width) is placed exactly in the middle between the coils. The distance between the edges of coils is 174 mm. The internal walls of the incubator Q-cell 240 are made from plastic. It is not known which kind of material is behind the internal plastic wall, so the simulations are made for the worst situation. Thus, the case behind the internal plastic wall is made from iron. The incubator manufacturer did not provide a technical drawing of the internal composition of the incubator with a description of the material used. Therefore, the incubator is represented as a container with dimensions 1315 × 504 × 320 mm 3 (height × width × depth) in the numerical simulation. The walls of the incubator are modelled by a metal sheet with a thickness of 2 mm; the shielding is created from a metal sheet with a thickness of 3 mm. The electromagnetic parameters of the metal sheets at a frequency of 50 Hz are the ones from the simulation software library: This type of metal sheet is usually used in the manufacturing of refrigerators. Ninety evaluation points, nine in each AOI, are considered for calculating the homogeneity of B-field (HBF). The B-field as RMS is calculated at these evaluation points and based on these values the HBF is calculated in the exposure coil for each sample separately: where B C4 is the value of B-field exactly in the middle of the coil, sample C, point 4 (p4), Figure 2, since the reference value B C4 also represents the highest value of the B-field from all evaluation points, the value of B is substituted by the minimum value within a particular sample, so HBF represents the worst case of each sample's homogeneity. The whole numerical space is discretized using a tetrahedral mesh in the CST Studio, and the low-frequency domain solver employing a magnetoquasistatic equation is applied. Built-in function Symmetry Planes cannot be applied; the boundary condition in XY, YZ, and YX planes are set to the open ones. For this reason and due to available computing capacity, the simulations are performed with the accuracy of 10 −3 . However, simulation complexity and computational capacity limit the simulation accuracy. In COMSOL, the model is created in a 3D spatial layout discretized using an automatic generated tetrahedral fine mesh. Then, the multiphysical interface electromagnetic heating and the frequency-transient study are applied. The values of HBF calculated in CST Studio and COMSOL in the exposure coil are listed in Table 1. The results show that the applied B-field uniformity of the AOI of three middle samples B, C, and D in the exposure coil is better than 99.16%, in sample A is minimum homogeneity of 92.55%, in sample E is the homogeneity of 92.75%. Therefore, the three middle samples would be a suitable solution for the research focused on electromagnetic biocompatibility studies at a cellular level. The two edge samples should be considered in further experiments as the non-homogeneous application of B-field. However, the existence of two sets of samples-homogeneous and inhomogeneous-can offer interesting and important results, namely how the less homogeneity of the applied field affects cell growth. Since the control samples represent a state without intervention, the main objective for the control coil is to reach the magnitude of the B-field lower than the Earth's magnetic flux density value of 39 ± 5 µT, measured at the original experimental sample's location. The geomagnetic field measurement is performed with a Vernier Labquest3 measuring instrument incorporating a Vernier MG-BTA probe. The maximal values of B-field RMS values obtained by simulation in CST and COMSOL are listed in Table 2. The highest value of the B-field appeared in sample A at the level of 11.16 µT, which represents approximately 29% of the earth's magnetic field value. The results from both simulation programs are comparable in the center of the exposure coil and control coil, but values from the upper edge of the exposure coil differ significantly. The RMS values of the B-field obtained from both simulation programs differed in all evaluating points at most at 84 µT (4.94%) in the exposure coil and 4 µT (47.75%) in the control coil. The difference in the obtained values is acceptable and is caused by a different calculation method of the applied software programs, finite integration technique vs finite element method. The CST Studio is used for further verification of the EMF parameters of the exposure system. Figure 3 shows the variation of B-field distribution in the solenoid on the axis at the input current of 0.7 A versus distance of the sample's position in the coil cavity. The exposure coil meets the homogeneity criteria only in the three central samples, B, C, and D. The control coil meets the criteria of the maximum value of the B-field in all five samples. The variation of E-field distribution in the coil on the axis at the current input of 0.7 A versus distance of the sample's position in the coil cavity is shown in Figure 4. The E-field induced by the magnetic field (MF) generated by the exposure coil is of the order of 10 −4 V·m −1 in both coils. In the peripheral parts of the samples, there is a sharp increase and decrease in the monitored parameter due to the induced eddy currents in the samples, Figure 5. The E-field induced in the cell culture medium by the MF generated by the developed coil is in the order of 10 −4 V·m −1 . The reported threshold for an E-field to elicit a biological effect is estimated to be around 10 −4 V·m −1 for neutrophil cells [8] and used in research into umbilical cord blood lymphocytes [40]. Thus, it is questionable if the EF generated by the new exposure system can be neglected, and the results obtained for biocompatibility studies can be referred to be caused by the MF. On the other hand, the E-field in the presented exposure system is 10-11 orders of magnitude lower than the E-field across a 7.5 nm thick cell membrane [41] with a typical transmembrane potential of 50-120 mV, which turns out to be 6.67 MV·m −1 . Therefore, for the E-field calculation, the yeast cell is considered as a spherical capacitor according to: where U is the transmembrane potential, considered as 50 mV, R 2 is the external diameter of the cell representing the external electrode, and R 1 is the internal diameter of the cell representing the internal electrode. The R 1 value is calculated as a difference between R 2 and the thickness of the yeast cell membrane with a value of 7.5 nm. According to the calculation described above, the E-field induced by the new exposure system in the cell culture medium should not affect the processes in the cell membrane. In the next step of simulations, the frequency change's effect on the of the B-field's homogeneity is verified. Eight frequencies are used: 50, 100, 500, 1000, 1500, 2000, 2500, 3000 Hz. The results show that the change in frequency does not affect the homogeneity of the B-field. The temperature field distribution is evaluated using COMSOL in the samples after two, four, six, and eight hours from the start of the experiment, while the initial temperature is set to 30 • C. A graphical representation of the spatial temperature distribution in the AOI after eight hours of irradiation is shown in Figure 6. Any temperature variation and notable temperature differences between exposure and control AOIs are observed at a current of 0.7 A and a magnetic flux density of maximal value of 2.4 mT for the eight hours duration of the experiment. The temperature distribution is uniform. The exposure coil does not cause temperature increase due to Joule heating of the content in the samples. The proposed exposure system is suitable for cell electromagnetic biocompatibility research studies. The crucial findings from the simulations are that there is a negligible influence of the iron shielding and of the incubator's iron walls on the exposure level or homogeneity. Construction of Exposure System The construction of a designed exposure system for cell electromagnetic biocompatibility studies consists of constructing of a support system for Erlenmeyer flasks and of the construction of coils, the exposure one and the control one. The construction is followed by measuring of the magnetic flux density of both coils for verification of the numerical results. The resistance to temperature, acids, alkalis, alcohol, and inflammability is considered when selecting the material of the coil support structure. Polytetrafluoroethylene (PTFE) meets the specified requirements-it has high thermal stability from −200 • C to +260 • C, is chemically inert, nonflammable, and tolerated by living organisms [42]. The location of the irradiated samples is defined by exact positions by a support system (SpS). The SpS is designed and modelled in Inventor ® 3D CAD software. Therefore, it is necessary to make two support systems for each coil separately. The individual parts of the SpS are made of polyethylene (PE). Since the PTFE has a high surface tension, the SpS made from PE can be easily inserted into the coil cavity. The whole construction of SpS consists of four guide rails and five shelves. The SpS should secure safe handling with the yeast samples in the Erlenmeyer flasks. After the suitable simulation of the exposure system, technical documentation is prepared for processing the material and producing a supporting structure of the coils. The supporting structure of the coil is made of PTFE, composed of parts-semi-finished products in the form of eight hollow cylinders. The purchased semi-finished parts are processed on a lathe to the required size and then assembled into the final shape. Enameled copper wire of 1 mm diameter in a total length of 912 m for each coil is wound onto the coils supporting structure using the lathe. The completed pair of coils is shown in Figure 7. The location of the irradiated samples is defined on exact positions by a support system (SpS), also presented in Figure 7. The parts of the SpS are not permanently connected, the construction can be divided into individual parts and possibly modified. The SpS also provides easy and safe handling of the samples. Measurement of Magnetic Flux Density The magnetic flux density measurement is performed using the Narda NBM-550 EMF analyzer with the EPH-50D probe in the center of each coil for each sample position in the incubator with the door closed. The incubator contains the control coil, the exposure coil, and shielding at the time of measurement. The exposure coil is powered by an arbitrary wave generator RIGOL DG4162; the signal is amplified by a HUBERT KEYSIGHT Infinii Vision MSO-X 3012A linear amplifier while the driving current is controlled by a 6 1 /2 local digital multimeter Aligent 34401A. The exposure coil is powered by I = 0.7 A (time course of sine shape, RMS), f = 50 Hz. The conductors are attached to the coil through a drain channel at the rear of the incubator, which is used to drain excess water when the incubator is used for cooling so that the integrity of the incubator is not compromised. The measured data are compared with the results obtained from the numerical simulation at the evaluation curve corresponding to the size of the probe. The measured data of magnetic flux density are compared with the values obtained from the simulations, Figure 8. The percentage difference between the measured and simulated values does not exceed 11%, Figure 9. The maximum of measured B-field RMS values in the control coil is 18.47 µT, while the exposure coil is powered by I = 0.7 A (time course of sine shape, RMS), f = 50 Hz, with the incubator's door closed. The measured magnetic flux density inside the control coil is 90 times lower compared to the value measured in the same position in the exposure coil. For further using the exposure system, the values obtained by measurement are considered the reference values. A calibration curve is constructed to correctly determine the value of current required to obtain the desired value of magnetic flux density. Cell Electromagnetic Biocompatibility Experiment To examine the effect of LF MF on cells and verify the functionality of the proposed exposure system, 12 paired experiments are performed. The biological material used in the experiment is a pure strain, BY4741 yeast cells Saccharomyces cerevisiae. The yeast cells are obtained from the Slovak Academy of Science in Bratislava and refrigerated at 4 C in a Petri dish. The theoretical basis of the experiment comes from Lednev's IPR theory. The principle is in applying the MF generated by a parallel combination of static B DC and time-varying B AC MF. The theory of IPR suggests the nonthermal effect of MF on ions in living cells such that a protein-bound ion of mass m and charge q can be considered as a harmonic oscillator [12]. The correct application of the theory presupposes the application of timevarying MF with a frequency corresponding to the cyclotron frequency of the target bound ion, calculated according to the formula: where q is the electric charge of the target ion, m is the molecular mass of the target ion, and B DC is the value of static magnetic flux density-the geomagnetic field in this case. The target is calcium ion 40 Ca 2+ with a cyclotron frequency of 29.89 Hz. According to the IPR, the maximal biological effect should be reached at the B AC /B DC ratio of 1.8. For the earth's magnetic field value of 39 µT measured at the location of the experiment's realization, the corresponding value of B AC is 70.2 µT to maintain the ratio. The exposure coil is powered by an arbitrary wave generator RIGOL DG4162; the signal is amplified by a HUBERT KEYSIGHT Infinii Vision MSO-X 3012A linear amplifier. The driving current is controlled by a digital multimeter Agilent 34401A. The exposure coil is powered by I = 79.5 mA (time course of sine shape, RMS), f = 29.89 Hz. The multimeter is connected for the whole experiment duration to monitor the value of the supply current continuously. The control coil is without a power supply, and it is shielded. The control coil represents the state without intervention, and its aim is to show the difference in the results between the natural process of multiplication and the growth of yeasts affected by the MF. The YPD medium is used in the experiment; the solution consists of 95% distilled water, 2% peptone, 2% dextrose, 1% yeast extract. After the medium is prepared, a 24 h pre-cultivation period is followed, starting with the inoculation of yeast cells into a flask with 20 mL of prepared YPD. The yeast medium is placed on a shaker and cultured for 24 h at 180 rpm at an ambient room temperature of 23 • C. Pre-cultured yeast solution with a volume of 0.05 mL is pipetted into each of ten prepared Erlenmeyer flasks, 5 control, and 5 exposed samples containing 25 mL of medium. The flasks are closed with cotton-wool plugs to ensure air access. The samples containing cultured cells in the YPD medium are placed into a coil cavity and then moved into the incubator. The incubator ensures the maintenance of a stable temperature of 30 • C during the whole experiment. The temperature is continuously measured in both cavities of the incubator using temperature sensors DS18B20. The time of exposure to the MF is 8 h. To quantify the effect of applied MF, the numbers of cells in exposed vs control samples are compared. Cells are counted before and after each experiment in each sample using the Bürker chamber. The cells are photographed using Axiocam ERc 5 s in an inverted microscope Zeiss Primovert. The procedure for determining the effect of MF using the count of cells is depicted in Figure 10. If X = 1, the cells in exposure and control coil grow comparable. If the parameter X is more than 1, the excitation/growth stimulation effect of the applied field appeared. If the parameter X is less than 1, the inhibition effect of the applied field appeared. The realization of experiments corresponds to the usual laboratory practice with an emphasis on accuracy. To be able to repeat and verify the experiments, an experimental protocol incorporating a manual for implementing these experiments is developed. In addition, the necessary parameters are recorded for further comparison or repetition: the conductivity and temperature of the YPD medium before and after the experiment, the temperature during the whole experiment, and the photos of samples on the Bürker chamber of each sample before and after the experiment, and the count of cells of each sample before and after the experiment. The results of experiments for all samples with respect to parameter X are reported in Table 3. The observed effects of MF exposure in the first experiment represent, for samples D, C, and A stimulation 16%, 23%, 2%, and for samples E and B inhibition 43%, 35%. The reason why MF with the same parameters induces inhibition in one sample and stimulation in the other will be the subject of further research. Results and Discussion The exposure system allows exposure to homogeneous LF MF of 10 samples, of which five control/shielded and five are exposed, with yeast Saccharomyces cerevisiae simultaneously, while maintaining the same temperature conditions for both groupsexposure and control one has been developed. The exposure system is designed using numerical simulations. The key parameters are measured using constructed facility to confirm the design. The magnetic flux density field, electric field, and total current density field distributions are characterized along with the magnetic flux density field homogeneity and overall exposure uncertainty. The applied B-field uniformity of three middle samples in the exposure coil is better than 99.16%, and the maximum magnetic flux density in the control coil does not exceed 18.47 µT (0.75% of the maximum value of magnetic flux density in the exposure coil). The level of homogeneity of the three middle samples can be considered outstanding [43,44], and the homogeneity of samples A and E is satisfactory [45]. As for the value of the magnetic flux density field, these results cannot be confronted with the available data because other authors do not state its value; it has not been measured. The situation for the control samples is generally described as the control samples are not exposed to the field or the control samples are shielded. The impact of the switched-off apparatus was not considered [46][47][48]. To the shading of the B-field generated by the exposure system theoretically to zero, it would be necessary to use high-quality shading, e.g., MuMetal. However, it would result in the shading of the terrestrial field. This state is not desirable because it does not represent natural conditions. The natural external EMF, such as the terrestrial E-field and B-field, or the atmospheric EMF-Schumann resonances are vital/necessary for any cell's, microorganism's, animal's, and human's physiological functioning [49]. The functionality of the developed exposure system is tested and confirmed during the 12 experiments, each lasting eight hours. The coil current and temperature in the cavities are continuously monitored during the experiments. The experiments are based on the IPR theory. The effect of applied MF is determined based on the number of cells, as explained in the previous section. Cell numbers are determined by direct counting in a hemocytometer, the Bürker chamber. Results of these experiments have shown significant differences between the control and exposed group of samples, Table 3. Mild inhibition and excitation effects of the applied MF are observed. Other groups reported an inhibitory effect on the growth of yeast cells [45,50], no effect [51], or both effects depending on B-field magnitude [52]. The questions of how the MF affects the yeasts or cells in general, what frequencies, B-field magnitudes, or their combinations are responsible for inhibition, or acceleration growth are far from answered. This area needs systematic, wide, and in-depth research as well. To increase the validity of experiments performed using this exposure system, it is desirable to use several evaluation methods for cell culture growth [46]. In the time domain of the experiment, the provided exposure system offers an additional benefit. For example, in the previously used experimental set-up [53] with one exposed and one control sample, the time required to perform five experiments correctly was 128 h, including pre-cultivation. Using the developed exposure system presented within this article, the time required to perform the same number of experiments, including pre-cultivation, is reduced to 32 h, resulting in significant time savings and thus effectivity enhancement. It is also appropriate to point out the possible disadvantages of the proposed system. The proposed system is primarily intended for use in the area of low frequencies and low intensities of MF. The higher intensity MF could introduce a specific limitation to the presented system as the maximal current carrying capacity limits the coil design due to the copper wire. The solution that increases the MF in the sample area is to use a thicker wire and the Helmholtz configuration. It is questionable whether more accurate results can be reached using the multipurpose system (for low and high intensities) or a specific system designed directly for low and extremely low intensities. Due to the total cost of producing two coils and two SpS in the amount of 2153 €, it is more advantageous to use two different systems designed separately for low and extremely low intensities and the high intensities of the MF. As mentioned in Section 2, comparable systems for EMF irradiation are designed for irradiation of only one sample or multiple samples with a minimal volume using a microplate and do not consider the inhomogeneity of the B-field. Other systems offer possible advantages of lower costs in terms of materials use and power consumption. However, they cannot be compared in terms of homogeneity. Temperature changes are important experimental confounders in the research with yeast in general. Therefore, a stable and good controlled temperature is essential for the credible interpretation of results. To achieve minimum temperature fluctuation, a commercially available incubator is used. The incubator maintains a stable temperature of 30 • C. The temperature of the internal incubator environment (air temperature) is additionally monitored continuously during in vitro experiments by the temperature measurement system. The temperature measurement system consists of two temperature sensors, DS18B20, connected to the Arduino platform. The temperature is recorded on an SD card and shown on an OLED display every 10 min. Figure 11 shows the incubator air temperature course for eight hours lasting experiment; each point represents the average temperature from 12 experiments. The low temperature of 24 • C occurring at the start of each experiment is caused due to the opening of the incubator door and manipulation with samples. During the 8 h lasting experiment, the door remains closed. The temperature difference stabilizes after 4 h (∆T260) from the start of the experiment, where the temperature difference between control and exposed samples is less than 0.5 • C, and it is no longer increasing. The temperature of the YPD medium in exposed and control samples is measured before and after each experiment using the conductometer PC 70 Vio. The average YPD temperature difference between control and exposed samples A-E in 12 experiments before and after the experiment is shown in Figure 12. Figure 12. The average YPD temperature difference between control and exposed samples before and after the experiment. The average temperature difference between the exposed and control samples does not exceed 0.5 • C in both temperature sensors and conductometer. Achieved temperature differences are sufficient for the credible interpretation of yeast research results [54]. To the best of our knowledge, the system developed within this work, using live Saccharomyces cerevisiae cells intended for controlled LF MF exposure investigations based on the IPR theory, has not been described elsewhere and hence constitutes a new tool for cell electromagnetic biocompatibility studies. In addition, the system can be applied in both health risk assessment and investigations of claimed therapeutic effects of LF MF. Further research using the presented system will be aimed at performing a sensitivity analysis of the frequency response of the biological system in the form of yeast Saccharomyces Cerevisiae according to the IPR theory. Author Contributions: Conceptualization, methodology, software, validation, investigation, manuscript writing, visualization, Z.J. and L.C.; formal analysis, methodology, manuscript editing, supervision, L.J.; investigation, manuscript editing, R.R. All authors have read and agreed to the published version of the manuscript.	7,459.2	2022-07-06T00:00:00.000	[ "Physics" ]
Contribution of Pyk2 pathway and reactive oxygen species (ROS) to the anti-cancer effects of eicosapentaenoic acid (EPA) in PC3 prostate cancer cells Background n-3 polyunsaturated fatty acids (n-3 PUFAs), including eicosapentaenoic acid (EPA) and docosahexaenoic acid (DHA), are thought to exert protective effects in cardiovascular diseases. In addition, n-3 PUFAs have demonstrated anti-cancer effects in vitro and in vivo. Objective We investigated the anti-cancer effects and mechanism of action of EPA on PC3 prostate cancer cells in vitro. Methods PC3 cells were treated with various concentrations of EPA, and cell survival and the abilities of migration and invasion were evaluated. The time course of the growth inhibitory effect of EPA on PC3 cells was also assessed. The mechanism underlying the anti-cancer effects of EPA was investigated by human phosphokinase and human apoptosis antibody arrays, and confirmed by western blot analysis. We also examined the contribution of reactive oxygen species (ROS) to the effects of EPA using the ROS inhibitor N-acetyl cysteine. Results EPA decreased the survival of PC3 cells in a dose-dependent manner within 3 h of application, with an effective concentration of 500 μmol/L. EPA inhibited proline-rich tyrosine kinase (Pyk)2 and extracellular signal-regulated kinase 1/2 phosphorylation as determined by western blotting and the antibody arrays. The growth of PC3 cells was inhibited by EPA, which was dependent on ROS induction, while EPA inhibited Pyk2 phosphorylation independent of ROS production. Conclusions Inhibition of Pyk2 phosphorylation and ROS production contribute to the anticancer effects of EPA on PC3 cells. Introduction Prostate cancer (PC) is the second most common cancer in men worldwide, with the incidence increasing in Asian countries, including Japan [1,2]. Standard treatments for PC include surgery, radiation, hormone therapy, and chemotherapy. Advanced PC with metastasis is treated with androgen deprivation therapy (ADT) in association with medical or surgical castration [3]. However, after several years of treatment, patients ultimately develop castration resistance; non-metastatic or metastatic castration-resistant (CR) PC is refractory to ADT and develops mechanisms to proliferate irrespective of castration. Furthermore, anti-androgen therapy can worsen the patient's condition by stimulating PC growth. This has been demonstrated with flutamide therapy and is known as anti-androgen withdrawal syndrome [4]. Only a few chemotherapeutics and radiation have been developed for the treatment of CRPC [5], and these have not improved the poor prognosis of this disease [6]. There is growing evidence that n-3 polyunsaturated fatty acids (n-3 PUFAs) found in fish oil-especially eicosapentaenoic acid (EPA) and docosahexaenoic acid (DHA)-can improve lipid metabolism and blood lipid profiles, prevent the progression of atherosclerosis, and reduce the incidence of cardiovascular diseases [7] as well as liver and plasma triglyceride levels [8]. We previously demonstrated that chronic oral administration of EPA prevented endothelial dysfunction in a mouse model of type 2 diabetes [9]. n-3 PUFAs have also been reported to have various anti-cancer effects in several types of malignancy in vitro and in vivo. n-3 PUFAs inhibit extracellular signal-regulated kinase (ERK) and Akt signaling pathways and show anti-cancer effects in breast cancer [10], reduce the incidence of liver cancer in hepatitis virus-infected patients [11], and lower the risk of pancreatic cancer [12]. Cis-unsaturated fatty acids including EPA are easily incorporated into cancer cells and induce free radical generation, inducing tumoricidal action [13][14][15][16]. However, other studies have suggested that there is insufficient evidence for a significant association between n-3 PUFAs and cancer incidence [17]. We previously reported that EPA and DHA have anti-proliferative, −migratory, and -invasive effects in the PC3 CRPC cell line [18]. We also observed that the combination of anti-cancer drugs and n-3 PUFAs synergistically inhibited the proliferation of PC3 cells (unpublished data). However, the mechanism of action of n-3 PUFAs, and specifically, the molecular basis for the inhibitory effects of EPA on PC3 cell proliferation, migration, and invasion has yet to be defined. Various signaling pathways control cell proliferation; their dysregulation can lead to over-proliferation and aberrant migration and invasion. Tyrosine kinase (TK) is an important mediator of cell proliferation. Receptor (R) TK is a transmembrane receptor with TK activity, whereas non-receptor (NR) TK is present in the cytoplasm and comprises Abl, Src, and prolinerich TK (Pyk)2 families. Various NRTKs are expressed in PC3 cells, and Src plays a role in multiple biological processes in PC cells [19]. Activation of these TKs leads to cell proliferation through activation of transcription factors that promote cell cycle progression. NRTKs enhance the proliferation of cancer cells to promote their survival, so suppressing these pathways can lead to cell growth inhibition or death. Most cancer cells are resistant to apoptosis and activate pathways that suppress the pro-apoptotic active form of caspases [20]. Moderate ROS levels may cause tumorigenesis through DNA damage in protumorigenic cells [21]. Thus, eliminating or reducing ROS production is a potential strategy for cancer therapy, since ROS also regulate the activation of several RTKs and NRTKs [22]. On the other hand, EPAinduced ROS overproduction stimulates apoptotic signals in HepG2 liver cancer cells [23], and oxidative stress causes PC3 cell death by stimulating mitochondrial ROS production and apoptosis [24]. DHA has been found to cause cell death by inducing ROS production in PC3 and DU145 PC cells [25]. In the present study, we investigated the mechanism underlying the anti-cancer effects of n-3 PUFA EPA, first by characterizing the optimal concentration and time required for these effects in PC3 cells, and then by analyzing protein expression by western blotting and antibody arrays. To clarify the effects of ROS production induced by EPA, we used the ROS inhibitor N-acetyl cysteine (NAC). Cell culture PC3 cells (Riken BRC Cell Bank, Tsukuba, Japan) were grown in DMEM (Gibco) supplemented with 10% heatinactivated fetal bovine serum (FBS), 100 U/mL of penicillin, and 100 μg/mL of streptomycin (Gibco). The cells were cultured at 37°C in a humidified 5% CO 2 atmosphere. invasion. Double-chambered cell culture dishes with a transwell insert separating the two chambers were used to evaluate PC3 cell migration and invasion. Cells were seeded in the upper chamber, which was uncoated (migration) or coated (invasion) with Matrigel, while the lower chamber was filled with DMEM containing 10% FBS. Data represent mean + SEM (n = 3). P < 0.05 Cell proliferation assay PC3 cells were seeded in 35-mm dishes at a density of 3 × 10 5 cells/dish in DMEM containing 10% FBS and incubated for 24 h, and the culture medium was replaced with serum-free DMEM after washing with PBS. After an additional 24 h, the cells were washed with PBS, and serum-free medium containing 3% BSA and various concentrations of EPA (100-500 μM) were added. After 24 h, the number of cells was counted with a hemocytometer. Time course of cell proliferation PC3 cells were seeded in a 35-mm high μ-Dish Grid-500 (Ibidi, Planegg, Germany) at a density of 3 × 10 5 cells/dish in DMEM containing 10% FBS. After 24 h, the cells were washed with PBS, and serum-free DMEM was added; after another 24 h, the cells were washed with PBS, and serum-free medium containing 3% BSA with or without EPA (500 μM) was added. Cells were imaged with a microscope and counted at the indicated times. Migration and invasion assay Migration and invasion were assessed as described in our previous study [18]. Briefly, we used Transwell membranes (8-μm pore size; Corning Inc., Corning, NY, USA) with or without Matrigel coating. PC3 cells were cultured in the upper chamber of a Transwell insert in serum-free medium containing various concentrations of EPA. The lower chamber was filled with DMEM containing 10% FBS. For the migration assay, Matrigel-uncoated upper chambers were used; cells were seeded at a density of 2 × 10 5 cells/mL, and after 24 h, the number of cells that had migrated into the lower chamber was counted. For the invasion assay, Matrigel-coated upper chambers were used. Before seeding the cells, Matrigel was fixed by incubation for 90 min, and cells were seeded at a density of 1 × 10 5 cells/mL. After 48 h, cells that had invaded into the lower chamber were counted. Antibody arrays PC3 cells were cultured in DMEM containing 10% FBS for 24 h, then cultured in serum-free medium for Fig. 4 Results of the apoptosis antibody array. PC3 cells with or without EPA treatment were analyzed. The y axis shows the pixel density (%). The expression level of each protein in the control was adjusted to 100% an additional 24 h before treatment with EPA (500 μM) or vehicle as a control. After 2 h of incubation with EPA or vehicle, samples were prepared for the antibody array according to the manufacturers' protocols. Chemiluminescence detection was performed on a Lumi Cube (Liponics, Tokyo, Japan) using Clarity Western ECL Substrate (Bio-Rad, Hercules, CA, USA). Images of the arrays were analyzed using Image J software (National Institutes of Health, Bethesda, MD, USA). Lumi Cube and Clarity Western ECL substrate were used for signal detection, and protein band intensity was analyzed using Image J software. Statistical analysis Statistical analyses were performed using R v.3.4.3 (https://www.r-project.org/). Data are presented as means ± standard error, and differences between means were evaluated with Dunnett's test. A P value less than 0.05 was considered statistically significant. EPA suppresses PC3 cell proliferation EPA inhibited the proliferation of PC3 cells in a dosedependent manner (Fig. 1). However, only the highest concentration of EPA (500 μM) significantly reduced the number of cells to 50% of the control value. In serum-free medium, the number of PC3 cells in the control group did not change over a 24-h period, but gradually decreased with EPA treatment, with significant differences detected after 3 h. After 12 h, the number of cells was 50% of the control value and remained constant up to 24 h (Fig. 2). EPA inhibits PC3 cell migration and invasion EPA inhibited both PC3 cell migration and invasion in a dose-dependent manner, and the rate of migration and invasion relative to the control was 43% (Fig. 3a) and 26% (Fig. 3b), respectively, at 200 μmol/L EPA. Thus, PC3 cell invasion was inhibited to a greater extent than migration by EPA. Proteins associated with phosphokinase and apoptosis pathways are activated by EPA treatment Using two types of antibody arrays, we determined that proteins involved in the phosphokinase and apoptosis pathways were activated by EPA treatment. Based on the results shown in Fig. 2, protein samples for antibody arrays and western blotting were prepared 2 h after adding 500 μmol/L EPA. More significant results were obtained between the EPA-treated and untreated Table 2 Differences in the expression levels of various proteins in EPA-treated PC3 cells relative to the control (related to Fig. 5) 1. Differences relative to the control are shown as ± ± ± ± > 20% 2. Differences relative to the control are shown as ± ± ± > 10% 3. Differences relative to the control are shown as ± ± > 5% control groups using the phosphokinase as compared to the apoptosis array (Figs. 4 and 5). The expression of apoptosis-related proteins including soluble tumor necrosis factor receptor (sTNF-R)2, second mitochondriaderived activators of caspase (SMAC), Survivin, tumor necrosis factor related apoptosis-inducing ligand receptor (TRAIL-R)1, caspase 8, B-cell lymphoma (Bcl)-2, and p53 differed by more than 20% between the two groups (Table 1), with sTNF-R2 and Bcl-2 being downregulated and the other proteins being upregulated by EPA treatment. The phosphokinase array data showed increased or decreased phosphorylation of several kinases (Fig. 5 and Table 2). EPA reduced the phosphorylation of several proteins by over 20% relative to the control, including Pyk2, endothelial nitric oxide synthase (eNOS), c-Jun, and ribosomal S6 kinases (RSK)1/2/3. Contribution of ROS to PC3 cell death We next analyzed the contribution of ROS to the effects of EPA on PC3 cell survival and found that EPA treatment decreased survival to 60% of the control rate, whereas NAC (5 mmol/L) had no effect. Treatment with both EPA and NAC restored the survival rate to the control level. Thus, NAC abrogates the anti-cancer effect of EPA against PC3 cells (Fig. 6). EPA inhibits ERK and Pyk2 phosphorylation To confirm the mechanism of action of EPA, cells were treated at a concentration of 500 μmol/L, and the expression of native and phosphorylated ERK and Pyk2 was examined by western blotting. EPA significantly suppressed ERK phosphorylation at 1 and 2 h, although there was no difference relative to the control group after 3 h (Fig. 7). A similar significant inhibitory effect was observed on Pyk2 phosphorylation at 2 and 4 h (Fig. 8). Treatment with the ROS inhibitor NAC (5 mM) alone did not affect Pyk2 phosphorylation, while NAC slightly enhanced the inhibitory effect of EPA on Pyk2 phosphorylation (Fig. 9). Discussion The anti-cancer effects of n-3 PUFAs in vitro and in vivo are well known [13,14,18,[26][27][28][29][30][31], and these effects have been observed in PC cells. Our previous study showed that the n-3 PUFAs EPA and DHA inhibit PC3 cell proliferation, migration, and invasion [18]. Several reports have indicated that n-3 PUFAsinduced apoptosis or oxidative stress suppresses the growth of various types of cancer cells, including PC cells [32,33]. Activation of NRTKs may be related to PC cell growth, and clinical trials of therapeutics that target NRTK are currently underway [34,35]. Other mechanisms of action have been proposed for the anti-cancer effects of n-3 PUFAs [36,37]. In this study, we addressed this question using phosphokinase and apoptosis antibody arrays and examined the contribution of ROS to the anti-cancer effect of EPA on PC3 cells. For the antibody array analysis, the cell cycle of PC3 cells was synchronized by culturing in serum-free medium for 24 h. PC3 cells are relatively resistant to the anti-cancer effects of EPA. We treated cells with a moderately high concentration of EPA that was nonetheless within the physiological range (500 μM). PC3 cell proliferation did not change over 24 h in the control group, which is consistent with the notion that the rates of proliferation and death are equivalent in PC3 cells grown under conditions of moderate nutrient starvation. In contrast, the number of cells in the EPA-treated group decreased over time, with significant differences relative to the control group observed from 3 h to 24 h. The results of the apoptosis array showed that EPA increased the expression of the pro-apoptotic proteins TRAIL-R1, caspase 8, and p53, and decreased that of the anti-apoptotic protein Bcl-2. However, EPA also increased the levels of the anti-apoptotic proteins SMAC and Survivin while having no effect on caspase 3. We speculated that ROS induced cell death via a pathway independent of caspase-induced apoptosis. To test this hypothesis, we evaluated the effect of the ROS inhibitor NAC on EPA-treated cells. First, we measured ROS generation at two time points, 2 and 24 h. In these experiments, we observed relevant (not significant) ROS generation induced by EPA only at 2 h, and NAC reduced the ROS level to the control (data not shown). We observed that the addition of NAC abrogated the anti-cancer effects of EPA (Fig. 6), suggesting that ROS induced by EPA suppressed PC3 cell proliferation. This is supported by previous reports showing that the anti-cancer effects of n-3 PUFAs are mediated by ROS in liver cancer and PC cells [23,24]. We also investigated the contribution of cell proliferation pathways to the anti-cancer effects of EPA using a phosphokinase antibody array. The results showed that Pyk2, eNOS, c-Jun, and RSK1/2/3 were downregulated by over 20% by EPA treatment as compared to the control group. Western blot analysis revealed that ERK and Pyk2 phosphorylation were inhibited by EPA (Figs. 7 and 8). Suppressing ROS production with NAC did not affect the inhibitory effects of EPA on Pyk2 phosphorylation, suggesting that the suppression of Pyk2 phosphorylation by EPA does not involve modulation of ROS. Pyk2 is a member of the focal adhesion kinase (FAK) family of NRTK and plays important roles in cell survival, proliferation, and migration [38][39][40][41]. In our experiments, FAK phosphorylation was also suppressed by EPA treatment (Table 2). FAK is localized at the site of cell adhesion with the extracellular matrix along with integrins, whereas Pyk2 associates with integrin within Fig. 7 Effect of EPA on ERK1/2 phosphorylation. Upper panel: Representative results of western blot analysis. Lower panel: Pixel density of the control group at the indicated times was adjust to 100%; relative pixel density of EPA-treated PC3 cells is shown. Data represent mean + SEM (n = 3). P < 0.05 the cell [42,43]. Since several types of integrins are expressed in PC3 cells and interact with extracellular matrix, inhibition of activated FAK at these sites may prevent cell adhesion and migration [44]. The Nterminal domain of FAK has phosphorylation sites for the cell proliferation-related factors epidermal growth factor receptor, c-Met, and Src [45]. The antibody used in the antibody array detected phosphorylated tyrosine 402 of Pyk2, which is necessary for Src activation [40] and was slightly reduced in our experiment (Table 2). It was reported that Pyk2 activation induced smooth muscle contraction by activating Rho-associated kinase signaling [46], and FAK and Pyk2 show similar activities on this pathway in the regulation of cell migration [38,47,48]. Activation of both factors leads to abnormalities in cell morphology and movement associated with malignant transformation. Our results suggest that EPA negatively regulates the proliferation, migration, and invasion of PC3 cells via inhibition of Pyk2 and FAK [18]. Pyk2 is an NRTK that is activated by increased intracellular Ca 2+ concentrations [49,50], which are in turn increased by n-3 PUFAs and ROS [23,[51][52][53]. In our results, the n-3 PUFA EPA inhibited Pyk2 phosphorylation, whereas ROS inhibition by NAC did not affect the inhibitory effects of EPA, suggesting that EPA blocks Pyk2 activation independent of intracellular Ca 2+ levels. Conclusions Our results demonstrate that EPA induced cell death by stimulating ROS production in PC3 cells, and suppressed Pyk2 phosphorylation inducing inhibition of cell proliferation, migration, and invasion partly controlled by downstream Pyk2 signaling. Our findings were summarized in Fig. 10. These results provide a basis for the development of novel therapeutics for the treatment of PC.	4,188.8	2020-01-31T00:00:00.000	[ "Medicine", "Biology", "Environmental Science" ]
Nanoscale nonlinear plasmonics in photonic waveguides and circuits Optical waveguides are the key building block of optical fiber and photonic integrated circuit technology, which can benefit from active photonic manipulation to complement their passive guiding mechanisms. A number of emerging applications will require faster nanoscale waveguide circuits that produce stronger light-matter interactions and consume less power. Functionalities that rely on nonlinear optics are particularly attractive in terms of their femtosecond response times and terahertz bandwidth, but typically demand high powers or large footprints when using dielectrics alone. Plasmonic nanostructures have long promised to harness metals for truly nanoscale, energy-efficient nonlinear optics. Early excitement has settled into cautious optimism, and recent years have been marked by remarkable progress in enhancing a number of photonic circuit functions with nonlinear plasmonic waveguides across several application areas. This work presents an introductory review of nonlinear plasmonics in the context of guided-wave structures, followed by a comprehensive overview of related experiments and applications covering nonlinear light generation, all-optical signal processing, terahertz generation/detection, electro optics, quantum optics, and molecular sensing. Introduction Photonic waveguides are a ubiquitous building block of optical circuits, used from passive long-haul data transfer in optical fibers, to active nanoscale signal processing on miniaturized planar architectures. The idea of using micro-scale circuits to manipulate optical signals from lasers dates back to the 1960s [1], and is now an established and powerful technological platform [2,3]. Such photonic integrated circuits (PICs) now routinely carry, route, and process light via guided waves-using both passive [4] and active [5] components-on a convenient monolithic chip, shown in the Schematic of Fig. 1a. PICs can be implemented using a number of dielectric platforms, including III-V semiconductors [6,7] lithium niobate [8], silicon [3,9] and silicon nitride [10], to name a few-some of which are compatible with complementary metal-oxidesemiconductor (CMOS) fabrication processes [11]. PICs find numerous applications across multiple disciplines [12] including telecommunications [13], quantum technologies [14,15], sensing [16], and machine learning via programmable PICs [5]. Inspired by this approach, optical fibers are also increasingly expanding their traditional guidance capabilities to include active components via metallic, semiconductor, or highly nonlinear materials. A concept schematic of such a hybrid optical fiber (HOF) [17], is shown in Fig. 1b. The main advantage of waveguide-based PICs over their electronic counterparts is their ability to directly manipulate analogue information that is encoded in photons, which are stable, robust to noise, and have high bandwidth. In recent years, the density of components which perform various functions, shown in the schematic of Fig. 1c, has rapidly increased, up to thousands of photonic components per chip [22], integrated with millions of electronic transistors [23]. Some operations, such as splitting [24], coupling [25], polarization rotation [26], filtering [27], and phase shifting [28], can [17], including nonlinear (blue) and plasmonic (purple) waveguides considered in this review. Adapted from Ref. [17] with permission. c Schematic example of PIC linear-and nonlinear-elements. Adapted from Ref. [15] with permission (Copyright The Optical Society). In this review, we consider d hybrid, guided wave nonlinear plasmonics structures (left) where extreme, nanoscale field enhancements near the metal (middle) can lead to nanometre-scale nanophotonics via highly nonlinear materials (HNLM) [19]. Adapted with permission from Ref. [19] (Copyright The Optical Society) and Ref. [20,21] (Copyright (2018) American Chemical Society) be entirely passive, agnostic to the amount of power guided by the PIC. In contrast, functions such all-optical switching [29] and light generation [30], are intrinsically active. Since photons do not interact with each other, this manipulation requires an interaction with the optical medium itself, which in its fastest incarnation occurs through the nonlinear optical response at the atomic or molecular level [31]. Although nonlinear responses naturally occur at ultrafast timescales and favour high-bandwidth applications, they are also exceedingly weak, and only become significant for large field intensities. Much effort has been dedicated to finding ways to increase optical nonlinear effects, either by developing new materials with intrinsically high nonlinearities [32] or by appropriately engineering highly nonlinear waveguides [33]. As one perspective describes [34], silica fibers proved to be a valuable platform for many early nonlinear waveguide experiments: although the nonlinearity of silica is low, the development of ultra-low-loss fibers in the 1970s allowed the observation of numerous nonlinear effects including stimulated Raman scattering, self-phase modulation, four-wave mixing and stimulated Brillouin scattering, as well as the first observation of solitons [35] and supercontinuum [36]. All these effects typically require long fiber lengths, and their operational principles crucially rely on the subtle interplay between nonlinear-and dispersive-effects after metres or even kilometres of propagation. In the past decade or so, much effort has thus been dedicated to miniaturizing and integrating these nonlinear functions on readily available chip-scale waveguide elements and circuits composed of highly nonlinear materials such as silicon [37,38] or chalcogenide [32]. Although progress in fabrication has resulted in low linear losses over typical propagation lengths, nonlinear performance is often limited by the materials' nonlinear losses (e.g., two-photon-and free-carrier-absorption), though mitigation strategies have been proposed [37]. One obvious advantage of waveguide systems over their bulk counterparts is their ability to maintain a constant spot size upon propagation, via the guided mode: since nonlinear effects demand high field intensities, they are strongest in devices supporting small mode areas. Pushing this concept to its limit, the degree to which any alldielectric PIC can be miniaturized is inherently restricted to approximately half the wavelength in the medium: if a waveguide lateral dimension falls below this limit, light is no longer tightly confined inside the waveguide and leaks externally [39]. In siliconbased PICs, for example, the lateral dimensions used for guiding telecommunications wavelengths are ∼ 0.5 µm. We refer the reader to Ref. [40] for a detailed discussion on important matters relating to all-dielectric nonlinear sub-wavelength photonic circuits. Truly nanoscale modal confinement can thus only be achieved by using metals: photons can couple to oscillating charges at metallic surfaces, giving rise to surface-plasmon polaritons (SPPs) which can have extremely small effective modal areas-orders of magnitude below the diffraction limit [41]. As such, SPPs have long been eyed as prime candidates for nano-PIC building blocks [42,43]. In this case, holding back immediate uptake is the large linear optical loss that accompanies extreme confinement, due to intrinsic electron damping [44]. In the worst case scenario, propagation lengths at metal-dielectric surfaces can be smaller than the wavelength itself. Despite this significant disadvantage, plasmonics continues to attract a lot of attention [45], and is frequently pointed to as the transformative platform for addressing inherent limitations of all-dielectric nonlinear devices [46]. The hope is that, although longrange propagation is out of the question, perhaps local field amplitudes can be large enough to make it all worthwhile. Researchers have thus harnessed localized SPPs that oscillate on individual metallic nano-elements without propagating [47]. Indeed, nonlinear plasmonic nanoantennas [48], metasurfaces [49], and metamaterials [50] have all been the subject of intense theoretical and experimental investigations. For overviews of nonlinear plasmonics, we refer the reader to Refs. [51][52][53][54]; for a comprehensive review of plasmonics in photonic integrated circuits, we refer the reader to Refs. [55][56][57][58]. But what are the prospects for integrating nonlinear plasmonic functionality on a chip for nanoscale nonlinear optics? In first instance, the answer is simple: place a plasmonic element close to a dielectric waveguide [59] and harness the resulting non-linear process via the localized surface plasmon. Although this approach can enhance the nonlinear performance of dielectric waveguides [59], only a small fraction of total power guided by the dielectric is used. An alternative approach takes a seemingly long-winded route: the diffraction-limited photonic mode can be transformed into a sub-diffraction plasmonic mode (e.g., via a directional coupler [60,61], adiabatic transformers [62,63], or end-fire [64,65] and perpendicular [66][67][68] couplers), which all guide light to a nano-volume. Photonic-to-plasmonic mode conversion schemes typically require as little as one wavelength of propagation, but still transfer a high fraction of power to a plasmonic nano-concentrator (close to 100%, when combined with mode-matching schemes [65]). For reviews on photonic-to-plasmonic nanocoupling schemes, see for example Refs. [69,70]. Owing to the hybrid nature of the waveguides involved, the vastly different optical properties of each participating material, and the co-existence of two mutually opposing effects (namely, high intensities and large losses), describing the nonlinear effects in plasmonic demands careful consideration. With a number of excellent reviews on nonlinear plasmonics [51,54] and nonlinear metasurfaces [71], here we concentrate on nonlinear plasmonics in guided-wave systems, with an eye on photonic integrated circuits. One example structure, formed by a metal-dielectric-metal nonlinear gap on top of a guiding silicon nanowire, is shown in Fig. 1d: it can produce extreme field enhancements in a guided chip platform, potentially enabling giant nonlinear optics when combined with highly nonlinear materials [19,72,73]. Although not all structures discussed will be on PICs, we have selected theory and experiments which reveal the underlying physics that should be considered in the context of propagating nonlinear SPPs, and is thus relevant to photonic integration. The outline of this review is as follows. In Sect. 2 we review the linear properties of several representative plasmonic waveguides, and introduce some important parameters impacting their nonlinear performance. In Sect. 3 we give a general overview of nonlinear optics, with particular attention to the Kerr nonlinear response of lossy, hybrid, guided wave systems. We also discuss the relative influence of typical materials, and other nonlinear effects. In Sect. 4 we present salient experiments in guided-wave nonlinear plasmonics. In Sect. 5 we present an experimental overview of photonic-plasmonic nonlinear circuits for nanoscale nonlinear light generation, alloptical switching, electro-optic functions, terahertz generation/detection, and Raman spectroscopy. In Sect. 6 we provide a brief perspective on nonlinear plasmonics in the context of quantum PICs, and conclude in Sect. 7. Fundamentals of plasmonic waveguides We begin by reviewing the fundamentals of plasmonics waveguides, with particular attention to those parameters that are most relevant for enhancing nonlinear lightmatter interactions. With a large number of excellent recent reviews on plasmonic waveguides, we hope to avoid redundancy by concentrating on those parameters most relevant to our later discussion on nonlinear optics: linear propagation loss, group velocity, and effective modal width. In first instance, we can distinguish two common classes of chip-scale plasmonic structures: (1) by one metal and one dielectric; (2) hybrid plasmonic waveguides, which harness multiple materials in often sophisticated arrangements, with the ultimate objective of reducing losses and maintaining nanoscale confinement. Unless otherwise stated, in this Section we consider waveguides supporting 1D modes and 2D propagation. This approach allows rapid calculations of both propagation constants and associated modes via numerical solutions of analytical functions [74], retaining much of the underlying physics while reducing the number of degrees of freedom to choose from. Pure plasmonic waveguides The archetypal plasmonic waveguides supporting deep sub-wavelength plasmon modes [75] are the metal-dielectric (MD), the dielectric-metal-dielectric (DMD), and the metal-dielectric-metal (MDM) waveguides. We revisit their most important mode properties, taking the opportunity to compare with their dielectric counterparts where appropriate. Bulk surface plasmon polaritons (MD) We start with the simplest plasmonic waveguide, shown in the schematic of Fig. 2a: a semi-infinite metal/dielectric interface supporting a transverse magnetic (TM) surface plasmon polariton (SPP) mode, propagating in z. The dispersion relation of SPP modes 200 A. Tuniz has a closed-form expression given by [41] where β is the propagation constant, from which the effective index n eff can be obtained via the vacuum wave number k 0 = 2π/λ (vacuum wavelength: λ), and ε m (ε d ) are the relative dielectric permittivity of the metal (dielectric). In these calculations, we consider the metal to be gold, one of the most commonly used plasmonic materials as a result of its high stability and relatively low loss, taking the measured values for ε m (λ) shown in Fig. 2b [76]. Fig. 2c shows the real part of n eff and associated attenuation length L att = 1/[2 m(β)] as a function of wavelength. At long wavelengths, ε m is large and negative, so that n eff ∼ k 0 √ ε d . Approaching the visible, ε m + ε d → 0 leads to an increase in n eff , limited by material losses via m(ε m ). Fig. 2d shows a colorplot of the associated electric field magnitude as a function of position and wavelength for modes of equal power: in the near-infrared, the electric field is weakly transversely confined to the metal (metal penetration depth: 20-30 nm); towards visible wavelengths, the field is increasingly confined at the metal/dielectric interface and produces a local intensity enhancement. Note that this effect occurs for both transverse-and longitudinal-field components [77]. We quantify this by calculating the group velocity v g = ∂ω/∂β and effective modal width w eff , respectively. Low v g is associated with slow light [78], which leads to longitudinal enhancement via the trailing edge of a pulse's field catching up with its leading edge; a small effective modal width w eff also enhances the electric field via transverse confinement [79]. The group velocity v g (normalized to the speed of light c) and w eff (here taken as the 1/e width of \|E\|) are shown in Fig. 2d: both have a global minimum close to resonance where ε m = −ε d . Although the SPP mode is a valuable starting point for the discussion, it approaches a weakly-guided surface wave at longer near-infrared wavelengths where many PICs operate. Field enhancements occur by reducing the waveguide features to subwavelength dimensions, as we now discuss. Thin metal plasmonic waveguides (DMD) We now consider the salient properties of modes supported on thin metallic films at the standard telecommunication wavelength λ = 1.55 µm [80]. Here the complex propagation constant is obtained from the numerical solution of an analytical transcendental equation [74]. As the infinite gold film of Fig. 2a transitions into a finite thickness nanofilm, the two supported modes on either side of the film can couple via their evanescent tails, giving rise to anti-symmetric-and symmetric modes (with respect to H ), analogously to what occurs for two coupled dielectric waveguides. These are referred to as the short-range (SR-) and long-range (LR-) SPPs, respectively, although other nomenclatures exist [81]. The SR-SPP possesses the most striking characteristics: Fig. 3a shows calculated e(n eff ) and associated attenuation lengths L att as a function of film thickness t = 1−50 nm. As the phase velocity decreases (large n eff ), the losses also increase (short L att ). This, in turn, is accompanied by a dramatic reduction in both the w eff and v g (Fig. 3b) indicating omnidirectional field enhancements at the metal-dielectric boundary as per the SPP. Figure 3c shows the associated Poynting vector magnitude \|S\| on a logarithmic scale, illustrating the dramatic increase in confinement of SR-SPPs for nanoscale metal thickness. The increased losses are a direct result of at larger fraction of modal power in the metal, although the largest fraction of power is in the surrounding dielectric. Note that the smallest effective width here corresponds to λ/20, one order of magnitude below the diffraction limit in free space. For comparison, Fig. 3d shows that the LR-SPP e(n eff ) decreases as the film thickness is reduced, and its attenuation length increases. As Fig. 3e illustrates however, the effective lateral modal width increases to several wavelengths, and v g /c approaches unity. As the Poynting vector colorplot of Fig. 3f reveals, here the field is not confined to the metal surface as the film thickness is reduced, in sharp contrast to the SR-SPP. As final comparison, Fig. 3g-i show equivalent calculations for the fundamental mode supported by an all-dielectric air-clad silicon waveguide (refractive index: 3.5; t = 100−600 nm). Reducing the waveguide width below 100 nm results in an effective index approaching unity (Fig. 3g), and a local minimum in v g and w eff at t ∼ 200 nm (Fig. 3h). Though this minimum is associated with field enhancements in the dielectric (shown in the Poynting vector colourplot of Fig. 3i), w eff , v g , and t are orders of (e) Fig. 4 a e(n eff ) (red) and attenuation length L att (blue) of the fundamental mode a metal-dielectric-metal (gold/air/gold) waveguide as a function of air gap width t at λ = 1.55 µm. b Effective width w eff and group velocity \|v g \|/c as a function of t. c Colourplot of the Poynting vector magnitude (log 10 \|S\|) as a function of t. d-f Equivalent calculations for the fundamental mode of a dielectric slot waveguide (silicon/air/silicon). g-i Equivalent calculations for the fundamental mode of a hybrid plasmonic waveguide (silicon/air/gold). All calculations are performed with silicon or gold slabs of width w = 300 nm magnitude larger than those of the SR-SPP. The absence of material losses comes at the cost of increased physical dimensions: the relative trade-offs between device footprint and associated losses are a recurring motif when comparing dielectric-and plasmonicwaveguides [43], which is especially relevant for integrated nonlinear plasmonics [82]. Plasmonic slot waveguides (MDM) The last pure plasmonic strucure we discuss is the plasmonic slot waveguide [83]. We consider the fundamental mode of a sub-wavelength air slot surrounded by two optically thick gold films at λ = 1.55 µm. Here the gold/air SPP modes on either surface also couple as they are brought together, giving rise to symmetric-and anti-symmetric modes (with respect to the magnetic field): the former produce sub-wavelength lateral confinement and low group velocity. Figure 4a shows calculated e(n eff ) and associated L att of the fundamental MDM mode as a function of sub-wavelength gap thickness (t = 1−200 nm): both increase as t approaches the single-nanometre scale, showing a dramatic reduction in both the w eff and v g , as plotted in Fig. 4b. Figure 4c shows the associated normalized Poynting vector magnitude: the majority of the power remains inside the slot, and the effective width nominally corresponds to the width of the plasmonic gap, which can be orders of magnitude below the diffraction limit. Although a significant portion enters the metal leading to large absorption and short L att (Fig. 4a, blue line), one theoretical study of tapered MDM waveguides [84] showed that, for certain tapering angles, a nonlinear dielectric in the slot could significantly mitigate mode attenuation by exciting a spatial plasmon soliton [85]. A comparable all-dielectric structure is the dielectric slot waveguide [86] shown in Fig. 4d, which uses a high-index dielectric (refractive index: 3.5, w = 300 nm) instead of gold. The continuity of the displacement field leads to an enhancement of the electric field inside the slot, by a factor corresponding to the ratio of the permittivity of each dielectric [86]. Incorporating a low-index, high-n 2 organic dielectric in a silicon slot can thus already significantly enhance its nonlinear optical properties [87]. Figure 4e shows that w eff and v g decrease as t approaches nanometre dimensions, and the corresponding intensity colourplot in Fig. 2f indicates that the fraction of the field in the gap also increases. However, relative to the plasmonic slot, the associated field enhancements are orders of magnitude weaker. The extremely low loss of such structures still makes them very attractive for nonlinear applications, but also demand millimetre-scale propagation lengths under typical experimental conditions [87]. Hybrid plasmonic waveguides The final relevant structure to consider is the so-called hybrid plasmonic waveguide (HPWG) [88][89][90], shown in the schematic of Fig. 4g: it is formed by a metal structure adjacent to a high-index dielectric, separated by a low-index spacer. This device exhibits properties that are akin to both plasmonic-and dielectric-slot waveguides, retaining some advantages of each when decreasing spacer thickness t. For example, while it possesses a low effective width (here achieving a minimum w eff = λ/30, see Fig. 4h), its group velocity does not change as significantly. However, it possesses lower linear losses than the plasmonic slot waveguide, by about one order of magnitude. A colourplot of the field intensity as a function of spacer thickness, shown in Fig. 4i, reveals that much of thie field is in the sub-wavelength low-index spacer. The combination of low losses and large confinement thus makes them candidates for enhancing the nonlinearity of optical waveguides [91]. A "jungle" of plasmonic waveguides So far we have assumed 1D waveguides and 2D propagation; in practice, any waveguide will have a 2D mode profile and propagate in 3D. As a simple example, cylindrical wires support modes which can be also described by an analytic transcendental equation [92], and the SR-and LR-SPP modes are the radially polarized (TM 0 ) and linearly polarized (HE 1 ) modes respectively, each possessing similar properties to those shown in Fig. 3a-f. More complicated profiles demand full calculations [92]. Owing to the large number of associated dimensional and rotational degrees of freedom, there is a vast "jungle" of reported 2D plasmonic waveguide designs, including so-called wedge- [93], channel- [94], gap- [83], and dielectric-loaded [95] plasmonic waveguides, to name a few. All such waveguides form a library of PIC-compatible structures provid- Representative 2D modes supported by plasmonic waveguides with sub-wavelength spatial features. These include a radially polarized (TM 0 ) SR-SPPs b linearly polarized (HE 1 ) LR-SPPs on cylindrical nanowires [92], c aperture modes [92], d gap plasmons [83] e dielectric-loaded plasmons [95] and f their long-range equivalent [96,97]. Also shown are the modes of g a hybrid- [89] h channel- [94] and i wedge- [93] plasmonic waveguides. D dielectric, L lower-index dielectric. A air, M metal. Calculations performed by the Author ing omnidirectional field enhancements via their sub-diffraction modes, which in turn strongly depend on the spatial distribution of the higher-and lower-index dielectrics, spacers, and metals involved. A summary figure of commonly reported plasmonic structures and associated modes is shown in Fig. 5. We refer the reader to Ref. [56] for an example review of the linear-modal properties of 2D plasmonic waveguides. Nonlinear optics in lossy media Having presented the fundamental linear properties of plasmonic waveguides, we now discuss their nonlinear properties, which began attracting increased attention starting in the 1980s [98,99]. We first review some relevant theoretical tools and results, and begin considering the simple textbook case [31,51] of a homogeneous, isotropic material, which responds to a scalar electromagnetic field E via a polarization where ε 0 is the permittivity of vacuum and χ (n) is the material's n-th order electric susceptibility. More generally, this expression can contain E oscillating at different frequencies ω i to produce a polarization P(ω), in which case χ (n) depends on the frequencies involved. Since electric and polarization fields are most generally vectors, χ (n) are generally tensors. Linear optical processes (e.g, refraction and absorption) are described by the χ (1) term in Eq. (2) alone, valid for small field amplitudes, and involving one frequency at a time. Optical processes at larger field amplitudes can only be described by including higher-order terms, which result in more complicated interactions involving multiple frequencies. χ (2) is responsible for several important effects such as second harmonic generation (SHG), optical rectification (OR), and sum/difference frequency generation (SFG/DFG); χ (3) can give rise to even more nonlinear processes, but the most commonly considered are the Kerr effect, third-harmonic generation (THG), four-wave-mixing (FWM), self-phase modulation (SPM). All these effects are described in great detail in several textbooks [31,80] and reviews [51,54]. Since χ (2) nonlinear processes are prohibited in centro-symmetric structures, in the context of plasmonic waveguides they most commonly occur at metal/dielectric interfaces where centro-symmetries are trivially broken [100], although many plasmonic waveguide designs also include non-centrosymmetric structures adjacent to the metal [19,101]. In contrast, all materials have non-zero third-order susceptibility, making χ (3) effects always relevant at high intensities. The most important third-order nonlinear process is arguably the Kerr effect, which is responsible for the nonlinear polarization at the incoming frequency. We now consider it in some detail, with particular attention to hybrid waveguide structures containing lossy materials. The nonlinear Kerr coefficient A plane wave with wavenumber k = nk 0 propagating in a bulk medium with complex refractive index n, induces a nonlinear refractive index change in the medium at high intensity I . The nonlinear refractive index n 2 quantifies the change in refractive index per unit intensity: where n 0 = lim I →0 n is the linear refractive index. For bulk lossy materials, n 2 is related to χ (3) via [31,102] and is most commonly measured using the z-scan technique [103]. With knowledge of materials' n 0 and n 2 , we now consider multi-material waveguides that support modes with propagation constant β = n eff k 0 and power P. In this case, the change in propagation constant is quantified by a nonlinear coefficient γ via where β 0 = lim P→0 β is the linear propagation constant. The parameter γ is required to simulate high-intensity light propagation in waveguides using a nonlinear equation (NLE) [80]. In the simple case of extremely lossy waveguides with short, wavelength-scale propagation distances, the NLE is given by where α 0 = 2 m(β 0 ) is the linear absorption coefficient of the waveguide, and A is a field amplitude. Note that γ is a complex number-its real part is associated with the nonlinear phase shift, and its imaginary part is associated with optical limiting or saturable absorption. Generalizations of Eq. (6) may contain additional nonlinear or dispersive effects [36], and can be extended to describe extended coupled pump, signal, and idler fields [82]. Equation (6) can also be generalized to include the transverse field dependence [104], which is necessary to describe plasmon-solitons whose spatiotemporal profile does not change with z even in a transversely infinite medium due to self-focusing effects [85]. Parameters n 2 and γ are analogous in that their real parts give the nonlinear phase shift and their imaginary parts give rise to optical limiting or saturable absorption, depending on sign. Calculating γ is generally difficult, especially in waveguides formed by multiple, high-index materials that induce optical losses. In the simple case of low-loss single mode optical fibers with low index contrasts, which possess similar n 2 in the core and cladding and support scalar modes, γ = k 0 n 2 /A eff where A eff is an effective mode area [36]. Until recently, it remained unclear which of the many expressions for γ [105][106][107][108][109] were valid for hybrid waveguides formed by extremely lossy materials. Following a systematic analysis and comparison with full numerical calculations, the most general expression for γ was ultimately established to be [110] where e, h are electric-and magnetic-modal fields respectively, ω is the angular frequency,ẑ points in the propagation direction, and the x y plane is transverse. Equation (7) was independently obtained by Im et al. [111] and Li et al. [109], and although it appears complicated, it can be immediately calculated using any linear mode solver, requiring only knowledge of the linear-and nonlinear-properties of an arbitrary waveguide's constituent materials. Schematic comparing perturbative and non-perturbative approaches. a In the linear regime, β is constant and γ = 0. In the nonlinear regime, P increases the propagation constant by γ P. b In perturbative treatments, the nonlinear permittivity changes but the mode profile is assumed not to and γ is constant. c In non-perturbative treatments γ depends on P. Orange line: intensity. Adapted from Ref. [110] Equation (7) reduces to Eq. (39) in Ref. [108], valid for arbitrary lossless waveguides, and can factorized in terms of more physically intuitive properties [77,79]. This factorization is not unique: one choice, shown to be valid for lossless waveguides, is given by [91] where χ (3) is the average of nonlinear susceptibility over the constitutive materials, weighted by the magnitude of the electric field. The definition of effective area A eff is also not unique [79]: one frequently used choice is given by the area of longitudinal power flow [108] The factorization of Eq. (8) provides valuable physical intuition: v g enhances the transverse electric field due to slow-light effects, and A eff gives rise to longitudinal field enhancement. Both can drive nonlinear changes in the refractive index of the waveguide's constituent materials, modifying the propagation constant. A similar factorization was recently shown to provide useful insights even for extremely lossy plasmonic waveguides [77]. With the factorization of Eq. (8), we can go back and estimate that the SR-SPP and MDM structures of Figs. 3b and 4b would possess the largest γ amongst the structures considered in Sect. 2, although material properties also play an important role via χ (3) . Note that both Eqs. (4) and (7) consider nonlinear changes in the refractive index to be small perturbations, so that the propagation constant of the mode changes but the fields do not, as illustrated in the schematics of Fig. 7a, b. For large relative nonlinear index changes, non-perturbative approaches are necessary [112], which account for changes in both the optical medium and the modal profile [113] as illustrated in the schematic of Fig. 7c. This results in a power-dependent γ [110,[113][114][115]. This compli- Im Fig. 8 Complex nonlinear susceptibility χ (3) of gold a as a function of wavelength [105] and b plotted in the complex plane [110]. c Associated gold n 2 [Eq. (4)] plotted in the complex plane. d Associated SPP γ [Eq. (7)] plotted in the complex plane. Note the scaling and rotation of gold n 2 and SPP γ with respect to gold's χ (3) . Colourbar represents wavelength. a is adapted from Ref. [105] under a creative commons license (CC BY-NC-SA 3.0) cated nonlinear problem can be addressed by numerically iterating a series of simple linear problems [114]: the calculated linear mode at a given power changes the local refractive index, resulting in a graded index profile supporting a new mode, which is then calculated. This process can be iterated until the propagation constant converges, although this is not guaranteed. The change in propagation constant Δβ(P) is linear only at low powers as shown in Fig. 7c, and the nonlinear coefficient of Eq. (7) is given by γ = lim P→0 dβ/d P. At sufficiently high powers, local changes in the materials' refractive index can be strong enough to induce modal bifurcations, for example in nonlinear plasmonic slot waveguides [116,117]. Non-perturbative approaches have recently emerged to interpret experiments in so-called epsilon-near-zero materials [112], which exhibit extremely large nonlinear refractive index changes [118] and are increasingly relevant for ultra-compact nonlinear devices applications [77,119]-see also Sect. 4. Unless otherwise stated, all present discussions relate to non-perturbative conditions. Relating (3) and : a complex matter The relationship between complex n 2 and complex χ (3) for bulk media has some interesting and counter-intuitive consequences [120]. To illustrate this, we consider the χ (3) dispersion for gold, theoretically considered in Ref. [105] and shown in Fig. 8a. The wavelength dependence of χ (3) can also be represented in the complex plane as shown in Fig. 8b. Equation (4) then indicates a rotation of n 2 with respect to χ (3) in the complex plane, as shown in Fig 8c. The analytical relation between γ and χ (3) for arbitrary plasmonic waveguides is not so simple, although for the case of a SPP γ can be calculated analytically [110], and is shown graphically in Fig. 8d. In waveguides with no linear loss, i.e. when the linear permittivity is purely real, the real part of γ is proportional to e χ (3) and the nonlinear absorption is proportional to m χ (3) . However, this proportionality fails for lossy waveguides, i.e. when the linear permittivity is complex. Indeed, note that γ can be purely real, corresponding only to a nonlinear phase shift and no nonlinear absorption, even when both real and imaginary parts of χ (3) are negative. Similarly, γ can be purely imaginary, i.e. only nonlinear absorption and no nonlinear phase shift, even when both real and imaginary parts of χ (3) are positive. Figure 8 also demonstrates that there is no straightforward correlation between the complex phase of γ and that of χ (3) , and that the full complex nature of both the linear and nonlinear quantities plays an important role both in bulk metals [102,120,121] and in plasmonic waveguides [110,111]. Figures of merit of Kerr nonlinear performance With knowledge in hand of both linear losses and nonlinear coefficients, we now consider the nonlinear performance of Kerr plasmonic waveguides, and discuss how figures of merit can guide their designs. Since attenuation lengths in plasmonic waveguides are quite short-typically a few wavelengths, see for example Fig. 3a-phase matching (PM) is not as crucial as for low-loss systems. This can be understood by examining Fig. 9, which schematically illustrates how much nonlinear power P NL is generated by a driving pump under different conditions. Phase matching (blue curve) leads to the phase fronts of the pump-and nonlinear-fields to advance synchronously, and the nonlinear fields to add up coherently upon propagation, conserving momentum. PM is crucial in for the efficient build-up of nonlinear power over optically long distances, because the nonlinear response of dielectric materials is weak, and high conversion efficiencies require careful design [31]. In the absence of phase matching (red curve), the resulting nonlinear fields can have different relative phases during propagation, which limits the amount of nonlinear power produced. In extremely lossy plasmonic systems, absorption has the effect of both reducing pump power at long lengths (preventing nonlinear light generation), and attenuating the intensity of the generated nonlinear signal (removing the generated signal). In this scenario, the phase matching requirement is moot, since at long lengths loss is the dominant mechanism limiting nonlinear effects. This is more quantitatively highlighted by full calculations of conversion efficiencies for the specific case of near-degenerate four wave mixing in the lossless-and lossy-case, shown in Fig. 9b, c respectively [82]. An important quantity to consider in Kerr nonlinear waveguides is the nonlinear phase shift Δφ NL (t), induced by changes in the propagation constant at high powers, as described by Eq. (5). In the case of a temporally varying ultrashort optical pulse of power P(t) centered around a frequency ω 0 propagating inside a lossy medium, the nonlinear phase shift is given by [122] φ where γ = γ R + iγ I can be calculated from Eq. (7), is the effective length, and L att = 1/α 0 . In the absence of loss, this reverts to the familiar form [80] φ NL = γ P L. Equation (10) leads, for example, to the nonlinear generation of new frequencies via self-phase modulation through ω(t) = ω 0 + dφ NL (t)/dt [80]. The effectiveness of a nonlinear waveguide is commonly quantified by a figure of merit (FOM), chosen to compare the performance of different systems. A commonly used FOM is F = γ L att [123], which roughly computes the inverse power required to obtain one radian of phase shift over one attenuation length. Note that Eq. (11) deceptively suggests that the nonlinear effects increase indefinitely with power; a more complete analysis should account for material damage effects at high powers. To illustrate this, Fig. 10a shows a schematic summary of the achievable nonlinear phase shift in a bulk material and a waveguide containing it. The blue curve shows an initial linear increase in the nonlinear phase shift with driving power following Eq. (11), reaching a maximum before material damage, associated with a maximum nonlinear index change Δn max . The red curve shows the equivalent effect in a nonlinear waveguide: the slope, given by γ , can be much larger than its bulk counterpart due to the omnidirectional field enhancements discussed. However, this is accompanied by a lower damage threshold. This effect is general, but particularly severe in plasmonic waveguides due to the potential presence of localized "hot spots" at the metal surface (see for example where P 0,th is the maximum power supported by the mode before damage occurs, and can be estimated from modal calculations of the electric fields around plasmonic hot-spots, combined with experimental measurements of material damage thresholds [124]. Once the above FOM is known, Li et al. showed that the maximum achievable nonlinear phase shift is given by ΔΦ max NL = 2F 2 /3, at an optimum device length L OPT = ln3· L att ≈ 1.1L att . For the specific case of nearly-degenerate four-wave mixing [127], this corresponds to a signal-to-idler conversion efficiency of η = 4F 2 /27. An illustrative full calculation comparing lossless and lossy waveguides, originally presented in Ref. [82], is shown in Fig. 9b, c. Subsequent work [125] proposed the concept "nonlinear effectiveness", which quantifies a mode's capacity to use a certain material's maximum nonlinearity: it was shown that this requires a strong electric energy confinement, and broadband slow light effects. A comprehensive comparison of several material and geometry combinations suggested that MDM structures perform best for compact efficient nonlinear optics [128]. We are now in a position to discuss typical recent experimental configurations for on-chip nonlinear plasmonics, illustrated in the Fig. 10b schematic: light from a linear dielectric waveguide is coupled into a subwavelength plasmonic region containing a highly nonlinear material. Here, the intense fields provide nonlinear optical effects over ∼ L att , and the resulting nonlinear light is out-coupled into the dielectric waveguide. In such a way, low-power and low-footprint nonlinear effects are concentrated to a dedicated region, and losses are minimized. It is thus worthwhile reflecting on the requirements for achieving the large F in Eq. (12) in the context of plasmonic systems. Since L att is typically of the order of a few wavelengths, one can compensate the small propagation loss with a large γ or using a higher power. However, the omnidirectional field enhancement producing a large γ for a certain χ (3) lowers the damage threshold P 0,th . Moreso than for all-dielectric devices which can accumulate nonlinear effects using longer lengths, plasmonic nonlinear devices crucially require both a large χ (3) and a high damage threshold. If used in hybrid structures, they should also posses a lower refractive index than the adjacent semiconductor, and ideally be compatible with industrially scalable fabrication. Recent experiments have shown compact nonlinear functions using commercially available polymers such as JRD1 [19] and MEH-PPV [127], (possessing a large χ (2) and χ (3) , respectively), spin coated on a number of hybrid MDM waveguides on a silicon-on-insulator (SOI) platform (see also Sects. 4 and 5). We note that an early theoretical analysis [129] came to the conclusion that nonlinear plasmonics was not well suited for applications requiring high conversion efficiency (e.g., all-optical switching and frequency conversion), since the maximum achievable nonlinear phase shift was calculated to be at most 0.1 rad, with nonlinear conversion efficiencies of order −30 dB, assuming that the maximum achievable index change was 1%. Applications which do not require high conversion efficiencies, such as nonlinear sensing and imaging which benefit from smaller mode volumes, were seen as more suitable. Recent developments in device designs have shown MDM plasmonic structures with −13 dB FWM conversion efficiency [130] over wavelength-scale propagation, and epsilon-near zero materials with nonlinear refractive index changes of 170% [118]. Material considerations Due to the hybrid nature of nonlinear plasmonic waveguides, it is also important to consider how each constituent material contributes to the total γ . We may re-write Eq. (7) as γ = m γ m , where γ m is the contribution of a material m with nonlinear susceptibility χ (3) m to the total γ of a mode. The ratio γ m /χ (3) m thus quantifies the degree of concentration of light to a particular medium for that mode. Figure 11a shows γ m /χ (3) m for each material of the HPWG geometry considered in Fig. 4g as a function of the gap thickness t. Note that for large values of t, the larger ratio is in the underlying dielectric waveguide; for smaller t, the ratio is largest in the subwavelength spacer. Overall, the degree of concentration of light in the metal is always orders of magnitude less: this motivated early theoretical investigations to neglect the metal's contribution to the total nonlinear response in similar systems [116]. Calculating γ m , i.e., each material's contribution to the total γ , shown in Fig. 11b, paints a different picture: since air has a χ (3) that is seven orders of magnitude smaller than that of silicon [31], its contribution to γ is negligible. On the other hand, gold's χ (3) is orders of magnitude larger, so that its contribution approaches that of silicon for smaller separations as the field overlap with gold increases. Overall however, silicon is the dominant contributor to the total γ for this particular HPWG configuration. Table 1 for material parameters used Including a material with a large χ (3) inside the spacer (e.g., DDMEBT [72]) can dramatically increase the total γ , as shown in Fig. 11c, d: for subwavelength t, the large field fraction in the spacer, in unison with its large χ (3) , dominates the contribution to the total γ , enhancing the performance of the underlying waveguide by at least an order of magnitude. Table 1 shows the linear and nonlinear parameters used. For equivalent calculations in 2D waveguides, see for example Ref. [91]. Beyond this illustrative example, the relative contributions to the total nonlinear response will depend on the materials' permittivities, susceptibilities, and geometric parameters. Such relationships were rigorously addressed by Baron et al. [131] for the simple case of a semi-infinite metal/dielectric SPP, where modes have an analytical form. To identify whether the metal-or the dielectric-contributions dominate, a figure of merit ρ was proposed and shown in Fig. 11e. Here, ρ depends on both the ratio of metal/dielectric permittivites, nonlinear susceptibilites, as well as intrinsic modal characteristics: ρ < 0 indicates that dielectric dominates the nonlinear response, whereas for ρ > 0 the gold dominates. Overall, low-index and low-susceptibility configurations (e.g., air, silica, and aluminum oxide) are metal-dominated; otherwise, the large fields at the metal surface enhance the dielectric's nonlinear response. Si 3.5 1 + 0.3i (Fig. 11a) 1.55 [133] 25 + 0.0039i (Fig. 11b Harmonics generation In the case of second-and third-harmonic generation, and non-degenerate four wave mixing, developing general analytic guidelines for optimal device length and maximum conversion efficiencies is more challenging. In such cases, designs are highly dependent on the mode overlap profiles and losses of the participating the pump-and harmonic-modes, which can be vastly different, and thus require analyses on a caseby-case basis. To quote a few examples, a theoretical study [141] of second-harmonic generation in a χ (2) -polymer plasmonic-nanoslot structure at 1550 nm predicted maximum conversion efficiency η ∼ 10 −4 after propagating a length corresponding to the attenuation length (∼ 20 µm). A HPWG using a χ (2) material as the waveguide [142] or spacer [143] can yield a higher conversion efficiency (up to ∼ 8%), at the cost of a longer propagation length (> 100 µm). Similar conclusions can be drawn from THG via χ (3) effects [144]. Optical limiting and saturable absorption In bulk media, the transmitted power is associated with m(n 2 ) as per Eq. (4); in waveguides, it is due to m(γ ) as per Eq. (7) via Eq. (6). In lossless systems, m[χ (3) ], m(n 2 ) and m(γ ) all have the same sign. In plasmonic systems, which possess complex propagation constants, these quantities can have either a positive or nega- tive value, leading to a reduction-or increase-in the transmission at high intensities (i.e., optical limiting and saturable absorption (SA), respectively). Although nonlinear absorption is commonly seen as a limiting factor to nonlinear optical devices [122], it can be harnessed in nonlinear plasmonic devices in the context of "active plasmonics" [145], whereby changes in the absorption properties close to the metal/dielectric interface, driven by an external signal, can modulate the plasmonic mode, most recently shown to provide a means of providing low-power all-optical switching by integrating graphene on a MDM slot [146]. Nonlinear absorption effects in metals are strongly dependent on the pulse duration of the incoming light, even at constant wavelength. This pulse-length dependent absorption has been measured in detail for gold [147], and is due to the complex electron dynamics induced by an incoming optical pulse, although this effect is weaker away from the interband region in the near-infrared [148]-see Ref. [102] and Ref. [149] for related experimental and theoretical reviews. The Pockels effect We have so far considered the Kerr nonlinearity-whereby changes in the refractive index are proportional to quadratic fields (i.e., the intensity)-as a representative degenerate case when considering nonlinear plasmonics in chip-compatible structures. The above discussion, and much of the underlying physics, can be extended to linear electro-optics (EO) effects, i.e., the Pockels effect, whereby changes in the refractive index are proportional to linear fields via χ (2) . Most notably, a metal nanoslot containing a χ (2) medium leads to a strong nano-scale Pockels effects via large modal overlap between the short-wavelength optical fields E OF and long-wavelength fields E RF , as shown in Fig. 12 [150]. As a result, such fields efficiently interact via the underlying nonlinear medium: the propagation constant of the optical field changes via Δn eff ∝ χ (2) E RF E OF 2 dxdy [151]-while this mode overlap is small in dielectric waveguides, it can be large in plasmonic structures, leading to more compact electrooptic devices operating at low powers [152,153]. Here the effective index change of the optical mode is given by [151] Δβ = k 0 Δn eff = 1 2 where r 33 is the electro-optic coefficient [31], and where Γ and n s are a optical mode-dependent field-power interaction factor and a slow-down factor, respectively, defined in Ref. [151]. Equation (13) assumes that the dominant nonlinear effects occur in the slot, in a non-perturbative regime, and neglects losses, but it demonstrates how nonlinear plasmonics effects are enhanced via the same physical mechanisms underpinning the heuristic formula of Eq. (8). The most commonly used electro-optic material is LiNbO 3 [101], and organic electro-optic (OEO) materials have recently been developed and included in dielectric-plasmonic devices [19], for field sensing at GHz and THz frequencies [150] and electro-optic data modulation with extremely low footprints (∼ 2.4 Tb/s/mm 2 [154]). Nonlinear experiments with plasmonic waveguides With the widespread use of commercially available numerical solvers (e.g., finite element, finite-difference time domain, and beam propagation techniques, to name a few), plasmonic waveguide structures have been the focus of a large number of numerical studies. Many nonlinear plasmonics experiments consider planar substrates containing metal nanostructured arrays [53], whose linearly and nonlinearly coupled modes are typically excited through external, diffraction limited illumination. The waveguideequivalent version of such structures often rely on placing such nano-antennas on top of [59,155] or at the endface of [156] a waveguide. More efficient nano-coupling requires careful design [69,157], and such structures often require multiple fabrication steps that demand nanometre-precision alignment [55]. Early nonlinear plasmonic waveguides tended to relatively weak nonlinear responses and, being a few wavelengths long, characterizing them was challenging, often requiring sensitive measurements [158]. We now provide an introductory overview of nonlinear experiments in plasmonic waveguides. We first consider wave-guiding structures formed by a single metal/dielectric interfaces to achieve their nonlinear function, before moving to hybrid systems. The nonlinear effects considered are due to guided surface plasmons that are compatible with photonic circuitry, although most experiments rely on free-space excitation. Surface plasmon polaritons The pioneering experimental work on nonlinear plasmonics can be traced back to the 1970s with the first observation of second harmonic generation by exciting SPPs on a bulk silver film [159], measuring more than an order-of-magnitude enhancement in SHG emerging from propagating plasmon excitation, when compared to front-surface reflection. Later fundamental studies used a wavevector-space spectroscopy technique to observe this process in more detail [160], directly measuring the annihilation of two surface plasmons and creation of second-harmonic photons. The first device-driven nonlinear plasmonics experiments targeted nonlinear switching: in first instance, this can be achieved by inducing nonlinear changes in the dielectric permittivity ε m at the metal's surface, which alters the propagation constant in Eq. Early experiments with metal/semiconductor waveguides used aluminium grating structures adjacent to silicon, and showed high-contrast switching operation [161], but operated near silicon's absorption edge at λ = 1.064 µm, where the response is dominated by free-carrier generation and lattice heating, which is in the nanosecond to millisecond range. Ultrafast nonlinear modulation enabled by plasmonics started emerging from the mid-2000s. In one notable experiment [145], summarized in Fig. 13, the transmission of ultrafast surface plasmon polariton pulses propagating on an aluminium/silica interface could be modulated by an external probe, with response times of ∼ 200 fs. This was enabled by operating at the absorption peak of aluminium (λ = 780 nm), where changes in the real-and imaginary-parts of its permittivity were due to ultrafast interband transitions. In particular, these were due to nonlinear changes at the metal surface, and occurred only for a polarization parallel to the propagation direction; a slower, thermally-driven polarization-independent response was also identified. Rich nonlinear electron dynamics at metal surfaces can also lead to the external excitation of surface plasmon polaritons directly on a gold film-typically disallowed due to lack of phase-matching between plasmonic-and free-space beams-via the formation of an effective "nonlinear grating" [162]. Nonlinear plasmonic modulation can alternatively be addressed via nonlinear changes in the permittivity of the adjacent dielectric: typically, gold/silicon bulk SPPs [163] and gold/polymer waveguide SPPs [164] enable modulation speeds of 0.1-1 ms. Related studies explored plasmonic coupling due to the nonlinear interactions between modes of different harmonics in plasmonic films. Palomba et al. experimentally demonstrated the nonlinear excitation of surface plasmons at λ = 613 nm via four-wave mixing of ultrashort infrared in a Kretschmann configuration [165]. These fundamental results, which highlight the potential for nonlinear manipulation of surface plasmons, highlighted how important surface effects are: despite the fact that gold possesses a bulk χ (3) , the surface χ (3) at the gold/dielectric interface was the dominant nonlinear source. Subsequent experiments on the same structure measured three distinct four-wave mixing effects, including nonlinear reflection off the gold surface, the excitation of evanescent fields, as well as the excitation of the nonlinear surface plasmon [166]. The nonlinear conversion was later improved by nanostructuring the gold surfaces, where local field enhancements improved the conversion efficiency with respect to a smooth film by a factor of ∼ 25 [167]-2000 [168] times. These pioneering studies showed novel chip-compatible excitation mechanisms as a result of the large nonlinearities at gold surfaces, driven by large local intensities. A series of subsequent experiments investigated the intrinsic χ (3) of gold by probing the nonlinear "self-action" effects of SPPs, whereby a SPP modifies its own propagation characteristics. De Leon et al. [136] investigated intensity-dependent SPP propagation on a gold film, and used it to obtain the complex χ (3) experimentally (this is challenging, and most experiments estimate its magnitude [102]). The authors measured a power-dependent reflection spectra of the Kretschmann configuration as shown in Fig. 14a; a nonlinear transfer matrix model was then used to obtain χ (3) = 4.67 × i3.03 as a single fitting parameter at λ = 800 nm. A review by the same authors [102] found that measured values of χ (3) of gold can vary by several orders of magnitude, depending on wavelength, pulse duration, or the nature of the nonlinear experiment. Most strikingly, similar measurements of the nonlinear absorption of SPPs at gold/air interfaces [169] resulted in χ (3) values which were three orders of magnitude larger. In this case, the authors excited SPPs on a gold film using asymmetric gratings, which also collected the light, as shown in Fig. 14b(left). χ (3) was then deduced from systematic optical limiting measurements, also in Fig. 14b(right). The authors attributed the apparent χ (3) discrepancy to potential differences in the structure's surface roughness. These examples also serve to illustrate the difficulties in obtaining reliable and consistent nonlinear parameters for metals. Long-range surface plasmon polaritons In the 1980's, the first SHG experiments on nonlinear LR-SPPs were reported, which sought to observe some of the emerging theoretical predictions [171], and first investigated the trade-off between confinement and propagating distance. For example, in 1983 Quail et al. [172] showed that the field excitation on both surfaces of the film leads to a two-order of magnitude improvement in harmonic generation compared to an equivalent bulk film. Experiments targeting the χ (3) response of gold via nonlinear absorption of LR-SPPs on both thin metal films [106,173] and metal nanowires [170] were recently performed. In this case, nonlinear effects were measured after mm-and cm-scale propagation distances. Lysenko et al. measured nonlinear absorption of plasmonic modes in waveguides formed by gold nanofilms of different thickness (22-35 nm) surrounded by bulk SiO 2 and Ta 2 O 5 nanolayers. The authors measured a thicknessdependent nonlinear absorption induced by 200 fs pulses at 1064 nm (Fig. 15a), and developed a nonlinear wave equation that generalizes Eq. (6) to include gold's temporal response, which accounted for non-instantaneous contributions from free electrons. Their model indicated that χ (3) nearly doubles as the film thickness is halved. The authors suggested that these changes in χ (3) are due to increased collisions of electrons in thin gold layers. Such quantum size effects are significant for thinner metal layers: for example, Qian et al. [174] showed more than a radian nonlinear phase shift for a bulk 3 nm gold film under similar conditions. Tuniz et al. observed nonlinear absorption of long-range plasmons on gold nanowires (diameter: 100 nm) integrated within the core of a step-index silica fiber [170], after centimetre-scale propagation. The integration of sub-micron metal wires in fibers [175] typically leads to wire break-up; the authors overcame this limita-220 A. Tuniz Fig. 16 Second-and third-harmonic generation on single Au nanowires. a Schematic representation of the experiment: a hole array is illuminated TM-polarized femtosecond pulses, (λ = 1550 nm). SPPs are adiabatically converted to SR-SPPs by reducing the width of the wire. Visible light, corresponding to harmonics-generation, is scattered from the wire. b Typical measured spectrum showing peaks at 775 nm (SHG) and 517 nm (THG). Inset: solid line power fits to the measured yield as function of power. The slopes of 2 (red), and 3 (green) respectively confirm the required power dependence for SHG and THG. c Wavelength dependent SHG and THG. Adapted with permission from Ref. [176]. Copyright (2016) American Chemical Society tion by including the gold nanowire in the core of a single mode optical fiber, which allowed to access a unique regime where the plasmonic mode was the only effectively propagating mode, which directly interfaced with a single mode fiber. In the regions where the wire breaks up, light was recaptured by the fiber, and then re-coupled into the plasmonic mode at the subsequent wire junction. This approach solved the problem of detrimental wire discontinuities and fabrication imperfections along the gold nanowires, by preventing the light from scattering away, and enabled measurements of ultrafast nonlinear absorption (30fs pulse duration, 1560 nm wavelength). Selfphase modulation effects, on the other hand, were dominated by the silica matrix. The nonlinear absorption coefficient obtained was in agreement with expectations from experimental trends [102]. Short-range surface plasmon polaritons Experimental observations of nonlinear effects on propagating short-range surface plasmon polaritons are uncommon, due to the short attenuation lengths (typically, of the order of a few µm), and due to challenges in efficiently coupling to such nanoscale modes [69]. In 2016 De Hoogh et al. [176] were the first to report nanoscale nonlinear optics with propagating plasmonic modes on a photonic chip. They showed both second-and third-harmonic generation due to surface-and bulk-nonlinearities on single gold nanowires. The short-range SPP modes were excited using a previously reported adiabatic taper approach [177] shown in the schematic of Fig. 16a. The authors conclude that the measured THG and SHG, shown in Fig. 16b, c, emerge both from the local enhancement induced by plasmonic nano-focusing before being launched into the nanowire, and from the modes propagating on the nanowires themselves. Note that although the pump-and harmonic-modes are not phase matched, this did not preclude higher harmonic generation in this lossy system. More recently, Chen et al. [178] showed that a coupled plasmonic two-wire system-formed by two 6 µmlong 100 nm gold nanowires, separated by 100 nm-can selectively generate both symmetric-and anti-symmetric second-harmonic modes by judicious mixtures of the 1560 nm pump modes, tailored via the input coupling conditions. This approach might find use in providing additional degrees of freedom for nonlinear circuit designs, such polarization control, waveform shaping, and selective routing. Nanofocused surface plasmon polaritons The appeal of plasmonics-based approaches is the ability to guide and then concentrate light to deep subwavelength volumes, which can be achieved by tapering a waveguide to the nanoscale, as shown in Fig. 3. In this case, the short-range plasmons concentrate light in all directions, potentially within mode areas of less than (λ/100) 2 , and the region in this case, the region in close proximity to the sharp tip of the tapered plasmonic strucures gives rise to the field enhancements that further favour nonlinear processes compared to the bulk (non-tapered) case. This approach has been shown to enhance nonlinear processes inside the metal and in the surrounding dielectric region, with applications in nonlinear imaging and nonlinear light generation. In one experiment, Verhagen et al. [177] experimentally showed the enhancement of nonlinear multi-photon processes associated with energy levels of Erbium, which surrounded a tapered silver plasmonic waveguide pumped at 1.49 µm. The measured far-field intensity enhancement due to this nonlinear process provided evidence of local near-field enhancements, which would otherwise be difficult to observe without using near-field techniques. Other experiments have utilized the intensity enhancements inside a hollow metal cantilever taper-as shown in Fig. 17a-to produce high-frequency harmonics. Despite the fact that small apertures formed by perfect conductors cut off and do not support propagating modes, Park et al. [179] harnessed a peak increase in the field intensity near a taper's aperture, shown in Fig. 17b, as a result of a subtle interaction between the incoming field, and the forward-and backward-propagating surface plasmons. The local field was enhanced by a factor of up to 350, which the authors use to produce up to 43 harmonics of Xenon gas, into the extreme ultraviolet (UV), pumping with near-infrared (NIR) radiation. The experimental results showcasing these results are plotted in Fig. 17c. Other approaches use the metal itself as the nonlinear medium, driving the nonlinear processes upon tapering of the metal waveguide. Having previously demonstrated the ability to guide arbitrary femtosecond short-range plasmons pulse to a plasmonic nanofocus (directly revealed by SHG-assisted interferometric cross-correlation measurements [181]) Raschke and collaborators [180] used four-wave mixing effects for nonlinear imaging (apex radius: 15 nm). The measured conversion efficiency was 10 −5 , which was enough to observe the plasmonic hot-spot dynamics of a separate gold surface with 50 nm resolution. A number of different experiments on the same geometry revealed several intriguing nanoscale nonlinear effects, including electron emission from the tip [182], and a nanostructure-induced enhancement of χ (3) of gold for sharper metal tips via longitudinal field gradients [183]. These results highlight the many opportunities provided by guided-wave nonlinear plasmonics due to localized strong field effects, even in the face of low nonlinear conversion efficiencies. We refer the reader to Ref. [184] for a recent and comprehensive review of strong-field nonlinear nano-optics. Hybrid plasmonic waveguides While the nonlinear plasmonic experiments presented so far relate to guided-wave structures, they are one step away from being compatible with photonic integrated circuits, where they would interface with dielectric waveguides [43,57,58,186,187]. Sederberg et al. [185] bridged silicon photonics [3] with nonlinear plasmonics, reporting optical third harmonic generation enhanced by plasmonics on a silicon nanowire, as summarized in Fig. 18a. In this experiment, a gold film was deposited on top of a silicon waveguide, shown in the scanning electron microscope (SEM) image of Fig. 18 a Schematic silicon/gold nanoplasmonic waveguide. Inset: calculated intensity of its fundamental (left) and third-harmonic (right) modes. b SEM of the fabricated silicon plasmonic waveguide. c Optical microscope image of visible light emission from the nanoplasmonic waveguides due to third harmonic generation. d Comparison of the third-harmonic spectrum obtained for a bare-and plasmonic-Si waveguide for the same incident power, showing plasmonic enhancement of ∼ 30%. Adapted with permission from Ref. [185], Copyright (2015) by the American Physical Society. e FWM power P NL transmitted by a HPWG for varying input power P L and spacer gap thickness as labelled, compared with a bare SOI waveguide. Insets show SOI and HPWG mode calculations, and a SEM image of one of the waveguides. Adapted with permission from Ref. [158], Copyright The Optical Society. e Schematic of the associated measurement approach [148]: a pulse is cut using a pulse shaper (PS) and coupled into a WG. The resulting spectral broadening is spectrally filtered using a long pass filter (LPF), and P NL is detected using a spectrometer Fig. 18b. Light was launched and collected via end-fire coupling, with NIR pulses (λ = 1.55 µm) driving third-harmonic generation (λ = 517 nm) in a waveguide of length 5 µm, as shown in Fig. 18c. Note the significant experimental challenges associated with this measurement: the short attenuation length of silicon at visible frequencies (L att ∼ 600 nm) makes phase-matching unnecessary (see Fig. 9a). Compared with a bare silicon waveguide, the THG signal from the plasmonic-enhanced waveguide was approximately 27% stronger (as shown in Fig. 18d) in a device that was three times shorter, resulting in a maximum conversion efficiency of 2.3 × 10 −5 . More recently, the high confinement and low losses of HPWGs were exploited for compact SHG and sum frequency generation (SFG). One experiment [188] measured a SHG conversion efficiency in a HPWG waveguide formed by CdSe (length: 5 µm; width: 360 nm) deposited on a gold film, separated by a 10 nm Al 2 O 3 spacer, and pumped at 800 nm [188]. In this case, the dominant nonlinear effect originated from the CdSe, and was enhanced by the excited HPWG modes. The authors selectively coupled to the photonic-and plasmonic-modes of this multi-mode system: the latter showed a 20-fold SHG enhancement with respect to the former, with a maximum conversion efficiency of 4×10 −5 W −1 , and with several prospects for further improvement (e.g., higher quality gold/silver films, better nonlinear mode overlaps, and by optimizing nanowire cavity effects.) In this particular experiment, phase matching also did not play a role due to the large loss of the second harmonic mode. A subsequent experiment on an AlGaInP-based HPWG structures [189] directly measured the evolution of second-harmonic-and sum-frequency-generation (SFG) in phase-matched ∼ 15 µm length waveguides and ∼ 1 µm HPWG microresonator disks, with peak SHG conversion efficiencies up to 2.6 × 10 −6 . A comparison with all-dielectric waveguides showed more than a 1000-times enhancement, and the efficiencies per unit length were claimed to be competitive with state-of-the-art lithium niobate devices. Most notably, a broadband SFG processes-wherein multiple combinations of phase-matched nonlinear frequencies could be addressed via a supercontinuum source-were three to five times more efficient than SHG as a result of the lower losses of the modes involved. Measuring Kerr nonlinearities in comparable micrometre-length waveguides is more challenging, since the phase shifts can be as low as ∼ 10 −4 rad [91], resulting in negligible spectral broadening due to self-phase modulation. Nevertheless, measuring such effects can be important for benchmarking the performance of plasmonicallyenhanced HPWGs. To address this requirement, Diaz et al. [158] presented a method to sensitively measure self-phase modulation in micro-scale waveguides. The experimental procedure relies on shaping each pulse via an all-reflective waveshaper, such that long wavelength are completely removed, leading to a sharp spectral edge. Such spectrally cut pulses are then coupled to the waveguide, where the small nonlinear signals generated in the cut region can be detected after removing the pump light with a spectral filter. This background free measurement enables sensitive measurements of Kerr nonlinear effects. A comparison between a silicon waveguide and a hybrid plasmonic waveguide with a silicon nitride spacer, shown in Fig. 18e, reveal no significant improvement, since the χ (3) of silicon nitride is too low to boost γ above that of the bare silicon waveguide, despite the sub-wavelength mode area. A later theoretical analysis [91] revealed that DDMEBT in HPWG can enhance the SOI γ by an order of magnitude (see also Fig. 11d). Finally, we highlight a recent experiment which revealed self-focusing effects in a hybrid gold/silica/chalcogenide structure at telecommunication wavelengths over distances of ∼ 100 µm, harnessing the field enhancements and the large nonlinearities in chalcogenide [190]. In spite of the early promise of hybrid plasmonic waveguides for nonlinear applications [191], and their potential to enhance the performance of the underlying dielectric waveguide [91], HPWGs have enjoyed limited use in PICs, perhaps because the associated fabrication/design difficulties to be overcome are too large, and the expected performance improvement too little. A number of recent experiments provide compelling evidence that metal-dielectric-metal waveguides [83] are easier to fabricate, can be immediately interfaced with dielectric waveguides, and provide giant nonlinear effects after wavelength-scale guidance. We now discuss nonlinear MDM waveguides, starting with their Kerr nonlinear performance. Additional circuit-integrated MDM nonlinear effects are discussed in Sect. 5. Kerr plasmonic slot waveguides Early nonlinear experiments with MDM waveguides showed evidence of all-optical switching in plasmonic directional couplers [192] formed by adjacent 80 nm wide plasmonic slots that were a only a few micrometres long [193], operating at 1550 nm. Despite the low footprint, these switches were reliant on the metal nonlinearity, were prone to optical damage, and required 5 kW of peak power. Fig. 19 a SEM image of a gap plasmon waveguide on a SOI substrate, of length L = 2 µm and gap width w = 25 nm. Arrows indicate a schematic of the FWM experiment, whereby a pump at frequency ω p (green) and a signal at ω s (blue) generate an idler at ω i (red). b Mode intensity calculations at the input taper region and c in the plasmonic slot region (right), used for coupling via an adiabatic transition. d Example pump/signal spectrum and generated idler spectrum, showing -13 dB conversion signal-to-idler conversion. e Conversion efficiency versus waveguide length for different lengths of the waveguide (black markers), compared with theory (red line)-see also Fig. 9c. From Ref. [127]. Reprinted with permission from AAAS More recent approaches have relied on incorporating high-index dielectrics inside the plasmonic slots. The highly nonlinear plasmonic modes are accessed from dielectric waveguides via efficient modal conversion schemes, e.g., by placing the plasmonic slot either on top of [127] or adjacent to [194] the waveguide, most commonly with a tapered section to assist the mode transformation [195,196]. Compared to the HPWG shown in Fig. 5g, the plasmonic slot geometry enables evaporation or spin-coating of a highly nonlinear material as a very last fabrication step. Nielsen et al. [127] used this approach to report giant four-wave mixing (FWM) conversion efficiencies in a plasmonic slot waveguide of 2 µm in length. The waveguide is shown in Fig. 19a, and consists of the commercially available, highly nonlinear polymer MEH-PPV, which is sandwiched in a gold nano-slot (gap width: 25 nm). Light was coupled into the waveguide and collected via gratings and tapers, and the entire device was on a siliconon-insulator substrate covered by a thin silica spacer (total device length: 25 µm). The FWM process was attributed to the plasmonic slot mode profile: Fig. 19c, since the γ of the modes guided by all other plasmonic elements-such as the taper region shown in Fig. 19b-was negligible. The authors measured a maximum signal-to-idler conversion efficiency of −13.3 dB, (i.e., 4.7%), as shown in Fig. 19d, and longer device lengths led to a decrease in the conversion efficiency as shown in Fig. 19e, in agreement with theoretical predictions. Plasmonic waveguides with epsilon-near-zero materials Before moving to the next section, we briefly discuss a recent development in nonlinear plasmonics that has attracted much attention, namely the realization that bulk materials possessing a real part of the permittivity ε = √ n 0 that is close to zero (i.e., "epsilonnear-zero" (ENZ) materials) have an extremely large Kerr nonlinearity [197]. At first glance, when e(n 0 ) → 0 in Eq. (4), n 2 diverges-in fact, this is an artefact of the perturbative approach that was used to derive it [31]. In this case, changes in the intensity-dependent refractive index are more accurately described directly by [112] Experimentally, ENZ materials have been show to yield extraordinarily large refractive index changes of 170% in Indium Tin Oxide (ITO) [118], and similar effects were measured in Aluminium-Doped Zinc Oxide (AZO) [198], and artificial metamaterials [199]-see for example Ref. [197] for a recent review of ENZ media. Figure 20a shows the relative electric permittivity of ITO, and Fig. 20b shows the associated n 2 according to Eq. (4), with the largest n 2 occurring where e(ε m ) = 0. But how to to harness ENZ materials for guided-wave devices with extreme nonlinearities? Reported approaches include operating a waveguide containing an ENZ material at the frequency where ε m = 0 [201][202][203][204], or operating the waveguide with effective mode permittivity near cutoff such that e(ε eff ) = e(n 2 eff ) = 0 [205][206][207]. Further insight can be obtained by noting that, according to Eq. (14), large nonlinear changes in n can also be driven by large \|E\| 2 . This is the case for bulk ENZ media [77]: the transverse field has a local maximum at the ENZ wavelength, since it corresponds to a local minimum in the group velocity [77]. Furthermore, the longitudinal field can be further enhanced for TM polarization at angled incidence [118] due to the continuity of the normal component of the displacement field [197]. For waveguides formed by ENZ media, evaluating the nonlinear response requires calculating the nonlinear coefficient γ via Eq. (7), although insights can also be obtained from the factorization of Eq. (8). It is valuable to consider the simple case of a bulk SPP propagating at an air/ITO interface: Fig. 20c shows its real-and imaginaryparts as a function of frequency, and Fig. 20d shows the calculated associated γ according to Eq. (7). In contrast to the bulk case, the largest Kerr nonlinearities here occur at frequencies near ε m = −1, which is the point of the lossless electrostatic surface plasmon polariton [41]. A recent study also computed the associated v g and effective modal area, showing that these two parameters are indeed simultaneously minimized near this electrostatic plasmon resonance condition [77]. Similar calculations on other plasmonic waveguides led to the same conclusion. One key message of this analysis was that the enhanced Kerr nonlinearity in both bulk ENZ media and guided-wave structures can be understood in this unified framework of omnidirectional field enhancement. In all cases, the associated losses are quite large, even by plasmonic standards: the calculated attenuation lengths for ITO nanowires/nano-apertures are of the order of 50-100 nm, suggesting that, rather than wavelength-scale waveguides, sub-wavelength-thickness metasurface arrays (e.g., pillars and nanoholes) are most appropriate for boosting Kerr nonlinear responses of ENZ media. A number of experiments have been performed on similar planar ENZ metamaterials [199] and metasurfaces [119], extending the available wavelengths where giant optical nonlinearities can be harnessed. In the present context of guided-wave structures, ultrafast all-optical switching was most recently measured using bulk ITO surface plasmons near the ENZ wavelength using a Kretschmann configuration [200], as shown in Fig. 20e. Analogous experiments in thin films showed third harmonic generation enhancements [208]. Such materials and geometries are compatible with CMOS fabrication technologies. Given these promising results, future studies will undoubtedly elucidate the subtle and counter-intuitive physics underlying the large nonlinearities of ENZ materials, clarifying their feasibility as mass-producible components for chipcompatible sub-wavelength nonlinear devices. Nonlinear plasmonic circuits The structures in Sect. 4 show the impressive potential of guided-wave nonlinear plasmonic applications of individual, self-standing devices. Integrating or post-processing similar nonlinear plasmonic structures on readily available off-the-shelf dielectric waveguides has the power to grant them with additional, previously absent plasmonic functionalities while retaining a compact footprint. Recently for example, Tuniz et al. developed a HPWG circuit formed by two back-to-back hybrid plasmonic modules (namely, a plasmonic rotator and focuser, shown in Fig. 21a), both of which were integrated on a standard silicon photonic waveguide. Over the length of the 9 µm HPWG device, the authors show modal rotation (from TE to TM) and subsequent nanofocusing (via a tapered plasmonic tip), which leads to an enhancement of second harmonic generation due to the surface χ (2) effects of gold. The authors harness the enhancement of nonlinear light generation to experimentally demonstrate a field enhancement of more than 100× scattered from increasingly sharp tips, as shown in Fig. 21b, c, down to an estimated mode area of 100 nm 2 . Although the SHG conversion efficiency was only ∼ 10 −11 , these proof-of-concept experiments exemplify pathways for enhancing existing networks of photonic circuits with multiple sub-wavelength plasmonic nonlinear functions. A number of dielectric-plasmonic waveguide circuits, designed ab-initio, have unlocked wavelength-scale all-optical switching, electro-optics, and terahertz detection and generation, as we now discuss. All-optical switching Recently, Ono et al. used nonlinear plasmonic slot waveguides to address the well-known tradeoffs between all-optical switching speeds and associated energy requirements [29,146], using graphene as the nonlinear material in the slot. Their structures interface a silicon photonic circuit and a plasmonic slot waveguide with a graphene layer directly on top of the metal, as shown in Fig. 21a. While two-dimensional materials such as graphene [210] have extreme nonlinear optical properties, the optical interactions are still relatively weak due to the short moleculescale lengths over which nonlinear interactions occur. The authors overcome this limitation by combining the plasmonic hotspots at the edge of the gold metal (shown in the Fig. 21e calculations) and the high photonic-to-plasmonic efficiency of the plasmonic taper section [211], over micron-scale interaction lengths. Graphene's ultrafast saturable absorption (SA) thereby leads to the transmission of a signal pulse when a control pulse overlapped with it. Figure 21f shows the associated experimental transmission through the entire device as a function of pulse delay, highlighting the ultrafast response time of 260 fs. Electro-optics Several chip-compatible hybrid plasmonic devices that harness the χ (2) linear electrooptic effect have also been reported, enabling compact, low-power, and high-speed data modulation [151], terahertz detection [150] and generation [212]. Pockels-effect nonlinear modulators compete with those harnessing free-carrier [213], thermo-optic [214], or mechanical effects [215], due to the wide bandwidth and reduced power consumption in a micro-scale physical footprint. The driving physical principles are analogous to those described so far: a dielectric photonic waveguide funnels light to one-or multiple-plasmonic element-most commonly, a nanometre-scale plasmonic slot waveguide-containing a large χ (2) material [19]. Besides providing a large mode overlap between optical fields and a low-frequency (typically, GHz or THz) fields, the plasmonic slot waveguide is also a capacitor, providing a natural bridge between nanooptics and micro-electronics. Whether it be induced by external electrical signals [194], or external THz radiation [150], electric fields inside the slot can modulate the index change inside the slots, encoded as phase changes of an incoming constant-wave (CW) laser. The outgoing optical signals can then be detected with conventional spectrum analyzers or coherent receivers. The bandwidth of such devices is more than 1 THz, with the dielectric material itself having a response time of a few femtoseconds [194]. In this context, Melyikan et al. [194] reported the first experimental demonstration of a high-speed plasmonics phase modulator (40 Gbit/s) over a 29 µm-length MDM slot waveguide containing an electro-optic polymer. The concept was then extended to a two-arm configuration forming a Mach-Zehnder modulator (MZM), shown in the Fig. 22a, and whose performance is exemplified by the simulations in Fig. 22b. Here, two out-of-phase plasmonic waveguides ("off" states) are brought in phase ("on" states) via external electrical signals, directly encoding the external electrical signals on the incoming laser intensity via the power transfer function shown in Fig. 22c and with 70 GHz bandwidth. In these experiments, the plasmonic slot interfaced with a dielectric waveguide via an adiabatic taper, with the whole process being compatible with CMOS fabrication. In the spirit of relaxing fabrication requirements while maintaining high performance, this concept was used in an all-metallic device surrounded by the same nonlinear polymer [216]. In this case, the polarization of the electric field of the grating-coupled surface plasmon rotates-from the upper surface gold layer into the lateral plasmonic slots-and an external 116 GB/s electrical data stream was encoded into the optical signal. In a more recent resonant-switch design [217], the overall losses of the dielectric-plasmonic modulator were reduced by ensuring that the "on" state remains in the dielectric, while the "off" state couples to the lossy plasmonic mode, thereby harnessing the advantages of both dielectrics and plasmonics. Related designs are increasingly being included on monolithic chips of increasing sophistication [154]. The field of plasmonic-organic hybrid integration is rapidly developing; we point the reader to Refs. [19,152,153] for recent related reviews. Terahertz detection and generation The THz bandwidth associated with the nonlinear electro-optic devices presented above can also be harnessed for all-optical detection of electromagnetic fields at terahertz frequencies. Terahertz radiation is an enabling and rapidly developing multidisciplinary technology serving many diverse areas including security, telecommunications, and sensing [218]. However, as a relatively new technology, THz sources and detectors are less developed, typically bulky due to the relative large millimetre scale wavelengths involved, and are not particularly efficient in interfacing with conventional optical elements and photonic circuitry. Plasmonic nonlinear devices are increasingly bridging these technological gaps using χ (2) effects. Salamin et al. [150] experimentally demonstrated wirelessly driven plasmonic phase modulator that can directly encode a data from an external millimetre wave (0.06 THz) incident electric field on an optical carrier within an optical waveguide circuit, enhancing the low modal overlap between the incoming field and the optical wave via an appropriately designed resonance. This technology was recently adapted to even higher THz frequencies [219], and formed the basis for a low-footprint monolithic terahertz field detector [220]. This technology is rapidly moving out of the laboratory and into practical settings [221]-for example, Mach-Zehnder plasmonic configurations have been used as wireless THz-to-optical wireless receivers with 0.36 THz 3 dB bandwidth for 50 Gbit/s data streams [222]. Such architectures make terahertz technology more accessible, since it can be interfaced with conventional photonic structures (including optical fibers), and will likely be key in next-generation THz communications and portable low-cost THz detectors and terahertz imaging systems. The generation of broadband terahertz radiation, on the other hand, most commonly relies on transient currents in a biased photoconductor micro-antenna illuminated by femtosecond pulses [218]. As an alternative, all-optical terahertz sources can use difference frequency generation, a χ (2) process wherein two intense electric fields at THz-spaced frequencies generate a nonlinear polarization in the medium at the difference frequency. These schemes typically require phase-matching between the terahertz envelope and the beating optical waves, e.g., in mm-thickness crystals [223]. Yao et al. presented a microscale, chip-based structures using two-layer gated graphene heterostructure (each graphene layer separated by AlO 3 ), placed on top of a SiN 3 waveguide; conceptually, graphene forms an atom-thick plasmonic waveguide with a gate-tuneable permittivity, and large χ (2) . Counter propagating pump and signal photons phase match with the supported graphene plasmons, which can be appropriately externally tuned. Note the extreme properties of the plasmons involved: the generated terahertz plasmons have frequencies of 4-9 THz, and effective wavelength of 460-770 nm, corresponding to n eff = 50−120. Here the conversion efficiency is ∼ 10 −4 , limited by the propagation length of the graphene plasmons. An alternative approach for generating THz radiation is optical rectification (OR), whereby ultrashort optical pulses generate terahertz pulses in a χ (2) media as a result of the nonlinear interactions between the pulse's constituent THz-bandwidth frequencies [224]. OR has been harnessed to generate terahertz radiation on planar metal nanofilms [225] plasmonic nanoparticle arrays [226], and metamaterial arrays [227] but the best of our knowledge has yet to be reported in photonic-plasmonic waveguides. Surface-enhanced Raman scattering Finally, we mention one of the most widely-used nonlinear effects in plasmonics: Surface-Enhanced Raman Scattering (SERS) [228], wherein the large field enhancements enabled by plasmonics boosts the spectral fingerprints emerging from inelastic scattering processes between light and a molecule's vibrational modes. For many decades, SERS used localized, non-guided surface plasmon polaritons, e.g., via rough surfaces [229] and nanoparticles [230]. More recently, "remote" SERS has been developed [231][232][233], combining propagating SPPs (e.g., on a nanowire) with neighbouring localized SPPs, e.g., at its extremity. In such nonlinear plasmonic structures, the objective is to locally generate extreme fields in the smallest possible volume, and detect the Raman-shifted fingerprint at longer wavelengths; conventional guided guided surface plasmons generally do not provide sufficient enhancement, and plasmonic tips (i.e., tapered and terminated plasmonic waveguides) are used [234]. These provide the important advantage of a background-free nonlinear Raman signal originating from a nanoscale volume of interest. More recently, chip-compatible SERS devices that integrate plasmonic antennas [59,235] and plasmonic slot waveguides [236,237] have All pumps are in the NIR also emerged, whose modes are optimized to ensure the dominant Raman contribution comes from the slot by limiting the modal overlap with the dielectric waveguide [237]. Such sensors will also benefit from more efficient plasmonic coupling designs concentrating light to ever decreasing nanoscale volumes. We refer the reader to Ref. [228] for a recent review on SERS, which includes a comprehensive section on waveguide-based approaches. Nonlinear quantum plasmonics Photonics is one of the more promising platforms underpinning next-generation quantum-based technologies, e.g., quantum computing [130,238], secure communications [239], and quantum-enhanced metrology [240]. Light has been a workhorse for investigating quantum mechanics since the early days [241,242]; most recently, integrated optical platforms are playing an increasingly important role, promising to provide a noise-free monolithic means of conveniently and reliably generating, manipulating, and detecting single-and entangled-photons [243][244][245]. In keeping with the theme of this review, we now briefly discuss nonlinear plasmonics for quantum applications in the specific context of integrated waveguides. The nonlinear effects considered thus far operate at high (pump) photon numbers and weak nonlinearities (grey box of Fig. 24 [246]). A material's χ (2) or χ (3) nonlinearity can also produce entangled photon states at frequencies far from the pump, via spontaneous parametric downconversion (SPDC) and spontaneous four- Classical nonlinear optics (dark grey) relies on higher photon numbers, but materials' intrinsic interaction strength per photon is low. Quantum nonlinear enters the picture at larger interaction strengths per photons becomes large: if the photon number is small, photon-photon nonlinear optics takes place at the quantum emitter level (blue); if the photon number is large, and the interaction strength per photon is large, many photons interact simultaneously to produce strongly correlated many-body states. Adapted with permission from Ref. [246], Copyright (2018) Springer Nature wave mixing (SFWM) respectively [244]. On the other hand, nonlinear effects at the single-photon level provide a means of generating quantum states and rely on strong interaction strengths with matter per photon, as shown in the blue box of Fig. 24. At low photon numbers, the nonlinear interaction between two photons can be mediated by each photon strongly interacting with a quantum emitter (blue box). Here, typical schemes require photons to interact sequentially with a quantum emitter [247][248][249]the presence of the first photon is imprinted on the quantum emitter by changing its internal state, which influences the second photon, so that the quantum emitter induces a photon-photon interaction. Reaching the realm of quantum many-body nonlinear optics (yellow box) can open the door for creating entangled many-body states of photons. This requires both a large number of photons and a large nonlinearity per photon. One proposal has shown that many-body states of light can be generated by unidirectionally coupling many quantum emitters to a waveguide [250]. This can in principle also be achieved in plasmonic waveguides [251], although practical implementations may be limited by loss. Another possible route to reach this limit can potentially be achieved by using the material response of a given waveguide configuration, analogously to the requirement of a large nonlinear phase shift (Eq. (11)). What role, if any, can nonlinear plasmonics play in all this? One advantage is that that quantum nonlinear effects are more likely when the effective volume occupied by photons approaches the deep sub-wavelength scale, provided by nano-focused plasmonic modes. Complementarily, plasmonic nanostructures increase the density of available optical states [252], increasing the probability of photon emission, so that photons emitted by a quantum emitter can couple to surface plasmon modes near-deterministically [253]. This requirement is key both for efficient on-chip photon sources and for strong photon-photon nonlinearities [254]. However, care should be taken in ensuring that the associated enhancement in emission does not couple to a non-radiating channel (e.g., loss due to damping): proper emitter placement near the metal is extremely important to avoid detrimental quenching effects [252]. At a fundamental level, any useful single photon state is immediately destroyed by the loss of any photon, which often raises eyebrows when suggesting lossy plasmonic systems as viable quantum platforms. However, a number of recent experiments of on-chip quantum emitters [255], complemented by analytical theories [256] indicate that quantum plasmonics [257,258] can enhance the capabilities of all-dielectric architectures [252,259,260]. With ever improving circuit designs for coupling dielectric waveguide modes to single quantum emitters [261], one advantage of plasmonically coupled emitters over their all-dielectric counterparts is their broadband, non-resonant, enhanced emission rate [256] and thus shorter emitter lifetime, which could facilitate the generation of a coherent source of single photons that is required for most quantum protocols. One perspective [262] is that plasmonic devices reduce the spontaneous emission time t sp times below the characteristic dephasing times t deph at room temperature; dielectricbased approaches instead increase t deph by reducing the temperature as illustrated in Fig. 25a. An example feasibility study of efficient room-temperature sources of indistinguishable single photons using plasmonic cavities was reported in Ref. [263]. Several recent experiments have shown the promise of photonic-plasmonic quantum architectures. For example, Gong et al. used three-dimensional guided plasmonic nanofocusing on a deterministally positioned quantum emitter to enhance its spontaneous emission by a factor of ∼ 22. Most recently, a single-molecule nonlinearity was experimentally shown via a dye molecule inside a plasmonic waveguide [264], and the resulting single-photon fluoresence showed a one-order of magnitude reduction in emission lifetime compared to the non-plasmonic case. Grandi et al. [265] included a single molecule into a hybrid gap plasmon waveguide akin to that shown in Fig. 19a, showing single molecule emission from the output of the entire device, which originated from the plasmonic nano-gap, although the plasmonic gap of 200 nm was too wide to reduce the decay rate. With ever-improving techniques for deterministic placements of quantum emitters [266], and the ability to controllably pattern nanometre-scale metallic channels [267], similar geometries might provide the building block for fast room-temperature single-photon emitters that coupled to low-loss dielectric guides assited by plasmonics, as per the schematic of Fig. 10b. Guided-wave multiphoton nonlinearities have been recently theoretically and experimentally revisited for guided lossy media in the context of quantum applications. In 2016, Poddubny et al. [269] developed general theoretical framework of integrated nonlinear parametric photon-plasmons guided waves, accounting for material dispersion and losses. Such realistic studies suggested relatively high efficiency of 70%, and even presented novel enhancement mechanisms due to the anisotropic eigenmode topology of metal/dielectric multilayers. New toolkits for dealing with nonlinear quantum processes in lossy media are continuously being developed [270,271]. Experiments that rely on nonlinear plasmonics processes to generate quantum states are rare: guiding entangeld multi-photon states through the lossy media too easily destroys them. Recent efforts have attempted to use guided surface plasmon polaritons to enhance spontaneous parametric downconversion [272], and some initial steps have Fig. 25 a Schematic illustrating plasmonic speedup of single-photon emitters. "Coherent" photons from quantum emitters occur from spontaneous transitions from an excited state \|e > to a ground state \|g >. Coherence requires that the spontaneous emission time t sp be shorter than dephasing events with characteristic time scale t deph . Dielectrics optain long t deph by cooling; plasmonics achieve short t sp via a fast spontaneous emission rate (i.e., large Purcell factor). Adapted from Ref. [262]. Reprinted with permission from AAAS. b Schematic of a quantum dot array deterministically placed near a plasmonic nanofocus. Inset: SEM image of the fabricated device (Scale bar: 500 nm). d Concept schematic of device principle: a three-dimensional nanofocused plasmonic mode is aligned with a single emitter at the nanofocus. e Example decay dynamics of the emitter: the decay time is 4 ns for the single QD without a silver film (blue), and 0.14 ns for the QD with a silver film (red). (Black line: instrument response function.) Adapted from Ref. [268]. Copyright (2015) National Academy of Sciences been made [273]; stronger nonlinearities, lower losses, and hybrid waveguide designs [271], could potentially overcome current limitations. Although light-matter interactions are weaker in all-dielectric structures, the library of photonic elements (e.g., couplers, splitters, etc.) is better established, more flexible, and thus provides a more convenient platform for more advanced early experiments. Integrated plasmonics could potentially miniaturize these systems to the nanoscale, lower the energy requirements, and provide faster room temperature operation; currently however, the majority of quantum photonic experiments are still confined to research laboratories, where the absence of such characteristics do not preclude fundamental studies of chip-scale quantum interactions in these early research stages. Plasmonics-based approaches might however become the go-to later-generation technology for quantum photonic architectures, once they become more widespread. Conclusions and outlook We have provided an introductory overview of nonlinear plasmonic in guided wave systems, which we believe will play an important role in the next generation of compact, ultrafast, low-power photonic integrated devices. We have mentioned a few notable applications, including all-optical switching, terahertz generation, electrooptics, single-molecule sensing, and quantum optics, but this list is by no means exhaustive [51]. While plasmonics-based guided-wave structures are capable of extreme nonlinear optics inside deep sub-diffraction volumes, they push nanofabrication demands to the limit of current capabilities, and demand a lot from the materials involved-often operating at the edge of their breaking point (albeit at lower powers). However, recent years have been marked by the explosion of a huge family of highly nonlinear twodimensional (2D) materials, some of which have been mentioned here. The most famous of these, graphene, supports plasmonic modes [274][275][276] and can also act as a highly nonlinear medium for enhancing dielectric waveguides [277]. 2D materials have large nonlinear susceptibilities, but under standard illumination the interaction length is only a few atoms thick: guided-wave plasmonics [278] can provide a way of concentrating the light to a volume comparable to the thickness of the material itselfnot to mention interaction lengths orders of magnitude longer than the width of a few atoms! We have already seen the power of these combined features in the device of Fig. 21d-f, although a complete description at such scales must also account for nonlocal effects [279]. The role of plasmonics in enhancing the performance of such 2D materials has been the topic of recent reviews [280,281], and it is only a matter of time before guided-wave hybrid nonlinear plasmonic devices, enhanced by 2D materials, integrate with PICs to unlock record-level ultrafast nonlinear effects in an accessible manner. Photonic-plasmonic-2D circuits are now starting to appear [282], albeit in a different context, and current fabrication capabilities enable a scalable approach for including 2D materials on large-area waveguides [283,284]. Complementary to approaching improvements from a material perspective, it may be that other waveguide geometries may provide enhanced nonlinear interactions as a pathway for investigating new physics-for example, non-Hermitian systems [285], accessible via plasmonic waveguides [286], exhibit slow light effects at their exceptional point [287], where they are also extremely sensitive to their environment [288]. Related concepts [289] might prove a worthwhile avenue for chip-based nonlinear sensing of nanoscale events. In conclusion we hope that, as alternate avenues for nonlinear enhancement emerge, as fabrication techniques develop, and as material science further matures, this tutorialstyle review may provide a useful introductory conceptual toolkit for approaching this exciting and powerful field. Acknowledgements The author sincerely thanks C. Martijn de Sterke, Guangyuan (Clark) Li, Gordon H. Li, Stefano Palomba, Fernando J. Diaz, and Loris Marini for countless stimulating discussions on several topics contained in this review. The author also thanks Sahand Mahmoodian for valuable insights on nonlinear quantum photonics, and Birgit Stiller for providing feedback on a version of this manuscript. This work was in part supported by the University of Sydney Postdoctoral Fellowship scheme at the University of Sydney Nano Institute. The author is the recipient of an Australian Research Council Discovery Early Career Award (project number DE200101041) funded by the Australian Government. Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.	21,861	2021-04-01T00:00:00.000	[ "Physics" ]
Polysiloxanes and Silanes with Various Functional Groups—New Compounds for Flax Fibers’ Modification There is an increasing desire to use natural products that will be both effective and biodegradable. The aim of this work is to investigate the effect of modifying flax fibers with silicon compounds (silanes and polysiloxanes), as well as examining the effect of the mercerization process on their properties. Two types of polysiloxanes have been synthesized and confirmed by infrared spectroscopy (FTIR) and nuclear magnetic resonance spectroscopy (NMR). Scanning electron microscopy (SEM), FTIR, thermogravimetry analysis (TGA) and pyrolysis-combustion flow calorimetry (PCFC) tests of the fibers were performed. On the SEM pictures, flax fibers purified and covered with silanes were visible after treatment. FTIR analysis showed stable bonds between the fibers and the silicon compounds. Promising results of thermal stability were obtained. It was also found that modification had a positive effect on the flammability. The conducted research showed that the use of such modifications, in the context of using flax fibers for composites, can yield very good results. Introduction Natural fibers have been gaining popularity for a long time due to their biodegradability, and thus, their environmentally friendly nature. As a renewable raw material, natural fibers are used to produce recyclable ecological products, thereby reducing carbon dioxide emissions and the amount of waste generated by industrial processes. Flax fibers are used on a large scale by various industries, mainly for the production of textiles, paper and composites. This work focused on the appropriate, effective two-step modification (mercerization followed by silanization) of flax fibers using specially synthesized polysiloxanes and commercially available silanes with different functional groups for their later use in composites. Such modification is necessary to obtain good adhesion in the composite. This has been a known problem for many years-the different polarity of the fibers and polymers-hydrophilic and hydrophobic, respectively-makes it difficult to combine them effectively [1]. To obtain a change in the polarity of natural fibers, some chemical modifications are used, such as: mercerization [2], acetylation [3], acrylation [4], benzoylation [5], silanization [6], peroxide treatment [7], isocyanate treatment [8,9] and enzymatic treatment [10]. Two methods of modification, mercerization and silanization, were used in this work. In the process of mercerization, the reactivity of cellulose is increased by breaking the hydrogen bonds due to of the action of NaOH. That allows for a better wetting of the fibers. Moreover, NaOH can transform cellulose-I to cellulose-II [11,12]. Alkaline treatment Formulas of the modifiers are shown in Table 1. Synthesis of Polysiloxanes Synthesis of difunctional polysiloxanes was carried out in two steps. The schematic reaction is presented on the Scheme 1 below. Synthesis of Polysiloxanes Synthesis of difunctional polysiloxanes was carried out in two steps. The schematic reaction is presented on the Scheme 1 below. Polysiloxane with alkoxy groups PS1 Synthesis of Polysiloxanes Synthesis of difunctional polysiloxanes was carried out in two steps. The schematic reaction is presented on the Scheme 1 below. z = 50, y = 10, x = 15, R = (CH 2 ) 7 CH 3 Synthesis of Polysiloxanes Synthesis of difunctional polysiloxanes was carried out in two steps. The schematic reaction is presented on the Scheme 1 below. Polysiloxanes containing long hydrocarbon chain groups and vinyltrimethoxysilane groups were synthesized by the hydrosilylation reaction of poly(dimethyl-co-hydromethyl)siloxane, 1-octene, and subsequently vinyltrimethoxysilane. The process was carried out in the presence of catalyst-Karstedt complex [Pt 2 {(CH 2 =CHSiMe 2 ) 2 O} 3 ]. At first, poly(dimethyl-cohydromethyl)siloxane, 1-octene and the catalyst (5 × 10 −5 mol Pt per mol Si-H) were put into a three-neck round-bottom flask with a thermometer, reflux condenser and magnetic bar at room temperature. The solution was then heated to appropriate temperature. Olefin conversion was monitored by FTIR analysis. Upon completion, the appropriate amount of the second olefin, vinyltrimethoxysilane, was added with a 10% excess. The solution was then kept at the set temperature for another hour. The reaction mixture was then cooled and the excess olefin was evaporated under reduced pressure. The structure of the obtained products was confirmed by NMR analysis. In the case of vinyltrimethoxysilane (PS1), the synthesis proceeded in the same way without 1-octene substitution. Fibers Preparation and Modification Flax fibers were dried at 50-55 • C for 24 h. Then, the material was disintegrated on a knife mill Retsch SM-200 (Haan, Germany) with a sieve that had a mesh size of 3 mm. Mercerization The flax fibers were treated with 10% (w/w) NaOH aqueous solution for 10 min at room temperature. The NaOH/fibers weight ratio was 10:1. The fibers were then washed repeatedly in fresh distilled water until a neutral pH was obtained. Finally, all treated fibers were dried at 50 • C for 48 h. Modification with Silanes After the mercerization process, one method of modifying the fibers was the reaction with two silanes with different properties-the more polar N-(2-aminoethyl)-3aminopropyltrimethoxysilane and the less polar vinyl trimethoxysilane. An ethanol/water solution in the ratio of 6/4 (v/v) was prepared, to which the appropriate silane was added in the amount of 5% (w/w). Then, in the case of vinyl silane, the solution was acidified with acetic acid to pH 4.5. The hydrolysis process was carried out for 1 h. Fibers were then added and modification was carried out for 2 h at room temperature. The fibers were then placed in an oven set at 40 • C and dried. They were then cured for 10 min at 105 • C. Modifiaction with Polysiloxanes An alternative method of fiber modification before and after the mercerization process (other than in the Section 2.3.2.) was silanization with two polysilooxanes with different functional groups and properties. The first was polisiloxane with alkoxy groups, and the second was difunctional polysilaxane with akloxy groups and alkyl chains. Modification with polysiloxanes was carried out in two ways: under the same conditions as modifications with silanes (ethanol/water solution in the ratio of 6/4 (v/v)) and in the isopropanol/water solution in the ratio of 19/1 (v/v) (samples marked by "ip"). The appropriate polysiloxane was added in the amount of 5% (w/w). Then, the solution was acidified with acetic acid to pH 4.5. The rest of the procedure was performed as in Section 2.3.2. A one-hour hydrolysis process was carried out. The fibers were then added, and a two-hour reaction was carried out at room temperature. Next, the fibers were placed in an oven set at 40 • C and dried. Curing was carried out at a temperature of 105 • C for 10 min. The analysis of FTIR spectroscopy of modified fibers was performed in KBr pellets using a BRUKER IFS 66v/S spectroscope (Billerica, MA, USA) in the mid-infrared range of 4000-400 cm −1 with a resolution of 2 cm −1 . Thermal Stability Tests Thermogravimetric study (TGA) was performed with TA Instruments, Analyser Q50 (New Castle, DE, USA). A 15 ± 1 mg fiber sample was heated to 700 • C at a heating rate of 10 • C·min −1 under a nitrogen atmosphere with a constant gas flow rate of 90 mL·min −1 . The mass loss curve and the first derivative of TG (DTG) were determined. Flammability Tests Flammability tests were performed using a pyrolytic combustion flow calorimeter (PCFC) by FTT (Grinstead, UK) for fiber samples weighing 5±1 mg. Testing was carried out in accordance with ASTM D7309-2007. The heating rate was 1 • C·s −1 . The pyrolysis temperature range was 75-500 • C, and the combustion temperature was 900 • C. The flow was a mixture of oxygen and nitrogen gases at a ratio of 20:80 cm 3 ·min −1 . The maximum heat release rate (HRR max ) was determined. Scanning Electron Microscopy Microscopic test photos of longitudinal views of flax fibers were made with a Hitachi S-3400N scanning electron microscope (SEM) using a secondary electron detector SE in a high vacuum mode. Prior to the tests, the fibers were sprayed with a gold layer. The value of the accelerating voltage was 20 kV, and the working distance was 20 mm. Magnifications of 500× were selected. Results First, the obtained polysiloxanes were analyzed by FTIR and NMR methods. Subsequently, analyses of raw and modified flax fibers were carried out. The results were divided into two sections: the results of polysiloxanes synthesis and the results of flax fibers. For better understanding, Table 2 containing all samples can be found below: Table 2. All samples prepared in the research. Scanning Electron Microscopy Images of Flax Fibers The effect of the modification and the presence of silanes/polysiloxanes on the surface of flax fibers was confirmed by SEM. Surface morphology of flax fibers before and after modifications were investigated. Figure 2 shows SEM images of unmodified and modified flax fibers. Longitudinal electron micrograph images of the fibers were taken in two magnifications: 500 and 1500 times. Scanning Electron Microscopy Images of Flax Fibers The effect of the modification and the presence of silanes/polysiloxanes on the surface of flax fibers was confirmed by SEM. Surface morphology of flax fibers before and after modifications were investigated. Figure 2 shows SEM images of unmodified and modified flax fibers. Longitudinal electron micrograph images of the fibers were taken in two magnifications: 500 and 1500 times. Flax fibers without mercerization process Flax fibers with mercerization process res 1. Microscopic analysis of the surface morphology of the fibers is of great importance for characterizing the structural changes that have occurred after treatment. The main role of NaOH treatment is to clean the fiber of impurities and prepare the fiber for further processing. This is a widely used chemical process which removes noncellulosic components and part of the amorphous cellulose [25]. In turn, silane treatment is very helpful in Microscopic analysis of the surface morphology of the fibers is of great importance for characterizing the structural changes that have occurred after treatment. The main role of NaOH treatment is to clean the fiber of impurities and prepare the fiber for further processing. This is a widely used chemical process which removes noncellulosic components and part of the amorphous cellulose [25]. In turn, silane treatment is very helpful in removing lignin and hemicelluloses from natural fibers. In the photos of raw flax fibers, impurities on the fiber are clearly visible. After the mercerization process, the fibers are visibly cleaned. Aminosilane and vinylsilane modified fibers, both with and without prior mercerization, are also purified and covered with a thin layer of silanes. However, there is a clear difference in the photos of fibers modified with polysiloxanes. The fibers have been evenly covered with a layer of polysiloxanes, and their surface is smooth and clean. It was observed that the modifier covered the surface of the fibers in a relatively thick but smooth layer. Many researchers have noticed that the fibers after NaOH treatment become purified and more susceptible to the action of silanes. In turn, modification with silane allows the fibers to be covered with a uniform layer, and as such, the fibers' microscopic images show a smoothened surface on the fibers [26]. Puglia et al. [27] reported that NaOH can remove surface impurities from fibers while silane makes them smoother. SEM images from Liu et al. [28] showed that the surface morphology of treated corn stalk waste fibers was slightly rough and relatively clean after silane treatments. Scanning electron microscopy is a great method for characterizing fibers and the effects of modifications on their texture. However, to confirm that, in addition to physical changes on the surface of the fibers, stable bonds between the fibers and the silane have been formed, it is worthwhile to carry out other analyses. The mere adsorption of silane to the fiber will not improve the adhesion between the silane and the polymer in the composite [29]. Accuracy of diameter measurement of natural fibers is very difficult to achieve because natural fibers are irregular in shape and thickness [30]. Mercerization is a process that "cleans" the fiber of waxes, pectins, etc., which naturally leads to a reduction in the diameter of the fibers. On the other hand, modification with silanes or polysiloxanes causes the modifier to bind to the fiber and form a layer covering the fiber, which can lead to an increase in the diameter of the fibers. The diameters were tested based on images from a scanning electron microscope. Despite large divergencies in the size of the fibers (Figure 3), it was observed that after mercerization, the average diameter of the fibers slightly decreases, while in the case of modification with silicon compounds, it increases. The largest increase in average diameter was observed in the case of aminosilane modification. Interestingly, with the mercerization and polysiloxane 1 modification, a decrease in the diameter of the fibers was observed. FTIR Tests Using FTIR spectroscopy, structural changes on the fiber surface after silane treatment were examined, which confirmed that the silicon compounds were chemically grafted onto the fiber surface. The following Figure 4a shows the IR spectra for samples modified in one step (silanization only) and Figure 4b shows the IR spectra for samples modified in two steps (mercerization and silanization). Characteristic absorption bands for cellulose molecules appear in all tested samples [31][32][33]. The wide band, ranging from about 3000 cm −1 to 3500 cm −1 , comes from the stretching vibrations of the O-H groups in the cellulose. In all cases, the band after modification is less intense, which indicates the occurrence of bonding between alkoxysilanes and the fibers. The diameters were tested based on images from a scanning electron microscope. Despite large divergencies in the size of the fibers (Figure 3), it was observed that after mercerization, the average diameter of the fibers slightly decreases, while in the case of modification with silicon compounds, it increases. The largest increase in average diameter was observed in the case of aminosilane modification. Interestingly, with the mercerization and polysiloxane 1 modification, a decrease in the diameter of the fibers was observed. FTIR Tests Using FTIR spectroscopy, structural changes on the fiber surface after silane treatment were examined, which confirmed that the silicon compounds were chemically grafted onto the fiber surface. The following Figure 4a shows the IR spectra for samples modified in one step (silanization only) and Figure 4b shows the IR spectra for samples modified in two steps (mercerization and silanization). The bands in the spectral region of 2800-3000 cm −1 assigned to C-H stretching vibration are characteristic of alkylene (-CH 2 -) and alkyl (CH 3 ) groups, which were bound to fibers because of the modification with silicon compounds. The decrease in the intensity of these bands in some fiber samples may be caused by the reduction of the crystalline structure of cellulose. On the other hand, intensity of these bands increases because of the increased content of CH 2 groups in the silanes and polysiloxanes. This band increases especially after fiber modification with polysiloxane with a long alkyl chain (PS2). Moreover, in the spectra of samples modified with both types of polysiloxanes, an additional band appeared at 2962 cm −1 that is ascribed to CH 3 groups. The vibration band visible at 1734 cm −1 , resulting from the C=O stretching vibrations of the acetyl group in hemicellulose and aldehydes in lignin [34], disappears or is slightly reduced in the case of mercerized fibers. This is due to the degradation of hemicellulose and the dissolution of lignin during the alkali treatment of fibers. An absorption band in the range 1630-1650 cm −1 originates from the stretching vibrations of the O-H group and correspond to absorbed water in crystalline cellulose [35]. Other characteristic bands resulting from vibrations in the cellulose molecule (CH 2 bending vibrations) can be observed in all tested samples with wave number values of approx. 1430 cm −1 and 1370 cm −1 . In the spectra of fibers modified with aminosilane, a weak band at 1570 cm −1 , characteristic of primary amino groups, is seen. Furthermore, the spectra of the samples modified with polysiloxanes contain bands at 801 and 1260 cm −1 , originating from Si-O-Si symmetric stretching vibrations and the Si-O-C stretching vibration shoulder, respectively. In the region of 1000-1200 cm −1 , three characteristic bands appear for the cellulose molecule. At approx. 1160 cm −1 , a band of the asymmetric C−O−C stretching vibrations in cellulose was observed. At approx. 1110 cm −1 , a band of the C−OH skeletal vibration in cellulose was observed. At approx. 1050 cm −1 , C−O−C pyranose ring skeletal vibrations were ascribed to cellulose. These bands are reduced in all tested samples after modification, and can be attributed to the reduction in the crystalline structure of the cellulose after treatment. The band expected at 1018 cm −1 (Si-O-Si) overlapped with this broad region that corresponds to the characteristic peaks of cellulose [21]. Characteristic absorption bands for cellulose molecules appear in all tested samples [31][32][33]. The wide band, ranging from about 3000 cm −1 to 3500 cm −1 , comes from the stretching vibrations of the O-H groups in the cellulose. In all cases, the band after modification is less intense, which indicates the occurrence of bonding between alkoxysilanes and the fibers. The bands in the spectral region of 2800-3000 cm −1 assigned to C-H stretching vibration are characteristic of alkylene (-CH2-) and alkyl (CH3) groups, which were bound to fibers because of the modification with silicon compounds. The decrease in the intensity of these bands in some fiber samples may be caused by the reduction of the crystalline structure of cellulose. On the other hand, intensity of these bands increases because of the increased content of CH2 groups in the silanes and polysiloxanes. This band increases especially after fiber modification with polysiloxane with a long alkyl chain (PS2). Moreover, Thermal Stability Tests of Flax Fibers Natural fibers consist of hemicellulose, cellulose and lignin. Other ingredients, such as pectin and waxy substances, are not important in this context. The decomposition of natural fibers can be divided into four main stages (Yang et al. [36]). The first is the evaporation of moisture, followed by the decomposition of hemicellulose, and then the decomposition of cellulose and lignin. For a better understanding of the thermal properties of the fibers before and after silane treatment, information on the pyrolysis properties of these three main components is important. The analysis of the TGA/DTG curves showed that the flax fiber decomposition process can be divided into three main stages (see Figure 5). In addition, the second stage can be divided into two substages, which are not clearly visible in all cases since these stages overlap, but it was decided to determine them for all fiber samples. These steps vary depending on the modification used. the fibers before and after silane treatment, information on the pyrolysis properties of these three main components is important. The analysis of the TGA/DTG curves showed that the flax fiber decomposition process can be divided into three main stages (see Figure 5). In addition, the second stage can be divided into two substages, which are not clearly visible in all cases since these stages overlap, but it was decided to determine them for all fiber samples. These steps vary depending on the modification used. All characteristic points and stages of decomposition as well as the first derivative peak temperature (DTG peak) were included in Table 3 below. The first point to be considered is the temperature at which the fibers begin to de- All characteristic points and stages of decomposition as well as the first derivative peak temperature (DTG peak) were included in Table 3 below. The first point to be considered is the temperature at which the fibers begin to decompose. That was assumed as an onset temperature (T onset ). The temperature at this point was significantly higher for mercerized fibers, especially those additionally modified with polysiloxanes. The first stage of decomposition, reaching a temperature of about 150-170 • C, was the evaporation of water, and was characterized by a weight loss of 4.95-6.66%. Fibers modified with amine silane were characterized by the greatest weight loss in this area, but the differences between individual samples were small. The second stage-in which hemicellulose, amorphous cellulose and low-molecular compounds, i.e., waxes, pectin, etc., are decomposed-was separated for the temperature range of 176-227 • C to 305-320 • C, depending on the sample. It can be clearly seen that the beginning of this region shifts towards higher temperatures for mercerized fibers, except for fibers also modified with aminosilane. The weight loss in this step was 6.08-12.49, depending on the sample. The highest mass loss in this range is observed for samples modified with amine silane (both with and without prior mercerization). The third stage, which was the main stage of decomposition with the greatest mass loss (mainly cellulose degradation), was within the temperature range of 305-320 • C to 384-393 • C. At this stage, in the range 360-367 • C, depending on the type of fiber modification, the first derivative peak temperature (DTG peak) occurred. The peak of the first derivative indicates the point of greatest rate of change on the weight loss curve. This is also known as the inflection point. This point did not differ significantly between individual fiber samples. The mass of the tested samples decreased at the third stage by 41.48-63.9%. In this range, for samples modified with aminosilane, the lowest mass loss is observed (opposite to stage II). The fourth and last stage is the longest stage of decomposition, associated most probably with the slow degradation of lignin. The residue after the process ranged from 14.56-24.64%, depending on the sample. It can be noticed that the highest amounts (23.93% and 24.64%) of residual char belonged to the fibers modified with aminosilane. It is clearly visible that the use of different silanes with different functional groups, as well as different polysiloxanes, significantly affect the thermal stability of the tested fibers. An increase in thermal stability was observed at the initial decomposition temperature, and then at the shift of the second stage of decomposition for mercerized samples. Modification of natural fibers with silicon compounds resulted in the formation of a silica layer on the fiber surface. This layer can create a protective barrier from the thermal radiation and stamp out the release of combustible gases. Mercerization facilitated the bonding of silicon compounds with flax fibers. The flax fibers used in this research are very similar to another natural fiber-hemp fibers. They are also lignocellulosic fibers, and their modification occurs in a similar manner. Similar results can be expected with regard to thermal stability. In the literature, one can find that the use of the same two-step modification process on hemp fibers can lead to similar results. Rachini et al. [37] conducted research on the thermal stability of hemp fibers and the impact of mercerization and silanization on their thermal properties. He showed that both alkali treatment and silanization can improve the thermal decomposition of the hemp fibers. showed that both alkali treatment and silanization can improve the thermal decomposition of the hemp fibers. Figure 6 shows the HRR curves from the pyrolysis and combustion flow calorimeter (PCFC) test for samples of fibers modified only with silicon compounds (a), and first mercerized and then modified with silicon compounds (b). The use of an amine silane usually results in a lower HRR peak due to the presence of amino groups in its structure. Interestingly, the use of a polysiloxane with a long alkyl chain also resulted in a decrease in HRR to similar values, especially when using isopropanol as a solvent in the silanization reaction. In these cases, there was an approx. 25% reduction in HRRmax. For samples modified with polysiloxane without an alkyl chain, this reduction was slightly lower and amounted to about 10%. No significant differences in the height of the HRR curves were observed when comparing the mercerized samples ( Figure 6b) and those that did not undergo this process (Figure 6a). Modification with vinyl silane caused an approx. 10% increase in HRR, which is a normal phenomenon for this type of modification [24]. Microcalorimeter Tests of Flax Fibers The use of 95% isopropanol and only 5% water as a solvent for the polysiloxanes was preferable to the use of 60% ethanol and 40% of water as a solvent because polysiloxanes are very sensitive to water. If as much as 40% water was used, premature condensation of these compounds could occur, which in turn could hinder proper silanization. There are few literature reports in which the flammability of natural fibers modified only with silicon compounds without the use of additional flame retardants is tested. Most often, researchers use phosphorous agents to reduce the flammability of natural fibers [38,39]. Various combinations of flame retardants are used in the literature to achieve a synergistic effect, further reducing the flammability of the fibers [40]. Unfortunately, when these compounds are used, the thermal stability of natural fibers is often reduced [41]. This phenomenon can be a serious problem, especially if the fibers are ultimately to be used in composites, the processing temperatures for which are often above 150 °C. In turn, the use of silicon compounds to modify natural fibers has a positive effect on thermal stability [42]. Alkali and silane/polysiloxane treatment used in our study showed that this type of modification can positively affect the flammability properties of the fibers without decreasing their thermal stability. The use of an amine silane usually results in a lower HRR peak due to the presence of amino groups in its structure. Interestingly, the use of a polysiloxane with a long alkyl chain also resulted in a decrease in HRR to similar values, especially when using isopropanol as a solvent in the silanization reaction. In these cases, there was an approx. 25% reduction in HRR max . For samples modified with polysiloxane without an alkyl chain, this reduction was slightly lower and amounted to about 10%. No significant differences in the height of the HRR curves were observed when comparing the mercerized samples ( Figure 6b) and those that did not undergo this process (Figure 6a). Modification with vinyl silane caused an approx. 10% increase in HRR, which is a normal phenomenon for this type of modification [24]. The use of 95% isopropanol and only 5% water as a solvent for the polysiloxanes was preferable to the use of 60% ethanol and 40% of water as a solvent because polysiloxanes are very sensitive to water. If as much as 40% water was used, premature condensation of these compounds could occur, which in turn could hinder proper silanization. There are few literature reports in which the flammability of natural fibers modified only with silicon compounds without the use of additional flame retardants is tested. Most often, researchers use phosphorous agents to reduce the flammability of natural fibers [38,39]. Various combinations of flame retardants are used in the literature to achieve a synergistic effect, further reducing the flammability of the fibers [40]. Unfortunately, when these compounds are used, the thermal stability of natural fibers is often reduced [41]. This phenomenon can be a serious problem, especially if the fibers are ultimately to be used in composites, the processing temperatures for which are often above 150 • C. In turn, the use of silicon compounds to modify natural fibers has a positive effect on thermal stability [42]. Alkali and silane/polysiloxane treatment used in our study showed that this type of modification can positively affect the flammability properties of the fibers without decreasing their thermal stability. Conclusions Promising results of modification of natural fibers with silicon compounds were obtained in the work: • The successful synthesis of polysiloxanes was performed and confirmed by FTIR and NMR results; • SEM photos of the fibers showed that they were cleaned because of mercerization and their diameter was reduced, and during silanization, were covered with a thin, uniform layer of silicon compounds; • FTIR analysis showed that stable bonds between silanes or polysiloxanes and fibers were formed because of the modification; • A two-step modification of flax fibers (Na OH treatment and then modification with polysiloxanes) increased the thermal stability of the fibers and increased the temperature of the initial fiber decomposition; • Improvement of flammability properties was also obtained for the modifications with aminosilane and difunctional polysiloxane with a long alkyl chain. In summary, it was observed in the conducted research that alkali treatment was important in the modification of flax fibers. The positive effect of the performed mercerization can be seen both in SEM images and in thermal stability studies. In addition, the functional groups of the silanes and polysiloxanes used also have a significant impact on the obtained test results. The use of an amino group in silane visibly increases its flammability properties, but also lowers thermal stability and leaves the largest residue after the TGA test. The incorporation of an alkyl chain into the polysiloxane structure allowed for an excellent improvement in flammability properties. Particular attention should be paid to the fact that the use of polysiloxanes as an alternative to silanes is promising in the context of using of fibers in composites and obtaining good adhesion with the polymer matrix, due to the uniform coverage of the fibers and the formation of permanent bonds with them.	6,805.4	2023-05-01T00:00:00.000	[ "Materials Science" ]
Information-Theoretic Method for Assessing the Quality of Translations In recent years, the task of translating from one language to another has attracted wide attention from researchers due to numerous practical uses, ranging from the translation of various texts and speeches, including the so-called “machine” translation, to the dubbing of films and numerous other video materials. To study this problem, we propose to use the information-theoretic method for assessing the quality of translations. We based our approach on the classification of sources of text variability proposed by A.N. Kolmogorov: information content, form, and unconscious author’s style. It is clear that the unconscious “author’s” style is influenced by the translator. So researchers need special methods to determine how accurately the author’s style is conveyed, because it, in a sense, determines the quality of the translation. In this paper, we propose a method that allows us to estimate the quality of translation from different translators. The method is used to study translations of classical English-language works into Russian and, conversely, Russian classics into English. We successfully used this method to determine the attribution of literary texts. Introduction Translations from one language to another play an important role in the modern information society. Translations of a wide variety of data, including literary works, texts from newspapers and TV programs, and the content of various social networks, are used for dubbing films to maintain dialogues when participants speak different languages. Naturally, such a widespread use of translations arouses interest and contributes to the development of new translation methods, for example, various machine translation programs, and, in turn, the comparison of their quality [1][2][3]. It should be noted here that for centuries the task of comparing the quality of translations of literary works was solved by literary scholars using their traditional methods, and by now, significant material has been accumulated both on the analysis of specific individual translations of classical authors and on the development of general principles for the analysis of translations [4][5][6][7][8]. It is important to note the appearance in recent years of formal, mathematical methods for analyzing texts [9][10][11][12][13][14]. Nevertheless, despite a significant number of works devoted to the analysis of literary works, including translations, there are few generally accepted concepts and formal definitions in this area, for example, what a quality translation is. Among the fairly recognized and well-known theories, we present the classification of text variability sources proposed by Kolmogorov [15]: information content, form, and unconscious author's style. Moreover, the concept of unconscious author's style can also be applied to a separate literary work. In the case of translated works, the question of unconscious author's style is far from clear. The fact is that, apparently, the "style" of the text of the translation is created by both the author and the translator. It is interesting to note that the question of the influence of the translator on the style of the translated work has been studied by literary critics. However, they did not offer any approaches to quantify the author's and translator's contribution to the style of translation. In this paper, we propose a new method for the comparative analysis of two characteristics of translation quality, based on a comparative quantitative analysis of the unconscious style of translation texts. First, we quantify the contribution of translators to the unconscious author's style by comparing different translations of the same work. Secondly, we indirectly compare the contribution of the author of the work and the translator by analyzing translations of the same work by different authors. The proposed method is based on the approach developed by the authors [14] for the quantitative assessment of the writer's unconscious author's style. This method has been successfully applied to the attribution of literary works [14]. This article provides a description of the method and its discussion based on the analysis of the translation of English-language literary works into Russian, and, conversely, Russian classics into English. The choice is also due to the fact that the results obtained by the proposed method can be compared with the opinions of literary critics. In general, we can say that the results obtained by the proposed formal method are confirmed by the opinion of literary critics. Main Idea of the Method Modern information technology often uses archivers, or data compression methods. Archivers are based on so-called universal codes, which in turn are based on information theory, formal grammars, and some other concepts Archivers take text data as input and "compress" it, i.e., convert it into shorter text so that it can be decompressed into the original text. To compress it, archivers find unequal frequencies of characters and character combinations and use other text patterns. This paper describes a scheme for applying the information-theoretic approach. Let there be three texts T 1 , T 2 , and T 3 . We know that T 1 and T 2 were generated by different sources of information, respectively, S 1 , S 2 , and T 3 were generated by either S 1 or S 2 . We combine the texts into pairs T 1 T 3 and T 2 T 3 and compress both pairs. We also separately compress the files T 1 and T 2 , after which we calculate the differences in the lengths of the compressed files T 1 T 3 and T 1 and, similarly, T 2 T 3 and T 2 . If the difference between the lengths of the compressed files T 1 T 3 and T 1 is less than T 2 T 3 and T 2 , then we conclude that the text T 3 was generated by the information source S 1 . If the difference between the lengths of the compressed files T 2 T 3 and T 2 is less, then the pair T 2 and T 3 was generated by S 2 . This result occurs because the archiver, when compressing the appended part, that is T 3 , uses the statistical features found by it when compressing the first part, namely T 1 or T 2 . Therefore, T 3 is compressed better with text that generated by the same source of information. This approach was proposed by Tehan [16] and further developed in [17][18][19] . In [20,21], this idea was applied to construct a statistical method for classifying texts, which makes it possible to determine the reliability of the obtained conclusions. This scheme was quite successfully used by the authors to solve problems of text attribution in [14], where it was experimentally shown that each author has their own individual style, which is quite accurately manifested in their text of 4 kB (about two pages). Based on this fact, we apply the same scheme for solving the problems of analyzing the quality of translations as for attribution. Description of the Method on the Example of Establishing Unconscious Author's Style for English-Speaking Writers Let us move on to the description of the use of the scheme presented above: there are N different texts T 1 , T 2 , . . . , T N (for example, works of different writers). Each text T i is presented in the form of two samples, called training (X i , i = 1, . . . , N) and experimental, which, in turn, consists of m parts (slices), which we denote by Y ij , i = 1, . . . , N; j = 1, . . . , m. We compiled a sample of texts from the works of Poe, London, Lawrence, Kipling, Dickens, and Stevenson; N = 6, m = 16. From the works of these authors, we made 6 training samples X 1 , X 2 , . . . , X 6 of 64 kB each. Then, we made test samples-16 files Y 1j, j = 1, . . . , 16, 4 kB each, from Poe's works, Y 2 j, j = 1, . . . , 16, from London's works, . . . , and Y 6j , j = 1, . . . , 16, from Stevenson's writings. Then, we alternately "compressed" the file Y 11 with the training samples X 1 , X 2 , . . . , X 6 . Next, we determined with which of the samples the file is "better" compressed (i.e., we calculated d( 16. Let us explain the meaning of these numbers: 16 in the upper left corner means that out of 16 files Y 1j , j = 1, . . . , 16, all files "compressed" better with X 1 . In other words, all 16 "slices" from Poe's works "compressed" better with a training sample from his own work. Thus, it became clear that the author's style of Edgar Poe is uniquely determined by a slice of 4 kB in a training sample of 64 kB. The numbers in the second row of the table mean that out of 16 files Y 2j , j = 1, . . . , 16, 12 "compressed" better with X 2 . In other words, 12 "slices" from London's works are better "compressed" with his own training sample, however, two slices are more similar to Dickens' texts, one slice is more similar to Kipling's text and one slice is similar to Stevenson's text. There is a "recognizability" of the text of 12 slices out of 16. Then, for the Table 1 (and all tables below), we calculated the chi-square statistics and the Cramer coefficient V as follows: for a table see [22]. x 2 can be used to test the null hypothesis of homogeneity, since this value asymptotically obeys the chi-square distribution with (m − 1)(n − 1) degrees of freedom [22]. For example, the homogeneity hypothesis for Table 1 is rejected with a significance level of 0.0001. The Cramer coefficient V varies from 0 (corresponding to no association between the variables) to 1 (complete association). We call the whole process of transition from the source texts T 1 , T 2 , . . . , T N to a table (of size NxN) the construction of a contingency table, and the contingency table itself is denoted W(T 1 , T 2 , . . . , T N ) or W (depending on the context). As we have seen, in the cells of the contingency table, the numbers indicate the number of text slices. The authorship of each slice was attributed to one or another writer. If the method works "correctly", i.e., the method correctly determines the style of the author by slices, then the values in the table will be concentrated mainly on the main diagonal (in the ideal case of the method, the matrix will be diagonal). Otherwise, when the slices do not give an idea of the style of the author, the values in the tables will be evenly distributed. We quantified this effect by the Cramer coefficient V: with a diagonal placement, the Cramer coefficient is 1, and with a uniform distribution, V is close to 0. Selecting Method Parameters It is important to note that we determined all the parameters during the preliminary experiments in order to choose those that maximize the Cramer coefficient. First, various archivers were considered. Table 2 shows that LZMA gives the maximum value of the Cramer coefficient, so we used this archiver in the study. When identifying the authorship of literary texts, some researchers use the so-called text preprocessing. When solving our problem, we also used the text preprocessing method to find the maximum Cramer coefficient based on experiments. The experimental results are given in the following Table 3: It can be seen that the highest criterion value was achieved for texts from which only punctuation was removed. Therefore, this preprocessing was used in all other experiments. At the next stage, we experimentally determined the volumes of the training sample, one slice, and their number, focusing on the value of the Cramer coefficient, chi-square, and the total required amount of data. We explored slice sizes of 2 kB, 4 kB, and 8 kB with several training sample sizes. According to the results of the study, the size of the training sample was 64 Kb, with 16 slices of 4 Kb each. Interestingly, these values coincide with previously determined parameters for Russian-speaking writers [14]. Interpenetration of the Style of the Translator and the Style of the Author in Translation In this short section, we show experimentally that the style of translation of a literary work really depends on both the author's style and the translator's, but the translation style does not coincide with either one or the other. To solve this problem, we consider the works of K. Chukovsky and M. Engelhardt, who are both famous writers and famous translators. In the following table, we show the results of a study of their unconscious author's style as writers. (The homogeneity hypothesis for Table 1 is rejected with a significance level of 0.0001) We see in the table that the unconscious author's style is reliably determined from their own novels. Let us now consider the translations of Chukovsky and Engelhardt of M. Twain's novel The Adventures of Tom Sawyer. We preprocessed these translations (see Tables 3 and 4) in the same way as we processed their own works, shown earlier in Table 4. Comparing these tables, we see that the situations are completely different. Table 4 shows that the style of the author is determined almost unmistakably per slice of 4 kB (training sample 64 kB). In all cases, the author is correctly identified, and the Cramer coefficient is equal to 1. On the contrary, Table 5 indicates an extremely unreliable definition of the translator's style. The Cramer coefficient is significantly less than in the previous test. It is interesting to note that Table 5 also provides some additional information-the table shows that in Chukovsky's translations his own author's writing style manifests itself more than in Engelhardt's translations. Table 5. Results of translator style identification of Chukovsky and Engelhardt. The Cramer coefficient V = 0.43. K. Chukovsky (Translator) M. Engelhardt (Translator) The Adventures of Tom Sawyer Table 6. Table 6. Results of translator style identification. The Cramer coefficient V = 0.62. K. Chukovsky (Writer) M. Engelhardt (Writer) The Adventures of Tom Sawyer Table 4, we see that the results are radically different: Engelhardt's translations are more similar to Chukovsky's texts than to his own works. In other words, with the same parameters, the definition of the translation style, if possible, is with a very large number of errors. In our opinion, this fact indicates that the style of the writer differs significantly from the style of his own translations. The Degree of Preservation of the Author's Style by the Translator In this section, we consider the issue of preserving the styles of original authors when translating their works by a particular translator. We describe the proposed method step by step, illustrating with examples from different situations. 1. Step-Input data. We collect translations from one language into another of various texts made by one translator. 2. Step-Algorithm operation. Based on these translations, we make training and testing samples, build contingency tables W, and calculate the Cramer coefficient V. 3. Step-Interpretation. In general, the smaller the V, the weaker the differences in the style of translations. The content of W is interesting because the "own" style of some writers in translations can be revealed to be much weaker than other writers. For example, as we see below, the style of Gogol's works in English translations is much less preserved than, for instance, the style of Dostoevsky's works. 4. Step-Comparison of translators. Similarly, we can process the translations of several translators and compare them using the received W and V. Let us start with a study first of the translations of British and American classics into Russian made by the above-mentioned well-known Russian translators K. Chukovsky and M. Engelhardt and then the translations of Russian classics by famous English-speaking translators: Garnett, Piviar, Volokhonskaya, and Hogardt. Here and below, the volume of the training sample is 64 kB, and there are 16 slices, each 4 kB. (The homogeneity hypothesis for Table 7 (and following Tables 8-10) is rejected with a significance level of 0.0001). Table 9. Translations by C. Garnett. The Cramer coefficient V = 0.88. The table allows us to conclude that the translations of K. Chukovsky's works by O'Henry, Twain, and Wilde differ in style quite reliably, that is, the translator retains the unconscious author's style of the works. Tolstoy Dostoevsky Gogol Let us now consider the translations of the works of the Russian classics Dostoevsky, Turgenev, and Gogol by the famous translator from Russian into English, Charles James Hogarth. The contingency table is presented below. From the table, we see that the style of Dostoevsky's translations differs perfectly from the style of translated texts of other writers: 16 out of 16 of his slices are closest to the texts of his translations. Turgenev's style is transmitted somewhat worse-two slices from translations of his works are more similar to Gogol's works. In addition, Gogol's style is conveyed the worst of all-six slices from his works are more similar to other writers. On the whole, the author's style is worse preserved in Hogarth's translation than in Chukovsky's translations (Table 7). This fact is confirmed by the values of the Cramer coefficient-0.79 and 0.94, respectively. It should be noted that, apparently, not only translators preserve the style of the author in translation in different ways, but there are "difficult" authors whose style is difficult for all translators to maintain in translation. Among Russian writers of the 19th century, Gogol's authorial style was and remains the most difficult to translate, which is confirmed by the opinion of many literary critics [23]. Thus, the American critic K. Proffer [24] very briefly and peculiarly called Gogol's style "a nightmare for a translator", arguing that a feature of Gogol's style is a sense of humor and irony, a slight hint of sarcasm, which is difficult to translate into other languages. Gogol is a master of puns. Gogol's works are considered by all translators to be the most difficult to translate, and it is almost impossible to achieve an adequate translation. Difficulties in translating Gogol's texts are shown in the following contingency tables, which show data on the translations of C. Garnett and Pivear and Volokhonskaya. We can see from the above tables that, on average, the style of translations of works is preserved quite well (as evidenced by the high values of the Cramer coefficient), but in Gogol's translations into English, slices of texts from his works are often "closer" to translations of other authors. This fact indicates that his author's style is transmitted by the translator much worse (in other words, the author's style is worse preserved in translation). We found a similar effect in translations from English into Russian. In Table 11, we can see that Engelhart's translations convey Dickens' authorial style to a lesser extent than Doyle's style and Twain's style. Comparative Analysis of the Influence of the Translator's Style on the Translated Text As we noted earlier, each translator has their own unconscious author's style, based on individual personal vocabulary and syntax, sentence construction, the use of figures of speech, and idiomatic phrases. There is a widespread opinion among literary scholars, and indeed among a wide range of readers, that if the literary style of the translator is less recognizable or noticeable, then the quality of the translation is better (and, conversely, the translation is bad if the style of the translator "overshadows" the style of the author). In this section, we propose a method for quantitatively comparing the "contribution" of translators to the style of translation. In a sense, this allows you to compare the quality of translations of one work by different authors. As before, we first describe the proposed method step by step, and then we give examples that demonstrate different situations. Input data: Translations from one language to another of the same text made by different translators. First step: We process translations of one work made by different translators to obtain a contingency table W and calculate the Cramer coefficient V. Second step: We analyze and interpret the received data. At the same time, it should be noted that if V is less, then the differences in the style of translation of a given work are weaker and the styles of translators are less noticeable. The content of W is also of interest because the contribution of some translators to the value of V can be expressed much mroe weakly than others. Let us first consider an example of the translation of M. Twain's novel into Russian by Chukovsky and Yasinsky in Table 12. As we have already noted, if the contribution of translators to the style of translation was insignificant (ideally, the contribution of the translator was completely absent), then the values in all cells of the table would be 8. In this table, the situation is close to ideal, as evidenced by the values of the Cramer coefficient V. Therefore, the translator's contribution to the translated text is rather small. We discovered an interesting fact: the situation with translations of Russian writers into English is completely different. First, consider the calculation results shown in the following Tables 13-15. From the tables, we see that the precise identification of the translator's style in translations of Russian classics into English clearly contrasts with the situation with translations of English literature into Russian. This somewhat unexpected fact is confirmed by the opinion of many writers and literary critics who are fluent in both languages. Thus, the well-known writer, Nobel laureate I. Brodsky, said that the dominance of the translator's style in translations of Russian classics leads to the unification of the styles of all Russian writers [25]. Conclusions In this paper, we propose a method for quantitative assessment of the quality of translation, which solves two problems: (1) quantitative assessment of the degree of preservation of the author's style in translations and (2) quantitative assessment of the translator's contribution to the translated text. The method is "tuned" to the analysis of translations of classical English-language literature into Russian and, conversely, of Russian classic writers into English. Comparison of the obtained results with the opinion of literary critics, who have numerous works in these areas, allows us to conclude that the proposed method is adequate and effective. In our opinion, the proposed method, with appropriate parameter settings, can be applied to the analysis of "machine translation", the quality of film duplication, and in other areas where the task of evaluating the quality of translation arises.	5,287.8	2022-11-29T00:00:00.000	[ "Computer Science" ]
A STUDY ON NPAS OF SELECTED PRIVATE & PUBLIC SECTOR BANKS IN INDIA Banks play a very important role in any Financial System. It is the backbone of the Indian Financial System. The Rising NPA’s of the Banks in India for the last 5 years has really posed a threat to the Indian financial system. Recently Standard & Poor Global Rating agency has expected the NPA’s of Indian banks to remain elevated at 11.5 %. NPA helps to measure the Performance of any bank. It is quite evident that the Recent Covid Pandemic has badly hit not only India but the entire world to a greater extent. The Public and Private sector banks in India both have been adversely affected by the Rising NPA. Through this research, it has been observed that Public sector banks are more adversely affected than Private sector banks. My Study Focus on the Trend & Differences in the Non-Performing Assets of the Selected Indian Public and Private Sector Banks. INTRODUCTION The Indian Banking sector plays a very important role in the Indian Financial system. It acts as a link between lender and borrower. Banking in India has played a very important role to develop the saving habits among the masses. It helped to channelize the savings of the people into investment. In the past three decades Indian banking has outperformed and helped to develop the financial system in the country. Nevertheless Indian banking system has witnessed many revolutionary changes, for e.g. Nationalization of 14 major Private Banks in the year 1969. The Banking sector in India currently consist LITERATURE REVIEW Joseph and Prakash (2014) this paper deals with the comparative analysis of advances & Non-Performing Assets of Public & Private sector banks. 5 years data from the year 2008-2013 has been used for analyzing. It was observed that Public sector banks are having more NPAs than Private sector banks. Secondary Data was used for the Research Purpose. Chaudhary and Sharma (2011) the Paper focus on comparing NPA data of Public & Private sector banks. The trend of NPA also have been studied and it was observed that, Public sector banks are unable to compete with private sector banks and needs to improve its performance in Management Information system and also there is a larger need of Imparting training to employees of Public sector banks to make them competitive in comparison with private sector bank employees. Das and Dutta (2014) the study is done on comparing the NPAs of Public sector banks only. 26 public sector bank data on NPA was used. The comparison was done between SBI Associates and other Public Sector banks. Using the Anova test it was found that there is no significant difference in NPAs of SBI Associates and other Public sector banks. Kaur and Saddy (2011) the paper mainly focused on understanding the concept of NPA & factors contributing towards NPA. It also studies how NPA affect banking operations. Mittal and Suneja (2017) the paper mainly examine the level of NPAs in the Indian banking sector and then analyzing the causes for Increasing NPAs. The study also concludes that the magnitude of NPA in Public sector banks is more than Private sector banks. Miyan (2017) comparative analysis of selected Private and Public sector bank has been done on various performance parameters such as GNPA, ROA, and NNPA. 5 years data was use for analysis from 2011-2016. It was found that Performance of PSU banks is way behind the private sector banks. T-Test was used to find the significant difference. Kumar et al. (2021) the study examines the impact of NPA on Profitability of banks. Only 2 banks HDFC & SBI were used for comparison. It was found that Increase in the provisions for NPA declines the profitability of banks. It was also found that If NPAs are reduced then the Public sector banks could have a higher profitability. Singh (2013) the magnitude of NPA is comparatively higher in public sectors banks than private sector banks. To improve the efficiency and profitability of banks the NPA need to be reduced and controlled. Das and Dutta (2014) the study is done on comparing the NPAs of Public sector banks only. 26 public sector bank data on NPA was used. The comparison was done between SBI Associates and other Public Sector banks. Using the Anova test it was found that there is no significant difference in NPAs of SBI Associates and other Public sector banks. Kaur and Saddy (2011) the paper mainly focused on understanding the concept of NPA & factors contributing towards NPA. It also studies how NPA affect banking operations. Mittal and Suneja (2017) the The paper mainly examine the level of NPAs in the Indian banking sector and then analyzing the causes for Increasing NPAs. The study also concludes that the magnitude of NPA in Public sector banks is more than Private sector banks. Miyan (2017) comparative analysis of selected Private and Public sector bank has been done on various performance parameters such as GNPA, ROA, and NNPA. 5 years data was use for analysis from 2011-2016. It was found that Performance of PSU banks is way behind the private sector banks. T-Test was used to find the significant difference. Boddu (2019) this study compares the Loans and Advances, NPAs of both public and private sector banks in India to explore the preventive measures to control the rising NPAs. Suitable preventive measures help banks to decrease the level of NPAs in India. A lower level of NPAs helps the banks in consolidating their position, increasing confidence to depositors and increasing market share of the banks. Kumar et al. (2021) the study examines the impact of NPA on Profitability of banks. Only 2 banks HDFC & SBI were used for comparison. It was found that Increase in the provisions for NPA declines the profitability of banks. It was also found that If NPAs are reduced then the Public sector banks could have a higher profitability. Sahoo and Majhi (2020) the Paper analyze the recovery mechanism of NPAs with its three important wings i.e. recovery through Lok Adalat, Debt Recovery Tribunals (DRTs) and Securitization and Reconstruction of Financial Assets and Enforcement of Securities Interest Act (SARFASEI Act) and its impact on NPA.  To suggest few measures to improve the level of NPAs of banks. RESEARCH METHODOLOGY As per the literature review it has been observed that the present study mainly focus on NPAs comparison of Public sector banks and Private sector banks. My study also focus on comparison of NPA level of Private and Public sector banks. Only selected prominent banks from each Private & Public sector is taken for the study. Last 10 years data has been used for the meaningful study on NPAs. The secondary data has been used for analysis which is mainly taken from authentic sources such as RBI Publications etc. The data has been analyzed in tabular form, Trend analysis and ANOVA test has been used to find out significant difference.  Gross NPAs of HDFC Bank are comparatively Low than that of than ICICI & Axis bank, whereas ICICI NPA are highest in comparison of other two banks.  All these bank have the higher NPAs for the year 2019 & 2020 mainly due to corona pandemic.  All the Banks have seen decrease in NPAs in the year 2020 in comparison to the year 2019.  The financial position of HDFC in terms of Gross advances and NPAs is far better than other two banks.  All the banks whether Private or Public sector banks could see increasing trend in the level of NPAs since the year 2016 except HDFC Bank whose Level of NPAs is stable throughout the periods. T-Test Analysis A t-test is a type of inferential statistic used to determine if there is a significant difference between the means of two groups, which may be related in certain features. A t-test is used as a hypothesis testing tool, which allows testing of an assumption applicable to a population. Hypothesis Ho: There is no significant difference in the average values of the selected Private and Public sector banks. H1: There is significant difference in the average values of the selected Private and Public sector banks. As P value is less than 0.05 it can be interpreted that Ho will be rejected and H1 would be accepted as there is a significant difference between the Average values of NPAs of Selected Public & Private sector banks. Measures to Reduce NPAs of the Banks  Conducting Credible Credit appraisal of the company before giving loan.  Continuously assessing the financial position of the company after giving Loan to avoid further bankruptcy and take timely action  Selling off NPAs  Use of Sarfaesi act, 2002  Use of Insolvency Bankruptcy Code, 2016 Importance of Insolvency & Bankruptcy Code, 2016 in Recovery of NPAs of the Financial Institutions Insolvency & Bankruptcy Code, 2016 was introduced to resolve the claims which involved Insolvent companies. Although many other law were present in India to deal with the insolvency problem such as Sarfaesi Act, 2002, Debt Recovery Tribunals and Lok adalats. The average recovery rate after enactment of IBC.2016 has increased to around 45%. The average recovery time has also been decreased dramatically from 4 years earlier to this act. Insolvency & Bankruptcy code has remarkably help in speedy recovery of loans on the basis of which Indian ranking in "Ease of Doing Business" 2020 Report of the World Bank has improved to 63 rd Position from 142 nd Position prior to the Introduction of this act. FINDINGS  The study makes it clear that all the above banks both private and public sector banks are facing serious problem of NPAs.  Rising level of NPAs are impacting the Profitability and Liquidity of these banks.  Condition of Public sector banks is very poor due to Rising NPAs.  Performance of Private Banks is better than Public sector Banks in case of NPAs.  HDFC Bank is best performer when it comes to NPAs. The level of NPAs is stable throughout the years. SUGGESTIONS  Evaluate CIBIL score of the borrower before giving any loan.  Circulating Information of Defaulters in the Society.  Continuously assessing the financial position by the lender of the borrower for timely recovery or avoiding huge losses.  Using the speedy dispute settlement mechanism such as Insolvency & Bankruptcy Code, 2016 for the recovery of bad Loans.  Lending more to the Growing Sectors in the economy. CONCLUSION or less stable since last 10 years. Rising NPAs since last 5 years is a matter of concern for Indian banking system and also Indian financial system. It is equally important to take necessary steps by the banks and at government level to bring the NPAs down to restore the confidence in the banks. These banks should more focused on providing Quality Loan as "Prevention is always better than Cure" yes recent steps taken by government in respect of Insolvency code has got some relief to the banking sector to get fast recovery of loans but this mechanism of recovery of loans is post mortem and cannot enhance the economy, Ultimately giving Quality Loans is essential for the development of the economy and meeting its need. FUTURE SCOPE OF THE STUDY  The Research can be extended to other private and public sector banks.  Foreign banks can also be included in comparison of level of NPAs.  Many Variables other than NPAs can be used for meaningful comparison between banks such as Capital Adequacy Ratio, Liquidity Ratio, Profitability Ratio, etc.  Other statistical test could also be used to analyse the data.	2,725.8	2022-02-13T00:00:00.000	[ "Economics" ]
Th2NiC2: a low density of states superconductor The metallic carbides exhibit many novel prototypes of crystalline structure. Among these compounds Th2NiC2 was reported in 1991 as a new carbide which crystallizes in the U2IrC2 prototype structure. In this work we report a reinvestigation of the synthesis of this compound. We find that Th2NiC2 is a new superconductor. Our results suggest that this phase is stable only at high temperatures in the system Th–Ni–C. The substitution of Th by Sc stabilizes the phase and improves the superconducting properties. The highest superconducting critical temperature occurs at 11.2 K with nominal composition Th1.8Sc0.2NiC2. The electronic coefficient determined by specific heat measurements is close to zero. This unusual result can be explained by covalent bonding in the compound. Introduction One of the highlights of solid state chemistry in recent years has been the discovery of diverse types of transition metal carbides which exhibit many novel prototype crystalline structures [1][2][3]. An important characteristic of these special carbides is the presence of single and/or multiple metal-carbon bonds. In past years a variety of solid state ternary transition metal carbides that have been characterized are found to contain related types of metal-carbon bonds, as indicated by their interatomic distances. The other element, other than carbon or a transition metal, is generally a highly electropositive multivalent metal such as lanthanide (Ln), yttrium, uranium or thorium. The complete ionization of the electropositive metal to the stable ions Ln 3+ , Y 3+ or Th 4+ leads to a negatively charged transition metal subnetwork, which can be considered as organometallic net. There are many ternary compounds crystallizing in many different prototype structures but of particular interest in this paper is the U 2 IrC 2 prototype structure, also known as Na 2 HgO 2 . This prototype has a relatively simple body centered tetragonal crystal structure which was first determined for U 2 IrC 2 [4]. Moss et al [5] reported the discovery of two new compounds Th 2 NiC 2 and Th 3 Ni 5 C 5 . The Th 2 NiC 2 compound crystallizes in the tetragonal symmetry with space group I4/mmm and prototype structure U 2 IrC 2 , with lattice parameters a = 3.75Å and c = 12.35Å. The carbide Th 2 NiC 2 contains isolated linear C-Ni-C units in which the carbon atoms are only singly bonded to the nickel atom so that the C 4− ligand is isoelectronic with an F − ligand. The isolated linear [C-Ni-C] 8− anions with 26 valence electrons has Ni-C distance of about 1.93Å, which is suggestive of Ni-C single bonding [1,6]. In this compound each nickel atom has the 14 electrons required to fill seven orbital cylindrical spd 5 manifolds [1]. So assignment of the usual +4 and −4 oxidation states to the Th and isolated carbon atoms, respectively, in Th 2 NiC 2 leads to the formal oxidation state of zero for the nickel, which corresponds to a d 10 metal configuration. Uranium platinum metal carbides U 2 TC 2 (T = Ru, Os, Rh, Ir and Pt) also crystallize in the same prototype structure as Th 2 NiC 2 [7]. In this family of uranium platinum metal carbides the U 2 PtC 2 represents a superconducting [8] nearly-heavy-fermion system which has been investigated intensively [9][10][11][12]. The Th 2 NiC 2 and U 2 PtC 2 are electronically similar and in this work we present results which indicate that Th 2 NiC 2 is a superconducting material with a superconducting critical temperature close to 8.5 K. The partial substitution of Th by Sc decreases the lattice parameters and increases the superconducting critical temperature from 8.5 up to 11.2 K. We can observe three phases in equilibrium: the Th 2 NiC 2 with the U 2 IrC 2 prototype structure represented by Miller indices, α-ThC 2 with monoclinic symmetry and γ -ThC 2 with cubic symmetry. The inset shows a schematic unit cell where red spheres represent Th atoms, blue represents Ni atoms and black spheres represent carbon atoms. Experimental procedure The samples of nominal composition Th 2 NiC 2 and Th 2−x Sc x NiC 2 (x = 0.0, 0.1, 0.2 and 0.3) were made by arc-melting together the high purity elements, taken in the appropriate amounts, in a Zr gettered arc furnace on a water-cooled Cu hearth under high purity argon. The samples were remelted five times to ensure good homogeneity. Due to the low vapor pressure of these constituent elements at their melting temperatures, the weight losses during arc-melting were negligible (<0.5%). A microcomputer-controlled diffractometer equipped with a copper target for Cu Kα (λ = 1.540 56Å) radiation was used to obtain the powder x-ray diffraction patterns. The lattice parameters were determined using the PowderCell software [13]. Magnetic data was obtained using a commercial VSM-SQUID from Quantum Design. The temperature dependence of magnetization was obtained using zero-field-cooling (ZFC) and field-cooling (FC) conditions, under an applied magnetic field of 50 Oe. After both ZFC and FC measurements, M versus H data was acquired at 1.8 K. Electrical resistivity measurements were performed between 1.8 and 300 K using a conventional four-probe system. The samples were irregular in shape and gold wires were spot-welded to the samples and served as the voltage and current leads. These measurements were carried out with and without an applied magnetic field in order to estimate the upper critical field, using a PPMS apparatus (Quantum Design). The heat capacity (relaxation method) of a piece cut from one sample was determined in the 1.8-10 K range using a calorimetric probe inserted in the PPMS apparatus. Figure 1 shows the x-ray diffraction pattern of the as-cast Th 2 NiC 2 . The majority peaks can be indexed to the tetragonal structure belonging to the space group I4/mmm and prototype structure U 2 IrC 2 . The lattice parameters a = 3.748Å and c = 12.349Å are consistent with values reported in the literature [7]. The inset shows a schematic unit cell where the red spheres represent Th atoms, blue represents Ni atoms and black spheres represent carbon atoms. This pattern also shows two variations of ThC 2 which are monoclinic and cubic in equilibrium with Th 2 NiC 2 . The Th-C binary system has two basic phases, mono (ThC) and dicarbide (ThC 2 ) [14]. The dicarbide exists as three polymorphic modifications called α, β and γ . The α possesses the monoclinic symmetry stable at low temperature. The intermediate β-ThC 2 form has a tetragonal structure, while the high temperature γ -ThC 2 phase is cubic. However, of these, only γ -ThC 2 (cubic phase) is a superconductor with critical temperature close to 4.1 K [15]. The magnetization of our sample as a function of temperature shows a clear superconducting transition close to 8.5 K, as shown in figure 2. The inset of this figure shows type II superconducting behavior. We attribute this superconducting transition to the Th 2 NiC 2 compound, because the other carbides are not superconductors in this temperature range. We observed that this compound is very unstable in air; a few hours after preparation the sample decomposes and the superconducting behavior disappears. In order to stabilize this compound we tried one annealing at 900 • C for two days. After this annealing the superconductivity at 8.5 K disappears and a small fraction of Th 3 Ni 5 C 5 appears in equilibrium with the other binary carbides. The critical temperature observed in this case, via magnetization measurements, shows a small superconducting fraction with critical temperature close to 5.0 K. In fact, we recently reported bulk superconductivity in Th 3 Ni 5 C 5 with this same critical temperature [16]. These results suggest that Th 2 NiC 2 is a high temperature phase and is only metastable at room temperature. A way to stabilize this phase might be to 'create' chemical pressure in the structure, which is possible through substitution of Th by smaller atoms. One possibility would be by making Th 2−x Sc x NiC 2 , Sc being much smaller than Th [17][18][19]. We verified this possibility, making compositions with 0.1, 0.2 and 0.3 of Sc content in the nominal composition Th 2−x Sc x NiC 2 . For Th 1.9 Sc 0.1 NiC 2 composition, the x-ray results show Th 2 NiC 2 and α-ThC 2 with monoclinic symmetries in equilibrium. This result suggests that, at this level of substitution of Th by Sc, the phase of interest is stabilized. The magnetization versus temperature, shows that for this composition the critical temperature increases, from 8.5 K of the pure compound to 10.0 K (see figure 3). Furthermore, the compound is more stable in air than the pure compound (without Sc). This sample survives being exposed in air for about three months. Indeed the other compositions mentioned above also form in this structure. However, for nominal composition of scandium with x > 0.3 the superconductivity disappears and there arises a ferromagnetic behavior, where the secondary phase (unidentified) appears to be responsible for the magnetic behavior. The dependence of the superconducting critical temperature on Sc content is shown in figure 4, where we see dome-like behavior. The highest superconducting critical temperature occurs at a composition Th 1.8 Sc 0.2 NiC 2 . We chose this composition to complete the characterization. Figure 5 shown x-ray results for this composition (Th 1.8 Sc 0.2 NiC 2 ), where we observe that the sample appears more single phase than pure compound presented in figure 1. There is only a monoclinic dicarbide in equilibrium with Th 2 NiC 2 , as occurs for the other compositions for which the critical temperature behavior is shown in figure 4. This result strongly suggests that Sc stabilizes the tetragonal phase with the U 2 IrC 2 prototype structure. The inset of this figure shows the behavior of the lattice parameter with Sc doping. The systematic decrease of the lattice parameter is consistent with the fact that Sc is smaller than Th. The critical temperature is about 11.2 K, as shown in figure 6. The difference in results between ZFC and FC conditions indicates a type II superconductor and points to a weak Meissner effect due to strong flux pinning. The , where the λ L parameter represents the penetration depth which yields λ L ∼ 138 nm at 1.8 K. Results and discussion The resistance as a function of temperature showed, unambiguously, the sharp superconducting transition close to 11.2 K (figure 7), which is consistent with the result of figure 6. The inset displays the shifting of the critical temperature with applied magnetic field. We show resistance instead of resistivity because the shape of the sample is completely irregular, making the determination of the This estimate can be made using the WHH formula [20] in the limit of short electronic mean-free path (dirty limit): Figure 8 shows the curve estimated by WHH which follows the data points very closely and gives a µ 0 H c2 (0) value of 11.6 T. On the other hand, the temperature dependence of the upper critical field can either be explained by Pauli paramagnetism with extremely strong spin-orbit scattering or with a completely dominating orbital field effect. The Pauli limiting field is given by . This κ is consistent with the behavior of a type II superconductor, as seen in the M versus H curve (inset of figure 6). The specific heat measurements show a rather small jump close to 11.0 K (figure 9). The inset shows the C/T against T 2 where we can observe the normal state specific heat which can be fitted to the expression C n = γ T + βT 3 by a least-squares analysis. This fitting yields the value γ ∼ = 0.02 (mJ mol −1 K −2 ) and β ∼ = 0.271 (mJ mol −1 K −4 ). This β value corresponds to a Debye temperature of D ∼ 330 K. The γ value represents the Sommerfeld coefficient per mole formula unit and is proportional to the density of states at the Fermi level. This value is surprisingly suggesting almost no electrons at the Fermi level. This small value of the Sommerfeld coefficient is consistent with the small jump observed at the superconducting critical temperature. If we consider a weak coupled BCS superconductor, the jump in specific heat at T c is C/γ T c = 1.43, which implies C ∼ 0.32 mJ mol −1 K −1 . This value is close to the noise level of a commercial Quantum Design Physical Properties Measurement System. For phonon-mediated superconductivity, superconducting critical temperature (T c ) is given approximately by T c = (θ D /1.45) exp[−1.04(1 + λ)/λ]. Taking the Debye temperature estimated from figure 9 (330 K), λ is approximately 0.5, which is in the weak electron-phonon coupling regime. However, the strong covalent bond between Ni and C could give a large value of ω 2 , which represents the characteristic phonon frequency averaged over the phonon spectrum. This important parameter could play an important role in the mechanism of the superconductivity in this material. Indeed other materials also present low density of the states at the Fermi level where the jump in specific heat is very hard to see. As an example, we can mention diamond doped with boron which exhibits a superconducting critical temperature close to 4.0 K. In this material the authors were not able to observe a jump in specific heat due to a pretty low Sommerfeld coefficient [21]. Another example is superconductivity in the perovskite phase BaPb 1−x Bi x O 3 , which can exhibit superconductivity at 11.1 K [22]. Although the authors were not able to explain why they did not observe a jump in the specific heat measurement, the Sommerfeld coefficient is too low as also occurs in our sample. But it is surprising that this compound has very low density of states at the Fermi level when compared with U 2 PtC 2 , which crystallizes in the same prototype structure and displays a γ value of approximately 75 mJ mol −1 K −2 , almost a heavy-fermion material. In our results on the superconductivity in Th 3 Ni 5 C 5 we found an electronic coefficient (Sommerfeld constant) of about 38.84 mJ mol −1 K −2 . In our view these results are very unusual, but although the jump seen in our specific heat measurement is small, we can say that this compound is a bulk superconductor. Conclusion In this work we have shown that Th 2 NiC 2 is a superconductor with superconducting critical temperature close to 8.5 K. Our results also suggest that this phase is stable at high temperature and is extremely sensitive in air. The substitution of Th by Sc stabilizes the structure and decreases the instability in air. The substitution of Th by Sc also improves the superconducting properties and the superconducting critical temperature is optimized for the nominal composition Th 1.8 Sc 0.2 NiC 2 . The low value of the Sommerfeld coefficient is a surprising and unusual result since generally compounds with Th and Ni present values of the electronic coefficient which are relatively high. Our results shown unambiguously that Th 1.8 Sc 0.2 NiC 2 is a new bulk superconductor material.	3,565.8	2012-04-01T00:00:00.000	[ "Physics" ]
Nuclear data correlation between different isotopes via integral information . This paper presents a Bayesian approach based on integral experiments to create correlations between different isotopes which do not appear with differential data. A simple Bayesian set of equations is presented with random nuclear data, similarly to the usual methods applied with differential data. As a consequence, updated nuclear data (cross sections, n , ﬁ ssion neutron spectra and covariance matrices) are obtained, leading to better integral results. An example for 235 U and 238 U is proposed taking into account the Bigten criticality benchmark. Introduction It was recently demonstrated that an uncertainty decrease and non-zero correlation terms between different nuclear data reactions can be obtained when using integral information such as criticality benchmarks [1] (see Refs. [2][3][4] for other examples). In reference [1], cross-correlation terms between n (emitted neutrons per fission), x (fission neutron spectra) and s (n,f) (fission cross section) were calculated in the case of the 239 PU isotope with specific Pu benchmarks in the fast neutron range. Such approach can be useful to lower calculated uncertainties on integral quantities based on nuclear data covariance matrices, without artificially decreasing cross section uncertainties below reasonable and unjustified values. This is appropriate when the propagation of uncertainties from differential data to large-scale systems indicates an apparent discrepancies between uncertainties on measured integral data (neutron multiplication factor, boron concentration, isotopic contents) and the calculated ones. In this reference, the correlation terms between reactions for a specific isotope and the decrease of differential uncertainties were calculated using a simple Bayesian Monte Carlo method. In the present work, the same method is applied (1) to obtain correlation terms this time between different isotopes, and (2) to decrease the uncertainties for important reactions, using again criticality-safety benchmarks. The approach and the equations used in the present work are the same as in [1]. In the following, the case of the 235 U and 238 U isotopes will be considered and the Bayesian update will be performed using a specific criticality benchmark with high sensitivity to these isotopes: the intermediate metal fast number 7 benchmark, or imf7 (also known as Bigten) [5]. First the method will be recalled in simple terms, then the application with the imf7 benchmark will be presented. The updated benchmark value, cross sections, correlations and uncertainties will be compared to the prior values, thus demonstrating the results for the differential quantities. This is of interest in the context of nuclear data evaluations, where both nominal values and covariance matrices can reflect the present results. Correlation from integral benchmarks The basic principles of the method were already presented in [1]. We will outline here the major equations. The Bayesian updates of the prior information is obtained using a Monte Carlo process: random nuclear data are produced following specific probability density functions (pdf). Such pdf were obtained as follows: starting from uniform distributions, comparisons between calculations and differential measurements (from EXFOR) were performed. Following the description of reference [6] (and as presented below for integral data), weights are derived from such comparisons and pdf of TALYS model parameters are updated. The next step is to sample from these specific parameter pdf to produce random nuclear data; each random nuclear data is used in the benchmark simulation; the random calculated quantities are compared to the measured one, and; -finally each random nuclear data is weighted according to the agreement between the calculated and measured quantities (see below for details on the definition of such weights). In the present work, the k eff value of the imf7 benchmark is used as the only integral quantity: the reported value in [5] is k exp = 1.00450 with an experimental uncertainty of Dk = 70 pcm. As a prior for the nuclear data, the random 235 U and 238 U cross sections (and emitted particles and spectra) are obtained from the TENDL-2014 library [7]. The T6 system [8] was used to generate so-called random ENDF-6 and ACE files, containing all necessary random nuclear data. This way, the same file production and processing is followed, based on TALYS and NJOY [8,9]. In the case of the imf7 benchmark, the k eff value is very sensitive to the unresolved resonance range [10] and the ENDF-6 files are processed with the PURR module of NJOY. Each ENDF-6 and ACE files are similar in format, but different in content. They are based on sampling of model parameters of the different nuclear models according to specific independent probability distributions (see the TMC, BMC, UMC-B and BFMC methods [6], [11][12][13] for details). Model parameters are sampled a large number of times (with the index i = 1 … n) to generate full cross sections and other nuclear data quantities for 235 U and 238 U from 0 to 20 MeV (see for instance [14] for the testing of such file distributions). The sampling between these two isotopes is performed in independent manner, so that no correlation between 235 U and 238 U can exist other than from the model themselves. The prior correlation matrices for 235 U and 238 U are simply obtained from the n random files, using the conventional covariance and standard deviation formula. The n random ACE files are then used in n MCNP6 simulations [15], leading to n values of calculated k eff,i with i varying from 1 to n. The comparison between n random calculated k eff,i=1...n and the experimental value k exp is performed with the simplified chi-2 Q i values and associated weights w i (here, chi-2 is called Q i to differentiate it from the neutron spectra x): Such formulation can easily be linked to the usual Bayesian likelihood [13,16]. The weights are then assigned to the corresponding 235 U and 238 U nuclear data files (for both isotopes together) which lead to k eff,i . Considering n random files for each isotopes, there is n 2 possible combinations; in the following, we will consider only n combinations such as (1,1), (2,2),…(i,i). Examples for the weights of the random 235 U and 238 U nuclear data are presented in Figure 1. In this example, one iteration i corresponds to the use of one specific random file for 235 U and another one for 238 U. As observed, the distribution of the weights w i strongly varies from values close to 1 (for Q i ≈ 0, indicating a good performance of the random files i) to very small values (almost 0 for large discrepancies between k exp and k eff,i ). Due to this large range of weights, a large number of random files is necessary to obtain meaningful results. In the case of 7000 random files for each U isotope, about 18% of the weights are higher than 0.01. The final quantity for a specific benchmark consists of a matrix containing [i, s i ( 235 U), s i ( 238 U), w i ] for i = 1 … n, where s i stands for all nuclear data quantities as a function of energy. As previously mentioned, the value of n = 7000 is considered in this work. The correlation r(s a , s b ) can be calculated for specific values of the incident neutron energies for s a (E k ) and s b (E p ). For instance, s a is the fission cross section of 235 U and s b is the capture cross section of 238 U, both at a specific energy E k and E p , respectively. Considering the vector [i, s i ( 235 U), s i ( 238 U), w i ], r can be calculated as follows. Using the definition of weighted averages: the correlation r(s a , s b ) between s a and s b is given by Such correlation r can be obtained for different E K and E p , thus defining a full correlation matrix between the same cross section and the same isotope, between different cross sections for the same isotopes, and between isotopes. As quantities in these equations (average cross sections, standard deviations and correlation factors) come from a Monte Carlo process, one has to check their convergence as a function of the iteration number, as presented in Figure 2. One can see that in both cases (considering or not weights w i ), the final correlation values are different, and the difference is outside the standard errors (defined as ffiffiffiffiffiffiffiffi q for the non weighted case). As it can be seen on this figure, the non weighted running correlation evolves smoothly with the increasing number of samples, while the weighted running correlation exhibits large jumps for low iteration i where high weight samples are added to the calculation (as seen in [16] showing same kind of behavior). In the following, more details will be given on the imf7 benchmark together with the results regarding the prior and posterior information for the uranium isotopes. Application to 235 U and 238 U The work presented in [1] was limited to the single 239 Pu isotope, since it was applied to integral experiments from the PMF subtype (Plutonium Metal Fast) of the ICSBEP collection [5], for which only 239 Pu nuclear data dominate the benchmark calculation result. Following the same idea, the imf7 benchmark is selected as its k eff is highly impacted by both 235 U and 238 U. The imf7 benchmark The imf7 benchmark (intermediate enrichment uranium metallic fast number 7), also known as Bigten, is a highly enriched uranium core, surrounded by a massive natural uranium reflector. It is characterized as a fast system, as the majority of the neutron spectrum is above 100 keV. Bigten is a cylindrical assembly with a core composed entirely of fissionable material in metal form. There are three distinct regions: a nearly homogeneous cylindrical central core made of uranium enriched at 10% in 235 U, surrounded by a heterogeneous core volume made of natural uranium and highly enriched uranium (93%) and a cylindrical reflector, made of depleted uranium, completely surrounding the core. Figure 3 shows the neutron spectrum averaged over imf7, calculated using MCNP6 with TENDL-14 nuclear data, and average energies for fission and capture are presented in Table 1. It has a typical fast spectrum with an average neutron energy of 530 keV. This imf7 configuration has long been known by evaluators to be sensitive to nuclear data for both 235 U and 238 U isotopes. This double dependency is so strong that mixing nuclear data for 235 U from one source (e.g. ENDF/B-VII.1 [17]) with data for 238 U from another source (e.g. JEFF-3.3) in a imf7 benchmark calculation, results in a poor restitution of the measured k eff value. Some examples are presented in Table 2 by repeating the benchmark calculation with different nuclear data evaluations for 235 U and 238 U. As observed, if both uranium isotopes come from the same library, the calculated k eff is close to the experimental value. On the other hand, a mixture of the library of origin leads to very different calculated k eff. These cases can be interpreted as the effective presence of correlated isotopes in current evaluated nuclear data libraries. Correlations By extending the methodology described in reference [1], such cross-isotopes correlations can be rigorously quantified. All combinations of neutron incident energy, observables (cross sections, prompt fission neutron spectra, nubar, etc.), and target isotopes are possible, as illustrated in Figure 4. Correlation matrices for a selection of cross sections, nubar and pfns in the case of 235 U and 238 U. Top: correlation without taking into account the imf7 benchmark; bottom: same, but taking into account imf7. See text for details. In each sub-block, the cross sections are presented as a function of the incident neutron energy (the lower-left part corresponds to the lower neutron energy range, whereas the higher-right part corresponds to the higher neutron energy). The upper panel of Figure 4 shows the full 235 U-238 U correlation matrix for the prior (unweighted), Total Monte-Carlo (TMC) [11] samples for 235 U and 238 U, as computed from the TENDL-2014 library. Four blocks are separated by two red lines, each block represents the correlation and cross-correlation for these isotopes: bottom-left: 235 U-235 U, bottom-right: 235 U-238 U, top-left: 238 U-235 U and top-right: 238 U-238 U. As it can be seen, cross-isotopes correlations between isotopes are zero, since model parameters for both isotopes were independently sampled in this study. The lower panel shows the full 235 U-238 U correlation matrix for the TMC samples of 235 U and 238 U, weighted according to equation (2), where k exp is the experimental value of the imf7 benchmark, and k eff,i that derived from the 235 U and 238 U sampled files, indexed by i. Obviously, that lower panel exhibits cross-isotopes correlations contrary to the upper one, and it also exhibits correlations between different types of observables like those discussed in [1]. Although the TMC treatment allows the constructions of covariance matrices between all the nuclear data observables, the matrices shown in Figure 4 sub-matrices display some stronger correlations, mostly along the diagonal, but also for observables derived from the optical model potential (total, non elastic and elastic cross sections), highlighting the role played by that model in inducing correlations in nuclear data. As expected, similarly to the conclusions of references [1,16], a weak negative correlation for the posterior is observed (see Fig. 5 for an enlarged sub-matrix) between the n of 235 U and its fission cross section, for energies close to the mean energy of neutrons causing fission in 235 U (Tab. 1). This anti-correlation results from n and s (n,f) being two factors in the product describing the neutron source term in the neutronic transport equation: a stronger s (n,f) is exactly compensated by a weaker n. The correlation matrix between the 235 U capture and fission cross sections (Fig. 6) is harder to interpret, since it exhibits a complex structure. Although the crosses materializing the mean energies leading to fission and capture reactions in the core and blanket regions of the assembly both sit in the weak correlation region of the map (close to the negligible correlations zone (white), there are regions of stronger correlation, both positive and negative, nearby. The moderate positive correlation for neutron energies seen above 500 keV can be understood as 235 U(n,f) driving the source term of the neutronic transport equation and 235 U (n,g) being a contributor to the absorption term of that equation. For lower neutron energies, two zones of moderate negative correlation are observed, one for low (E < 200keV) neutron energy inducing fission, and one for low neutron energy inducing capture. That complex structure of the 235 U capture and fission correlation might result from the interplay between 235 U in the core region (fast spectrum) and the blanket region (slower neutronic spectrum). From Figure 4, one can also note two important aspects: anti-correlation for 235 U between x and (n,g): in order to compensate for a higher neutron capture, the fission spectrum becomes harder, thus producing more neutrons at higher energy; especially in the case of 238 U, anti-correlation appears in the updated matrices between the inelastic cross sections themselves. Again, this can be understood in order to compensate for the loss of neutrons caused from a specific inelastic cross section (for instance (n,inl)) by another one (for instance (n,inl 2 )). In the off-diagonal cross-isotope correlation blocks, a prevalent weak positive correlations can be observed between 235 U(n,f) and 238 U(n,g) at energies where the neutronic spectrum is strong (see Fig. 7 for an enlarged submatrix). Again, that positive correlation is explained by 235 U(n,f) driving the source term and 238 U(n,g) being the other strong contributor to the absorption term of the neutronic transport equation. A very prevalent weak anti-correlation can also be observed between the fission cross section of 235 U and the total elastic cross section of 238 U (presented in an enlarged format in Fig. 8). They are anti-correlated since a weaker fission cross section of 235 U can be compensated by a more efficient neutron reflector ( 238 U(n,el)), which reflects leaking neutrons back into the 235 U core for another attempt to fission 235 U. Updated cross sections and variances The weighting of TMC samples according to equations (1) and (2) not only introduces correlations between observables, but it also leads to modifications of the central values The general observation is that the cross sections (including n and x) are moderately updated (maximum of 1.0% for the 235 U(n,inl) cross section) whereas the variances are strongly reduced (see for instance 235 U(n,f)). The changes in the posterior cross sections are to some extent depending on the prior uncertainties. If the prior uncertainties are small, the changes will also be small. Therefore the changes presented in Figure 10 can be different for different prior. In the case of 235 U, that reduction brings the variance in the same order of magnitude as that of the existing experimental differential data. However, for 238 U, the reduced standard deviation is still larger than that of existing differential data: a further Bayesian update with that differential data would further reduce the calculated uncertainty of the 238 U n (see for instance [18,19] for details). A limited set of cross section uncertainties is strongly affected by the Bayesian update: with a decrease for 235 U(n, f), 235 U(n,inl), 235 U(n,g), 238 U(n,inl), 238 U(n,inl) and 238 U (n,el) and an increase for 238 U(n,inl). One should notice that the (n,inl) cross section for 238 U is relatively small, with a maximum at 400 mb, compared to the (n,inl) cross section (with a maximum of 1.5 b). Such change could be explained by statistical fluctuations, but a dedicated study on this effect would be necessary to clarify its origin. The increase of this cross section uncertainty has therefore a limited impact. It is difficult to assess the relative importance of these cross sections in the decrease of the k eff uncertainty, but the mentioned reactions are important for the account of neutrons in the energy region of interest. The value of the 235 U posterior fission cross section is modified by a factor as large as 1.003 relatively to that of the prior, and its standard deviation is strongly reduced. When compared with the international cross section standard [19] for the 235 U fission cross section (see As a final remark, since the Bayesian weighting of samples applies to sets of complete ENDF-6 formatted files (one set including an ENDF-6 file for 235 U and a file for 238 U), that weighting process produces adjustments and variance reduction for all the observables included in these files, from the inelastic and elastic cross sections, which do play role in the calculation of imf7, to cross sections like (n,p) or (n,a), which are hardly constrained by the benchmark. Resulting k eff distributions The final result of the Bayesian weighting process, driven by the experimental k exp of the imf7 benchmark, is the simulated k eff distribution, calculated by MCNP6, using the weighted correlated 235 U-238 U samples, and how it compares to the one calculated with the initial unweighted samples from TENDL-2014. Table 2 shows the averages and standard deviations of the calculated k eff distributions, compared with the experimental value, with unweighted sampled labeled as "prior", and weighted samples labeled as "posterior". Those distributions of k eff are also displayed on Figure 12. In Table 3 and Figure 12, the posterior distribution can be observed to agree very well with the experimental result and its uncertainties, while the average k eff resulting from the unweighted prior is lower, with a much wider distribution. Discussions As mentioned in the introduction, the goal of this type of work is to reduce the calculated uncertainties on integral quantities while keeping realistic uncertainties and correlations for the differential data. Additionally, as showen in Table 3 for imf7, the updated 238 U and 235 U nuclear data provide k eff which is in better agreement with the experimental value. Such method can be extended by including more benchmarks in the definition of Q i (and also by including other quantities such as spectra indexes), but prior to the continuation, two tests can be performed. The first one is partially presented in Figures 9 and 10, showing that the updated nuclear data are still in agreement with the differential data (i.e. pointwise cross sections, or pointwise n). This is not explicitly shown in these figures, but the fact that the updated cross sections are very close to the prior values indicates that the method does not produce very different cross sections compared to the prior. And as it was mentioned, the agreement with the standard cross section is still respected, given the large variances of the TENDL curves. The second test concerns the predictive power of the method: by choosing a benchmark with similar characteristics than imf7, is its calculated k eff improved? If this is the case, one can consider that the indications provided by the updated cross sections are general enough to be exported to outside the case of imf7. To answer this question, three additional benchmarks are calculated with the same random 238 U and 235 U nuclear data files: using or not the weights from imf7. Two of these benchmarks are relatively close to imf7: 4, 7 (2018) hmf1 (or Godiva being a metallic sphere of 235 U) and imf1-1 (or Jemima, being metallic cylindrical arrangement of 235 U). A third benchmark is on purpose chosen to be very different than imf7: it is a thermal system of low-enriches UO fuel rods with a high water-to-fuel ratio: lct6-1. For this benchmark, the modifications of the 238 U and 235 U nuclear data in the fast neutron range from imf7 are expected to have little impacts on the calculated k eff . The results of these calculations are presented in Table 3. First, the k eff values for hmf1, imf1-1 and lct6-1 calculated with weights from imf7 (posterior in Tab. 3), are not in worse agreement with experiment than the ones calculated without weights (prior in Tab. 2). This suggests that weighting random samples according to one given benchmark does not produce a distribution that is only good for that benchmark. Moreover, introducing the imf7derived weights seems to slightly improve the agreement of all three of our test cases with experimental values, suggesting that the changes due to that weighting carry some real physics and are not just a better local optimization. However, while the weighted imf1-1 and lct6-1 calculation results are within experimental uncertainties, that of hmf1 is still well outside of experimental uncertainties, suggesting that the imf7 specific weighting is missing some of the physics that is essential for the hmf1 case. Now, looking at the calculated uncertainties for the weighted hmf1 and imf1-1 cases, we observe that their widths are reduced compared to those of the unweighted calculations, suggesting again that imf7-derived weights carry some real physical information. However, the widths resulting from weighted calculations are much larger than experimental uncertainties. In the case of the lct6-1 benchmark, the uncertainties are not reduced: the changes generated at high energy do not impact the uncertainties for this thermal system. This indicates that in the case of a general evaluation of nuclear data, one needs to include benchmarks spanning over a wide energy range. In order to confirm the conclusions from the above test, it should be repeated on a more extensive set of benchmark cases. The next step in this process would then be to calculate weights from all those benchmark cases, to combine them (maybe through a simple product), and test whether the resulting weighted distribution provides a good restitution of all the experimental benchmark data used to determine those weights (see for instance the work performed in [3,20]). There is also no reason to restrict the benchmark data used to calculate weights to only k eff , and other types of data, like spectral indices or differential measurements, are likely to carry information that constraints nuclear data in a different manner. Conclusion It has been shown that including integral constraints from experiments that are sensitive to two isotopes introduces effective cross-correlations between the nuclear data of these isotopes. It was demonstrated that it is possible to quantify such cross-correlation between isotopes using an integral benchmark, based on a Bayesian method and a set of random nuclear data. The case under study concerns the 235 U and 238 U isotopes and the Bigten (imf7) benchmark. Additionally, the updated nuclear data and their covariance matrices lead to a better agreement with the calculated and measured integral data, for the central values and for uncertainties, while keeping the original good agreement with differential data. This is an extension of the method previously proposed for 239 Pu [1] and is a confirmation that such method allows (1) to be part of the evaluation process of nuclear data, and (2) to obtain reasonable integral and differential uncertainties. In the future, the method will be applied taking into account a larger set of integral data and exploring applications below the fast neutron range. Our limited testing is suggesting that weighting with respect to one benchmark experiment does not negatively affect the agreements with other experiments and even improves them slightly. A more extensive testing is needed to confirm that combining weights calculated from different benchmark experiments leads to a weighted sampling that simultaneously accounts for all those benchmarks and their associated uncertainties. Such a combination of weights originating from different benchmarks will be the subject of a forthcoming article. Like in [1], the present work is at the "proof of concept" stage: the methodology seems to work with a reduced set of integral constraints and the rather simple models used to produce the TENDL-2014 library. In order to produce evaluations of the quality of the best evaluated nuclear data libraries, that method will have to be extended to: include a larger and more representative set of integral experimental constraints, spanning a wide range of neutronic spectra and applications; include integral constraints other than k eff in the calculation of weights, include differential constraints as well as international cross sections standards [19] in the calculations of weights; apply that methodology to the more sophisticated models [21][22][23] used to evaluate the nuclear data of the best international data libraries; completely implementing the above extensions would produce fully updated nuclear data and covariance matrices, including cross-isotopes and cross-observables correlations, following a well defined reproducible scheme. These files should allow for accurate simulation of application, including calculated uncertainties. Such work would then be part of the elaboration of a nuclear data library based on models (for differential data), realistic model parameter distributions and integral constraints, as presented in [24].	6,267.8	2018-05-01T00:00:00.000	[ "Physics" ]
Geographic Range and Nest Architecture of Cephalotrigona Capitata Smith, 1854 (Apidae: Meliponini) in the State of Bahia, Northeastern Brazil Abstract The bees of the genus Cephalotrigona (locally known as “mombucas”) play a key role in natural environments but their bioecological features, required to design proper management and conservation strategies, are scarce in most species. Thus, the goal of the present study was to map the occurrence sites of C. capitata in the state of Bahia, north-eastern Brazil, and to provide useful information about nest architecture to their technical management. This species was recorded in fifteen municipalities in Bahia, totaling forty-one nests. The range of C. capitata varied from locations at sea level to seasonal ombrophilous forests at an altitude of 600 m high. The nests were built in trees with a mean diameter of 19.8 ±3.0 cm. The nest architecture was similar to that reported in other stingless bee species, with a variation in analyzed parameters. The thermoregulation was more efficient in highly populated boxes. The present results can be used for the conservation and management of this species, which represents a potential source of income for local farmers. INTRODUCTION The Cephalotrigona stingless bees (Apidae: Meliponina), popularly known as "mombucas" in Brazil, are recorded widespread throughout Central and South America, in Mexico, Nicaragua, Costa Rica, Panama and Colombia. They are typically found in such areas of dense vegetation as Atlantic Rainforest, Seasonal Forests and Amazon Rainforest (Roubik, 1983;Quezada-Euán, 2005). In Brazil, two species C. capitata Smith, 1874 and C. femorata Smith, 1854 have been reported (Schwarz, 1948;Silveira, Melo, & Almeida, 2002). The range of both species include the state of Amapá, Pará (northern region), Ceará (northeastern region), Mato Grosso (Central Brazil), Espírito Santo, Minas Gerais, São Paulo (southeastern region), Paraná, and Santa Catarina (Southern region) (Silveira, Melo, & Almeida, 2002;Camargo & Pedro, 2017). The occurrence of these species in Bahia has not been recorded so far. The first report about C. capitata (cited as Trigona capitata) in Brazil was carried out by Ihering (1903) who described their nesting sites and nest traits. Bionomic data about Cephalotrigona reported by Zwaal (1992) was based on a species found in Mexico. Authors Sakagami, Beig, & Kyan (1964) analyzed the oviposition process in C. Capitata femorata. Even though reports about the occurrence of Cephalotrigona spp. in Brazil are available, little is known about the ar- Geographic range of Cephalotrigona capitata chitecture of their nests. Therefore, the goal of the present study was to report the occurrence and determine the range of C. capitata in the state of Bahia, northeastern Brazil and provide information about the nest architecture to be used in the sustainable management and conservation of this species. MATERIAL AND METHODS The records of C. capitata were based on interviews, location and rescue of nests, as well as collection of specimens on flowers with the use entomological nets. The analyzed nests were found in trees and rustic boxes, and the coordinates of their locations were obtained through GPS (Global Position System) and SEI (2015) dataset. Dr. J.M.F. Camargo (USP-FFCLRP-1992) as C. capitata Smith, 1874, identified the collected bee samples from Canavieiras-BA while Dr. Favizia Freitas de Oliveira (UFBA) identified the specimens from Mundo Novo-BA. The vegetal species herein cited was sampled and sent to identification in the IBGE-RADAM herbarium. Nest architecture was evaluated based on nesting substrate (diameter and size of trunk cavities and wood thickness), diameter of honey combs, number of food pots and population estimates. The dimensions and internal components of nests were measured with a tape measure , honey amount in pots was estimated with graduate syringes and pollen amounts in pots were estimated with a portable digital scale (Alves, Carvalho, & Souza, 2003;Barbosa et al., 2013). The mean and standard deviation values were calculated for each parameter using the BioEstat v. 5.0 software. Distribution range Cephalotrigona capitata was sampled in fifteen municipalities from the state of Bahia (Tab. 1). The estimated range comprises 3.59% of Bahia territory (Fig. 1). Forty-one nests, the largest abundance, were reported in Mundo Novo, Gandu, Igrapiúna, and Wenceslau Guimarães. Eight of those from trees plus two from rustic boxes were rescued and evaluated in the municipalities of Amargosa, Canavieiras, Gandu, Mundo Novo, Taperoá, and Wenceslau Guimarães (Tab. 1; Fig. 1). The altitude in which the species was sampled ranged from the sea level (Canavieiras) to 600 m high (Mundo Novo) in seasonal forest domains. Nesting Nests of C. capitata were found in wide hollows of living trees with a large diameter and thick wood (Tab. 2). Only a single colony was observed in openings previously used by Melipona mondury in a dead tree of the species Tapirira guianense Aubl. Their nests were found in the following tree species characterized by great heights, large trunk diameters and thick wood: Mimosa sp. in Cravolândia, Parkia pendula and Sclerolobium sp. in Taperoá and Igrapiúna, Tapirira guianense in Gandu and Wenceslau Guimarães, and Jacaranda caroba in Igrapiúna and Camamu. Nest entrance The entrance of the evaluated nests were hardly visible, with elongated to dished external structures depending on the colony. Nonetheless, most nests did not present obvious entrances, composed of solid dark resin (Fig. 2). The entrance tube was short and narrow, built from dark cerumen, opening into the nest periphery. In this study, many bees were observed around the tube. Internally, the tube would end in a gallery 21 cm in length and 18 cm in diameter formed by a highly compact and hard material. The entrances of colonies were located between 0.5 to 5.0 m above the ground, with a mean value of 2.83± 0.25 m, thus showing the preference of large trees to their nesting. The orifice diameter presented a mean value of 0.88 ± 0.05 cm, usually guarded by a single bee, always located on the internal part. Defenses The bees usually defend themselves when disturbed, even in highly populated colonies with large amounts of honey. Some individuals of C. captata were observed on animal feces in Cocos-BA, which usually occurs when there is resources scarcity Caption: rescued nests, located nests, **records of species presence, # nests in rustic boxes. Involucre The involucre in most analyzed nests contained two to three thin layers of dark cerumen covering the majority of brood combs, even in trees with increased wood thickness. In the rustic boxes, the involucre consisted of a few layers separated by air pockets. Less populated colonies had thicker involucres. Nest The C. capitata nest contained several brood combs, which varied in diameter and number according to the hollow dimensions and the stage of colony development (Tab. 2). The food pots are frequently arranged in the ends of combs, with honey pots close to the boundaries of the hollow. The length of the nests ranged from 28 to 98 ± 7.5 cm, while the breeding and feeding areas varied from 18 to 35 cm and from 15 to 50 cm, respectively. Brood Combs The horizontal combs were usually rounded, even though the combs of two colonies raised in rustic boxes had a helical shape. The size of brood combs varied from 8.5 to 20.0 cm in diameter, with a mean value of 13. 15 ± 1 cm. The number of combs ranged from six to nineteen, with a mean value of 12.9 ± 0. Larger cells were detected in combs and up to two princesses were observed in the food pots. Propolis and resin Resin deposits were observed in the end of involucres in some nests. These egg-shaped deposits store a sticky resin of greenish to dark coloration. The resin is the basis for nest building. When the resin is transferred, for example, to a rational box, the process of nest regeneration is facilitated, because the workers reuse it. After rescue, the transference of the resin to the rustic boxes favored the fast organization of nests by bees, by using the resin/propolis to caulking the orifices. Cerumen (material used by bees in the production of involucre) when new, is fine and flexible yet rigid, but the cerumen reused by the workers becomes thicker due to the accumulation of layers. The coloration results from the resin compounds and their reutilization turns the resins darker. Scutellum In the boxes, the scutellum could be observed in the inferior and superior parts of nests while this structure was present only in the inferior end of nests in trunks. The mean measurements of the thickest portion of scutellum were 8 cm in thickness and 10.9 cm in length, being smaller in trunks of living trees, since scutellum has the function of thermoregulation, which is not so Food pots The food pots, large and egg-shaped, formed a compact mass in developed colonies. The pollen pots were located near the nest, as reported in other species, in lower numbers than honey pots depending on the season. However, a greater number of pollen pots than honey pots were observed in some colonies. The mean number of honey pots was 21.3 units/nest. The pots were egg-shaped and built with thick and dark cerumen, and their height ranged from 2.5 to 4.5 cm. Production of honey and pollen The honey production in the studied colonies ranged from 1.0 to 3.5 liters, and local people reported from 10 to 15 liters of honey removed from a single colony. The honey coloration ranged from white to pale yellow with a sour to sweet taste. The honey has high humidity, around 60 to 71% (ºBx). The size of pollen pots ranged from 3.0 to 3.5 cm in diameter and 3.5 to 4.5 cm in height. The mean weight of pollen pots was 6.76 g with a variation between 4.25 and 9.27g. The estimated pollen production varied from 67 to 210 g. Natural enemies In colonies transferred to wooden boxes without the replacement of scutellum, pollen pots and brood combs containing destroyed eggs or larvae, the number of phorids decreased, thus avoiding the loss of colonies. Population The number of individuals in ten evaluated colonies ranged from 6,561 to 21,600, with a mean population of 13,315. 1 ± 3,961.5. Distribution range Knowledge about the actual distribution range of species is essential for bioecological studies and the design of proper conservation and management strategies since it increases protection of populations from potentially threatened groups (Hey et al., 2003). In the northeastern Brazilian state of Bahia, C. capitata was first reported to be found rainforests or transition zones, characterized by temperatures from 15 to 30°C and thermal amplitude between 3 and 5°C, with cool nights and warm days, high humidity (70 to 90%), and rainfall ranging from 700 to 2300 mm/ year (SEI, 2015). Canavieiras, 4 m above sea level), contains mainly vegetation typical of an ombrophilous forest domain with high humidity and narrow thermal amplitude. On the other hand, Cravolândia, Jequié, Mundo Novo, and Amargosa are characterized by lower humidity but the barriers of hills between high altitudes (400 to 600 m) accounts for the increased humidity and the presence of seasonal ombrophilous forests in these localities. The municipalities of Presidente Tancredo Neves, Gandu, Wenceslau Guimarães, Ubaira, and Taperoá are mostly covered by Atlantic rainforest with a humid climate, ombrophilous vegetation and highly varied altitude from the sea level to 500 m (SEI, 2015). The municipality of Cocos represented the range limit of C. capitata in the State of Bahia. It is located in the western portion of Bahia with a dry climate typical of cerrado (Brazilian savannah), gallery forests and mean rainfall of 900 mm/ year with a large thermal amplitude. In spite of the presence of forested areas, this region presents low humidity during extremely dry periods. These features differ from the other localities where C. capitata was recorded, which probably hinder their dispersal beyond this point since the environmental conditions are not adequate for their survival. According to Tab. 1, C. capitata nests were found in altitudes ranging from four to 700 metres. Thus, altitude does not seem to limit their distribution. On the other hand, the studied region has a warmer daytime temperature and colder night-time temperature, so daily temperature range seems to be a more relevant factor. This species was found occupying thick tree holes, suggesting this is a species thermoregulatory requirement. Nesting The nest analysis indicates that trunk thickness is essential to their nesting, since no nests of C. capitata were detected in thin trees. Contrary to Quezada-Euán (2005) from Mexico, colonies with nests in soil were not observed in the present study. Because of their high specificity in relation to nesting conditions and the increased human impacts of the studied region, this species is restricted to the well-preserved humid forest fragments. Apparently, C. capitata is not restricted to the height of nests in relation to the ground since Kerr et al. (1967) reported nests 2 m above the ground in living trees from the Amazon forest, while Nogueira-Neto (1970) described nests in trunk hollows close to the tree basis. On the other hand, large trees that provide wide hollows with thick walls are preferred by the studied species. Most likely, these conditions favor the development of high-populated colonies with a large number of brood combs. Nest entrance Roubik (1983) reported that aged nests had a protuberance below their entrance in which bees deposited resin. Nogueira-Neto (1970) proposed that the nest entrances represented an orifice built with dark resin mixed with wax, forming an external hardly visible and small projection without an entrance tube. Instead, the entrance is a simple orifice formed of cerumen in the knots or crevices of a trunk Geographic range of Cephalotrigona capitata (Wille & Michner, 1973). Kerr et al. (1967) highlighted that the upper part of the entrance tube in C. femorata colonies was rounded (0.8 cm in diameter), while the tube was 2 cm in length, dark brown in coloration, formed from cerumen and resin and guarded by seven docile guards around it, The entrance was an ornamented orifice with dimensions slightly larger than the bee heads. Wille & Michner (1973) reported that the entrance tube in nests could reach up to 65 cm being connected to the food pots. Although in this study the entrances of evaluated nests were not very visible and although the characteristics of the involucre may differ at different times of the year and in different environments, in general, our observations for C. capitata are in agreement with reports by cited authors, for other species. Defenses Possibly, the collection of feces by individual C. capitata is associated with their nest-protection behavior in dry and low humidity periods. Nogueira-Neto (1970) highlighted that this species is quite docile showing no defense strategies against disturbances. Kerr et al. (1967) reported that the bees bent their abdomen over their thorax assuming a putative defensive. Involucre Stingless bees do not thermoregulate their nests as precisely as honeybees do, so active behavioural efforts play a role in this activity (Vollet-Neto et al., 2015). Although meliponini species show an array of nesting and nest thermoregulatory strategies (Viana et al., 2015), their mechanisms are self-organized and arise from simple rules followed by each worker (Jones & Oldroyd, 2006). Sakagami (1982) stressed that the involucre played a key role in the thermoregulation of colonies, particularly in the breeding areas. Nogueira-Neto (2002); Cortopassi-Laurino (2003) and Viana et al. (2015) observed that the most populous colonies to have involucres with a smaller area than the less populated, suggesting that thermoregulation is more efficient in high-density populations. The vast majority of the analyzed nests in this study presented involucres with two to three thin layers of dark cerumen covering the brood combs. Similarly, Wille & Michner (1973) reported that the involucre had three thin layers above and a lateral layer below the combs. In the present study, the involucre structure described for natural nests is similar to that observed in rustic boxes. The involucre is placed below a dark layer composed of cerumen and hard resin. When the boxes are opened and the upper cerumen layer is ruptured, workers come quickly carrying resin in their mandibles to repair the involucres, as reported in other bee species (Zwaal, 1992). Although the involucre characteristics may vary at different times of the year and under different environmental conditions, in general our findings for C. capitata are in agreement with those reported by other authors for other meliponini species. Nest and Captive rearing Based on the evaluated colonies (Tab. 1), we recommend that the boxes for the captive rearing of C. capitata should be 18.0 cm wide x 18.0 cm long with a depth of 10.0 cm per handle or module. With these dimensions and a minimum thickness of 4 cm, the boxes are able to maintain an enhanced temperature balance in the nest area, which favors colony development, reduces scutellum formation and forces the bees to build honey pots. Moreover, it is useful to place another compartment below the box (18.0 x 18.0 x 3.0 cm in height) to serve as waste disposal. This recommendation is based on trunk thickness data presented in Tab. 1, in which nests were found in live trees with trunks thicker than this value. We suggested this thickness for the boxes in order to guarantee adequated conditions for thermoregulation, since live trees allow nests to maintain a constant temperature. Kerr et al. (1967) observed colonies with a maximum diameter of 28 cm, length of to 40 cm, and a storage area of 25 cm x 10 cm. In our study, the nests were built in trees with a mean trunk cavity diameter of 19.8 ± 3.0 cm and mean length of nest area of 57.7 ± 7.5. In this way, our study's recordings are in agreement with the findings of these authors, since they suggest that these species require large vertical spaces for their development, while wood thickness is important for colony maintenance. However, not only the thickness of the wood provides adequate thermoregulation, but also technical management which includes feeding and reducing wind and rain incidence to the farming/rearing box. Brood Combs In our study, the size of brood combs varied from 8.5 to 20.0 cm in diameter (Tab. 2), which was influenced by nest development and the diameter of hollows ranging between 15.0 and 28.0 cm (Tab. 2). Some nests had d from six to nineteen small combs, with an average of 12.9 (Tab. 2) because they had been built in small-diametered hollows, while well-developed nests with a large population built in big/bigger tree hollows had larger combs. These data were in agreement with findings by Kerr et al. (1967), who had analyzed the nests of this species in Amazonas and had found up to seventeen horizontal combs that reached up to 25 cm of diameter. These data are essential to define the proper dimensions of rustic boxes with the intention to rear this bee species in captivity. According to Nogueira-Neto (1970), C. capitata build both helical and horizontal brood combs. Similarly, in the present study combs both horizontal (most combs) and helical and real cells were detected. Zwaal (1992) observed the occurrence of royals cells built separately from the brood combs in natural colonies but not in artificial boxes. Ihering (1903) reported no royal cells, even though up to twenty princesses were found in nests, suggesting that the royal cells could have been destroyed after their emergence. Propolis and resin According to Kerr et al. (1967), propolis has sticky consistency covering the cerumen and stored in small amounts. Nogueira-Neto (1970) reported that stored propolis is slightly viscous, while the batumen and caulking are composed of thick layers of dark cerumen. Thus, the observations described in our study, which also identified sticky resin of greenish to dark coloration are in agreement with other studies. Scutellum The scutellum is composed by residues accumulated by bees (pollen remnants, bees carcasses, broom residues and debris). This material ferments and assists in thermoregulation, and has been found at various sites within the nest (at the bottom or at the top of the nest, and sometimes close to the breeding combs). The scutellum has yellowish coloration and gives off a strong smell when managed (Zwaal, 1992). Zwaal (1992) observed that the scutellum in natural nests surrounds the breeding area, putatively acting as efficient passive thermoregulators when compared to the active thermoregulation performed by workers (Quezada-Euán, 2005). In this work, no bees were seen carrying waste away from the nests. On the other hand, we find a lot of scutellum within the nests. This suggests that workers were using the colony residues in the scutellum production, for thermoregulation. C. capitata accumulates debris inside the colony in order to form scutellum, because otherwise the workers would discard it. This result is similar to the report by Zwaal (1992), while Roubik (1983) suggested that the scutellum might be involved in the protection and isolation of nests, besides serving as waste disposal. Food pots Nogueira-Neto (1970) described egg-shaped pots, and Kerr et al. (1967) observed that this species builds large pots about 4 cm in height, 2 cm in diameter and 3.0 cm in length. Similarly to these authors, we also found honey pots ranging from 2.5 to 4.5 cm in height and from 2.0 to 3.5 cm in diameter and pollen pots ranging from 3.0 to 4.5 cm in height and from 3.0 to 3.5 cm in diameter (Tab. 2). Geographic range of Cephalotrigona capitata Production of honey and pollen Ihering (1903) and Nogueira-Neto (1970) found d that C. capitata produced high amounts of honey and pollen, and Roubik (1983) reported in Panama a level of honey production ranging from 175 mL to 2,230 mL but poor in taste. In our study, we observed that during the period of data collection, according to the period of the year, the number of pollen pots was superior to those of honey. Thus, the reduced honey production by C. capitata colonies of coincided with the reduced pasture areas for these bees. In general, the lowest number, diameter, weight, volume and estimated production of the honey pots presented in the Tab. 2 corresponded to the last years of data collection, period coinciding with the increase in deforestation in the studied region as well as the expansion of agriculture and livestock. Deforestation and agricultural expansion also have been cited by Costa et al. (2014) and Brown & Albrecht (2001) as factors that affected the conservation and management of Brazilian native bee species. Natural enemies The phorid Pseudohypocera kerteszi is considered the main pest of nests of Cephalotrigona spp. and other stingless bee species (Sakagami, Beig, & Kyan, 1964). To prevent attacks from these enemies, Cephalotrigona capitata build a narrow entrance orifice guarded by one or two bees and a long entrance tube with watchers along it. Some species build cavities that reduce exposure to predators by constructing narrow and long tubes (Roubik, 1983). However, because guard bees move slowly, phorids are able to invade less populous colonies, which do not defend as well as colonies more populous. The removal of waste would help to avoid attracting parasitic phorids, which can harm a colony. Waste provides a substrate on which potentially harmful microorganisms thrive (Medina, Hart, & Ratnieks, 2014). It contains scutellum stored by the bees, and phorids are attracted by the smell of fermented scutellum. Other species throw the waste but not C. capitata. Therefore, the loss of swarms is commonly due to phorid attacks. In our study, in several colonies were not attacked when waste was removed. Population The highly populated colonies in this study was associated with such environmental conditions as natural or slightly disturbed forested areas with a large number of plants producing trophic resources (Zwaal, 1992). Lindauer & Kerr (1960) estimated the nest population from 1,000 to 2,000, and Nogueira-Neto (1970) reported that in strong colonies this number could be much higher. Among the species of meliponinins, bees that present larger body size tend to present relatively smaller population size. However, this did not occur with C. capitata, which, despite being considered a large body size meliponine, presented populous colonies (Tab. 2) when compared to other meliponins. Captive rearing To keep these bees in artificial boxes, the wood thickness and isolation protective mechanisms must be maintained. Roubik (1983) stressed the importance of choosing a hard and thick wood type. Many adaptation issues have been reported in colonies kept in wooden boxs, particularly related to thermoregulation. Zwaal (1992) observed that colonies would disappear shortly when kept in 2 cm thick boxes. If the room temperature remained stable at 21°C the nests did not perish, but when the temperature was below 21°C, the mortality of offspring increased because of their deficiency in maintaining the internal temperature. Therefore, insufficient thermal insulation in boxes is a limiting factor for the captive rearing of this species. Damasco, Nunes, & Jarduli (2012) reported that in a nest of C. capitata captured in Paraná, the swarm was transferred to a wooden box equipped with a waste disposal for residues and formation of scutellum. This adaptation was successful, since the "trash basket" could be replaced by a clean one when it was infested with Phorid larvae, keeping the brood combs safety. Cephalotrigona capitata were found in wide hollows and thick wood of living trees. Most did not have obvious entrances and were located between 0.5 and 5.0 m above the ground. This species occupied well preserved stretches of forests that corresponded to the distribution of forested areas of semi-humid to humid climate, characterized by wide thermal variation, with colonies found from sea level to high altitude sites. The involucre in most analyzed nests had two to three thin layers of dark cerumen covering the majority of brood combs, which were thicker in less populated colonies. The brood combs' diameter and number varied according to the hollow dimensions and the stage of colony development. The food pots were large and egg-shaped and the mean number of individuals in ten evaluated colonies was of 13,315. 1 ± 3,961.5. From this information, efforts can be made to reproduce in wooden boxs the closest conditions possible to the natural in order to obtain better colony productivity in captive rearing. Based on the plant species most occupied by C. capitata and on the aspects related to their bionomics, it will be possible to establish strategies for the conservation of this bee species.	6,118	2019-06-01T00:00:00.000	[ "Environmental Science", "Biology" ]
LORETA With Cortical Constraint: Choosing an Adequate Surface Laplacian Operator Low resolution electromagnetic tomography (LORETA) is a well-known method for the solution of the l2-based minimization problem for EEG/MEG source reconstruction. LORETA with a volume-based source space is widely used and much effort has been invested in the theory and the application of the method in an experimental context. However, it is especially interesting to use anatomical prior knowledge and constrain the LORETA's solution to the cortical surface. This strongly reduces the number of unknowns in the inverse approach. Unlike the Laplace operator in the volume case with a rectangular and regular grid, the mesh is triangulated and highly irregular in the surface case. Thus, it is not trivial to choose or construct a Laplace operator (termed Laplace-Beltrami operator when applied to surfaces) that has the desired properties and takes into account the geometry of the mesh. In this paper, the basic methodology behind cortical LORETA is discussed and the method is applied for source reconstruction of simulated data using different Laplace-Beltrami operators in the smoothing term. The results achieved with the different operators are compared with respect to their accuracy using various measures. Conclusions about the choice of an appropriate operator are deduced from the results. INTRODUCTION Neuroscience is a fast growing field posing many challenges and requiring expertise in various different scientific areas (Markram, 2013). In order to be able to satisfy these needs the neuroscience community is asked to constantly develop new strategies for data analysis, design new experiments, improve techniques used in the past, identify previous errors and correct them etc. Where neurophysiological measurements are concerned, at the beginning of the electroencephalographic (EEG) era (Berger, 1929) scientists and physicians were content with looking merely at the signals measured on the scalp. However, a few decades later the question on where in the brain these signals originate gained more relevance (Brazier, 1949;Wilson and Bayley, 1950). This was the time when source reconstruction of brain activity was born. From then onwards, strategies and methods of source analysis were rapidly developing and many different approaches came into existence. The dipole model was improved to use spatio-temporal information (Scherg and von Cramon, 1985), distributed source models with a linear inverse appeared (Hämäläinen and Ilmoniemi, 1984), their iterative application was discussed (Gorodnitsky et al., 1995), and later also Bayesian methods for source reconstruction (Schmidt et al., 1999) were introduced. The class of discrete, 3D-distributed, linear inverses contains a large variety of available methods. One very popular and widely used method is low resolution electromagnetic tomography (LORETA) (Pascual-Marqui et al., 1994;Pascual-Marqui, 1999), or also called "Laplacian weighted Minimum Norm" (Michel et al., 2004), which uses the Laplace operator for modeling the source-space correlation of the data. For a critical view on the capabilities of LORETA the reader can refer to (de Peralta Menendez and Andino, 2000). Initially, this method was applied with three-dimensional volume source space without taking into account the structure of the cerebral cortex. Soon after this, an improvement of this method was suggested by constraining the solution to the cortex and using a Laplace operator that smoothes along the two-dimensional cortical surface instead of the 3D brains space (Skrandies et al., 1995). This improvement was suggested earlier for the Minimum norm approach (Dale and Sereno, 1993) and is considered meaningful, since it is based on anatomy and helps to reduce the undesired spreading of the estimated source activity to adjacent gyri or fissures. In the first actual implementation of cortical LORETA it was suggested to use a weighted graph Laplace operator (Wagner et al., 1996). No reason was given for this particular choice of a Laplace operator and the properties of the Laplace operators were not investigated in detail. In this publication, we introduce four different surface-based Laplace operators which correspond to four different classes of operators. Laplace operators for surfaces are also termed Laplace-Beltrami operators and overviews can be found in Reuter et al. (2009), Dakov and Venkov (2014). Here, we present results on the unweighted graph Laplacian (Levy, 2006), the weighted graph Laplacian (Wagner et al., 1996), the unweighted geometric Laplacian (Pinkall and Polthier, 1993) and the weighted geometric Laplacian (Meyer et al., 2003) as a special case for a finite elements discretization of the operator (Dziuk, 1988). In the following these four operators are compared with respect to their basic mathematical properties, application and performance in the source reconstruction with cortical LORETA. Formulation of the LORETA Solution Let DǫR n×1 be a matrix containing measured EEG data at n channels for one time sample, and LǫR n×3m be the lead field matrix which is the solution of the so-called electromagnetic forward problem and contains the information about the data measured at the n channels if there is a unit strength current dipole at a given position (x, y, z) in the source space. The number of discrete source positions in the brain is denoted by m. L can be written in matrix form as: where, l ij = λ xij , λ yij , λ zij T ǫR 3×1 is the lead field for the source location with coordinates (x, y, z). The problem of EEG source reconstruction using the l2-norm can be formulated as follows: Find the source current density distribution JǫR 3m×1 which satisfies the following condition: Here WǫR 3m×3m is the weighting matrix in the regularization term and can be chosen according to the assumed properties of the true solution. If W is the identity matrix then we have the original minimum norm solution derived by Hämäläinen and Ilmoniemi (1984), if W is a diagonal matrix containing depth weights then the depth-weighted minimum norm is provided. In the case of LORETA the matrix W is defined as a combination of the diagonal depth weighting and the discrete Laplace operator. The definition of W due to Pascual-Marqui (1999) is given by: where I 3 ∈ R 3×3 is the 3 × 3 identity matrix, ⊗ is the Kronecker product (Laub, 2005), ǫR m×m is a diagonal weighting matrix (Pascual-Marqui, 1999) defined as and B ∈ R m×m is the stiffness matrix of the discrete Laplace operator (Skrandies et al., 1995) defined as Here d is the distance between two neighboring grid points, g p i is the value of a function g defined on a regular 3D grid at the point p i from that grid. N(i) is the set of indices corresponding to the direct neighbors of p i . B is described below in more detail. The regularization parameter α > 0 represents the balance between minimizing the residual norm and minimizing the regularization term. Given the above definitions the solution of the problem (1) is given by: where T ∈ R 3m×n is called the linear inverse operator and can be calculated as: for a square invertible matrix W (Hansen, 1998). The parameter α is the same as in equation (1). The equation (6) is similar to the LORETA solution for volume source space given in Pascual-Marqui (1999). The exact equation for LORETA is where pinv is the pseudoinverse (Björck, 1996) of the matrix LW −1 L T with tolerance βµ, i.e. setting all singular values of the matrix LW −1 L T less than the tolerance to zero. This resembles the truncated singular value decomposition introduced to source analysis in Wolters et al. (1999). Here β is an alternative parameter to α with values also in the interval (0 1]. The parameter µ is the largest singular value of the matrix LW −1 L T . LORETA With Cortical Constraint The next step is to develop a solution for LORETA on the cerebral cortex, i.e. not for a volume source space but for a twodimensional surface in the 3D space. First of all, the Laplace operator used before has to be changed to be able to operate on the surface. Suggestions for such Laplace operators were already described in the literature (Skrandies et al., 1995;Wagner et al., 1996). However, in these publications it is not taken into account that a large number of discretizations of the Laplace operator exists on two-dimensional surfaces, with different properties (Wardetzky et al., 2007;Belkin et al., 2008) and, consequently, yielding different results as smoothing operators (Desbrun et al., 1999). In this publication, four classes of Laplace-Beltrami operators are considered: unweighted graph Laplacians, weighted graph Laplacians, geometric Laplacians without area weights, and geometric Laplacians with area weights. Let S be the continuous cortical surface (2D-manifold) in R 3 . Its discrete form is then The general form of the discrete Laplace-Beltrami operator applied to a function f on M can be written as: where p i is the i-th node of M, f p i is the value of the function f at the node p i , w ij is the weight of the connection between the nodes p i and p j , d i is the area weight assigned to the node p i and N(i) is the set of indices corresponding to the direct (also called "1-ring") neighbors of p i . The stiffness maxtrix B contains the coefficients of the Laplace operator to be applied to the values f(p i ). Example for a simple closed mesh with four points p 1 ,..,p 4 which are all neighbors of one another: i = 1, N (1) = (2, 3, 4) : With analogous equations for f(p 2 ), f(p 3 ), and f(p 4 ), one ends up with All Laplace-Beltrami operators in this publication can be introduced on the basis of equation (9). The unweighted graph Laplacian (UW GrL) (Levy, 2006) is the simplest one. It takes into account only the adjacency of the nodes and not the geometry of the mesh (Figure 1a): The weighted graph Laplacian (W GrL) defined in (Wagner et al., 1996) is similar to the unweighted graph Laplacian but with different weights for the different connections (Figure 1b). In order to represent it in the general form (9) we define the corresponding weights as where, dist p i , p j is the distance between the nodes p i and p j . The geometric Laplacian without area weights (UW GeL) (Pinkall and Polthier, 1993) takes into account not only the connectivity between the nodes but also the geometry of the mesh by including the cotangents of the angles into the weights w ij : where α ij and β ij denote the two angles opposite to the edge (i, j) (Figure 1c). The geometric Laplacian with mixed area weights (W GeL) (Meyer et al., 2003) takes into account not only the angles but also the areas of the triangles in the mesh (Figure 1d): Due to (Meyer et al., 2003) the area A mixed is defined as the Voronoi area if the triangle is not obtuse and in the case of an obtuse triangle the area connected with the midpoint of the edge opposite to the obtuse angle. A mixed can be calculated with the following algorithm written in pseudocode: A mixed = 0 For each triangle T from the 1-ring neighborhood of p i do: If the angle of T at p i is obtuse A mixed + = area(T)/2 Else A mixed + = area(T)/4 (14) The Voronoi region of p i in T is the set of points in T such that the distance to p i is not greater than the distance to any other two nodes forming the triangle T. There are many other possible choices for the area around the nodes, however, A mixed is identified as the best-choice finitevolume region corresponding to the infinitesimal neighborhood on a continuous surface patch (Meyer et al., 2003). The areas of the triangles were calculated with a stabilized Heron's formula (Kahan, 2014) in order to avoid problems for needle-like triangles. In the following, the operators described by equations (10-13) will be compared with respect to their performance as regularization term for cortical LORETA. Beforehand, it is appropriate to look at some important properties of the Laplace operator (Chung, 1996). 1) The Laplace operator is diagonally dominant 2) The Laplace operator is positive-semidefinite (all eigenvalues are non-negative) 3) There is exactly one eigenvalue which is equal to zero. 4) The Laplacian matrix is singular. Since the Laplacian matrix is singular it cannot be used directly in equation (7) for calculating the inverse operator T. In order to be able to use the Laplace operator into (7) the matrix B T B is replaced by its approximation B T B + σ I m . In that case the weighting matrix W obtains a new form The parameter σ is in the interval (0 η 2 ], where η is the largest singular value of the matrix B T B. In order to demonstrate the effect of the parameter σ on the LORETA solution the equation (7) was modified such that depth weighting was not used and only the effect of the Laplacian Frontiers in Neuroscience \| www.frontiersin.org could be observed. The unweighted minimum norm solution and the unweighted LORETA solution for different σ 's were calculated for a simulated bilateral activation of auditory cortex, and compared for the graph Laplacian. A simplified form of the cortex was used for a better visualization of the smoothing effect. For σ = η 2 the solution with Laplace showed high similarity to the solution without Laplace. This means that the effect of the Laplace operator attenuates for large σ . For σ approaching zero the smoothness of the solution increases, until finally both auditory sources are fused together in the frontal area (Figure 2). This means that the impact of the Laplace operator becomes stronger for smaller values of σ . Intuitively, this behavior can be explained as follows: when we add large values to the diagonal of the Laplace matrix the diagonal elements become very large compared to the offdiagonal elements and the matrix behaves similar to a diagonal matrix. A diagonal matrix acts merely as a scaling matrix and does not affect the neighboring nodes. The problem remains how to determine the optimal value for the parameter σ . Until now the knowledge was derived that for too large values of σ the solution with Laplacian "converges" to the solution without Laplacian because of the attenuated effect of the operator and for too small values the smoothing effect is very strong, resulting in a solution which does not match the measured data very well. Consequently, the optimal value should be a tradeoff between these two extremes. For that reason two measures are used for estimating the optimal value: the goodness of fit (GOF) and the similarity to the solution without Laplace. The GOF is defined as in Hämäläinen et al. (1993) where, RV is the residual variance and is given by here b 1 , · · · , b n are the measured data at n channels and b 1 , · · · ,b n are the data reconstructed by the linear solver (with or without Laplacian). If GOF = 100%, the model explains the data perfectly, whereas GOF = 0% means that the model does not match the data at all. The similarity between solutions is calculated using the same equation as for the residual variance with the only difference that b 1 , · · · , b m are the normed source power values (i.e. ǫ[0 1]) for m source space points calculated without using a Laplacian andb 1 , · · · ,b m are the corresponding values calculated with a Laplacian. A value of 0% would mean that the solutions are identical. Values larger than 100% are also possible. In order to derive an optimal value for σ a data set was simulated with 2 bilateral auditory sources and noise from a real EEG measurement such that the signal-to-noise ratio (SNR) was 20. For that data set both measures (GOF and similarity) were calculated for different σ and different α values. The results were plotted as graphs with the x-axis denoting the values for σ and the y-axis denoting either GOF or the similarity to the solution FIGURE 2 \| Smoothing effect of the surface Laplacian. The top-most row and the second row show the simulated auditory sources together with a butterfly plot of the signal created at the sensors, the third row shows the unweighted minimum-norm solution, the fourth row the cortical LORETA solution for σ = 65, the fifth row for σ = 6.7e-2, the sixth for σ = 6.7e-3, and the bottom-most row shows the cortical LORETA solution for σ = 2.1e-3. without Laplacian. For each α value one line with different color was added ( Figure 3A). From Figure 3A one can choose an interval for the values of σ in which the trade-off between similarity and GOF is satisfied. At the left it is apparent that for σ > 1 the effect of the Laplacian becomes weaker and the solution is similar (deviation < 5%) to that without Laplacian. In the right figure one can see that in order to get a solution with GOF ≥ 90% the value of σ should not be smaller than 10 − ². The only value which satisfies the conditions GOF ≥ 90% and deviation to the solution without Laplacian ≥ 10% is σ = 0.04. Since the values in the first calculation were exponentially sampled now a subinterval [10 −2 10 −1 ] is chosen with linear sampling. Figure 3B shows the results after recalculating both measures for the smaller interval. Here, the values [0.07, 0.06, 0.05, 0.04, 0.03] were identified to satisfy the required conditions. The mean value of that interval was taken for further calculations, i.e., σ = 0.05. For that value the GOF is larger than 90% for all α and the deviation from the solution without Laplacian is larger than 17%. The cortical LORETA result for the optimal value is shown in Figure 4 together with the corresponding solution without a Laplacian for a simulated data set with two bilateral auditory sources. In Figure 4 one can see that the cortical LORETA solution is not more widely distributed than the solution without a Laplacian. Furthermore, the regions with maximal activity are shifted toward the simulated activity regions which are shown at the top of the figure. This procedure was repeated for all four Laplace operators and for three different types of cortical meshes. Anatomical MRI Data The procedure for determining the optimal value for σ was applied for three types of cortical meshes ( Figure 5): a simplified cortical mesh with 750 nodes, a more realistic cortical form with 3,709 nodes resulting from a non-linear co-registration of 10 individual MRI data sets, and an individual cortical mesh with 3,973 nodes. The simplified cortical surface was extracted from an averaged MRI created based on 50 T1-weighted MRIs linearly aligned in Talairach space. An initial, coarse triangular cortical surface was manually deformed to fit to the inner bone surface and finally shrunk by a small amount to approximate the cortex. For the averaged MRI resulting from the non-linearly coregistered data sets and for the individual MRI data set T1weighted structural magnetic resonance images (MRI) were obtained from 10 participants using a Philips Gyroscan 1.5 T FIGURE 4 \| Visual comparison of the simulated auditory sources (top), the depth weighted minimum-norm solution (center) and the cortical LORETA solution (bottom) for the optimal value of σ. (Jordanov et al., 2010). The non-linearly averaged cortex and the individual cortex were automatically segmented using the BESA MRI 2.0 software (BESA GmbH, Gräfelfing, Germany). Based on the provided T1-weighted MRI, first a voxel-based classification was performed (Zhang et al., 2001). Next, an implicit surface representation of the white matter-gray matter interface was reconstructed from the classification result. The implicit surface was finally slightly inflated to approximate the cortical surface in the middle of the gray matter layer (Chan and Vese, 2001;Lanfer, 2014). Simulated Data For the comparison between the different Laplace operators with respect to various performance measures 1,000 cortical sources were simulated (here, the realistic cortex with 3,709 nodes was used), however, colocalization with nodes of the cortical mesh used for the calculation of the leadfields was prevented in order to avoid an inverse crime. In that way 1,000 simulations without noise were generated for 1,000 different source locations. In every simulated data set only one source was active. For all simulations a 4 shell ellipsoidal head model as implemented in the software package BESA Research 6.1 (Berg and Scherg, 1994) was used. Leadfields were computed for a sensor configuration comprising 127 electrodes on a dense grid in an equidistant layout as defined by the geodesic sensor net (Tucker, 1993). Measures for the Comparison The measures used for the comparison were localization error, source depth, depth shift, number of local maxima, residual variance and computation time. For the estimation of the localization error and of the depth shift two different approaches were applied. The first approach was to take the mesh node with the maximal amplitude (MA) as the location of the estimated activity and the second approach was to take the center of mass (COM) as the location of the estimated activity. In both cases the image resulting from the source reconstruction was prepared in such a way that all active nodes with activity below 50% of the maximal activity were set to zero. In that way it was ensured that only the nodes with significant contribution to the source reconstruction were used for the comparison (Lin et al., 2006). The localization error was defined as the distance between the estimated and the simulated source location (Lin et al., 2006;Lucka et al., 2012). Smaller values for the localization error mean better solution. The source depth was defined as the distance between the source (simulated or estimated) and the nearest sensor. A solution was considered good if the depth of the estimated source was approximately the same as the depth of the simulated source. The depth shift was defined as the difference between the depth of the simulated source and the depth of the estimated source (Lin et al., 2006;Lucka et al., 2012). Positive values for the depth shift mean that the simulated source was deeper than the estimated one. The number of local maxima was investigated to provide an estimate of how many false positives were found in the estimated solution. A local maximum was defined as a mesh node which had larger amplitude than all of its direct neighbor nodes. For the statistical comparison of the measures one-way ANOVA was applied as implemented in MATLAB and Statistics Toolbox Release 2012b, The MathWorks, Inc., Natick, Massachusetts, United States. For the pairwise comparison of the measures in a post hoc statistical step, Tukey's 'Honest Significant Difference' method was used as implemented in R (R Core Team, 2015). Identifying an Appropriate α Value After the optimal value for the parameter σ was determined it was necessary to look for an appropriate α value to be used for the calculations. There are many automatic procedures for doing that, e.g., the L-curve. However, these procedures do not always yield the "best" value. Therefore, another procedure was used here. The first step was to determine the α value for data without noise. Since in equation (7) the pseudo-inverse is used instead of the inverse, α is not a continuous quantity and can only change with the singular values of the given matrix. For that reason an additional parameter is introduced which determines how many singular values should be set to 0 for the inversion. This parameter, called SVD cutoff index (SCI), can have only integer values between 1 and the number of total singular values. If the value is 1 then only the last (smallest) singular value is set to zero, if e.g., the value is 10 then the ten smallest singular values are set to zero. After the SCI is chosen the corresponding SVD cutoff value (SCV) is chosen to be between the singular value with index SCI and the one with index SCI+1. The value for the parameter α can be calculated as the ratio: The procedure of determining the optimal α value is to calculate the estimation once with and once without Laplacian for many different SCIs and then to take the index with the best results with respect to localization error and number of local maxima. RESULTS The first step in the analysis of the different Laplace-Beltrami operators was to determine their optimal values for σ . This was done by the procedure described in the section "LORETA with cortical constraint." The results are shown in Table 1. In the case of individual cortex, for two of the values for α (0.01, 0.005) there were no acceptable σ values for any of the weighted graph Laplacian, unweighted geometric Laplacian and weighted geometric Laplacian. Therefore, the maximal α value used in these cases was 0.0025. The values in Table 1 for σ were used for further calculations. In the following, the optimal value for α in the case of data without noise and for the case of the realistic cortex with 3,709 nodes was determined. The source estimations were calculated with and without Laplacian for one randomly selected dataset from the simulated data without noise for SCIs 1, 20, 40, 60, 80, and 100. For every SCI the localization error (point with max. amplitude) and number of local maxima were calculated. The results are shown in Table 2. From Table 2 one can see that the localization error and the number of local maxima for both using no Laplacian and for unweighted graph Laplacian, are best for the smallest SCI, in that case 1. Consequently, the α-value for the further calculation was chosen to be 9.9e-13 for the operator without Laplacian and 7.5e-15 for the operator with the unweighted graph Laplacian. It was assumed that the behavior of the inverse operator with the remaining Laplacians would be similar to those shown in Table 2. Therefore, for all Laplacians only the smallest singular value was set to zero. This resulted in the following α values: α NoL = 9.9E-13, α UW GrL = 7.5E-15, α W GrL = 5.2E-15, α UW GeL = 6.5E-15, α W GeL = 6.3E-15. These values were used for further calculations. The first measure investigated for the accuracy of LORETA with different Laplace operators was the correlation of the estimated source depth with the simulated source depth. This correlation was computed for two different approaches: once the center of mass (COM) was taken as the point for the estimated source, and once the grid point with the maximal amplitude (MA) was taken. The correlation coefficients for the different Laplace operators are depicted in Table 3. The comparison of the regression slopes of the source depth by the mean of one-way analysis of covariance (ANCOVA) resulted in statistically significant differences in both cases: COM (F(4, 4990) = 6.637, p < 0.0001), and MA (F(4, 4990) = 53.52, p < 0.0001). The p-values obtained for the pairwise comparison of the slopes are shown in Table 4. Figures 6, 7 visualize the data together with the corresponding regression lines. There was no statistically significant difference between the slopes of the different Laplace operators. However, it emerged that the slopes for the source depth correlation in the case of reconstruction without Laplacian were statistically different from the slopes in the cases with Laplacian (Table 4). In order to statistically investigate the different Laplace operators with respect to the depth bias the mean values of the depth shifts were compared (Figures 8A,B). For the depth shift there was a statistically significant difference between the results using different Laplace operators and no Laplace at all as determined by one-way ANOVA for both COM[F (4,4995) = 282.52, p < 0.0001] ( Figure 8A) and MA [F (4,4995) = 101.74, p < 0.0001] (Figure 8B). A Tukey post-hoc test revealed that the source estimations without Laplacian showed significantly larger depth shift than the source estimations with all other Laplacians (p < 0.0001) ( Table 5, columns 4 and 5). Between the Laplace operators only the depth shift with the unweighted graph Laplacian was significantly different from the one with the weighted Laplace operator in the case of COM (p = 0.009); all other pairwise comparisons did not yield significant differences ( Table 5, column 4). In the case of MA the difference was significant only between the weighted graph Laplacian and the unweighted geometric Laplacian (p = 0.002) ( Table 5, column 5). Another measure was the number of the local maxima as an indicator for false positives in the solution. Again the comparison manifested statistically significant results [F (4, 4995) = 233.37, p < 0.0001] (Figure 9A). Additionally, all but one of the pairwise comparisons were significant too (Table 5 column 6). The exception was the comparison of the weighted graph and weighted geometric Laplacians (p = 0.95). These two operators manifested a smaller number of local maxima than any other Laplace operators. For the residual variance also a statistically significant difference was manifested [F (4, 4995) = 1083.76, p = 0.0] ( Figure 9B). The solution without Laplacian yielded the lowest residual variance compared to the solutions with Laplace operator. The weighted geometric Laplacian revealed the lowest residual variance of the Laplacians. The last measure used for the comparison between the Laplace operators was the computation time in Matlab. The computation time investigated considered only the Laplace operator steps, not the computation time for the inverse operator or for the final solution. The main trend was that the graph Laplacians were calculated faster than the geometric Laplacians (Figure 10). DISCUSSION The aim of this study was to introduce the source localization method LORETA with a cortical constraint and identify an appropriate Laplace-Beltrami operator for optimizing the source reconstruction. The operators in this study were chosen in such a way that they represent an entire family of operators, have sparse structure and operate on a triangular mesh. These operators comprised the unweighted graph Laplacian, the weighted graph Laplacian, the unweighted geometric Laplacian and the weighted geometric Laplacian. The selected operators were investigated with respect to different measures, namely localization error, depth shift, number of local maxima, residual variance and computation time. The optimal solution of the current problem would be the one which minimizes the investigated measures. For the measures involving location of estimated activity two approaches were followed: using the center of mass (COM) as an estimated source activity location, or using the mesh point with the maximal activity (MA). These two approaches are both common for simulation studies and both have their advantages and drawbacks. The mesh node with the maximal amplitude seems to be more intuitive from the view of the experimental scientist who visually identifies the location of the estimated activity by means of a color map. The idea of using the center of mass follows the line of thought that the solution gained with an l2-norm based linear inverse solver is very smooth and widely distributed, and thus, taking only the maximum of this solution does not take into account the distributed nature of the solution. Since these two approaches yielded different values for the performance measures in this study, both of them were included in order to produce results comparable with other publications using only one of these approaches. One major property of the minimum norm solution without a Laplacian is that it has a bias toward superficial sources (Skrandies et al., 1995;Lin et al., 2006). It was shown that even the usage of an additional depth weighting term in the linear operator was not always successful in compensating for that bias, and suggested that also the Laplace operator can contribute to the correction of that bias (Skrandies et al., 1995). Therefore, the depth of the estimated sources was investigated in order to be able to detect if there is an advantage in using the Laplace operator with respect to the depth bias. The measures used in this study for the investigation of the depth bias influence were the correlation of the simulated sources depth with the estimated sources depth, as well as the depth shift. It was manifested that using the Laplacian in the inverse operator improved the correlation of depth between simulated and estimated sources, thus confirming previous findings in the literature about the 3D Laplacian (Skrandies et al., 1995;Pascual-Marqui, 1999) to be valid also for the cortical case. However, there was not a significant difference between the various operators. A similar result was observed also for the depth shift where all solutions using the Laplace operator performed significantly better than the solution without Laplace, however, none of the operators was superior to the others. Contrary to our expectations to find significant differences in the performance of cortical LORETA with respect to the different Laplace-Beltrami operators, no operator emerged to be superior to the others. One possible reason for that behavior could be, that the triangular meshes used for discrete representation of the cortical surfaces were rather regular (or "not irregular enough"), even in the case of individual cortices, and applying operators with weights aimed to correct for irregularity of the meshes did not significantly improve the solution. Another possibility for the missing differences could be the relatively low resolution of the l2-norm solution of the inverse problem. Testing the performance of the source activity estimation using an iterative method like CLARA (Jordanov et al., 2014;Beniczky et al., 2016) could provide a different insight into the source reconstruction capabilities using a Laplace-Beltrami-operator. A standard measure for the performance of a source estimation method is the localization error. The main question to be answered was how reliable is the result. Again, for this measure, the results did not indicate one Laplacian which performed better than all others, however, using a Laplacian was again shown to be advantageous for the source estimation. A completely different approach for the comparison of the solutions was to explore the number of false positives in the estimations. It was known that the simulations contained only one active source and it was expected that the estimated solutions contained also only one source. For the investigation of this assumption, the number of local maxima was considered to be an appropriate measure. In all cases the number of estimated source locations was higher than the number of simulated sources. Again the application of a Laplace operator yielded less local maxima than in the case without Laplace. This finding can be explained by the smoothing effect of the Laplacian, which leads to a reduction of high-frequency spatial components in the reconstructed source activity. With respect to this measure, two of the Laplace operators manifested a significantly smaller number of local maxima than the other Laplacians. These were the weighted graph and the weighted geometric Laplacians. The operator which best minimized the residual variance was the weighted geometric Laplacian. However, it performed not as well with regard to this measure as the case without Laplace operator. The solution without Laplacian was expected to yield lower residual variance than the solutions with Laplacian, since LORETA minimizes not only the norm of the solution and the difference between the modeled and the measured data but also strives for a smooth cortical solution. This is an additional condition which has to be fulfilled and this is done at the cost of increased residual variance. The last measure used for the comparison was the computation time. The main finding here was that the graph Laplacians are calculated faster than the geometric Laplacians. The fastest operator was the unweighted graph Laplacian and the slowest one was the weighted geometric Laplacian. The maximal computation time was less than 4 s for the used meshes, consequently, it was not considered as a crucial measure for the choice of the operator. However, if in the future considerably finer grids are used the computation time is going to be an important factor for the decision. Regarding all measures investigated in this simulation study, one can conclude that it is generally preferable to use a Laplacian in the linear inverse operator. Methods with Laplacian yielded better results for all used measures except for the residual variance. The identification of one best Laplace operator turned out to be nontrivial, since there was no operator which performed considerably superior to all others. However, two of the operators (weighted graph Laplacian and weighted geometric Laplacian) manifested a smaller amount of false positives than the others, which was considered as an advantage and, consequently, these two operators can be identified as the best candidates for LORETA with cortical constraint. Limitations of the Current Study and Future Direction Although this study was designed to be complete and selfcontained, it was not possible to take into account all possibly relevant aspects. Since there are many different discretizations of the Laplace operator, it was not possible to investigate all of them. For further types of Laplace-Beltrami operators please refer e.g., to Reuter et al. (2009), Dakov and Venkov (2014). The measures used for the comparison of the operators were applied only for simulated data without noise. Additional investigation of the operators' behavior in the presence of different signal-to-noise ratios would be an interesting topic of further studies. An application of cortical LORETA on measured EEG or MEG data and a comparison study with other, more recent localization methods, would be meaningful after the best operator has been determined. However, this is out of the scope of the current study and worth of future investigation. Additionally, it is important to mention that there exists an alternative formulation of the linear inverse operator T: which is equivalent to the operator given by equation (6) (Hansen, 1998). The difference between T given by equation (6) and T 2 is that T 2 can be calculated also for a singular matrix W. In the case of cortical LORETA the discrete Laplace operator is always singular. Consequently, the factor α would fulfill two tasks simultaneously: 1) The balance between minimizing the residual norm and minimizing the regularization term and 2) Correcting the rank of the matrix to be inverted. If we choose a value for α, which is more appropriate for task 1), e.g., 0.005, then this value would be far too high for correcting the rank of the Laplace operator. Consequently, the effect of using the Laplacian would be extremely reduced, resulting in a solution more similar to Minimum norm than to LORETA. If, on the other hand, we choose α in such a way that it is more appropriate for correcting the rank of the Laplacian, e.g., 1e-7, then the Laplacian's rank is going to be corrected but the balance between minimizing the residual norm, and minimizing the regularization term, is not going to be optimal. α and σ depend on different data (σ is dependent on the mesh, whereas α is dependent on the functional data -EEG), thus it cannot be guaranteed that a value for α which is appropriate for both purposes always exists. These considerations led to our idea to introduce the additional factor σ . Furthermore, we use σ only for the discrete Laplacian instead of for the entire matrix W, since all other matrices participating in the calculation of W are regular; changing their singular values using σ would distort the solution. AUTHOR CONTRIBUTIONS All authors listed have made a substantial, direct and intellectual contribution to the work, and approved it for publication.	9,219	2018-10-30T00:00:00.000	[ "Engineering", "Physics" ]
Could Noise Spectra of Strange Attractors Better Explained Wealth and Income Inequalities ? Evidence from the S & P500 Index SUMMARY: Inequity in wealth and income distributions is ubiquitous and persistent in markets economies. Economists have long suspected that this might be due to the workings of a power law. But studies in financial economics have focused mainly on tail exponent while attempting to recover the Pareto and Zipf’s laws. The estimation of tail exponents from log-log plots, as in stock market returns, produces biased estimators and has little impact on policy. This paper argues that economic time series are output signals of a multifractal process driven by strange attractors. Consequently, estimating noise spectra thrown-up by strange attractors stands to produce a much richer set of information, including the lower and upper bounds of unequal income distribution. 1-INTRODUCTION The World Inequality Report 2018 (Alveredo, Chancel, et al. 2017) is out.It shows that during 2015-16 the top 10% of individuals captured some 37% of total wealth in the European Union that has the best score among other huge economies such as China (41%), Russia (46%), Brazil (55%), the USA (47%), India (55%), etc.).In terms of income inequality, the report also shows that from 1980 to 2018, the world's richest 1% captured about 27% of new income, while the poorest 50% of the world population received only 13%.Thus, compared to the years preceding globalization, income inequality has increased almost everywhere, that is, in developing or developed countries, including those that enjoy the highest level of development.At first sight, one would tend to suspect similar policy structure as the culprit.However, observing that the top 10% had captured some 61% of total wealth in Sub-Sahara Africa, wealth and income inequality then appears to be an ubiquitous phenomenon.It is then reasonable to suspect that indeed some homogeneous power law lurks behind the distribution of wealth and income in market economies. A homogeneous power law can be represented as: F(x) = K x  , where F (.) is a variable, K is a constant, x is the other variable of interest, and  is the scale exponent.Financial economists have attempted to study the impact of power law identified in stock market activities, such as firms' size distributions, returns on investments, trading volume, etc. (Gabais, 1999;Gabais et al., 2003, among others).But, their studies attempt to reproduce Zipft' hyperbolic law, which relates word rank and word frequency in natural languages, or they are simply influenced by the so-called Pareto Law, expressed as: (p(x) > x) = K / f  .Moreover, estimates are obtained from log-log plots (Kopikrisnan et al., 2000).Experimental data of quantities that follow a power law are usually very noisy and therefore obtaining reliable estimates for the exponent  is difficult.In fact, estimates obtained from graphical methods based on linear least squares fit of some empirical data points, such as a time series, produce biased estimators.If precision is a requirement, then it is advised to use maximum likelihood methods as an alternative, as e. g. the so-called Hill's estimator that gives the inverse of the exponent of the Pareto distribution.In practice, however, one never know for certain whether an observed quantity is drawn from a power law distribution.At any rate, the information obtained either from graphical methods or others is of little use in policy formulation.For example, an  = 2, say, means that if F (.) is doubled, then x will go up by 2 2 . Even if a power law is really present, the researcher will not know how it manifests itself. Moreover, if the output signal is from a multifractal endowed with many scale exponents and driven by a strange attractor, what scale is being estimated?Clearly, graphical methods are of little use in such situations. Yet, power laws with fractional exponents showing scale invariance are perhaps the only way to study many critical phenomena in the real-world settings.The values of exponents reflect a large number of regularities found in physics, biology, psychology, etc., and in many human constructs such as music, economics, etc.Whether the exponent is an integer or a non-integer, relations between variables are characterized by the notion of self-similarity.And we contend that by knowing why self-similarity or dissimilarity exists is to know much more about the phenomenon that we wish to study in this paper. For tractability, however, let us first recall that self-similarity is more than a welcomed attribute, because it is fundamental to our nature whether we are aware of it or not, and also because it is the concept that underlies fractals, complexity, and many laws of nature.In this paper we intend to look at income distribution through the lens of fractals, and there, self-similarity plays a huge role, as evidenced by the difficulty of distinguishing between minute-by-minute or second-by-second stock averages (Mandelbrot, 1982;Priya, 2010;Koulakis and Ruseckas, et al., 2006;West and Shlesinger, 1990).Indeed, power laws are found in many scientific and human constructs.Proper analyses in these areas requires some generalization of the fractal concept either with many exponents (where they are called multifractals), or fractal 'tout court' with one scaling exponent; at any rate, self-similarity may appear on many scales.Hence, it would really be surprising if some power law did not underlie stocks and commodity exchanges. Power laws also govern the noise spectra of various processes.But, interpreting noise spectra has its own pitfalls, because non-stationarity in the data can produce fake scaling behavior, or as evidenced by the excessive association of the Zipf's law with pink noise or 1/f noise.The latter is a signal that gives a power spectral density (power per frequency interval) that is inversely proportional to the frequency of the signal.Maybe, the attachment of economists to 1/ f noise ((where  (= 1) is the squared modulus of the Fourier transform) is from the fact that 1/ f noise is connected to systems that are near their equilibrium.Whereas  > 1 is associated with nonequilibrium driven systems.Thus, values of  = 1 or 3 found in the studies mentioned above probably will not occur in spectral analyses, because economic systems are most often far from their equilibrium, and economic data are very noisy.Spectral analysis provides a representation of the output of a noise-free system, from which on can derive precious quantitative and qualitative information concerning the system's behavior. The stochastic behavior of a deterministic dynamic system could show a Power Spectrum Density (PSD) that is qualitatively similar to that of a truly random time series, as in a broad-band power spectrum.On the other hand, a purely stochastic system may also contain a flicker noise, because low frequency components dominate high frequency ones; this is known as the intermittency problem.That is why we believe that it is safer to start by first characterizing the attractor in view of ascertaining whether or not the system under study is stochastic or one that has structure. Power laws abound in the human experience.Its operation can easily be inferred from a few casual observations such as: Distributively, the frequency of 'smalls' exceeds that of 'bigs'; the fact that there are more small stars than big ones; the fact that the frequency of unstable configurations is higher than that of stable configurations in the universe; or why the frequency of small words exceeds that of long words, etc.It then follows that it might be natural that the frequency of smaller income earners would exceed the frequency of big earners, if market economies are driven by power laws.However, one must be careful in dealing with spectral power that is continuous and diverges to low frequencies.There exist a plethora of methods, such as the multitaper method, to handle these situations.In this study we are less interested in precise measurements than in determining the upper and lower bounds of power spectra of multifractals whose fractal attractors are characterized by fractional exponent. If the S&P-500 Index is the output signal of a multifractal, it is then worth investigating whether income distributions (which is of interest here) are better studied by noise spectra (their power spectrum) than by reconstructed distributions from tails exponents.To do so, however, it is necessary to first express the power law in frequency domain, i. e., f (f) = k / f  , where k is a constant,  is the power spectrum, and f (t) is the power function that measures the power of the signal per unit of time.We next characterize the attractor of the index to determine its fractal dimension.We will attempt that in the next two sections.The second will be devoted to the data, to preliminaries to the computation of the singularity spectrum (to demonstrate both its multifractality and its non-randomness), and to the noise spectra of the S&P-500.Our findings will be summarized in the Third Section. 2-DATA AND PRELIMINARIES We used the Grand Microsoft Excel series of closing prices of the S&P-500 from January 3 rd 1961 to February 28 th , sampled at daily intervals, and expressed as a Mixed Fractional Brownian Motion ((MfBm), see Appendix A), assuming its non-randomness.The series was next truncated into 7 segments (the b's in Appendix A) that were previously determined in Dominique and Rivera (2012), and each segment was de-trended using 3 logarithmic differences and filtered for white noise.Segment length varies from 2 9 to 2 11 .The analysis will be done in two stages.In the first, we will use the wavelet multi-resolution software of Trusoft International, the Benoit version, to determine the boundaries of the observational range or the limit of the Hurst exponents (H), as well as the Hausdorff dimensions ((D0 ), see Appendix B).In the second stage, D0 being the first scale exponent will be used as the starting point in the determination of the generalized fractal dimensions or the singularity spectra of the segments. As shown in Appendix B, the Hausdorff dimension (D0) is a more efficient measure than either the topological or the box-counting dimension, and also a more natural measure within the multifractal formalism.For, if a closed bounded set E  ℝ n (where ℝ is the real line) is a manifold, then the value of its dimension must accordingly be either i) an integer or a non-integer; and ii) points and countable unions of points of zero volume must have zero dimension.It can then be seen that the topological dimension fails on these conditions since it is always an integer, giving zero for the Cantor set for example, which is obviously not true.By a similar argument, the boxcounting measure also fail on ii), whereas D0 satisfies both i) and ii), and D0 (E) ≤ dim box (E). Recalling, at the same time, that if V ⊆ X and G is a collection of subsets of X whose union contains V, then G is a cover of V.If further X is a topological space, then G is an open cover if each of its subsets is an open set.Therefore, the term 'fractal dimension' , more generally referred to as the capacity dimension of fractal sets, is also the exponent D0 in the expression  (e) = e -D 0 , where  (e) is the minimum number of open sets of diameter (e) needed to cover the set (Rpowerasban,1990).The D0 given by the software will be the starting point in the computation of the singularity spectrum, including the correlation dimension. The Singularity Spectrum The method of multifractal cascades is now known as the multifractal formalism.Mandelbrot (1974) and Frisch (1995) introduced it in response to systematic experimental deviations observed in the Kolmogorov theory of homogeneous and isotropic turbulence.It has since undergone considerable theoretical development and practical applications in many disciplines as it seems well adapted to reveal the hierarchy governing special distributions of singularities of multifractal measures. In this paper it is referred to as the Mandelbrot Method (Mandelbrot, 2003) which is a simple iterative construction that asymptotically models strange attractors.It consists of an 'initiator' (the unit interval) on which a unit mass is uniformly distributed, and a 'generalized generator' () with two intervals (ei), i  (1, 2).The initiator is first divided into two bins with equal probability (pi). Next, the exponent q is assigned to the probabilities, while the exponent  is assigned to the support intervals.The exponent (q) is the Renyi's scaling exponent, and q is a real parameter that can take positive as well as negative values.In the case of monofractal (or self-affine),  (q) depends linearly on q; otherwise the process is a multifractal. Quadratic maps have the same structure, but different intervals.Experimentally, Schroeder (2009) has found that an interval size e1 = 0.400 to be a good approximation of ei for the logistic map.By using Equation (3) below, our e1 = 0.408903, which is equivalent to a generalized generator of  = 2.445564 instead of the approximated value of 2.5, chosen initially by Schroeder.The difference is due to the fact that the logistic map is not exactly self-similar.Since the approximate map of a given process might not be known in advance, one should appeal to equation (1) below to yield the size of the generator and intervals from the Hausdorff dimension obtained from the wavelet. Once e1 is known, then all the Renyi's generalized fractal dimensions, except D1 (the information dimension), can be computed.But beforehand, the Legendre Transform posits: (q) = -d(q) / dq; df () / d = q, and f () = q d/ dq- (q).While the Holder analysis decomposes a measure into a sum of measures, where each is characterized by a value of the Holder exponent ; the latter measures the strength of the local singularity or roughness.Everything is then on hand to construct the multifractal spectrum; thus, the generality of D0 in this approach cannot be over-emphasized. For D1, we have: but, for D and D-, it is easier to expand the numerator of (6).That is, Dq = -log 2 [log 2 (1 + q) ] / (1 -q) log e1; (8) using log 2 and letting q   or (-), we have: What is needed here is the correlation dimension that can be computed for q =2; that is: The importance of D2 lies in its relation with the concept of correlation.It can be shown that as e  0, the sum in ( 9) equals the total counts, defined as C(r) and used in the method proposed by Grassberger and Procaccia (1983).One of the many roles played by the correlation dimension lies in its ability to distinguish between chaos and random determinism.In the Grassberger and Procaccia method, one builds a d-dimensional data vector from d measurements spaced equidistantly in time, and determines D2 of the d-dimensional point set.If the data were random, then as d increases, D2 would increase continuously with d.However, if the system is deterministic, D2 will not increase any more once the embedding dimension exceed D2.For more on this, the reader is referred to Appendix C. 2.2-The Power Spectrum The wavelet multi-resolution software computes the Hurst exponent (H) for each segment of the S&P-500 Index, and the segments are used to calculate the power spectrum as shown in Appendix D. It should stressed at this juncture that self-similarity is at play, indicating that there exist relations between variables, and D0 is at hand.The power spectrum can then be computed for each segment of the Index.The Hurst exponent, defined as: H = log (R/S)/ log (∆t), is also a measure of persistence (H > 0.5) and anti-persistence (H < 0.5) in statistical time series.Persistence is related to long memory in time series, meaning that an increase in values is most likely be followed by another increase; while anti-persistence (H < ½) relates to short-term memory or return to the mean, meaning that an increase will most likely be followed by a decrease, and vice versa.An H = 0.5 is taken to mean randomness as in Brown noise spectrum. Interestingly, the rescaled range and segment sizes follow a power law, and H is its exponent.The intensity of fluctuations in anti-persistence mode increases as H moves closer to zero; hence, its connection to frequency. 3-THE RESULTS The results are summarized in Table 1, while Table 2 provides some additional information that might be apparent in Table 1.As it can be seen, the power spectrum that describes how much different frequencies contribute to the average power of the signal, fluctuates from segment to segment.Values between 1 and 2 reflect anti-persistence in the index, and values between 2 and 3 reflect long-term memory or persistence.Readers interested in knowing the wave length of the memory are referred to the excellent paper by Peters (1991).Thus, over the whole period, 4 segments reflect anti-persistence and three reflect persistence. The interesting observation for the present purpose is that the power spectrum lying between 1 and 2 reflects dark-pink noise spectra; and that implies thereby a deterioration in income distribution. While values lying between 2 and 3 reflect dark-brown spectra coinciding to improvement. Interestingly, over the whole range of the data considered, no brown noise was detected; and we note in passing that this should be significant for studies based on Brownian motion. For an additional verification, we consider some values obtained from another measure, called the Gini Index.The latter has a few interpretative limitations.For example, it measures relative income; thus two countries could have the same Gini value and yet are very different in terms of economic status.Or the Gini index may exceed a value of 1.0 when some individuals make a negative contribution to the total income, etc.However, these limitations do not apply in the present case.In essence, a value of zero in the Gini Index reflects perfect equality, while a value of 1 reflects perfect inequality.Now consider how the Gini Indices of the US economy vary over time.Over the period 1961-72, the S&P-500 Index was in persistence mode with a value of  = 2.044; the Gini index went down from 0.52 in 1961 to 0.42 in 1972.In contrast, from1972 to 1983, when the S&P-500 was in antipersistence mode, the power spectrum was 1.4, probably due to changes in the status of the US dollar, war, and the oil shock.We would then expect a deterioration in the Gini index over that period.That is what happens; the Gini coefficient went from 0.42 to 0.46.During the whole period 2003-2011, the system was again in anti-persistence mode, the Gini coefficient again went from 0.51 to 0.53.During the brief period 1998-2002, when the S&P-500 index was in persistence mode, the Gini coefficient remained at 0.50; that is the only glitch observed.But from 2003 to 2008, it increased from 0.49 to 0.50.From 2007 to 2008, the Gini index increased from 0.50 to 0.52.Thus, during the economic meltdown from 2007 to 2011, the index was in anti-persistence mode and the Gini coefficient increased from 0.50 to 0.53.If we were to examine the Gini indices for other countries, we would most likely observe a similar situation, except where it is mitigated by equality-like policies. Turning now to the correlation dimension D2 in 2-D, it remained between 2 and 3 over the whole period under study, implying that the S&P-500 was never a random process.The other interesting result is that the correlation dimension (which detects probabilistic structure among variables) of each segment is a non-integer, implying that its dynamics should show a countable set of periodic orbits of arbitrary long periods, and an uncountable set of non-periodic orbits.Such a situation might appear random to the naked eyes, but in fact the process is deterministic.Furthermore, when the system was in anti-persistence mode, D2 increased, and it decreased in persistence mode, as can be seen in the last column of Table 1.This means that there is a sort of phase shift that occurred at the fold at H = ½.This is explained by an enlargement or a shrinking of the singular spectrum at the values of the b's in Appendix A. For example, during the period 1961-72, the process was in persistence mode, while during the period 1972-83, it was in anti-persistence mode. Consequently, the information dimension D1 went from 2.4192 to 2.7083; D0 went from 2.4780 to 2.7791, and so on.That is, the size of the attractor increases in anti-persistence mode and decreases in persistence mode, as we would expect.Further verifications come from similar studies as in Medio (1992, 130).He found that for 3-D fractal attractors of continuous-time dissipative systems, the non-integer fractal dimension is between 2 < D2 < 3, as found in the last column of Table 1.To take yet another example, consider the findings of Edgar Peters (1991) who used the Grassberger/Procaccia procedure to compute D2 of the S&P-500 Index, sampled at monthly intervals from January 1980 to July 1989.He found that the embedding dimension was 2.33.Even though we do not have the same series' length nor the same sampling interval, nevertheless this study arrives a value of 2.3345 for the period 1983-87.This might be due the fact that D2 remained constant over the time interval, or due to the fact that segment as well as all the others were filtered for white noise prior to the analysis since both methods are sensitive to noise, or due to a combination of both; at any rate, this kind of concordance in that statistic is rather rare in economics. CONCLUDING REMARKS Our initial contention was that wealth and income inequalities in market economies are too ubiquitous and systematic not to be driven by some power law.To verify that assertion, the S&P-500 Index, sampled daily over a span of 50 years, was examined.It was found that the index varies from anti-persistence to persistence modes during the period studied.Consequently, its noise spectrum varies from dark-pink, when the power spectrum was between 1 and 2, to dark-brown, when the power spectrum lied between 2 and 3. On the assumption that the Index is the output of a multifractal, its singularity spectrum, including the correlation dimension of each segment, was also computed using the method proposed by Mandelbrot.The value computed for the correlation Table 1 : The power Spectrum and the Correlation Dimension of the S&P Index: 1961-2011 The symbols  and  indicate improvement or deterioration in ordinal space or increase and decrease in real space.D2 indicates the frequency of orbit's visits to different subspaces in the attractor.	5,174.4	2018-01-24T00:00:00.000	[ "Economics" ]
Efficient algorithm to compute the Berry conductivity We propose and construct a numerical algorithm to calculate the Berry conductivity in topological band insulators. The method is applicable to cold atom systems as well as solid state setups, both for the insulating case where the Fermi energy lies in the gap between two bulk bands as well as in the metallic regime and interpolates smoothly between both regimes. The algorithm is gauge-invariant by construction, efficient and yields the Berry conductivity with known and controllable statistical error bars. We apply the algorithm to several paradigmatic models in the field of topological insulators, including Haldane's model on the honeycomb lattice, the multi-band Hofstadter model and the BHZ model, which describes the 2D spin Hall effect observed in CdTe/HgTe/CdTe quantum well heterostructures. I. INTRODUCTION Topological insulators (TI) are a topological state of quantum matter which constitutes a new paradigm in condensed matter physics [1][2][3][4] . These recently discovered new materials exhibit unique fascinating properties such as currentcarrying surface or edge states that are strongly protected against perturbations in either the bulk or the surface of the material 5-10 and non-standard exchange statistics of quasiparticle excitations, which offer potential applications in the context of quantum computation [11][12][13] . The question what happens in topological insulators when the Fermi energy does no longer lie inside the gap between two energy bands, is by no means rhetoric but of high practical importance: in fact, this situation naturally occurs in the experimental process of production of candidate samples of topological insulators such as Bi 2 Se 3 and Bi 2 Te 3 compounds. These are used for instance in cooling devices due to their favorable thermoelectric properties. The chemical composition can be well-controlled and adjusted to the one of the desired topological insulator. However, it is much more demanding to control the level of the Fermi energy, which for many samples lies within the bulk energy bands instead of the insulating energy gap, thereby invalidating them as true TIs. This difficulty has motivated the development of sophisticated molecular beam epitaxy (MBE) techniques to precisely control the growth of ultra-thin Bi 2 Se 3 and Bi 2 Te 3 films 14,15 . Likewise, in two-dimensional TIs it is possible to adjust the Fermi energy to lie either in the band gap or the bulk bands. Experimentally, in CdTe/HgTe/CdTe quantum wells, formed by a thin layer of HgTe embedded between two CdTe layers, this can be achieved by an elaborated MBE technique that allows one to control the thickness of the intermediate HgTe layer and thereby tune the position Fermi energy with respect to the bands 16,17 . For an appropriate thickness, the Fermi energy lies in the gap between the bulk bands and the heterostructure shows the desired characteristic topological insulating behavior with a quantized spin conductivity of 2e 2 /h. Complementary to solid-state realizations, cold atoms in optical lattices have been proposed as a realistic plat-form to experimentally explore the new physics of TIs under controllable conditions [18][19][20][21][22][23][24][25][26][27][28][29][30][31][32][33][34] . In particular, in these systems the Fermi energy can be controlled directly by the filling of atoms in the lattice. In contrast, in condensed matter systems such as the above-mentioned chemical compounds the pinning of the Fermi level to a value inside the bulk bands typically arises due to external causes like crystal defects and other sources which are not straightforward to control. As a consequence, in transport properties and measurements bulk carriers often dominate over the contribution stemming from surface or edge states. Finally, this question plays as well a fundamental role in the physics of the anomalous quantum Hall effect (AHE) 35,36 , which precedes the upsurge of topological insulators as a prominent field in condensed matter. In the standard quantum Hall effect (QHE), which can be observed in non-magnetic materials, there is a linear dependence of the Hall resitivity ρ x y on an externally applied perpendicular magnetic field. In contrast, in the AHE an anomalous deviation from the linear law is observed in ferromagnetic materials. A complete theory for the AHE has remained elusive for more than a century, largely due to the complications arising from the fact that there are three main mechanisms that influence the electronic motion and can give rise to an AHE. Here, we shall be interested in the so called intrinsic mechanism for the AHE, which is the contribution that can be expressed in terms of the Berry-phase curvature and thereby represents an intrinsic quantum mechanical property of a perfect crystal. This intrinsic contribution, which is dominant in metallic ferromagnets with moderate conductivity, depends only on band structure properties and is largely independent of scattering that affects other AHE mechanisms. Understanding of this intrinsic and anomalous contribution has become possible with the seminal work by Haldane 37 who uncovered by a fully quantum-mechanical treatment, unlike precedent work based on semiclassical methods 38 , the topological origin of this contribution and its relation to the physics taking place at the Fermi surface. Haldane showed that the intrinsic contribution to the AHE conductivity stems from a combination of an integer-valued FIG. 1. a. Generic energy spectrum of a system with an energy gap ∆. In the displayed situation the Fermi energy falls into the first energy band and defines the Fermi surface as the equipotential energy line at E α (k) = E F (solid line). The projection of the energy dispersion of the first band is shown as a color-coded plot in the horizontal k x − k y -plane. b. For the numerical calculation of the Berry conductivity, the Brillouin zone is discretized by a finite grid. Momentum space plaquettes with energies E α (k) entirely below (above) the Fermi energy contribute entirely (not at all) to the Berry conductivity, whereas plaquettes which cut the Fermi surface contribute partially. c. Schematic summary of the numerical algorithm to calculate the Berry conductivity: After fixing the discretization grid of momentum space and calculating the Berry curvature contributions by means of the FHS algorithm for each plaquette of the Brillouin zone, a classical Monte Carlo sampling method is used to determine the weights with which the individual plaquettes contribute to the conductivity. Statistical uncertainties in the sampling process result in controlled and statistical errors in the Berry conductivity. part stemming from the contribution of filled bands and a part originating from the Fermi surface, i.e. from the cuts of a partially filled band at the Fermi energy E F (non-integer valued contribution). It is crucial to realize that in order to directly apply Haldane's equations 37 to a given problem, one needs to know precisely the form of the Fermi surface. In practice, except in very simple model cases, this is not possible since the band structure of real materials is obtained from detailed numerical calculations and one is typically given a numerical data set about the bands instead of an explicit formula. Thus, in practice it is highly desirable to have at one's disposal a numerical method, which is (i) gauge-invariant, (ii) efficient and (iii) outputs numerical results with controllable and known error intervals. In this work, we develop such a general and efficient numerical algorithm to compute the Berry conductivity when the Fermi energy does not lie within the band gap. In the following, we shall refer to Berry conductivity as the non-quantized conductivity associated to the Chern number according to the TKNN formula 39 when the position of the Fermi level lies in the conduction band, so that we recover the TKNN quantized conductivity for the standard insulating case if the Fermi energy lies in the energy gap between two bands. Our main results are: i/ We present a new method to compute the Berry conductivity when the Fermi energy level is located outside the band gap. We outline the algorithm (schematically summarized in Fig. 1), discuss its ingredients and show that it is gauge invariant and efficient (Sec. II). ii/ We emphasize that a central feature of the presented method is that it is endowed with known and controllable error bars for the non-integer value of the conductivity. This is essential since when the Berry conductivity is not integervalued, errors due to approximations need to be under control in order to distinguish two different values of the conductivity observable, so that one safely distinguish a topological phase from a trivial phase. iii/ To test and benchmark the performance of the algorithm we first apply it to the paradigmatic Haldane model 40 , which has a simple enough structure so that the analytic form of the two-band energy spectrum is known (Sec. III A). Subsequently, we apply the method to the more complex case of the Hofstadter model 41 , which belongs to the class of multi-band topological insulators, where the band structure information is obtained numerically (Sec. III B). These models are both of importance and have attracted interest in the field of quantum simulation of topological insulators with cold atoms in optical lattices. Here, our method provides the theoretical tools that allow one to map out the phase diagrams in future experiments. Finally, we also apply our method to the BHZ model 16 which is a realistic model which captures the physics of 2D spin Hall effect present in systems such as the above-mentioned CdTe/HgTe/CdTe quantum well compounds (Sec. III C). We conclude with a short summary and a discussion of possible future extensions of the presented method (Sec. IV). A. Generalized Berry conductivity Before presenting our numerical algorithm to calculate the Berry conductivity, in this section we briefly review the expressions for the intrinsic Hall conductivity both for the insulating case where the value of the Fermi energy lies in the gap between two bands, as well as the generalized result for the situation in which the Fermi energy falls lies in a partially filled band 37 . In the insulating case, the Hall conductivity is quantized and proportional to the sum of the Chern numbers of the occupied energy bands, The Chern numbers C α are integer-valued topological invariants, defined in terms of the integral of the Berry curvature F α x y (k) over the whole Brillouin Zone (B.Z.) 39,42 : The latter is expressed by the exterior derivative of the Berry connection where u α (k) is the eigenvector corresponding to the energy band E α (k). In the case that the Fermi energy does not lie in an energy gap between bands, as schematically shown in Fig. 1a, the intrinsic Hall conductivity generalizes to 36,43 with where Θ(E ) denotes the Heaviside function. Thus, the conductivity is the sum of the integer-valued Chern numbers corresponding to fully-occupied energy bands below the Fermi energy and a non-quantized contribution which depends on the Fermi surface, i.e. it stems from the integral over energy band(s), which are partially filled at a given Fermi energy E F . For systems with a particularly simple band structure, as e.g. in two-band systems, the expressions for the eigenenergies and eigenvectors of the bands are given in explicit form, and hence the Chern values C α can be calculated analytically. In general, however, the system Hamiltonian cannot be diagonalized analytically and an efficient numerical method to compute the Chern values is needed. B. Construction and properties of the algorithm The algorithm we propose to numerically compute the Chern values of Eq. (12) and thereby the Berry conductivity of Eq. (4) is based on a series of controlled approximations: First, we discretize the two-dimensional Brillouin zone by a finite n B × n B grid of small plaquettes at discrete momenta k l (see Fig. 1b and Appendix A for details), so that the integral over the (partially filled) band becomes with the Berry curvature contribution from a small two-dimensional plaquette of size ∆ k x ∆ k y , and the weighting factors The weights p α l (E F ) correspond to the partial area of the plaquette, which is covered by the Fermi sea, thus p α l (E F ) = 0 (p α l (E F ) = 1) for squares with energies completely above (below) the Fermi energy E F , and 0 < p α l (E F ) < 1 for momentum space plaquettes which are cut by the Fermi surface (see Fig. 1b). The choice the value n B , i.e. the resolution of the momentum space grid, is important: it can be motivated either by given physical conditions, such as a finite experimental energy resolution or e.g. the finite size of real-space optical lattices, which in turn induces a smallest characteristic scale in momentum space; or it can be chosen according to given numerical resources. The key of the numerical algorithm is now to evaluate reliably and under controlled approximations the discretized sum of Eq. (6), whose value converges to Eq. (12) for increasingly finer grids. (i) Gauge-invariant calculation of the Berry curvature: To numerically calculate the Berry curvature contributions F α x y,l we employ a numerical algorithm proposed by Fukui, Hatsugai and Suzuki 44 (FHS algorithm). It is highly efficient and the discrete sum 1/(2πi ) {k l }F α x y,l converges rapidly to the correct integer-valued Chern numbers C α , even for a very coarse-grained discretization of the Brillouin zone. This behavior is rooted in the fact that the algorithm is based on a lattice gauge formulation 45,46 instead of a finite difference discretization of the Berry curvature. In Appendix A we provide a brief summary of the FHS algorithm and the explicit expressions for the lattice strengthF α x y,l calculated with the FHS method. (ii) Efficient estimation of the weights p α l (E F ): To decide whether a given plaquette in momentum space contributes entirely, partially or not at all, we use a simple and rapid classical Monte-Carlo technique: for each plaquette of the grid localized around the discrete momentum k l , we generate n R uniformly distributed random points k R and compute E α (k R ) which takes the discrete values zero or one. Based on the latter we define the estimator for the weighting factors p α l (E F ). (iii) Statistical confidence interval and controlled numerical error of the Berry conductivity: Note that the randomness of this estimation procedure introduces a statistical uncertainty. Note that the value of the estimatorsp α l (E F ) is bounded between zero and one. However, it is clear that the statistical error will be largest for partially contributing plaquettes withp α l (E F ) ∼ 1/2, whereas the uncertainty inp α l (E F ) for plaquettes with energies completely above or completely below the Fermi energy is expected to be much smaller. In order to have a known and minimal statistical error inp α l (E F ), and thus in the Berry conductivity, it is highly desirable that the numerical algorithm takes this effect into account and provides statistical errors which depend on the actual value of the Fermi energy. The quantityp α l (E F ) is the estimator of the fixed though unknown parameter p of a binomial distribution B(n R , p), corresponding to the process of tossing n R times a biased coin. As is discussed in detail in Appendix B, using the normal approximation and for a fixed number of runs n R and a desired value < 1 this allows one to derive a confidence interval [p α l ,min , p α l ,max ] for p α l (E F ), called the Wilson interval 47 with modified boundary conditions. This means that with a probability 1 − the "true" value p α l lies in this interval. The key point is that the width of this interval depends on the actual value of the estimatorp α l and is typically significantly smaller than the trivial upper bound of one. After symmetrizing the interval by taking the maximum ∆p α l (E F ) = max(p α l − p min , p max −p α l ), each momentum space plaquette of the grid is associated to a probability valuep l (E F ) ± ∆p l (E F ) with a confidence of at least 1 − . As mentioned above, for even moderately fine grids the FHS algorithm provides essentially exact values for the Berry curvature contributions (see 44 and Appendix A). Thus, the statistical uncertainty ofp α l directly translates into an uncertainty in the Berry conductivity contributions, Finally, the estimated Berry conductivity is given bỹ with an error ±∆C α (E F ) of with confidence of at least 1 − . We remark that controllable error bars in combination are particularly important and valuable outside of the insulating regime, i.e. where the Fermi energy cuts a partially filled energy band, as in this case the Berry conductivity is not quantized and can assume continuous non-integer values. III. PRACTICAL APPLICATION OF THE ALGORITHM In this section, we apply the algorithm to different models. We first start with the Haldane model, a two band model that can realize both topological and trivial phases. We then go to the Hofstadter model, a multi-band model characterized by non zero Chern number and finish with the BHZ model, a two band realistic model realizing a quantum spin Hall effect in condensed matter physics. A. The Haldane model The model proposed by Haldane in 40 is a tight-binding Hamiltonian of spinless fermions on a honeycomb lattice, with dynamics governed by nearest-neighbor (N.N.) realvalued hopping term of amplitude J and an imaginary nextto-nearest neighbor (N.N.N.) hopping term J 2 (see Fig. 2a). In addition, the fermions are exposed to an onsite staggering potential β, which induces a chemical potential difference between nearest-neighbors sites of the bi-partite hexagonal lattice (φ and ψ sites). The model is exactly solvable and represents a paradigmatic model in the field of topological phases of matter, as it hosts a quantum AHE phase even in the absence of an external magnetic field. Recently, it has been proposed that the physics of this model could be observed experimentally in a quantum simulation with cold atoms in optical lattices 34 . The Hamiltonian of the system is given by Here, c † i and c i are fermionic creation and destruction operators, ν i j = sgn[(d 1 × d 2 ) z ] and s φ,ψ = ±1. The vectors d 1 and d 2 are oriented along the bonds of the hexagonal unit cell, as shown in Fig. 2a. The model can be readily solved by rewriting the Hamiltonian in terms of two-site basis cells (φ, ψ) (see e.g. 48 ) such that the hexagonal lattice becomes a triangular lattice of (φ, ψ) cells. In the Fourier space the Hamiltonian is then given by 40 is expressed in terms of the vectors between nearest neighbors δ 1 , δ 2 and δ 3 and f (k) = sin[a 1 ·k]+sin[a 3 · k]+sin[(a 1 +a 2 )·k] is expressed in terms of the lattice vectors a 1 and a 2 as shown in Fig. 2 and defined in Appendix C. Diagonalization of the Hamiltonian readily yields the two-band energy spectrum which is shown in Fig. 2b. For β = J 2 = 0, the Hamiltonian corresponds to pure nearest-neighbor hopping of fermions with the characteristic spectrum exhibiting the two inequivalent Dirac cones 49,50 . A non-zero staggering potential β = 0 induces an imbalance of the fermion density on φ and ψ lattice site. The formation of a charge-density-wave phase is associated to the opening of a topologically trivial insulating gap in the spectrum. On the other hand, a strong enough N.N.N. hopping term J 2 opens a topologically non-trivial energy gap that signals the transition of the system into a AHE phase characterized by a non-zero Chern number. The size of the energy gap is determined by the formula ∆ = 2\|β − 3 3J 2 \|, and for \|β\| < 3 3\|J 2 \| the system is in the topological phase. We will now illustrate the working principle of our algorithm by applying it step by step -as schematically summarized in Fig. 1c -to the Haldane model. To this end, we start by fixing the Hamiltonian parameters to J 2 = 0.1J , β = 0, i.e. deep in the topologically non-trivial phase. Next, we discretize the Brillouin zone (step 1), where we use for numerical convenience a rectangular-shaped B. Z. parametrization which is equivalent to the standard hexagonal form (see Appendix C for details). Then, we compute the field strengthF x y for each plaquette (step 2); the result is shown in the right column of Fig. 3. We fix the number of random points (we choose n R = 20) (step 3) and compute for each plaquette for n R randomly distributed momentum vectors E α (k R ) (step 4). Once the Fermi energy is fixed (step 5), here to a value of E F = −1.5J values of n B , illustrating how an increasingly finer grid of the Brillouin zone leads to an increased resolution and numerical precision. Finally, the estimated weightsp α l (E F ) and the Berry curvature contributionsF x y,l are combined to calculate the Berry conductivity (step 8) according to Eqs. (11) and (12) with an associated error bar (step 9) as given by Eq. (13). By applying the algorithm again for varying values of the Fermi energy, the Berry conductivity can be obtained as a function of the Fermi level energy. The obtained Berry conductivity is shown in Fig. 4a: starting from low conductivity values at the bottom of lower energy band, the conductivity increases up to its plateau value of one for Fermi energies lying in the topological insulating gap, before it subsequently starts to fall off again once the Fermi energy reaches the upper band. To test the behavior of the algorithm when the system undergoes a phase transition from the topological AHE phase to the trivial insulating phase, we increase the Hamiltonian parameter β to observe the competition of the N.N.N. hopping term with the staggering potential. The subplots in Fig. 4 show the transition from the topologically non-trivial phase characterized by a Chern number of one to the topologically trivial charge-density-wave phase with a vanishing Chern number. The algorithm correctly captures the closing of the gap as well as the jump of the conductivity plateau-value as the phase transition takes place. We emphasize that the algorithm automatically takes into account the fact that at the phase transition the Berry curvature is highly localized at the Dirac points and thus concentrated in only few plaquettes -a fact that the algorithm signals in the form of larger error bars of tem resides in the topological phase with a small topological gap opened. Here, the algorithm allows one to clearly verify numerically the 1/E F power law dependence of the Berry conductivity for Fermi energies close to the gap. The σ B e (E F ) = (e 2 /h) 3 3J 2 /\|E F \| behavior is predicted by the linear approximation of the spectrum around the Dirac points 43,51-53 . The results are shown and discussed in Fig. 5. B. The Hofstadter model Let us now apply the numerical algorithm to the Hofstadter model 41 , which describes spinless fermions on a square lattice, subjected to a uniform magnetic field of magnetic flux quanta per unit cell Φ. Only very recently, several groups have achieved to observe the characteristic physics including the fractal spectrum known as Hofstadter's butterfly in graphene superlattice systems [54][55][56] . This is complementary to ongoing experimental efforts to realize theoretical ideas 57 on how to implement the fermionic Hofstadter Hamiltonian with cold atoms in optical lattices [58][59][60][61] . The Hamiltonian in second-quantized form is given by where the sum is over nearest neighbor sites (see Fig. 6) and the phase factor exp(i θ i j ) corresponds to the Peierls substitution expressed in terms of the line integral over the vector potential along the link between two neighboring sites i and j of the square lattice. If Φ = p/q is a rational number, the energy spectrum of the bulk, described in the Fourier space, splits into q sub-bands, each one of them associated with a non-trivial integer-valued Chern number. Due to its multi-band structure the Hofstadter Hamiltonian can in general not be diagonalized analytically and thus represents an interesting testbed for the numerical algorithm. Fig. 7 shows the numerical results for the Berry conductivity for different values of the flux per plaquette (Φ = 1/3, 1/5 and 1/7). For Fermi energies lying in the energy gap between bulk bands, the algorithm correctly reproduces the constant Berry conductivity, which corresponds to the sum of the Chern numbers of completely filled bands. Once the Fermi energy falls into a bulk band the Berry conductivity is no longer quantized. Whereas for Φ = 1/3 the Berry conductivity interpolates monotonically between the gap plateau values, for Φ = 1/5 the conductivity displays an interesting feature for Fermi energy values in the second band: instead of showing a monotonic growth, it first decreases to a minimum value, before starting to increase until it reaches the plateau dictated by the quantized value of the conductivity in the gap. The same phenomenon occurs, even more pronounced, in the third band of the spectrum for Φ = 1/7. The small controlled statistical error bars of the numerical method ensure that the non-monotonic signature in the Berry conductivity is indeed a physical feature rather than a numerical artifact. C. The BHZ model In 2005, it was suggested that the quantum spin Hall effect (QSHE) could possibly be observed in graphene 51,62 , which however turned out to be impeded by too weak spin-orbit coupling in this system. Shortly later, a realization of the QSHE in HgTe/CdTe nanowell structures was proposed 16 and experimentally realized only one year later 17 : by varying the thickness of the different layers of the heterostructure, the material can exhibit a trivial insulating phase as well as a topological insulating phase, characterized by a Z 2 topological invariant. The physics can be described by an effective Hamiltonian valid close to the Γ point, derived by Bernevig, Hughes and Zhang (BHZ model) 16,63 . The Hamiltonian is a given by 4 × 4 matrix in momentum space, , where 1 is the two-dimensional identity matrix, σ i denote the Pauli matrices and The parameters A, B , C , D and M depend on material properties as well as the thickness of the layers and can be computed numerically 16,63 . The Hamiltonian decouples into 2×2 blocks, and the spin conductivity can be written as the difference of the conductivity for each spin orientation and it makes thus sense to study the conductivity of one of the orientations. Here, we apply our algorithm to the BHZ model with parameters as calculated in 16 . Figure 8a shows the energy spectrum that exhibits a small gap of 0.01eV , which renders the computation of the Berry conductivity in the non-insulating regime more demanding. Figure 8b -d shows the numerical results for the Berry conductivity for increasingly finer grids of the Brillouin zone. Whereas even for the roughest grid studied (n B = 40) the algorithm correctly captures the qualitative behavior and the conductivity minimum value value of -1 for the Fermi energy lying in the shallow energy gap. However, as signaled by considerably large error bars, only few plaquettes contribute large values of Berry curvature to the conductivity. Thus, finer grids (see Fig. 8c and d with n B = 160 and n B = 320) are required to quantitatively correctly describe the conductivity behavior in the vicinity of the gap. This effect illustrates the importance of a high enough resolution, both numerically and in an experiment. As the algorithm qualitatively captures the behavior even for rather coarsegrained grids, this can be helpful to predict observations in the case of restricted experimental resolution, e.g. originating from limited sizes of optical lattices for cold atoms, or finite temperature constraints in solid state experiments. Finally, we remark that the BHZ model is an effective model valid close to the Γ point, and thus the results of our analysis are also only valid in the vicinity of the energy gap. It is possible and will be an interesting extension of the present work to apply the numerical method to a more realistic, refined model which incorporates more information about the band structure of the system. IV. CONCLUSIONS AND OUTLOOK In this work we have proposed and constructed a numerical algorithm to calculate the Berry conductivity in topological band insulators. The algorithm works for both the insulating case where the Fermi energy lies in the gap between two bulk bands as well as the situation where it lies within a band. The algorithm is gauge-invariant by construction, efficient and outputs the Berry conductivity with known and controllable error bars. We have successfully applied the algorithm to several paradigmatic models of topological quantum matter, including Haldane's model on the honeycomb lattice 40 , the multi-band Hofstadter model 41 and the BHZ model 16 that describes the 2D spin Hall effect observed in CdTe/HgTe/CdTe quantum well compounds. In addition to its applicability to topological insulators, the numerical method to compute the Berry conductivity for arbitrary values of the Fermi energy level can be applied to several other important problems: It can be used to study new phases of matter such as topological Fermi liquids 37,64,65 which arise in interacting systems of fermions that realize a TI phase or an AHE phase. Mean field methods applied to these systems predict the existence of such phases 48,66,67 . Here, the efficient and controllable numerical method for computing the Berry conductivity provides the appropriate observable to map out the possible topological phases of those systems with the desired accuracy [68][69][70] . Recent experiments in which topological insulating phases 58,61 have been quantum simulated with cold atoms in optical lattices, provide another natural scenario where our new algorithm can be applied. Complementary to condensed matter systems, these experimental setups offer the possibility to study the intrinsic Berry conductivity in AHE systems under particularly clean and controllable conditions. Here, our algorithm can provide a precise observable to reliably and quantitatively distinguish symmetry protected topological phases from trivial phases and can predict some interesting features within the energy band. In fact, there have been proposed several ways to measure characteristic signatures of topological quantum phases in systems of cold atoms [71][72][73][74][75] . An experimentally useful extension of our work would be to generalize our numerical method to the case of three dimensional topological insulators under time-reversal symmetry protecting conditions. Finally, it is an interesting question how to generalize the controlled numerical method to an open quantum system scenario, such that it can be applied to topological insulators and topologically ordered systems coupled to an environment [76][77][78][79][80] . The continuous Brillouin Zone is discretized by a twodimensional lattice grid of n B points in each direction. For simplicity, we focus here on a rectangular grid, but the formalism can be readily extended to any polygonal grid 46 . The plaquettes of the momentum space lattice are then given by with 0 ≤ i , j ≤ n B − 1, The lattice field strengthF α x y (k l ) of band α on the grid is then defined in terms of the link variable U µ (k) as If the admissibility condition \|F α x y (k l )\| < π is satisfied 44,46 , the lattice gauge theory corresponds to the continuous gauge theory 44,46 and one can write: Based on these Berry curvature contributions, the Chern number can be computed as Appendix B: Choice and the computation of the statistical error In this section, we present the concept and the details of a confidence interval (C.I.) to characterize the statistical uncertainty of the estimated weightsp α l , as defined in Eq. (9). For simplicity of the notation, we suppress the band index α and momentum index l in the following. The problem of estimating the weights corresponds to determining the unknown, though fixed probability value p of a binomial distribution B(n R , p), based on the outcome of n R trials. The probability to observe k of the n R enquiries the value +1 is given by The goal is to associate a C.I. of a width much smaller than one to the estimated valuep, such that the true value p lies with a probability 1− inside the C.I. There are several ways to define the C.I., and we will in the following outline the advantages and inconveniences of some of them to motivate the necessity to adopt a simple and appropriate one that we use in our algorithm. To characterize and compare the quality of different conventions for the C.I, it is convenient to introduce the coverage probability: it corresponds to the effective probability to be inside the C.I. and can be compared to the expected probability 1 − . As a guiding principle, a "good" C.I. is an interval with p cov 1 − . On the contrary, for p cov < 1 − , the C.I. is "bad" as the statistical "guaranteeing functionality" of the interval fails. The other case p cov > 1 − is not dramatic in our context as this implies that the true value of the estimated quantity p actually lies in the C.I. with a probability even higher than the targeted value of 1 − . The construction of C.I. is based on the central limit theorem, which can be used to prove the convergence of the Binomial distribution to a normal distribution N , in our case: The central limit theorem and the definition of the C.I. of a normal distribution with an expected probability 1− permits us to write the C.I. ofp l as a self-consistent equation in terms of p: where z is the quantile function of the normal distribution 81 . A first way to define a C.I. is by maximizing the second term of the sum, yielding for p = 1/2. This relation highlights the typical 1/ n R dependence of the statistical error and can be used to provide a rough estimate of the size of the C.I. in terms of n R . However, as the length of the interval does not longer depend on the estimated valuep l itself, it does not satisfy our requirement. It will have a coverage probability p cov > 1 − and would output error bars that overestimate the actual uncertainty of the observable of interest. Another commonly used C.I. is constructed using the approximation p(1− p) p l (1−p l ) in Eq. (B3), thereby replacing the unknown "true" value by the estimator value, so that This C.I. is known as the Wald interval 47 . Despite its simplicity, this convention suffers from several problems: for p l 0 orp l 1, the Wald interval shrinks to zero, implying a bad a coverage probability for p-values close to one or zero. As discussed by Brown et al. 47 , a series of criteria has been used in the literature to test the region of validity of this C.I. However, these criteria can be misleading and do not always characterize correctly the C.I. In Fig. 9a we illustrate this problem for a fixed value of n R = 40 and by computing the coverage probability of the C.I. in terms of the value of p on 10000 samples. One notices at first glance the tendency of the curve to lie below the expected value of 1 − . Although the C.I. works rather well for values of p close to p = 0.5, it captures only poorly the situation at values close to the boundaries. Finally, the curve has a fast and significant oscillating behavior which gives rise to the phenomenon of so-called lucky/unlucky numbers: when increasing slightly the probability p, the coverage probability jumps from a good p cov to a poor p cov value as it is the case for instance around p = 0.8 in the shown example. The couple (p, n R ) defines the lucky/unlucky numbers. In Fig. 10a we fix the value p = 0.25 and vary the value of n R . Here one observes also significant fluctuations that are only stabilized at larger values of n R . This effect becomes is even more striking at small p, as illustrated in the Fig. 10 c. where a fixed value of p = 0.007 has been chosen: under an increase of n R , the C.I. seems to converge to a favorable value of p cov until reaching n R = 423 where p cov suddenly drops from 0.94 to 0.78. We thus exclude the Wald interval as a candidate to construct the C.I. for thep α l estimators in our algorithm. Most of the mentioned problems can be avoided if the approximation p(1 − p) p l (1 −p l ) is not applied in Eq. (B3). Instead, one can exactly solve Eq. (B3), which is a quadratic equation forp. This yields the so-called Wilson interval 47,82 As illustrated in the Fig. 9b, the Wilson interval is much more stable and the coverage probability is oscillating around the value 1− . Figure 10 b. shows that the Wilson interval reaches rapidly and in a stable way the expected value 1 − . The only problem still to be cured is at the boundaries, at p-values around zero or one, where the coverage probability drops. Figure 10 d. illustrates the convergence at small p, here fixed to p = 0.007 and indicates that the effect of lucky/unlucky numbers is much less important than for the Wald interval. Brown et al. 47 propose to replace the lower (upper) boundary of the C.I. obtained by the normal approximation by a lower (upper) boundary obtained from a Poisson approximation for small (big) values ofp. This indeed stabilizes the behavior of the C.I. even close to the boundaries but complicates the expression of the C.I. Here, we propose a simpler patch, which has the same desired ef- fect: we use the following replacement: including x = 3 and x = n R − 3 if n R > 40. Finally, merely for convenience to obtain symmetric error bars, we symmetrize the C.I. aroundp by choosing a width which corresponds to twice the value of max{p max −p,p − p min }. While keeping the C.I. narrow, this only leads to a modest over-estimation of the actual uncertainty of the estimator. The C.I. interval defined in this form has a simple analytical form in combination with a good coverage probability, even for small n R 47 . We will use this construction of the C.I. in the Monte Carlo sampling part of the algorithm, and refer to it as Wilson interval with modified boundaries in the main text. In this section, we illustrate the importance of an appropriate momentum space resolution, parametrized by the discretization number n B . Figure 12 presents a zoom of Fig. 4c of the main text, showing the numerically estimated Berry curvature for different grids n B . One finds that all graphs have the same behavior until reaching a value around E F = −0.33. There, the behavior of the estimator of the Berry curvature becomes jerky. This is a characteristics which shows up when some few momentum-space plaquettes have an important Berry curvature contribution. The error bars are signal this effect. The situation improves for increasing values of n B : the curves converging to one sharp curve, showing that the main contribution of the Berry curvature is stems from states with an energy close to zero, and the error bars decrease significantly. FIG. 13. Numerically estimated Berry conductivityσ Be with error bars for two values of the number of random points n R = 20 and 160 for a system with J 2 = 0.1J , β = 0.5J and n B = 20. As expected, the computation with n R = 160 is much more precise, resulting in significantly smaller error bars. Note that as desired the n R = 160 curve is entirely comprised in the region spanned by the error bars of the computation with n R = 20. Another way to reduce size of the error bars is to increase the value of n R , the number of random points used to computep l in each plaquette. Figure 13 displays the estimated Berry conductivity for a fixed value n B = 20 and for the two values n R = 20 and n R = 160. As expected, the curve for n R = 160 is much more stable than the curve obtained for n R = 20: we see here a better interpolation in terms of the Fermi energy at this resolution. We emphasize the fact that the curve corresponding to n R = 160 is contained completely in the region spanned by the error bars of the n R = 20. This is an important point of the chosen construction of the confidence interval, as described in Appendix B. Appendix E: Importance of the choice of the error bars In this section, we compare the Wilson interval with modified boundaries with the C.I. defined in Eq. (B4) by examining the final error interval obtained in the Haldane model using both methods. We work here with J 2 = 0.1J , β = 0.5J such that the Berry curvature is really sharp and localized. The Figure 14 shows the results for both types of error interval with the parameters n B = 20, n R = 100. The error interval as obtained by using the Wilson interval (see Appendix B) captures correctly the fact that the main contribution to the Berry curvature is strongly localized in momentum space. This gives rise to an increased statistical error in the energy region in which the Fermi energy crosses plaquettes with a large contribution to the Berry curvature. The error obtained with the other C.I. , presented in Fig. b., is constant and independent of the value of the Fermi en-	9,895.2	2014-01-08T00:00:00.000	[ "Physics" ]
Low Energy Implantation of Carbon into Elastic Polyurethane Ion modification of polymeric materials requires gentle regimens and subsequent investigation of mechanical and deformation behavior of the surfaces. Polyurethane is a synthetic block copolymer: A fibrillar hard phase is inhomogeneoulsy distributed in a matrix of soft phase. Implantation of carbon ions into this polymer by deep oscillation magnetron sputtering (energy—0.1–1 keV and dose of ions—1014–1015 ion/cm2) forms graphene-like nanolayer and causes heterogeneous changes in structural and mechanical properties of the surface: Topography, elastic modulus and depth of implantation for the hard/soft phase areas are different. As a result, after certain treatment regimens strain-induced defects (nanocracks in the areas of the modified soft phase, or folds in the hard phase) appear on the surfaces of stretched materials. Treated surfaces have increased hydrophobicity and free surface energy, and in some cases show good deformability without any defects. Introduction Materials with carbon or carbon-containing plasma-modified surface have found wide application in products with improved biomedical properties [1,2]. The treatment of soft polymers is of great interest. In this case, close attention should be paid to the mechanical and deformation properties of the coating. Regardless the nature of treatment, the stiffness of the coating (or the modified surface layer), even at the initial stage of formation [3], quickly becomes significantly higher than the polymeric substrate. A mechanical loading damages such surfaces, that is widely shown for different polymers and techniques of treatment: Carbon sputtering [4], cold plasma [5], plasma enhanced chemical vapor deposition [6], plasma ion implantation [7], etc. In a case of multi-cycle deformation, which is typical for real operating conditions, the cracks propagate to a depth exceeding the thickness of the modified layer, significantly damaging the substrate [8]. Whether the use of such materials is safe or not is an open question. Obvious, that the damage of the surface should be avoided. A low-energy ion implantation could be the solution. In this case, ions will be distributed both on the surface and in some surface layer. An appropriate dose of ions will provide low stiffness, i.e., high deformability of the modified surface. In comparison with sputtering, changes in the structure of the of the polymer (appearance of free radicals and hydrogen bonds) caused by the ion implantation will have a positive effect on the properties (high hydrophobicity and free surface energy) that are important for biomedical materials. Polyurethane is a widely used synthetic polymer. Depending on a formulation, its mechanical properties vary from viscous liquids to rigid plastics. In particular, elastic polyurethanes are used in the manufacture of biomedical products (tubes, implants, etc.). Polyurethane has a complex chemical structure. It is a block polymer with domains of hard phase in a matrix of soft phase [9,10]. The study of changes of such polymer at an initial stage of ion treatment is the key for the development of the materials with good deformability. Note, that prolonged treatment creates homogeneous and undesirable hard surface. The purpose of this work was to study the effect of low-energy carbon ion implantation on the structural and mechanical properties of elastic polyurethane, the formulating conditions of creating deformable and defect-free surfaces. Creating Polyurethane Polyurethane (PU) was manufactured from prepolymer (urethane based simple polyester and toluene diisocyanate) and cross-linking agent (hardener MOCA-16.5 weight parts and solvent polyfurite (polyoxytetramethylene glycol)). The components were heated up to 70 °C, vacuumed for 5 min, mixed, then vacuumed again and cured in a mold with an open top at 100 °C for 20 h. The thickness of the obtained plates was 2 mm. Initial elastic modulus measured by uniaxial mechanical tests-25 MPa, and elongation at break-800%. This PU consists of hard and soft blocks. Its chemical formula is shown in Figure 1. Carbon Implantation A flat balanced magnetron with a graphite target diameter of 80 mm was used as a source of ions, which worked in high-current modulation of pulsing current discharge (deep oscillation magnetron sputtering, DOMS). The ions coming from the plasma of magnetron discharge are accelerated near the sample by a metal mesh with grid size of 1.2 by 1.2 mm. The potential U of the mesh was set to 0.1, 0.3, 0.5 and 1.0 keV. The samples were placed in a vacuum chamber on a water-cooled holder. The chamber was pumped to the pressure 5 × 10 −5 Torr. The partial pressure of argon was set to 2 × 10 -3 Torr and the surface of the magnetron target was cleaned in plasma of its own discharge for 5 min. During the cleaning, a shield was installed between the magnetron and the sample. The treatment was carried out continuously in a pulsed mode: The amplitude of current discharge-40 A, pulse duration-8 μs, time between pulses-10 ms. The surface temperature of the samples was controlled by infrared pyrometer and did not exceed 30 °C. Carbon Implantation A flat balanced magnetron with a graphite target diameter of 80 mm was used as a source of ions, which worked in high-current modulation of pulsing current discharge (deep oscillation magnetron sputtering, DOMS). The ions coming from the plasma of magnetron discharge are accelerated near the sample by a metal mesh with grid size of 1.2 by 1.2 mm. The potential U of the mesh was set to 0.1, 0.3, 0.5 and 1.0 keV. The samples were placed in a vacuum chamber on a water-cooled holder. The chamber was pumped to the pressure 5 × 10 −5 Torr. The partial pressure of argon was set to 2 × 10 −3 Torr and the surface of the magnetron target was cleaned in plasma of its own discharge for 5 min. During the cleaning, a shield was installed between the magnetron and the sample. The treatment was carried out continuously in a pulsed mode: The amplitude of current discharge-40 A, pulse duration-8 µs, time between pulses-10 ms. The surface temperature of the samples was controlled by infrared pyrometer and did not exceed 30 • C. Choice of treatment parameters is based on the prevention of surface heating. In addition, preliminary studies have shown that higher energy or dose of ions produces stiff wrinkled surface that does not meet the deformability requirements. Atomic Force Microscopy An atomic force microscope (AFM) Ntegra Prima (NT-MDT B.V., Moscow, Russia) was used in a regime of dynamic nanoindentation: The probe indents a surface with high frequency (~1 kHz). Each point of the relief has its own dependence of load vs. depth of indentation. The indentation load was limited to 1.5-2 nN, so that the depth of indentation did not exceed 10-15 nm. Subsequent data processing gives a map of surface elastic modulus using a Johnson-Kendall-Roberts model. Obviously, the obtained values for the ion-treated surface are not the true modulus of the modified layer, but can be used for comparative analysis of the results. Probes with calibrated tip radius (5 nm) and cantilever stiffness (0.4 nN/nm) were used. The shapes of the probes were estimated by the blind reconstruction method based on images of a test sample (TipCheck). The obtained AFM-images were subjected to partial restoration of true relief [11]. The materials were studied in both undeformed and stretched state. In the latter case, samples (20 by 2 by 2 mm) were glued to the substrate. The middle part was stretched by placing inserts between the sample and substrate. As a result, the top central part of the sample is stretched to a deformation of~50%. Raman Spectroscopy Raman spectra were obtained by Bruker Senterra spectrometer with 532-nm excitation laser: Duration of accumulation-1 s; number of accumulations-1200; slot aperture-25 by 1000 µm. The radiation power was limited to 0.2 mW to prevent destruction of the soft polymer and to collect data from the very top layer of the surface. The base line of the spectra was leveled by adaptive method in Spectragryph software v1.2. Four spectra were obtained from different parts of the sample; the presented results are the average of these measurements. Free Surface Energy A wetting contact angle was determined by a sessile drop method. Water and diethylene glycol were used as test fluids. Free surface energy was calculated by an Owens-Wendt-Rabel-Kaelble method as the sum of dispersion (takes into account Van der Waals interactions) and polar (dipole interactions and energy of hydrogen bonds) components. Results and Discussion The surface of the untreated PU has a complex hierarchical morphology with structural-mechanical heterogeneities. Protruding surface irregularities visible on the microscale (Figure 2a) are agglomerates of the hard phase of the polymer: Elastic modulus of these elevations is higher that the surrounding area. At submicron level ( Figure 2b) the hard phase has a tangled fibrillar nanostructure. The concentration of fibrils in the agglomerates (area "B" in Figure 2b) is higher than the low-modulus lowlands (area "A"); the latter correspond to areas with high concentration of soft phase. The distributions of elastic modulus in relatively stiff E hard (bright areas of the modulus map) and soft E soft (dark areas) areas are shown in the insert in Figure 2a The implantation depth of carbon ions was estimated using a Trim software [12]. The sizes of structural-mechanical inhomogeneities of the surface are from units (thickness of fibrils) to hundreds of nanometers (agglomerates). Therefore, it is reasonable to perform the calculations separately for the hard and soft phases (the chemical structure is presented in Figure 1). The results of the simulation ( Figure 3) show that carbon ions penetrate deeper into the hard phase than into the soft phase. Therefore, the concentration of the implanted carbon in the unit volume of the soft phase is higher. This difference increases as the energy of ions rises. Representative AFM-images of treated surfaces are given in Figures 4 and 5. Elastic modulus shows clear contrast between the treated areas of hard (bright areas) and soft (dark areas on the modulus maps that correspond to the lowlands of the relief) phase. Roughness of the modified soft phase of the PU rises as the energy of ions increases (Figure 6a). In these areas (highlighted by arrows in Figure 4), local wrinkles start to form at the lowest energy of ion implantation. When the energy of ions reaches 1.0 keV, the entire surface of the material is covered with a chaotic wrinkled structure. The roughness of the modified hard phase does not change significantly up to the energy of 0.5 keV (Figure 6a), and then increases due to the wrinkling. Note The implantation depth of carbon ions was estimated using a Trim software [12]. The sizes of structural-mechanical inhomogeneities of the surface are from units (thickness of fibrils) to hundreds of nanometers (agglomerates). Therefore, it is reasonable to perform the calculations separately for the hard and soft phases (the chemical structure is presented in Figure 1). The results of the simulation ( Figure 3) show that carbon ions penetrate deeper into the hard phase than into the soft phase. Therefore, the concentration of the implanted carbon in the unit volume of the soft phase is higher. This difference increases as the energy of ions rises. The implantation depth of carbon ions was estimated using a Trim software [12]. The sizes of structural-mechanical inhomogeneities of the surface are from units (thickness of fibrils) to hundreds of nanometers (agglomerates). Therefore, it is reasonable to perform the calculations separately for the hard and soft phases (the chemical structure is presented in Figure 1). The results of the simulation ( Figure 3) show that carbon ions penetrate deeper into the hard phase than into the soft phase. Therefore, the concentration of the implanted carbon in the unit volume of the soft phase is higher. This difference increases as the energy of ions rises. Roughness of the modified soft phase of the PU rises as the energy of ions increases ( Figure 6a). In these areas (highlighted by arrows in Figure 4), local wrinkles start to form at the lowest energy of ion implantation. When the energy of ions reaches 1.0 keV, the entire surface of the material is covered with a chaotic wrinkled structure. The roughness of the modified hard phase does not change significantly up to the energy of 0.5 keV (Figure 6a), and then increases due to the wrinkling. Note Roughness of the modified soft phase of the PU rises as the energy of ions increases (Figure 6a). In these areas (highlighted by arrows in Figure 4), local wrinkles start to form at the lowest energy of ion implantation. When the energy of ions reaches 1.0 keV, the entire surface of the material is covered with a chaotic wrinkled structure. The roughness of the modified hard phase does not change significantly up to the energy of 0.5 keV (Figure 6a), and then increases due to the wrinkling. Note that the wrinkles of the hard phase after 250 impulses of ion implantation are formed at 0.5 keV (see Figure 5c), but for N = 100 the higher energy is required (Figure 4d). Wrinkling is caused by a loss of stability of the surface. A stress in a stiff layer on an elastic substrate at which the loss of stability occurs is proportional to the elastic modulus of the layer [13]. The values of Ehard depend of the dose of ions and for the energy of 0.5 keV, 250 pulses are enough to wrinkle the hard phase. At higher energy of ions, Ehard reaches one level regardless the ion dose (that is not the case for the modulus of modified soft phase), so the entire surface is covered with wrinkles. As for the modified soft phase, the rapid loss of stability can be explained by its properties (thickness and stiffness of the modified layer) and the influence of the surrounding hard phase. Elastic moduli of the treated hard and soft phases increase from 60 to 90 MPa (Figure 6b). The increase of energy of ions leads to the homogenization of the mechanical properties of the surface: Ehard/Esoft falls (see insert in Figure 6b). However, this trend is not satisfied for the materials with 100 impulses of ion implantation: At U = 0.5 keV a rise of mechanical heterogeneities of surfaces occurs. This slight jump of the properties will have a significant effect on the deformation behavior of the surface (see below). Wrinkling is caused by a loss of stability of the surface. A stress in a stiff layer on an elastic substrate at which the loss of stability occurs is proportional to the elastic modulus of the layer [13]. The values of Ehard depend of the dose of ions and for the energy of 0.5 keV, 250 pulses are enough to wrinkle the hard phase. At higher energy of ions, Ehard reaches one level regardless the ion dose (that is not the case for the modulus of modified soft phase), so the entire surface is covered with wrinkles. As for the modified soft phase, the rapid loss of stability can be explained by its properties (thickness and stiffness of the modified layer) and the influence of the surrounding hard phase. Elastic moduli of the treated hard and soft phases increase from 60 to 90 MPa (Figure 6b). The increase of energy of ions leads to the homogenization of the mechanical properties of the surface: Ehard/Esoft falls (see insert in Figure 6b). However, this trend is not satisfied for the materials with 100 impulses of ion implantation: At U = 0.5 keV a rise of mechanical heterogeneities of surfaces occurs. This slight jump of the properties will have a significant effect on the deformation behavior of the surface (see below). Wrinkling is caused by a loss of stability of the surface. A stress in a stiff layer on an elastic substrate at which the loss of stability occurs is proportional to the elastic modulus of the layer [13]. The values of E hard depend of the dose of ions and for the energy of 0.5 keV, 250 pulses are enough to wrinkle the hard phase. At higher energy of ions, E hard reaches one level regardless the ion dose (that is not the case for the modulus of modified soft phase), so the entire surface is covered with wrinkles. As for the modified soft phase, the rapid loss of stability can be explained by its properties (thickness and stiffness of the modified layer) and the influence of the surrounding hard phase. Elastic moduli of the treated hard and soft phases increase from 60 to 90 MPa (Figure 6b). The increase of energy of ions leads to the homogenization of the mechanical properties of the surface: E hard /E soft falls (see insert in Figure 6b). However, this trend is not satisfied for the materials with 100 impulses of ion implantation: At U = 0.5 keV a rise of mechanical heterogeneities of surfaces occurs. This slight jump of the properties will have a significant effect on the deformation behavior of the surface (see below). (Figures 7d, 8d). On the nano-scale (see inserts in Figures 7,8) at U ≤ 0.5 keV, mesh-like nanostructure covers the surfaces. In the literature, one can find analogues of similar structures formed at the initial stage of carbon-like coating growth [14]. A height of cell walls is 0.4-0.6 nm, the size of cells is 10-20 nm. Qualitative analysis established the growth of mesh size as energy of ions increases. In comparison with untreated polymer (Figure 2b), modulus maps of treated materials at nano-level do not correlate with the relief: Measurements are influenced by inhomogeneous mechanical properties of the modified surface layer of the material. Figures 9 and 10 show AFM images of 50% stretched materials. The local stiffness of the surfaces has a determining influence on the mechanical behavior of the material during deformation. The reliefs of stretched surfaces at U ≤ 0.3 keV are similar: Carbon-modified agglomerates of the hard phase are oriented along the axis of deformation (Figure 9a); identical AFM images of such surfaces are not presented. Submicron cracks are formed in soft areas of the stretched surface of the material treated with 0.5 keV and N = 100 (Figure 9b). The increase of energy to 1 keV leads to coalescence of cracks and increase of their width (Figure 9c), i.e., the fracture starts at lower critical tension. The average depth of cracks is 15-20 nm, that is more than the depth of ion implantation: High local strain (indicated by polymer strands connecting the crack edges, see Figure 9c) of the PU ruptures soft phase at the bottom of the cracks. Coatings 2020, 10, x FOR PEER REVIEW 7 of 11 In comparison with untreated polymer (Figure 2b), modulus maps of treated materials at nanolevel do not correlate with the relief: Measurements are influenced by inhomogeneous mechanical properties of the modified surface layer of the material. Figures 9 and 10 show AFM images of 50% stretched materials. The local stiffness of the surfaces has a determining influence on the mechanical behavior of the material during deformation. The The elastic modulus of a hard and soft phase after 250 impulses of implantation is higher (see Figure 6b) than for N = 100, however, damage of the surface does not occur ( Figure 10): Strain-induced folds, co-directed with the axis of deformation appear first in the stiff areas (Figure 10a), and then expanding on the entire surface (Figure 10b): The elastic moduli of the modified phases reach one value. Note, that the wrinkles, caused by loss of stability, disappear under the tension. The elastic modulus of a hard and soft phase after 250 impulses of implantation is higher (see Figure 6b) than for N = 100, however, damage of the surface does not occur ( Figure 10): Strain-induced folds, co-directed with the axis of deformation appear first in the stiff areas (Figure 10a), and then expanding on the entire surface (Figure 10b): The elastic moduli of the modified phases reach one value. Note, that the wrinkles, caused by loss of stability, disappear under the tension. The strain-induced folds ( Figure 10) are quite sharp. These are the areas of potential crack nucleation and propagation under multicycle loads. The deformation behavior is explained by the peculiarities of the elastic moduli of the treated hard and soft phase (see insert in Figure 6b): At 100 impulses of implantation the material is more favorable to break in less stiff and thinner modified layer of soft phase. At 250 pulses, the difference in stiffness is minimal, the stress field is more homogeneous, and, despite the higher stiffness of these surfaces, no damage occurs. Note, that no cracking was observed at U = 0.1 keV, when the difference in moduli is also significant. This can be explained by the small thickness of the modified layer (see Figure 3). The internal stresses in the modified layer are nonlinearly dependent on the (low) energy of the implanted ions [15] and can be both tensile and compressive. The wrinkling in the areas of the modified soft phase is the sign of local compressive stresses. These internal stresses are amplified by influence of the surrounding hard phase. Thus, inhomogeneous stress field on the material surface layer could be another cause of cracking. The treated materials have an increased free surface energy ( Figure 11a). The growth is due to the polar component of the energy (characterizes dipole interactions and energy of hydrogen bonds), that should have a positive effect on the sorption activity of proteins [16]. Besides, low surface hydrophilicity is important for many biomedical applications (Figure 11b) [17]. It correlates with the polar energy component and maximal for U = 1 keV and N = 100, however, this material showed the worst deformation behavior. The raise of the polar part is attributed to the increase of C=O groups and fall of dispersive partreduction of carbonyl and benzene bonds (see further). (a) (b) Figure 11. (a) free surface energy and (b) wetting contact angle. The strain-induced folds ( Figure 10) are quite sharp. These are the areas of potential crack nucleation and propagation under multicycle loads. The deformation behavior is explained by the peculiarities of the elastic moduli of the treated hard and soft phase (see insert in Figure 6b): At 100 impulses of implantation the material is more favorable to break in less stiff and thinner modified layer of soft phase. At 250 pulses, the difference in stiffness is minimal, the stress field is more homogeneous, and, despite the higher stiffness of these surfaces, no damage occurs. Note, that no cracking was observed at U = 0.1 keV, when the difference in moduli is also significant. This can be explained by the small thickness of the modified layer (see Figure 3). The internal stresses in the modified layer are nonlinearly dependent on the (low) energy of the implanted ions [15] and can be both tensile and compressive. The wrinkling in the areas of the modified soft phase is the sign of local compressive stresses. These internal stresses are amplified by influence of the surrounding hard phase. Thus, inhomogeneous stress field on the material surface layer could be another cause of cracking. The treated materials have an increased free surface energy ( Figure 11a). The growth is due to the polar component of the energy (characterizes dipole interactions and energy of hydrogen bonds), that should have a positive effect on the sorption activity of proteins [16]. Besides, low surface hydrophilicity is important for many biomedical applications (Figure 11b) [17]. It correlates with the polar energy component and maximal for U = 1 keV and N = 100, however, this material showed the worst deformation behavior. The strain-induced folds ( Figure 10) are quite sharp. These are the areas of potential crack nucleation and propagation under multicycle loads. The deformation behavior is explained by the peculiarities of the elastic moduli of the treated hard and soft phase (see insert in Figure 6b): At 100 impulses of implantation the material is more favorable to break in less stiff and thinner modified layer of soft phase. At 250 pulses, the difference in stiffness is minimal, the stress field is more homogeneous, and, despite the higher stiffness of these surfaces, no damage occurs. Note, that no cracking was observed at U = 0.1 keV, when the difference in moduli is also significant. This can be explained by the small thickness of the modified layer (see Figure 3). The internal stresses in the modified layer are nonlinearly dependent on the (low) energy of the implanted ions [15] and can be both tensile and compressive. The wrinkling in the areas of the modified soft phase is the sign of local compressive stresses. These internal stresses are amplified by influence of the surrounding hard phase. Thus, inhomogeneous stress field on the material surface layer could be another cause of cracking. The treated materials have an increased free surface energy ( Figure 11a). The growth is due to the polar component of the energy (characterizes dipole interactions and energy of hydrogen bonds), that should have a positive effect on the sorption activity of proteins [16]. Besides, low surface hydrophilicity is important for many biomedical applications (Figure 11b) [17]. It correlates with the polar energy component and maximal for U = 1 keV and N = 100, however, this material showed the worst deformation behavior. The raise of the polar part is attributed to the increase of C=O groups and fall of dispersive partreduction of carbonyl and benzene bonds (see further). The raise of the polar part is attributed to the increase of C=O groups and fall of dispersive part-reduction of carbonyl and benzene bonds (see further). The destruction of polymer macromolecules leads to an emergence of free radicals. Theory predicts the formation of two types of free radicals [18]: Short-lived free radicals in damaged macromolecules and long-lived free radicals trapped in carbonized clusters. Carbonization of a polymer is usually achieved by high dose (10 16 -10 17 ion/cm 2 ) of implanted plasma ions that recombine bonds between existing carbon atoms of the polymer macromolecules [19]. Such long-time treatment raises stiffness of the surface, making materials useless in deformation-related applications. In our case, carbon structures are formed by direct implantation of carbon, and the lower dose of ions creates softer surface. Coatings 2020, 10, x FOR PEER REVIEW 9 of 11 The destruction of polymer macromolecules leads to an emergence of free radicals. Theory predicts the formation of two types of free radicals [18]: Short-lived free radicals in damaged macromolecules and long-lived free radicals trapped in carbonized clusters. Carbonization of a polymer is usually achieved by high dose (10 16 -10 17 ion/cm 2 ) of implanted plasma ions that recombine bonds between existing carbon atoms of the polymer macromolecules [19]. Such long-time treatment raises stiffness of the surface, making materials useless in deformation-related applications. In our case, carbon structures are formed by direct implantation of carbon, and the lower dose of ions creates softer surface. Raman spectra of the materials are presented in Figure 12. The increased intensity in the region of 1800 cm −1 corresponds to vibrations of C=O bonds; 2000-2200 cm −1 : C≡C, C=C bonds [20]. The intensity of carbonyl (1712 cm −1 ) and benzene (1184, 1617 cm −1 ) bonds of hard phase as well as N-C-N (1318 cm −1 ) and C-O (1050 cm −1 ) bonds decreases [21]. The peaks in the frequency range >2300 cm −1 correspond to graphene-like structures [22][23][24]. The intensity of these peaks is higher at lower energy of ions. In this regime the carbon atoms have especially low implantation depth and, therefore, concentrate in a narrow surface layer forming single or stacking [24] graphene-containing structures: The monolayer of graphene [22] or disordered graphene [22] having sharp peaks at 2500 and 2700 cm −1 , graphene-oxide characterized by increased intensity at 2500-2700 cm −1 [23], and the G-peak (~1600 cm −1 ) of carbon is masked by the polyurethane. The mesh-like surface nanostructure (Figures 7a-c, 8a-c) could be the direct evidence of such formations. Note, that more precise analysis is complicated due to heterogeneities of the original substrate and the modified surface layer. As the energy increases, the distribution of carbon ions over the depth broadens that decreases the intensity of graphene. Conclusions Low energy implantation of carbon ions into the elastic polyurethane surface forms graphenelike structures and causes heterogeneous structural and mechanical changes of the surface properties associated with the phase separation of the initial polyurethane into hard and soft blocks. Depending on the energy and dose of the ions, the local regions of the surface is modified not uniformly. Wrinkles appear on the surface, first in the soft phase, then spreading over the whole surface. At the nano level, the fibrillar structure of the hard phase of the initial polymer is partially preserved up to the energy The peaks in the frequency range >2300 cm −1 correspond to graphene-like structures [22][23][24]. The intensity of these peaks is higher at lower energy of ions. In this regime the carbon atoms have especially low implantation depth and, therefore, concentrate in a narrow surface layer forming single or stacking [24] graphene-containing structures: The monolayer of graphene [22] or disordered graphene [22] having sharp peaks at 2500 and 2700 cm −1 , graphene-oxide characterized by increased intensity at 2500-2700 cm −1 [23], and the G-peak (~1600 cm −1 ) of carbon is masked by the polyurethane. The mesh-like surface nanostructure (Figure 7a-c, Figure 8a-c) could be the direct evidence of such formations. Note, that more precise analysis is complicated due to heterogeneities of the original substrate and the modified surface layer. As the energy increases, the distribution of carbon ions over the depth broadens that decreases the intensity of graphene. Conclusions Low energy implantation of carbon ions into the elastic polyurethane surface forms graphene-like structures and causes heterogeneous structural and mechanical changes of the surface properties associated with the phase separation of the initial polyurethane into hard and soft blocks. Depending on the energy and dose of the ions, the local regions of the surface is modified not uniformly. Wrinkles appear on the surface, first in the soft phase, then spreading over the whole surface. At the nano level, the fibrillar structure of the hard phase of the initial polymer is partially preserved up to the energy of 0.5 keV, and a mesh-like nanostructure appears on the surface. As the energy increases to 1.0 keV, the surfaces become homogenously stiff with a wrinkled relief. The stiffness of the treated surfaces increases with the energy and dose of ions. However, the ratio of local elastic moduli of the modified hard and soft phases changes nonlinearly: At 100 pulses of implantation and energy of ions 0.5 or 1.0 keV the leap of mechanical inhomogeneities was observed. This is the reason of submicron size cracks in the soft areas of the surface of the stretched material. Treatment with 250 pulses leads to the equalization of the local mechanical properties of the surface and at energies of 0.5 and 1.0 keV no cracking occurs. However, sharp folds, which are aligned with the deformation axis, were observed on the stretched surfaces: Areas of potential defects under multi-cycle loads. Materials treated with an energy 0.3 keV and less, did not show any strain-induced defects. Ion implantation decreases hydrophilicity and increases free surface energy, which makes these materials promising in developing flexible biomedical products. The optimal treatment regime for this polyurethane, in the terms of surface activity and deformability, can be considered as: Energy 0.3 keV and 250 impulses of implantation. In this case, an active and less hydrophilic surface with homogeneous stiffness is formed, deformable of at least to 50% without any defects.	7,358.4	2020-03-16T00:00:00.000	[ "Materials Science", "Physics" ]
Mesenchymal Stem Cells Reversed Morphine Tolerance and Opioid-induced Hyperalgesia More than 240 million opioid prescriptions are dispensed annually to treat pain in the US. The use of opioids is commonly associated with opioid tolerance (OT) and opioid-induced hyperalgesia (OIH), which limit efficacy and compromise safety. The dearth of effective way to prevent or treat OT and OIH is a major medical challenge. We hypothesized that mesenchymal stem cells (MSCs) attenuate OT and OIH in rats and mice based on the understanding that MSCs possess remarkable anti-inflammatory properties and that both OT and chronic pain are associated with neuroinflammation in the spinal cord. We found that the development of OT and OIH was effectively prevented by either intravenous or intrathecal MSC transplantation (MSC-TP), which was performed before morphine treatment. Remarkably, established OT and OIH were significantly reversed by either intravenous or intrathecal MSCs when cells were transplanted after repeated morphine injections. The animals did not show any abnormality in vital organs or functions. Immunohistochemistry revealed that the treatments significantly reduced activation level of microglia and astrocytes in the spinal cord. We have thus demonstrated that MSC-TP promises to be a potentially safe and effective way to prevent and reverse two of the major problems of opioid therapy. Chronic pain is a significant public health problem. It afflicts more than 100 million Americans and costs more than $635 billion annually 1,2 . Opioids, such as morphine, play an indispensable role in pain relief but are often associated with two major problems: opioid tolerance (OT) and opioid-induced hyperalgesia (OIH) [3][4][5][6][7] . OT is a physiological process where the body adjusts to a medication of frequent exposure and requires escalating doses to achieve the same effect. OIH is a phenomenon, in which individuals taking opioids to treat pain paradoxically develop an increased sensitivity to noxious stimuli. Both OT and OIH in animals have been validated in humans 8,9 . Nearly 50,000 people die every year of opioid overdose in the US, leading the Center for Disease Control and Prevention (CDC) to declare the problem an ongoing "national epidemic". These facts underscore an urgent need for finding effective therapies to treat pain and OT and to reduce the disastrous outcomes associated with opioid treatment. Distinct molecular mechanisms are indicated for the two closely related but different phenomena 9 . Neuroinflammation, mediated by immune cells and glial cells, appears to play a central role 10,11 . Opioids such as morphine can cause neuroinflammation 12 through acting on Toll-like receptor 4 on microglia and lead to development of OT 13 . Similarly, OIH is mediated by μ opioid receptor-dependent expression of P2X4 receptors on microglia and release of brain-derived neurotropic factor (BDNF) 14 . The P2X4-BDNF-TrkB pathway mediates microglia-to-neurons signaling and leads to sensitization of spinal lamina I neurons and OIH 14 . Thus, modulating neuroinflammation may prove to be an effective strategy to treat both OT and OIH. We aim to develop a safe and efficacious therapy for OT and OIH in clinical practice. We chose to use MSCs because of their powerful paracrine functions, as shown in animal models of diseases such as traumatic brain injury 15 , peripheral neuropathy 16,17 , and neuropathic pain 18 . Immunomodulatory and anti-inflammatory effects of MSCs were related to neuroprotection, neuroregeneration, and neuroneuromodulation in these studies. For example, intravenous (IV) injection of human adipose-derived MSCs (hAD-MSCs) induced a significant reduction in mechanical allodynia and complete reversal of thermal hyperalgesia in a dose-dependent fashion in a mouse model of diabetic neuropathy 19 . The treatment decreased the level of IL-1β and increased IL-10 in the lesioned nerve and restored normal inducible nitric oxide synthase (iNOS) expression in the spinal cord. More recently, it was shown that IT rMSCs inhibited neuropathic pain via secretion of transforming growth factor beta (TGF-β ) 20 . Thus, MSCs may release factors that promote tissue recovery through stimulating resident stem/progenitor cells, remodeling extracellular matrix, forming new blood vessels, and modulating immune functions [21][22][23] . We hypothesized that MSC transplantation (MSC-TP) attenuates chronic OT that is induced by long-term daily morphine injections. We further hypothesized that MSC-TP attenuates OIH that is developed as a consequence of chronic morphine injections. We tested these hypotheses by using intrathecal and intravenous routes of transplantation in rats and mice and studied the distribution of the transplanted cells and the level of activation of microglia and astrocytes in the spinal cord in response to morphine and MSC-TP. Results We first isolated MSCs from rat bone marrow and characterized the cells through flow cytometry (FACS) and induced differentiations. These cells showed morphological properties and cell markers characteristic of stem cells and differentiated into osteoblast cells and adipose cells in specific culture media (Fig. 1). We then tested the preventive and therapeutic effects of intrathecal and intravenous MSC-TPs on OT, which was induced by daily morphine injections. Acute OT was induced after 3 days of daily injections. Administration of cumulative doses of morphine on day 4 produced a dose-response curve with a maximum effect dose of 18 mg/kg (MS: 24.52 ± 0.48, n = 5), which was significantly higher than that of the control group (8.0 mg/kg) (NS: 25.00 ± 0, n = 3) (Supplementary Fig. 1a; P = 0.01). Chronic OT was induced by daily morphine injections for 3 to 4 weeks and evaluated by measuring two sets of paw withdrawal thresholds (PWTs) to mechanical and thermal stimulation. The first were measured before daily morphine injection and the second were measured 50 min after the injection. The differences between the two sets reflect the level of tolerance. A large difference indicates no or low tolerance while a small difference indicates high tolerance. The differences decreased gradually and significantly after 7 days of morphine treatment, reached a minimal difference at day 12, and maintained a small difference thereafter ( Supplementary Fig. 1b, df = 7, F = 45.97; P < 0.0001; n = 8). OT was further evaluated by the tail flick test. Maximum possible effect (MPE) of morphine was used to indicate tolerance; high MPE (%) indicates low or no tolerance while low MPE indicates high tolerance. MPE was significantly reduced after 7 days of morphine treatment, reaching ~10% of the baseline value by day 12 (Supplementary Fig. 1c; df = 6, F = 47.34; P < 0.0001; n = 12). Intrathecal or intravenous MSC-TP did not cause any behavioral change in normal rats ( Supplementary Fig. 1e,f, P > 0.05, n = 6 for NS, n = 8 for MSC). In contrast, it significantly and consistently attenuated the development of OT. Both intrathecal and intravenous MSCs were remarkably effective ( Fig. 2a-h). A one-time transplantation significantly mitigated OT for the whole course of the experiments of up to 26 days (Fig. 2b,c). The effects were almost identical when the transplantation was performed one day or 7 days before morphine treatment. These experiments were repeated separately by two groups of experimenters who were blinded to the treatments ( Supplementary Fig. 2). The effects were further evaluated by the tail flick tests (Fig. 2d,e. Figure 2d, df = 3, F = 202.6; P < 0.0001, n = 6 for each group; Fig. 2e: df = 3, F = 131.7; P < 0.0001, n = 6 for each group). Consistent with the paw withdrawal experiments, MSCs significantly increased the MPE (%) regardless of the route and time of the transplantation, although to a lesser extent (Fig. 2d,e). MSC-TP significantly reversed established OT when it was performed 14 days after daily morphine treatment ( Morphine-induced OT reached its peak at day 12 of daily morphine injections in both the mechanical and thermal tests (Fig. 2f-h). Both intrathecal and intravenous MSCs significantly and consistently restored the sensitivity to morphine. This therapeutic effect took place rather rapidly and lasted for the whole course of the experiment to day 28 with no sign of waning. Within 2 days of transplantation, the PWT differences increased significantly in both transplantation groups, compared to the control group (P < 0.001). Similarly, intrathecal or intravenous MSCs significantly increased tail flick MPE from below 15% to above 30% on day 16 and above 50% on day 24 ( Fig. 2g) (P < 0.05). The therapeutic effects were further tested in mice. Intravenous transplantation of MSCs significantly attenuated OT. Compared to the control group, the mean tail flick latencies were significantly increased in the transplantation group (Fig. 2h). Consistent with the rat experiments, this effect took place within 2 days of the transplantation. Next, we tested the preventive and therapeutic effects of MSCs on OIH, which was also induced by daily morphine injections. OIH was reflected by the progressive decline of PWTs from the baseline values established 3 days before morphine treatment ( Supplementary Fig. 1d, df = 17, F = 17.89; P < 0.0001. MS n = 1; NS n = 8). The decline took place over a course of 5-7 days (P < 0.001) and the hyperalgesia status persisted even after the cessation of daily morphine injections. Such decline was not seen in the control group, which received daily saline . (P < 0.05, # P < 0.01, ¥ P < 0.001 compared with the MS group n = 6 in each group). OT was induced by repeated daily MS injections at 7.5 mg/kg in rats (f,g) or 10 mg/kg in mice (h). MSC-TP was administrated at Day 14 when OT had fully developed. Pain-like behavior was evaluated by von Frey filament in rats (f) and tail flick tests in rats (g) and mice (h). (P < 0.05, # P < 0.01, ¥ P < 0.001 compared with the same day MS group; n = 6 in MS group; n = 9 in IT and IV groups). Data: mean ± s.e. IT: intrathecal; IV: intravenous. MS, morphine sulfate; PWT, paw withdrawal threshold. Scientific RepoRts \| 6:32096 \| DOI: 10.1038/srep32096 injections (P > 0.05). Intrathecal or intravenous MSC-TP (5 × 10 5 ) substantially prevented the development of OIH (Fig. 3). The effects were long-lasting with no sign of waning over time. Consistent results were observed whether the transplantation was performed one day or seven days before morphine treatment. These results were replicated by two groups of experimenter who were blinded to the treatments. To test the therapeutic effects, MSC-TP was performed once OIH had fully developed. The transplantation effectively and rapidly reversed OIH. This effect lasted for the whole duration of the experiments (Fig. 3c, df = 11, F = 116.5; P < 0.0001, n = 6 for MS; n = 7 for IT or IV MSC). The therapeutic effect was further tested in mice (Fig. 3d, df = 5, F = 75.82; P < 0.0001. MS n = 6; MSC n = 8). Intravenous MSC-TP at day 14 of daily morphine treatment significantly increased the mean tail flick latency (P < 0.001). All animals survived the entire course of the experiments up to 68 days and had normal locomotion, food and fluid intake, body weight gain, and biochemical parameters for liver and kidney functions ( Supplementary Fig. 3, P > 0.05. NS group: n = 6, MS group n = 11, MS/MSC group n = 12). Histopathology examination at necropsy did not reveal any abnormality in any major organs. Dil labeled MSCs in red were successfully traced to the surface of the spinal cord and the dorsal root ganglia (DRG) after intrathecal transplantation ( Fig. 4a-d). Double staining of Dil and DAPI (nucleus) confirmed viable MSCs on the dorsal side of the spinal cord and the DRGs at various time intervals. We did not find any Dil labeled MSCs in any of these tissues after intravenous transplantation. Discussion The search for effective preventive and therapeutic strategies to counteract OT and OIH has been invigorated by the gravity of the profound negative impacts of OT and OIH. Here we for the first time report a powerful anti-tolerance effect of MSC-TP. Both intrathecal and intravenous MSC-TPs effectively attenuated the development of OT when performed before the initiation of chronic daily morphine injections in rats. MSC-TP almost completely reversed chronic OT when performed after OT had been established, regardless of the route of transplantation. These findings were consistent in both rats and mice. Thus, we have provided several lines of evidence that MSC-TP is a promising preventive and therapeutic therapy for OT with great potentials for clinical translation. We chose to focus on chronic OT, rather than acute OT, because it resembles more closely to clinical practice. In addition to the traditional approach to assessing OT by demonstrating a hallmark rightward shift in the agonist dose-response curve after 3 to 5 days of daily morphine administration ( Supplementary Fig. 1a), we introduced a new paradigm to investigate chronic OT, which was induced by daily morphine injections for up to four weeks ( Supplementary Fig. 1b-d). Preemptive MSC-TP significantly and persistently attenuated the development of OT (Fig. 2a-e). The results from two groups of investigators were strikingly consistent. The first group performed intrathecal MSC-TP and found a significant anti-tolerance effect ( Supplementary Fig. 2). The second group further performed both intrathecal and intravenous MSC-TPs and confirmed the anti-tolerance effects by both routes of transplantation (Fig. 2a-e). In addition, we tested MSC-TP at two time points (1 day and 7 days) before morphine treatment and observed identical anti-tolerance effects (Fig. 2b,c). These data clearly indicate that MSC-TP, either intrathecally or intravenously, could effectively prevent the development of OT. In addition to the preventive effect, MSC-TP effectively reversed established OT and restored sensitivity to morphine (Fig. 2a,f-h). This finding is important because it suggests that the therapeutic effect may be applicable to an increasingly large population of patients with ongoing OT due to chronic use of opioids for a variety of cancer and non-cancer pain conditions. It is well known that analgesic tolerance is commonly developed but tolerance to opioid adverse effects, such as respiratory depression and constipation, does not readily develop. This differential tolerance is one of the most common reasons patients suffer from detrimental consequences including overdose and death when dose escalation is required to overcome analgesic tolerance. Managing this population is a daunting challenge even to well-trained pain specialists. MSC-TP promises to emerge as an effective therapy for OT. Its clinical translation may have a profound impact on improving the safety and efficacy of opioid therapy and reducing opioid overdose and death. A second important finding of this work is the remarkable preventive and therapeutic effects of MSC-TP on OIH. We found a remarkable anti-hyperalgesia effect of MSC-TP in rats and mice. Similarly, the two groups of investigators independently showed consistent results and came to the same conclusions. The first group tested intrathecal MSC-TP and demonstrated a significant attenuation of the development of OIH. The second group tested both intrathecal and intravenous MSC-TPs and confirmed the preventive and therapeutic effects on OIH (Fig. 3). OIH and OT are related but distinct biological phenomena and clinical entities 24 . There is convincing evidence to clinically differentiate the two entities 9 . Dose escalation is required to overcome OT but such a strategy only further exacerbates OIH. In contrast, MSC-TP attenuated both OT and OIH. Interestingly, intravenous MSC-TP resulted in similar degrees of anti-tolerance and anti-hyperalgesia effects compared to intrathecal MSC-TP (Figs 2 and 3). The former is clinically advantageous compared to the latter. However, since MSCs transplanted by this route are largely trapped in the lungs [25][26][27] and may be injured by activation of complements 28 , we expected a short, if any, duration of the therapeutic effects. Surprisingly, both routes of MSC-TP achieved long-lasting preventive and therapeutic effects. This finding is clinically important and mechanistically intriguing. It not only indicates a convenient route of clinical application but also suggests systemic mechanisms of action. It doesn't seem to be necessary to place the cells in close proximity to the spinal cord ( Fig. 4a-d). MSCs likely exert anti-tolerance and anti-hyperalgesia effects through their powerful paracrine function, regulating the sensitivity to noxious stimulation and opioid medications through modulation of immune and inflammatory processes in the peripheral and central nervous systems. Microglial activation in the spinal cord plays a prominent role in the development of both OT and OIH 14,29 . Attenuating this process by MSCs is a plausible explanation for the observed effects. Indeed, microglial activation, induced by daily morphine injections, was significantly attenuated by MSC-TP ( Fig. 4e-g). Also notable was the upregulation of GFAP expression after morphine treatment (Fig. 4e,f,h). MSC-TP partially and significantly restored GFAP expression. These data are consistent with the reports that chronic morphine injection activated spinal and cortical glia cells 30,31 . Morphine tolerance and hyperalgesia/allodynia have been associated with spinal microglial and astroglial activation 32 . Selective activation of an astrocyte JNK pathway after the stimulation of neuronal μ -opioid receptor (MOR) appears to mediate astrocyte-neuron signaling and contribute to OIH 33 . Inhibition of spinal glial activation by fluorocitrate, a nonselective metabolic inhibitor of astrocytes, partially reversed the development of morphine tolerance in rats 31 . These observations support the notion that both immune cells (microglia) and glial cells (astrocytes) are involved in the development of OT and OIH. MSC-TP may have achieved its therapeutic effects through acting on these cells. In addition, MSCs may modulate other cell types in the innate and adaptive arms of the immune system. For example, MSCs shifted the cytokine secretion profile of dendritic cells, naïve and effector T cells [T helper 1 (TH1) and 2 (TH2)], and natural killer cells to a more anti-inflammatory phenotype 34 . Undoubtedly, our current mechanistic understanding of the MSC therapy is in its infancy. It is important to appreciate its complexity and resist the temptation of attributing the therapeutic effects to a single molecular signaling pathway because MSCs may regulate immune cells, glial cells, and neurons by mechanisms that include both direct cell-to-cell contacts and release of a multitude of soluble factors. Our data suggest that MSC-TP is safe and practical. All animals survived the whole experiments up to 68 days and maintained normal locomotion, food and fluid intake, body weight gain, and liver and renal function parameters ( Supplementary Fig. 3). Histology at necropsy did not reveal any abnormality in any major part of the body. Thus, there was no evidence of toxicity even with long-term experiments up to 68 days. The intrathecally transplanted cells may have survived and maintained function in vivo for at least 34 days (Figs 2 and 3). This is consistent with our finding that viable MSCs were found in the pia mater of the spinal cord (Fig. 4). The long survival and long-lasting effects of MSCs are particularly important in clinical applications. Several factors may have contributed to this success. We used MSCs from the bone marrow in an early passage (passage #4) and the cells are essentially non-immunogenic [34][35][36] . A recent study used cells after 16 passages and failed to demonstrate any analgesic or anti-inflammatory effects 11 . Applications of human MSCs are being explored extensively 37 via multiple clinical trials on spinal cord injury 38,39 , cardiovascular disease 40,41 , Parkinson's disease 42 , and diabetes 43 . Consistent with our results of xenogeneic transplantation from rats to mice (Figs 2 and 3), immune rejection has not been a major concern because MSCs are immune-privileged due to their absent or low expression of major histocompatibility complex class II (MHC-II) and other co-stimulatory molecules 44 . Human MSCs are viable in tissues for months after systemic administration in sheep 45 . MSCs are known to have a strong immunosuppressive property and have been used successfully in autologous as well as allogeneic MSC-TP without pharmacological immunosuppression 46 . This unique capacity is being utilized in combating autoimmune diseases in clinical trials 47 . In addition, it is well accepted that MSCs have extremely low risk of tumorigenicity 48 ; MSCs could actually inhibit tumor growth 49 . Clinical studies have convincingly demonstrated that direct injection of MSCs does not produce unwanted side effects and is well tolerated and safe 50,51 . In summary, we report a powerful anti-tolerance effect and a remarkable anti-hyperalgesia effect of MSC-TP in rats and mice. These effects were consistently observed by two groups of investigators independently. Both the intrathecal and intravenous routes of transplantation were effective. Intrathecally transplanted cells homed in the pia mater of the spinal cord and the DRGs and appeared to have maintained long-term viability. The animals showed normal vital functions without any trace of toxicity. The inhibitory effects of MSCs on microglia and astrocytes appeared to be related to the anti-tolerance and anti-hyperalgesia functions. It may be tempting to uncover a specific molecular or cellular mechanism of MSC action. However, it is most likely that multiple mechanisms are involved. MSCs may regulate immune cells and neurons by mechanisms that include both direct cell contact and release of soluble factors such as interleukin 10 (IL-10), leukemia inhibitory factor (LIF), and transforming growth factor (TGF) through a paracrine mechanism 52,53 . Activation of opioid receptors by MSCs may also contribute 54 . Collectively, we have demonstrated for the first time that MSC-TP promises to be an innovative, safe, efficacious, and cost-effective therapy to prevent and treat OT and OIH. This emerging therapy has enormous potential to profoundly impact clinical practice. It may improve the efficacy of opioid therapy, reduce the risk of opioid overdose, and save lives. Methods The research protocols were approved by the Cleveland Clinic Institutional Animal Care and Use Committee. We used both rats and mice in this investigation in order to determine the consistency of our findings in different species with allogeneic and xenogeneic transplantations of MSCs. The methods were carried out in accordance with the approved guideline. C57BL/6J mice (8-10 weeks, The Jackson Laboratory, Bar Harbor, Maine. USA) were used and were group housed (4-5/cage) in standard cages in a colony room maintained on a 12-hour reversed light/dark cycle. All mice had continuous access to food and water throughout the study. Similarly, mice were handled and adapted to the testing environment for 1 week prior to initiation of the experiments. All behavioral testing procedures were conducted between 08:00 and 13:00 h. Animals used in this study were cared for in accordance with the guidelines of the Institutional Animal Care and Use Committee at Cleveland Clinic. Animals. Adult male Isolation and culture of MSCs from the bone marrow of rats. MSCs were isolated from the bone marrow as described 55 with minor modifications. Rats (n = 6) sacrificed by CO2 asphyxiation according to Institutional Animal Care and Use Committee (IACUC) guidelines. The femurs and tibiae were removed from six-week-old male Sprague-Dawley rats, and washed three times with sterilized 1xPBS. The ends of the tibia and femur were cut by sharp Scissors. A 25-gauge needle was inserted into the bone marrow to flush out the tissue with a-MEM and filtered through a 100-μ m filter mesh (BD Bioscience). The bone marrow (BM) cells were cultured in a-MEM with 16% fetal bovine serum (FBS), 1% L-glutamine, and antibiotic solution (100 u/ml penicillin-streptomycin) in culture flask and incubated at 37 °C with 5% CO2. The medium was replaced after 24 h and every 3-4 days thereafter. MSCs were passaged when they reached 90% confluency by 1 min treatment with 0.05% trypsin and 0.02% EDTA at 37 °C. MSCs were characterized by surface markers through flow cytometry and by differentiation into adipogenic and osteogenic cells. Cell differentiation was tested at passage 4 according to manufacturer's protocol (Rat MSC differentiation kit, sc020, Fisher). MSCs used in all experiments were controlled within passage 5. Characterization of MSCs by flow cytometry. MSCs were expanded to passage four and were examined for surface marker expression using flow cytometry as previously described 56,57 with modifications. Briefly, cultured MSCs were harvested, washed, and re-suspended in FACS buffer (1% FCS and 0.1% sodium azide in 1xHBSS). After blocking with CD16/CD32 Abs at 4 °C for 30 min, cells were stained for surface markers with directly conjugated Abs in FACS buffer at 4 °C for 30 min. Cells were washed twice and re-suspended in 200-400 μ l of PBS for flow cytometry analysis as described before 57 Osteogenic differentiation. MSCs were cultured in the completed culture medium supplemented with Rat MSC differentiation kit (sc020, Fisher). After three weeks of osteogenic differentiation, cells were fixed in 4% PFA (paraformaldehyde, Sigma) for 1 h at 4 °C and stained with 40 mM Alizarin Red (pH 4.1, Sigma) for 10 minutes to visualize calcium deposition. Induction of OT and OIH. Acute opioid tolerance. Rats were injected with equal volume of normal saline (control) or 7.5 mg/kg morphine (experimental group) subcutaneously for 3 days. On the fourth day, rats of both groups received cumulative doses of morphine (0.0, 3.2, 5.6, 8.0, 10.0, and 18.0 mg/kg) as described 58 . Thermal plantar test was used to construct the agonist dose-response curves. Chronic OT and OIH. Morphine was diluted in sterile saline solution (0.9% NaCl) to a concentration of 7.5 mg/ml. Control group received an equal volume of sterile saline. Morphine sulfate (7.5 mg/kg) was injected subcutaneously daily to induce opioid tolerance. Either saline or morphine was injected at 10:00-11:00 AM. To evaluate OT, paw withdrawal thresholds (PWTs) (in grams) to mechanical stimulation were measured before and 50 min after each daily morphine injection. The differences between the two measurements reflect the level of tolerance to morphine. A large difference indicates no or low tolerance while a small difference indicates high tolerance. Similarly, thermal stimulation (Hargreave's test) and tail flick test were further used to evaluate opioid tolerance. The response latencies (in seconds) were measured before the initiation of daily morphine injections to establish baseline control values and then measured in various time points after the initiation of daily morphine injections (Figs 2, 3 and 4). The measurements were done in the morning, 30 min after each morphine injection. OIH was assessed by measuring PWTs to mechanical stimulation. Baseline values were first established 3 days before the initiation of daily morphine injections. PWT values were then measured every other day after the initiation of morphine injections. All measurements were taken in the morning before each morphine injection. OT and OIH in mice. To induce OT, mice were injected with 10 mg/kg of morphine for 14 days as previously reported with modification 59 . Tail flick tests were performed before the initiation of daily morphine injections to establish baseline values. The response latencies (in seconds) were further measured after initiation of daily morphine injections. The measurements were taken in the morning before and 30 min after each morphine injection. The differences between the two measurements reflect the level of tolerance to morphine. A large difference indicates no or low tolerance while a small difference indicates high tolerance. OIH was assessed by measuring tail flick test. Baseline values were first established 3 days before the initiation of daily morphine injections. Tail flick latencies were then measured at defined intervals after the initiation of morphine injections. Intrathecal and intravenous transplantation of MSCs. Intrathecal transplantation of MSCs was per- formed in the lumbar region, as described by Lu et al. 60 with modification. The rat was shaved in lumbar region of the back under anesthesia (40 mg/kg Phenobarbital Sodium), and placed on a rolled pad so the back was arched. After skin disinfection, the L4-L5 lumbar interspace was identified by palpating spinous processes. A needle (27G) was slowly advanced through the skin over the L4-L5 interspace until it reaches the subarachnoid space. Upon confirmation of needle placement in the subarachnoid space by a tail flick, a single dose of cells in 10 ul PBS was injected over a period of one minute. Intravenous transplantation of MSCs was performed through the tail vein in rats and a lateral retro-orbital approach in mice 61 . One dose of MSCs in 100 ul PBS was injected in each animal. MSC-TP was performed before the initiation of daily morphine injections to evaluate the preventive effect or after the initiation of daily morphine injections to evaluate the therapeutic effect. For preventive effect, transplantation was performed one day or 7 days before the initiation of daily morphine injections. At each experimental paradigm, rats were divided into three groups, each receiving PBS (Control group), 0.5 million MSCs intrathecally (IT MSC group), or 0.5 million MSCs intravenously (IV MSC group). Each group had 13-14 animals. Behavioral tests were performed on the days as indicated (Fig. 3a). For therapeutic effect, transplantation was performed 14 days after the initiation of daily morphine injections when opioid tolerance and opioid-induced hyperalgesia have been established. Similarly, rats were divided into three groups, each receiving PBS (Control group), 0.5 million MSCs intrathecally (IT MSC group), or 0.5 million MSCs intravenously (IV MSC group). Each group had 13-14 animals. Mechanical test and tail flick for thermal test were performed on the days as indicated (Fig. 4a). Behavioral tests. The sensitivity to noxious stimulation was determined by measuring the paw withdrawal thresholds (PWTs) in response to mechanical and thermal stimulation to the right hind paw or by measuring the tail flick latencies in response to a set temperature thermal stimulation to the tail. Animals were handled and habituated before behavioral testing to familiarize them with the environment and to minimize stress. All behavioral tests were performed in the Behavior Core Facility by experienced experimenters who were blinded to the treatment. The behavioral measurements began prior to morphine treatment to determine the baseline values and continued as indicated (Figs 3a and 4a). Mechanical Sensitivity Testing. Animals were placed in individual 10 × 10 × 15 cm plastic boxes on an elevated metal mesh floor and allowed to acclimate for at least 30 min before test. Mechanical sensitivity was tested using von Frey sensory evaluator filaments (Stoelting, Wood Dale, IL, USA) 62,63 . Filaments were applied to the plantar surface of the right hind paw in ascending order of force (0.4-60 grams) until the filament bent and was held there for ~3 seconds or until a paw withdrawal response took place. Upon a paw withdrawal response, the Scientific RepoRts \| 6:32096 \| DOI: 10.1038/srep32096 filaments were applied in descending order, beginning with the next thinner filament until there was no withdrawal response. The threshold was the thinnest filament to evoke a paw withdrawal response. The procedure was repeated three times at 5 min intervals to avoid sensitization and the withdrawal thresholds were averaged and recorded (mean ± SEM). Thermal Sensitivity testing (Tail Flick Test). Sensitivity to thermal stimulation was evaluated by the hot water tail-flick test 64 with modification. The temperature of a digital water bath was set up (52 °C) and confirmed with a glass thermometer. Animals were allowed to acclimate to the laboratory environment for 1 h before testing. After subcutaneously injection of morphine or saline, the animals were returned to their cage. For tail flick test, the animals were restrained with a restrainer (Harvard Bioscience, Inc., Whitehall, PA). Approximately 5 min before the test, animals were removed from their cage and allowed to crawl into the restrainer. The tail was marked with ink at one third from the distal tip and submerged in the hot water to the marked level. The cut-off time was limited to 10 s to avoid the tissue damage as measured on a digital laboratory timer. The flick of the tail was recorded as the tail-flick latency in seconds. The test was performed 30 min after morphine injection. Data were calculated as percent maximal possible effect (100% MPE), which was calculated by the following formula: 100% × [(treatment response time -basal response time)/(10 s -basal response time)] = 100% MPE. The group data were expressed as mean MPE ± SEM. The entire behavioral testing was blinded with respect to the treatment groups. Thermal plantar testing (Hargreaves test). Rats were allowed to habituate in the environment for at least 60 min prior to the behavioral test 65 . Each rat was placed in a box (22 × 12 × 12 cm) containing a smooth glass floor (Stoelting, Wood Dale, IL, USA) 62,66 . The temperature of the glass was measured and maintained at 27 °C ± 0.5 °C. A heat source (Stoelting, Wood Dale, IL, USA) was focused on a portion of the hindpaw and a radiant thermal stimulus was delivered. The stimulus shut off automatically when the hindpaw moved or 20 seconds had passed to prevent tissue injury. The intensity of the radiant heat stimuli was adjusted to obtain either short or long baseline latencies. This allowed quantization of the treatment effect (lengthening of the latency, relative to the baseline values and control groups). In this study, latencies for Hargreaves stimuli at baseline ranged from 7 to 11 s. The experimenters were blinded to group assignments. The safety measures. The safety measures included food and fluid intake, locomotion, body weight (g), and liver and renal function tests. Since the locomotor function and food and fluid intake were not noticeably different between different treatment groups in our preliminary experiments, we focused on measuring and recording body weight gain (expressed as mean ± SEM) over time among different experimental groups. In addition, we also measured the levels of blood glucose and alanine aminotransferase (ALT) to monitor the liver function and blood urea nitrogen (BUN) and creatinine to monitor the kidney function. Immunohistochemistry and quantitation of immunoreactivity staining. Histology and immunohistochemistry of microglia and astrocytes in the spinal cord were performed as we have described previously 61,67 . Briefly, animals were sacrificed by intracardiac perfusion with ice-cold PBS, followed by 4% PFA solution under deep anesthesia (Phenobarbital Sodium). The lumbar segment of the spinal cord was rapidly dissected and post-fixed in 4% PFA. It was washed with PBS and placed in cryoprotection buffer overnight at 4 °C until it sank. Afterwards, 30-um-thick coronal sections of the spinal cord were cut on a sliding microtome or cryostat machine (Leica Microsystems) and kept in cryostorage buffer at − 20 °C. Quantitation of immunoreactivity and cell counting. The quantification of percentage area occupied by immunoreactivity, the number of cells in predetermined area of the superficial dorsal horn (number per unit area), or percentage of immunoreactive cells among total cells were performed as previously described 61,67 . Briefly, 40x pictures of the superficial area of the lumbar spinal cord dorsal horn (L4-6) were captured by confocal microscopy and used for quantification. Digitized images were analyzed with National Institutes of Health ImageJ1.34s software. A thresholding procedure was established to determine the proportion of immunoreactive area within each fixed field of view. These parameters were then held constant for each set of images obtained at equal objectives and light intensities. The data represent the mean immunoreactivity area in the spinal cord horn. The percentage of total number of microglial cells was counted in defined area of the spinal cord dorsal horn (Fig. 11). MSCs labeling and tracing in vivo. MSCs were labeled with Dil (Invitrogen, D-3911) according to manufacturer's instructions. In brief, 2 to 3 million of MSCs were collected and incubated with Dil dye (2 mg/ml) at 1XPBS w/o Ca 2+ and Mg2 + buffer (Invitrogen) at 37 °C water bath for 5 minutes with 2-3 times of shaking. Cells were washed with the 1xPBS and re-suspended in 1xPBS for transplantation. Rats lumbar spinal cord segments, L1-L6 DRGs and sciatic nerve were perfused with 4% PFA indicated below after MSC-TP 1, 5 and 8 days. Low magnification pictures (10x) for the whole spinal cord were captured by anatomy fluorescence microscope with digital camera (Leica, Image Core, CCF, LRI). The tissues were cut into 20 uM section to exam Dil histologically. Scientific RepoRts \| 6:32096 \| DOI: 10.1038/srep32096 Statistical analysis. The paw withdrawal thresholds to mechanical stimulation were expressed as mean ± SEM. The withdrawal thresholds to thermal stimulation (Hargreaves test or tail flick test) were used to calculate % MPE as described and expressed as the mean % MPE ± SEM. The use of % MPE took into account the inevitable differences in baseline values between different groups of animals so that such variability would not affect comparisons between different groups. Body weight and blood tests for liver and kidney functions were expressed as mean ± SEM. Immunoreactivity was expressed as mean ± SEM. Statistical analyses were made using one way or two way analysis of variance (ANOVA) followed by paired comparisons with Bonferroni corrections when comparisons were made between more than 3 groups. One way analysis of variance (ANOVA) followed by Turkey's multiple comparison test for immunohistochemistry analysis. Graphpad Prism was used for all the analysis. P < 0.05 was considered statistically significant (*P < 0.05; # P < 0.01; ¥ P < 0.001).	8,535.8	2016-08-24T00:00:00.000	[ "Biology", "Medicine" ]
Assessment and Certification of Neonatal Incubator Sensors through an Inferential Neural Network Measurement and diagnostic systems based on electronic sensors have been increasingly essential in the standardization of hospital equipment. The technical standard IEC (International Electrotechnical Commission) 60601-2-19 establishes requirements for neonatal incubators and specifies the calibration procedure and validation tests for such devices using sensors systems. This paper proposes a new procedure based on an inferential neural network to evaluate and calibrate a neonatal incubator. The proposal presents significant advantages over the standard calibration process, i.e., the number of sensors is drastically reduced, and it runs with the incubator under operation. Since the sensors used in the new calibration process are already installed in the commercial incubator, no additional hardware is necessary; and the calibration necessity can be diagnosed in real time without the presence of technical professionals in the neonatal intensive care unit (NICU). Experimental tests involving the aforementioned calibration system are carried out in a commercial incubator in order to validate the proposal. Introduction Newborns with health complications and premature birth have great difficulty in regulating their body temperature because of various reasons, such as a high metabolism rate caused by the conditions of an illness, low birth weight and a high rate of surface to body volume, which causes a high amount of energy per kilogram to be lost when compared to an adult. Within this context, a neonatal incubator helps in taking care of the health of newborns [1]. A neonatal incubator must be seen as a thermoneutral environment, which provides favorable conditions that assure the minimum energy expenditure of the newborn while body temperature is within a safe range [2]. A neonatal incubator is one of the most important pieces of neonatal intensive care unit (NICU) equipment [3,4]. For every 1,000 Brazilian children born in 2010, about 8.7 individuals died in the first week of life, while 2.6 individuals died between the seventh and 27th day of life, according to the Brazilian Ministry of Health [5]. Within this period of early life, the incubators are fundamental tools for reducing the risk of mortality and diseases. The incubator provides an adequate microclimate, so that newborn infants can overcome the first occurrences of diseases, and it also controls skin temperature and the relative humidity [6]. As with any other electro-medical equipment, a neonatal incubator must be calibrated periodically, because its malfunction may cause serious damage to the newborn's health or even lead to the newborn's death. The technical standard IEC 60601-2-19 establishes operating specifications for neonatal incubators, so that a safe environment can be offered for newborns [7]. Such specifications are verified by performing several tests, including the application of input signals to the temperature, humidity and air flow actuators, with the aim of analyzing the behavior of the aforementioned variables at specific points inside the incubator. In order to perform the aforementioned tests, the incubator is removed from service, and a calibration system is installed, so that temperature and humidity data are provided for a standardized calibration procedure. Considering that it is necessary to turn off the incubator to run the tests and, also, that the number of available incubators in neonatal centers is limited in most Brazilian hospitals, a maintenance strategy that minimizes the incubator down time will prove to be very useful. In this work, a literature survey is carried out by considering the published studies addressing the identification and control of the dynamics involving temperature and humidity in neonatal incubators. Zermani et al. [8] use genetic algorithms (GA) to estimate the parameters of NARMA (nonlinear auto-regressive moving average) and ARX (autoregressive model with external input) models, which identify the dynamic properties of humidity in the newborn incubator. Furthermore, a comparative study was made between a proportional integral derivative (PID) controller and a model-based predictive control (MPC), where the parameters of the PID controller and the cost function of MPC were optimized using GA [8]. Such controllers are used to adjust the humidity inside the incubator. The simulated results presented by the researchers have demonstrated that the MPC controller has a better performance when compared with the PID one. Zermani et al. [9] design an indirect adaptive generalized predictive controller (IAGPC) for the temperature loop in a newborn incubator. In this case, the process is identified by an ARX structure, whose parameters are updated in real time to fit the process changes. The authors establish a comparison among three controllers: (i) ON-OFF; (ii) PID and (iii) IAGPC. Experimental results show that IAGPC is the most efficient controller. In Neto et al. [10], the researchers proposed the application of multivariable PI control strategies for the humidity and temperature control loops in the neonatal incubator. The evaluated TITO (two input two output) system was decoupled into four independent loops. In order to ensure the robustness and stability of the whole control system, PI tuning-based methods, such as modified Ziegler-Nichols and revised BLT (Biggest Log-Module Tuning) were then developed. Amer and Al-Aubidy [11] declared that it is essential to detect any abnormal condition in the premature birth incubator system as soon as possible. In their work, a novel technique using an artificial neural network (ANN) is used in order to simulate adequate incubator control for taking care of premature births. The outputs of the sensors that indicate the temperature, humidity and oxygen concentration of the incubator internal environment are the inputs of an ANN, which identifies the corresponding case and decides the suitable reaction based upon previous training. According to the authors, the ANNs have been consolidated as a powerful tool to be used in the control and identification of dynamic systems. ANNs are massively-distributed parallel structures, consisting of simple processing units known as neurons, which have a natural propensity for storing experimental knowledge and making it available for use [12]. Among the ANN properties, some of the most relevant ones are the intrinsic nonlinearity, the generalization capabilities and adaptability, fault tolerance and the universal approximation capability [13]. In the literature, several studies report applications of ANNs as a potential technique for identification of the dynamical behavior for temperature and humidity inside controlled environments. Although these next briefly described works are not directly related to neonatal incubators, there is some similarity with the scope of this paper. Martnez et al. [14] presented a system based on an ANN for the estimation and prediction of environmental variables related to the tobacco drying process. This system was validated with temperature and relative humidity data obtained from a real tobacco dryer through a wireless sensor network (WSN). In another paper, Ferreira et al. [15] developed an intelligent, light-weight and portable sensor, using ANN models as a time-series predictor system in order to obtain accurate measurements for global solar radiation and atmospheric temperature. This sensor can be applied in several areas, such as in agriculture, renewable energy and energy management or thermal comfort in buildings. ANNs were also applied by Salazar et al. [16] in order to predict the temperature and relative humidity inside an interconnected polyethylene greenhouse with tomato cultivation. The authors performed a feasibility study, which has once again demonstrated the potential of ANNs as an accurate forecasting tool. The inputs for the three neural network models were chosen as the outside temperature, relative humidity, solar radiation and wind speed. In the first two neural models analyzed in the paper, only one output is considered, i.e., temperature or relative humidity. The last model takes into account both variables as outputs at the same time. Various aspects described in the aforementioned papers may be considered in the proposed study. The present work introduces a new method for calibrating neonatal incubators based on multilayer perceptron (MLP) neural network models, with a reduced number of sensors if compared with that used in the traditional calibration procedure. The basic idea is the use of an ANN to infer the temperature and humidity of the extra sensors positioned inside the equipment according to standard IEC 60601-2-19 by using only the sensors included in the commercial neonatal incubators. The proposed method has four significant advantages over the standard calibration process: (i) while the standard technique requires five temperature sensors and one humidity sensor, the proposed method dramatically reduces the number of sensors to one for temperature and one for humidity; (ii) the calibration process works with the incubator under operation; (iii) no hardware calibration is necessary, and the sensors used in the new process are already installed in the traditional incubator; and (iv) the necessity for calibration is diagnosed in real time, and the presence of technical experts in the NICU is not necessary. This paper is organized as follows. The next section presents a brief description of the incubator used in this research and the traditional calibration procedure for such equipment. In Section 3, a summary of the main points established by the IEC standard is presented. The theoretical referential of MLP and inferential systems is presented in Section 4. Finally, the simulations and relevant results are discussed in Section 5, while the main conclusions are given in Section 6. Theoretical Background of Neonatal Incubator and Calibration System A neonatal incubator, which is represented in Figure 1, consists of a rigid box built in fiber and steel, where an infant may be kept in a controlled environment for medical care. The device includes an AC-powered heater, an electrical motor fan to circulate the warmed air, a water container to add humidity, a mechanical filter through which the oxygen flows and an access port for nursing care. The electric motor allows the air to circulate into the neonatal incubator through an air inlet at the bottom of the equipment. It influences the temperature and humidity levels inside the incubator dome, as well as the oxygen level. The air is renewed by a set composed of an electrical exhaust fan and an air inlet. The arrangement is a multivariable system characterized by nonlinearities and interaction among the variables. Within this context, two variables are the most important ones: temperature and humidity, which are regulated by controlling the electric current that flows through two resistors. Temperature and humidity are measured by sensors positioned on the dome. Figure 2 shows a schematic diagram of the overall operation of a neonatal incubator. The relative humidity sensor, SHT11, represented in Figure 3a, is used, which is placed inside the incubator as shown in Figure 3b. SHT11 integrates a sensor element and a signal processor in a tiny footprint, providing a fully-calibrated output. This sensor is composed of two sensor elements: a capacitive sensor to measure the relative humidity (resolution = 0.05%, accuracy = ±3.0%) and a band-gap sensor to measure the temperature (resolution = 0.01 • C, accuracy = ±0.4 • C). According to the manufacturer (Sensirion), CMOS (complementary metal-oxide-semiconductor) technology associated with an analog-to-digital (A/D) converter and a serial interface circuit provides superior signal quality, a fast response time and insensitivity to external disturbances. To perform the standardized calibration procedure, an embedded digital system was designed. The system is composed of a data acquisition platform based on the microcontroller PIC18LF2620 and a data communication wireless link based on ZigBee protocols. The embedded system is connected to a PC (personal computer) through a ZigBee radio link. This embedded system is responsible for monitoring the sensor states and sending commands to the power drivers that control temperature, humidity and oxygen levels and the air flowing inside the neonatal incubator. The main objective of the embedded system is to provide the data acquisition and the communication link with the PC. The communication link is a ZigBee network, which is a low-cost, low-powered wireless network standard (IEEE-Institute of Electrical and Electronics Engineers-802.15.4). The reduced cost allows the technology to be widely used in wireless control and monitoring applications. The low-powered ZigBee devices often transmit data over long distances by transferring data through an intermediate device in a mesh configuration. ZigBee operates in industrial, science and medical radio bands. The prototype uses the ZigBee radio MRF24J40MA (Microchip), which is shown in Figure 4. The designed microcontrolled system is based on standard IEC 60601-2-19. This standard specifies the behavior of the temperature in specific points inside the incubator. The aforementioned points are outlined in Figure 5a and named as A, B, C, D and E. Besides temperature, relative humidity must be also measured in central point A. The mechanical part of the calibration system is comprise of a rigid base made of acrylic, five supports for sensors arranged according to the standard recommendation (given in Section 3) and a control board. At points B, C, D and E, the temperature sensors, MCP9808 (Microchip), are placed. Point A integrates a temperature and humidity sensor, SHT11. The acrylic base with sensors for calibration purposes must be positioned inside the incubator during the tests and removed from the dome when the incubator is under normal operation. The sensor system that integrates the commercial incubator is maintained during the test. Figure 5b presents the experimental setup for the data acquisition system. Overview of Standard IEC 60601-2-19 The parameters for quality and safety in Brazilian neonatal incubators must comply with standard IEC 60601-2-19 "Particular requirements for the safety of neonatal incubator". This standard determines the variation range of environmental variables in the neonatal incubator, such as temperature, humidity, velocity of air flow, and noise level. According to the referred IEC standard, five positions within the incubator are defined to carry out the measurements of temperature and relative humidity, as shown in Figure 6. The standard defines a measurement plane placed 10 cm above of the incubator mattress, where a set of five measurement points (A, B, C, D, and E) exist. Five temperature sensors must be positioned at points A, B, C, D, and E, while the humidity sensor must be positioned at central point A. ccording to standard IEC 60601-2-19, besides the correct positioning of the sensors, a set of six steps must be followed to calibrate the incubator, which are: 1. Before starting the tests, verify if the environment temperature is between 21 • C and 26 • C. 2. Adjust the incubator temperature control (set by the operator) to 11 • C over the environment temperature. The incubator must reach the temperature control at the time instant specified by the manufacturer with a tolerance of up to 20%. The incubator temperature control is measured by the incubator sensor. 3. Adjust the incubator temperature control to two operation points, i.e., 32 • C and 36 • C. At each point, wait for the stabilization temperature condition (the measured incubator temperature at point A must not vary more than 1 • C within the period of one hour). 4. In step number three, the average temperature measured by the sensors at points A, B, C, D and E cannot differ by more than 0.8 • C from the average temperature of the incubator within the period of one hour. Moreover, the average temperature cannot differ more than ±1.5 • C from the incubator control temperature. The incubator average temperature is obtained by means of temperatures measured in regular intervals. 5. With the incubator temperature at 32 • C, set the temperature control reference in steady state to 36 • C and verify if the overshoot is below 2 • C and that the setting is reached in less than 15 min. 6. The humidity value shown by the incubator sensor must not differ by more than 10% of the value measured by the sensor at position A during the whole period of the incubator functioning. The aforementioned procedures must be performed periodically after each maintenance procedure or when the incubator sensors are found to be uncalibrated. The accuracy of temperature and humidity measured by the designed embedded digital system (Section 2) are given by the IEC standard and listed in Table 1, whose specifications are based on certified equipment. The measured temperature must not differ from the standard temperature thermometer in more than 0.8 • C. The accuracy of the standard thermometer must be at least 0.05 • C Measured of humidity inside the incubator. IEC 601-2-19, Section 8, Item 50.110. The relative humidity must not differ more than 10 % from the value measured by a standard equipment. Inferential System Based on MLP Inferential systems (or soft sensors) represent an attractive approach to estimate the primary process variables [17], particularly when conventional hardware sensors are not available or when high cost or technical limitations complicate on-line use [18]. Inferential estimators make use of easily available process knowledge, including a process model and measurements of secondary process variables, so that primary variables of interest can be estimated in real time [19]. Inferential systems are mathematical algorithms that have been used in a wide range of applications in industry, such as supervision, control and optimization of processes due to their advantages over some physical sensors that present measurement problems. Such systems provide faster measurements with more accuracy and reliability at a lower cost in terms of development, implementation and maintenance. According to Warne et al. [20], despite the wide use of inferential models in practical applications, only a few techniques are discussed in the available literature. Generally speaking, there are mainly three approaches for building inferential models: mechanistic modeling (first principles), statistical regression and artificial intelligence modeling. In this work, the intelligent technique of a multilayer perceptron (MLP) neural network is applied to design the inferential system. The MLP neural network architecture consists of a set of neurons arranged in layers, where a layer sends its outputs only to the neurons of the next layer. In addition, all neurons in a given layer are connected to all neurons of the next layer through adjustable parameters of the network, called synaptic weights. These connections are responsible for the MLP's ability to learn and store the knowledge extracted from the problem context. Figure 7 shows the basic structure of an MLP. An MLP has at least one intermediate layer, called a hidden layer. If only one hidden layer is used, the MLP is able to solve problems of low complexity. Cybenko [13] demonstrated that any continuous function can be approximated by an MLP consisting of a single hidden layer with sigmoid activation functions and an output layer formed by linear neurons. The author concludes that there is no need to use more than one hidden layer and/or a mixture of different types of activation functions. However, the use of more sophisticated architectures may be more appropriate in cases of high complexity mappings. The synaptic weights between the MLP layers are adjusted through a process called learning or training. The supervised training process can be applied to the MLP architecture to solve mapping problems. In this learning process, the synaptic weights are adjusted based on a set of experimental input-output pairs of the problem and a cost function [12]. A commonly used cost function in the training algorithm is the MSE (mean squared error). In this case, the algorithm tries to minimize the MSE between the neural network output and the target value over all the training examples. Algorithms, such as backpropagation, Levenberg-Marquardt and the Kalman filter, among others, can be used to minimize the cost function in the learning process of this network [21,22]. A supervised neural network requires an external source to provide information about the analyzed problem. This information is given by a set of N pairs of vectors, {x(n), d(n)}, n = 1, 2, . . . , N , where x(n) ∈ R (p+1) represents the input vector at discrete time n and d(n) ∈ R m denotes the desired vector of outputs (responses) for such an input vector. Each input vector is represented as: . . . where p > 0 denotes the effective number of input variables used in the problem. The term "effective" is used here, because the input component, x 0 (n), is not really a variable in the usual sense, since it has a fixed value equal to −1 and is known as the threshold or (bias). Similarly, the n − th output vector at a discrete time is represented as follows: where m > 0 indicates the number of output variables in the neural network. Any component of the input vector is symbolized by x j (n) ∈ R, while the component of a given vector output is symbolized as d k (n) ∈ R. For a given problem, it is considered that vectors x(n) and d(n) are related according to some unknown mathematical relationship, f (·): For this purpose, one can use a supervised neural network, such as MLP, by adjusting its synaptic weights during the training procedure to generate a mathematical model that acts as an approximation of f (·), denoted byf (·): where one expects output y(n) generated by the neural network to be very close to the desired output d(n), even when the inputs presented to the network are different from those used in the training process. This is possible because of the generalization ability of neural networks. The main goal of this work is to use the supervised neural network MLP to infer the humidity and temperature values at the points specified by IEC 60601-2-19. Thus, in the learning process, vector d(n) represents the real behavior of such variables and y(n) correspond to the values inferred by the network. According to IEC 60601-2-19, the humidity and temperature must be measured by a set of sensors positioned inside the neonatal incubator during its calibration. In the traditional procedure, the incubator must be put out of service, which reduces the working time of this equipment. Within this context, it is expected that the inferential systems can be used to evaluate the incubator conditions and the calibration in an optimized way. It is then reasonable to assume that the inferential system can replace the traditional calibration system based on sensors installed inside the incubator. The proposed neural inferential system adopt the values of humidity and temperature (secondary variables) as inputs, which were instantaneously measured by the sensor, SHT11, available in a commercial incubator. By using measurements and a trained MLP, the system is able to estimate primary variables, which are: humidity at point A and temperature at points A, B, C, D and E, as shown in Figure 6. The general structure of MLP inferential systems is shown in Figure 8, where T s and H s are the measurements of temperature and humidity at the incubator's dome output, respectively, d is the transport delay of the system and, finally, m1 and m2 are the dimensions of the regressors applied to the incubator's secondary variables. At the output of the structure,ŷ i is the estimation of the i-th primary variable provided by the i-th inferential system, according to Table 2. Results and Discussion The results and the methodology used are detailed in the next two subsections. In Section 5.1, the analysis of the results obtained by the neural network is presented. The results are related to the inferential values of the temperature and humidity inside the incubator. In Section 5.2, the procedural steps for the evaluation and calibration of the incubator with the inferential MLP system are presented. Results Obtained by Neural Network For the training and testing of the MLP network, 900 samples of temperature and humidity were acquired from points A, B, C, D and E and the incubator sensor, with sampling periods of 24 s. The collected data were divided into two sets: one for training, containing 750 samples and 150 samples for tests. The samples were collected by applying two PRBS (pseudo-random binary sequence) electric current signals to the incubator resistors. The amplitude limits of the PRBS electric currents are set by IEC 60601-2-19 for humidity and temperature inside the incubator. Main text paragraph. The humidity (H S ) and temperature (T S ) values measured by the incubator sensors at the air outlet of the dome were used as input data in the network, aiming to infer the magnitudes measured by the humidity sensors at point A (H A ) and temperature sensors at points A, B, C, D and E (T A , T B , T C , T D and T E , respectively). Therefore, the IEC 60601-2-19 standardized temperature and humidity points can be determined by the inferential MLP process using only the temperature and humidity values (H S and T S ) measured by the incubator sensor. The values of H S and T S are used to assemble the arrays containing the data to be used as network inputs, which must be associated with the desired outputs, H A , T A , T B , T C , T D and T E . The structures for training and validation are presented in Figure 9. Parameters T S (n), H S (n) and H S (n − 1) are used in Figure 9a to estimate humidityĤ A (n). Figure 9b uses H S (n), T S (n) and T S (n − 1) to estimate humidityT i (n)), where i is the reference temperature at points A, B, C, D and E. Such values are used in the training process of the network based on a backpropagation algorithm [23]. After the training of all neural networks, the next step consists of evaluating the performance of inferential networks through the MSE of the inferred variables with respect to their desired values. For this purpose, a tool known as boxplot is used to qualitatively evaluate the distribution of errors. Boxplot (or box diagram) is a graphical way to represent numerical data aligned horizontally or vertically through five quantities: the smallest observation, lower quartile, median, upper quartile and largest observation [24]. The boxplots also show the outliers, which are represented here by the symbol "+". Outliers are values numerically distant compared to other existing data, which may assume values too small or too large that are located beyond the percentiles. Figure 10a presents the boxplots for the MSE values of variable humidity in the incubator obtained from a variation in the number of neurons in the hidden layer. These results demonstrate that two or four neurons represent the best model. However, the adoption of four neurons is more interesting, because the model with two neurons has outliers. Figure 10b shows the result for variable T a . It can be seen that the boxplots for this variable provide results that are very close to those for variable H a . By analyzing the boxplots in Figure 10c-f, for the temperature at points B, C, D and E, one can notice that there are smaller errors for two, four and six neurons. The choice of six neurons can be disregarded by presenting a median slightly larger than the other choices and the presence of extreme values. Thus, in order to infer the temperatures, it can be stated that the choice of four neurons in the hidden layer is the best option to achieve the best models. Therefore, the test results for the aforementioned structures are given in Figure 11a-f. Table 3, which shows the average MSE and the error variance for both the training and validation of the networks. These values are obtained from the completion of 30 training and validation steps for each variable to be estimated. The results show higher values of MSE for the temperature and humidity at point A. The temperature MSE at this point is 4.18 × 10 −3 and the MSE for humidity is 4.55 × 10 −3 . Point A corresponds to the central position of the incubator, as it tends to present higher nonlinearity in the dynamics by being more susceptible to variations. This is less likely to happen at points B, C, D and E, which are located at the ends of the incubator and are less susceptible to temperature variation. Regarding the analysis of the aforementioned results, the calculation of the relative percentage error between the desired values and the inferred network was performed. In terms of this error, there is the same trend as the MSE, i.e., the humidity and temperature at point A represents a variation of temperature slightly higher if compared to the other specified points. At point A, the moisture presented a relative percentage error ranging from −5% to +5% and a temperature error ranging from −7.5% to +8.0%. At other points, the variation is always within the range of −3.0% to +6.0%, showing that the difference between the desired and inferred temperatures is small. This demonstrates that the MLP achieved satisfactory performance, serving as an effective tool for evaluating the calibration of sensors in the incubator. Assessment of the Proposed Procedure The aforementioned inferential neural network system can be used to implement a software-based calibration system as a substitute or a complementary strategy to hardware calibration. The steps described in the previous section can be summarized as the following instructions in the form of the flowchart presented in Figure 12. When the software detects that the incubator control loops are not within the standard specified range through the MLP inferential sensors, a warning is automatically sent to the NICU computer and to the responsible nurse's mobile phone number, notifying that the incubator needs to be calibrated. This trigger may also occur manually, when the operator detects that the incubator is operating under a nonstandard state. After sending the alert message, the software checks if there is a newborn in the incubator. This check is performed by means of a load sensor. If the presence of a newborn is detected, an alarm signal is triggered, which persists, until the newborn is removed. Then, the software automatically sends a command that starts the calibration procedure of the temperature and humidity loops. Finally, a test is performed to check whether the temperatures at points A, B, C, D and E and, also, humidity at point A are adequate after the calibration process. Numerically, this test results in Tables 4 and 5. According to the results presented in the aforementioned tables, the proposed system achieved a satisfactory performance, satisfying the requirements of the standard IEC 60601-2-19. In other words, the temperature inferred by the MLP network at points A, B, C, D and E does not differ from the neonatal incubator average temperature by more than 0.8 • C, and this temperature does not differ more than ± 1.5 • C from the incubator control temperature. Compared to the established technique, the estimation error is negligible, as shown in the fourth column of Tables 4 and 5. Regarding the humidity, its variation was below 10% of the reference value, which satisfies the IEC 60601-2-19 standard. According to this standard, the reference value for relative humidity is 48%. Conclusions The proposed neural network inferential calibration process of neonatal incubators can be used as an option other than the traditional calibration system based on hardware sensors. This inferential method based on standard IEC 60601-2-19 has shown the advantages of such a method over the hardware calibration process, such as the fact that only one temperature and one humidity sensor are used, that it is not necessary to turn off the incubator to evaluate the incubator condition, that if a more accurate calibration is necessary, it can be detected in runtime, and that during the occurrence of a process warning, alert signals can be sent directly to NICU personnel for repair. Therefore, the intelligent sensor system based on an inferential neural network can reduce the period of the evaluation and calibration of the incubator. Regarding the accuracy of the estimated values, the neural network architecture has shown small errors with respect to the actual values for humidity and temperature. Therefore, it can be stated that this technique allows for the accurate identification of the dynamic model of the incubator, which is capable of providing estimation in possible advanced risk cases, such as the temperature being out of a safe range and failures in the heater system and/or air flow in the equipment. With the use of the incubator over the years, it may be required that an MLP retraining be performed, due to natural changes in the operational conditions of this equipment. This is supposed to be the unique limitation of the proposed system. This limitation could be overcome with the implementation of a more sophisticated self-training MLP algorithm.	7,540.4	2013-11-01T00:00:00.000	[ "Computer Science" ]
Cyclin-dependent Kinases Phosphorylate the Cytomegalovirus RNA Export Protein pUL69 and Modulate Its Nuclear Localization and Activity* Replication of human cytomegalovirus (HCMV) is subject to regulation by cellular protein kinases. Recently, we and others reported that inhibition of cyclin-dependent protein kinases (CDKs) or the viral CDK ortholog pUL97 can induce intranuclear speckled aggregation of the viral mRNA export factor, pUL69. Here we provide the first evidence for a direct regulatory role of CDKs on pUL69 functionality. Although replication of all HCMV strains was dependent on CDK activity, we found strain-specific differences in the amount of CDK inhibitor-induced pUL69 aggregate formation. In all cases analyzed, the inhibitor-induced pUL69 aggregates were clearly localized within viral replication centers but not subnuclear splicing, pore complex, or aggresome structures. The CDK9 and cyclin T1 proteins colocalized with these pUL69 aggregates, whereas other CDKs behaved differently. Phosphorylation analyses in vivo and in vitro demonstrated pUL69 was strongly phosphorylated in HCMV-infected fibroblasts and that CDKs represent a novel class of pUL69-phosphorylating kinases. Moreover, the analysis of CDK inhibitors in a pUL69-dependent nuclear mRNA export assay provided evidence for functional impairment of pUL69 under suppression of CDK activity. Thus, our data underline the crucial importance of CDKs for HCMV replication, and indicate a direct impact of CDK9-cyclin T1 on the nuclear localization and activity of the viral regulator pUL69. Human cytomegalovirus (HCMV) 2 is a member of the Herpesviridae family and a human pathogen with worldwide distribution. Primary HCMV infection of the immunocompetent host is usually asymptomatic, whereas severe disease can occur upon infection of the immunocompromised and immunonaive. HCMV is a leading cause of complications in transplant recipients and AIDS patients, and congenital infection may result in mental impairment and hearing loss (1). HCMV replication is differentially regulated in host cell types, and viral replication is dependent on regulation of the cell cycle (2). HCMV infection induces cell cycle arrest, while simultaneously the virus sustains an active cellular metabolic state supporting productive infection (3). Infected cells arrest in a pseudo-G 1 state with high levels of cyclin E and cyclin E-associated kinase activity (4 -6). A number of additional alterations of cyclin-dependent protein kinase (CDK) activity have also been described, such as increased synthesis and reduced degradation of cyclin B1, as well as cytoplasmic translocation of CDK1 in HCMV-infected cells (7). The up-regulation of CDK activity during HCMV replication implies that viral replication requires CDK activity to create an environment favorable for efficient viral transcription, genome replication, and assembly of viral particles. Several regulatory steps in HCMV replication are dependent on CDK activity, particularly those involving CDK1, -2, -7, and -9 (8 -12). Additionally, inhibition of CDK activity affects replication of HCMV and other herpesviruses (13). Roscovitine, a purine analog that preferentially inhibits CDK1, -2, -5, -7 and -9, has been shown to decrease viral DNA synthesis and production of late viral protein and infectious virus (8,9,12,14). Roscovitine is therefore a useful tool to investigate the impact of CDK activity on viral replication and to understand inter-regulation between CDKs and viral proteins. Cross-talk between CDKs and other protein kinases during HCMV replication is one issue of current interest (15). CDKs, particular serine/threonine kinases that become activated upon binding to cyclins, are involved in the regulation of multiple cellular processes. They can be subdivided into two major functional groups, cell cycle-associated CDKs and transcriptionally regulating CDKs. A prototype of the transcriptionally regulating CDKs is the positive transcription elongation factor b (P-TEFb), which is composed of CDK9 and cyclin T1 (cycT1). This complex is an important regulator of transcription through phosphorylation of the C-terminal domain of the large subunit of RNA polymerase II, thus allowing for transcription elongation (16). The expression of many genes is regulated at the level of transcription elongation, and the activity of the P-TEFb complex is tightly controlled. For example, the association of 7SK small nuclear RNA and HEXIM1 acts as an inhibitor to P-TEFb (17)(18)(19)(20), whereas autophosphorylation of phospho-acceptor sites at the CDK9 C terminus acts to stimulate and promote nuclear translocation of the P-TEFb complex (21). Among the viral proteins identified as substrates of pUL97, the pluripotent regulator pUL69 appears functionally relevant. pUL69 acts as a transcriptional activator (34,35), a nuclear mRNA export factor (36), and a mediator of cell cycle arrest (37,38). Recent studies show pUL69 binds RNA, has nucleocytoplasmic shuttling activity, and recruits the cellular mRNA export machinery via interaction with the cellular mRNA export factor UAP56/URH49. This latter activity promotes cytoplasmic accumulation of unspliced mRNA (36,39,40). pUL69 is a phosphoprotein subject to phosphorylation by the pUL97 viral kinase (41), although it is still unclear whether CDKs also play an important role in its phosphorylation. Against this background, it is significant that the CDK inhibitor roscovitine influences the intranuclear localization of pUL69 in HCMV-infected fibroblasts by changing pUL69 homogeneous nuclear distribution toward speckled aggregation (9). In this study, we provide evidence for direct targeting of pUL69 by CDKs, which modulates pUL69 nuclear localization and activity. Findings in support of this concept are the speckled nuclear aggregation of pUL69 induced by CDK inhibitors, colocalization of CDKs and pUL69 in HCMV-infected cells, and the direct in vitro phosphorylation of pUL69 by CDK-cyclin complexes. qPCR-Quantitative real time PCR (qPCR) was performed (49) using the ABI Prism 7700 sequence detector (Applied Biosystems, Foster City, CA) and corresponding software SDS (sequence detection system version 1.9). qPCR was used to quantify viral genome copies normalized against cellular genome equivalents in initial viral stocks, and supernatant was taken at various time points post-infection. Total DNA was extracted from virus-infected cells using the Wizard DNA purification kit (Promega, Mannheim, Germany). qPCR was performed in a 25-l reaction mixture containing 5 l of either the sample or the standard DNA solution. Additional components of the reaction mixture were 12.5 l of 2ϫ TaqMan PCR Mastermix (Applied Biosystems), 7.5 pmol of each primer complementary to a region within exon 4 of the IE1 gene locus (5Ј-AAGCGGCCTCTGATAACCAAG-3Ј and 5Ј-GAGCA-GACTCTCAGAGGATCG-3Ј), and 5 pmol of probe directed against the HCMV MIE region exon 4 (5Ј-CATGCAGATCTC-CTCAATGCGCGC-3Ј). To calculate the ratio of viral DNA per cellular DNA equivalent, the cellular DNA was quantified in parallel using primers complementary to a region within the albumin gene locus (5Ј-GTGAACAGGCGACCATGCT-3Ј and 5Ј-GCATGGAAGGTGAATGTTTCAG-3Ј) together with an albumin gene-specific probe (5Ј-TCAGTGGAAGAT-GAAACATACGTTC-3Ј). Both probes used were labeled with 6-carboxyfluorescein reporter dye and 6-carboxytetramethylrhodamine quencher dye. The DNA standard for quantification of viral genome copies was prepared by serial dilutions of plasmid pHM123 containing the IE1 cDNA (50). The cellular standard was composed of albumin PCR products obtained from lymphocyte cell extracts (51). The thermal cycling conditions consisted of two initial steps of 2 min at 50°C and 10 min at 95°C followed by 40 amplification cycles (15 s 95°C, 1 min 60°C). DNA extracts were analyzed in triplicate for each sample. Yeast Two-hybrid Analysis-Protein interactions were analyzed using GAL4 fusion proteins (GAL4-BD, DNA binding domain; GAL4-AD, activation domain) in the yeast two-hybrid system as described previously (24). An expression plasmid for cyclin T1 was provided by L. Lania (52). Expression plasmids for CDK9 and cycT1::CDK9 were subcloned from pACTII-T1 and pRc/CMV-PITALRE-HA (52,53). Saccharomyces cerevisiae strain Y153 was used for interactor analysis, and the selection of clones was achieved by cultivation on media restricting growth to combined tryptophan/leucine prototrophy. Selected colonies were analyzed for ␤-galactosidase activity by filter lift tests. In Vivo Labeling Assay-Proteins were labeled in vivo in HCMV-infected HFFs by incubation with [␥-33 P]orthophosphate (150 Ci/4.5 ϫ 10 5 cells) in the culture media. Cells were lysed in RIPA buffer and supernatants subject to immunoprecipitation (pAb-UL69). Samples were analyzed by SDS-PAGE/ Western blot as described for the in vitro kinase assay. Nuclear mRNA Export Assay for pUL69-A nuclear mRNA export assay, based on the export activity of recombinantly expressed pUL69, was performed with lysates from transfected HeLa cells as described previously (39). CAT reporter assays were performed as described by Farjot et al. (54). The plasmid construct pDM128/CMV/RRE, encoding an intron-containing mRNA with the coding sequence of CAT, was used as a reporter of nuclear export activity. CDK inhibitors were added 16 h post-transfection. CAT protein expression was quantified following cell lysis at 48 h post-transfection and analyzed in triplicate using a CAT enzyme-linked immunosorbent assay (Roche Applied Science). HCMV Replication Is Regulated by CDK Activity, CDK Inhibitors Induce a Pronounced Aggregate Formation of pUL69 That Is Quantitatively Different for Variants of HCMV-As described previously, the intranuclear localization of the viral regulator pUL69 is significantly altered in the presence of the CDK inhibitor roscovitine, with the formation of speckled nuclear aggregates induced in the late phase of replication (9). This phenomenon was detectable for several strains of HCMV such as Towne (9), AD169 (41) (Fig. 1A), and clinical isolates. 3 For strain AD169, roscovitine treatment produced pUL69 aggregation in 67.3% of virus-positive cells (Fig. 1C). This phenotype varied in quantitative terms between AD169-derived virus variants GDGrXbaF4, GDGrP53, and 759rD100. As depicted in Fig. 1B, GDGrXbaF4 and 759rD100 contain a GCV resistance-conferring mutation in open reading frame UL97 (deletion 590 -593) (44). GDGrP53 and 759rD100 contain a point mutation in open reading frame UL54 (A987G) resulting in cidofovir resistance (43,45) and high level GCV resistance (Fig. 1D, GCV). When analyzing the nuclear localization of pUL69, variants GDGrXbaF4 and GDGrP53 were similar to AD169 with pUL69 aggregate formation in the presence of roscovitine (Fig. 1A, panels a-d and f-i). However in quantitative terms, the pUL69 aggregation of GDGrP53 was significantly reduced (38% of virus-positive cells, p Ͻ 0.01; Fig. 1C) compared with the parental strain AD169. As shown for AD169 and the variants GDGrXbaF4 and GDGrP53, an additional CDK inhibitor, R58, also induced a pUL69 phenotype similar to that induced by roscovitine (Fig. 1A, panels k-n). R58 is a strong inhibitor of CDK2, CDK5 (IC 50 Ͻ 1 M in vitro), and possibly other CDKs. The double mutant 759rD100 behaved differently, showing very little speckled aggregation of pUL69 in infected cells treated with roscovitine or R58 (Fig. 1A, panels e, j, and o). Quantitative immunofluorescence analysis demonstrated only 15% of 759rD100-infected fibroblasts showed pUL69 aggregates under roscovitine treatment, which represented a highly significant reduction (p Ͻ 0.0001; Fig. 1C). Similar results were obtained with inhibitor R58. The UL97/UL54 mutant 759rD100 showed reduced replicative sensitivity toward roscovitine, demonstrated on plaque reduction assay (Fig. 1D, Rosco), but showed a strong sensitivity toward the second CDK inhibitor R58. This indicated that R58, although inefficient in inducing pUL69 aggregates in 759rD100-infected cells, mediated an inhibitory effect on viral replication, albeit through a mode of action that seemed independent from pUL69. The question whether observed differences between HCMV variants were due to major alterations in viral replication characteristics was addressed by qPCR. HFFs were infected with DNA-normalized viral stocks (i.e. viral DNA copy numbers per cell eq), and the kinetics of viral genomic DNA synthesis were determined as depicted in supplemental Fig. S1. The genomic replication curves of AD169 and GDGrXbaF4 were very similar, whereas 759rD100 and GDGrP53 showed a clear replication deficit throughout the period analyzed. This most likely can be attributed to the pUL54 DNA polymerase mutation of 759rD100 and GDGrP53. However, the qPCR data cannot fully explain the roscovitineinduced pUL69 aggregation phenotypes of the virus variants. Whereas GDGrP53 showed the lowest replication efficiency, 759rD100, but not GDGrP53, showed a clear lack of pUL69 aggregate formation. Thus, the phenotype peculiarity of variant 759rD100, with its double mutation in pUL54 DNA polymerase and pUL97 protein kinase, needs to be further analyzed on a mechanistic basis. In this context, it was interesting to observe that a known pUL97 kinase inhibitor, Gö6976, but not an unrelated tyrosine kinase inhibitor, AG490, produced a pattern of speckled pUL69 aggregate formation very similar to roscovitine or R58. Parental AD169 and variants GDGrXbaF4 and GDGrP53 showed a pronounced Gö6976-induced pUL69 aggregation (supplemental Fig. S2, t, v, and x), although very little aggregate formation was observed for 759rD100 in response to Gö6976 treatment (supplemental Fig. S2, z; statistically significant, p Ͻ 0.01). Of note, pUL97 was never observed in colocalization with pUL69 aggregates but remained in a nonspeckled, homogeneous nuclear distribution. Additionally, the presence of Gö6976 led to some exclusion of pUL97 from viral replication centers (supplemental Fig. S2, c and u), an effect that had been described before (55). Thus, inhibition of pUL97 as well as CDKs can induce the pUL69 aggregation phenotype in several variants of HCMV. To narrow down the number of CDKs associated with the formation of pUL69 aggregates, a series of novel inhibitors with strong inhibitory potential against CDKs in vitro were utilized. These inhibitors fell into two groups with respect to their ability to produce pUL69 speckled aggregates: five compounds (A14, A43, A79, R25, and R58) induced strong effects comparable with roscovitine, whereas two other compounds (A50 and A98) failed to alter pUL69 distribution (supplemental Table S1). These CDK inhibitors all share a strong inhibitory potential against CDK1 and CKD2 in vitro (IC 50 Ͻ 1 M) and possibly further inhibitory effects against other CDKs. Thus, the pattern of CDK inhibition required for pUL69 aggregation could not be deduced from this experiment. However, it is highly suggestive that CDK1/2 inhibition is not sufficient to confer the phenotype of pUL69 aggregation, and additional inhibitory activity is required. CDK Inhibitor-induced Intranuclear Aggregates of pUL69 Are Localized within Viral Replication Centers-To characterize the speckled aggregation of pUL69 more closely, HCMVinfected fibroblasts were analyzed under roscovitine treatment by costaining of pUL69 with a selection of viral and cellular nuclear proteins (Fig. 2). Viral DNA polymerase pUL54 and its processivity factor pUL44 are prominent markers of viral replication centers, and pUL69 is typically also detectable within these compartments. Following infection of HFFs with HCMV AD169 in the presence of roscovitine, pUL44 (Fig. 2, c and g) and pUL54 (data not shown) did not alter their localization, but markedly, the speckled aggregates of pUL69 fully localized within the area of replication centers (Fig. 2, e-h). This suggests an accumulation of the replication center-associated fraction of pUL69 into subnuclear speckles under conditions of inhibited CDK activity. Additionally, further types of prominent intranuclear structures were examined, such as splicing compartments (marked by splicing factor SC-35), the nuclear pore complex (marked by NUP62/152/90), or aggresome structures (marked by heat shock cognate protein 70, HSC70 (56)). As illustrated in Fig. 2, neither SC-35 (k and o), NUP62/152/90 (s and w), nor HSC70 (data not shown) displayed alterations in their localization or colocalized with pUL69. CDK Inhibitor-induced Aggregates of pUL69 Colocalize with CDK9 and Cyclin T1-The investigation of CDK distribution patterns in HCMV-infected HFFs under inhibitor treatment showed a specific association of CDKs with pUL69 speckled aggregates ( Fig. 3A and supplemental Fig. S3). CDK1, -2, -7, and -9 were analyzed under roscovitine, R58, or Gö6976 treatment. AG490 and Gö7874, which affect neither pUL97 nor CDK activity, served as specificity controls. Strikingly, CDK9 was found to undergo changes in localization similar to pUL69. CDK9 accumulated in replication centers of HCMV-infected cells as shown by colocalization with pUL69 in the absence of inhibitor (Fig. 3A, panels e-h). This colocalization was further developed in the presence of roscovitine or R58, i.e. CDK9 also aggregated in the form of nuclear speckles (Fig. 3A, panels i-p). CDK9 aggregation was only marginally detectable in the presence of the pUL97 inhibitor Gö6976 (Fig. 3A, panels q-t). Both controls, Gö7874 (Fig. 3A, panels u-x) and AG490 (data not shown), had no impact on the localization of pUL69 and CDK9. In a next step, the regulatory subunit of CDK9, cyclin T1 was analyzed. As depicted in Fig. 3B, cyclin T1 was not only recruited to HCMV replication centers (panels e-h) but also formed speckled aggregates in colocalization with pUL69 under treatment with roscovitine (panels i-l) or R58 but not Gö6976 (data not shown). These findings strongly suggest that although CDK inhibitors as well as pUL97 inhibitors induce a very similar speckled aggregation of pUL69, the composition of the structures and the underlying mechanisms are different. Thus, roscovitine-mediated aggregation is likely to be associated with CDK9-cycT1 activity, although Gö6976-mediated aggregation appears to be CDK-independent. Interestingly, closer investigation of protein-protein interactions by yeast two-hybrid analysis revealed direct interactions between pUL69 and cyclin T1 as well as a cycT::CDK9 fusion construct, but not with CDK9 alone (Fig. 3C). Signal intensity of the positive scores of the filter lift staining remained at a mod- erate level compared with the positive control (CDK9 and cyclin T1) indicating a dynamic mode of low affinity interaction between pUL69 and cyclin T1. A putative colocalization between pUL69 and other CDKs was further analyzed. CDK7, functionally related to CDK9, accumulated in replication centers similar to CDK9 and cyclin T1 in HCMV-infected cells (supplemental Fig. S3, e-h). However, neither CDK nor pUL97 inhibitors induced speckled aggregation of CDK7 (supplemental Fig. S3, i-t). For CDK2, HCMV infection did not lead to an incorporation into viral replication centers or any other detectable changes in intranuclear localization. Also, the addition of inhibitors had no impact on the pattern of nuclear localization of CDK2 (data not shown). On the other hand, a previously described translocation of CDK1 from the nucleus to the cytoplasm was detected in HCMV-infected cells (7). Hence, colocalization between CDK1 and pUL69 was not detected in the presence or absence of inhibitors. Phosphorylation of pUL69 in Vitro and in Vivo-The strong inter-regulation of CDKs with the nuclear localization of pUL69 initiated investigation of CDK-mediated phosphorylation of pUL69. To this end, in vitro kinase assays were performed to analyze the ability of recombinant CDK-cyclin complexes (CDK1-cycB1, CDK2-cycE, CDK7-cycH-MAT1, and CDK9-cycT) to phosphorylate pUL69 immunoprecipitated from transfected 293T cells. The activity of the CDK-cyclin complexes were confirmed via phosphorylation of a reference substrate, RB-CTF (data not shown). Importantly, a clear signal for direct phosphorylation of pUL69 by CDK1-cycB1, CDK7-cycH-MAT1, and CDK9-cycT was detected (Fig. 4A, upper panel, lanes 1, 3, and 4). Nonspecific phosphorylation activity was excluded by the lack of measurable phosphorylation of HCMV pUL26, used as a specificity control (Fig. 4A, lower panel). Low level base-line phosphorylation of pUL69 could be detected without addition of CDK-cyclin complexes, which probably indicates traces of pUL69-phosphorylating kinase activity in the immunoprecipitates (Fig. 4A, upper panel, lane 5). The fold increase in phosphorylation of pUL69 by CDKcyclin complexes was then determined via densitometry. As shown in Fig. 4B, the strongest phosphorylation was mediated by CDK1-cycB1 (12.4-fold increase). Pronounced levels of phosphorylation were also measured for CDK9-cycT (7.2-fold) and CDK7-cycH-MAT1 (6.4-fold), whereas CDK2-cycE-mediated phosphorylation was lower (4.1-fold). Thus, these in vitro data indicate that direct phosphorylation of pUL69 can be mediated by several CDK-cyclin complexes. To assess phosphorylation of pUL69 in vivo, we infected HFFs with HCMV AD169 for 2 days and incubated cells with [␥-33 P]orthophosphate to allow for in vivo labeling of proteins. An evaluation of pUL69 immunoprecipitated from these cells revealed a strong signal of phosphorylation (Fig. 4C). Phosphorylation could be partly inhibited by the treatment of infected cells with 15 M roscovitine (signal reduction of ϳ66 Ϯ 13%; data not shown). However, this inhibition of phosphorylation did not occur in a CDK-specific manner and was also observed for other protein kinase inhibitors. This points to a complex regulation of the phosphorylation of pUL69. Thus, the in vivo phosphorylation of pUL69 appears to be dependent on CDK and other protein kinase activities. The particularly strong CDK1-cycB1-mediated phosphorylation of pUL69, as demonstrated by in vitro data (Fig. 4, A and B), raised questions about the nucleo-cytoplasmic translocation of CDK1. We determined whether CDK1 was detectable in pUL69-positive nuclei of HCMV-infected fibroblasts by performing immunofluorescence analysis, including confocal laser-scanning microscopy. For this purpose, a kinetic study was performed to investigate localization patterns during the immediate early and early phases of HCMV replication (supplemental Fig. S4). Nuclear pUL69 was observed from 4 hpi, and the percentage of pUL69-positive cells increased continu-ously over time. Under roscovitine treatment, a transient delay of pUL69 expression was detected (supplemental Fig. S4A), with reduced immunofluorescence signal intensities confirming a slightly lower level of pUL69 production. Interestingly, the beginning of nucleo-cytoplasmic translocation of CDK1 was observed at 8 hpi in both roscovitine-treated and untreated cells (supplemental Fig. S4B). Over the period analyzed, the fraction of pUL69-positive cells showing CDK1 in a nucleo-cytoplasmic or cytoplasmic localization steadily increased, with a completion of the translocation at about 24 hpi. Of note, this translocation was slightly retarded through the inhibition of CDK activity by roscovitine. Thus, although a direct colocalization between CDK1 and pUL69 was not detectable, the presence of both CDK1 and pUL69 in the nuclei of HCMV-infected fibroblasts may allow an inter-regulation of the two proteins at early time points of infection. Inhibition of CDK Activity Reduces mRNA Export Activity of pUL69-A nuclear mRNA export assay was performed to investigate functional aspects of pUL69. This assay determined the ability of pUL69 to export intron-containing CAT mRNA. As shown in Fig. 5, a decline in nuclear export activity was observed when CDKs were inhibited by either roscovitine or R58. The pUL69-mediated export signal was reduced to 43% under roscovitine treatment (statistically significant, p Ͻ 0.01) and to 63% under R58 treatment. An inhibitory effect on the pUL69 nuclear export function was also detected for the pUL97-directed inhibitor, Gö6976 (41). Thus, CDK as well as pUL97 activity is required for the full functionality of pUL69 with regard to mRNA export. DISCUSSION The HCMV replication strategy has evolved to an elaborate inter-regulation with factors controlling the cell cycle. On the one hand, HCMV ensures that the regulatory state of the cellular environment efficiently supports viral reproduction, and on the other hand, HCMV reprograms the cellular factors such as regulatory protein kinases from their original function toward virus-specific regulatory pathways (2,3). A number of studies have shown that HCMV replication is functionally linked with CDK activity at various regulatory junctures. In this study, we provide novel insights into the link between cellular CDK activity and the intranuclear localization and functionality of the viral mRNA export factor pUL69. Our findings indicate the following: (i) HCMV-infected fibroblasts treated with CDK inhibitors show an intranuclear speckled aggregation of pUL69. (ii) Variants of HCMV are differentially sensitive to inhibitors inducing pUL69 aggregation. (iii) Speckled pUL69 aggregates are mainly localized within viral replication centers. (iv) CDK9 and cyclin T1 strictly colocalize with the inhibitor-induced speckled aggregates, whereas other CDKs behave differently. (v) The HCMV-triggered nucleo-cytoplasmic translocation of CDK1 does not exclude a putative early nuclear interaction with pUL69. (vi) pUL69 is phosphorylated in vivo and in vitro, identifying CDKs (mainly CDK1 and -9) as novel pUL69-phosphorylating kinases. (vii) CDK activity is required to stimulate a high level of nuclear mRNA export activity of pUL69 in a reporter assay. The importance of CDK activity in the replication cycle of HCMV has been well documented (3, 4, 6 -9, 11, 12). However, for most of these investigations, the description of molecular mechanisms linking CDK activity with viral regulation of replication was still unresolved. We have identified the viral regulatory protein pUL69 as one target of CDK-mediated regulation. This protein contributes to an HCMV-induced cell cycle arrest that may result from interaction with CDKs and/or cyclins but is poorly understood so far (37,38). Additionally, pUL69 acts as a transcriptional transactivator via interaction with the cellular transcription elongation factor hSPT6 (57,58) and as a nuclear RNA export factor via interaction with UAP56, a component of the cellular mRNA export machinery (36,39,40). As pUL69 is a phosphoprotein (35,57), it has been speculated that its activity might be partly regulated through phosphorylation. In this study, we provide evidence for the phosphorylation of pUL69 by CDKs. pUL69 acted as a specific substrate in in vitro kinase reactions with all four analyzed CDKs (CDK1, -2, -7, and -9), whereby CDK1 and -9 exerted the highest pUL69-phosphorylating activity. Our data point to a combined impact of more than one CDK on the phosphorylation and activity of pUL69, based on findings that CDK1-cycB1 and CDK9-cycT exerted main activities in a pUL69-specific in vitro kinase assay, whereas CDK9-cycT exclusively showed colocalization with pUL69 during late phase of infection. In addition, both CDK inhibitors roscovitine and R58 induced speckled aggregates of pUL69, although they possess partly different inhibitory profiles toward individual CDKs. Thus, it remains speculative which of the analyzed CDKs are the key determinants for pUL69 regulation. The impact of CDKs on HCMV replication may be an ordered sequence of events with pUL69-directed activity of CDK1 an early event during viral replication. Consistent with this, both pUL69 and CDK1 were localized in the nucleus prior to CDK1 nucleo-cytoplasmic translocation. This illustrates that CDK1, although not colocalizing with pUL69, may contribute to the regulatory phosphorylation of pUL69 at early time points of infection. In contrast, CDK9 might be required at later time points for regulation of pUL69 activity, as demonstrated by direct colocalization of CDK9-cycT and pUL69 in late phase replication centers. Integrating these findings, this indicates a regulatory impact of CDK9 and CDK1 and, possibly, further CDKs on the functionality of pUL69. Interestingly, the pUL97 viral protein kinase was characterized as a CDK-related kinase possessing similar functional properties (23). Our studies of other protein kinases involved in the phosphorylation of pUL69 was compatible with these findings. Recently, we provided evidence that the CDK ortholog pUL97 phosphorylates pUL69 (41). Combined with the data of this present study, we hypothesize that cellular CDK and viral pUL97 activities are required to modulate the nuclear localization and function of pUL69 during cytomegalovirus replication. There is only a limited number of examples describing functional cross-talk between CDKs and herpesviral protein kinases. One prominent example is the sequentially ordered inter-regulation between CDK1-cdc25C and herpes simplex type 1 (HSV-1)-encoded protein kinases UL13 and US3 (59,60). In this case, the two HSV kinases, UL13 and US3, are capable of phosphorylating, and thereby activating the regulatory cellular phosphatase cdc25C, which normally activates CDK1 (cdc2), by removing two inhibitory phosphates. However, in HSV-1infected cells the function of cdc25C is reduced to modulating CDK1 function. In infected cells, CDK1 acquires a new binding partner, the HSV DNA polymerase processivity factor UL42. The CDK1-UL42 complex then functions as an activator of late viral gene expression. In addition, CDK1 is able to phosphorylate UL42 and therefore possibly stimulate its activity in the UL42-DNA polymerase complex (61). In summary, our data imply that CDK-specific regulatory pathways modulate the multiple functions of pUL69 in HCMV-infected fibroblasts. Further studies will be required to gain a deeper insight into the respective molecular mechanisms and to learn more about contact points of the CDK-HCMV interaction.	6,154.4	2009-03-27T00:00:00.000	[ "Biology" ]
Simulation of a Multi-Carrier System in a Non-Linear Flat Fading Channel This paper is on multi-carrier wireless communication system, focusing on the physical layer in such modulation system. Multi-Carrier MC system has received much attention in modern communication technology and is finding attractive applications. The paper introduces and discusses the discrete cosine transform DCT-based MC. Similarities and differences with respect to the Fourier transform-based system are pointed out. It was found that a reduction up to 5dB in peak power can be gained over the conventional system. Also, the simulation over flat fading channel depicted that QAM signaling in MC scheme based on DCT performed better than that in FFT-based scheme, although it depends strongly on the channel parameters. Introduction The Multi-Carrier MC technology has received much attention in modern communication systems.In MC systems a number of data symbols are transmitted at separated sub-carriers in parallel thus increasing the symbol length, which reduces the sensitivity to interference.Multi-Carrier Modulation MCM divides the available frequency band into a large number of orthogonal tones which can be implemented in all digital realization by exploiting Discrete Fourier Transform DFT methods as in DMT and OFDM multi-carrier modulation Hnzo et al. 2004;Reimers 2001 andAl-Dhahir andMinn, 2006).Discrete Multi-tone DMT a currently standard for Asynchronous Digital Subscriber Line ADSL units manages bandwidth for data transmission.It is one of the technologies that provides high speed Internet access in residence and offices economically via wire technology (Gagnaire, 2003 andDaly 2003).Orthogonal Frequency Division Multiplexing OFDM gives high resistance to fre-_____________________________________________ Corresponding author's e-mail<EMAIL_ADDRESS>selective fading.It is often used in mobile radio systems such as digital video broadcasting DVB/ digital audio broadcasting DAB (Reimers, 2001 andAl-Dhahir andMinn, 2006).It is proposed for 4 th generation wireless communication systems (Bria, 2001).However, due to the large number of sub carriers, the MC signal owns a large Peak to Average Power Ratio PAPR.Thus, the transmitted signal is sensitive to non-linear distortions that will degrade the error performance and introduce high adjacent channel interference with spectrum regrowth.Several techniques for reducing PAPR of multi-carriers signal have appeared in the literature (Mustafa, 2007 andFriese, 1996).However, most of these techniques introduce additional complexity.Moreover, there have been intense research efforts aimed at designing different multi-carrier transceiver scheme with different spectral containment.Among those designs are the Discrete Cosine Transform DCT-MCM (Al-Dhahir andMinn, 2006 andAk-Dhahir andMinn, 2005) and Discrete Wavelet Multi-Tone DWMT (Daly, 2003 andFarhang-Boroujeny andChin, 2000). In this paper DCT-based multi-carrier system was con- .ƒà°ùeƒÑN hP IÉaeb ^ ¢S Ω ΩÉ ¶f ' QG »H …G »H sidered.Many simulations have been carried to outline the differences over linear AWGN, non-linear AWGN and a flat faded channel with respect to the DFT-based system.Section II reviews in brief the properties and representations of the DCT, and section III depicts DCT-based scheme.Simulation results and some further discussions of the error performance over different channel parameters are presented in section IV.Finally, section V concludes the results. Discrete Cosine Transform The DCT has been considered as one of the best tools in digital signal processing and therefore, it has many applications, e.g., in the area of multimedia and telecommunications.In this section, an introduction to the DCT is given in order to provide background and motivation for our work. The orthogonal DCT is classified into many different types with slightly different even/odd boundary conditions at the two ends of the matrix.The first definitions of the forward and inverse DCT were given by: The DCT operators map an N-size real sequence into another N-size real sequence.It is a linear Fourier-related transformation similar to the DFT using only real numbers of the DFT.By repeating the samples in a time reversed order and performing a DFT on the length sample set a DCT is obtained.Also, if we represent the N-size sequence as X=[x 1 ,…., x N ] and Y=[y 1 ,…., y N ] and denote the NXN DCT matrix by: (2) Thus Eq. 1 can be rewritten as Y=XC and X=YC -1 .DCT is computationally simpler and the DCT matrices are orthogonal, ie.CC T =I, the inverse transform matrix are obtained with a matrix transpose (where T denotes transposition and I is the identity matrix) (Strang, 1999;Oppenheim, et al. 1999 andRao andYip, 2003).In MC-based transceiver, the binary data is encoded into a set of M-PSK or M-QAM symbols, called sub-symbols, and converted into lower rate sequences via serial to parallel conversion, which are then multiplexed by an IDCT.The outputs are serialized and transmitted.At the receiver, inverse operations take place in reverse order.In Fourier-based system, the Hermition symmetry is enforced to ensure a real waveform at the output of the IFFT.This can be achieved by conjugate mirroring the complex symbols and transmitting zero signals on the DC and Nyquist tones of IFFT block sequence.However, as complex signals are mapped onto orthogonal sub-channels without imposing the conjugate symmetry condition, the multi-carrier system transmits twice the data in twice the bandwidth required when the condition is imposed.Also, it is not necessary in our proposed system, as even a real signal can be mapped onto orthogonal sub-channels. MC System Model DCT converts real signals to real signals (as depicted earlier), and hence binary signaling or parsed in-phase and quadrature components of complex data word must be used in each sub-channel individually to keep the same data rate in DCT systems. Simulation Results Much of this paper compares the error performance of DCT-based MC system, which transmits a real waveform with that of the Fourier-based; therefore the conjugate symmetry condition is imposed on FFT-based system under the same transformation size N=256 to provide fairest comparison possible.In the simulation we also considered a system operating without guard space in the resulted time domain symbols. The scheme was simulated in Matlab ver.7 with a constant bit allocation of log 2 (M) bits per sub-channel is assumed.The number of simulated bits was about 10 6 bits and the results was computed and averaged over 3 iterations for the demodulator. Linear AWGN Channel The error performance curves presented in Fig. 2 are based on the signal-to-noise ratio SNR of the channel, assuming the channel is a pure AWGN channel.As seen in the figure, a differential phase shift keying (DPSK) in both MC system will limit the number of bits per symbol and results in an 11dB loss in SNR.By comparison the error curves, a gain in SNR can be observed at the low However, a DCT scheme can be used with all the subcarriers allocated to transmit only the estimated phase of the different constellation points (QPSK, DPSK), where the amplitude is constant, thereby improving the data rate to the double. Non-Linear AWGN Channel We also analyzed the distribution of time domain transformed amplitude as depicted in Fig. 3 & Fig. 4. The probability density function (PDF) depends on the probability of occurrence of each discrete sample level. The distribution of MC signal with 256 sub-carriers and 16-QAM is founded where the amplitude S n has Rayleigh distribution with PDF given by: (3) where S n the transformed (using IFFT or IDCT) time domain samples have real parts only as stated before. For Rayleigh distribution, as seen the signal levels around the mean value have higher probability than other levels, while the occurrence of the large signals has the smallest probability and moreover different level in DCT and FFT-based schemes, as shown in Fig. 3 & Fig. 4. It is reasonable cast a way to change the statistic of the amplitude for the benefit of PAPR reduction.Notice that PAPR is a random variable for each transmitted block.It has been found from a large number of independent runs for the DCT and FFT-based MC scheme that within each block, the peak power is reduced using DCT-based system of about 5dB and a reduction of PAPR of about 1.3dB are achieved. For the simulated error in Fig. 5 a non-linear power amplifier is applied with AM/AM response of clipping scheme at the saturation point at three different Output Back Off OBO; 4, 5 & 6dB with no compensation of the non-linearity in the conventional MC system.OBO is defined as the ratio of the maximum possible amplifier output power to the average output power.This indicates the power efficiency of the amplifier (Al-Dhahir and Minn, 2006).OBO=6dB guarantees linear transmission while OBO=4 & 5dB in FFT-MCM scheme generates an almost flattened error curves indicating saturation of the power amplifier.These results can be directly compared with the results for MC scheme using DCT.Error curves with OBO=4, 5 & 6dB are also measured.It is found that when OBO=5dB the simulated approaches the linear transmission due to the overall net improvement in the PAPR using DCT, while when OBO decreases to 4dB the performance diverges from the linear case also, indicating non-linear distortion occurs. Flat Fading Channel Flat fading has an impulse response given by: (4) where g (t, τ) is the impulse response at observation time t to an impulse applied at time t-τ.This channel is considered to be slowly time-varying such that the amplitude β(t) and the phase shift θ(t) can be considered constant during one symbol interval. To compare and contrast the performance of the schemes over fading channel, the error curves are examined at similar parameters.Fig. 6 shows that, for QAM multi-carrier signal the performance are incompatible exactly for 16-QAM signal where the DCT-based system approaches the linear transmission.The distortion arose from fading channel can be compensated by increasing SNR, while multi-amplitude signaling based on FFT without channel estimation induces irreducible error curves.As depicted the DPSK signals in both MC systems have the same performance depending on the phase tracking of the decoder.The simulated error rates as shown in Fig. 7 are depicted for various flat fading parameters, which affects on the overall DCT-based MC system performance and the error performance limit.It is clearly observed that the amplitude and phase shift of the reflected signals have a much stronger influence on the systems using QAM than on DPSK modulation systems. The results could be greatly improved to meet the specification requirement that could be implemented in the future to further enhance the simulator, where the results can be extended to any system which uses the MCM technique. Conclusions Extensive computer simulations have been carried out to demonstrate and compare the performance of DCT multi-carrier scheme with that of FFT scheme.The results indicate quite similar performance over linear AWGN channel.Concerning the overall system performance, QAM is often more robust and more spectrally efficient than DPSK modulation.Performance enhancement of the proposed system has been recorded over non-linear AWGN channel due to the reduction in PAPR power of the DCT multi-carrier signal.It is more flexible to present linear transmission with no compensation of the non-linearity in the system.Moreover, the simulated results concludes that the new scheme over flat faded channel without channel estimation performs better corresponding to that of the traditional scheme, especially with QAM modulated tones depending too much on the reflected path amplitude and phase.Such modulated tones in FFT-based scheme results in an irreducible error even at high SNR.However, the results could be greatly improved to meet the specification requirement that could be implemented in the future to further enhance the simulator, where the results can be extended to any system which uses the MCM technique. Figure 1 Figure 1 illustrates the MCM system used for multicarrier modulation.In DCT-based scheme, the IFFT & FFT blocks are simply replaced by an IDCT & DCT or vice versa, respectively, since IDCT matrix is the transpose of DCT matrix. Figure 1 . Figure 1.Block diagram of DCT-based MC modulator Figure 2 .Figure Figure 2. Pe in DCT-based and FFT-based MC scheme versus SNR of linear AWGN channel	2,703.4	2009-12-01T00:00:00.000	[ "Computer Science", "Engineering" ]
Phosphorylation of tyrosine 319 of the angiotensin II type 1 receptor mediates angiotensin II-induced trans-activation of the epidermal growth factor receptor. Although tyrosine kinases are critically involved in the angiotensin II (Ang II) type 1 (AT1) receptor signaling, how AT1 receptors activate tyrosine kinases is not fully understood. We examined the structural requirements of the AT1 receptor for transactivation of the epidermal growth factor (EGF) receptor (EGFR). Studies using carboxyl terminal-truncated AT1 receptors indicated that the amino acid sequence between 312 and 337 is required for activation of EGFR. The role of the conserved YIPP motif in this sequence in transactivation of EGFR was investigated by mutating tyrosine 319. Ang II failed to activate EGFR in cells expressing AT1-Y319F, whereas EGFR was activated even without Ang II in cells expressing AT1-Y319E, which mimics the AT1 receptor phosphorylated at Tyr-319. Immunoblot analyses using anti-phospho Tyr-319-specific antibody showed that Ang II increased phosphorylation of Tyr-319. EGFR interacted with the AT1 receptor but not with AT1-Y319F in response to Ang II stimulation, whereas the EGFR-AT1 receptor interaction was inhibited in the presence of dominant negative SHP-2. The requirement of Tyr-319 seems specific for EGFR because Ang II-induced activation of other tyrosine kinases, including Src and JAK2, was preserved in cells expressing AT1-Y319F. Extracellular signal-regulated kinase activation was also maintained in AT1-Y319F through activation of Src. Overexpression of wild type AT1 receptor in cardiac fibroblasts enhanced Ang II-induced proliferation. By contrast, overexpression of AT1-Y319F failed to enhance cell proliferation. In summary, Tyr-319 of the AT1 receptor is phosphorylated in response to Ang II and plays a key role in mediating Ang II-induced transactivation of EGFR and cell proliferation, possibly through its interaction with SHP-2 and EGFR. The signaling mechanism of the angiotensin II (Ang II) 1 type 1 (AT1) receptor has traditionally been portrayed to be depend-ent on heterotrimeric G proteins, including G␣ q and G␣ i proteins and their downstream targets, primarily phospholipase C (1). This results in inositol triphosphate generation, which in turn causes an increase in intracellular calcium concentrations and diacylglycerol formation, leading to activation of protein kinase C. However, recent investigations revealed that tyrosine phosphorylation is also intimately involved in AT1 receptor signaling (2)(3)(4)(5)(6). Ang II-induced ERK1/2 activation, for example, requires tyrosine kinase activation, including Src family tyrosine kinases (7,8) and epidermal growth factor receptor (EGFR) (9,10). It is unclear, however, how AT1 receptors, which lack intrinsic tyrosine kinase activities, are able to stimulate tyrosine kinases. We have recently shown that an AT1 receptor second intracellular loop mutant, lacking heterotrimeric G protein coupling, is able to activate Src tyrosine kinase (11). This suggests that heterotrimeric G protein-independent mechanisms are able to activate Src. Furthermore, increasing lines of evidence suggest that the carboxyl terminus (C-tail) of the AT1 receptor plays an important role in the AT1 receptor signaling (11,12). For example, ligand binding to the AT1 receptor induces physical association of the C-tail of the AT1 receptor with Jak2, thereby causing phosphorylation and translocation of STAT to the nucleus (13). Other signaling molecules, including phospholipase C␥ and SHP-2, also have been shown to interact with the C-tail of the AT1 receptor (14,15). These results suggest that direct interaction between the heterotrimeric G protein-coupled receptor and intracellular signaling molecules may play an important role in mediating activation of downstream-signaling mechanisms. Accumulating data suggests that EGFR is involved in signal transduction of many G protein-coupled receptors, including the AT1 receptor (9, 10) (for review, see Ref. 16). Ang II induces tyrosine phosphorylation of EGFR and its association with Shc and Grb2, leading to subsequent activation of the Ras-Raf-MEK-ERK1/2 pathway (9). Although several signaling mechanisms are involved in Ang II-induced activation of EGFR (9,(17)(18)(19)(20), whether or not direct interaction between the AT1 receptor and intracellular signaling molecules is required for EGFR activation and, if so, the amino acid sequence of the AT1 receptor mediating Ang II-induced EGFR activation has not been identified. To elucidate the molecular mechanism of Ang II-induced EGFR activation, we investigated the structural requirements of the AT1 receptor and the associating signaling mechanism leading to transactivation of EGFR. Our results indicate that tyrosine 319 at the conserved YIPP motif in the carboxyl terminus of the AT1 receptor plays an essential role in mediating Ang II-induced transactivation of EGFR and cell proliferation. EXPERIMENTAL PROCEDURES Materials-Ang II was purchased from Peninsula. Anti-FLAG M2 affinity gel was from Sigma. Horseradish peroxidase-conjugated antiphosphotyrosine monoclonal antibody (RC20H) and anti-EGF receptor monoclonal antibody were from Transduction Laboratories. Anti-v-Src monoclonal antibody was from Calbiochem. Anti-EGF receptor sheep polyclonal antibody was from Upstate Biotechnology. Rabbit anti-ERK1/2 polyclonal antibody was from Zymed Laboratories Inc., and rabbit anti-active ERK1/2 polyclonal antibody was from Promega. Horseradish peroxidase-conjugated anti-rabbit IgG and anti-mouse IgG antibodies were from Cell Signaling Technology. Anti-AT1 receptor antibody was from Santa Cruz Biotechnology. Dowex AG1-X8 formate resin was from Bio-Rad. 3-[4-Iodotyrosyl-125 I]Ang II and myo-[ 3 H]inositol were from Amersham Biosciences. Enolase was from Roche Molecular Biochemicals. AG1478 was from Biomol. Cell Cultures, Transfection, and Receptor Binding Assays-COS-7 cells were maintained in Dulbecco's modified Eagle's medium (DMEM) supplemented with 10% fetal bovine serum, 100 units/ml penicillin G, and 100 g/ml streptomycin at 37°C in a humidified 5% CO 2 atmosphere. Transfections were performed on 70% confluent monolayers in 60-mm dishes for immunoprecipitation and Src kinase assays or in 35-mm dishes for ERK1/2 assays. For transient transfection, 2.5 ml of Opti-MEM I (Invitrogen) containing 4 g of DNA, 8 l of Lipo-fectAMINE Plus reagent, and 12 l of LipofectAMINE was used for a 60-mm dish. One ml of Opti-MEM I containing 2 g of DNA, 6 l of LipofectAMINE Plus reagent, and 4 l of LipofectAMINE was used for a 35-mm dish. Empty pcDNA 3.0 plasmid was added as needed to keep the total amount of DNA constant per transfection. Cells were incubated in serum-free Opti-MEM at 37°C for 3 h. COS-7 cells were then incubated with 10% fetal bovine serum in DMEM and incubated overnight. Transfected cells were serum-starved in serum-free DMEM for 24 -36 h before stimulation. Assays were performed 48 h after transfection. Saturation binding curves were determined using a modification of the whole cell receptor binding assay (16) as described previously (11). The B max and the dissociation constant (K d ) 3-[4-iodotyrosyl-125 I]AngII binding was determined by using Prism 3.0 (GraphPad). Protein assay was performed on each sample using the Bio-Rad protein assay kit. Cardiac fibroblast cultures were prepared as described previously (17). In brief, hearts were removed from 1-day-old Crl:(WI)BR-Wistar rats (Charles River Laboratories) and subjected to digestion with collagenase type IV (Sigma), 0.1% trypsin (Invitrogen), and 15 g/ml DNase I (Sigma). Cell suspensions were applied on a discontinuous Percoll gradient (1.060/1.086 g/ml) and subjected to centrifugation at 3000 rpm for 30 min (18). The layer containing primarily non-cardiac myocytes was removed and subjected to the preplating procedure for 1 h. The supernatant was discarded, and the attached cells were cultured in the media containing DMEM/F-12 supplemented with 10% fetal bovine serum. Cells were passed twice to enrich for cardiac fibroblasts. Cells were cultured in serum-free conditions for 48 h before experiments. Immunoprecipitation and Immunoblotting-Cell stimulation was carried out at 37°C in serum-free medium. After stimulation, COS-7 cells were scraped and lysed in CHAPS buffer (150 mM NaCl, 40 mM HCl-Tris, pH 7.5, 1% Triton X-100, 0.1% CHAPS, 10% glycerol, 2 mM EDTA, 0.1 mM NaVO 4 , 1 mM NaF, 0.5 mM 4-(2-aminoethyl)benzenesulfonyl fluoride, 0.5 g/ml aprotinin, 0.5 g/ml leupeptin) for immunoprecipitation or hypoosmotic lysis buffer (25 mM NaCl, 25 mM Tris, pH 7.5, 0.5 mM EGTA, 10 mM sodium pyrophosphate, 1 mM Na 3 VO 4 , 10 mM NaF, 0.5 mM 4-(2-aminoethyl)benzenesulfonyl fluoride, 0.5 g/ml aprotinin, 0.5 g/ml leupeptin) for ERK1/2 kinase assays. Cell lysates were incubated on ice for 10 min and subjected to centrifugation for 30 min. Protein concentrations of the supernatants were adjusted to be 1 mg/ml with the lysis buffer. For immunoprecipitation of EGFR-FLAG, the cell lysates (500 g) were incubated with 40 l of anti-FLAG M2 affinity gel at 4°C for 2 h. For immunoprecipitation of endogenous EGFR, the cell lysates were incubated with 4 g of anti-EGFR monoclonal antibody for 1 h followed by protein G-agarose (30 l of slurry) for 45 min. The immune complexes were washed 3 times with lysis buffer and denatured in Laemmli sample buffer. After SDS-PAGE, samples were transferred onto polyvinylidene fluoride microporous membranes (Millipore). Immunoblots were performed as described previously. Phosphorylated levels of the EGFR, the AT1 receptor, or ERK1/2 were analyzed by immunoblotting with anti-phosphospecific antibody and scanning densitometry, and the results were expressed as fold increase compared with the control. Phosphoinositide Production-Measurement of inositol phosphates (IPx) was based upon the method of Berridge et al. (19) as described previously (11). Cells were incubated with myo-[ 3 H]inositol (10 Ci/ml) in DMEM for 24 h at 37°C. Labeling was terminated by aspirating the medium, rinsing cells with oxygenated reaction buffer (142 mM NaCl, 30 mM Hepes buffer, pH 7.4, 5.6 mM KCl, 3.6 mM NaHCO 3 , 2.2 mM CaCl 2 , 1.0 mM MgCl 2 , and 1 mg/ml D-glucose), and harvesting cells with phosphate-buffered saline, 0.02%EDTA. Cells were centrifuged twice (300 ϫ g, 5 min) in reaction buffer, and the pellet was resuspended in an equal volume of reaction buffer containing 60 mM LiCl. Stimulation of IPx production was initiated by mixing 0.25 ml of cell suspension with 0.25 ml of 0 -100 nM Ang II in reaction buffer (without LiCl). The mixture was incubated for 30 min at 37°C, then 0.5 ml of ice-cold 20% trichloroacetic acid was added. Precipitates were pelleted (4100 ϫ g, 20 min), and the trichloroacetic acid-soluble fraction was transferred to new tubes, washed with water-saturated diethyl ether, and neutralized with NaHCO 3 . IPx were isolated by adsorption to 0.5 ml of Dowex AG1-X8 formate resin slurry and rinsed 5 times with 3 ml of unlabeled 5 mM myoinositol followed by elution with 1 ml of 1.2 M ammonium formate, 0.1 M formic acid. The elutes were counted by liquid scintillation counter in 5 ml of ScintiVerse. Src Kinase Assay-The tyrosine kinase activity of Src was determined by the immune complex kinase assay using enolase as a substrate as described previously (7,11). Cell lysates were prepared in a lysis buffer (150 mM NaCl, 15 mM HEPES, pH 7.0, 1% deoxycholic acid, 1% IGEPAL, 0.1% SDS, 0.1 mM NaVO 4 , 1 mM NaF, 0.5 mM 4-(2aminoethyl)benzenesulfonyl fluoride, 0.5 g/ml aprotinin, 0.5 g/ml leupeptin). The cell lysates containing equal amount of protein (750 g) were incubated with anti-v-Src monoclonal antibody at 4°C for 1 h. Protein G-Sepharose was then added. The immunoprecipitates were washed twice with lysis buffer without SDS or deoxycholic acid and then washed once with kinase buffer (50 mM HEPES, pH 7.6, 0.1 mM EDTA, 10 mM MnCl 2 , 0.015% Brig 35). Pellets were incubated for 15 min at 37°C in the kinase buffer with 1 Ci of [␥-32 P]ATP and 0.25 g of enolase as a substrate. The reaction was terminated by the addition of Laemmli sample buffer on ice. Reaction mixtures were boiled and subjected to 12% SDS-PAGE followed by autoradiography. Results were analyzed by densitometry. Adenovirus Vectors-Adenovirus-mediated transduction was performed as described previously (20). Cells grown in 60-mm dishes were transduced with an adenovirus vector harboring dominant-negative Ras (Ad5.N17Ras) (courtesy of Dr. M. Schneider, Baylor College of Medicine, Houston, TX) at a multiplicity of infection of 100. For a control study, Ad5/⌬E1sp1B (courtesy of Dr. B. French, University of Virginia, Charlottesville, VA) was used. Adenovirus vectors harboring the wild type AT1 receptor or AT1-Y319F were generated by using the AdEasy system (21). All experiments were performed 48 h after transduction. Cell Proliferation Experiments-Cells were plated at a density of 0.3 ϫ 10 6 /well in six-well plates. Twelve hours after plating, cells were serum-starved for 12 h and then stimulated with Ang II (10 Ϫ7 M) in the presence or absence of AG1478 (250 nM) for 36 h. Ang II or AG1478 was added every 12 h. After stimulation, cells were washed twice with phosphate-buffered saline. The cell layer was scraped with 1 ml of standard sodium citrate containing 0.25% SDS and vortexed extensively. Total DNA content was determined by the Hoechst dye method as described previously (17). Statistics-Data are given as the mean Ϯ S.E. Statistical analyses were performed using the analysis of variance. The post-test comparison was performed by the method of Tukey. Significance was accepted at p Ͻ 0.05 level. Ang II Activates EGFR in COS-7 Cells Expressing the Wild Type AT1 Receptor-Ang II stimulation of COS-7 cells without transfection of the AT1 receptor did not activate either endogenous or transfected EGFRs (not shown). By contrast, in COS-7 cells transfected with the AT1 receptor, Ang II, caused timedependent increases in tyrosine phosphorylation of either endogenous (not shown) or transfected EGFRs (Fig. 1A). Tyrosine phosphorylation of EGFR by Ang II was observed within 3 min, reached a peak around 5 min, and lasted for more than 60 min ( Fig. 1B (n ϭ 7) and 2.8 Ϯ 0.1-fold at 60 min (n ϭ 3)). This suggests that stimulation of the AT1 receptor causes transactivation of EGFR in COS-7 cells. Tyrosine 319 in the Carboxyl Terminus of the AT1 Receptor Plays an Important Role in Ang II-induced Activation of EGFR-Because it has been suggested that the C-tail of the AT1 receptor plays an important role in mediating cell-signaling mechanisms of the AT1 receptor (11), we examined the role of the AT1 receptor C-tail in Ang II-induced transactivation of EGFR. We co-transfected EGFR-FLAG and carboxyl-terminaltruncated AT1 receptors into COS-7 cells, and EGFR was immunoprecipitated with anti-FLAG antibody. Although Ang II caused significant increases in tyrosine phosphorylation of EGFR in cells transfected with AT1-(1-338), it failed to do so in cells transfected with AT1-(1-311) ( Fig. 2A). These results suggest that the amino acid sequence located between amino acid 312 and 337 of the AT1 receptor is required for activation of EGFR by Ang II stimulation. Between amino acids 312 and 337 of the AT1 receptor, the YIPP-(319 -322) motif is evolutionarily conserved in AT1 receptors cloned from many species (Fig. 2B). It has been shown that several signaling molecules directly or indirectly associate with the YIPP motif in the AT1 receptor (13,14). To test the role of this conserved motif in Ang II-induced EGFR activation, we made a mutant where a tyrosine residue at position 319 in this motif was mutated to phenylalanine (AT1-Y319F). Interestingly, Ang II-induced activation of EGFR was abolished in COS-7 cells transfected with AT1-Y319F (Fig. 2C). II stimulation failed to increase phosphorylation of EGFR (Fig. 3A). These results suggest that phosphorylation of tyrosine 319 may be involved in Ang II-induced activation of EGFR. Tyrosine 319 Is Phosphorylated in Response to Ang II Stimulation-To test if tyrosine 319 of the AT1 receptor is phosphorylated in vivo in response to Ang II stimulation, we generated phosphotyrosine 319-specific anti-AT1 receptor antibody (antiphosphotyrosine 319 antibody). Either wild type AT1 receptor (AT1-WT) or AT1-Y319F was transfected in COS-7 cells. The AT1 receptor phosphorylated at tyrosine 319 was immunoprecipitated by the anti-phosphotyrosine 319 antibody and immunoblotted with the same antibody. Although AT1-WT was not detected by anti-phosphotyrosine 319 antibody in unstimulated cells, it was detected after the cells were stimulated with Ang II for 3 min (Fig. 3B). Phosphorylation of tyrosine 319 was transient and returned to the basal level at 5 min. No apparent signals of phosphotyrosine 319 were detected in samples obtained from cells expressing AT1-Y319F (Fig. 3B), suggesting that the signal found in AT1-WT was most likely from phos-phorylated Tyr-319. Duplicate samples were subjected to immunoprecipitation using anti-(total)-AT1 receptor antibody and immunoblotted with the same antibody. The result confirmed that similar amounts of the AT1 receptor were solubilized in each sample (Fig. 3C). These results indicate that tyrosine 319 in AT1-WT is transiently phosphorylated in response to Ang II stimulation. Ang II-induced Activation of EGFR in AT1-WT Is Inhibited in the Presence of AT1 Carboxyl Terminus Minigene-To confirm that tyrosine 319 of the AT1 receptor is important for Ang II-induced transactivation of EGFR, we examined the effect of overexpression of a mini-gene-containing AT1 receptor carboxyl terminus peptide (HA-AT1-C) upon Ang II-induced EGFR activation in COS-7 cells. Expression of HA-AT1-C inhibited Ang II-induced activation of EGFR in COS-7 cells (Fig. 4A). Overexpression of HA-AT1-C alone did not inhibit activation of EGFR by EGF treatment (Fig. 4B), suggesting that the effect of HA-AT1-C is stimulus-specific. Furthermore, a minigene-containing HA-AT1-C, whose tyrosine 319 is mutated to phenylalanine (HA-AT1-C-Y319F), failed to inhibit Ang II-induced EGFR activation in COS-7 cells (Fig. 4A). These results further support the notion that tyrosine 319 in the AT1 receptor plays an essential role in mediating EGFR activation by the AT1 receptor. Ang II Stimulates Interaction between AT1 Receptors and EGFR, Which Was Inhibited in the Presence of Dominant Negative SHP-2-We next examined if the AT1 receptor and EGFR physically associate with each other. We co-transfected AT1 receptors and EGFR into COS-7 cells and stimulated the cells with Ang II. EGFR immunoprecipitates were immunoblotted with anti-AT1 receptor antibody. Interestingly, AT1 receptors were co-immunoprecipitated with EGFR in samples stimulated with Ang II for 3 min (2.7 Ϯ 0.2-fold versus 0 min, n ϭ 4, Fig. 5A). This interaction between the AT1 receptor and EGFR was transient, since it was not observed at 5 min. Equal amounts of EGFRs were immunoprecipitated in each sample, and the immunoprecipitated EGFR was tyrosine-phosphorylated in response to Ang II stimulation, consistent with the results shown in Fig. 1 (Fig. 5A). Because it has been shown that SHP-2 is associated with AT1 receptor at the YIPP motif (14,15), we examined if disrupting interaction between the AT1 receptor and SHP-2 affects interaction between the AT1 receptor and EGFR and abolishes Ang II-induced activation of EGFR. Ang II-dependent interaction between the AT1 receptor and EGFR and activation of EGFR were both inhibited in the presence of dominant negative SHP-2 (22) (Fig. 5A). These results suggest that SHP-2 plays an essential role in mediating Ang II-induced activation of EGFR. Because tyrosine 319 plays an essential role in mediating Ang II-induced activation of EGFR, we attempted to determine if EGFR can interact with AT1-Y319F. EGFR and AT-Y319F were co-transfected into COS-7 cells, and cells were stimulated with Ang II. EGFR was immunoprecipitated and then immunoblotted with anti-AT1 receptor antibody. We did not detect the AT1 receptor in the EGFR immunoprecipitates (Fig. 5B). As expected, EGFR was not phosphorylated in response to Ang II stimulation in these experiments. These results suggest that tyrosine 319 plays an essential role in mediating Ang II-dependent interaction between the AT1 receptor and EGFR. Mutation of Tyrosine 319 Does Not Affect Other Signaling Mechanisms-We examined if other signaling mechanisms are also affected in the AT1-Y319F mutant. Ang II caused significant levels of IPx accumulation in cells transfected with AT1-Y319F, which was not statistically different from those in cells transfected with AT1-WT (Fig. 6A). This suggests that AT1-Y319F maintains coupling with the G␣ q -phospholipase C pathway. This also suggests that activation of IPx alone is not sufficient for the AT1 receptor to mediate EGFR activation. We attempted to determine if activation of other tyrosine kinases is also affected in AT1-Y319F. Ang II caused significant increases in Src activities in cells transfected with either AT1-WT or AT1-Y319F (2.9 Ϯ 0.1-fold in WT, 2.7 Ϯ 0.2-fold in Y319F at 5 min) (Fig. 6B). Ang II also caused similar levels of increases in tyrosine phosphorylation of transfected JAK2-FLAG in cells expressing either AT1-WT or AT1-Y319F (3.5 Ϯ 0.1-fold in WT, 3.4 Ϯ 0.2-fold in Y319F at 5 min) (Fig. 6C). These results suggest that Ang II-induced activation of some tyrosine kinases, including Src and JAK2, is preserved in cells expressing AT1-Y319F. Besides EGFR, tyrosine kinase Src also plays an important role in mediating ERK activation in some cell types. To identify the (back-up) mechanism by which AT1-Y319F activates ERKs, we examined the role of Src in Ang II-induced ERK activation. Expression of dominant negative Src did not affect Ang IIinduced ERK activation (3.6 Ϯ 0.2-fold, p Ͻ 0.05 versus control) in COS-7 cells expressing AT1-WT. By contrast, dominant negative Src abolished Ang II-induced ERK activation in COS-7 cells expressing AT1-Y319F (1.0 Ϯ 0.1-fold, not significant versus control) (Fig. 7B). This suggests that Src compensates the loss of EGFR activation for Ang II-induced ERK activation by AT1-Y319F. It has been shown that both EGFR-and Src-dependent activation of ERKs is mediated by Ras. Ang II-induced activation of ERKs in COS-7 cells expressing either AT1-WT or AT1-Y319F was abolished in the presence of dominant negative Ras (Fig. 7C), suggesting that the mechanisms of ERK activation by AT1-WT and AT1-Y319F are likely to converge at the level of or upstream of Ras. Overexpression of the AT1-Y319F Mutant Abolishes Ang IIinduced Cell Proliferation in Cardiac Fibroblasts-Our results presented thus far indicated that AT1-Y319F selectively lacks Ang II-induced transactivation of EGFR, whereas it maintains activation of some, if not all, signaling molecules, including IPx, Src, JAK2, and ERKs. To test if AT1-Y319F mutant has cellular functions different from AT1-WT, we expressed either AT1-WT or AT1-Y319F in primary cultured cardiac fibroblasts by adenovirus-mediated gene delivery. An adenovirus harboring an irrelevant sequence was used as a control vector. Cell cultures with comparable levels of expression of either AT1-WT or AT1-Y319F were used for these experiments. Overexpression of AT1-WT in cardiac fibroblasts increased Ang II-induced EGFR activation. By contrast, overexpression of AT1-Y319F failed to enhance Ang II-induced activation of EGFR, suggesting that Tyr-319 plays an essential role in mediating Ang II-induced EGFR activation in cardiac fibroblasts (Fig. 8A). Under these conditions, overexpression of the AT1-WT in cardiac fibroblasts significantly enhanced Ang II-induced cell proliferation, which was determined by the total DNA content, compared with control virus transduced cells (Fig. 8B). By contrast, overexpression of AT1-Y319F abolished small increases in cell proliferation found in control virus-infected cardiac fibroblasts (Fig. 8B). To determine the role of EGFR activation in Ang II-induced cardiac fibroblast proliferation, we treated the cells with AG1478. Treatment with AG1478 completely abolished the Ang II-induced cell proliferation in both control virus-transduced and AT1-WT-transduced cardiac fi- broblasts (Fig. 8B). These results suggest that the cell-signaling mechanism mediated by Tyr-319 in the AT1 receptor, including activation of EGFR, is required for Ang II-induced cell proliferation in cardiac fibroblasts. Tyrosine 319 Is Specifically Required for Ang II-induced EGF Receptor Activation-The YIPP motif found in the AT1 receptor is conserved in all members of the AT1 receptor family so far cloned, suggesting that this motif is involved in important cellular functions of the AT1 receptor. This motif is also found in the platelet-derived growth factor ␣ and ␤ receptors and is involved in ligand-dependent activation of phospholipase C␥ (23). Although it has been shown that several signaling molecules associate with the motif in the AT1 receptor (13-15, 24, 25), the specific requirement of tyrosine 319 for activation of downstream protein kinases has not been clearly demonstrated in vivo. Because Ang II-induced activation of Src and JAK2 was not affected in AT1-Y319F, the requirement of tyrosine 319 in the AT1 receptor seems specific for activation of EGFR among the activation of major tyrosine kinases. The structural requirements of the AT1 receptor in Ang II-induced EGFR have not been previously determined. It has been shown that the AT1 receptor is tyrosine-phosphorylated by ligand binding (14,26). However, the tyrosine residue phosphorylated by ligand binding has not been identified in vivo. Because signaling molecules containing the SH2 domain interact with the YIPP motif of the AT1 receptor, it has been speculated that tyrosine 319 is phosphorylated (13)(14)(15). By using anti-phosphotyrosine 319-specific AT1 receptor antibody, we demonstrated that tyrosine 319 is phosphorylated in response to Ang II. Although it has been speculated that ligand-dependent phosphorylation of the AT1 receptor may modulate the activities of downstream signaling molecules besides internalization of the receptor, this has not been clearly demonstrated. In fact, it has been recently shown that phosphorylation of the AT1 receptor by G protein-coupled receptor kinase does not play an essential role in Ang II-induced cell signaling (27). In our study, because expression of AT1-Y319E increased basal levels of EGFR phosphorylation and because Ang II failed to show additive effects on EGFR activation by Y319E, phosphorylation of tyrosine 319 seems to mediate Ang II-induced activation of the EGFR. At present we do not know which tyrosine kinase is responsible for phosphorylation of tyrosine 319. The Mechanism of EGFR Activation by the AT1 Receptor-Ca 2ϩ -dependent mechanisms (9), other tyrosine kinases, such as Src and Pyk2 (28 -30), metalloproteases (for review, see Ref. 31), and reactive oxygen species (32) have been proposed as mechanisms of Ang II-induced activation of EGFR in various cell types. The requirement of tyrosine 319 in the AT1 receptor for EGFR activation found in the present investigation may not contradict these previous observations but may represent another requirement for EGFR activation by AT1 receptors. Because production of IPx and activation of other kinases such as Src and JAK2 by Ang II are preserved in AT1-Y319F, activation of these molecules alone is not sufficient for Ang II-induced EGFR activation. Growing lines of evidence suggest that heterotrimeric G protein-coupled receptors directly interact with intracellular signaling molecules (33,34). The mini-gene containing AT1-C, but not AT1-C-Y319F, effectively blocked Ang II-induced activation of EGFR. Thus, protein-protein interaction at the AT1-C containing tyrosine 319 may mediate Ang II-induced EGFR activation. It has been shown in vitro that the SH2 domain of SHP2 interacts with the YIPP motif, where Tyr-319 is located, once tyrosine is phosphorylated (13,15). Our results suggest that the AT1 receptor and EGFR transiently interact with each other in a ligand binding-dependent manner, and the timing of their interaction coincides with that of Tyr-319 phosphorylation. Therefore, we speculate that SHP2 binds to Tyr-319 when Tyr-319 is phosphorylated, thereby acting as a scaffold protein. Because dominant negative SHP-2 (22) was able to inhibit both AT1 receptor-EGFR interaction and Ang II-induced transactivation of EGFR, it is likely that SHP-2 mediates AT1 receptor-EGFR interaction. It has been recently shown that EGFR associates with ␤-adrenergic receptor in a ligand binding-dependent manner, possibly through a scaffold protein, ␤-arrestin (35). Thus, both SHP-2 and ␤-arrestin work as scaffold proteins to induce interaction between G protein-coupled receptors and EGFR for transactivation of EGFR. This hypothesis is consistent with a recent observation that assembling signaling molecules, including the AT1 receptor and EGFR, by scaffolding proteins such as caveolin is required for Ang IIinduced transactivation of EGFR (36). It should be noted that phosphorylation of EGFR persists even after interaction between the AT1R and EGFR is lost. We speculate that once phosphorylation of EGFR is initiated, the activity of EGFR may be maintained through autophosphorylation until other mechanisms of inactivation are activated. Hypothesis about the Back-up Mechanism-Although Ang II binding to the AT1-WT activates ERKs through an EGFR-dependent mechanism, binding to AT1-Y319F still activates ERKs through an Src-dependent mechanism despite the fact that AT1-Y319F failed to activate EGFR. The cellular mechanism by which the Src-dependent mechanism compensates for the loss of EGFR activation to activate ERKs only in cells expressing AT1-Y319F is unclear at present. One possible explanation would be that EGFR may sequestrate molecules leading to Ras activation, such as Grb2, when tyrosine 319 of the AT1 receptor is available. It should be noted that it has been previously shown that tyrosine kinases are able to mediate Ang II-induced ERK activation only when a protein kinase C-dependent mechanism of ERK activation is blocked (37). Thus, tyrosine kinases may in general work as a back-up mechanism for the AT1 receptor to maintain the activity of ERKs. The Role of Tyrosine 319 in Cell Proliferation in Cardiac Fibroblasts-Our results show that tyrosine 319 of the AT1 receptor plays an essential role in mediating Ang II-induced cell proliferation in cardiac fibroblasts. Although we have shown in this work that tyrosine 319 of the AT1 receptor plays a key role in Ang II-induced EGFR activation in cardiac fibroblasts, this does not exclude the possibility that activation of other molecules may also depend upon tyrosine 319. However, considering the fact that the EGFR-specific inhibitor AG1478 completely abolished Ang II-induced cell proliferation, it is likely that the effect of the mutation at tyrosine 319 is primarily mediated through its effect upon Ang II-induced EGFR activation. The signaling mechanisms of Ang II-induced cellular responses have been primarily studied by using either specific chemical inhibitors or inhibitor molecules such as dominant negatives (10). Our results suggest that the AT1 receptor mutant, which has a selective defect in the signaling mechanism, can be used to elucidate the cellular function of the signaling mechanisms activated by the AT1 receptors.	6,678	2003-03-14T00:00:00.000	[ "Biology", "Medicine" ]
Thermodynamic equilibrium analysis of entrained flow gasification of spent pulping liquors The main goal of this work is to investigate if thermodynamic equilibrium calculations can be useful for understanding and predicting process performance and product composition for entrained flow gasification of spent pulping liquors, such as black liquor. Model sensitivity to input data is studied and model results are compared to published pilot plant data. The high temperature and the catalytic activity of feedstock alkali make thermodynamic equilibrium a better predictor of product composition than for many other types of biomass and gasification technologies. Thermodynamic equilibrium calculations can predict the flows of the main syngas and slag products with high accuracy as shown by comparison with experimental data with small measurement errors. The main process deviations from equilibrium are methane formation and sulfur distribution between gas and slag. In order to study real process deviations from equilibrium, it is very important to use consistent experimental data. Relatively small errors in the model input, primarily related to fuel composition, can lead to grossly erroneous conclusions. The model sensitivity to fuel composition also shows that the gasification process is sensitive to naturally occurring feedstock variations. Simulations of a commercial-scale gasification process show that cold gas efficiency on sulfur-free basis can reach over 80 % and that greatly improved efficiency can be obtained by reducing ballast present in the form of water or inorganics. Introduction Biomass gasification is a promising technology for production of second-generation biofuels and green chemicals. Many gasification technologies produce syngas with high concentrations of tars, which leads to extensive gas cleaning requirements. Entrained flow (EF) gasifiers, using higher temperatures and short residence times, often generate a relatively clean gas and can use no or much simpler gas cleaning. A potential drawback of EF gasification (EFG) is the Electronic supplementary material The online version of this article (doi:10.1007/s13399-016-0225-7) contains supplementary material, which is available to authorized users. requirement to feed solid fuels in a pulverized form, while liquid feedstocks can be atomized. An EFG technology that has come a long way towards commercialization, through demonstration of an integrated biomass-to-biofuel process, is black liquor gasification (BLG) [1]. Black liquor (BL) is a by-product of Kraft pulping that contains dissolved lignin and hemicellulose fragments together with spent inorganic pulping chemicals. It is normally combusted in a recovery boiler (RB) that recovers inorganic pulping chemicals and generates steam. It has been shown that higher overall efficiency and better profitability can be obtained by gasifying BL and generating the process steam from other fuels [2][3][4][5]. The EF BLG process has been demonstrated and investigated in pilot scale using Kraft BL [1,[6][7][8][9][10] and sulfite thick liquor (STL) [11,12]. The latter is the spent pulping liquor from sodium-based sulfite delignification. EF gasification of BL and STL is carried out under slagging conditions at 1000-1100°C. The inorganic smelt (slag) leaving the gasifier is dissolved in water to form a solution called green liquor (GL), which is used to recover the inorganic pulping chemicals. It has generally been established that thermodynamic equilibrium calculations (TECs) are useful for understanding the behavior of biomass gasification under varying process conditions for a range of fuel compositions [13][14][15]. In practice, deviations from equilibrium are common and relate for example to char conversion, tar formation, methane formation, and the water gas shift reaction, especially for low-temperature processes [16][17][18]. Cold gas efficiency (CGE), which controls the potential biofuels yield, is of general interest to predict and understand in biomass gasification while the sulfur distribution between gas and slag phases, which is very important for the recovery of the pulping chemicals (below referred to as the sulfur split), is more specific for BLG. Ash is typically either disregarded or considered inert when developing thermodynamic equilibrium models for nonslagging fixed bed and fluidized bed gasifiers [18][19][20]. For entrained flow gasification of solid biomass, however, slag properties are important and TECs have proven useful for understanding and predicting such properties [21,22]. Treating ash as inert in gasification of sodium-rich fuels such as BL or STL would give totally misleading results since approximately 20 % of the carbon leaves the reactor in the inorganic smelt [6]. Previously, Berglin [23] used a simplified thermodynamic equilibrium model only for the gas phase, aiming to study the performance of air-blown BLG for electricity production. Other thermodynamic equilibrium studies of BLG included slag species in the models [7,10,[24][25][26]. However, the few studies that compared their results to experimental data only focused on equilibrium gas composition and did not validate the composition of the inorganic phase [7,10]. The properties of the inorganic smelt are very important in gasification of BL and STL since it is used to form the aqueous GL that enables to reuse of the pulping chemicals [6,12]. In addition, the inorganic phase influences the mass and energy balances for the gasifier substantially. In summary, there are no studies applying TECs to STL gasification and only two studies comparing TECs for BL gasification to experimental syngas data, neither of which validated prediction of the inorganic phase against experimental results. Hence, better validation is needed under a range of process conditions to reliably and consistently use TECs for studying spent pulping liquor gasification. The main goals of this work are to assess if TECs can be useful for predicting process performance and product composition for EFG of spent pulping liquors as well as quantifying and understanding deviations between model and experimental data. The techno-economic studies of BLG-based biorefineries use fixed BL properties and process design relevant for the cases studied [5,27]. However, many parameters with substantial impact on the process performance can be influenced through the process design. A second goal of this work is to use a validated thermodynamic equilibrium model to quantify the impact of important design parameters on the gasification performance and syngas properties for a commercial-scale gasification process. In principle, any observed differences between TEC results and experimental data can be due to real deviations from equilibrium in the process, erroneous experimental data, and erroneous inputs to the equilibrium model or any combination of these. In order to understand the influence of each of potential cause of deviation, we start by investigating the sensitivity of the TEC to input data followed by an analysis of the deviations for the pilot plant cases studied. The last part of this study uses TEC to predict the behavior of a commercial-scale gasification process. Pilot plant data This work uses previously published data from three studies [6,10,12] of the BLG pilot plant in Piteå, Sweden, to obtain relevant operating conditions for TECs and to compare predictions to experimental results. The reader is referred to those studies as well as other experimental studies [8,9] for a description of the pilot plant and the process. The experiments described and analyzed by Jafri et al. [6] used BL from the Smurfit Kappa Kraftliner mill (Piteå, Sweden) as feedstock. The data includes five operating points (OPs) at 2.6-3.1 MW th load and 24-29 bar. The data in Wiinikka et al. [10] is based on experiments using the same feedstock source. Six OPs with a load around 3 MW th and varying pressure (25-29 bar) and temperature were investigated. The experiments presented by Furusjö et al. [12] used STL from the Domsjö Fabriker (Örnsköldsvik, Sweden) sodium sulfite-based delignification process, which has significantly different properties from BL as discussed further below. The Furusjö and Jafri data quantifies both syngas and inorganic GL components and thus allows full mass and energy balances to be made. Wiinikka et al. only presents data for the syngas. Hence, no balances can be made and no comparison between model and experiment are possible for inorganic components in this case. It is difficult to measure process temperatures accurately [6,12], but the shielded thermocouples used are still considered to give reasonable temperature values that can be used to observe trends and differences between OPs. In practice, syngas methane concentration is used as a temperature proxy and is set to approximately 1 % for Bstandard^operation by varying the amount of oxygen added to the gasifier. The measured reactor temperatures used in this work for the Jafri et al. data set are slightly different (10-20°C higher) from the values presented in the original paper. This is due to the fact that an average of two thermocouples positioned in the mid-level of the reactor was used, whereas Jafri et al. presented values for one of these. The Wiinikka et al. temperature data is based on a thermocouple in the same position. For the Furusjö et al. data, temperatures from the thermocouple positioned close to the bottom of the reactor were used. Thermodynamic equilibrium calculations TECs were carried out using FactSage (GTT-Technologies, Aachen, Germany) as well as an in-house tool called SIMGAS, which was developed in the Matlab environment (Mathworks, Natick, MA, USA). Both tools use a nonstoichiometric approach since stoichiometric methods are not suitable for complex multicomponent systems. In comparison to FactSage, SIMGAS uses a simpler thermodynamic model, including only ideal mixtures of components for both the gas and the inorganic smelt phases. Included components are listed in Table 1; this selection is based on species for which significant concentrations were found during numerous runs with varying process conditions. The pure component data for gas components and solid carbon are taken from the NIST Chemistry WebBook while data for pure inorganic smelt components are based on Lindberg [28]. In SIMGAS, Gibbs energy minimization is accomplished using an activeset method in order to be able to include both linear and nonlinear constraints. Initially, a comparison between FactSage and SIMGAS results was made in order to assess the influence of the simpler thermodynamic model in SIMGAS. This comparison between the two thermodynamic models is described in Section 2 of the Supplementary Material. The conclusion is that the simpler implementation in SIMGAS is sufficient to describe the process for the purposes of this study. No significant differences were observed for major species. Hence, the remainder of the work described in this paper is based on the SIMGAS model. However, if the aim is to study smelt properties or smelt reactions, the more sophisticated smelt model of FactSage would be required. The primary motivation for using the Matlab-based SIMGAS tool is the ease with which constraints can be implemented in order to be able to model deviations from equilibrium in some cases. In practical application of EF biomass gasification, temperature is controlled by oxygen addition, i.e., the process is autothermal. For simulation of pilot plant OPs, TECs are executed in a mode where the temperature of the products is calculated based on an energy balance over the gasification process. For the simulation of commercial biorefinery gasification operation, a certain gasifier temperature is specified and the required amount of oxygen to reach this temperature is calculated through the gasifier energy balance. For the gas phase, comparison between model results and experimental data is straightforward. The inorganic smelt is, however, dissolved to form GL. In this work, the experimentally determined concentrations of inorganic components in GL are compared to TEC smelt predictions. The components are sulfur, carbonate, and total inorganic carbon (TIC). TIC is the sum of carbon present as carbonate and hydrogen carbonate. Hydrogen carbonate is not present in the smelt leaving the gasifier but is formed in the quench as discussed below. Sodium and potassium are not compared since according to both model and experimental data, they are found solely in GL. Feedstock heat of formation Solving the gasifier energy balance during TECs requires knowledge of the enthalpies of all streams entering and leaving the process. The enthalpy can be defined with respect to any reference condition. In this work, the selected reference condition is the elements at standard reference state, which means that the enthalpy describing chemical energy is the heat of formation. For the oxygen, nitrogen, and gasification products, enthalpy estimation is straightforward since the chemical composition is known. However, for the spent pulping liquors, which are highly complex mixtures of organic and inorganic species, it is not possible to use tabulated data for heat of formation. Instead, their enthalpies of formation are calculated from higher heating values (HHV) determined by bomb calorimetry using Hess' law in conjunction with assumptions about the bomb combustion products as described earlier [6]. When this approach is used, it is very important to know exactly which heating value is measured, i.e., what are the final combustion products in the bomb calorimeter? Spent pulping liquors of the type studied in this work are characterized by high sodium and sulfur contents, which are typically 20 and 5 % on a dry-weight basis, respectively. The bomb product chemistry is investigated in some detail in the Supplementary Material. The main conclusion is that due to the presence of high amounts of inorganics in the fuel, carbon ends up as a mixture of carbon dioxide and sodium carbonate. Sulfur ends up as sodium sulfate or sodium hydrogen sulfate depending on the sodium/sulfur ratio. This is different from assumptions typically used for low ash fuels. Model sensitivity to input data The inputs to a thermodynamic equilibrium model for a gasification process consist of the flow, temperature and composition of the streams entering the gasification reactor, and the reactor heat loss in combination with the thermodynamic data/assumptions. In the present case, there are three streams: spent pulping liquor, oxygen, and nitrogen. The nitrogen flow is very small compared to the other flows and does not have a direct impact on the process (except as thermal ballast). The pilot plant from which all experimental data used in this work comes is equipped with Coriolis type mass flow meters for both oxygen (Krohne Optimass 1300; KROHNE Messtechnik GmbH, Duisburg, Germany) and feedstock (Yokogawa Rotamass 3 series; Yokogawa Electric Corporation, Tokyo, Japan). According to the instrument specifications, the measurement error is typically around 0.2 %. In our experience, a larger uncertainty can be expected in practice; we have used a 95 % confidence interval of ±1 % in the sensitivity analysis below. The oxygen and nitrogen compositions are well known since they are of high purity. Feedstock elemental analysis for BL and STL is associated with a relatively large uncertainty as summarized in Table 2 based on uncertainty estimates from the experimental studies. Uncertainty estimates are only available for BL and do not agree fully between the two studies. A pooled relative standard deviation for each element was calculated based on the 95 % confidence limits provided in the earlier work and assuming these to have the same number of degrees of freedom as shown in Table 2 (rightmost column). From the standard deviation presented in Table 2, it is clear that the uncertainty in most elements is relatively large. Figure 1 shows results from varying selected inputs to the upper and lower 95 % confidence limits (as approximated by ±2 pooled standard deviations) for a selected operating point (Jafri et al. OP 5). It can be concluded that black liquor composition uncertainty contributes significantly to variation in the model outputs, while the contribution from flow measurement uncertainty seems to be less important. As an example, the carbon content uncertainty leads to predicted temperatures in a range of 1077-1286°C. This range is much larger than the practical operating envelope of the gasifier [6], which should have significant consequences on agreement between model outputs and experiments, as discussed further below. The variation in cold gas efficiency due to carbon analysis uncertainty is approximately ±5 % units, which is a very significant number and corresponds to approximately ±9 % relative uncertainty. When compared with the ±2.6 % relative uncertainty (2s rel ) for the BL C analysis, it is easy to conclude that input errors are amplified in the model. The influence of BL sodium content uncertainty is even larger than that of carbon while it is slightly smaller for hydrogen. Figure 1 also shows the influence of composition measurement uncertainty on the sulfur split and the hydrogen/carbon monoxide ratio in the syngas, which are important properties for the design of a spent pulping liquor gasification-based biorefinery. From the above results, it is clear that BL composition is a major contributory factor to the uncertainty in TEC results. It is, however, likely that the influence of individual component uncertainty is not a good measure of the total influence on the TECs. In order to estimate the effect of the combined uncertainty of all BL constituents, a Monte Carlo (MC) simulation was carried out. All BL elements except oxygen were assumed to have independent errors with a normal distribution and a standard deviation according to the pooled relative standard deviation presented in Table 2. Feedstock oxygen content was calculated by difference as it is typically done in BL element analysis. In this study, 8000 cases were simulated using TECs with stochastic input data to generate a distribution of all model results, as shown for temperature in Fig. 2. Only BL composition was accounted for in the MC study, although it would have been feasible to include also other uncertain model inputs simultaneously. From the MC simulation results presented in Fig. 1, it is clear that the uncertainty from the combined BL constituents is greater than that from individual components or other factors studied. The results presented in Fig. 1 [12] that the uncertainty in fuel composition creates an uncertainty in oxygen content and stoichiometric oxygen demand. The Monte Carlo simulation gives a 95 % confidence interval for stoichiometric oxygen demand of 0.75-0.86 kg O/kg BL, which clearly indicates that lambda is not always a welldetermined parameter for experimental data. The BL oxygen content distribution had a standard deviation of 1.6 % (absolute) or 4.0 % (relative). The investigations described in this section show that input parameter uncertainty influences the TEC results to a great extent. This can be problematic when using TECs to predict gasification process performance as shown further below. Sensitivity to feedstock heat of formation estimation An important thermodynamic assumption made is related to the calculation of feedstock heat of formation estimation. For low ash fuels, the assumptions about bomb calorimeter combustion products are not so important, but for high ash and high sulfur fuels, such as the spent pulping liquors studied in this work, the issue is more important and complex. The approach described in Section 2.3 and in more detail in the Supplementary material is typically used for BL, but we also evaluated a simplified approach based on the assumption that alkali metals form oxides, as is typically done in ash analysis of biomass. Assuming that sodium and potassium form oxides (as opposed to carbonates and sulfates), all sulfur forms sulfur dioxide (as opposed to sulfate) and all carbon forms carbon dioxide (as opposed to a mix of carbonate and carbonate dioxide); the calculated BL heat of formation values are 1.6-2.3 MJ/kg solids higher for the liquor compositions studied in this work, which corresponds to approximately 10-20 % of HHV. The resulting error completely changes the results of the TECs. This highlights how critical it is to make correct assumptions about the bomb calorimeter combustion products. According to Table 2, the relative standard deviation for HHV is 1 %. Hence, the assumption about combustion products has a much larger impact on TECs than the actual HHV measurement error. Non-constrained equilibrium calculations Operating data in the form of fuel flow rate, fuel composition, oxygen flow rate, nitrogen flow rate, and process pressure were used in TECs representing each OP in the data sources, with the exception of OP2 of Furusjö et al., which was excluded due to very poor experimental balance closures (−18 % for total mass, −30 % for carbon) [12]. Comparisons between experimental data and model predictions are shown in Figs. 3, 4, and 5. The difference between measured and calculated temperatures, shown in Fig. 3, is up to 185°C. This is very high compared to the span in the experimental data but not greater than the deviations shown possible to result from uncertainty in BL composition (cf. Fig. 1). It is very clear that while the correlation between model and experimental temperature within each data set is very strong, there are large differences between the data sets. The two BL fuel data sets that use BL from the same mill have deviations of opposite sign. This points clearly to that reasons for the temperature deviations are specific to a whole data set but not to the fuel type. Such a factor can be data set wide systematic errors in flow measurements or fuel composition; a single fuel composition is used for all OPs in each data set as discussed above. Considering the simulation results in Fig. 1 and the magnitude of the differences in Fig. 3, fuel composition is a likely reason. This is further discussed below. It should be noted that the TEC temperature predictions presented in this work do not agree with those presented by Wiinikka et al. [10] for their data. The reason for this is not clear, but one possibility is estimation of BL heat of formation. Wiinikka et al. [10] do not specify what assumptions they use regarding combustion products. As discussed in Section 3.1, this can have a very large influence on the calculations. The predicted flows of major syngas species are fairly consistent with experimental data as shown in Fig. 4, despite the large deviations between model and experimental temperatures. The largest differences are observed for hydrogen but once again with different sign for the two BL data sets (black and white diamonds in Fig. 4a). Data for minor syngas species are shown in Fig. 5. As expected, methane concentrations do not agree well. In fact, TEC methane flows are very close to zero for the Furusjö and Jafri data sets while experimental data shows 0.2-1.5 % of total fuel carbon as methane. For the Wiinikka data set, the model methane flows are more in agreement with experimental data (white squares in Fig. 5a), but this is explained by the very low predicted temperatures for this data set (black squares in Fig. 3) that are not consistent with experimental temperatures. Hence, in agreement with previous studies [7,10], it can be concluded that methane does not follow thermodynamic equilibrium. From Fig. 4b, it can be seen that experimental GL TIC is higher than predicted by TECs. This, in combination with the fact that hydrogen carbonate is found in GL but not in the smelt leaving the gasifier, points to the fact that GL absorbs carbon dioxide from the gas, which leads to formation of hydrogen carbonate as shown in Eq. 1. The absorbed amount is, however, not large enough to have any large effect on the carbon dioxide syngas flows as shown in Fig. 4a. Predictions of GL sulfur are fairly consistent with experimental data as shown in Fig. 4b. It is nevertheless clear from Fig. 5a), which is predicted well, but this (accidental) agreement is an effect of the erroneously high predicted temperature as shown below. For the Furusjö et al. data set, the explanation for the fact that GL sulfur is well predicted but not gas phase sulfur is that the experimental sulfur balances do not close well [12]. Hence, the sulfur split is uncertain for that data set. There is no GL data reported for the data set of Wiinikka et al., so no comparison to experimental GL sulfur is possible. COS experimental flows (Fig. 5b) are much lower than TEC predictions. The likely reason for the lower measured values is not only that equilibrium is not reached in the gasifier but also that COS is known to be hydrolyzed in the quench [9]. Hence, the The three data sets are represented by coloring. No COS flow data available for Wiinikka et al. [10] COS flow measured in the cool syngas is not representative of the gas leaving the gasifier. Constraining methane It has been reported that constraining methane concentration, i.e., accounting for this deviation from equilibrium, in CFD and thermodynamic modeling of BLG has a large effect on predictions of temperatures and major syngas species, especially hydrogen [7]. Constraining the model methane production to the experimental values changes the predicted temperature by 15 ± 4°C for the BL cases (Jafri et al., Wiinikka et al.) and 4 ± 1.6°C for the STL case (Furusjö et al.). Considering the differences between model and experimental data in Fig. 3, this difference is not significant. This is partly explained by the fact that methane only accounts for 0. The major effect of constraining the model methane production, aside from the methane flows themselves, is on hydrogen production in agreement with previous work [7], which decreases by 11 ± 3 % for the BL data sets (Jafri et al., Wiinikka et al.) but only by 1.6 ± 0.6 % for the STL case (Furusjö et al.). Hence, constraining methane improves the agreement between model and experimental values for hydrogen for the Wiinikka et al. data but does the opposite for the Jafri data. Overall, the agreement between model and experimental values for major syngas components is not improved significantly compared to Fig. 4a. The other major deviation from equilibrium, related to sulfur distribution, also has a smaller influence on predictions of other parameters. Clearly, there are other reasons for the deviations between model and experimental data in Figs. 3 and 4. It should be noted that an adjustment of the Gibbs free energy for methane or the introduction of an activity coefficient can potentially be good alternatives if the aim is to develop a model that can predict methane formation. Such a modification would preferably be based on fitting experimentally observed methane formation for well-validated experimental data. However, initial tests with the data sets used in this work indicate that it is not possible to find a single modification that enables reasonable methane prediction for more than a single data set simultaneously. Similarly, other nonequilibrium species, such as the sulfur species discussed above, can potentially be predicted based on modifications of the thermodynamic equilibrium model. BL composition adjustment A systematic deviation of 5-6 % in the experimental carbon balance is present for all OPs in the results of Jafri et al. [6]. No such systematic deviation is present for the other elements investigated (Na, K, S). Hence, flow measurement errors are not likely the reason. The 5-6 % balance error is about twice as large as expected based on reported BL analysis uncertainty and 95 % confidence, cf. Table 2, but fuel analysis error is still a likely main contributor to the observed balance deviation. To test this hypothesis, we adjusted the fuel carbon content for the Jafri et al. data set from that in Table 2 to a higher value that allows the carbon balance to close using experimental data. Figures 6 and 7 shows the results for the Jafri data set after this adjustment together with the last three OPs of Furusjö et al., which all showed carbon balance closures within 2 % [12]. The data of Wiinikka et al. does not allow balances to be calculated since they did not quantify the inorganic phase and are not included for this reason. The data in Figs. 6 and 7 allows comparison between simulation results and experimental data that does not suffer from large measurement errors (as shown through the good overall and carbon balance closures for these OPs). Hence, the risk of influence from deviations caused by poor experimental data is lower, which means that the real deviation of the process from equilibrium can be studied. From the temperatures shown in Fig. 6, it is clear that the agreement between measured and model temperatures is now excellent for both data sets. The influence of this fuel composition change is dramatic (in comparison with Fig. 3), which is in agreement with the sensitivity to fuel composition observed in Section 3.1 and points out the risk of totally misleading conclusions due to uncertain fuel analysis data. Figure 7a shows that the BL carbon composition adjustment also gives an improved prediction for major gas components; as an example, the average relative prediction error for hydrogen is reduced from 18 to 5 % despite the fact that no constraint is used for methane. However, the adjustment of fuel composition for the Jafri et al. data does not improve the prediction of minor gas components (not shown) compared to the unadjusted data (Fig. 5). In fact, the hydrogen sulfide flow predictions deviates more after fuel composition adjustment due to the change in model temperature. The Fig. 6 The relation between measured reactor temperatures and model predictions for selected/adjusted data with good carbon balance closure (see text). Data sets represented by coloring deviation for GL inorganic carbon species does not change and still indicates carbon dioxide absorption in the GL. These results indicate clearly that the deviation in sulfur species and methane are real process deviations from equilibrium. The much improved agreement between model and experimental data for the Jafri et al. data set after adjusting the fuel C content to match the experimental carbon balance does not prove that fuel analysis error is the sole source of error. It is, however, a strong indication that fuel analysis is a main contributor and once again emphasizes the large influence of fuel analysis on the TEC. In order to investigate the influence of the non-equilibrium methane behavior on the rest of the components, the same selected data set was used for TECs with methane flows constrained to equal experimental data. The predicted temperature does not change greatly (not shown) and the agreement between model and experimental data for syngas components is improved marginally as shown in Fig. 8. The overall consistency between model and experimental data indicates that TECs can be used to both understand and predict the behavior of EF gasification of spent pulping liquors. The difference between model predictions and experimental data for the major syngas species, as shown in Fig. 8a, is 3 % relative on average with a maximum individual value of 7 %. However, the sulfur gas-smelt distribution still deviates substantially from equilibrium, cf. Fig. 8b. Influence of process conditions on process performance This section investigates the effect on gasifier performance of selected parameters important in the design of a commercial BL or STL gasification-based biorefinery: reactor heat loss, feedstock pre-heat temperature, feedstock DS content, and process pressure. Base case values used are 0.7 % heat loss, 150°C feedstock pre-heat, 75 % DS, and 30 bar, which are similar to the data from the pilot-scale experiments discussed in Section 3.3 with the exception of heat loss, as discussed below. Theoretically, the optimum point for operating a gasifier is when exactly enough oxidant is added to avoid formation of elemental carbon and achieve complete gasification [29]. For the gasification process treated in this work, a practical lower temperature limit is around 1000°C, mainly determined by the fact that very high carbon conversion is required to ensure that pulping chemicals can be recovered [6,8]. It has been Fig. 7 The relation between measured molar flows for major species and model predictions for selected/adjusted data with good carbon balance closure (see text). Data sets represented by coloring. Plot (a) shows major syngas species. Plot (b) shows inorganic species: carbonate, sulfur, and total inorganic carbon (TIC) shown that an optimal temperature with respect to CGE is in the range in which the pilot-scale gasifier is normally operated [10]. Based on data from the previous section, a temperature of 1050°C is assumed and the required amount of oxygen to reach this temperature is calculated through the gasifier energy balance for the various cases simulated. Thus, the cases with lower heat loss and higher DS content will require a lower oxygen addition to give the required temperature. Moreover, gasification at 1050°C is assumed to give a methane concentration of 1 vol.% in syngas for BL [6] and 0.2 vol.% for STL [12]. Predictions of fraction of sulfur in the gas cannot be interpreted quantitatively due to the deviations discussed above, but it has been shown that changes in the process lead to changes in the same direction as dictated by equilibrium [7,8], so trends in model predictions are still useful. For STL cases, the fuel composition in Table 2 was used. Due to the disagreement between simulations and experimental results for the BL pilot-scale data from Jafri et al. and Wiinikka et al., those BL compositions were not used in this section. Instead, typical BL compositions according to Table 3 were used. The difference between the two BL compositions in Table 3 is electrostatic precipitator (ESP) recycling. Due to fly ash formation in a black liquor RB, a recirculation of ESP ash, up to 10 % of the BL DS flow [30], is normally practiced in Kraft pulp mills. This gives the BL fed to the RB a higher ash content than would otherwise be the case. Since there is no fly ash formation in the gasification process, the rightmost column in Table 3 is relevant for a case where the RB is replaced by a gasifier while the left column can be relevant for a case in which there is an RB and a gasifier in parallel operation. Potassium and chlorine enrichment in ESP ash was disregarded since they are not important for the gasification process. From a gasification point of view, the main effect of removing the ash recycle is that the heating value of the BL increases due to a lower inorganic fraction. Figure 9 shows results from TECs with varying heat loss in the interval 0-5 %, where 4.2 % represents the pilot plant cases and 0.5-1 % is a realistic interval for a commercial plant. The influence on CGE is 2-5 % units for different CGEs and different feedstocks. This shows that energy efficiency will be better in commercial scale. The influence on H 2 /CO and sulfur split is very limited (not shown), which indicates that pilotscale experiments can be relevant for predicting at least H 2 / CO. For the sulfur split, it is not known if the deviation from equilibrium, which is substantial, is dependent on scale. Successive simulations, discussed below, were done at 0.7 % heat loss, which is considered relevant for a commercial-scale plant. Varying feedstock pre-heat (not shown) is analogous to changing the reactor heat loss from a TEC point of view; heat loss is a negative term in the gasifier energy balance while feedstock sensible heat is a positive term. The 100°C temperature interval investigated for feedstock pre-heat corresponds to a 1.9-2.7 % difference in sensible energy compared to the feedstock heating value (HHV), which is about half of the interval studied for heat loss. Consequently, the effect on CGE is about half of that in Fig. 9. The CGE for HHV in Fig. 9 (lines without symbols) show that practically 100 % of any heat that is added through either increased pre-heating or decreased heat loss is converted to chemical energy in the gas. This is a consequence of the energy balance of the gasifier since the smelt chemical and sensible energy is largely unaffected by the change due to the constant temperature used in these simulations. As noted above, less oxygen is required to reach 1050°C if heat losses are lower. Decreasing heat loss or increasing pre-heating can thus be viewed as converting thermal energy to chemical energy with 100 % efficiency. It should be noted that varying feedstock pre-heat can also have other effects on the Fig. 9 Influence of reactor heat loss on cold gas efficiency on HHV basis (lines) and H 2 + CO S-free LHV basis (diamonds) for the three feedstock compositions BL (solid), ash recycle adjusted BL (dashed), and STL (dotted) gasification, e.g., gas phase sulfur and methane formation [8], but these effects cannot be predicted by TECs. An industrially relevant interval for BL DS content is 65-85 %, with 65 % representing old pulp mills or mills with liquors that are particularly difficult to concentrate and 85 % representing a future scenario. Modern mills typically reach 80 % DS [30]. It is evident from Fig. 10 that feedstock DS content has a strong influence on the process, which is not surprising given that a 65 % DS liquor contains three times more water per kilogram of solids than an 85 % liquor. Due to the high process temperature, the penalty for added thermal ballast is significant. The only major effect of pressure in the interval 5-60 bar is on the sulfur distribution. For BL, the sulfur split increases from 20 to 60 % when increasing the pressure from 5 to 60 bar, as shown in Fig. 11. The influence on HHV CGE seen in Fig. 11a is simply an effect of this sulfur shift since syngas HHV includes contribution from S species, as evidenced by the lack of effect on CGE on S-free LHV basis. Wiinikka et al. [10] showed that equilibrium methane concentrations increase with pressure. Our results show the same behavior, but the effect is relatively small and, as discussed above, thermodynamic equilibrium methane predictions are far from experimental values. In general, the results in Figs. 9, 10, and 11 clearly show the effect of inorganic ballast; the STL feedstock with approximately half the inorganic content of BL has a significantly higher CGE in all cases. There is also a substantial effect from the recycled ash, as shown by the different CGE values for the typical BL and the ash recycle adjusted BL, which has approximately 10 % lower sodium content. In addition, the fraction of sulfur that ends up in the gas phase is much higher for the STL case. This is explained by the proportions between sulfur and sodium/potassium in the different feedstocks. For STL, the molar S/(Na + K) 2 ratio is 1.6 which means that even if all sodium formed sodium sulfide, 40 % of the sulfur would go into the gas phase. In practice, more sulfur ends up in syngas due to the equilibrium between sodium sulfide and sodium carbonate. Oxygen is a significant cost for operating a gasifier of this type [5]. Simulation results show a large effect on oxygen consumption from some of the parameters studied, e.g., a reduction of 33 % when increasing DS from 65 to 85 % for the ash-corrected BL. Generally, any change that improves CGE also decreases specific oxygen consumption. This is simply a consequence of the fact that added oxygen is used to oxidize feedstock to obtain the required heat for the process. Conclusions The high temperature and the catalytic activity of feedstock alkali makes thermodynamic equilibrium a better predictor of product composition in EFG of spent pulping liquors than for many other types of biomass and gasification technologies. TECs can predict the flows of the main syngas and slag products with high accuracy as shown by comparison with experimental data with small measurement errors. Small changes in feedstock composition have a relatively large influence on the thermodynamic equilibrium model predictions. This means that the model is sensitive to any errors in input data due to analytical uncertainty. However, since it was also concluded that the process follows thermodynamic equilibrium fairly well, this also means that the process itself is sensitive to naturally occurring variations in feedstock composition, e.g., due to seasonal changes, which can influence important operating parameters such as oxygen consumption and CGE. This is important knowledge for further research and process design. The main process deviations from equilibrium are methane formation and sulfur distribution between gas and slag. Constraining methane to the experimental value improves prediction accuracy of other gas species. Investigating the possibility of implementing empirical modifications to the equilibrium model in order to predict methane and sulfur split are important areas for further research. The full carbon conversion predicted by the thermodynamic equilibrium model is not a major deviation from the real process, which is in contrast to modeling other biomass gasification technologies. The gasification of BL and STL gives a different product distribution, due to the different feedstock compositions, but can be described with the same thermodynamic equilibrium model with the same main deviations from experimental data. This indicates that gasification of other fuels with similar properties, e.g., high alkali content, may also be possible to describe using the same model. The simulations of a commercial-scale gasification process show that CGE on S-free LHV basis can reach over 80 %. There are large efficiency gains connected to reducing ballast, in the form of water or inorganics, as well as reducing heat losses. The on-going work with BL/pyrolysis oil blends is one possible way to decrease inorganic ballast while maintaining the high reactivity and carbon conversion. Open Access This article is distributed under the terms of the Creative Comm ons Attribution 4.0 International License (http:// creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.	9,563.4	2016-11-11T00:00:00.000	[ "Environmental Science", "Engineering", "Chemistry" ]
Superconductivity in non-centrosymmetric ThCoC2 Superconductivity in compounds whose crystal structure lacks inversion symmetry are known to display intriguing properties that deviate from conventional BCS superconducting behavior. Here we report magnetization, resistivity, and heat capacity measurements on polycrystalline samples of ThCoC2, which has been reported to crystallize in the non-centrosymmetric CeNiC2 prototype structure, and show clear evidence of bulk superconductivity in ThCoC2 with a critical temperature of Tc = 2.65 K. From the specific heat data we find a Sommerfeld coefficient of γ = 8.38 mJ mol−1 K−2 and a Debye temperature of ΘD = 449 K. Interestingly the Hc2 superconducting phase diagram displays positive curvature, and the specific heat at low temperature deviates from conventional exponential temperature dependence, which is suggestive of possible unconventional superconducting behavior in ThCoC2, similar to that seen in the isostructural and isoelectronic non-centrosymmetric superconductor LaNiC2. Introduction One of the highlights of Solid State Chemistry in recent decades has been the discovery of many types of ternary transition metal carbides that crystallize in different novel prototype structures and exhibit interesting magnetic and electrical properties. These compounds by definition are composed of carbon, a transition metal, and a highly electropositive multivalent metal such as a lanthanide (Ln), Sc, Y, or Th. The complete ionization of the electropositive metal to stable ions Ln 3+ , Y 3+ , or Th 4+ leads to a negatively charged transition metal-carbon sub-network, which can be regarded as an organometallic net [1][2][3]. An important characteristic of these carbide compounds is the presence of single and/or multiple metal-carbon and carbon-carbon bonds. Compounds of particular interest in this paper are those that crystallize in the CeNiC 2 prototype structure, which is base centered orthorhombic space group Amm2 (No. 38) [4]. The structure is formed from the stacking of alternating two-dimensional (2D) NiC 2 and distorted hexagonal rare-earth-metal sheets, with the structure of ThCoC 2 , shown in figure 1. This kind of prototype structure is selective in electron count with the transition metals limited to group VIII Fe, Co, and Ni atoms [5]. In these compounds the stoichiometric formula is given by LnTC 2 , YTC 2 , or ThCoC 2 (T = Fe, Ni, or Co) and interestingly have short bond lengths between adjacent carbon atoms in the range of 1.32-1.47 Å, which is suggestive of carbon-carbon double bonds in the organometallic net, as shown figure 1. More than 30 compounds have been found to crystallize in the same CeNiC 2 prototype structure, particularly from a series of rare-earth (R) carbides RNiC 2 and RCoC 2 [6][7][8][9][10][11][12][13]. Neutron diffraction studies on these series (R = Pr, Nd, Tb, Dy, Ho, Er, and Tm) [8][9][10][11][12][13] found different magnetic order between the series, with RCoC 2 compounds generally becoming ferromagnets and RNiC 2 compounds ordering antiferromagnetically in different spin configurations [13]. The studies found that the magnetic order arises from the 4f electrons confined on the rare earth sites, whereas the Co and Ni sites remain non-magnetic, and is also evident by a very weak temperature-dependent magnetic susceptibility in YCoC 2 [8], YNiC 2 [7], and LaNiC 2 [14]. Interestingly LaNiC 2 becomes superconducting below 2.7 K [14] and is therefore classified as a non-centrosymmetric (NCS) superconductor since the CeNiC 2 prototype structure lacks an inversion center. This structural characteristic brings important implications for the physics in a superconducting state. Most superconducting materials possess an inversion center in their structures and have the Cooper pairs classified as either an even-parity spin-singlet, or an odd-parity spintriplet state in accordance to the Pauli exclusion principle and parity conservation. However, the classification of the Cooper pairs into even or odd-parity states relies on the presence of an inversion center in the crystal structure. The absence of inversion symmetry introduces an antisymmetric spin-orbit coupling (ASOC) that lifts the spin degeneracy with spin-split Fermi surfaces, and importantly allows a mixing of spin-singlet and spin-triplet pairing states into a so called mixed-parity pairing state. Exotic properties can therefore emerge in parity violated superconductors such as time reversal symmetry (TRS) breaking, nodal gap structure, and other unconventional behavior attributed to the lack of inversion symmetry [15][16][17][18][19][20]. Indeed, it was recently demonstrated that LaNiC 2 shows TRS breaking in muon spin relaxation (µSR) measurements [21], and evidence for nodal gap structure in magnetic penetration depth measurements [22]. Therefore, in this work we examine the compound ThCoC 2 that is reported to be iso-structural [23] and iso-electronic to LaNiC 2 , and show that ThCoC 2 is likely a new non-centrosymmetric superconductor with a critical temperature of T c = 2.65 K. The results extracted from specific heat, magnetization, and resistivity measurements suggest that the superconducting state is similar to that seen in LaNiC 2 and possibly unconventional in origin. Experiment The polycrystalline ThCoC 2 sample was synthesized by arc melting together the stoichiometric amounts of the elements Th (3N), Co (4N), and graphite C (5N) on a water cooled Cu hearth in an arc-furnace under UHP (6N) Argon atmosphere and using a Zr getter. The samples were flipped over and re-melted 5 times to ensure good homogeneity, with minimal resultant weight loss (<0.7%) from arc melting. The sample was subsequently wrapped in tantalum foil and sealed in a quartz tube under vacuum for annealing treatment at 1100 • C for 14 days to improve sample quality and remove any unreacted magnetic impurities. Powder x-ray diffraction (XRD) patterns were obtained using a Rigaku MultiFlex diffractometer equipped with a monochromator providing Cu Kα (λ = 1.540 56 Å) radiation. The diffraction pattern analysis and phase identification were done using MDI Jade software. Magnetization measurements were performed down to T = 1.8 K using a commercial VSM-SQUID magnetometer by Quantum Design. Electrical resistivity and heat capacity measurements were obtained using a physical property measurement system (PPMS) also by Quantum Design. The specific heat of a sample piece polished flat (∼17 mg) was measured in the temperature range of 0.45-20 K with a He 3 calorimeter using the relaxation method. The resistivity sample was polished into a flat block (2.6 mm × 0.8 mm × 0.25 mm), and fine platinum wires were spot-welded to the sample as voltage and current leads for the standard four-probe method. The AC resistivity was measured down to 0.6 K with an applied current of 10 mA using a Linear Research Inc. model LR700 AC resistance-bridge and controlled by Lab View data acquisition software. Figure 2 shows the XRD pattern and corresponding Miller indices for the annealed polycrystalline ThCoC 2 sample. The peaks are indexed to the orthorhombic CeNiC 2 prototype structure with space group Amm2 (38), and the refinement of the lattice parameters yields a = 3.8053 Å, b = 4.5321 Å, and c = 6.1424 Å, in good agreement with results reported in literature [23]. These results suggest that the ThCoC 2 sample is phase pure at least within the standard detection limit for powder XRD method (∼5%), and is consistent with the CeNiC 2 prototype structure. To our knowledge the investigation of ThCoC 2 compound has been rather limited [23], and therefore it would be highly desirable to have additional in-depth structure determination by single crystal x-ray or neutron diffraction experiments to definitively demonstrate ThCoC 2 crystallizes in the non-centrosymmetric CeNiC 2 prototype structure, as has been shown for many of the other tertiary rare-earth carbides RTC 2 by neutron diffraction studies [8][9][10][11][12][13]. Results and discussion The temperature dependent magnetization using zerofield cooled (ZFC) and field cooled (FC) method using an applied 50 Oe magnetic field shows a clear diamagnetic transition onset around 2.55 K in figure 3. It should be noted that the as-cast samples slowly decomposed in air and had a small residual ferromagnetic signal, likely from small inclusions of unreacted Cobalt or a Cobalt binary, which makes the annealing treatment critical to stabilize the sample and eliminate any magnetic impurities. The hysteresis between ZFC and FC regimes indicate ThCoC 2 is a type II superconductor. Without correcting for demagnetization or sample size effects, we estimate the superconducting volume fraction (ZFC) to be 110% of perfect diamagnetism, which suggests possible bulk superconductivity. The Meissner flux expulsion (FC) is about 5% of the diamagnetic flux expulsion, a characteristic of strong flux pinning. The inset in figure 3 displays 4-quadrant M versus H data at T = 1.8 K and shows weak type II superconducting behavior. Sample resistivity from 0.6-300 K is presented in figure 4, with the figure inset showing the superconducting transition suppressed with applied magnetic field. The polycrystalline ThCoC 2 sample is a very good metal with resistivity increasing linearly above T > 100 K at a rate of 0.33 µ cm K −1 , and an extremely low residual resistivity of 0.37 µ cm at T = 4 K. The resultant residual resistivity ratio (RRR = ρ 300 /ρ 4 ) is thus a very large RRR = 218, indicating excellent alloy quality of the sample. The zero magnetic field data shows a sharp superconducting transition centered at T c = 2.65 K and a zero-resistance temperature of T c = 2.55 K. The sharp superconducting transition ( T C ∼ 0.2 K) is consistent with the magnetization measurements and is also indicative of good sample quality. While the magnetization and resistivity data suggest bulk superconductivity in ThCoC 2 , an anomaly in the specific heat measurement is necessary for confirmation. Figure 5 displays the specific heat divided by temperature (C/T ) versus T 2 in various applied magnetic fields. A jump in the specific heat is clearly observed with the midpoint of the transition at T c = 2.53 K and a small transition interval T C ∼ 0.3 K . The consistency between the magnetization, resistivity, and heat capacity transitions is clear evidence of bulk superconductivity in ThCoC 2 . The expected suppression of the superconducting transition with applied magnetic field is also seen in figure 5, along with a broadening of the transition that is not unusual for polycrystalline samples. Analysis of the superconducting state first requires a comparison to the normal state specific heat C n . The normal state specific heat is comprised of the electronic contribution C e and the lattice contribution C lattice , such that C n = C e + C lattice . The electronic contribution is the linear Fermi liquid term C e = γ T , with γ being the Sommerfeld coefficient. The lattice contribution to the heat capacity at low temperature is usually described just by the cubic Debye model term βT 3 ; however an increasing deviation above T c requires the addition of the next higher order phonon dispersion term, such that C lattice = βT 3 + δT 5 . Fitting the normal state specific heat data to the expression C n = γ T + βT 3 + δT 5 results in the fit parameters γ = 8.38 mJ mol −1 K −2 , β = 0.0856 mJ mol −1 K −4 , and δ = 5.59 × 10 −4 mJ mol −1 K −6 . The β value corresponds to a Debye temperature of D ∼ 449 K. The Sommerfeld coefficient γ suggests a low density of states at the Fermi level typical of other transition metal superconductors, with corresponding LaNiC 2 compound having γ = 7.83 mJ mol −1 K −2 and D ∼ 496 K for comparison [14]. Subtraction of the phonon contribution C lattice from the total specific heat allows analysis of just the electronic contribution C e , displayed here as C e /T versus T in figure 6. The analysis of the specific heat anomaly shows the magnitude of the jump at T c to be C e /γ n T c ∼ 0.86, which is significantly smaller than the weak-coupling BCS limit of 1.43. Since the superconducting transition is a second order phase transition, thermodynamics requires the entropies of the normal and superconducting states to be equal at the transition temperature T c . This equality has the convenient geometric representation in figure 6 that the two enclosed areas above and below the γ = 8.38 mJ mol −1 K −2 baseline must be equal. The entropy balance is demonstrated in figure 6 with good agreement between the upper area of 6.07 mJ mol −1 K −1 , and lower area of 6.09 mJ mol −1 K −1 after the low temperature data is extrapolated to the origin. This thermodynamic consistency shows the lack of a residual heat capacity from impurity phases and demonstrates the excellent sample quality with full superconducting volume. The microscopic theory of superconductivity predicts the electronic specific heat of conventional fully gapped s-wave superconductors to display a low-temperature exponential behavior of the form exp[− (0)/k B T ]. The electronic specific heat for ThCoC 2 does not appear to follow this exponential temperature dependence that is expected for conventional BCS superconductors. This deviation from exponential behavior is visualized in figure 7, which displays the logarithmic graph of the electronic specific heat versus the inverse reduced temperature (T c /T ). Clearly the data shows a marked deviation at low temperature from a linear plot in figure 7 that would be expected for conventional superconductivity. Indeed it was shown that the specific heat of the analog LaNiC 2 compound also deviates from conventional exponential behavior with a low-temperature T 3 dependence [14] that is consistent with nodes in the energy gap, in addition to other evidence of non-conventional behavior of time reversal symmetry breaking Figure 6. Temperature dependence of the electronic specific heat contribution divided by temperature (C e /T ). The calculated upper (6.07 mJ mol −1 K −1 ) and lower (6.09 mJ mol −1 K −1 ) area between the data and the value demonstrates entropy balance. in µSR measurements [21], and evidence for nodal gap structure with T 2 dependence in magnetic penetration depth measurements [22]. It is worth noting that earlier conflicting reports on LaNiC 2 conclude conventional BCS behavior in heat capacity [24] and Nuclear quadrupole resonance (NQR) measurements [25]; however the recent report by Bonalde et al [22] shows that is likely a result of the measurements sensitivity to magnetic Fe impurities introduced by lower quality Ni (3N) and the insufficient low temperature evaluation of the superconducting state in those studies. They correctly point out that temperatures below 0.3T c are needed to properly determine the structure of the energy gap, in which they measured magnetic penetration depth down to ∼0.017T c in LaNiC 2 . The ThCoC 2 samples examined here are comparative to the higher quality LaNiC 2 samples, with the characteristic measurements indicating excellent ThCoC 2 sample quality without the signature of impurities being readily apparent. It is therefore not unreasonable to expect possible unconventional behavior for ThCoC 2 in future lower temperature examinations of the superconducting state. The upper critical field H C2 as a function of temperature from the resistivity and specific heat data is plotted in figure 8, showing good agreement between the measurements. Unexpectedly the superconducting phase diagram has positive curvature instead of the conventional dome like plot separating the superconducting and normal state domains. Both the midpoint of the resistivity transition and the zero resistance values are plotted to emphasize this unusual behavior, showing positive curvature close to T c then transitioning into an apparent linear behavior below T < 1.4 K. An accurate estimate of the upper critical field H C2 (0) at absolute zero is difficult without conventional modeling, but it is evident that superconductivity in ThCoC 2 is quickly suppressed in magnetic field with H C2 only 3000 Oe at T = 0.6 K. Interestingly, examples of positive curvature in the upper critical field diagram include the borocarbides LnNi 2 B 2 C [26,27], magnesium diboride MgB 2 [28], the non-centrosymmetric Li 2 (Pd, Pt) 3 B [29], and the non-centrosymmetric heavyfermion superconductor CeRhSi 3 [30]. This kind of behavior can be attributed to interband coupling that can occur within a multiband model [27,31], with the small gap contributing at low temperature. In fact, a recent study on the superconducting state of LaNiC 2 also displays positive curvature near T c and an enhancement of H c2 (0) [31]. Chen et al systematically measured the low temperature London penetration depth, heat capacity, and resistivity, and argue the superconducting state in LaNiC 2 is best described by a two-gap BCS model, which is the likely result of a moderate value of the ASOC that splits the spin degenerate bands. Within this scenario the analog non-centrosymmetric superconductor ThCoC 2 could be a new example of a multiband compound, making additional low temperature characterizations of the superconducting state necessary to determine the exact nature of the energy gap. Conclusion In summary, polycrystalline samples of the reported noncentrosymmetric ThCoC 2 compound were characterized by magnetization, resistivity, and heat capacity measurements that unambiguously show bulk superconductivity with a critical temperature of T c = 2.65 K. The superconducting state shows interesting behavior with a deviation from conventional BCS exponential temperature dependence in the specific heat, and positive curvature in the superconducting phase diagram. These results, in conjunction with those of analog noncentrosymmetric superconductor LaNiC 2 , suggest possible unconventional superconducting behavior and warrant future low temperature investigations of the superconducting state in ThCoC 2 . The limited reports on ThCoC 2 also make it a good candidate for precise structure analysis by such methods as single crystal x-ray or neutron diffraction experiments.	4,068.4	2014-01-01T00:00:00.000	[ "Physics" ]
Selection of Unmanned Aerial Vehicles by Using Multicriteria Decision-Making for Defence eunmanned systems have been seeing a significant boom in the last ten years in different areas togetherwith technological developments. One of the unmanned systems is unmanned aerial vehicles (UAVs).UAVs are used for reconnaissance and observation in themilitary areas and play critical role in attack and destroymissions.ese vehicles have beenwinningmore features together with developing technology in todays world. In addition, they have been varying with different features. A systematic and efficient approach for the selection of theUAV is necessary to choose a best alternative for the critical tasks under consideration.emulticriteria decision-making (MCDM) approaches that are analytic processes are well suited to deal intricacy in selection of alternative vehicles.is study also proposes an integratedmethodology based on the analytic hierarch process (AHP) and technique for order preference by similarity to ideal solution (TOPSIS) to evaluate UAV alternatives for selection process. Firstly, AHP, aMCDMmethod, is used to determine the weights of each critical factor. Subsequently, it is utilized with the TOPSIS approach to rank the vehicle alternatives in the decision problem. Result of the study shows that UAV-1 was selected as the most suitable vehicle. In results, it is seen that the weights of the evaluation criteria found by using AHP affect the decisionmaking process. Finally, the validation and sensitivity analysis of the solution are made and discussed. Introduction Unmanned aerial vehicles (UAVs) are named such as remotely piloted vehicles, drones, robot planes, and pilotless aircraft, and in many ways. UAVs can fly autonomously or semiautonomously [1]. UAVs became an important element of many modern militaries and various civilian areas. Demands have been increasing for a variety of types of unmanned vehicles due to their success on battlefields. At the same time, they have capabilities such as persistent surveillance, tactical and combat reconnaissance, resilience together with its low benefit risk, and low cost. Increasing use of these vehicles is rising alongside increasing demands for battlefield intelligence, tighter defence budgets of countries, faster operation tempos, and lowered tolerance for casualties in the defence area. Besides, UAVs are also emerging as a suitable vehicle for a wide range of civil applications such as disaster monitoring and atmospheric observation [2]. As in developed countries, Turkey is also developing UAVs for defence and civilian usage. UAVs had been mainly developed for military applications and military areas a few years ago. However, now, they have composed serious enhancements in the design and capabilities of UAVs with their varying size due to technological advancements in robotics. Today, UAVs' numerous civil applications have recently emerged due to their reliability and operation with a very good level of flight stability. UAVs have some advantages according to their alternatives, such as a helicopter. ese advantages are required to reach places that are difficult to access, long periods of repetitive work, or operating in dangerous conditions, especially for extended periods of time or under stressful conditions and in risky tasks [3,4]. So, these advantages had ensured to use these vehicles widely. e architecture of a normal UAV consists of different five components that are the flight system, the control system, the monitoring system, the data processing system, and the landing system. e selection of the best UAVs with developing technology is very important in defence and affects the ability and the necessity of modern armies as well. Being able to perform reconnaissance and surveillance missions in the most effective and efficient way with well-chosen UAVs is depended on decisions of planners and managers. It is very important to create and invent them, but they are varied with developing technology day by day. us, a multicriteria selection process is needed for the best UAV decision with developing technology. e main objective of this study is to propose a systematic MCDM model to help in the defence area for the selection of the most suitable UAV among a set of available alternatives. Selecting a new UAV is a strategic decision-making process that has a high complexity because of the criteria, which must be considered simultaneously. Besides, most of these criteria affect each other, are contradictory, and an increase in one of the criteria's compliance might reduce the compliance of the other. So, in order to select the best UAV among the alternatives on the market, we need to consider various evaluation criteria. e MCDM methods help to choose the best alternative by considering various criteria and by evaluating all the alternatives. e academic literature has some examples of the application of the MCDM in the defence field. Some of them are evaluating naval tactical missile systems with fuzzy AHP [5]; selection of attack helicopters by AHP [6]; evaluation of the best main battle tank with fuzzy decision theory [7,8]; weapon selection by using AHP and TOPSIS [9], goal programming [10]; ELECTRE (elimination and choice translating reality) [11]; evaluating the military training aircraft by using MCDM under fuzzy logic [12,13]; using MCDM methods for determination of the best military cargo aircraft [14]; selection of investment projects in the defence industry [15]; and military airport location selection by using the hybrid application of AHP-PROMETHEE-(Preference Ranking Organisation Method for Enrichment Evaluation-) VIKOR (Serbian: Vise Kriterijumska Optimizacija I Kompromisno Resenje) methods [16]. Although UAVs play an important role in the design of an effective defence system for the military area, the academic literature about the selection of UAVs is limited. e study described in this paper has two specific objectives: (1) to introduce and describe the importance of UAV technologies and their applications; (2) to offer an analytic process that is based on AHP and TOPSIS methods for the best selection among the alternative UAVs to help decision makers in the defense area. e remainder of this study is structured as follows: the literature review about UAVs is presented in Section 2. In Section 3, AHP and TOPSIS, multicriteria decision-making methods, are explained. e selection process by using AHP and TOPSIS is made among the various UAVs in Section 4. Conclusions are presented in Section 5. UAVs in the Literature Although UAVs are developed for military purposes, including reconnaissance and attack roles, they are quickly gaining popularity worldwide in various civil areas. However, now, they are used in lots of areas. UAV-based systems have many advantages according to manned air vehicles: first, the cost of use related to the low purchase, management, and operation costs. ey can yield high-resolution images which are useful for traffic analysis based on video image processing. UAVs have some disadvantages such as low battery duration, battery life, limited UAV payload, vegetation, buildings, urban canyons, limited take-off mass, limited wing surface, limited wing loading, and no-fly zones which affect negatively. So, their applications for this field are limited and influenced by some factors affecting their performance, such as weather conditions should be mentioned, technical instrumental problems, and physical obstacles [17]. UAVs are widely used, especially in traffic and the construction industry. UAVs have been used for various aims such as safety/security monitoring, inspections, surveying, and aerial photography in the construction industry [18]. In traffic, there are applications such as to obtain detailed traffic information on real time [19]; to evaluate traffic flow conditions in urban areas using videos through the UAV [20]; and to estimate traffic flow parameters of a road intersection through a video image processing technique using an UAV [21]. Lee et al. mentioned about traffic and roadway incident monitoring via the UAV [22]. Khan et al. presented an extensive systematic and practical study on how to conduct an UAV-based traffic study [23]. Salvo et al. developed a new methodology to evaluate the real traffic flow conditions. In their methodology, they used the videos acquired by UAVs [20]. Barmpounakis et al. aimed to review research dedicated to using unmanned aerial systems in transportation in their study. ey mentioned the advantages of the airborne video as a means for acquiring highquality naturalistic data for both practitioners and researchers in their study [24]. Besides, these UAV applications are applied in different areas such as inspection of critical linear infrastructure such as oil and gas pipelines or electrical transmission lines and inspection of wind turbine blades by UAVs with imaging. UAVs are also used in the real-estate industry on the purpose of conducting aerial surveys and mapping of planned developments or to document transactions. At the same time, UAVs are being used in decision-making in agriculture on whether crops need to be watered and where to apply the fertilizer. Also, Gül [25] mapping operations in open-pit mines are made by using the UAV Erdelj and Natalizio search review of main disaster management applications with UAV [26]. ere are some papers about UAVs and their various applications in the literature. Hassanalian and Abdelkefi gave some brief information related to the civil uses of UAV/ drones and their fields of use. Also, they provided comprehensive information on the current status of the legal frames and regulations in Turkey and in the world [27]. Akgül et al. evaluated UAV and its systems. At the same time, they evaluated the usage of these vehicles for the forestry area [28]. Wu et al. discussed the development of a multiobjective mission flight planning algorithm for unmanned aerial system (UAS) operations within the National Airspace System (NAS) in their study [29]. Wu et al. presented a system for automated mission planning with a view to operate UAVs. Decision variables for their system were fuel consumption, flight time, wind and weather conditions, terrain elevation, airspace classification, and the flight trajectories of other aircrafts [30]. Arıca et al. presented a multicriteria path planning model for UAVs. eir model helps in the planning of optimal paths in terms of time, distance, and fuel consumption [31]. Caner demonstrated the pros and cons of UAVs compared with manned aircraft in his study. Besides, he provided information about the historical developments that have occurred and new technologies in the UAV [32]. Kiracı and Bakır focused on the selection of the aircraft to determine the most suitable aircraft for airline companies with different flight networks and different flight destinations with the TOPSIS method [33]. Ercan and Gencer investigated the literature on "dynamic route planning" for unmanned aerial systems [34]. Lin and Hung selected military UAVs using the fuzzy weighted average algorithm. ey used three main criteria in their study that are mission flexibility, operational suitability, and operational assessment [35]. Besides, some researchers, as distinct from the selection of UAVs, studied about aircraft type selection with multicriteria decision-making methods [36][37][38]. See et al. presented a multiattribute methodology for selecting the best aircraft among a set of alternatives. e authors used the method of hypothetical equivalents and inequivalents. ey used three criteria: speed, range, and the number of passengers [39]. Yeh and Chang proposed a fuzzy multicriteria decision-making algorithm with group decisionmaking for evaluation of the performance of each aircraft. eir study that the performance of each aircraft is evaluated through fuzzy rating used three main criteria and eleven subcriteria (main criteria: technological advances, social responsibility, and economic efficiency; subcriteria: aircraft maintenance capability, pilot adaptability, aircraft reliability, maximum range, passengers' preference, the level of noise, operational productivity, airline fleet economy of scale, operating cost, purchasing price, and corporate strategy) [40]. Gomes et al. proposed a fuzzy stochastic approach for the selection of aircraft. NAIADE method (Novel Approach to Imprecise Assessment and Decision Environments) was used in their evaluation process based on three criteria (financial, logistics, and quality), further articulated to twelve subcriteria (acquisition cost, liquidity, operating costs, range, flexibility, cruising speed, replacement parts availability, landing and take-off distance, comfort, avionics, availability, and safety) [41]. Bruno et al. proposed a model aircraft evaluation in their study. e model proposed includes four main criteria (economic performance, technical performance, aircraft interior quality, and environmental impact) and eight subcriteria such as aircraft price, operative cost, cruise speed, autonomy, seat comfort, cabin luggage compartment size, noise, and environmental pollution. eir aim is to propose a novel model for aircraft evaluation according to the airlines' needs [42]. Dožić et al. proposed a new methodology for the aircraft type selection problem with regard to different criteria that involve quantitative and qualitative aspects, three main criteria (aircraft characteristic, cost, and added value indicators), and ten subcriteria. ey used opinions of experts from different airlines and universities for the analytic evaluation process. Fuzzy AHP was used due to the uncertainty of decision problems in their study [43]. Petkovics et al. studied different subjects from other academics. ey selected the appropriate drone for the specific needs of farmers to collect the necessary data for precision agriculture using drones. ey selected the best drone among the two drone types for data collection in agriculture [44]. In this context, this paper proposes a novel model for UAV selection, which aims to overcome the complexity of the UAVs' evaluation process. e integrated AHP-TOPSIS methods that are multicriteria decision-making methods are used for the selection process. ese methods include a simple analytic process and basic calculations. e hybrid model proposed has a lower level of computational complexity, which facilitates its practical application. So, a systematic decision-making process was developed via the usability of the model and its applicability that helps decision makers in the UAV selection process. e contribution of the paper could be perceived through two main issues: the first one is related to the selection criteria used as the model input, which is of crucial importance for decision-making and modeling, while the second one refers to the MCDM technique. Based on the analysis of the relevant literature, the criteria used in the decision-making process of aircraft type selection are identified. Also, the contribution of the paper is the proposed methodology for UAV selection among the aircraft selection problem based on the integrated MCDM methods for the first time. We see various studies of aircraft selection using MCDM methods done in our literature research. And we see a few studies such as planning of the UAV route and its mission planning. However, literature is limited to the UAV selection process using MCDM methods. Besides, numerous articles have recently been published about various UAV applications in military use and research, have been a major driver for advancing UAV technology, and have made a big contribution. However, academic studies on decision-making about UAVs are limited. is study will contribute to the literature in this respect. Multicriteria Decision-Making In this section, the AHP and TOPSIS methods are presented. is study utilizes two MCDM methods, AHP to determine the weights of criteria and TOPSIS to rank alternatives and select the best alternative UAV. A brief description and steps of each method are, respectively, provided as follows. 3.1. Analytic Hierarchy Process. Analytic hierarchy process (AHP) is a flexible and effective decision-making process, developed by omas L. Saaty. is method is useful, making the best decision for quantitative and qualitative aspects of a decision in establishing priorities. e AHP method has been applied in various fields: management, production, transportation, agriculture, industry, allocation, and distribution of resources in the complex decision problems solving strategic decisions. ere are some causes of common usage of the AHP: (1) helps decision makers to find important degree in making simultaneous evaluation in the decision-making problem; (2) includes relatively basic mathematical calculation as compared to other analytic methods; (3) is flexible to be integrated in various decisionmaking methods such as ranking and programming models; and (4) has the ability to control mathematical calculations and judgment of decision makers [45]. A decision hierarchy structure of AHP includes three levels that are the goal, the criteria, subcriteria, and the alternatives. e hierarchy structure makes the problem more understandable and clearer for the decision makers at the decision process [46,47]. e selection process or calculating the weights in AHP has five major steps [48,49]: Step 1: determining decision problems, alternatives, and criteria. Establishing a matrix comparing the criteria and alternative pair wisely by using Saaty's scale. Saaty's scale: Extreme Importance-9; Very Strong Importance-7; Strong Importance-5; Moderate Importance-3; and Equal Importance-1. Step 2: calculate the criteria weights with pairwise comparisons. So, find relative importance of weights of evaluation criteria in the hierarchy by using the scale, 1-9 points, of Saaty. And pairwise comparison matrices are created. For example, n × n comparison matrix a is created for n criteria to the relative importance of the criterion i and the criterion j. Among the a matrix elements is the following connection: Step 3: then, the normalized decision matrix is created Step 4: calculate the consistency index (CI) measured as follows: Firstly, calculate the ʎ max value for total consistency; Step 5: calculate a consistency ratio (CR). If the CR is less than 0.10 (CR < 0.1), then the ratio shows an acceptable level of consistency in the AHP. If CR is more than 0.10 (CR > 0.1), the ratio is inconsistent as follows: with random index (RI) as given in Table 1. TOPSIS. e technique for order preference by similarity to ideal solution (TOPSIS) technique was established by Hwang and Yoon, an approach which presumes that each criterion tends toward a monotonically decreasing or increasing utility [50,51]. e necessitation of having the shortest distance to the positive ideal solution/the farthest distance from the negative ideal solution for the selection of alternatives is the fundamental concept of this technique [52]. It suggests the Euclidean distance strategy for this process that assesses the relative closeness of the selected alternatives to the ideal solution. us, a series of comparisons of these relative distances can be obtained with the preference order of the alternatives [53]. e first step includes the creation of an evaluation matrix which consists of m alternatives and n criteria. e intersection of each alternative with each criterion is given as x ij , and therefore, the matrix can be described as (x ij ) m ×n . e second step includes the normalization of the matrix: I � 1, 2, . . . , m; j � 1, 2, . . . , n. In the decision process, equation (7) is used to determine the distance between each alternative and the positive ideal point. e distance between each alternative and the negative ideal point can be determined with equation (8) utilizing the same separation measure [54]: where the CC * j index value lies between 0 and 1. e larger the index value means the better the performance of the alternative. e TOPSIS technique usually deals with benefit and cost data. In this paper, the positive ideal solution (PIS) is the one with the lowest cost and most benefits of all alternatives, and the negative ideal solution (NIS) is the one with the highest cost and lowest benefits of all alternatives. Selection of the Unmanned Aerial Vehicles e proposed decision model, composed of AHP and TOPSIS methods, consists of three stages. Firstly, identifying the criteria to be used in the model; second, finding the weight of criteria using AHP computations; and third, the final ranking of UAVs with TOPSIS. To fulfill this selection process, the AHP-TOPSIS hybrid model has been selected in this research due to three causes. ese three causes are simple mathematical and computational process, ranking the alternative locations based on their overall performance, and finally, the information requirements of the proposed framework are stratified into a hierarchy to simplify the information input and allow a selection problem to focus on a small area of the large problem. Besides, inconsistencies of the experts can be measured with CR values in these decision processes. A three-step methodology has been used in this study to the UAV selection process (see Figure 1). Identification of the Criteria and Expert Team. UAVs have experienced major development and gained fastgrowing popularity worldwide during the last several decades. Nowadays, these vehicles are extensively used in various critical military and defence applications for reconnaissance, surveillance, and security reinforcement [68]. Evaluation of UAVs plays an important role in the design of an effective defence system. e producers of UAV are constantly innovating and improving their UAVs. ese developing processes aim to answer to both needs of themselves and their potential customers. Besides, these processes are improved of technical features that deal with flight parameters such as flying range, speed, load capacity, and autonomous. In this scope, we evaluated six alternative UAVs for the defence area. We use seven performance criteria that are important criteria which belong to UAVs. ese criteria are wingspan (C1), length (C2), payload capacity (C3), endurance (C4), cruise speed (C5), operational altitude (C6), and range (C7). ese criteria are important for UAVs. For example, the criterion of the payload is one of the most important criteria for UAVs. With its capable payloads onboard, UAV can not only detect a target but also able to mark it with its laser designator and can attack. UAVs can take off, land, and fly fully autonomously for a mission without assistance from a pilot fully autonomously without assistance from a pilot. e payload capacity and variations are available for both civil and military applications. Criteria to be considered in evaluating UAVs are determined by the expert team. Furthermore, the judgments of experts are used in this methodology to evaluate the UAV selection process. e utilization of expert opinions is proposed in this methodology to define the problem statement, to identify the criteria for evaluating UAVs, and to the weight of the criteria in decision-making. e expert team determined six possible Table 2. is study aims selection of the best UAV as in the goal shown in the hierarchy's top. Six alternatives and seven criteria are shown in the hierarchy structure in the same figure. In AHP, the pairwise comparison matrix (Table 3) is formed to determine the criteria weights. Academic experts make evaluations using Saaty's 1-9 scale to determine the values of the criteria of pairwise comparison matrices. Table 4 shows the weight of criteria that are found as the result of comparisons with AHP. e criteria weights in the descending order are shown in Table 4 as can be seen that C3 and C5 were rated more important than the remaining criteria. Besides, C4 and C7 were rated relatively lower, and C1, C2, and C6 became the lowest-rated criteria. e consistency ratio is found by using formulas (1) and (2) and random index (RI). is value is an acceptable level. e decision hierarchy for selection of the best UAVs is shown in Figure 2. It is to obtain a weighted decision table in the TOPSIS method using the criteria weights calculated by AHP. e resulting weighted decision matrix in TOPSIS process is shown in Table 5. Positive ideal and negative ideal solution values are shown in Table 6. In this problem, C1 and C2 are cost criteria, whereas the other criteria, C3, C4, C5, C6, and C7, are benefit criteria. Finally, the experts evaluated the six UAV alternatives for each evaluation criterion with AHP. e weight of the criteria has been calculated to ensure of the TOPSIS weighted valuation matrix for the UAVs. e AHP-TOPSIS methodology, the decision matrix described, is made in Table 2. In the following process, the decision matrix is normalized and is shown in Table 4, together with weighted values. Table 5 shows the calculated weighted normalized matrix. In the following step, the ideal and negative ideal solutions are determined using equations (3) and (4), shown in Table 6. en, with equation (5), the distance from PIS and NIS is calculated, respectively, as their results are shown in Table 7. Eventually, the ranking has been calculated, and the score of each alternative is shown in Table 7. Consequently, the distance of each vehicle from A+ to A− can be currently calculated. e last step solves the similarities to an ideal vehicle. Based on CCi values in Table 7, the ranking of the UAVs in order is UAV-5, UAV-6, UAV-1, UAV-2, UAV-3, and UAV-4. AHP-TOPSIS model results indicate that UAV-4 is the best vehicle with the CCi value of 0.6173. Two different decision processes are applied as the TOPSIS method and the AHP-weighted TOPSIS method. e CCi values obtained in this condition are presented in Table 7, with their comparisons with previous values. Based on unweighted CCi values, the ranking of the UAV systems in order is UAV-1, UAV-5, UAV-2, UAV-6, UAV-4, and UAV-3. e results changed according to the unweighted ranking. e rankings of the considered alternatives as derived by employing these two methods are exhibited in Figure 3. Validation and Sensitivity Analysis of the Solution. TOPSIS and VIKOR (Vise Kriterijumska Optimizacija I Kompromisno Resenje) methods are distance-based similar methods. ere is a difference between the decision process. While the VIKOR method uses linear normalization, TOPSIS uses vector normalization procedure. However, two methods have simple mathematical calculations and have been widely used for the complex decision process. In this section, we use the VIKOR method to validate TOPSIS results. Besides, Spearman's rank correlation test is used for control of the statistical similarities among the results of ranking methods. See, for a more detailed description of the VIKOR method and its decision-making process, [69][70][71][72]. e decision problem is solved separately with the VIKOR method, and AHP-weighted VIKOR solution results are given comparatively in Table 8. e results of all methods are also shown in Figure 4 graphically. e scores of criteria obtained by the AHP method are also used for two rankings. e rating scores of the TOPSIS applications are then compared with the ones obtained using the VIKOR method results by using Spearman's rank correlation test and proposed in Table 9. As the outcome of Spearman's rank correlation test, the results are provided in the same table. In our study, the critical Z value is 1.645 which is selected at the level of significance of α � 0.05. Each Z value (TOPSIS, AHP-TOPSIS, VIKOR, and AHP-VIKOR) is higher than 1.645. It can be stated that the ranking provided by VIKOR applications is statistically similar to the other TOPSIS applications. In conclusion, the most suitable UAV is selected by using AHP-TOPSIS. VIKOR and its applications support each result. However, the results show that there are small differences in ranking of methods. Conclusion e UAV selection is very important in decision-making process in terms of the success of the defense area. e objective of this paper is also to present an integrated MCDM approach for determination of the best UAV. erefore, AHP and TOPSIS methods are used together. e UAVs are an increasingly important element of many modern militaries and various civilian areas. So, there is a need for selection process for the UAVs among various technologies. erefore, there are a lot of criteria affecting Importance of is Study. Importance of this study is to help developing countries in the defence area for their decisions to select among the UAV alternatives. UAVs are project-based applications and are strategical vehicles for national security. So, this study will be a good guide which helps developing countries. Recommendations for Future Studies. In future studies, the other MCDM methods such as analytic network process (ANP), ANP-TOPSIS, or fuzzy methods can be used, and the obtained results can be compared. e optimal solution of UAV selection can be done by mathematical models like goal programming or integer programming under resource constraints such as budget. It can be used as the zero-one goal programming model, by which the AHP/ANP priority weights can be combined with the objective function. Besides, one can focus on selection of specific criteria that are key performance drivers that can lead to informed selection of the UAV for successful decision-making. e criteria used in the proposed model can be improved with additional criteria. e proposed model can be also used in the other important decision processes such as drone selection and attack helicopter selection. e model can also be used with minor modifications in other decision-making processes in the defence area. In addition, mathematical models such as goal programming can be combined with this model. Besides, the weapon and the weapon system selection for UAVs can be done with MCDM. e qualitative criteria together with the quantitative criteria such as human factor, flying, and handling qualities can be added to the proposed model. At the same time, UAV selection process with MCDM can be used in traffic control and surveillance, infrastructure inspection, maintenance, security, precision agriculture, and also smart cities. Especially, the unmanned vehicle selection for traffic control and surveillance is a very important point for megacities. Data Availability e data used to support the findings of this study are available from the corresponding author upon request.	6,634.6	2020-06-17T00:00:00.000	[ "Engineering", "Computer Science" ]
Acute extracellular fluid volume changes increase ileocolonic resistance to saline flow in anesthetized dogs We determined the effect of acute extracellular fluid volume changes on saline flow through 4 gut segments (ileocolonic, ileal, ileocolonic sphincter and proximal colon), perfused at constant pressure in anesthetized dogs. Two different experimental protocols were used: hypervolemia (iv saline infusion, 0.9% NaCl, 20 ml/min, volume up to 5% body weight) and controlled hemorrhage (up to a 50% drop in mean arterial pressure). Mean ileocolonic flow (N = 6) was gradually and significantly decreased during the expansion (17.1%, P<0.05) and expanded (44.9%, P<0.05) periods while mean ileal flow (N = 7) was significantly decreased only during the expanded period (38%, P<0.05). Mean colonic flow (N = 7) was decreased during expansion (12%, P<0.05) but returned to control levels during the expanded period. Mean ileocolonic sphincter flow (N = 6) was not significantly modified. Mean ileocolonic flow (N = 10) was also decreased after hemorrhage (retracted period) by 17% (P<0.05), but saline flow was not modified in the other separate circuits (N = 6, 5 and 4 for ileal, ileocolonic sphincter and colonic groups, respectively). The expansion effect was blocked by atropine (0.5 mg/kg, iv) both on the ileocolonic (N = 6) and ileal (N = 5) circuits. Acute extracellular fluid volume retraction and expansion increased the lower gastrointestinal resistances to saline flow. These effects, which could physiologically decrease the liquid volume being supplied to the colon, are possible mechanisms activated to acutely balance liquid volume deficit and excess. Correspondence F.H. Rola Departamento de Fisiologia e Farmacologia, UFC Rua Coronel Nunes de Melo, 1127 Caixa Postal 3157 60430-270 Fortaleza, CE Brasil Fax: 55 (085) 243-9333 Presented at the IX Annual Meeting of the Federação de Sociedades de Biologia Experimental, Caxambu, MG, Brasil, August 24-27, 1994. This work is part of a Masters thesis presented by A.T. Santiago Jr. to the Departamento de Farmacologia, Universidade Federal do Ceará, Fortaleza, CE. Research supported by CNPq, CAPES and UFC. Received March 21, 1996 Accepted June 6, 1997 Introduction The luminal fluid in the gastrointestinal (GI) tract is in dynamic equilibrium with the extracellular fluid (ECF) volume (1).Acute ECF volume retraction by hemorrhage, dehydration or reduction in salt intake increases intestinal salt and water absorption (2).In contrast, acute ECF volume expansion reduces or even abolishes intestinal absorption while increasing secretion (3). Acute ECF volume changes also modify the motor pattern of the upper GI tract.The gastroduodenal resistance to liquid flow increases after ECF volume expansion and decreases by retraction (4,5).ECF volume expansion by intravenous infusion of saline also decreases both gastric and jejunal compliance while hemorrhage increases it (6,7).These observations suggest that the GI tract, at least in its upper portions, behaves like a regulatory liquid reservoir capable of adjusting its volumetric, absorptive and secretory capacities to cope with organic needs, i.e., accommodating and absorbing more liquid volume after hemorrhage and, contrarily, becoming less receptive to liquid volume distension and less absorptive after acute hypervolemia.This evidence is supported by the functional interaction between motor changes and absorption; increased tone reduces absorption while decreased tone stimulates salt and water absorption (8). However, despite the contributions of the ileal and large intestine to liquid volume regulation (9), the effect of ECF volume changes on the lower GI tract motor behavior has not been studied.Under basal conditions, i.e., when the large bowel absorbs about 20% of the total ingested fluids and gut secretion (10), patients with ileostomies are more susceptible to dehydration when placed on a low sodium diet or during an acute diarrheal illness (11).Therefore, we decided to study the effect of acute ECF volume expansion and retraction on the resistance to liquid flow offered by the ileocolonic segment as a whole, as well as by its segments in anesthetized dogs. Animals and surgical procedures Mongrel dogs of either sex (12.7 ± 3.2 kg, N = 62) fasted for 16 h with free acess to water were utilized.The animals were anesthetized with sodium pentobarbital (30 mg/ kg, iv) and the femoral vessels were cannulated.The right vein was used for anesthetic infusion when necessary or for isotonic saline (0.9% NaCl) infusion to expand the ECF volume.The right artery was utilized for bleeding and to provide blood samples for arterial hematocrit determinations.The left artery was cannulated with a heparinized (35 IU/ml) glass cannula and connected to an Hg manometer.Mean arterial pressure (MAP) and breathing were recorded with a Palmer kymograph.Central venous pressure (CVP) in cmH 2 O was measured visually using a catheter introduced via the external jugular vein, positioned in the right atrium and connected to an H 2 O manometer zeroed with the normal ictus cordis. Perfusion circuits After laparotomy, one of the following 4 different types of perfusion circuits was created (Figure 1). An oral probe (0.8 cm ID, 1.0 cm OD) introduced through an ileal fistula was positioned and fixed with an obstructive ligature in the distal ileum.An aboral probe (1.0 cm ID, 1.3 cm OD) introduced through a colonic fistula was positioned and fixed with another obstructive ligature in the proximal colon.The tips of both cannulas were positioned 4 cm from the ileocolonic sphincter.Care was taken to preserve regional circulation and innervation in all perfusion groups. The oral probe (0.8 cm ID, 1.0 cm OD) was introduced through an ileal fistula (15 cm from the ileocolonic sphincter), and the tip was positioned approximately 10 cm from the ileocolonic sphincter.The aboral probe (1.0 cm ID, 1.3 cm OD) introduced through a colonic fistula (2 cm from the ileocolonic sphincter) was positioned in the terminal ileum, near the ileocolonic sphincter. The oral probe (0.8 cm ID, 1.0 cm OD) was introduced through an ileal fistula (5 cm from the ileocolonic sphincter), and the tip was positioned and fixed near the ileocolonic sphincter.The aboral probe (1.0 cm ID, 1.3 cm OD) introduced through a colonic fistula (5 cm from the ileocolonic sphincter) was positioned in the terminal ileum, near the ileocolonic sphincter. The oral probe (0.8 cm ID, 1.0 cm OD) was introduced through a colonic fistula near the ileocolonic sphincter and the tip was positioned and fixed 2 cm from the ileocolonic sphincter.The aboral probe (1.0 cm ID, 1.3 cm OD) introduced through a colonic fistula (10 cm from the ileocolonic sphincter) was positioned in the proximal colon, 8 cm from the ileocolonic sphincter. The free end of each probe (oral and aboral) was connected to the barostat bottom which was then filled with isotonic saline solution (0.9% NaCl, w/v) and maintained at a constant liquid level throughout the experiment (Figure 1).The oral barostat liquid level, placed 16 cm above the xiphoid appendix, was always kept above the aboral barostat in order to ensure an aboral ileocolonic flow due to the resulting pressure gradient.The result was a communicating vessel system in which the perfused segment represented the only resistance offered to the liquid flow which could vary.The liquid flowing out of the aboral barostat was manually collected at 2-min intervals and measured and the rate is reported as the ileocolonic, ileal, ileocolonic sphincter and colonic flows in ml/min.To obtain an ileoco- Ileocolonic perfusion Flow (ml/min) lonic flow of about 50 ml/min and 70 ml/min in the other separate segments (ileal, ileocolonic sphincter and colonic segments) a variation in the pressure gradient was performed: 4.7 ± 0.3, 3.4 ± 1.3, 2.5 ± 0.6 and 1.6 ± 0.5 cmH 2 O, respectively.The isotonic saline utilized in the perfusing system was heated to 37 o C before reaching the gut by circulating it in a glass coil immersed in a water bath at 38 o C. Experimental design After surgery, the animals were allowed to rest for a period of 20-30 min to stabilize hemodynamic and respiratory parameters.After this interval, the mean rate of saline flowing from the aboral barostat over a period of 20 min represented the normovolemic control flow.The animals were submitted to acute ECF volume expansion by iv isotonic saline infusion (0.9% NaCl solution, 20 ml/min) up to a volume equivalent to 5% of the body weight or acute ECF volume retraction by controlled hemorrhage until a 50% drop in MAP levels, followed by reinfusion of the removed blood at a rate of 20 ml/min.The perfusion flow was measured during saline infusion (expansion period), 30 min after it was completed (expanded period), during hemorrhage and 20 min thereafter (ECF volume retraction and retracted, respectively), as well as during blood volume reinfusion (ECF volume restoration). At the end of each experiment, the animals were sacrificed under deep anesthesia by iv KCl injection. Drugs In another set of experiments, atropine sulfate (Sigma Chemical Co., St. Louis, MO) was injected iv (0.5 mg/kg) after normovolemic control into ileal (N = 5) and ileocolonic (N = 6) perfusion circuits.Saline flow measurements were then recorded for 30 min (drug control) and ECF volume expansion was performed according to the previous protocol. Statistical analysis The results are reported as mean ± SEM.One-way repeated measures ANOVA and Dunnett's test were used to compare the differences in saline flow between each experimental period and the Student t-test for paired means was used to compare percent differences in hematocrit.Statistical differences were considered to be significant at P<0.05. Results Effect of acute ECF volume expansion on ileocolonic, ileal, colonic and ileocolonic sphincter segment perfusion flow, MAP, CVP and hematocrit Figure 2 shows the effect of acute expansion on the ileocolonic (2A), ileal (2B), ileocolonic sphincter (2C) and colonic (2D) segment flow, MAP and CVP.As can be seen in Figure 2A, the mean ileocolonic flow (N = 6) was gradually decreased during the expansion period until the difference was statistically significant when a volume up to 5% body weight was infused (P<0.05),persisting during the first 20 min of the expanded period (P<0.05).The mean percent flow reductions during the expansion and expanded periods were 23 and 45%, respectively (from 46.7 ± 2.8 to 38.7 ± 4.4 and 25.7 ± 6.3 ml/ min, respectively, P<0.05).MAP was not significantly modified while CVP levels increased at the beginning of expansion (1% body weight), remaining stable at least during the first 20 min of the expanded period (P<0.05). Figure 2B shows that the mean ileal flow (N = 7) slowly decreased during the expansion period, reaching a statistically significant difference only during the expanded The mean ileocolonic sphincter flow (N = 6) was not significantly modified either during the expansion or during the expanded period (from 81 ± 2.7 to 76.4 ± 6.4 and 73.8 ± 5.2 ml/min, Figure 2C).MAP was also not significantly modified.Even when we analyzed the changes in flow separately on the basis of the percent body weight of volume infused (as in the ileocolonic and ileal flows), no significant flow reduction was observed.However, CVP levels were significantly increased throughout the expansion and expanded periods (P<0.05). Figure 2D shows that the mean colonic flow (N = 7) decreased during the expansion and expanded periods, but attained statistical significance only during expansion and not during the expanded period.The mean percent flow decrease during the expansion period was 12% (from 67.7 ± 3.4 to 59.8 ± 3.5 and 66.3 ± 3.9 ml/min, P<0.05).MAP and CVP were significantly increased throughout the expansion and expanded periods (P<0.05). Table 1 shows that the arterial hematocrit values were significantly decreased during the expanded period when compared to control values (P<0.001). Effect of atropine infusion on acute ECF volume expansion, ileocolonic and ileal flows, MAP and CVP Figure 3 shows that atropine injection blocked the effect of ECF volume expansion on the ileocolonic (3A) and ileal saline flows (3B).Atropine per se did not modify saline flow during the atropine control period (from 68.9 ± 5.2 to 71.4 ± 4.9 and from 64.2 ± 3.3 to 64.6 ± 2.7 ml/min) and no additional saline flow decrease was observed during the expansion or expanded periods (to 64.6 ± 7.5 and 70.5 ± 4.8 ml/min and to 64.3 ± 3.2 and 61.8 ± 2.9 ml/min, for ileocolonic and ileal perfusions, respectively).MAP and CVP levels were significantly increased during the expansion and expanded periods (P<0.05) and MAP levels were also increased during the atropine control period in the ileal perfusion group (P<0.05). Ileocolonic flow was significantly decreased by 17% after ECF volume retraction during the retracted period (from 52.8 ± 3.9 to 45.5 ± 5 ml/min, P<0.05), but returned to control levels during removed blood reinfusion (Figure 4A). Figure 4B, 4C and 4D show that saline flow was not modified during or after retraction (retracted period) or after blood volume reinfusion in the other perfusion groups.MAP levels decreased during and after retraction and during removed blood volume reinfusion (P<0.05) in the ileocolonic, ileocolonic sphincter and colonic perfusion groups.In ileal perfusion, however, MAP returned to control levels after reinfusion of removed blood.CVP levels were decreased during and after retraction in the ileocolonic and colonic groups (P<0.05) and only during retraction in the ileocolonic sphincter group (P<0.05) but were not modified in the ileal group.Table 1 shows that acute ECF volume retraction did not modify the arterial hematocrit values. Discussion We have previously demonstrated that acute volume expansion increases gas-troduodenal resistance to saline flow in rats and dogs, while hypovolemia due to hemorrhage decreases it (4,5).Since ECF volume retraction increases intestinal salt and water absorption (2) while expansion reduces or even abolishes intestinal absorption and increases secretion (3), we proposed a role for ECF volume in the modulation of liquid flow through the upper GI tract, i.e., coupled modulation of the GI tract contractile activity, salt and water transport and ultimately luminal liquid transit to cope with organic needs. The present study, using the same experimental protocol (ileocolonic segment and separate ileal, ileocolonic sphincter and colonic portions were perfused under constant pressure and changes in flow were assumed to represent modifications in motor activity, tone and/or motility), extends this idea to the lower gut, showing that acute volume imbalances also modify the motor behavior of ileocolonic segments.This approach has been frequently utilized to study resistances through discrete portions of the gut (4,5,12). In constrast to the opposite flow changes (reduction and increase) which we observed after opposite volume imbalances (expansion and retraction) in the gastroduodenal segment (4), both acute ECF volume expansion and retraction were associated with decreased ileocolonic saline flow. Similar to the gastroduodenal segment, ECF volume expansion gradually and persistently decreased the saline flow through the ileocolonic segment during the expan-sion and expanded periods.However, when searching for the main sites of resistance, we observed some peculiarities.In the separate ileal perfusion, we observed a late saline flow decrease, which was significant only during the expanded period.In contrast, saline flow in the separate ileocolonic sphincter was quite stable both during the expansion and expanded peri- ods, while colonic flow reduction was slight and more pronounced only during the expansion period.Thus, analyzing Figure 2, the flow decrease observed in Figure 2A appears to be the final result of the sum contribution of Figure 2B and 2D.These findings point to the ileum as the major contributor to the marked flow reduction observed during the expanded period while colonic segment contribution appears to be especially important for the earlier flow decrease during the expansion period, observed throughout ileocolonic segment perfusion. In contrast to the increased gastroduodenal flow after ECF retraction that we have previously observed ( 5), ECF volume retraction, like expansion, was followed by a smaller but still significant drop in ileocolonic flow.However, when we attempted to identify the main sites of resistance we were unable to find them since no saline flow decrease either during or after ECF volume retraction was observed in the other separate perfusion circuits (ileal, ileocolonic sphincter and colonic circuits).These findings suggest that the integrated sum of the ileal, ileocolonic sphincter and colonic resistances appears to be necessary for the complete expression of the ECF volume retraction effect on the ileocolonic segment. In addition to reducing saline flow, ECF volume expansion induced hemodilution and increased CVP and MAP (except in ileocolonic perfusion).ECF volume retraction was also followed by major circulatory changes, with MAP and CVP being significantly reduced.Hematocrit values were decreased after ECF volume expansion but remained unchanged immediately after hemorrhage.These findings suggest that modifications in CVP levels are closely related to saline flow changes, but MAP was not modified in the ileocolonic perfusion group, where a significant saline flow reduction was observed.In addition, an opposite effect was observed, i.e., ileocolonic flow was reduced while CVP and MAP were increased simultaneously. ECF volume expansion with isotonic, isotonic-isooncotic or isotonic-isooncoticisohemic infusion produced similar reductions in gastroduodenal flow (4).These results indicate that the effect of hypervolemia on gut motility does not appear to be related to factors other than acute volemic expansion. The cholinergic role in the decrease in saline ileocolonic and ileal flows due to acute ECF volume expansion was further investigated.For this purpose, the effect of atropine (which is a competitive muscarinic antagonist in smooth muscle) was studied.Atropine per se did not modify ileal or ileocolonic flows.In addition, no saline flow change was observed during or after ECF volume expansion in animals pretreated with atropine.Thus, the effect of acute ECF volume expansion was blocked by atropine.Since atropine decreases intestinal contractile activity (13), the saline flow decrease due to acute ECF volume expansion in the lower GI tract may involve an intestinal contractile activity increase by cholinergic activation, which was antagonized by atropine. In summary, in contrast to our previous results in the upper GI tract portions (4-7), we observed that both opposite volemic changes (expansion and retraction) increased the ileocolonic resistance to saline flow.The physiological relevance of these findings is speculative and yet to be established, especially because of the small number of supporting reports in this field.However, the decrease of liquid volume being supplied to the colon could indicate that the small bowel is more apt to promptly deal with acute volemic challenges.This idea is reinforced by the results of other studies, which demonstrate increased small bowel fluid and electrolyte absorption during retraction (2), decreased absorption/increased secretion during expansion (3) and apparent large bowel incapacity to readily and efficiently modify those patterns under the same volemic conditions (14).Consequently, the persistence of basal ileocolonic transit could contribute to further aggravate liquid volume excess or deficit. Figure 2 - Figure 2 -Effect of acute extracellular fluid (ECF) volume expansion on the ileocolonic (A), ileal (B), ileocolonic sphincter (C) and colonic (D) flows of saline (in ml/min), central venous pressure (CVP) and mean arterial pressure (MAP) of anesthetized dogs.After a normovolemic control period (20 min), the animals were expanded by iv isotonic saline infusion (20 ml/min) up to 5% body weight.The saline flow was monitored during expansion and 30 min thereafter.Bars represent the mean values of saline flow during each period.The expansion period was increased stepwise at 1% intervals from 1 to 5% body weight and is reported for each step.The expanded period was divided into three 10-min intervals.Cont, Normovolemic control period.Upper dots indicate MAP (mmHg) and lower dots CVP (cmH 2 O).Vertical lines indicate standard error of the mean.P<0.05 compared to control period (Dunnetts test). .05), and remaining stable for at least 20 min.The mean percent flow reductions in the expansion and expanded periods were 19.3 and 38%, respectively (from 70 ± 4.4 to 56.5 ± 8.5 and 43.4 ± 11.7 ml/min, respectively, P<0.05).CVP levels increased significantly during expansion when a volume up to 2% body weight was infused, remaining increased until the end of expansion (P<0.05).MAP levels increased significantly throughout the expansion period and during the first 10 min of the expanded period (P<0.05). Figure 3 - Figure 3 -Effect of atropine on acute extracellular fluid (ECF) volume expansion, ileocolonic (A) and ileal (B) saline flow (ml/min), central venous pressure (CVP) and mean arterial pressure (MAP) of anesthetized dogs.After normovolemic control period (Cont, 20 min), the animals received an iv injection of atropine (0.5 mg/kg).The perfusion saline flow was measured for 30 min (Atrop) after the animals were expanded by iv isotonic saline infusion (20 ml/min) up to 5% body weight.The perfusion flow was monitored during expansion (Exp) and for 30 min thereafter (Expd).Bars represent the mean perfusion flow values.Continuous lines represent MAP (mmHg) and broken lines represent CVP (cmH 2O).Vertical lines indicate standard error of the mean.P<0.05 compared to control period (Dunnetts test). Figure 4 - Figure4-Effect of acute extracellular fluid (ECF) volume retraction on the ileocolonic (A), ileal (B), ileocolonic sphincter (C) and colonic (D) flows of saline, central venous pressure (CVP) and mean arterial pressure (MAP) of anesthetized dogs.After a normovolemic control period (20 min), the ECF volume was retracted by bleeding the animals up to a 50% drop in MAP.The animals were then brought back to the euvolemic condition by reinfusing the blood volume previously removed at a rate of 20 ml/min.Besides the normovolemic control period, ileocolonic, ileal, ileocolonic sphincter and colonic saline flows (ml/min) were measured during bleeding and for a 20-min period thereafter (ECF volume retraction and retracted, respectively) as well as during autologous blood reinfusion (restoration).Bars represent average of flow values during the normovolemic control (Cont), retraction (Ret), ECF volume retracted (Retd) and restoration (Rest) periods.Continuous and interrupted lines represent MAP (mmHg) and CVP (cmH 2 O), respectively.Vertical lines indicate the standard error of the mean.P<0.05 compared to control period (Dunnetts test). Table 1 - Effect of ECF volume expansion and retraction on arterial hematocrit.Expansion was obtained by isotonic saline infusion of 1% to 5% stepwise at 20 ml/ min.Retraction was obtained by bleeding until there was a 50% reduction in arterial pressure.Data are reported as means ± SEM for 51 dogs.Number of animals in each group reported in parenthesis.*P<0.01 compared to control values (Student t-test).	5,024.6	1997-08-01T00:00:00.000	[ "Medicine", "Biology" ]
Partial passive ownership holdings and R&D risk choices in a differentiated duopoly Abstract This study investigates the R&D risk choices in a differentiated duopoly, in which a firm has partial passive ownership holdings (P.P.O.s) in its rival. Firms perform R&D projects with identical expected outcomes but different risk degrees. It mainly finds that: (1) the P.P.O.s make both firms more willing to take R&D risks; (2) compared with the firm which owns a share of its rival, its partially owned rival is more willing to take R&D risks; (3) for both firms, their private incentive for R&D risk is lower than the social incentive. However, the P.P.O.s may make the private optimum closer to the social optimum. Introduction Partial passive ownership holdings (P.P.O.s), also known as partial passive shareholding, refer to the phenomenon that a firm holds non-controlling minority shares of other firms (Dong & Chang, 2020;Leonardos et al., 2021).This phenomenon of P.P.O.s began in Japan in the 1950s.Later, there were many cases of firms owning shares in competitors in western developed countries, such as Microsoft's acquisition of a 7% stake in its rival, Apple, in 1997.At present, the P.P.O.s are also becoming more and more extensive in many developing countries (e.g., China).R&D is crucial for manufacturing firms.It is the inexhaustible driving force for the long-term and sustainable development of firms.The P.P.O.s make the interests of firms related, which may affect their R&D incentives.Since a firm that invests resources in developing a new technology (or product) does not determine whether it will succeed, the results of R&D activities are generally uncertain (Silipo & Weiss, 2005).In facing such uncertainty, the optimal choice of risk associated with R&D projects is a variable of interest (Xing, 2019a). Based on the above background, we intend to answer the following questions in this study: (1) How do the P.P.O.s affect the optimal R&D risk choices in a duopoly?; (2) Which firm is more willing to take R&D risks, the firm that owns the rival's shares or its partially owned rival?; (3) Whether the firms can achieve the socially optimal R&D risk level?If not, how do the P.P.O.s affect the extent of inefficiency? To investigate the above questions, we develop a differentiated duopoly model to explore the optimal R&D risk choices when a firm holds the minority of shares of its rival.The model has two stages.In the first stage, firms determine the type of R&D project from a series of projects with the same expected outcome but different risks.In this stage, they are uncertain about the outcome of R&D, but they know the probability distribution it obeys.In the second stage, firms have already known the R&D outcome and they compete in quantity (i.e., conduct Cournot competition).The main findings are as follows.First, the P.P.O.s have a positive effect on the R&D risk-taking of both firms. 1 Second, the firm that does not own the rival's shares takes higher R&D risks than the one that owns its shares.Third, the private optimum of R&D risk is too low from the perspective of social welfare, while the P.P.O.s may reduce this inefficiency. The main contributions of this study are as follows.First, to the best of our knowledge, this study is the first to investigate the effect of P.P.O.s on strategic R&D risk choices.Second, the comparison of R&D incentives of firms in an asymmetric situation has been a topic of interest in previous literature.We compare the optimal R&D risks of the firm which owns a share of its rival and its partially owned rival.This study significantly supplements the relevant research.Third, we compare private and social incentives for R&D risk and analyse the possible influence of P.P.O.s on the difference between them. This article also relates to the literature that investigates the effects of P.P.O.s on the R&D incentives.Shelegia and Spiegel (2015), L opez and Vives (2019), Vives (2020), Anton et al. (2021), Bayona and L opez (2018) and Brito et al. (2020) consider how the P.P.O.s affect cost-reducing R&D (quality-enhancing R&D) and give the conditions under which they have positive effects on this type of R&D investments.In addition, Liu (2019) examines the impact of PPOs on R&D aiming at improving product differentiation in a mixed duopoly and thinks that they enhance the possibility of the private firm to make such R&D investment.However, these studies do not involve the R&D risk and thus do not consider how the PPOs affect the R&D risk choices. The rest of this study is organised as follows.Section 2 describes the basic model, and section 3 (section 4) analyses the private (social) optimum of R&D risk.Section 5 extends the basic model and the final section concludes. The basic model Consider an industry with two firms producing differentiated goods.Each firm provides only one type of product and firm i provides product i.Following the studies of Singh and Vives (1984), Qiu (1997) and Lin and Saggi (2002), we give the following inverse demand function: In (1), p i (q i ) is the price (quantity) of product i, a>0 and 0<r<1: The larger value of parameter r implies the higher (lower) degree of substitutability (differentiation) between two products. Firms undertake process R&D investment to lower their marginal production cost.After R&D, the marginal production cost of firm i is: c i ¼ cÀx i , where c is the initial marginal production cost and x i is the R&D effort of firm i. 2 It follows that the firm i's production cost after R&D is: The R&D outcomes are uncertain when firms conduct process innovation (Xing, 2014).We assume that x i obeys the probability distribution: The covariance of x i and x j is assumed to be zero (i.e., Covðx i , x j Þ ¼ 0, i 6 ¼ j).That is, they are independent. 3 The fixed costs of each firm are only generated by its R&D investment.Therefore, the profit function of firm i can be given by: In (2), Iðl i , r i Þ is the R&D cost of firm i. 4 In order to ensure the existence of interior solutions in the R&D stage, Iðl i , r i Þ is assumed to satisfy: (1) it is strictly convex about r i 5 ; and (2) it is enough large if r i tends to infinity.Following Tishler (2008), Xing (2017), Lee and Cho (2020) and Xing et al. (2021), we use the variance of R&D outcome to represent the risk of R&D project. Firm 1 has P.P.O.holdings in firm 2, owning a share k (0<k<0:5) of the rival's profits (Fanti, 2015;Leonardos et al., 2021;Papadopoulos et al., 2019).Although firm 1 owns a share of firm 2, it does not have decision-making power over firm 2 (B arcena-Ruiz & Sagasta, 2021).We assume that firm 1 maximises its total profit.Its objective function is: In addition, firm 2 only obtains a share 1Àk of its profit.It follows that the objective function of firm 2 is: Social welfare comprises the producer surplus (p 1 þ p 2 ) and consumer surplus (CS), which is given by: We consider a two-stage game.The timing of it is as follows.In the first stage, each firm engages in R&D and determines the R&D risk levels.In this stage, it chooses the type of its R&D project from a series of projects with identical expected outcome but different risks.A firm choosing the type of R&D project is equivalent to choosing the R&D risk level in this study.Both firms are uncertain about the outcome of R&D (i.e., x i ), despite they are aware of the probability distribution it obeys.Therefore, the expected value and variance of R&D outcome (i.e., l i and r i ) are common knowledge for firms 1 and 2 (Xing et al., 2021).Following Tishler (2008), Xing (2017) and Zhang (2020), we assume that firms are risk-neutral.In the second stage, firms compete in quantity (i.e., conduct Cournot competition).In this stage, each firm knows the R&D outcomes of itself and its competitor. The private optimum In the second stage, given the R&D outcomes of firms (i.e., x 1 and x 2 ), each firm chooses the production level (q i ) that maximises its objective function (see ( 3) and ( 4)).The first-order conditions (F.O.C.s) for the optimal outputs are given as follows: According to F.O.C.s, we obtain the following best-reply function: Obviously, as the value of k increases, firm 1 is less aggressive in the product market (Papadopoulos et al., 2019).We solve the system of F.O.C.s (see ( 6) and ( 7)) and give the following equilibrium outputs: The second-order conditions (S.O.C.s) are satisfied because of o 2 U 1 =oq 2 1 ¼ À2<0 and o 2 U 2 =oq 2 2 ¼ À2ð1ÀkÞ<0: Submitting (8) and ( 9) into (3) and ( 4) respectively, we obtain the profit functions on x 1 and x 2 : Now we turn to the first stage of the game.Firms choose their project type from a series of R&D projects.The expected outcomes of these projects are the same, but their risks are different (Tishler, 2008;Xing, 2014;Zhang et al., 2013).In this situation, a firm choosing the project type is equivalent to choosing the risk (i.e., variance) of R&D project.Using (10) and (11), we derive the following expected profit of firm i: According to ( 12) and ( 13), we derive the expected gross profit of firm i ( By analysing the impact of R&D risks on them, we get the following Lemma. Lemma 1 states that the R&D risk of a firm has positive effect on its (or its rival's) expected gross profit and the expected industrial gross profit.When investigating how corporate social responsibility (emission tax) affects the R&D (environmental R&D) risk choices, Lee and Cho (2020) and Xing et al. (2021) obtain similar results.However, they do not consider the cross ownership between firms. The reason for Lemma 1 is as follows.If the marginal production cost changes from symmetry (ex ante) to asymmetry (ex post), the existence of product substitution will transfer production from the firm with high marginal cost (ex post) to the one with low marginal cost (ex post) (Kitahara & Matsumura, 2006;Matsumura, 2003).This substitution between products can save the total production costs of the industry.It follows that the higher marginal production cost asymmetry implies the more industrial gross profits.That is, there exists the industry-profit-increasing production substitution effect.A firm that chooses a R&D project with higher risk (given the other firm's choice) is more conducive to the marginal production cost asymmetry (Xing et al., 2021).Thus, each firm's R&D risk has positive effect on the expected industrial gross profit.In addition, if a firm chooses a R&D project with higher risk, it will expect to get more from the total profits of the industry.The competitor of this firm also expects to get more due to it can take advantage of differentiation incentives if the marginal production cost asymmetry is highly volatile (d 'Aspremont et al., 1979;Hotelling, 1929).Thus, a firm's R&D risk has positive effect on its (or its rival's) expected gross profits. In the R&D stage, both firms are uncertain about the R&D outcomes of their project.They choose the R&D risk (r i ) to maximise the expected value of their objective function.Thus, the F.O.C.s can be given by: Due to l 1 and l 2 are the same constant, only r i is variable in oIðl i , r i Þ=or i (i ¼ 1, 2).For clarity of expression, we set: Combining with ( 14) and ( 15), the equilibrium R&D risks (i.e., r C 1 and r C 2 ) satisfy the following equations: Now we examine the impact of P.P.O.s on equilibrium R&D risks.Using ( 17) and ( 18), we can prove the following proposition. Proposition 1. (i) r C 1 increases as k increases, and (ii) r C 2 increases as k increases. Proof.See Appendix B. Proposition 1 implies that, the P.P.O.s lead both the firm that owns the rival's share and its partially owned rival to be more willing to take R&D risks.Further, when we give the specific form of the R&D cost function, then we can compare the effects of PPOs on r C 1 and r C In this situation, the positive effect of P.P.O.s on the optimal R&D risk of firm 2 is stronger than that of firm 1.The intuition of Proposition 1 is as follows.First, we consider the part (i).The P.P.O.s can achieve a certain degree of collusion (Azar et al., 2018;Brito et al., 2019;Leonardos et al., 2021), which strengthens the industry-profit-increasing production substitution effect.Combining with Lemma 1, a higher value of k implies higher positive effect of the firm 1's R&D risk on the expected industrial gross profits.In addition, the increase of k leads firm 1 to get more from the industrial gross profits.Thus, firm 1 takes more R&D risks if k increases.Second, we consider the part (ii).With a higher value of k, on the one hand, it leads to higher positive effect of the firm 2's R&D risk on its expected gross profits because of weaker market competition.On the other hand, it leads firm 2 to get less from its expected gross profits.The former effect dominates the latter one.Thus, firm 2 takes more R&D risks if k increases. Next we turn to analyse which firm (firm 1 or firm 2) is more willing to take R&D risks when it chooses the R&D project.According to ( 17) and ( 18), we derive the following result. Proposition 2 shows that firm 2 chooses riskier R&D project than firm 1 in equilibrium.This result implies that, compared with the firm that owes the rival's shares, its partially owned rival is more willing to take R&D risks.The intuition is as following.In contrast with firm 2, firm 1 has some shares in its rival and thus can internalise the market competition to a certain extent (Fanti, 2015).It follows that a higher k tends to reduce the firm 1's incentive to produce but improve the firm 2's incentive to produce (B arcena-Ruiz & Sagasta, 2021).The above effect is considered by each firm when it chooses the type of its R&D project (i.e., determining the R&D risk level).In order to reduce the marginal production cost to a greater extent, firm 2 is more aggressive than its rival.Thus, firm 2 is more willing to take R&D risks than firm 1. The social optimum This section considers the (second-best) socially optimal R&D risk that maximises the expected social welfare first and then takes a comparison of the private and social incentives for R&D risk.To derive the expected social welfare, we need to give the expected consumer surplus first.Combining (8, 9) and CS ¼ ðq 2 1 þ 2rq 1 q 2 þ q 2 2 Þ=2 gives the following expected consumer surplus: Lemma 2 indicates that each firm's R&D risk has positive effect on the expected consumer surplus.The reason for this lemma is as follows.If the marginal production cost changes from symmetry (ex ante) to asymmetry (ex post), the firm with low (high) marginal cost (ex post) increases (reduces) the output through strategic interaction between two firms (i.e., strategic substitutes) (Kitahara & Matsumura, 2006;Matsumura, 2003).Compared with the situation of cost symmetry (ex ante), cost asymmetry (ex post) makes the firm with low marginal cost more willing to reduce prices.It follows that the higher marginal cost asymmetry implies the more consumer surplus.That is, there exists consumer-surplus-improving production substitution effect.A firm choosing a project with higher R&D risk (given the other firm's choice) is more conducive to the marginal production cost asymmetry (Xing et al., 2021).Thus, each firm's R&D risk has positive effect on the expected consumer surplus. Submitting (12, 13) and ( 19) into (5) yields the following expected social welfare: According to (20), we derive the following F.O.C.s: Combining the above F.O.C.s and ( 16), the socially optimal R&D risk of firm i (i.e., r SC i ) satisfies the following equations: Using ( 23) and ( 24), we obtain 4r 2 ð2ÀkÞ g 0 ðr SC 2 Þ½4Àð1þkÞr 2 3 : It follows that or SC i ok >0 because of g 0 ðr SC i Þ>0, 4À2kr 2 þ r 2 >0, 2Àk>0 and 4Àð1 þ kÞr 2 >0: Thus, r SC i increases as k increases.The reason for this result is that the P.P.O.s can achieve collusion to some extent, which enhances the industry-profitincreasing production substitution effect.However, the P.P.O.s may weaken the consumer-surplus-improving production substitution effect.The former effect of k dominates its later effect.With an increase of k, the social gain from R&D risk is higher.Thus, the socially optimal R&D risk level also increases. Proof.See Appendix E. Proposition 3 shows that the equilibrium R&D risk for each firm is too low from the viewpoint of social welfare.In other words, each firm has an insufficient incentive for taking R&D risks.Further, if we give the following form of the R&D cost function , we can prove that oðr SC 1 Àr C 1 Þ=ok<0 and oðr SC 2 Àr C 2 Þ=ok<0: In this situation, comparing with the situation without P.P.O.s, the P.P.O.s make the private optimum of each firm closer to the social optimum (i.e., the P.P.O.s lead the level of inefficiency to decrease). The reason for Proposition 3 is as follows.The private optimum of R&D risk is different from the social optimum because the social planner and firms pursue inconsistent goals when determining the R&D risk level.For firm 1, it only cares for its own expected profit and partially cares for its rival's expected profit.However, it does not consider the positive effect of its R&D risk on the expected consumer surplus (see Lemma 2).In addition, for firm 2, it only cares for its own expected profit in a certain proportion.However, it does not consider the positive effect of its R&D risk on the rival's expected profit and the expected consumer surplus (see Lemmas 1 and 2).The social planner not only considers a given firm's expected profit, but also considers that of its rival and the expected consumer surplus.Thus, from the perspective of social welfare, firm 1 (or firm 2) has an insufficient incentive for taking R&D risks. Extensions This section considers several extensions of the basic model and checks the robustness of the main findings in Sections 3 and 4. Allowing firms to hold each other's shares In the basic model, only one firm owns some shares of its rival (i.e., unilateral cross ownership).However, two firms may hold each other's shares (i.e., bilateral cross ownership) in reality.We now consider the situation that each firm is allowed to have a minority of shares in its rival.In this section, we assume that firm i acquires equities on a share k j (0<k j <0:5) of firm j's profits (i 6 ¼ j, i, j ¼ 1, 2).Thus, the objective function of firm i is given by: U i ¼ ð1Àk j Þp i þ k i p j (i 6 ¼ j, i, j ¼ 1, 2) (Bayona & L opez, 2018).We assume that k 1 6 ¼ k 2 : That is, we extend into the asymmetric P.P.O.s where each firm owns different shares of its rival.The analysis in the main context is one of examples where a firm has partial P.P.O.s in its rival while the rival does not have.Other assumptions are the same as those in the basic model.Similar to the derivation of the basic model, we obtain the private optimum (i.e., r Ã I i ) and social optimum (i.e., r s I i ) and then prove the following results. This implies that when two competing firms own different shares of its rival, the results of Propositions 1-3 in the benchmark case still hold. Allowing firms to compete in price in market stage In the second stage of the basic model, two firms compete in quantity (i.e., conduct Cournot competition), whose choices are strategic substitutes.However, the firms' choices are strategic complements in price competition (i.e., conduct Bertrand competition).Are the main results of the basic model (Propositions 1$3) robust under price competition?To answer this question, we need to consider the situation that firms compete in price in the market stage.The only difference between this section and the basic model is the form of market competition.We assume that firms decide on price in the second stage in this section.According to (1), we give the following demand function: q i ¼ 1 1Àr 2 ½ð1ÀrÞaÀp i þ rp j (i 6 ¼ j, i, j ¼ 1, 2).Similar as in sections 3 and 4, we derive the private optimum (i.e., r B i ) and social optimum (i.e., r SB i ) and then prove the following proposition. This implies that the main results of Propositions 1-3 in the benchmark case are robust to the mode of competition (Cournot or Bertrand). Proof.See Appendix H. In the literature on the comparisons of R&D investments between Cournot and Bertrand, it is already well-known that Cournot firms invest more in R&D than Bertrand firms (Chen & Lee, 2022;Hinloopen & Vandekerckhove, 2009;Qiu,1997). 6 However, this phenomenon can be reversed in this study where the R&D outcome (i.e., x i ) is a parameter that has a probability distribution with the same expected mean while the R&D risk is a choice variable (see Proposition 6).That is, given the same expectation of R&D outcome between Cournot and Bertrand, the strategic effect of R&D on the profit disappears, and only the level of R&D risk matters in determining the firm's choices. The reason for Proposition 6 is as follows.The higher marginal production cost asymmetry implies more industrial gross profits (see the explanation of lemma 1).Due to the fact that the outputs are strategic substitutes (the prices are strategic complements) under Cournot (Bertrand), the cost asymmetry between firms increases industrial gross profits more significantly under Bertrand (Xing et al., 2021).It follows that Bertrand firm's R&D risk has a higher positive effect on the expected industrial gross profit.Similar reason as in Lemma 1, for a given r and k, if a firm chooses a R&D project with higher risk, it will expect to get more from the total profits of the industry.Thus, Bertrand firm's R&D risk has a higher positive effect on its expected gross profits.This leads Bertrand firm to choose higher R&D risk behaviour, compared to Cournot firm.Lee and Cho (2020) and Xing et al. (2021) analyse the R&D risk choices under different competition modes by considering corporate social responsibility and emission tax respectively.They think that the R&D risk level in the Bertrand competition case is higher than in the Cournot competition case.Proposition 6 states that their result still holds even in the situation of P.P.O.s. Allowing R&D spillovers There might exist research spillovers between the firm's R&D investments.This section tests the robustness of the main propositions when considering R&D spillovers. We consider the following marginal production cost of firm i is: c i ¼ cÀx i Àbx j (i 6 ¼ j, i, j ¼ 1, 2), where c is the initial marginal production cost, x i (x j ) is the R&D effort of firm i (firm j), and b (0 b 1) is the R&D spillover parameter (D 'Aspremont & Jacquemin, 1988).In the basic model, b ¼ 0: Other assumptions are the same as those in the basic model.Similar to the derivation of the basic model, we obtain the private optimum (i.e., r Ã II i ) and social optimum (i.e., r s II i ).We set b In addition, b also lies in ð0, 1Þ (see Figure 1).Then, we can prove the following results. It is worth noting that: ).The reason is as follows.Similar reason as Lemma 1, the R&D risk of firm 1 has a positive effect on its (or its rival's) expected gross profit (i.e., or 1 >0) when considering the R&D spillovers.Further, we find that when b is moderate, the P.P.O.s have a negative (or weak positive) effect on or 1 , and have a negative effect on oEðP 2 Þ or 1 : It follows that the P.P.O.s have a negative effect on : With an increase of k, firm 1 pays more attention to industry profits when choosing R&D projects, and thus it may choose lower risk level of R&D project if b is moderate; and (ii) r Ã II 1 may be higher than r Ã II 2 if r is small and b is large.The reason is as follows.When r is small, the degree of product differentiation is large and the competition in the market stage is weak.Firm 1 partially considers the expected profit of firm 2 when selecting R&D projects.In the situation of weak market competition, once its R&D is successful, it will also bring great benefits to firm 2 if b is large, which in turn will increase firm 1's total profit.The above effect does not exist for firm 2. Thus, firm 1 may take higher risks than firm 2 if r is small and b is large. The above results remind us that, if the research spillovers are significant, we should be cautious when analysing the impact of P.P.O.s on the R&D risk choices of the firm who owns a share of its rival, or comparing the R&D risk levels of the firm who owns a share of its rival and its partially owned rival. Allowing products to be complementary In the basic model, we assume that the products of different firms have a certain degree of substitutability.However, there also exist the P.P.O.s in some industrial chains.In this situation, the products of upstream and downstream firms are complementary.This section allows products to have a certain degree of complementarity, which is also measured by parameter r: However, unlike the basic model r is assumed to satisfy À1 r<0 in this section.Other assumptions are the same as those in the basic model.We can prove that the privately optimal R&D risk levels of firms 1 and 2 satisfy ( 17) and ( 18) respectively, and their socially optimal R&D risk satisfy ( 23) and ( 24) respectively.Obviously, we can find similar results as in the basic model. 7 That is, the results of Propositions 1$3 are robust in the situation that the products provided by firms have a certain complementarity. Allowing R&D risks correlation In the basic model, we assume that the covariance of x i and x j is zero (i.e., Covðx i , x j Þ ¼ 0, i 6 ¼ j, i, j ¼ 1, 2).This assumption implies that x i and x j are independent.However, x i and x j may be dependent under certain conditions.In this situation, there may be Covðx i , x j Þ 6 ¼ 0 (i 6 ¼ j, i, j ¼ 1, 2).Even in the R&D competitive environment, the P.P.O.s may lead to a certain positive correlation between the R&D risks of the firms. 8In this section, we assume that Covðx i , , where h is a constant.Note that h is not very large to ensure that the expected profits of each firm are positive.Other assumptions are the same as those in the basic model.We can prove that the privately (socially) optimal R&D risk levels of firms 1 and 2 satisfy (17) and (18, 23) and ( 24)), respectively.Obviously, Propositions 1$3 in the benchmark case still hold in the situation that each firm's R&D risks have correlations according to the firm's R&D investment levels. 9 Conclusions This study examines the optimal risk choices from a series R&D projects with different risks in a differentiated duopoly market when one firm has a minority of its rival's shares but the other one does not.It indicates that P.P.O.holdings increase the optimal R&D risk chosen by both firms, while the firm that does not own the rival's shares is more willing to choose the higher R&D risk than the one that owns its shares.In addition, for each firm its privately optimal risk is always lower than the socially optimal risk, and the P.P.O.s play an important role in determining the extent of this inefficiency.Finally, it extends to allow bilateral cross-ownership, Bertrand competition, R&D spillovers, product complementarity and R&D risks correlation.The practical implications are as follows.Firm managers should consider the P.P.O.s when they choose the R&D projects in a competitive environment: the larger level the P.P.O.s, the more volatile should be the outcomes of their selected R&D projects.In addition, policymakers should be aware that the P.P.O.s are conducive to promoting firms to choose high-risk R&D projects and may reduce the inefficiency of private R&D incentives. In future studies, we can consider the government policy such as R&D risk sharing program or R&D subsidies, and consider the heterogeneous firms such as mixed duopoly or firms with different level of C.S.R. Notes 1.For example: (1) in May 2009, Daimler, a German carmaker, said that the company had acquired nearly 10% of Tesla Motors Inc, an electromagnetic drive car manufacturer in California.After that, both companies aimed at the research and development of new battery-driven cars with higher risk; and (2) after Microsoft held the stock of apple in 1997, both of them carried out the research and development of more advanced operating systems (Windows 98 and Mac OS X) with higher risk.2. Note that we do not consider the R&D spillover effects.This simplifying assumption aims to show the only impact of PPOs on the R&D risk choices.However, there might exist research spillovers between the firm's R&D investments, which is also an important key factor in the literature of R&D (Banal-Estañol et al., 2022;Zhuang & Zhao, 2022).We will test the robustness of the main propositions when considering R&D spillovers in section 5.3.3. We think that the assumption of Covðx i , x j Þ ¼ 0 (i 6 ¼ j) is reasonable in the R&D competition environment.The reason is as follows.To prevent the disclosure of R&D secrets, firms generally formulate strict R&D confidentiality regulations, which prevent competitors from obtaining relevant information (Shen et al., 2010).When there is R&D competition among firms, they will avoid the disclosure of R&D information.However, x i and x j may be dependent under certain conditions.We will test the robustness of the main propositions when Covðx i , x j Þ 6 ¼ 0 (i 6 ¼ j) in section 5.5.4. Zhu (2011) considers the risk costs of technological innovation. 5. We further assume that the function Iðl i , r i Þ is twice continuously differentiable about r i and meets o 2 Iðl i , riÞ or i 2 >0: 6.This is because the strategic effect of R&D on the profit in relation to its rival's output is positive under Cournot competition (due to the fact that the outputs are strategic substitutes, if Cournot firm invests more and produces more, its rival firm invests less and produces less) while that effect of R&D on the profit in relation to its rival's output is negative under Bertrand competition (due to the fact that the prices are strategic complements, if Bertrand firm invests more and produces more or sets lower price, it rival firm invest more and produces more or sets lower price) (Chen & Lee, 2022).7. The proof of the main results in this section is the same as that of Propositions 1 $ 3. To avoid repetition, we omitted their proof. 8. Note that the conceptual difference exists R&D spillovers (Section 5.3) and R&D risk correlation (Section 5.5).Following D 'Aspremont and Jacquemin (1988), we think that the R&D spillovers imply that some benefits of each firm's final R&D results (e.g.successful inventions) flow without payment to other firms.This refers to the spillovers of the final R&D results of a firm, rather than the leakage (or disclosure) of relevant information in the R&D process.Thus, the R&D spillovers do not imply that x 1 and x 2 are dependent in this study.In addition, 'each firm's R&D risks have correlations according to the firm's R&D efforts' means that x 1 and x 2 are not independent of each other.However, if each firm obtains the R&D information of its competitor in the R&D process, the final R&D results may be related and this may lead to Covðx 1 , x 2 Þ 6 ¼ 0: 9.The proof of the main results in this section is the same as that of Propositions 1 $ 3. To avoid repetition, we omitted their proof. Disclosure statement No potential conflict of interest was reported by the authors. Funding We Appendix A Proof of Lemma 1. Appendix G Proof of Proposition 5. Similar as in sections 3 and 4, we can prove that r B i and r SB i satisfy: Thus, we can prove the following results: (i) According to ( 29) and (30), gðr B 1 Þ ¼ are very grateful for the support of Natural Science Foundation of Shandong Province [grant number ZR2020MG072], Soft Science Project of Shandong Provincial Key R&D Plan [grant number 2021RKY05095] and MOE (Ministry of Education in China) Project of Humanities and Social Sciences [grant number 17YJC790137].	7,659.4	2022-11-03T00:00:00.000	[ "Business", "Economics" ]
The Role of Xenobiotic Transporters in Ophthalmic Drug Delivery - The eye is a very complex sensory organ consisting of numerous structures to coordinate the function of sight. It has a series of physical and chemical barriers to help maintain its homeostasis, and mediate environmental exposures. Transporters in the eye play a very important role in maintaining homeostasis by facilitating the movement of ions, nutrients and xenobiotics to various tissues in the eye, especially to non-vascular tissues like the lens and cornea. They also ensure proper cell signaling by shuttling neurotransmitters within the retina. Thus, they are expected to play an important role in determining the ocular exposure of drugs and other pharmacotherapeutics. However, the role of ocular transporters in ophthalmic drug delivery and their clinical relevance has not been well characterized. The purpose of the present review is to summarize the current evidence in the literature on ocular drug transporters and their role in ocular drug delivery, with the emphasis predominantly on their role in ocular pharmacokinetics. This article is open to POST-PUBLICATION REVIEW . Registered readers (see "For Readers") may comment by clicking on ABSTRACT on the issue's contents page. Higher biotin-GCV permeability into the retina-choroid and slower elimination from vitreous GLUT1 on the HRPE cells Glu-dopamine Transporter recognizes prodrug, not the parent drug OPT oligopeptide transporter, SMVT sodium-dependent multiple vitamin transporter, B 0,+ amino acid transporter, GLUT glucose transporter, rPCEC rabbit primary corneal epithelial cells, HRPE human retinal pigment epithelium cells, RCS retina-choroid-sclera, ACV acyclovir, GCV ganciclovir, RPE retinal pigment epithelium a Substrate of B 0,+ INTRODUCTION The role of drug transporters in the gastrointestinal (GI) tract, lymphatic system, blood brain barrier (BBB), liver and kidneys is well known and documented [1][2][3][4][5][6]. Transporters can affect multiple aspects of drug disposition and can result in potential drug-drug interactions, lack of efficacy, toxicity and drug related adverse events relative to exposure levels. Drug transporters have the potential to alter the efficacy of a molecule at the site of action, for example, in the liver or the brain [1]. The brain parenchymal cells contain transporters including Pglycoprotein (P-gp), which can efflux drugs, thus lowering their concentration at the site of action and ultimately decreasing their efficacy [2][3][4][5]. Conversely, facilitative transporters like organic cation transporters (OCTs) or dopamine transporter (DAT) would improve drug efficacy by transporting the molecule into or out of neurons, thus, increasing their target site concentration [5][6][7]. Although topical ocular administration is the primary method for delivery of therapeutics to the eye, not much is understood about the clinical relevance of transporters in ocular drug disposition. The eye, being responsible for the sense of vision, has evolved into an organ with complex anatomy and physiology to maintain homeostasis and ensure effective functioning. The mechanisms include transporters, ion channels, and physical barriers which act as defense mechanisms and help maintain concentrations of essentials nutrients. While there are good reviews on ocular drug transporters, little is known about their role in ocular drug efficacy, ocular pharmacokinetics and ocular safety issues [8][9][10]. In the present article, we review the current evidence in the literature on ocular drug transporters and their role in ocular drug delivery, with the emphasis predominantly on their role in ocular pharmacokinetics. We will also discuss their future relevance and the need to better understand their role in ocular drug delivery. ANATOMY OF THE EYE The eye is a very complex structure that is closely connected to the rest of the body through its vascular and neural networks [11,12]. Its unique anatomy and physiology place several constraints on the delivery of drugs to the eye because of its innate nature to protect itself and, ultimately, the vision from exogenous substances [11]. Therefore, it is critical to understand the anatomy and physiology of the eye to design adequate and effective drug delivery systems. The objective of this review is not to cover in detail the anatomy of the eye since there are excellent reviews and articles that cover this topic in detail. We recommend the reader review these references for further detail [11,. Ocular drug disposition is not only influenced by the complex anatomy, but also by lacrimation, tear film dilution and tear turnover mechanisms [34]. The presence of melanin and ocular transporters, as well as the blood-aqueous barrier (BAB) and blood-retinal barrier (BRB) significantly affect ocular drug disposition [12,[35][36][37][38][39] [40][41][42][43][44][45]. There are several routes of ocular drug delivery ( Table 1). The benefits and challenges of Anesthesia, prevention of endophthalmitis, inflammation and pupil dilation the various routes of administration [12] as well as the complexity of ocular pharmacokinetics [46] have been reviewed elsewhere. Briefly, topical ocular administration is generally the preferable route for patients based on ease of administration and can be used for the treatment of a variety of ocular diseases. However, invasive techniques (i.e. intravitreal or intracameral injections) need to be employed for diseases such as age-related macular degeneration (AMD), retinal vein occlusion, or macular edema. TRANSPORTERS AND THEIR ROLE IN FUNCTIONING OF THE EYE The eye is an important sensory organ and is one of the very few which come in direct contact with the environment. One can therefore appreciate the role of transporters as a defense mechanism preventing the entry of foreign and/or possibly toxic xenobiotics into the eye, which could disrupt vision [8][9][10]. To date transporters in the eye and their role in ophthalmic drug delivery have not been well characterized. In addition to serving as a defensive mechanism, transporters also help maintain pH, ionic and osmotic equilibrium in the eye [47][48][49][50]. Membrane transporters regulate the levels of ions, glucose and vitamins to maintain homeostasis in the eye [51]. A disruption of these mechanisms leads to various conditions including cataract [51]. In a similar fashion, inhibition and activation of transporters in various matrices of the eye could lead to safety issues. For example, inhibition of the glucose transporter, found in the lens and the conjunctiva could lead to disruption of the homeostasis of the lens [51]. Cellular transporters play an important role in the disposition of drugs at the site of therapeutic action. An efflux transporter could limit the amount of drug reaching the target ocular tissue limiting their efficacy, while an uptake transporter could result in elevated levels of a drug in a particular tissue. Ophthalmic drug transporters have been the subject of investigation and the published literature indicates that they play a role in ophthalmic drug delivery. In the following sections we present information on ocular transporters known to be expressed in the eye, dividing them into anterior and posterior segments. A brief review is provided, of in vitro and in vivo techniques used to study transporters and evidence of the role of ocular transporters. In vitro Methodologies A variety of in vitro techniques have been employed to investigate the expression, localization and function of transporters. Techniques like PCR, Western blotting and immunohistochemistry are used to study the expression of transporters and their localization in various tissues [52,53]. To determine their functionality, current technologies on transporters utilize either inhibition or substrate transport studies (done as a monolayer or in suspension). Transport studies with compounds in transfected cell lines, vesicles from insects, transporter cDNA expressing insect membranes and oocytes are used to identify the substrate nature of compounds and to determine their kinetics [54][55][56][57]. Inhibition studies can also be conducted using the afore mentioned techniques. These studies reveal the substrate or inhibitor nature of a compound, as well as the nature of inhibition (competitive against non-competitive). Corneal cell lines from humans and animal species [49,52,53,[58][59][60][61][62][63], are available to study drug permeability and drug transporters. The role of transporters in conjunctival cell lines like HCjE (human conjunctival epithelial cells), CJVE (rabbit conjunctival epithelial cells) has also been demonstrated [58,59,64]. The studies with HCLE (human cornea limbal epithelial cells) and HCjE cells were the first to show the expression of OCTN1 and OCTN2 in these cell lines and demonstrated the role of OCTN2 in the active transport of L-carnitine transport. The studies with CJVE cell line demonstrated the function of novel sodium dependent and sodium independent transport mechanisms for synthetic and endogenous opioids. The permeability of compounds across retinal cell lines like ARPE-19 and retinoblastoma cells [65,66] has also been investigated. The study revealed the presence and expression of a new oligopeptide transporter (SOPT2), which transported synthetic opioid (DADLE) with partial sodium dependence. Such in vitro systems become useful to determine the permeability of compounds and the role of transporters across these barriers. The use of in vitro studies to characterize the capacity of a compound to act as a substrate or inhibitor of a particular transporter provides information about its ability to reach the intended site of action. Furthermore, these studies provide understanding about the kinetics of the process, the possible interactions and the potential clinical relevance. In vivo and Ex vivo Methodologies Both in vivo and ex vivo methodologies are also used to study transporters. Inhibition and saturability studies are done with wild type animals to determine in vivo kinetics of transporters. Drugdrug interaction potential can also be tested using transporter knock-out or transgenic animals [67-71], which are deficient in a particular transporter compared to the wild type animal. Studying the pharmacokinetics of the compound in the two models simultaneously gives a measure of the contribution of the transporter to the maximal concentration (C max ) and area under the concentration-time curve (AUC) of the compound. In vivo studies to investigate the role of ocular transporters have been conducted in rabbits [72]. The role of P-gp was well demonstrated by investigating the pharmacokinetics of quinidine and erythromycin in rabbits [72,73]. The role of MRP5 in rabbits was also studied using acyclovir as a model substrate [53]. Potential ocular or ocularsystemic drug-drug interactions can also be investigated using such in vivo and ex vivo methodologies [74,75]. Some of the ocular tissues including cornea and retina can be excised and the permeability of compounds studied across the sections. Ex vivo studies with isolated cornea [49, 61, 62] and isolated retina [65,76] have been conducted to determine the permeability of molecules and the role of transporters. Some of the new technologies being explored to study transporters involve the use of antibodies to knock out transporters and study their role in drug pharmacokinetics [77]. A second technique is the inhibition of transporters, using siRNA [78,79], to study the role of transporters. These are fairly new techniques which have not been thoroughly investigated for ophthalmic drug delivery. In vitro and ex vivo techniques can be used to rank order compounds. In conjunction with in vivo studies, the in vitro and ex vivo data can be used to conduct in vitro-in vivo correlations (IVIVC). Further refinement of these techniques needs to be conducted to adapt them to study the role of transporters in the eye. Transporters in the Anterior Segment of the Eye The tissues/matrices in the anterior segment of the eye are conjunctiva, cornea, aqueous humor, lens, lens capsule, iris, ciliary body and trabecular meshwork. Transporters are known to be expressed in the conjunctiva, cornea, lens, iris and ciliary body. However, our major focus will be on transporters in the conjunctiva and cornea while only mentioning the transporters in the lens and ICB [47,49,51,[80][81][82][83][84]. Drug Transporters in the Cornea Transporters in the corneal epithelium are outlined in Figure 2 [9]. There are both uptake and efflux transporters, for both large and small molecules which can influence drug exposures. These transporters serve to transfer molecules across the epithelium in either direction (towards the tear film or towards the aqueous humor). Studies on acyclovir and acyclovir prodrugs demonstrated not only that peptide transporters like PepT1 are present on the corneal epithelium, but also play an active role in the transport of these prodrugs. The presence of a facilitative transporter, OATP2A1, has been demonstrated in cornea, conjunctiva, iris and ciliary body and its role in the permeability of both latanoprost and its free acid metabolite was reported Drug Transport in the Iris-ciliary Body (ICB) and the Lens The lens is rich in nutrient transporters and ion channels to maintain the osmotic and refractive nature. A glucose transporter was shown to be present in the ciliary body of human and rat eye Retina is considered to be the ocular tissue with the highest metabolic rate per weight [108]. This is primarily because the retina is considered an outgrowth of the developing brain and has similar neural constituents that require an active protection system as seen in brain tissue [109]. One of these protection systems is the blood-retinal barrier (BRB), which is formed by tight junctions between the cells of the RPE and the endothelial cells of the capillaries and prevents the leakage of protein or fluid from the vasculature into the retina [110]. For instance, Figure 3 represents the transporters at another barrier-a barrier to drug delivery to the back of the eye, more popularly known as the BRB [9]. This figure shows the transporters at the outer and inner blood-retinal barriers. There are multiple uptake and efflux transporters and the localization of P-gp on both surfaces of the epithelium is puzzling and interesting. The transport across this epithelium is also a combination of paracellular, transcellular and active transport. This barrier is more complex than the BAB and represents a major hurdle for drug delivery to the back of the eye, both from topical ocular and non-topical delivery [9]. Drug Transporters in the Retinal Pigmented Epithelium (RPE) and Retina Furthermore, various glucose transporters facilitate the transport of glucose across the bloodaqueous barrier (BAB) and blood-retinal barrier (BRB) [34,95]. For instance, GLUT1, GLUT3 and GLUT4 are high affinity glucose transporters, while GLUT 2 is considered a low affinity glucose transporter. GLUT5 is a high affinity fructose transporter [95]. In the case of GLUT 1 it was reported in RPE, choroid, par plasma, lens fiber cells and retinal Mueller cells [111]. It has also been reported that amino acid transporters including glutamate, glycine, GABA, proline and tryptophan are present on the retina [107, 112]. The GLUT1 glucose transporter is expressed in endothelial and epithelial barriers like the retinal capillary endothelium and RPE, which was studied in diabetic and nondiabetic human eyes [113]. Monocarboxylic acid transporters (MCTs), which transport pyruvate and lactate, among other carboxylic acids have been found in the retina [107, 114, 115]. Specifically, MCT1 has been reported on the apical membrane of rat RPE, while MCT3 has been described on the basolateral membrane [116]. Recently, a folate receptor (FR), which is a specialized carrier-mediated active transporter system, has been described in human derived retinoblastoma cell line (Y-79) [117]. Similarly, in the same cell line biotin has been reported to be transported via a human sodium dependent multivitamin transporter (hSMVT), which is a specialized carrier-mediated system for biotin uptake into retinoblastoma cells [118]. The same research group also reported a riboflavin transporter in Y-79 cells, which is a transporter system that is regulated by protein kinase A and Ca 2+ /calmodulin pathways [119]. It needs to be mentioned that the water transport across the RPE and other ocular tissues has been reviewed elsewhere [109]. Briefly, the water transport across the RPE is mediated via an active solute-linked water transport via monocarboxylate transporter (MCT1 and MCT3) and via osmotic and hydrostatic forces that determine the other two passive mechanisms for water transport [109]. ROLE OF TRANSPORTERS IN OCULAR DRUG DELIVERY -FRONT AND BACK OF THE EYE As stated above there is the presence of efflux and influx transporters in various ocular cell lines and tissues. While the efflux transporters lower the bioavailability of a drug by effluxing it out of the cell membrane and cytoplasm, the influx transporters will facilitate the translocation of a drug across biological membranes. Therefore, it can be understood that transporters play a crucial role for ocular drug delivery. It needs to be understood that the permeation of a drug through the eye will be dependent on the passive transport of the administered drug and its concentration gradient [9]. Keeping in mind that active transport is against the concentration gradient and needs energy, careful consideration needs to be taken for an ophthalmic drug, considering that transporter proteins may become saturated at high concentrations. This is highly relevant after intravitreal injection when there is a high local concentration in the vitreous humor. It needs to be mentioned that choroidal vessels present fenestrations from which a drug can escape this vasculature and leak out to the RPE and then reach the neural retina and vitreous humor. Thus, a drug can enter the vitreous humor and retina via the retinal capillaries or the blood stream of the choroid [9, 120]. The gene expression of both uptake and efflux transporters in different parts of the eye was measured by Zhang et al. [121]. Based on their findings, MRP1 seems to be a more important efflux transporter for the eye compared to P-gp. However, as is the usual case, more attention has been devoted to P-gp than to MRP1 in functional assays. As far as the facilitative transporters are concerned, OATPs, PEPT2, OCTs, OCTNs seem to be important for the eye. Though the expression, presence and localization of these transporters have been studied by Zhang et al. [121], the functional activity of all these transporters has not been well characterized. In other words, the presence or expression of a transporter indicates that the transporters might be functionally active, but does not guarantee it. The functional activity also needs to be demonstrated. As presented, many of the ocular transporters have been cloned and expressed in various ocular cell lines and tissues. Some of them are involved in various processes including absorption, distribution and excretion of ophthalmic drugs. Because this is a very active field of study, various drug delivery approaches are taken to develop more effective therapeutic agents. The challenges to effectively deliver drugs to the posterior part of the eye following topical ocular administration are well known. The first barrier is the high tear turnover rate which will wash off or dilute the dose of the administered drug causing precorneal loss [10]. Then, there is the presence of efflux and influx transporters in the eye, the blood-aqueous barrier (BAB) and blood-retinal barrier (BAB). Presented below will be various approaches that have been pursued to circumvent these issues. For example, transporter-targeted prodrug delivery has been utilized to improve bioavailability in the eye. To date, the role of P-gp has been the most studied, as was the case with oral route of drug delivery. The localization and functional activity of P-gp and MRP in porcine eyes was demonstrated [101]. The molecular evidence and functional expression of MRP2 in human corneal epithelium and rabbit cornea and its role in ocular drug efflux was demonstrated [125]. Dey et al. [73] demonstrated the effect of P-gp in erythromycin pharmacokinetics in rabbit and human cornea. The corneal AUC of erythromycin in the presence of testosterone, a P-gp inhibitor, was significantly increased indicating a role for P-gp in influencing corneal drug bioavailability. MRP1 has not received much attention despite the evidence presented by Zhang et al. [121] and more studies are warranted. Some ophthalmic relevant drugs with their mode of administration along with the uptake and efflux transporters affecting their disposition is given in Table 2 [9]. Influx Transporters In the case of ocular influx transporters there are mainly amino acid and peptide transporters [12]. For instance, ASCT1 (SLC1A4), a neutral amino acid transporter that belongs to the SLC1 gene family has been detected in rabbit cornea and in rPCEC (rabbit primary corneal epithelial cells) [63]. Similarly, ASCT2 (SLC1A5) was expressed in retinal Muller cells and it was suggested that this transporter also serves as an effluxer of D-serine [102]. The neutral and cationic amino acid transporter B 0,+ (SLC6A14) has been found to be expressed in rabbit cornea, rabbit corneal epithelium and human cornea and to be involved in the L-arginine transport across corneal epithelium [126] but also across pigmented rabbit conjunctiva [127]. Furthermore, the Na+-independent large neutral amino acid transporter LAT1 (SLC7A5) was identified in human and rabbit cornea [49], while LAT2 (SLC7A8) was identified in the posterior segment [using an in vitro human model using RPE cell line (hTERT-RPE)] [128] and in ARPE-19 cells where it was determined to be involved in L-phenylalanine transport [129]. The peptide transporters are proton coupled transporters that contribute to the translocation of di-and tripeptides across the epithelium [130] and are mainly classified into PepT1, PepT2 and peptide/histidine transporters (PHT1 and PHT2) As described above there are various transporters that have been identified in ocular cell lines and tissues. Therefore, various approaches have been pursued to circumvent efflux transporters or to take advantage of the influx transporters. The most common approach is using transportertargeted prodrugs (Table 3) [12]. This approach has led to improvements in ocular bioavailability of various drugs since it takes advantage of the ocular influx transporters or due to changes in physicochemical properties in the prodrugs or by a combination of these two factors. In general, the prodrugs are recognized by the ocular membrane transporters as substrates and allow their translocation across the epithelia. So far this approach has been undertaken for transportertargeted drug delivery to cornea, conjunctiva and RPE [12]. For instance, studies on acyclovir and acyclovir prodrugs demonstrated not only that peptide transporters like PepT1 are present on the corneal epithelium, but also play an active role in the transport of these prodrugs. The presence of a facilitative transporter, OATP2A1, was demonstrated in cornea, conjunctiva, iris and ciliary body and its role in the permeability of both latanoprost and its free acid metabolite was reported [87]. More examples are presented in the section below. Higher aqueous solubility of the prodrug along with the transporter recognition [245] B 0,+ on the cornea Phenylalanine-ACV and EACV The prodrugs inhibited the transport of L-arginine a across the cornea implied that they are substrates of B 0,+ [126] OPT system on the cornea L-valine ACV Three-fold higher transcorneal permeability of L-valine ACV compared to ACV [157] OPT system on the cornea Gly-Val-GCV, Val-Val-GCV and Tyr-Val-GCV Significant transcellular passive diffusion and transporter recognition resulted in higher AUC and C max [246,247] OPT system on rPCEC cells and the cornea Val-quinidine and Val-Val-quinidine Prodrugs were not recognized by P-gp efflux pump and further found to be substrates of peptide transporters [248] OPT system on the retina Gly-Val-GCV, Val-Val-GCV and Tyr-Val-GCV Two-fold higher RCS tissue permeability than that of GCV due to higher lipophilicity and translocation mediated by OPT across RPE [249] SMVT [149][150][151][152][153][154][155][156]. While these approaches seem to be common in oral drug delivery, such an investigation is lacking in the field of ophthalmic drug delivery. Some of the above mentioned excipients like xanthan gum, EDTA, cremophor and PEGs, which are used in eye drops for topical ocular delivery have been shown to inhibit transporters and can potentially play a role in increasing ocular tissue exposures. However, studies to determine the contribution of transporter inhibition and increased residence time/penetration enhancement have not been done. Investigations into approaches to avoid transporters would be useful to improve ocular drug bioavailability. Prodrugs which are metabolized to the active moiety in vivo have been used in ocular drug delivery (Table 2). Latanoprost, when administered via topical ocular route is metabolized in vivo to its active form. It was demonstrated that OATP2A1, which is expressed in both RPE/choroid and anterior segment tissues like cornea, conjunctiva, iris and ciliary body, plays a role in the permeability of both latanoprost and its free acid metabolite [87]. In another study, it was demonstrated that bimatoprost and latanoprost and their free acid metabolites were substrates of MRP1, MRP2 and MRP5, using MDCK cell lines over-expressing these transporters and excised rabbit cornea [82]. Furthermore, it has been reported that acyclovir prodrugs, which utilize both ocular transporter and enzyme interplay can be used to deliver drugs to the eye [157-159]. CURRENT AND FUTURE RELEVANCE OF OCULAR TRANSPORTERS IN DRUG DEVELOPMENT Many clinically available ophthalmic drugs are known to be either substrates or inhibitors of transporters, and are presented in this review. While the expression and presence of both facilitative and efflux transporters has been demonstrated, knowledge of their role in affecting ocular pharmacokinetics in preclinical species is limited. Their relevance to clinical ocular pharmacokinetics has not been demonstrated. A drug-drug interaction between oral/systemically administered compounds and topical ocular administered compounds can occur where either of the molecules can be a victim or perpetrator of the interaction. Oral and systemically administered drugs can distribute to the eye due to the action of transporters. These molecules can interact with topically administered drugs altering their local ocular pharmacokinetics, safety and efficacy. In the same way, topically administered molecules can affect the ocular distribution of orally administered compounds by inhibiting the efflux transporters in blood-ocular barriers, of which the oral/systemically administered compounds are substrates. Hippalgoankar et al. in fact showed that such drug-drug interactions are possible by studying the interaction between a topically administered and systemically administered P-gp substrates/inhibitors [74]. Thus, these transporters play an important role in affecting the ocular disposition of the drugs and affect their therapeutic action of the drugs. At the same time such oral and systemically administered drugs distributing to the eye due to transporter mediated drug-drug interactions can cause unwanted ocular effects. Interplay between enzymes and transporters, like CYP3A4 and P-gp has been well studied [160][161][162][163][164][165]. However, the same cannot be said of ophthalmic drug delivery. Some of the prodrug approaches used in ocular drug delivery, which utilize the interplay of enzymes and transporters, are mentioned in this review. These studies demonstrate that ocular tissue levels of the active metabolite can be modified by the action of the efflux and facilitative transporters. Such studies demonstrate that enzyme transporter interplay is possible in ocular drug delivery and needs to be further investigated, either to the advantage of drug development and delivery or to prevent possible pitfalls which can arise due to genetic differences in the populations of transporters and enzymes. Molecules which are designed to utilize body's existing transporter systems to improve their systemic and target organ/tissue bioavailability can be used to improve posterior segment drug exposures upon topical ocular delivery as also for decreasing systemic drug exposures upon intraocular (intravitreal or intracameral) administration. Such approaches are currently being tested in the clinic with oral drug delivery. CONCLUSIONS While the expression and presence of transporters in the eye has been well demonstrated, studies to understand their role in ocular drug delivery, vis-àvis their role in ocular pharmacokinetics, efficacy and safety, are only in their infancy. While the role of transporters in ocular homeostasis is known, a disruption of these mechanisms by drugs modifying transporter function needs to be studied. While the in vitro and in vivo techniques for investigating the role of transporters in ocular drug delivery already exist, intensive investigation needs to be carried out before their clinical relevance can be elucidated and understood	6,273.4	2013-12-03T00:00:00.000	[ "Chemistry", "Medicine" ]
Factors influencing the usage of XBRL tools , Introduction With the recent rapid advancement in digital financial repotting, not only changed the type of disclosure of financial statements but also changed the tools of disclosure previously were paper, Excel, PDF, or HTML now became XBRL (Al- Rawashdeh, 2011). XBRL gives companies the ability to comply with regulations using the XBRL tools that they prefer to work with companies now use a new tool for preparing the data to XBRL public company disclosure. Thus, XBRL tools are crucial for disclosure about the face of the financial statements and related footnotes (Blankespoor, 2019). The XBRL Tool is a software that supports the XBRL tagging process and preparation of XBRL documents. These tools have considerable potential that provides new possibilities for financial statements and footnotes that are not available in non-XBRL formats (Efimova et al., 2019). XBRL tools can enhance reporting and analysis quality and users' decision quality. XBRL is an XML-based vocabulary for electronic transmission of financial and non-financial data; each item is enclosed by a pair of XBRL tags. XBRL is an open standard that is maintained by XBRL International. With support from the governing bodies worldwide have secured XBRL as the official standard for financial reporting. Saudi Arabia is no exception, Tadawul has adopted XBRL as new technology to ensure equal access to corporate disclosures and enable users to analyse financial statements fairly among all users (Rawashdeh & Selamat, 2013). XBRL tool is a critical software of users, and thus its acceptance and usage receive growing interest for users, companies, and providers. Moreover, recent studies (Liu et al., 2017;Tohang & Lan, 2017;Ilias et al., 2020) have indicated that many countries have adopted XBRL. However, users' interest, acceptance, and usage of the XBRL tools are not well known unlike XBRL itself as a global framework for exchanging business information. Thus, investigating the factors affecting users' acceptance of XBRL tools and their intentions to use them in a wide way is critical. Therefore, this study examined the behavioural intentions of users to use XBRL tools. In the context of information systems and information technology, the widespread use of new technology tools relies on user acceptance (Davis, 1989). In recent years, several researchers have been able to develop various theories and models for predicting the behaviour of users towards adopting new technology or systems in different areas. The technology acceptance model (TAM) is among the most used models when reviewing the adoption and diffusion literature (Wu & Chen, 2005;Chouhan & Goswami, 2015; Althunibat et al., 2019;Baby & Kannammal, 2020). Although TAM has been used a lot by researchers (Wu & Chen, 2005;Chouhan & Goswami, 2015;Althunibat et al., 2019;Baby & Kannammal, 2020) and still is, in the end, some negative aspects appeared in this model. In a new attempt to overcome the negative aspects of the TAM model, to obtain a more realistic model Venkatesh et al. (2003) integrated several common factors from eight models and prominent theories. These include the social cognitive theory [SCT], the theory of planned behaviour [TPB], innovation diffusion theory [IDT], the TAM; the motivational model (MM), the combined TAM-TPB, the model of PC utilization [MPCU], and the theory of reasoned action [TRA]. To create a new technology adoption, acceptance, and usage model. This combination of carefully selected variables is called the unified theory of acceptance and use of technology (UTAUT) model. Recently, the UTAUT model is a widely used model for forecasting acceptance and usage of technology and new systems. The UTAUT model has been applied in various fields. For example, the adoption of mobile payment (Patil et al., 2020), The adoption of a social learning system (Khechine et al., 2020), acceptance of mobile health (Petersen et al., 2020), acceptance of Mobile learning system (Almaiah et al., 2019;Chao, 2019), web 2.0 tools (Gitau, 2016) and acceptance of software engineering tools (Wrycza et al., 2017). The previous researches are a sample of the widespread use of the UTAUT model, the UTAUT model provides high flexibility that enables it to be used in multiple fields. This flexibility stems from the ability of this model to integrate several theories and models into a fairly acceptable model. Despite the structure on which the UTAUT model is built, but the model has faced some criticism in its ability to explain individuals 'acceptance of technology-based systems, criticism makes sense if we consider that humans and technology are constantly evolving. Therefore, several researchers believe that the original UTAUT model should be expanded if necessary according to the nature of the research in question. Some of them added to it some variables as an attempt to calm doubts about the ability of the model to predict the acceptance of technology-based systems by individuals. For example, some researchers (Taiwo et al., 2012;Chao, 2019;Rahi et al., 2019;Manrai & Gupta, 2020) suggest adding other internal and external factors that could enhance the model's ability to predict the acceptance of technology-based systems. XBRL tools handle quantities of data. There are many types of XBRL tools, including coding financial data, validating data, and financial analysis. Therefore, there are various users of XBRL tools and they use XBRL for a variety of purposes. XBRL tools are used to process sensitive financial data, any error that may affect the financial position of the company, and used for financial analysis and any error that may lead to making a wrong investment decision. For purposes of this research, it is expected that adding trust and satisfaction factors to the original UTAUT model will play an important role in accepting and using XBRL tools. For example, the trust factor was incorporated as a predictor in the model by several researchers (Manrai & Gupta, 2020). According to a previous study on the adoption of online tax filing (Carter et al., 2011), trust is a vital factor in determining users ' behavioural intentions to adopt technology-based systems. For this research, the UTAUT has been adopted and expanded by incorporating trust and satisfaction factors. To investigate the behavioural intentions of users toward the use of XBRL tools in preparing and validating, and analyzing financial statements and footnotes. Adopting XBRL as a global framework for exchanging business information differs from adopting XBRL tools, in the case of XBRL, there is one framework called XBRL and there is no other option. However, in the case of XBRL tools, there are many tools available for a variety of purposes and thus the investigation of the factors affecting XBRL tools may differ from the investigation of the factors that affect the adoption of XBRL as a global framework for exchanging business information that can be adopted. Therefore, this research aims to investigate the factors affecting the behavioural intention to use XBRL tools. In addition to developing an expanded model, that contains trust and satisfaction. This research also aims to prove that satisfaction is a mediating and predictive variable of the behavioural intentions of users to use XBRL tools. Finally, the research aims to confirm the model experimentally in this context. To achieve the aims of the research, the following four research questions were formulated. (1) What factors affect the behavioural intentions of users affecting the use of XBRL tools? (2) Does trust affect the UTAUT model regarding XBRL tools? (3) Does satisfaction affect the UTAUT model regarding the use of XBRL tools? (4) How can both trust and satisfaction affect the expectation of behavioural intention to use XBRL tools? Most of the previous studies focus on adopting XPRL as a global framework for exchanging business information. Since XPRL tools are available in most companies, this research is projected to contribute the related literature to the acceptance and use of technology-based systems by determining trust as a precedent for the use of XBRL tools. Promote and deepen the theoretical understanding regarding the behavioural intention among users regarding the use of XBRL tools for analytical purposes. Provide empirical evidence on the role of satisfaction as a mediating variable that links the internal and external factors and behavioural intentions of users to use XBRL tools. Providing a reference to specialized developers of the XBRL tools on future directions of users related to using XBRL tools. XBRL Adoption Most of the previous studies in this region focus on adopting the XBRL itself and not focusing on adopting the XBRL tools. Ilias et al. (2020) investigated several factors that can effect on XBRL adoption in Malaysia, which are, optimism, innovativeness, discomfort, insecurity, and intention to Use. Uyob et al. (2019) analysed consumer attitude variables concerning MBRS use. The results of this research suggest that perceived usefulness and ease of use are factors that affect the accounting practitioners' attitude to use MBRS. Rawashdeh and Selamat (2013) examined the various normative, attitudinal, and control factors influencing the adopters' intention to accept XBRL in Saudi Arabia. Pinsker and Felden (2016) examine voluntary XBRL adoption intent in Germany; results indicate that professional role and normative pressure have positive relationships with XBRL adoption intent. Lakovic et al. (2018) investigated the determinants of XBRL adoption, their results refer that environmental factors, technical factors, and organizational factors influence the XBRL adoption. Muchlis et al. (2019)analysed the factors influencing early adopters in Indonesia during the early phase of XBRL. They found that the size and profitability of a firm are the financial characteristics of the firm, which determine the early adoption of XBRL. Alkhatib et al. (2019) examined the factors influencing the voluntary adoption of the statutory accounts digital reporting and the return of small private companies in the UK. Their findings indicate that the relative benefit of standardization, support from top management and the company's technology competence factors had a significant effect on the voluntary adoption of digital technologies. Previous studies try to explain what factors can influence XBRL adoption. However, Studies covering XPRL adoption are almost missing from previous studies. In a meta-analysis, Ansary et al. (2020) tried to determine the determinants of XBRL adoption. They divided determinants of XBRL adoption into three primary factor groups: Perceived usability regarding the technical aspect, perceived usefulness and ease-of-use. The second aspect considers environmental factors such as normative pressure. The third dimension is the organizational aspect, which consists of the managerial attitude and organizational expertise. Findings have seemed contrasting to date. Venkatesh et al. (2003) in an in-depth study of a reputable group of theories and models used in the framework of technology adoption. They studied a set of variables that were compiled from these theories and models that are called use in the context of information technology. Through this experimental study of these selected variables, the researchers came up with a model from their point of view, which is considered an ideal model for standardizing models and theories related to accepting technology into one unified model called the UTAUT model (Table 1). Table 1 Core factors and definitions Core factors Definitions Performance expectancy (PE) "The degree to which an individual believes that using the system will help him or her attain gains in job performance" (Venkatesh et al., 2003) Effort expectancy (EE) "The degree of ease associated with the use of the system" (Venkatesh et al., 2003) Social influence (SI) "The degree to which an individual perceives that important others believe he or she should use the new system" (Venkatesh et al., 2003) Facilitating conditions (FC) "The degree to which an individual believes that an organizational and technical infrastructure exists to support use of the system" (Venkatesh et al., 2003) UTAUT is a powerful model that provides a unified theoretical basis for technology adoption to explain user behaviour in several areas (Venkatesh et al., 2003). Several studies have shown that the model can explain 70% of the variance in the behavioural intention to use technology, 40% of the variance in the usage of technology. The model consists of four main variables: performance expectations, effort expectancy, social influence, and facilitating conditions. Gender, age, experience, and voluntary to use as moderators are expected to mitigate the effects of these basic behavioural beliefs on behavioural intentions and systems adoption, as shown in Fig. 1 Table 2 The eight component models and theories of the UTAUT model Trust Feeling insecure is one of the barriers to using new applications. Especially if you are dealing with sensitive data such as financial statements that any mistake in it leads to an impact on the financial position of the company or a mistake in an investment decision. Thus, the trust acts as a critical role in influencing users' using technology-based systems in question (e.g. XBRL tools). Therefore, the current study assumes that trust is an important factor in predicting the behavioural intentions of users towards the use of XBRL tools. In this research context, trust can be defined as the user's belief that the XBRL tools providers behave ethically in practising their business (Pavlou & Fygenson, 2006;Chao, 2019), XBRL tool shall be accredited and certified in compliance with the XBRL specifications. From the perspective of the model of adopting technology-based systems, the trust can influence the personal satisfaction of users to use a particular system (Asiati et al., 2019). For this reason, the trust in tools influences satisfaction in the XBRL tools in terms of accuracy, speed, time-saving which affects the user conviction in the XBRL tools (Al-Rawashdeh, 2011). Furthermore, it affects also user satisfaction emerging from the experience with the demo XBRL tools (Fei et al., 2017). Besides, trust is a central component of Social Exchange Theory (Roloff, 1981). If an XBRL tool is perceived to be trusted for the user more likely to maintain the exchange, relationship with the respective XBRL tools providers. This trend coincided with the relationship between trust and the intention to use XBRL tools, with a high level of trust between users contributing to a high level of intention to use the XBRL tool. Trust is critical in XBRL tools as it affects perceived tool quality and defines user trust in the use of XBRL tools. Investment decisions, for example, depending on the outcome of the disclosure, which in turn depends on the XBRL tools. Trust in applications was found to influence conviction in the previous study (Carter et al., 2011;Nourallah et al., 2019). Therefore, the hypotheses are formed as below: H1: Trust has a significant influence on the behaviour intentions of users to XBRL tools. H2: Trust has a significant influence on the satisfaction of users to use XBRL tools. Effort Expectancy and Performance Expectancy Variables in models differ insignificance. The UTAUT model, according to these criteria, is no exception. According to Venkatesh et al. (2003), Performance and effort expectations are considered to be strong variables in the model in their impact on users' behavioural intentions to adopt technology-based systems. Accordingly, this research assumes that performance expectations and effort expectancy are key determinants of behavioural intentions towards the acceptance and adoption of XBRL tools. From the available data, the use of performance expectations and effort expectations as a direct determinant of the behavioural purpose of the XBRL tools in this research is strong justification, leading to the following hypotheses: H3: Effort expectancy has a significant influence on the behavioural intentions of users to use XBRL tools. H4: Performance expectancy has a significant influence on the behavioural intentions of users to XBRL tools. Satisfaction Although trust is a state of personal conviction, satisfaction is a feeling that arises from user interaction with applications. Satisfaction factor tests the evaluation of the encounter, including the effort and performance expectancy associated with the application or tools used in the tagging or validation or analysing, which can be considered as one of the most important predictors of their subsequent behaviour (Taylor & Strutton, 2010). Users need to be satisfied with the positive value perceived by the XBRL tools. Satisfaction is a mediating factor for predicting users' behavioural intentions towards the use of XBRL tools. In the point of view of technology-based systems, user satisfaction may have a significant effect on user behaviour's desire to use a specific XBRL based system. (Bao, 2015;DeLone & McLean, 2016). Oliver (1981) defined satisfaction "as a summary of the psychological state resulting when the emotion surrounding disconfirmed expectations is coupled with the consumer's prior feelings about the consumption experience." According to DeLone and McLean (2016), This satisfaction is the degree to which users are satisfied with tagging, validation and analyses services. Consent meanings have different contexts, several linguistic definitions exist for pleasure, and together they make the meaning rich and simple. In this study, satisfaction is described as satisfaction with something and satisfaction with something. For example, satisfaction in a specific XBRL tool and satisfaction about a specific XBRL tool (continuous satisfaction). Chao (2019) stated Effort expectancy and effort expectancy had significant effects on satisfaction. Besides, Asiati et al. (2019) that trust affected satisfaction. As a result, this research hypothesized that user satisfaction rates could significantly influence users' behavioural intentions to use XBRL tools, the following hypotheses were suggested. H5: Effort expectancy has a significant influence on the user's satisfaction with and about XBRL tools. H6: Effort expectancy has a significant influence on the user's satisfaction with and about XBRL tools. H7: Satisfaction has a significant influence on the behavioural intentions of users to use XBRL tools. Research instrument To complete this study, the research instrument was built to be divided into two parts. In the first section, demographic statistics were given on a nominal scale. The questionnaire gathered basic information on the characteristics of the respondent, such as age, gender, employment, occupation and experience. The second part, containing 21 items, was used to calculate the five factors used in the research model (Fig. 2). These five factors have been identified as endogenous variables (effort expectancy, Effort expectancy, trust, satisfaction, and behavioural intentions to use XBRL tools). Each factor is measured by several items. In order to measure the factors involved in this study, a 5-point Likert scale was implemented to obtain questionnaire responses. The Likert scale consisted of five answer options, starting from "strongly disagree" (mapped to No. 1) to "strongly agree" (mapped to No. 5). After evaluating similar research using the UTAUT model as their theoretical basis, the study method was designed to collect data from the respondents with it (i.e., effort expectancy, trust, effort expectancy, satisfaction, and behaviour intentions to use XBRL tools). Similarly, the measures relevant to this study have been established. Some of the original factors have been taken from the UOTAUT model; other factors have been taken from related studies in the sense of the adoption of technology and information systems, see Table 3 below: (Venkatesh et al., 2003;Rawashdeh & Selamat, 2013;Chao, 2019) Detailed sections and items are displayed in sequence in Appendix 1. According to Al-Rawashdeh (2011), If the questionnaire has been modified, it is preferable to conduct a pilot study to improve the validity of the questionnaire and to verify the accuracy of all measurement elements. After that, the reliability test was performed using Alpha Cronbach, with an acceptable minimum of 0.7 (Bland & Altman, 1997). A total of 35 responses have been processed and analysed. Acceptable reliability values ranged from 0.75 to 0.84, which are compatible with the minimum values of Alpha Cronbach. Such findings showed that the questionnaire was primed, accurate and accessible to collect data from the respondents. The coronavirus (COVID-19) outbreak has stimulated many researchers to conduct online surveys, which mainly rely on online convenience samples, emails, or social media sites where participants choose themselves. Studies for generalizing the selection of a random sample representative of the general population are important, but this does not mean that work in which findings can not be generalized does not provide a value, and its utility can be much more than generalized study. Data Analysis and discussions To develop a new model, PLS-SEM can be used because it is used in the initial stages of model development or theory. Because it helps the researcher to explain the reliability and reliability of the measurement model and its structure, therefore PLS-SEM was used in this research. Empirical data were collected via an online survey, in which messages were sent to potential respondents to participate, and colleagues in the participating field were invited to participate, and others were encouraged to participate. Respondents are totally anonymous and no information relating to their identity has been obtained, except for demographic details, which is not connected to the identification of the respondent. The privacy of the respondents was therefore completely protected. Demographic Characteristics of respondents Respondents were told that all information was confidential and that no one could access it and in any way know the identity of the respondents. The questionnaire was not long to increase the response rate. 281 questionnaires were collected and 14 incomplete questionnaires were rejected, leaving 267 valid questionnaires to be used for statistical analysis. Respondent-related demographic data are shown in Table 3. The percentage of males participating in the survey and using the XBRL tool was higher than females, at a rate of 70.4% to 29.6%, respectively. The most common age group for the use of XBRL tools was between 17 and 34. Most of the survey respondents and users of XBRL tools were holders of a bachelor's degree. 59.6% described themselves as accountants 70.8%. XBRL is a newcomer in Saudi Arabia. 74.9 of the participants had a two-year period of use of XBRL (Table 4). Measurement Model Evaluation The factor analysis and reliability test had been carried. The rotated factor matrix that resulted from varimax rotated the principal axis factor of the independent and dependent variables, 1.0 eigenvalue cut-off scale was used. It suggests that all factors with eigenvalues > 1.0 are chosen. As presented in Table 4, indicating that three independent, one mediating variable and one dependent variable had been selected, in addition to listing their factor loadings (Table 5). The factor loadings for the information quality ranged from 0.73 to 0.90, for the effort expectancy from .667 to .763, for satisfaction from .679 to .749, for effort expectancy from .671 to .836, for trust from .578 to 0.86 and for intention .908 to .925. As all factor loadings were found to be at an acceptable and qualified further analysis. Construct reliability is used to evaluate the factor that is evaluated by its items using Cronbach's alpha and CR. The internal reliability was evaluated by examining the Cronbach's alpha and composite reliability (CR) values for all constructs. To measure the CV, the average variance extracted (AVE) was measured. The analysis results are presented in Table 4. The findings refer that estimating Cronbach's α which varied between 0.73 and 0.90 and CR started from .81 for effort expectancy to .91 for intention. Every item had been evaluated separately to ensure item reliability and convergent validity. As noted from the results, the values exceeded the cut-off point specified in this paper, which is 0.7 (Hair Jr et al., 2010). This, in turn, reflects high internal reliability, consistent with previous studies. Finally, with values exceeding the recommended values (.5), the AVE values for each factor in the model ranged from 1.09 to 1.41. In Table 4, the evaluated factor loadings range started from .578 to .925, and were statistically significant (p < 0.05). Model Fit Measures and Cutoff Criteria Reviewing Fit statistical data listed in Table 6, it has been found that all fit statistics suggest an excellent fit. Core fit statistics showed a Chi-square value equivalent to 276.851, 0.001 as p-value, 276.851 as CMIN, 176 as DF, 1.573 as CMIN / DF, 0.952 as CFI, 0.074 as SRMR, 0.046 as RMSEA, and 0.707 as PClose (Table 5). All such numbers suggest an excellent match according to the generally proposed criteria (Gaskin & Lim, 2016). In order to determine the viability of their measured values, it is essential to estimate the fitness of the different parameters for the research model in order to start with the expansion of the expected estimations of the parameters. Generally, the estimations of the parameter show that if the model is consistent with the underlying theory. According to Hu and Bentler (1999), in the analysis of covariance structure the cutoff criteria for fit indexes (Table 5) are a combination of CFI>0.95 and SRMR<0.08, add the RMSEA<0.06 to further solidify evidence. Results showed that the estimates were excellent (Table 5), based on the criteria proposed by Hu and Bentler (1999), this refers to the fact that the model is fit for this research. The critical ratio and significance of path coefficient, when the critical ratio (CR) is > 1.96 for a regression weight for a specific hypothesis, the path will be significant at the .05 level or less, this refers that estimated path parameter for such hypothesis is significant on this path. Tucker-Lewis coefficient measure also estimated, .943 as TLI. Tucker-Lewis coefficient was discussed in the context of analysis of moment structures and is known as the Bentler-Bonett non-normed fit index (NNFI) (Bentler, 1990). The typical range for TLI lies between zero and one, but it is not limited to that range. TLI values close to one indicate a very good fit. Statistical Findings and Hypotheses Testing PLS-SEM regression was used to examine the assumed effect in the model for Trust, effort expectancy and effort expectancy factors on the behavioural intentions of users to use XBRL tools (Fig. 2). For example, to examine the moderating effect in the model, effort expectancy, effort expectancy, and trust as predictors and satisfaction as a mediating factor to predict the behaviour intentions of users to use XBRL tools. After establishing an acceptable model for this research, the researcher assessed the structural model explaining the path diagram in Fig. 2. Table 7 reports the regression weights. The SEM analysis results indicate that H1, H5, H6 and H7 have been supported (Table 7). To test the statistical significance of the parameter estimates from SEM, the test statistic is the critical ratio (C.R) and p-values. The probability of getting a critical ratio as large as the critical ratio in absolute value is less than p-value. In other words, the regression weight for information in the prediction of specific factor is significantly different from zero at the p-value level (two-tailed). Concerning a significance level of 0.05 or less, the critical ratio (C.R) requires being greater than ±1.96 (Hair Jr et al., 2010). Any critical ratio level below this indicates that the parameter can be identified insignificant for the research model. Fig. 2. Research Model Estimating trust impact on intent (H1) was.191 when trust increases by 1. The intent increases by.191 (with C.R. = 2.695; p<0.001), which was significant, and this result is interesting (Table 6). It is consistent with the suggestion made by Carter et al. (2011) in the context of the exchange of sensitive data. XBRL tools are used to process sensitive financial data for disclosure or analysis. If the trust in the XBRL tools is not high, it will not be trusted, and this makes sense for the rational user. Often, trust (e.g. accuracy, error-free, up-to-date, certified) is helpful for users of XBRL tools to submit valid XBRL instance documents or to make more reliable alternative comparisons between financial statements. Estimating to impact trust (H2) on satisfaction was 0.02 when Interactivity goes up by 1, E-trust goes up by 0.02 (with C.R. =0.632; p=0.527), which was not significant (Table 6). Alfin et al. (2013) have shown that trust has no significant impact on satisfaction. XBRL has received considerable attention in recent years. Results show that trust in XBRL tools exceeds satisfaction and is a priority for users when deciding to use it. Users will use the licensed tool regardless of their satisfaction with the XBRL tools. To summarize the results, it has been shown that there is a direct effect of trust on users' intention to use XBRL tools, but not an indirect effect through satisfaction as a mediating factor. The third hypothesis (H3) has been rejected, the estimating was -0.037 with C.R. =-0.272and p=0.786 and the fifth (H5) accepted, estimating was 0.257with C.R. =04.109 and p<0.001 (Table 6). In other words, in the research model of this study, we may consider a complete mediation of satisfaction in the relationship between effort expectancy and intention of users to use XBRL tools, which is consistent with some findings (Miladinovic & Hong, 2016;Handoko et al., 2018;Chao, 2019). At the other hand, the effort expectancy is supposed to have an indirect effect on the intention by satisfaction. The results of this study are also consistent with those of the previous related research (Chao, 2019). The value of satisfaction during effort expectancy has a major effect on XBRL tools developers. Therefore, the creation and review of XBRL software for the tagging and analysis of financial statements will rely more on the effort expectancy of the XBRL tools. Because by using XBRL tools, they submit financial information or make a financial transaction that has an impact on their financial future. It may also make errors if the XBRL method is difficult to use, leading to financial losses and a waste of time. These findings indicate that effort expectancy is a critical factor in terms of satisfaction with the tools used, especially in the early stages of the use of XBRL tools. Apparently, this is a new contribution to the existing theory of acceptance of XBRL tools. The fourth hypothesis (H4) was rejected and the sixth (H6) accepted. Furthermore, a complete mediation of satisfaction in the relationship between performance expectancy and intention of users to use XBRL tools could be found in the research model of this study. Nevertheless, the relationship between performance expectancy and intention of users to use XBRL tools (H4) is not significant (C.R. = -0.565 and p=0.572), which means that there is no intention effect on performance expectancy (Table 6). However, the relationship between performance expectancy and intention of users to use XBRL tools (H6) is significant, with an estimation of 0.209 for C.R. = 4.25 and p<0.001, respectively. It may be due to the fact that users of XBRL tools will be comfortable with their use of the tool and that it will help them improve their job performance before they plan to use XBRL tools. This is illustrated in the H6 by a full mediation of satisfaction between performance expectations and the intention of users to use XBRL tools. The results are consistent with previous studies (Chao, 2019). Then, performance expectancy can have an indirect impact on the decision to use it by satisfaction. Basically, the purpose of XBRL tools is to conduct different activities, whether financial disclosure or data analysis. Thus, if the user is not satisfied with the expected results of this tool, the user will not use it. The final purpose of the current study is to investigate the complete mediation of satisfaction with the intention of users to use XBRL tools. Estimated effect satisfaction on intention (H7) was 1.08 when (with C.R. = 4.468; p<0.001) (Table 6), which was very significant (Bao, 2015;DeLone & McLean, 2016;Chao, 2019). This is an interesting result; it appears to be a new confirmation of what exists in the theory. In previous work, Alfin et al. (2013) found that trust had led to the intention of users to use XBRL tools. The Predictive and Explanatory Power of the Model The results of this study provide support for the research model. The major aim for satisfaction to use as a mediating variable of the intention of users to use XBRL tools and it is the ability to explain and predict user intention, based on merely three theoretical factors, namely, trust, performance expectancy and effort expectancy. Four hypotheses were supported by these interesting findings. Trust has a direct impact on the intention of users to use XBRL tools, not an indirect impact on satisfaction. Besides, performance expectations and effort expectations have shown a positive impact on satisfaction as an indirect impact on the intention of users to use XBRL tools. Regarding the predictive power of satisfaction, the results showed a strong interpretation of the intention of users to use XBRL tools by satisfaction. Such effects provide a number of essential insights into the determinants of the use of XBRL tools. First, its satisfaction with XBRL tools is the major determinant of the user's intent. Second, the expectation of effort is a significant, highly indirect determinant of intent. Third, the expectation of performance is important as an indirect determinant of intent. Last but not least, trust is a highly direct determinant of intent. In general, satisfaction as a mediating variable between the variable and intention of users to use XBRL tools, rather than having a direct impact from the variables of the independents. Additionally, future research may consider including such an alternative approach. Discussion The purpose of this study was to identify factors that influence users' behavioural intentions to use XBRL tools. According to the researcher's experience, the factors influencing the use of analytical instruments have not been examined before or at least in this way. Therefore, this research model incorporates specific elements such as trust and satisfaction in extending the original model to give a stronger predictive capacity to the model. The results of the analysis showed the factors affecting the behavioural intention of users to use XBRL tools. Satisfaction came in the first order in the effect, then immediately followed by effort expectancy the third arrangement was performance expectancy. The last order in effect was for trust. These results can be explained as follows. Four variables clarified the variation in behavioural intentions of the XPRL tools. Where trust had a direct effect on the behavioural intentions of using the XPRL tools, and these results were consistent with the previous findings as mentioned above. The results supported trust and satisfaction with the ability to be the main variables in the model. The nature of the relationship lies in the disappearance of the trust gap between users and developers, the existence of goodwill messages between them, and the disappearance of the factor of mistrust and fear among users, which leads them to form intent directly without the need for variable mediation, as the results have shown. Trust of the user in one of the XPRL tools does not mean his utter intention of users to use XBRL tools to use this tool. Instead, trust acts here as a filter for preferred tools in order to make more evaluations of them, such as performance expectancy and effort expectancy. When satisfaction is combined with trust, the aim is to use one of the XPRL tools in question. Satisfaction is not self-constructed, but satisfaction is created as a consequence of the knowledge or preconceived notion of the performance expectancy and effort expectancy of the tool. Therefore, in the beginning, and then later, they are happy with it. The consumer has often reported outcomes of desired satisfaction in order to prevent the occurrence of mistakes and waste of time or the ability to achieve constructive and motivating outcomes as a result of the use of the tool in question. For example, the consumer does not seek the approval of the developer. The willingness of the user to use a resource to achieve a particular task is measured by trust-based intentions and experience-based satisfaction. As assumed in the original model, expectation of performance and expected effort confirmed the existence of a strong influence on users' behavioural intentions to use XBRL tools through satisfaction. In addition, compatibility with other research was achieved with these famous variables. Among the outstanding results in this model is the role of the variable expectation of performance, and the expectation of effort is their relationship to satisfaction. This means that the use of XBRL tools is associated with the user's satisfaction of these tools whenever the users use them in their work. This result was also consistent, as with previous results, with a group of researchers in this context. Based on the findings from this study, XBRL tools are becoming an increasingly crucial tool for companies to disclose. When users find that XBRL tools have low expectations of effort and high expectations of performance. It is expected to consider the tool to improve their disclosure and decision, enhance their satisfaction towards XBRL tools and user behavioural intentions to use XBRL tools. For example, providing evidence explained in detail with adding pictures for each step in addition to providing comprehensive videos explaining the use of the XBRL tools in detail, and this is available to users before proceeding to purchase the XBRL tools. This would improve the satisfaction of potential users of the XBRL tools. Limitations and Future Research Due to the scarcity of previous studies which have examined the subject in general and in Saudi Arabia in particular. The study focuses on all users of XBRL tools in general, and studies can focus on independent auditors, accountants, analysts, and even business owners. It is also claimed that the time after Coronavirus (COVID-19) was not the same as before Coronavirus (COVID-19) (pre and post corona) the study was strongly affected by the isolation imposed on citizens, including researchers and respondents in this study. The online questionnaire was also used and this form of questionnaire allows the respondents to choose themselves to be a part of the survey. This does not make it possible to generalise the results. Potential studies are therefore advised to use paper questionnaires and random surveys over a long period, as opinions that shift due to the accumulation of respondents' experiences. Conclusion For the purposes of this study, a detailed model was built to describe user behavioural intentions to use XBRL resources where the dependency on the UTAUT model was depended on. Some of the variables that were mainly present in this model have been abandoned and, in exchange, new variables have been introduced that match the adoption of the XBRL tools. The model consists of five variables, the effort expectancy, performance expectancy, trust, satisfaction, and behavioural intentions of users to use the XBRL tools. After the questionnaire was developed for study purposes, it was used to collect data via an online questionnaire and 267 questionnaires were analysed. The results of the research showed that the study model had high internal reliability and accuracy, demonstrating the model's ability to clarify. The results showed that the variables applied to the model, such as trust and satisfaction, had a strong impact on the behavioural intentions of using the XBRL tools. In addition to the strong influence, effort expectancy and effort expectancy had on users' behavioural intentions to use XBRL tools through satisfaction. This study just knocks on the door, XBRL is still in progress and the XBRL tools that support its use are in progress. Understanding the factors that influence users' behavioural intentions to use XBRL tools could be helpful for potential developers of XBRL tools, as may be useful from the academic point of view for researchers interested in this matter.	9,052.4	2021-01-01T00:00:00.000	[ "Computer Science" ]
An Insight on Image Annotation Approaches and their Performances : Image Annotation (IA) followed by Image Retrieval (IR) plays a significant role in today’s computer vision world. As the manual IA is a tedious and time-consuming process, the automated IA became very predominant in the computer vision applications. IA deals with the assigning of meaningful labels to various objects in the image. The objective of this article is to represent the various IA approaches adopted in the last decade. Observation of the existing IA methods and their performances leads to identify the pitfalls the existing approaches. Few approaches used standard datasets and images downloaded from internet to evaluate the performance of the Image Annotation. Introduction Globally, automation is inevitable in every domain.In the perspective of computer technology, the boundaries of application keep prolonging.Nevertheless, the utility of the concept is definite.The Information Era provides huge data to humankind.The blend of such data with Artificial Intelligence boomed out with several vital applications like augmented reality, automatic speech recognition, and neural machine translation, image processing, health monitoring system, autonomous vehicles, facial recognition, unmanned drones and others.Image annotation, one of the image processing techniques, labels and classifies the images based on annotation tool or text by identifying the features considering the ultimate purpose of the model.The image annotation is an automatic system thus adding metadata to the dataset.Image annotation (IA) is also termed as data labeling, tagging, processing or transcribing [2] 2 Role of Image annotation IA plays a vital role in formulating the training data regarding computer vision and its applications.That is, to make the machine to recognize the surrounding objects, annotated images becomes mandatory for the machine learning (ML) algorithms or approaches to see the real world objects and train accordingly.According to the statement, 'the performance of Artificial Intelligence and its applications relies on the training data and its accuracy', labels are used to provide information about the various objects to computer vision (CV) model [1].Usually, the labels are pre-determined by the CV scientists or engineers.Later, based on the annotated data, the algorithms learn and recognize the identical patterns in the new data.The objective of IA is to allocate or assign the task specific and relevant labels to the objects, things or persons in the images.The possible labels include text-based (classes), localization of objects (using Bounding boxes) and even sometimes, the pixel-based labels. To annotate the images, the following are required (1) images (2) person to annotate the images and (3) the platform for image annotation. Following are the various techniques where IA plays a vital role in object recognition. (a) Two dimensional Bounding box where a box is created over the region of the interest (usually an object) in the image.For example, if the image has objects such as bicycles, person, cars then the boxes are drawn over those objects and subsequently the annotator performs the labeling of those boxes.(b) Three dimensional Bounding Boxes also represented as Cuboid-based labeling, where a box is created over the region of the interest (referred as object) in the image with its depth representations.(c) IA using Polygon Annotation (PA) where objects with irregular shaped and irregular sized objects in the images are labeled.Here, as the name indicates, the polygons are formed over the objects such that, the object's location and volume are determined in the images.(d) Poly lines based IA is adopted to annotate the splines, boundaries and lines in the images.Applications of poly line based IA includes, trajectories planning, annotating of power lines, road lanes, side walls and training of autonomous vehicles route (particularly warehouse robots to place the object or items in a conveyor belt).(e) Semantic Segmentation (SS) is a type of IA, where a precise and specific tag is specified for every pixel(s) in an image.Unlike other methods of IA, where object's boundaries (alone) or edges are considered.SS is used where pixel-wise annotation is required.For example, the environmental scenarios are made observed by autonomous vehicles and robots using the SS based IA.The sub-categories of SS are Instance Segmentation and Panoptic Segmentation.Instance Segmentation deals with the identification of every instances of every object at the pixel level in an image.On the other hand, panoptic segmentation integrates the functionalities of SS and Instance Segmentation, where every objects instance were identified localized and segmented after assigning the corresponding class labels.(f) Keypoint based IA is used to figure out the object's boundaries along with its position and size.For example, during the annotation of car, the objects such as mirrors, wheels, headlights are determined.While annotating the human being, the various parts namely head, eyes, nose, mouth, shoulders, arms, anklets, knees and foot are identified. To summarize, the applications of IA are not limited to image or object classification, image or object detection, and image or object segmentation with the corresponding instances.The following section depicts the various existing IA approaches and their performances. Theodosiou and Tsapatsoulis [1] analyzed Image annotation technique in terms of content, lexicon and annotation.The paper examined the factors influencing the quality of annotation by means of crowdsource platform.The examination was carried out using free keywords, preselected keywords and hierarchical vocabulary words on 500 images -an dataset of from Commandaria collections.Among the investigation, hierarchical vocabulary worked effectively and further, annotation was not based on the concepts which lead to inconsistency but it was a common problem. Sarin, Fahrmair, Wagner, and Kameyama [2] leveraged features of digital image from the salient regions and background to achieve automatic image annotation.Initially the salient regions and background are estranged without using prior knowledge from the datasets Corel5K and ESP game datasets.Subsequently, every estranged region of the digital image was compared to the whole digital image by computing the sign test with pvalue < 0.05.The performance of the approach was proved by comparing the result with other state-of-the-art techniques. Sangeetha, Anandakumar and Bharathi [3] surveyed the optimization techniques on Image annotation and retrieval.A detailed and comparative analysis was done on optimization algorithms with different feature selection algorithms and classifiers.Feature selection algorithms like Histogram analysis, Discrete Wavelet Transform, Discrete Cosine Transform in combination with classifiers such as K-Means, KNN, Fuzzy Feed forward Neural Network, SVM, Euclidean Distance and Similarity evaluation.To achieve maximum optimization, the feature weights were optimized through algorithms like Particle Swarm Optimization (PSO), Genetic algorithm (GA) and Firefly Algorithm (FA).From the survey, PSO based feature selection technique yielded fine results. Khainga and Yu [4] studied step by step methods in Deep Learning Model (DLM) based Image annotation techniques.The bottom-up approach of the image annotation that involve steps such identification of objects, words, sentences using ML were studied in-depth.The ML algorithms CNN, Recurrent NN and Long and Short Term Memory were analyzed in detail.Further, the attributes, image size, and sample size of the datasets -MSCOCO, FLICKR 8K, and FLICKR 30K were explained.Finally, the performance evaluation metrics such as Bilingual Evaluation Understudy (BLEU), Recall-Oriented Understudy for Gisting Evaluation (ROUGE), Metric for Evaluation based Image Description Evaluation (METEOR), Consensus-based Image Description Evaluation (CIDEr), and Semantic Propositional Image Caption Evaluation (SPICE) which compute the similarity index amid the ground truth and machine generated results were discussed in detail. Ashley, Barber, Flickner, Hafner, Lee, Niblack and Petkovic [5] developed a prototype system -Query By Image Content (QBIC) which contain two phases (a) query by color drawing (b) identification of image objects.The semi-automatic techniques such as Floodfill algorithm and Snake based Edge Following algorithm eased identification of images and retrieval of images from the database population. Bouyerbou, Oukid, Benblidia, and Bechkoum [6] discussed hybrid image representation techniquesblock, feature and region based automatic image annotation.The hybrid -global and local features considered for the study though used the benefits of both the features, revealed that images were represented clearly in spite of complexity in the scenario and multiple semantic meanings were explored from a single image.For the effective representation, the combination of features selected must be perfect. Caicedo, González and Romero [7] worked on content-based histopathological image retrieval using kernel and semantic annotation methods.The automatic image annotation involved extraction of multiple visual features from input image, representing data with all possible visual features using kernel function and detection of histopathological content using the representation.Finally, the results were used to explore alike images while annotation or just in indexing the retrieval work of input images.The retrieval performance of kernel function in terms of Precision and Recall were plotted to show the significance of visual retrieval especially SIFT.On comparison with the visual search, the acclaimed kernel based semantic technique depicted 57% more accuracy in identifying histopathological content. Bouchakwa, Ayadi and Amous [8] reviewed Visual Content based and Users' tags based Image annotation techniques.In Visual content-based method, both high and low level feature based annotation and the semantic gap in describing the images were analyzed.Similarly, semantic relationship between tags and structured knowledge resources in Users' tag based annotation were discussed.However, Region based image representation (RBIR) the feature extraction methods like low level feature extraction -color, shape and spatial relationships, feature descriptors -SIFT, SURF, GIST and deeper features were discussed for segmentation.Further, the in-depth study of Semantic learning included Supervised -KNN, DT, SVM and Bayesian Network, Unsupervised -Clustering, Hidden Markov and Neural Network along with Deep Learning-Convolutional Neural Network (CNN).The concept of Image captioning that involve object detection both one stage and two stage detectors and the algorithms related to it were investigated. Kılınc, and Alpkocak [9] retrieved annotation based images from the web by expansion and reranking approach.The preprocessed images were expanded in three phases WordNet (Miller, 1990) for both Document Expansion (DE) and Query Expansion (QE) phases.The results were narrowed down through similarity score and based on Cover Coefficient based Clustering (C3M) the final similarity score was evaluated.When investigated on web images, the sixth run of reranking exhibited best results with MAP and P@5 values are 0.2397and 0.5156 respectively. Chen, Zhu, Wang, Jin and Yu [10] annotated images by applying tag candidate retrieval and multi-facet annotation technique.The deployment of content based indexing and codebook using concepts eradicate noise issues in the images.Moreover, the relationships in-between facets pictured out in joint feature map while tag graph depicts tags in every annotation.The structured learning concept when examined on Flickr images the performance metrics Precision, Recall and F1 score showed 33% more improvement than other methods STRUCT, GIST, SHAPE and SIFT.Efficiency was also proved by comparing performance metrics with that of three semantic tag features such as co-occurrence (TC), com-monality (CT), and specialization (ST). Deselaers, Deserno and Müller [11] reviewed and discussed the results of automatic image annotation techniques in ImageCLEF2007.Among the 12000 images from RWTH Aachen University Hospital, 11000 images were used for training and 1000 images for testing.The IRMA code and the subsequent hierarchical classification annotated the images ranking 7. Gao, Yin and Uozumi [12] developed a hierarchical Image annotation technique by classifying the multiple labels through SVM and fine tuning the annotation by using Expectation Maximization (EM) algorithm.The 1300 images were pre-processed by semantic keywords into several labels, and the images were extracted Gaussian mixture model followed subsequently by feature extraction.The roughly annotated images by SVM were fine tuned by EM before evaluating the accuracy metrics.The finely corrected annotation using Contextual relationship involved 5 fold cross validation to deduce the errors. Guo, Jiang, Lin and Yao [13] combined Learning Vector Quantization (LVQ) technique and SVM classifier to gear up the annotation process without losing its accuracy.The drawback of SVM using extreme training samples was overthrown with Self Organizing Map and Affinity Propagation algorithm.By doing so, acceleration geared and cost was minimized as only representative samples were used.On par with other methods such as SVM with actual dataset, traditional SOM based LVQ with SVM, Quadratic Discriminant Analysis (QDA) classifier with AP based LVQ and QDA with actual data, the combined SOM+AP based LVQ with SVM performed better without losing accuracy. Harada, Nakayama, Kuniyoshi and Otsu [14] developed a novel approach to annotate and retrieve weakly labeled images by amalgamating Higher-order Local Auto-Correlation (HLAC) features and canonical correlation analysis.The well-defined intrinsic space between images in conceptual learning enhanced faster and accurate results.The performance of the approach was compared with JEC annotation technique to prove the superiority. Hatem and Rady [15] investigated different feature dimensionality reduction techniques to retrieve and annotate 120 sport images from the Leeds Sports Pose sport dataset.While JSEG algorithm segmented the images, 10 fold cross validation for classification accuracy and performance metrics were evaluated to prove the performance of LSA.The authors put forth a comparative study of SVM and other reduction methods such as Information Gain, Gain Ratio, Chi-Square, and Latent Semantic Analysis (LSA), in terms of accuracy, integrated LSA depicted 96% while SVM showed 76.4%. Weston, Bengio and Usunier [16] acclaim ML algorithms for image annotation that can scale testing and training and quantify less memory usage.Such model optimizes the precision at k using Weighted Approximate-Rank Pairwise loss (WARP) where semantic learning of both words and image were possible.The results were evaluated by sibling precision metric and MAP algorithm to prove the novelty. Hu, Shao and Guo [17] investigated the visual feature extraction methods namely Discrete Cosine Transform (DCT), Gabor Transform (GT) and Discrete Wavelet Transform (DWT) for annotating the images.The low level features extracted through afore mentioned techniques, high level semantic words were mapped for image annotation.The performance analysis of 2000 images from VOC2008 dataset with DCT, DWT and GT exhibited DCT was more efficient for Gaussian mixture model in automatic image annotation. Ismail, Alfaraj and Bchir [18] used PCMRM framework relied on visually similar image regions into homogeneous clusters, to evaluate the joint distribution of textual keywords and images.The results were compared with other state-of-the-art algorithms to show the superiority. Tiwari and Kamde [19] annotated and retrieved images with the aid of contextual information in the images.The entire model included four phases such as (a) Contextual Information Extraction (b) Text Processing (c) Term weighting (d) Image Retrieval.Further, the evaluation of the model with other image contextual extraction techniques like the N-Terms window (NT) extractor, the paragraph (PAR) extractor, the VIPS-based extractor (VIPS), the Monash (MON) extractor, and the Full-Text (FULL) extractor. Wang, Dawood, Yin, and Guo [20] investigated in detail the feature mapping techniques such as homogeneous and discriminative tree based methods using the FastTag algorithm.The investigation was examined in three datasets namely Corel5K, ESP Game and IAPRTC-12.5.Based on intensive investigation and tabulated results, the homogeneous feature mapping technique with X2 kernel performed better in precision when combined with the FastTag algorithm with longer operation time, in contrast to LDM with less execution time and low precision value. Li, Dawood, and Guo [21] compared several Linear Dimensionality Reduction (LDR) methods such as Principal Component Analysis (PCA), Random Projections (RP), and Locality Preserving Projections (LPP).With FastTag algorithm framework LDR methods, the efficiency, effectiveness and also memory usage were compared using Corel5k, IAPRTC-12 and ESP game datasets.The execution time taken by all LDRs were same for small dataset while PCA and LPP prolonged the execution time during huge data.RP performed better than other LDRs, irrespective of precision value and data density.Lee and Wang [22] deployed feature extraction methods to annotate images using text mining technique based on geographical location.Both labeled and unlabeled images of sample size 3600 from Tourism Bureau Kaohsiung website, Flickr, and blogs were investigated for the study. Tang, Zha, Tao and Chua [23] annotated multi-label images through Semantic-Gap-Oriented Active Learning.The combination of semantic gap measure in sample selection strategy improved the effectiveness and minimized manual intervention.Moreover, the quantitative measurement of the semantic gap by correlation sparse-graph in multi-labeled images improved the effectiveness in image annotation. Table 1 summarizes few of the techniques, datasets and their performances of the existing IA approaches. Applications of Image Annotation Image annotation is a process in Machine Learning and Artificial Intelligence where the images are labeled and classified exploiting texts or annotation tools through highlighting or identifying the features by recognizing them automatically.To recognize the objects of interest successfully, they are annotated using the metadata added to easily describe them.When huge data of same type are fed, then it is termed as trained model to identify the objects in real time.Summary of findings from the existing approaches are as follows: • IA can be accomplished in terms of content, lexicon and annotations • Optimization technique with feature selection has significant performance in annotating the images.Various approaches adopted in the existing methods for IA were QBIC, CNN, DLM, SIFT, SURF, GIST, LVQ, QDA, HLAC, LSA, DCT, DWT, GT, LDR and PCA • Hybrid feature extraction methods extracted both local and global image features which enhanced ----of IA process.• Various images from standard datasets and downloaded from internet were used to annotate the images. • Clustering of similar image features (such as texture, shape and color), noise reduction, optimization techniques and fusion of existing methods resulted in the improvement of annotation process. Conclusion This paper attempted to focus on various existing IA approaches in the last decade.Upon observing the performance of existing methods, the following were concluded: • Integrating the image's features namely texture, shape and color, forms the combine feature vectors for significant representation of images.• Denoising and hybrid image feature extraction has significant performance in labeling process. • Fusion of existing feature extraction approaches and optimization techniques made precise representation of image features. Even though this paper focused on various IA approaches and their performances, does not represent the mechanisms adopted in the concerned approaches.However, the study on various approaches led to determine the processing and pitfalls of existing IA approaches along with the need for hybrid framework for clustering and feature extraction processes. Figures 1 (Figure 1 (Figure 1 Figure 1 (Figure 1 (Figure 1 Figures1 (a) to (e) illustrate the various existing IA approaches for labeling the objects in the images summarized from[24]	4,110.4	2021-04-28T00:00:00.000	[ "Computer Science" ]
GEDpm-cg: Genome Editing Automated Design Platform for Point Mutation Construction in Corynebacterium glutamicum Advances in robotic system-assisted genome editing techniques and computer-aided design tools have significantly facilitated the development of microbial cell factories. Although multiple separate software solutions are available for vector DNA assembly, genome editing, and verification, by far there is still a lack of complete tool which can provide a one-stop service for the entire genome modification process. This makes the design of numerous genetic modifications, especially the construction of mutations that require strictly precise genetic manipulation, a laborious, time-consuming and error-prone process. Here, we developed a free online tool called GEDpm-cg for the design of genomic point mutations in C. glutamicum. The suicide plasmid-mediated counter-selection point mutation editing method and the overlap-based DNA assembly method were selected to ensure the editability of any single nucleotide at any locus in the C. glutamicum chromosome. Primers required for both DNA assembly of the vector for genetic modification and sequencing verification were provided as design results to meet all the experimental needs. An in-silico design task of over 10,000 single point mutations can be completed in 5 min. Finally, three independent point mutations were successfully constructed in C. glutamicum guided by GEDpm-cg, which confirms that the in-silico design results could accurately and seamlessly be bridged with in vivo or in vitro experiments. We believe this platform will provide a user-friendly, powerful and flexible tool for large-scale mutation analysis in the industrial workhorse C. glutamicum via robotic/software-assisted systems. INTRODUCTION Industrial biomanufacturing, using well-tailored microbial cell factories with economically competitive titers, synthesis rates and yields (TRY), offers a potentially green and economical alternative to current petroleum-based chemical synthesis (Clomburg et al., 2017). Corynebacterium glutamicum, the famous industrial workhorse for amino acid production with a current output of over 6 million tons per year (Lee et al., 2016), is increasingly being adapted as a promising chassis for the biosynthesis of other compounds (Becker et al., 2018). However, most microorganisms, including the industrial C. glutamicum, strains have not evolved to naturally and/or efficiently produce the majority of petrochemical compounds (Lee et al., 2012). Despite the substantial rational engineering efforts devoted to developing efficient cell factories, it is still arduous to achieve competitive TRY values due to our current limited understanding of cellular metabolism (Nielsen and Keasling, 2016;Ding et al., 2020). Instead, most industrial workhorses are developed without in-depth genetic knowledge by random mutagenesis strategies such as adaptive laboratory evolution and chemical/physical mutagenesis (Ikeda, 2003;Zhang et al., 2014;Sandberg et al., 2019). Over the past few decades, random mutagenesis strategies combined with applicable selection methods have led to the development of various industrial C. glutamicum strains as well as a valuable trove of genetic diversity (Ikeda, 2003;Zhang et al., 2018;Stella et al., 2019). With the development of genome sequencing and genetic engineering tools, novel synthetic biology elements such as enzyme variants have been identified through reverse engineering (Ikeda and Takeno, 2020), which can motivate further innovation in the development of industrial C. glutamicum strains. Since point mutations (single nucleotide substitutions, insertions or deletions) are the predominant mutation type identified in industrial/evolved strains (Kvitek and Sherlock, 2013;Lang et al., 2013), large-scale point mutation analysis is highly desired for further understanding the genetic basis responsible for the evolved C. glutamicum phenotypes (Bailey et al., 2002;Nielsen and Keasling, 2016). However, one major issue in the current point mutation analysis is the genetic modification of cells to introduce the enormous numbers of point mutations needed for highthroughput screening, which appears to be an impossible task for laboratory biologists. For the case of a saturation mutation library targeting the 10 bp sequence of the core region of a bacterial promoter (−35/−10 region), the strain library size can reach 4^10, nearly one million, which is impossible to accomplish with human labor alone. Recently, a robotic system-assisted C. glutamicum automation genome editing platform (MACBETH) has been developed, with the capacity to generate thousands of single nucleotide mutant strains per month (Wang Y. et al., 2018), which opens the possibility for future robot-assisted large-scale point mutation editing. Unfortunately, as a CRISPR/Casdeaminase-mediated base editing platform, MACBETH was unable to edit all the nucleotides of interest due to the limitation of genome-targeting scope, editing window, and base transition capability (Wang Y. et al., 2019). Genome point mutation editing is a much more precise genetic modification than gene deletion or insertion. Although there are many genome editing techniques, which can be simply divided into ones based on homologous-recombination (HR) and non-homologous end-joining (NHEJ), few are suitable for strictly precise genomic point mutation editing (Wang et al., 2021). The nuclease/integrase/transposon-mediated NHEJ system, which requires a specific recognition site at the targeted chromosomal locus in advance and will inevitably generate chromosomal scars (such as the loxP or attB sites) after genome editing (Hu et al., 2014;Marques et al., 2020), is obviously inapplicable for point mutation editing. The RNA-guided CRISPR/Cas-based HR systems can provide scarless chromosomal modifications (Jiang et al., 2017;Wang et al., 2018a). However, additional mutations in protospacer and PAM regions are usually prerequisite for avoiding re-cutting by endonucleases, and editable genome regions are restricted due to the limited availability of guide RNAs (Wang T. et al., 2019), which also limits its application in point mutation editing. The RecT-mediated ssDNA/dsDNA HR system, which can support the editing of any nucleotide of interest without the need to repeatedly construct editing vectors (Binder et al., 2013), seems like an ideal technique. However, the ssDNA/dsDNA electro-transformation efficiency in Gram-positive C. glutamicum might be problematic (Ruan et al., 2015), especially for the industrial recombinant strains. In addition, due to the relatively short homologous arms in ssDNA/dsDNA, the off-target risk is high, especially for point mutations inside sequences with additional copies in the chromosome. In the counter-marker-assisted HR system, which is based on two rounds of single crossover HR (Schäfer et al., 1994), any nucleotide in the genome is theoretically editable without risk of introducing additional mutations/scars. The application of an editing vector carrying relatively long homologous arms can not only significantly improve the electro-transformation efficiency, but also lower the off-target risk caused by mismatching. In addition, the editing vector can be used repeatedly, which reduces the cost of vector DNA assembly. Although the efficiency of HR is relatively low, conditional lethality mediated by counterselectable markers, such as the sucrose-lethal gene sacB (Schäfer et al., 1994) or streptomycin-sensitive gene rpsL (Kim et al., 2011), can ensure the occurrence of two rounds of singlecrossover HR. Therefore, the counter-marker-assisted HR system is the most promising chromosomal point mutation technique for C. glutamicum. The counter-marker HR mediated genome editing system is mostly implemented using non-replicating suicide plasmids (Wang et al., 2021), among which the suicide vector pK18mobsacB (GenBank: FJ437239.1) is the most widely used in C. glutamicum (Schäfer et al., 1994). For point mutation editing based on the pK18mobsacB vector, a pK18mobsacB-derived vector and a total of two rounds of crossover HR and mutant isolation are required ( Figure 1). Firstly, the upstream homologous arm (UHA) and the downstream homologous arm (DHA), which contain the corresponding point mutation, are amplified from the chromosome of C. glutamicum by primer pairs 1/2 and 3/4 (named as primer-1/2/3/4), respectively. Then, the homologous arms as well as the linearized pK18mobsacB vector fragment are assembled into the pK18mobsacB-derived suicide vector, containing homologous arms and the desired point mutation. Subsequently, the resulting vector is electroporated into the cells and integrated into the chromosome in the 1 st -round of single crossover, and the resulting transformants are selected in the 1 st -round isolation based on the positive selection marker, the kanamycin resistance gene kan R . Next, the integrated vector is eliminated in the 2 ndround of single crossover and the transformants are selected in the 2 nd -round of isolation based on a negative selection marker, the sucrose-lethal gene sacB. Finally, the correct mutants Frontiers in Bioengineering and Biotechnology \| www.frontiersin.org October 2021 \| Volume 9 \| Article 768289 containing the point mutation are confirmed by sequencing. Thus, a total of two homologous arms and at least two pairs of primers (primer-1/2/3/4) are required for the whole process of point mutation editing. Despite the development of computer-aided design (CAD) tools for the design of genetic modifications (Kalendar et al., 2011;Appleton et al., 2017a;Appleton et al., 2017b), no CAD tool is currently available for the design of counter-marker-assisted HR editing. Although some tools such as j5 (Hillson et al., 2012), Raven (Appleton et al., 2014) and FastPCR (Kalendar et al., 2017) can assist the design of optimal primers for amplifying UHA/ DHA, laboratory biologists still have to manually extract and input optional templates, which is laborious, time-consuming and error-prone if it needs to be done on a large scale. In addition, the design of primers for sequence verification, one of the most important experimental steps, is often neglected by laboratory FIGURE 1 \| Schematic diagram of GEDpm-cg and the point mutation editing technique based on uploaded sequences. GEDpm-cg is a web-based computeraided design tool for the construction of genomic point mutations in C. glutamicum. When input files containing information of point mutations, vector and targeted genome are uploaded, GEDpm-cg is able to provide precise and high-throughput in-silico designed results for in vitro editing vector DNA assembly and in vivo point mutation editing in C. glutamicum. The design of point mutations by GEDpm-cg can be divided into the overlap-based vector DNA assembly and the suicide plasmid-mediated counter-selection point mutation genome editing. For vector DNA assembly, design tools such as j5 or Raven are able to assist the design of vector DNA assembly for biologists, while no tool is available for the design of suicide plasmid-mediated counter-selection point mutation genome editing. UHA, upstream homologous arm; DHA, downstream homologous arm; green band, homologous end between UHA and vector; yellow band, homologous end between DHA and vector, blue diamond, base before point mutation; red diamond, base after point mutation. Underlined names indicate the easily-neglected verification primers. Frontiers in Bioengineering and Biotechnology \| www.frontiersin.org October 2021 \| Volume 9 \| Article 768289 beginners, which limits the ability to correct unexpected misediting/non-editing. For instance, if adding one pair of verification primers (test-primer-1/2, Figure 1), non-editing failures caused by false-positive transformants (Ma et al., 2015) during the 1 st -round of single crossover and isolation can be avoided. In addition, unexpected mutations located around the ends (∼100 bp) of UHA-DHA cannot be precisely sequenced if simply using primer-1/2 rather than test-primer-1/3 ( Figure 1) for the final sequence verification. Thus, a one-stop, comprehensive CAD tool for the whole design process of counter-marker HR mediated genome editing is highly desirable for automated and high-throughput point mutation editing in C. glutamicum. In this study, to reduce the effort and time needed for point mutation editing design and provide a comprehensive packaged result for laboratory biologists, we developed a user-friendly online tool (Figure 1), named the Genome Editing automated Design platform for point mutation construction in Corynebacterium glutamicum (GEDpm-cg, https://gedpm-cg. biodesign.ac.cn/). The counter-selection HR system (Schäfer et al., 1994;Tan et al., 2012) and the overlap-based assembly method (Casini et al., 2015) were chosen as the loading FIGURE 2 \| Workflow for the design of point mutations in GEDpm-cg. The design of primers and homologous arms is performed by three programs: In-house program, Primer3 and BLAST. Logical judgment flow is divided into detailed steps which are marked with corresponding inputs (blue numbers), parameters (purple numbers) and outputs (green numbers). Moreover, output files linked to experiments are marked with red numbers. First, three input files are checked by the in-house program and BLAST. If the input file does not meet the standards, it should be refined. Otherwise, they will be submitted to the back-end program with the parameters set by the user to go through the design process. The design process consists of three main operations: determination of the templates for primer design by the in-house program, design of primers based on the templates by Primer3 and evaluation of the mis-match likelihood of UHA/DHA by BLAST. Then, the design results will be provided in four output files. "Failed task" and 'Evaluation result' can guide the user to re-optimize the parameters. "Design results" and 'Primer order' are used to guide the experiment. IP, In-house program; P3, Primer3; BL, BLAST; UHA, upstream homologous arm; DHA, downstream homologous arm; PCR, polymerase chain reaction. Frontiers in Bioengineering and Biotechnology \| www.frontiersin.org October 2021 \| Volume 9 \| Article 768289 techniques. Homologous arms and primers required for genetic modification, vector DNA assembly and sequencing verification were provided as design results. Moreover, the GEDpm-cg was built in a novel, entirely serverless architecture, with computing, as well as data storage, done in a serverless manner, which ensures flexibility in allocating computing resources. Finally, to verify the accuracy of design results generated by GEDpm-cg, three independent point mutations were experimentally implemented. GEDpm-cg Service Implementation and Availability GEDpm-cg is written in the Python programming language (https://www.python.org). GEDpm-cg makes external calls to Primer3 (Untergasser et al., 2012) for primer and flanking homology sequence design, and to BLAST (Zhang et al., 2000) for checking redundant mutations inside target sequences from the input file and identifying putative mis-priming and flanking homology sequence incompatibility events. The point mutation design is conducted via the workflow integrating inhouse program, Primer3 and BLAST ( Figure 2). A detailed user manual for GEDpm-cg is provided online (https://gedpm-cg. biodesign.ac.cn/help). GEDpm-cg is freely available for noncommercial (e.g., academic, nonprofit, or governmental) users. The service is available through the public GEDpm-cg webserver (https:// gedpm-cg.biodesign.ac.cn/). Serverless Architecture of GEDpm-cg The serverless architecture of GEDpm-cg allows us to devote more time to core workflows and to build scalable, reliable systems more quickly and easily. We used three-tier architecture to build our website, which is a popular pattern for user-facing applications (Supplementary Figure S1). All tiers that comprise this architecture are deployed on Amazon Web Service, including the front presentation tier, logic computation tier, and data storage tier. The front presentation tier represents the component that users directly interact with (such as a web page, etc.), which is hosted by AWS S3 static website functionality and accelerated by AWS CloudFront. The logic computation tier of our website manages http requests from external systems and contains the core services such as AWS Lambda, AWS API Gateway and AWS Step Functions. AWS Lambda provides core computation functionality, which runs the point mutation design processing workflows. The API Gateway handles the http requests and routes them to the correct backends. AWS Step Functions orchestrate the serverless workflow by processing messages from the API Gateway and invoking AWS Lambda asynchronously. The data storage tier manages persistent storage from our website, including AWS DynamoDB, and AWS S3. On the GEDpm-cg home page, uploaded input files will be stored on an AWS S3 bucket. When the submit button is clicked after all parameters are set, a request is sent to the API Gateway, which passes the parameters to the AWS Step Function, and all parameters are stored on AWS DynamoDB. Then, the browser gets the response, jumps to the results page, and waits for the computing results. AWS lambda is invoked asynchronously by AWS Step Function event sources, which runs the logic code and uploads the result files to AWS S3. Each submission will trigger a computing process in parallel, regardless of how much demand there is on the website, showcasing the usefulness of serverless computing. Strains, Primers, and Reagents All strains and plasmids used in this study are listed in Supplementary Table S1. The primers (GENEWIZ, Suzhou, China) are listed in Supplementary Table S2. Plasmids were extracted using the TIANprep Mini Plasmid Kit (Tiangen, Beijing, China). DNA fragments were amplified by polymerase chain reaction (PCR) using the Q5 ® high-fidelity DNA polymerase purchased from NEB (Hitchin, UK). PCR products were purified using the TIANquick Midi Purification Kit (Tiangen, Beijing, China). DNA fragments were assembled using the ClonExpress II One Step Cloning Kit purchased from Vazyme (Nanjing, China). Yeast extract and tryptone were purchased from OXOID (Hants, UK). BHI broth was purchased from Hopebio (Qingdao, China). Other reagents were purchased from Solarbio (Beijing, China). Antibiotics were added to the media at the following concentrations when required: 50 μg/ ml kanamycin for E. coli, and 25 μg/ ml kanamycin for C. glutamicum. Construction of Plasmids and Strains E. coli DH5α was used as the host for plasmid construction, and was cultivated in Luria-Bertani (LB) medium containing (per liter) 10 g tryptone, 5 g yeast extract, and 10 g NaCl. The introduction of point mutations into the genome of C. glutamicum was achieved via a two-step homologous recombination procedure using the suicide vector pK18mobsacB (Schäfer et al., 1994). The starting strain was C. glutamicum ATCC 13032. To introduce the C568T mutation into the endogenous gene adhA, the vector pK18-adhA C568T was constructed as follows: the flanking regions of the adhA gene with relevant modifications were amplified from genomic DNA of C. glutamicum using the primer pairs adhA-1/adhA-2 and adhA-3/adhA-4. The corresponding products were assembled into the vector pK18mobsacB digested with BamHI based on the T5 exonuclease-dependent DNA assembly method (Xia et al., 2019) using the ClonExpress II One Step Cloning Kit (Vazyme, Nanjing, China), resulting in the vector pK18-adhA C568T . The plasmids pK18-ald C973T and pK18-ldhA C463T were constructed analogously. Design Principle of the Design of Homologous Arms and Primers for Point Mutation Editing by GEDpm-cg In order to seamlessly bridge in-silico design results with in vivo or in vitro experiments, DNA assembly methods should be loaded into GEDpm-cg. In comparison with the traditional restrictionligation methods, newly developed DNA assembly methods such as Gibson Assembly and Golden Gate have been increasingly favored as streamlined assembly workflows by biologists due to their simplicity, cost effectiveness and cloning efficiency (Casini et al., 2015). In this study, overlap-based assembly methods such as Gibson Assembly and CPEC (Casini et al., 2015), in which DNA fragments are assembled based on homologous ends (usually from 15 to 40 bp), were chosen as the loading technique for GEDpm-cg. In addition to the basic two pairs of primers (primer-1/2 and 3/4) for amplifying the UHA/DHA, verification PCR using the first pair of verification primers (testprimer-1/2) is required to avoid false-positives during the 1 stround of single crossover and isolation. The correct mutants containing the point mutation are finally confirmed by sequencing using the second pair of verification primers (testprimer-1/3). Thus, a total of two homologous arms and four pairs of primers are required for the whole process of point mutation editing (Figures 1, 2). The design principles of the UHA, DHA and primers are as follows. In order to save cost, the length of primers was expected to be short. As shown in Figure 3A, to correctly match template sequences, the length of the 3′-end sequences was limited between 18-25 bp for any primer. To ensure the assembly efficiency among UHA, DHA and empty vector, the length of overlap regions was set to 20 bp. For the overlap between UHA/ DHA and empty vector, a 20 bp sequence at the 5′-end of the linearized empty vector was added to the 5′-end of primer-1, and a 20 bp reverse complementary sequence of the 3′-end of the linearized empty vector was added to the 5′-end of primer-4. For the overlap between UHA and DHA, a 20 bp sequence covering the point mutation was added to the 5′-end of primer-3, and its reverse complementary sequence was added to the 5′-end of primer-2. For the verification of the 1 st single crossover ( Figure 3B), test-primer-1 was located upstream of the UHA in the targeted genome, and test-primer-2 was located downstream of the DHA in the pK18mobsacB-derived vector. Thus, if clones obtained in the 1st isolation round on kanamycin plates were unable to yield clear PCR products, they were identified as false positives that should be discarded. For the verification of the 2 nd single crossover ( Figure 3C), the verified test-primer-3 was located downstream of the DHA in the targeted FIGURE 3 \| Schematic diagram of the design principle of GEDpm-cg. Parameter settings for the design of primers and homologous arms for the vector DNA assembly (A), the 1 st -round of single crossover and isolation (B) and the 2 nd -round of single crossover and isolation (C). UHA, upstream homologous arm; DHA, downstream homologous arm; green band, homologous end between UHA and vector; yellow band, homologous end between DHA and vector, blue diamond, base before point mutation; red diamond, base after point mutation. UHA min-max , length range of upstream homologous arm; ODR UHA , optional design region of upstream homologous arm; DHA min-max , length range of downstream homologous arm; ODR DHA , optional design region of downstream homologous arm; UIS min-max , length range of upstream internal sequence (sequence from 5′-end of test-primer-1 (forward primer for verification of the 2 nd round of crossover) to the 5′-end of the upstream homologous arm); ODR UIS , optional design region of forward primer (test-primer-1) for verification of the 2 nd round of crossover; DIS 1st min-max , length range of downstream internal sequence for verification of the first single crossover (sequence from the 3′-end of the downstream homologous arm to the 5′-end of test-primer-2 (reverse primer for verification of the 1 st single crossover)); ODR 1st DIS , optional design region of reverse primer (test-primer-2) for the verification of the first single crossover; DIS 2nd min-max , length range of downstream internal sequence for the verification of the second single crossover (sequence from the 3′-end of the downstream homologous arm to the 5′-end of test-primer-2 (reverse primer for the verification of the 1 st single crossover)); ODR 2nd DIS , optional design region of the reverse primer (test-primer-3) for verification of the second single crossover. Frontiers in Bioengineering and Biotechnology \| www.frontiersin.org October 2021 \| Volume 9 \| Article 768289 Frontiers in Bioengineering and Biotechnology \| www.frontiersin.org October 2021 \| Volume 9 \| Article 768289 7 genome. The mutant strains are verified by sequencing the PCR products obtained using test-primer-1 and -3. It is worth noting that all seven primers were optimized within their respective optional design regions (ODRs, Figure 3A), which was determined by the user's parameter settings (Supplementary Figure S2), in order to reach scores as high as possible using the Primer3 algorithm. Automated Design of Point Mutation Editing Using the GEDpm-cg Web Server For the convenience of biologist users, GEDpm-cg is available across computer platforms via a common web-browser interface ( Figure 4A), and as such does not require the user to install or update the software. Compared with software built at a centralized server with potential load-balancing problems when many users are submitting their requests simultaneously, GEDpm-cg, built in a serverless manner, can invoke numerous simultaneous functions in parallel, automatically scaling with the size of the workload (Enes et al., 2020). Amazon lambda (https://aws.amazon.com/lambda/) was used as the core computing service due to its quite short startup time and flexibility. An online user manual provides an overview of GEDpm-cg's functionality, step-by-step how-to examples, indepth descriptions of input and output files, detailed documentation of GEDpm-cg, error-message explanations, and experimental protocols for the aforementioned point mutation editing. To begin the GEDpm-cg point mutation design process, the user needs to select/upload relevant information concerning the targeted genome, point mutation information and linear vector sequence. The targeted genome can be selected among 71 offered C. glutamicum strains available in the NCBI database or uploaded by the user as a FASTA-format sequence file (community standard FASTA). The linear vector sequence is uploaded by the user as a TEXT (TXT) format sequence file. Because the UHA and DHA are expected to be assembled with the linear vector at the 5′-end and 3′-end, respectively, the sequence direction of the linear vector is suggested to be verified repeatedly by the user. The point mutation information is provided by uploading a CSVformat sequence file containing the sequence ID, sequence without the point mutation, index of the targeted mutation site in the sequence and the targeted nucleotide before and after the mutation is introduced ( Figure 4B). Notably, in order to avoid repeatedly reading the large genome sequence file and reduce the design time, the length of the uploaded sequence must be larger than the sum of UHA max , DHA max , UIS max and DIS 2nd min , and the mutation site should be located around the middle of the uploaded sequence to satisfy the template needed for the design of the UHA, DHA and primers. In addition to the input files, the user can also alter the default DNA assembly parameters including the lengths of the UHA, DHA, UIS, DIS 1st and DIS 2nd , as well as the primer design parameters including thresholds for the melting temperature (T m ) and GC content (Supplementary Figure S2). After the user submits the three inputs, GEDpm-cg will firstly evaluate whether these inputs meet the standards (see the section on 'in-depth descriptions of input and output files' in the online user manual), then utilizes BLAST (Zhang et al., 2000) to check whether the uploaded sequences for the mutation information are strictly consistent with the targeted genome (Figure 2). If there is any error, GEDpmcg returns an error report (see the section on "error-message explanations" in the online user manual) to prompt the user to correct the uploaded files. Otherwise, GEDpm-cg will utilize Primer3 (Untergasser et al., 2012) to optimize the cloning primers (primer-1/2/3/4, concatenations of 20 bp overhanggenerating sequences and 18-25 bp template-matching sequences) required for generating the UHA and DHA fragments, and the verification primers (test-primer-1/2/3, 18-25 bp template-matching sequences) for two rounds of single crossover (Figure 2). The design results will be provided in four output files. The output file, named "Design results," contains the sequence IDs, primers, lengths of targeted PCR fragment and the T m of homodimer and heterodimer formation of primers ( Figure 4C). Another file, named "Primer order", contains a list of primers provided to the primer synthesis company ( Figure 4D). The output file, named "Failed task", contains the sequence IDs without accessible primers and their failure reasons judged by Primer3 ( Figure 4E). The user can re-set the given parameters (T m and/or GC content) for Primer3 to obtain feasible results. In addition, to avoid potential off-target events, the output file (Evaluation result) provides possible mis-matching sequences and positions for the uploaded sequences ( Figure 4F). The user can try to lower the mismatching possibilities by altering the lengths of the UHA/ DHA. If no problems occur, batch-designs containing 10,000 tasks can be completed within 5 min. To save cloud storage resources, these output files will be stored for no more than 1 week. Experimental Verification of Design Results for Point Mutation Editing by GEDpm-Cg To ensure the reliability of the design results generated by GEDpm-cg, three point-mutations (Supplementary Table S1) in three independent genes (adhA, ald and ldhA) were experimentally constructed in C. glutamicum ATCC 13032. The related input and output files are shown in the additional files. Recently, Xia et al. (2019) developed the T5 exonucleasemediated, overlap-based DNA assembly technique TEDA, which is relatively simple, cost-effective and highly efficient compared with currently popular overlap-based DNA assembly methods such as Gibson Assembly. The assembly of UHAs, DHAs and the linearized pK18mobsacB vector, and the two rounds of single crossovers and isolations were conducted according to the section "experimental protocols" in the online user manual. Agarose gel electrophoresis (Supplementary Figure S3) and Sanger sequencing ( Figure 4G) confirmed that the corresponding pK18-derived vectors were successfully assembled and the three mutations had been successfully introduced into the genome. The average editing efficiency for genomic point mutation editing based on the design of GEDpm-cg can reach Frontiers in Bioengineering and Biotechnology \| www.frontiersin.org October 2021 \| Volume 9 \| Article 768289 8 Table S3), which is consistent with the theoretical 50% editing efficiency after two rounds of single crossover and isolation (Schäfer et al., 1994). DISCUSSION The development of microbial cell factories has been greatly facilitated by computer-aided design tools (Appleton et al., 2017b;Hillson et al., 2019;Carbonell et al., 2020), among which design tools for genome editing play an important role in liberating biologists from laborious, repetitive and error-prone design work (Montague et al., 2014;Quintin et al., 2016;Wang et al., 2019b). However, most genetic modification design tools were specifically developed to handle a specific module for a single phase of the editing process that will be more programmable, such as CHOPCHOP (Montague et al., 2014) for designing CRISPR guide RNAs, PrimeDesign (Hsu et al., 2021) for designing specifically engineered guide RNAs (pegRNAs), Merlin (Quintin et al., 2016) for designing ssDNAs, as well as j5 (Hillson et al., 2012) and Raven (Appleton et al., 2014) for designing DNA assembly primers. To further improve the automaticity of MACBETH and evaluate the off-target risk, our group previously developed an online tool (gBIG, http:// gbig.ibiodesign.net/) for the high-throughput design of guide RNAs, which allowed sequence design for the base editingmediated inactivation of over 3,000 target genes within minutes (Wang Y. et al., 2019). However, no CAD tool is currently available for one-stop design covering all the experimental steps required for genetic modification. Although laboratory biologists can employ these specific CAD tools in a stepwise manner to assist their genetic modification design, truly automated and high-throughput design is still limited by the nonstandardized data exchange and input/output formats (Carbonell et al., 2020). Specifically, users have to manually extract and upload numerous templates (the optional up-and downstream homologous arms) for primer design using j5/Raven or other CAD DNA assembly tools (Hillson et al., 2012;Appleton et al., 2014), which is cumbersome in high-throughput approaches and error-prone. In this study, we developed the online tool GEDpmcg for the automated, rapid and precise design of genomic point mutation editing in C. glutamicum. For the first time, the design of functional elements (homologous arms required for the countermarker-assisted HR system), the vector DNA assembly (primers design for vector construction) and the sequencing verification are integrated in the single CAD tool GEDpm-cg ( Figure 2). As a result, it can provide automated and high-throughput design results covering all the experimental elements required for the constructing and verification of point mutations ( Figures 4C,D). To be biologist-friendly, GEDpm-cg provides an open and free web-service, and the step-by-step how-to examples as well as the in-depth descriptions of input and output files (see in the online user's manual) are all developed to suit the needs of our biologist colleagues. Moreover, to further ensure that the point mutation is being introduced as the user desires without off-target mutations, the alignment between the targeted sequences and targeted genome is checked in advance and the possibility of off-target events is also evaluated. Finally, a testing simulation of over 10,000 single point mutations could be completed within only 5 min, and three point-mutations in the genome of C. glutamicum were experimentally constructed guided by GEDpm-cg ( Figure 4G). Thus, the in-silico design results were seamlessly bridged with in-vitro vector construction and in-vivo C. glutamicum point mutation editing. Although the emerging CRISPR/Cas genome editing systems are increasingly favored for genetic manipulation in C. glutamicum (Wang et al., 2021), the counter-selection-based system is still a reliable genome editing technique, especially for the construction of mutations that require strictly precise nucleotide editing (Stella et al., 2019). To improve the editing efficiency of this technique, various variants were developed, for example by replacing the native promoter of sacB in classical pK18mobsacB with the 18-times stronger P lacM promoter (Tan et al., 2012), replacing the negative selection marker sacB with the novel streptomycin-sensitive gene rpsL (Kim et al., 2011) or 5-fluorouracil-lethal gene upp (Ma et al., 2015), as well as replacing single-copy and non-replicating pK18-derived vectors with multi-copy and temperature-sensitive pCGR2-derived vectors (Okibe et al., 2011). Notably, GEDpm-cg is able to flexibly support all these counterselection-based variants upon the uploading of specific linear vector sequences by the user. What's more, GEDpm-cg can also be used for genomic point mutation editing in other species beyond C. glutamicum based on this same approach if the users upload corresponding genome and vector sequences. Nevertheless, the upgrading of GEDpm-cg to support point mutation editing by other genome editing techniques such as CRISPR/Cas system and/or other editing types for fragment sequence editing is still expected to fulfil different users' preferences in our future work. With the recent technology advances in robotic/software-assisted strain engineering system, it has become feasible to enable an ultraefficient turnover rate of design-build-test-learn synthetic biology cycle (Chao et al., 2017;Hillson et al., 2019). The robotic system-assisted CRISPR/Cas-deaminase-mediated C. glutamicum genome base editing platform, MACBETH, was developed by our colleagues in 2018. The MACBETH enables a maximal editing capacity of up to 9,000 single nucleotide mutant strains within 1 month (Wang Y. et al., 2018), which obviously exceeds the ability of human labor alone to construct no more than one hundred mutant strains per month. Since the basic experimental operations will not be beyond vector construction, plating, cultivating and screening, the MACBETH platform, based on the CRISPR/Cas9 mediated base editing, can be feasibly modified to support automated point mutation editing based on the countermarker homologous-recombination. The combination of design automation based on GEDpm-cg and experiment operation automation based on MACBETH platform will be a superior tool for high-throughput point mutation editing of C. glutamicum. In conclusion, we developed GEDpm-cg with superior efficiency, user-friendliness and flexibility for the design of genomic point mutation editing in C. glutamicum, which can liberate biologists from laborious, repetitive and error-prone experimental design. We believe our platform can open the possibility for large-scale mutation mining via robotic/ software-assisted systems and consequently lead to a better understanding/engineering of cellular metabolism in the near future.	7,835.8	2021-10-15T00:00:00.000	[ "Engineering", "Computer Science", "Biology" ]
Effects of intravenous administration of allogeneic mesenchymal stromal cells, derived from adipose tissue, in five dogs with chronic kidney disease This study aimed to evaluate the safety of allogeneic adipose-derived mesenchymal stromal cell (aASC) treatment in dogs with chronic kidney disease (CKD) at the time of infusions and during the 120-day follow-up after the last infusion. Five dogs with CKD received three intravenous infusions of approximately 1×10 6  10% of aASCs per kilogram of body weight at 21-day intervals. Clinical and laboratory evaluations were performed at the time of each treatment and at 30 and 120 days after the last infusion. Adverse effects of the treatment were assessed using clinical observations, laboratory analyses, and owners’ answers about their dog’s behavior after infusions and during follow-up. The investigated animals did not present any adverse effects immediately after infusion or during the follow-up after the last infusion according to clinical and laboratory observations, as well as the dog owner’s descriptions. One treated animal showed a reduction in creatinine, from 3.5mg/dL to 2.4mg/dL from day 0 to day 153, gained 100g of body weight, and improved disposition. The study results demonstrate that aASC therapy is safe for dogs with CKD; however, further studies are needed to investigate these promising results. INTRODUCTION Chronic kidney disease (CKD) is an irreversible, progressive disease that often affects approximately 25% of dogs (Lund et al., 1999;Pelander et al., 2015).It is characterized by a structural or functional abnormality of the kidneys, leading to the loss of renal function over the course of at least 3 months (Polzin, 2011). Although CKD is chronic, it causes a constant inflammatory state in the patient, which, together with oxidative stress, leads to several serious consequences, such as anemia, nutritional changes, and endothelial and immunological dysfunction (Kogika et al., 2015). Clinical signs are usually nonspecific and are observed in more advanced stages of the disease.The most common manifestations are polyuria, polydipsia, weight loss, poor coat quality, halitosis, stomatitis, gastroenteritis, and dehydration (Cowgill et al., 2016). Treatment must be targeted at each stage of the disease; however, adaptations in therapy must accompany an individual clinical response.Treatment has two main goals: to slow the progression of the disease, preserve functional nephrons, and reduce clinical signs; and to improve the patient's quality of life.Among the main guidelines for treatment are discontinuing the use of nephrotoxic drugs; identifying and treating pre-and post-renal abnormalities; using specific diets; maintaining hydration; and treating urinary tract infections, hypertension, proteinuria, hyperphosphatemia, metabolic acidosis, and anemia (Iris, 2019). For kidney diseases, like with many other chronic diseases, the use of cell therapies has been proposed.The main objective of this therapeutic modality for patients with kidney disease is to improve their quality of life and delay disease progression.These goals are achieved through the administration of mesenchymal stromal cells (MSCs), as they can reduce local inflammation in different organs and tissues (Sutton et al., 2002). MSCs obtained from adipose tissue seem to be safe for allogeneic application without the risk of rejection by the recipient's immune system, as these cells appear to have a low degree of expression of class I major histocompatibility complex (MHC I) and no expression of MHC II on the cell surface (Le Blanc et al., 2003).Thus, it is possible to use allogeneic MSCs from a cell bank in urgent cases of application, such as in cases of acute renal failure or in the exacerbation of chronic renal disease. In this context, cell therapy using MSCs has been studied as a possible treatment for kidney disease.Recently, Villanueva et al. (2019) evaluated the efficacy and safety of intravenous infusion of autologous adipose tissue-derived mesenchymal stromal cells (ASCs) for the treatment of CKD in six human patients using an application protocol of 1×10 6 cells/kg of body mass.They did not find any adverse effects or possible benefits of ASC therapy in these patients. Similarly, Quimby et al. (2011) infused autologous ASCs in cats at three points in the renal cortex under ultrasound guidance; they observed a slight decrease in serum creatinine levels, but there were no adverse effects.In a study by Thomson et al. (2019), autologous MSCs were infused into cats via the renal artery.The study demonstrated a possible route of administration and reported that there were no adverse effects 3 months after treatment. In another study, Quimby et al. (2013) investigated intravenous infusion of 2 and 4×10 6 cryopreserved allogeneic ASCs and 4×10 6 cells cultured from cryopreserved adipose, every two weeks (total three infusions).Cats receiving a lower dose of ASC showed no adverse effects.However, there was no clinical improvement after treatment, suggesting that the effects of cell therapy should be evaluated in the medium-and long-term.The treatments caused immediate adverse effects in the treated animals receiving the higher dose, such as vomiting and increased respiratory rate.However, studies using MSCs in dogs with CKD are scarce and there is no information on its use in terms of safety, possible effects on kidney function, and quality of life in dogs with CKD.Therefore, considering the therapeutic potential of MSC treatment and the lack of studies proving its safety in CKD dogs, this study aimed to evaluate the safety of intravenous infusion of allogeneic ASCs (aASCs) in dogs with CKD Arq.Bras.Med.Vet.Zootec., v.74, n.2, p.310-318, 2022 immediately after each infusion and during 120 days of follow-up after the last infusion.1).CKD diagnosis was first performed by private veterinary clinicians who then referred dogs for participation in this study.CKD was staged according to the guidelines proposed by the IRIS (2019).Inclusion criteria included dogs who were not administered analgesics or antiinflammatory drugs immediately before or throughout the study period.Dog owners were informed of the study.After obtaining signed consent, the patients were administered cell therapy concomitantly with the clinical therapy previously established by the veterinarians responsible for the dogs.All investigated animals received three infusions of aASCs at 21-day intervals between consecutive infusions on days 0, 21, and 42.Clinical and laboratory evaluations were performed at the time of each infusion, as well as 30 and 120 days after the last infusion. This During the clinical evaluation, the owners were asked about clinical evolution, use of medications, exposure to toxins, appetite and type of feeding, behavior and urinary frequency, volume and color of the urine, and quantity and quality of water ingested.Physical examination included the evaluation of the state of consciousness, body mass, hydration status, mucosal membranes, peripheral lymph nodes, heart rate, respiratory rate, and rectal temperature.Systolic blood pressure (SBP) was measured using a model 811 B Doppler device (Parks Medical, USA). For laboratory evaluations, venous blood was collected from the jugular vein and was immediately analyzed at the Laboratório de Análises Clínicas -Clinilab (Curitiba, Brazil) to evaluate the total erythrocyte and leukocyte counts, serum urea level, creatinine level, phosphorus level, symmetrical dimethylarginine (SDMA) level, potassium level, sodium level, ionic calcium level, and blood gas analysis. Subcutaneous adipose tissue was collected from three healthy dogs from the region lateral to the origin of the tail (approximately 20 g).The culture and characterization of MSCs were performed as described by Falcão et al. (2020).aASCs were frozen in 0.5mL straws at a concentration of 1.25 million cells per straw in a freezing solution containing dimethylsulfoxide solution and fetal bovine serum.The obtained aASCs were further thawed and tested for the presence of mycoplasma, fungus, and bacteria using polymerase chain reaction (PCR) (VeritiThermalCycler, Thermo Fischer Scientific).The aASCs were then frozen in a nitrogen canister until the day of infusion.On the day of treatment, the cells were thawed and washed to remove the freezing solution.After washing, the cells were diluted in transport solution, stored in a syringe, and prepared for infusion. All patients involved in this study received an infusion of 1×10 6  10% ASCs per kg of body mass, diluted in 50 mL of lactated Ringer's solution (Fresenius Kabi Brasil Ltda., Brazil), given intravenously via the cephalic vein for 30-40 minutes; each patient randomly received cells from one of the three donors.Infusions were performed on days 0, 21, and 42.Some of the cells used were also analyzed for cell viability.For this, we used the vital dye 7-AAD (BD Pharmingen, USA) for 30 min, washed with PBS (Gibco Invitrogen, USA), and fixed with PBS containing 1% paraformaldehyde (Sigma Aldrich, Brazil).The samples were acquired using a FACSCalibur flow cytometer (BD Bioscience, USA) and analyzed using FlowJo software (FlowJo, Ashland, USA); the results are described as the percentage of dead cells. The data analysis was initially descriptive.The clinical and diagnostic (parametric) data collected in the study were analyzed for normality using the Shapiro-Wilk test, and then described as mean and standard deviation.The laboratory results of each moment were analyzed using repeated measures analysis of variance followed by the post hoc Bonferroni test, considering the significance level at 5% and pvalue >0.05, using the STATA software (version 14, College Station, Texas, USA). According to the dog's tutors' descriptions, none of the patients showed any adverse effects.There were no episodes of vomiting, diarrhea, fever, change in respiratory and/or heart rate, or anaphylactic reaction during and 1 hour after each infusion.In addition, there were no changes, such as apathy or lack of appetite, after aASC infusion. The results related to body weight, SBP, total erythrocyte and blood leukocyte counts, serum creatinine and SDMA levels, venous pH, urinary density, and UPC for each of the five treated dogs are shown in Table 2. Maintenance of most values was observed, as shown in Table 2.The number of leukocytes decreased in all dogs.In the dog 5, the serum creatinine level fell from 3.5 mg/dL to 2.4 mg/dL from day 0 to day 153; the levels in dogs 3 and 4 remained stable. Cylinduria was observed in one dog (dog 4) at the beginning of the study.This patient had 4-6 hyaline cylinders and 1-2 granular cylinders per field.During the treatment, a reduction in the quantification of cylinders was observed, and from day 63, no cylinders were observed. Considering the five treated animals, there was no statistical difference in blood cell counts; creatinine, urea, sodium, potassium, calcium, and phosphorus levels; or urinary density over the study period (p>0.05)(Table 3). The details of each of the treated animals and the information reported by the owners are shown in Table 4. Dog 1 gained 40 g of body weight, stopped vomiting, and had increased activity.Dog 3 had decreased urinary frequency and increased activity.Dog 4 gained 600 g of body weight.Dog 5 gained 100 g of body weight and had increased activity. DISCUSSION The main objective of the present study was to evaluate the safety of the administration of aASCs in dogs with CKD, since this has not yet been demonstrated in this species and in these specific clinical conditions.It was observed that the intravenous infusion of aASCs did not present any adverse effects during the immediate period after any of the three infusions or during the follow-up period of four months after it.It was decided to use allogeneic cells, as it is understood that in the case of patients with CKD, performing anesthetic and surgical procedures to collect adipose tissue could worsen the delicate clinical condition.We chose to prepare MSCs for administration by intravenous infusion as it is a simpler procedure and has a lower risk of complications in patients with CKD. The five animals investigated in the present study were followed up for a period of 120 days after treatment, which allowed us to assess the safety of aASC treatment.However, for the evaluation of the therapeutic effects in dogs with CKD, a larger sample would be necessary.Moreover, even though routine clinical markers were included, these were of limited value to prove the beneficial action of the aASCs.Nonroutine tests would be necessary, such as renal biopsy with histopathological evaluation and interleukin evaluation, or it would be necessary to evaluate a larger number of patients. Thus, the statistical evaluation of clinical findings and biomarkers of renal function and injury did not show any difference between the different assessments, but it is believed that some of the identified observations have clinical relevance and therefore deserve to be emphasized and discussed to serve as inspiration for future studies. Initially, dogs that showed clinical signs of CKD at the beginning of treatment showed improved symptoms.Dogs with CKD under conventional treatment may show clinical improvement but supporting medication must often be maintained (Roudebush et al., 2010).Therefore, it is suggested that the clinical improvement in the animals investigated in the present study might be due to the participation of ASCs in controlling the disease and improving their quality of life because they did not use any other medication during this study. In this study, there was a gain in body weight in dogs treated with ASCs, showing an important beneficial result of treatment with cell therapy, since the loss of body weight usually occurs with the progression of the disease.Weight loss can be caused by numerous factors that result from inflammation, such as increased protein catabolism and increased excretion of bicarbonate, which leads to metabolic acidosis that stimulates muscle cell breakdown and loss of lean muscle.In addition, the cytokines, IL-6 and TNF-α, also act by compromising appetite, leading to a spontaneous reduction in food intake (Oliveira et al., 2010).Loss of body weight is a risk factor that increases the rates of morbidity and mortality in people and dogs with chronic diseases (Freeman, 2012;Ineson, Freeman and Rush, 2019).In this study, the anti-inflammatory action of MSCs may have been the mechanism that underlay the improvement and maintenance of body weight in dogs with CKD; further studies are needed to better understand the mechanisms involved as this may influence the survival rate. In addition, the dogs were more active after infusion of ASCs according to their owners.This leads us to think that because they were elderly animals and may have other comorbidities, such as joint disease, they can benefit from the treatment performed.MSCs may have migrated to other inflammatory foci, such as the joints, resulting in improved disposition and activity; however, in addition to the possible control of inflammatory foci in the joints, we hypothesize that the kidney is also beneficial.This factor is known as homing, which is the ability of stem cells to find their destination in a target organ through the bloodstream, where it directs the migration of stem cells through different signaling pathways, mediated by cytokines or receptors for growth factors released on the surface of stem cells (Tao et al., 2018). Hematological tests did not show any significant changes.The number of leukocytes decreased during the study period, which may suggest further evidence of treatment safety in this study, due to the absence of any type of adverse inflammatory response resulting from allogeneic cell treatments. As for biomarkers, such as creatinine used to assess renal function, it was observed that there was a decrease in one of the treated dogs and in two dogs, the level remained stable.This may indicate a beneficial treatment effect on the control or even improvement of the clinical picture, which would require a new study with a larger number of animals for confirmation.A decrease in serum creatinine levels in patients with CKD stages II and III was reported in cats, at 7 and 60 days after the infusion of allogeneic amniotic membrane-derived MSCs (Vidane et al., 2017, showing a renoprotective effect of cell therapy in CKD. Cylinduria was observed in only one of the investigated dogs, with complete resolution at the end of the treatment.These results suggest an improvement in the condition of the tubular epithelium.The effects of ASCs on the formation of cylinders in kidney disease have not been previously demonstrated in domestic animals; however, ASCs attenuate markers of kidney injury and oxidative damage, protect the tubular epithelium from reperfusion injury, inhibit apoptosis in the injured region, and increase the proliferation of surviving cells in a study carried out in rats (Zhuo et al., 2013). Other limitations of this study would include the evaluation of UPC, which was not performed at all the stages for laboratory reasons, since this test was only performed if proteinuria was present.UPC would help us to establish a prognosis because persistent proteinuria is associated with the risk of uremic morbidity and mortality, and we suggest that it should be included consistently in future studies.SBP was not measured at all stages, as some patients were very stressed at the time of the clinical examination.However, two patients showed a reduction in SBP during treatment, and for these, we have two hypotheses: that they had already accustomed themselves to the outpatient clinic, staff, and treatment, or that it may have been another beneficial ASC effect.Additionally, future studies should include the evaluation of the glomerular filtration rate and analysis of urinary cytokines.The low number of animals surprised us because we believed that we would be able to recruit a larger number of cases to better answer the study questions; however, the adherence of owners and veterinarians was low, showing what can be inherent in a study involving routine cases. CONCLUSION In conclusion, this study showed that cell therapy with intravenous administration of aASCs in dogs with CKD was safe.Although some benefits were observed in the present study, this must be further investigated in a larger number of cases. Table 1 . Breed, age, and IRIS stage of dogs with chronic kidney disease treated with allogeneic adipose tissue-derived mesenchymal stromal cells YO, years old; IRIS, International Renal Interest Society.*IRIS stage on admission to the study. Table 2 . Clinical (body weight and systolic blood pressure) and laboratory examination (blood erythrocytes and leukocytes, creatinine, SDMA, venous pH, urinary specific gravity, and UPC) of dogs with chronic kidney disease treated with adipose tissue-derived allogeneic mesenchymal stromal cells creatinine, mg/dL; SDMA, symmetrical dimethylarginine, mg/dL; UPC, urine protein to creatinine ratio; U, unrealized. Table 4 . Clinical signs and behavior (according to owners) of dogs with chronic kidney disease before and after treatment with adipose tissue-derived allogeneic mesenchymal stromal cells BW, body weight; g, grams; ↑, increased; ↓, decreased; +, gained.	4,167.8	2022-03-01T00:00:00.000	[ "Biology" ]
Research on Cobot Action Decision-Making Method Based on Intuitionistic Fuzzy Set and Game Theory The bounded rational properties of humans in human-robot collaboration (HRC) is a fundamental reason for collisions in proximity HRC. As HRC scenarios in manufacturing become increasingly popular, robot action decision-making needs to consider such properties of humans. In previous studies, humans are usually regarded as rational agents whose behaviors are predictable and planned. Still, humans are susceptible to distractions caused by external disturbances, and different cognitive processes of the task can produce unpredictable behaviors. To better simulate human bounded rational behavior, based on an intuitionistic fuzzy (IF) multi-attribute decision algorithm, we propose a cobot action decision-making method that integrates human intention, safety, and efficiency, to produce a human-like decision. We use the IF set to calculate the score and accuracy values of two Nash equilibria of a static chicken game, which can predict human action intentions with collision risk and provide optimal action decisions for the robot simultaneously. We generated 10,000 sets of data using the Monte Carlo method and validated the effectiveness of our proposed method by comparing it with MDP and POMDP methods. The results showed that the decision-making method could effectively perform the task of making action decisions for the robot. Simulation experiments and Turing test results show that our proposed method can predict a human’s subjective action decision intention in a situation with potential collision risk with 85.62% accuracy. At the same time, the experimental participants believe that the robot can get 4.83 out of 5 satisfaction points for an action decision. human intention of the assessment is often vague and proba-82 bilistic, which is because human is not fully rational intelli-83 gence. The judgment and decisions made by the human are 84 often subjective and bounded rational, and when the human 85 and machine work in proximity in scenarios with potential 86 collision risk. The frequent judgment of collision risk will 87 significantly increase the cognitive load of the human work, 88 resulting in errors in judgment leading to increased collision 89 risk. Therefore, only the human-like decision-making action 90 of cobots can reduce the cognitive load of human judgment 91 on cobot behavior, reduce human psychological pressure, and 92 improve the efficiency of both the human and cobot while 93 ensuring safety. 94 For a proximity HRC scenario such as Figure 1, a rapid 95 decision must be made for the cobot's action. Most impor-96 tantly, the decision at this point needs to be based on an 97 accurate judgment of the human's action intention. Previ-98 ous studies on similar scenarios [7], [8], [9] proposed using 99 methods such as game theory to make decisions about the 100 cobot's actions. However, they were based on the other party 101 being a fully rational agent, which is against the bounded 102 rationality characteristic of humans in reality. Although some 103 studies have investigated the bounded rationality problem 104 [10], [11], they have not studied the scenarios with potential 105 collision risk in extremely proximity HRC scenarios. In the 106 scenario of proximity HRC, human thinking time is concise. 107 Most decisions are made through experience and intuition, 108 with a vital characteristic of limited rationality. Moreover, 109 if accurate judgments of human intentions cannot be made, 110 safer action decisions cannot be made for the robot, and trust 111 in the robot cannot be improved. 112 There is a great need for a method to measure human 113 fuzzy decision intentions and formulate decisions to solve 114 the cobot decision problem to improve cobot efficiency and 115 safety, reduce human cognitive load and enhance comfort and 116 trust. 117 Statistical methods have analyzed most studies on the prob-118 lem of prediction of human intentions and then expressed 119 them in the form of probabilities. However, in real scenarios, 120 human perceptions of such issues are vague and can only give 121 vague ranges, which cannot be expressed in a definite given 122 lv number. In contrast, in the scenario presented in this paper, 123 human intentions expressed in probability can only reflect 124 whether two intentions are passed or not. In reality, humans 125 will have a large proportion of hesitation states, which cannot 126 be expressed in probabilistic form, which results in misjudg-127 ment of human intentions and low satisfaction with robot 128 action decisions. On the other hand, the IF set uses the 129 subordination and non-subordination functions to express the 130 intention of passing well, not passing and hesitating that 131 appears in the human at this time, which provides more 132 reference for the development of the robot's action decision. 133 The decision is made according to the size of the hesita-134 tion state of the person at this time. For example, the cobot 135 decides to pass when the hesitation intention is significant to 136 improve efficiency, and vice versa, not to pass to ensure the 137 person's safety and ultimately enhance the satisfaction of the 138 person. 139 In addition, in the scenario studied in this paper, improving The main contributions of this paper are as follows: 193 This section presents the application of game theory to robot 194 motion planning and collision avoidance decision-making 195 problems. Different solutions to the issues of collision avoid-196 ance and interaction-aware modeling in HRC have been given 197 in autonomous driving and robot motion planning. In par-198 ticular, the interaction-aware modeling problem is increas-199 ingly attracting the attention of researchers in socially aware 200 robot navigation [7], [12]. Interaction-aware modeling is 201 the basis for solving HRC decision-making problems, and 202 decision-making methods model the relationships between 203 agents, actions, environments, and tasks [13]. Decision-204 making selects the best move for the robot based on the payoff 205 calculated from the utility function of each job. 206 Probabilistic methods, deep learning, and game theory are 207 among the most widespread decision-making methods. For 208 probabilistic methods, Markov decision processes (MDP), 209 Bayesian processes, and graph theory are some of the most 210 widely used methods. For instance, Roveda [10] et al. use 211 Hidden Markov Models (HMM) to teach the robot how to 212 achieve the task based on human demonstrations and use 213 Bayesian optimization-based algorithms to maximize task 214 performance. 215 However, probabilistic models do not satisfactorily deal 216 with bounded rational behavior and uncertainty problems. 217 Many studies have addressed similar issues using deep 218 learning or reinforcement learning approaches to enable the 219 correct processing of boundedly rational behavior. For exam-220 ple, Roveda [10] et al. used partially observable Markov 221 decision processes (POMDP) to develop a framework for 222 planning collaborative robot tasks in assembly, considering 223 both the designer and operator's intents. The designer's CAD 224 data automatically derives a set of potential assembly plans 225 and translates it into a state graph from which the operator's 226 intentions follow. However, the drawbacks of deep learning 227 and reinforcement methods are the computation cost and slow 228 learning speed. Game theory methods in decision-making 229 processes have only recently been exploited. They can model 230 most of the tasks of a group of agents (players) in collabora-231 tion or competition. Game-theoretic methods have been used 232 in different HRC applications. For example, Gabler [8] et al. 233 proposed a game-theoretic-based action selection framework 234 for HRC that allows robots to select appropriate actions based 235 on the behavior of their human colleagues during proximity 236 collaboration. The proposed framework models the HRC 237 scenario as a non-cooperative game model and selects action 238 strategies for the robot by the Nash equilibrium results. The 239 framework selects the optimal trajectory from the action set to 240 assign to the robot, completes the work, and avoids collisions. 241 However, the research considers people to be fully ratio-242 nal when building game models, which has a large gap 243 from the actual situation. People often make decisions with 244 bounded rationality in real work scenarios due to their 245 personality, work environment, and fatigue. Similar In our paper, we focus on the following three aspects: 1. 302 In the process of proximity HRC, there is no optimal path for 303 FIGURE 2. The framework of the proposed cobot action decision making method based on IF set and game theory. avoidance or inconvenient avoidance for human-cobot simul-304 taneous convergence to the same target. 2. Human action 305 intention under the combined effect of efficiency require-306 ments, subjective risk perception, and irrational factors. 3. 307 Robot integrated efficiency, safety, and human action inten-308 tion to make the optimal action strategy. 310 This section outlines our proposed cobot action decision-311 making method based on the IF set and game theory. First, 312 we outline the scenario and object to be studied, simplifying 313 the scenario to a static chicken game and treating the human 314 and cobot as agents. Second, the action decisions of the 315 human and cobot are modeled. We established four IF sets, 316 the IF set of human for efficiency, the IF set of human for 317 comfort, the IF set of cobot for efficiency and the IF set of 318 cobot for safety. Thirdly, the IF set of human action intention 319 is established by integrating the IF set of human for efficiency 320 and comfort. The IF set of collision avoidance is found by 321 integrating the IF set of human action intention and the IF set 322 of cobot for efficiency and safety. Finally, the cobot collision 323 avoidance decision-making method is established by calcu-324 lating the exact value of the IF set of the Nash equilibrium 325 solution. The cobot action decision method framework based 326 on the IF set and game theory is shown in Fig. 2. 328 Our focus is similar to the scenario shown in Fig. 1, where 329 a person and a cobot tend to grasp parts towards the same 330 target in a tight space simultaneously. Due to the extreme 331 reaction time, it is impossible to communicate between the 332 human and cobot through language to plan the sequence of 333 actions between the human and cobot in actual HRC scenar-334 ios. Such scenes also cause a potential collision risk between 335 the human and the cobot. Moreover, it is impossible to avoid 336 collision by optimizing the cobot's trajectories due to space 337 constraints, and one of two agents must adopt a temporary 338 yielding strategy to avoid a collision. 339 Previous experiments found that people prefer to accelerate 340 through potential collision areas to improve efficiency when 341 working in a seated position or in a state where they cannot 342 move their current position at will. And it is interesting to 343 note that in actual experiments and HRC scenes, humans also 344 The strategies adopted by both game agents A 1 , A 2 repre- to portray fuzziness [19]. It can simultaneously represent 393 three states of support, opposition, and neutrality, which can 394 describe the natural properties of objective phenomena more 395 delicately and comprehensively, so IF set are widely used in 396 economic management decision problems. We introduce the 397 basic concepts of IF set in the following. 398 Definition 1: Let X be a universe if there are two functions 399 on X , i.e., µÃ : X → [0, 1] and νÃ : define the degree of membership and the degree of 403 non-membership of an element x ∈ X , such that they satisfy 404 the following conditions: Then µÃ and νÃ determine an intuitionistic fuzzy set on 407 universe X , which can be abbreviated as: Definition 2: (Trapezoidal Intuitionistic Fuzzy Number): A 410 trapezoidal Intuitionistic Fuzzy Number(TIFN) denoted by, 411 a = (a, a 1 , a 2 ,ā) ; wã, uã is a special IF set on a real num-412 ber set , whose membership function and non-membership 413 functions are defined as follows: where wã, uã denote the maximum and minimum member-418 ship degree ofã respect, such that they satisfy the conditions 419 is called the measure of uncertainty. 422 Definition 3: 423 The sum of IF sets: The product of IF sets: 478 Therefore, we establish a model of human cognition of 479 efficiency based on the CPT and establish the IF set of human 480 for efficiency. Assume that the process of the human alone to 481 complete the assembly task, the time t s required for the maxi-482 mum speed of 1.8m/s is recorded as the maximum efficiency, 483 at which time the efficiency is 100%. The time t 0 needed 484 to reach the target from the current position at the current 485 speed is the reference point when the efficiency is t s t 0 . 486 Suppose one's perception of efficiency is entirely rational, 487 and the result of decreasing efficiency with increasing arrival 488 time is well understood. In that case, i.e., one maintains a 489 neutral attitude toward the relationship between efficiency 490 and speed, we construct the expected utility function of the 491 neutral attitude with these two efficiencies. Based on the expression for the value function in the CPT, 494 we determine the value function of efficiency as perceived by 495 the person. (2) 497 The decision weight function as perceived by the person: In Eqs. (2) to (4), α and β reflect the level of risk preference 501 of the decision-maker. In the value function f (t), the smaller 502 the value indicates, the higher the decision maker's sensitivity 503 to risk and p denotes the probability of reaching the target at 504 time t. λ is the loss aversion coefficient, and when λ > 1, 505 one values losses more than gains. t l is the maximum time 506 corresponding to the time when f E (t) =0. The w + (·) and 507 w − (·) represent the values of the decision weight function in 508 the gain and loss regions, respectively. The decision weight 509 function is inverted ''S'' shaped, as shown in Fig. 3. 510 The smaller the parameters γ , δ (0 < γ , δ < 1), the more 511 curved the function shape is, and the more decision-makers 512 tend to overestimate small probability events and underesti-513 mate significant probability events. 514 According to the CPT, the prospect of a person approach-515 ing the target area at the current speed, i.e., the model of 516 human perception of efficiency, can be expressed as: are as follows: The cobot is a fully rational intelligent unit, so the IF 529 set of cobot for efficiencyẼ R = µ R (t) , υ R (t) mem-530 bership and non-membership functions of the cobot are as 531 follow: The IF set of human and cobot for efficiency are shown in 535 Fig. 5 and 6, respectively. 537 The human perception of collision risk is different from the 538 cobot. The human perception of collision risk depends not 539 only on the relative distance and speed of the human and 540 the cobot but also on a variety of subjective factors such as 541 the human's perception of the cobot's appearance, experi-542 ence, and trust. The direct use of relative distance and speed 543 cannot accurately measure the human perception of collision 544 risk. Our previous study established a psychological safety 545 field(SE P ) model to calculate the psychological impact of a 546 cobot approaching different human body parts with a certain 547 speed, minimum separation distance, and direction. When the 548 psychological safety field strength is large enough, a person 549 perceives that they will choose to avoid the cobot. 550 The literature [22] and our previous research [23] found 551 that the acceptable cobot motion speed range is between 552 0.3m/s and 1m/s, and the range of speed considered com-553 fortable is between 0.5m/s and 0.8m/s. At this time, the 554 psychological safety field strength SE Ph can be calculated 555 VOLUME 10, 2022 comfort, and they usually can only give fuzzy conclusions 603 intuitively. Therefore, the IF set of various factors are given 604 different weights, and then the IF set of various factors 605 are weighted and aggregated by the weighted aggregation 606 method. 607 Since the human perception of the ranking of efficiency 608 and safety is vague, it is impossible to accurately deter-609 mine the importance ratio of efficiency and safety. Therefore, 610 to be closer to the real situation, we randomly interviewed 611 five team members and collected their perceived importance 612 weights for efficiency and comfort. The IF set of human for 613 efficiency and comfort are weighted together to obtain the 614 IF setK = µ K (x) , ν K (x) of human action intention on 615 whether to continue moving towards the target through the 616 potential collision area for the current situation. From the above analysis, it is concluded that the cobot 619 needs to consider three factors: safety, human action inten-620 tion, and work efficiency to consider whether to use the 621 original action of continuing to move towards the target or 622 choosing the strategy of temporarily yielding. For this rea-623 son, to ensure safety, we first set the weights of the three 624 factors as w S = 0.95, 0.05 , w K = 0.8, 0.1 and w E R = 625 0.6, 0.2 respectively. The IF weights w E w SE w S w K w E R 626 in the paper can be obtained by sensitivity analysis to get 627 the range of attribute weights changes when the ranking 628 of decision options' advantages and disadvantages is kept 629 constant. Using the sum and product of the IF set, the action 630 decisions made by the cobot considering the three factors 631 together are weighted and assembled to obtain the IF set of 632 collision avoidanceC = µ C (x) , ν C (x) . 633 When a person passes through the potential collision area 635 and the cobot temporarily yields, we record the IF set of 636 collision avoidance asC C = µ C C (x) , ν C C (x) ; The oppo-637 site strategy is recorded asC K = µ C K (x) , ν C K (x) . It is 638 necessary to give the ranking method of intuitionistic fuzzy 639 sets to judge the size. decision-making algorithm proposed in this paper, i.e., w E = 678 0.6, 0.3 w SE = 0.5, 0.4 . We set the discount factor γ d at a 679 fixed value in the MDP to a dynamic value, assigning a weight 680 of 0.6 to safety S M and a weight of 0.4 to efficiency E M . 681 Since no bounded rationality factor is involved in the MDP, 682 security is used as the probability in the state transfer matrix. 683 The decision is obtained by solving the MDP using the value 684 iteration method. The decision is obtained by solving the 685 POMDP using the Q-Learning algorithm. We are particularly concerned with the following two 690 scenarios. 691 1. When the cobot can reach the potential collision region 692 at the original speed, the human arrives after the cobot reaches 693 conservative and aggressive, respectively, and the POMDP 723 method is in between. To verify whether the action decisions 724 made by our proposed method can improve efficiency and 725 safety, we compared the efficiency and safety of the three 726 methods for 10,000 experiments. To show the efficiency and 727 safety more clearly in the figure, we averaged every 100 sets 728 of data. We then plotted them in the figure to reflect the 729 efficiency and safety comparison of the three methods. The 730 experimental results are shown in Fig. 9. The results show 731 that the MDP method is less efficient than our method and 732 the POMDP method, while the efficiency of our method is 733 slightly higher than that of the POMDP method. The three 734 methods do not differ much in terms of safety, and the 735 MDP method has the highest safety. In many scenarios, the 736 cobot action decisions made by MDP yield, while the safety 737 obtained by our method is similar to that of the POMDP 738 method. 739 The effect of our method on the weighting factors can be 740 determined by calculating the sensitivity of the weighting 741 factors to determine within what range changing the weights 742 will not change the ranking scheme of the IF set. 743 The results of the 10,000 sets of experiments generated 744 by the Monte Carlo method show that our method can 745 improve the smoothness and productivity of HRC and the 746 safety of HRC in the overall experimental process compared 747 to the MDP series method, which verifies the effectiveness of 748 the algorithm. We designed simulation experiments to verify 749 the accuracy of the human action intention model and human 750 satisfaction with cobot decisions in real scenarios. 751 We present some of the experimental results of the Monte 752 Carlo simulation experiments made by our method in the 753 article's attachment. 755 To verify the accuracy of the human intention model and 756 human satisfaction with cobot decisions in real scenar-757 ios, we designed the simulated experiment and Turing test. 758 We use the simulated experiment to verify the prediction 759 accuracy and the Turing test to test the satisfaction of the 760 decision method with human intentions and whether humans 761 can distinguish cobot decisions. 3. How much do you trust a cobot like a group B when 832 encountering a similar situation? From a score of 5 (very 833 trusting) to 1 (very distrustful). 834 4. After the test was completed, aspects that were not con-835 sidered by the robot were suggested for the robot's improve-836 ment by reviewing the test process. 837 A total of 42 volunteers participated in our experiment, 838 of which 32 participated in the simulation experiment, 839 29 males and three females. All ten volunteers, all males, 840 participated in the Turing test. The participants were 22 and 841 28 years old and were asked about their age, gender, and 842 experience with cobots and computer games. Most of the 843 participants had some experience with computer games, and 844 35% had some knowledge of cobots. When the scenario presented in the paper arises, we hope 847 to obtain the result that the cobot can accurately predict the 848 human intention of passing through each time. The cobot 849 makes accurate decisions by integrating the current situation 850 to ensure safety, improve efficiency, and maximize human 851 satisfaction. In the simulated experiments, participants did 852 not make unified decisions about the actions in each experi-853 ment, so we take the conclusion of the majority of participants 854 in each experiment as the standard. 855 For the simulation experiment, the sum n s of the same 856 decision as the standard decision action is added and divided 857 by the total number of participants n t to obtain the decision 858 unity d u of participants in each experiment, as shown in 859 formula (23). Then the average value and standard deviation 860 of the decision unity d u of 20 experiments are obtained. For the standard human decision in each experiment, the 863 cobot's action decision should be opposite to the human 864 action decision so that the collaborative task is smooth, effi-865 cient, and safe at the highest level. The number of times 866 the cobot's action decision is opposite to the human action 867 decision in all 20 experiments is the correctness of the cobot's 868 action choice S. 869 We counted the number n o of cobot decisions opposite the 870 standard human decision in each experiment. We calculated 871 the percentage n o in 20 experiments to obtain the degree 872 of correctness S of the cobot action choice, as shown in 873 equation (24). As for the prediction of human intention, only 874 our method can obtain the human action intention among the 875 three methods. We compare the IF setK = µ K (x) , ν K (x) 876 of action intentions in passing through and yielding inten-877 tions, get the human action intentions in each experiment by 878 ranking method, and compare the calculated action intentions 879 collected from simulated experiments to predict accuracy. We counted all participants' mean and standard deviation 882 for the unity of participant decision-making, the accuracy of 883 VOLUME 10, 2022 of the cobot. We believe that the higher satisfaction for 922 both our proposed method and the POMDP method is that 923 both types of methods incorporate a human factor of limited 924 rationality rather than simply calculating the gains in effi-925 ciency and safety. Such methods can more closely match the 926 human cognitive process for such scenarios while targeting 927 decisions more consistent with human bounded rationality 928 characteristics. 929 Statistically, the results for questions 1, 2, 3, and 4 in 930 the Turing test are as follows. For problem 1, group A is 931 most often considered a human decision, accounting for 50%. 932 It is closely followed by our proposed collaborative cobot 933 action decision-making method, i.e., Group B is regarded as 934 a decision made by humans, accounting for 40%. In contrast, 935 group C accounted for only 10%. The results show that our 936 proposed method has passed the Turing test, with a percent-937 age of decisions considered to be made by humans greater 938 than 30%. For questions 2 and 3, the mean scores are 4.4 and 939 4.3, respectively, demonstrating that the decisions made by 940 our proposed method are human-like and receive a high level 941 of confidence. For question 4, some participants suggested 942 that the robot did not fully understand human intentions. 943 There was not only passing, yielding, and hesitation in the 944 human-robot collaboration process but also the intention of 945 not actively participating, the neutral intention of only observ-946 ing the cobot's actions, and even the intention of refusing 947 to cooperate with the cobot. For this small part of other 948 intentions, we plan to make it one of our research directions 949 in the future as well. 950 From the experimental results, our proposed method not 951 only improves the robot's efficiency and human satisfaction 952 with the robot but also ensures the safety of HRC. We believe 953 that the reason for this result is that the IF set gives not 954 only the possibility of the human taking the intention to 955 pass but also the possibility of yielding and hesitation. The 956 IF set can better describe the human cognitive process and 957 provide a more accurate human action intention for cobot 958 decision-making. 960 The paper focuses on proximity collaboration scenarios 1 961 and 2, which are an in-depth study of similar studies in the 962 literature [14] that do not accurately predict scenarios. In such 963 scenarios, once both parties involved in the action decide to 964 pass, there is not enough time to take the avoiding action, 965 and it is easy to collide and cause danger. The action choices 966 of participants in this scenario often rely on bounded ratio-967 nal factors such as intuition and personality, and the action 968 decisions of different participants are highly divergent inter-969 nally, so how to better model human cognitive styles in this 970 scenario is the key to solving this problem. However, previous 971 studies lacked research on such situations and could not make 972 predictions and decisions based on human cognitive styles. 973 With the increasing popularity of HRC in both industrial and 974 life scenarios, the frequency of such situations will increase, 975 so it is necessary to model the human cognitive style in this 976 situation. 977 We propose the cobot action decision-making method 978 based on IF set and game theory for this problem. 979 efficiency and safety combined with the bounded rational-981 ity factor of humans and integrated simulation of a human 982 cognitive process. We try to get more accurate human action 983 intention prediction results through the human action inten-984 tion IF set, which is the first innovation point of our research 985 on such situations. We use the established cobot action 986 decision-making method to decide on cobot actions. Our fuzzy set of human action intention proposed in this paper 1036 better reflects the human decision-making method in such 1037 scenarios, and the prediction of human intention, whether 1038 through or not, is more accurate. The safety of cobot action 1039 is ensured from the intention perception level. The disad-1040 vantage of this paper is that our proposed method does not 1041 perform to a high degree in terms of safety, efficiency, and 1042 human satisfaction, with a prediction accuracy of only 86% 1043 for human intentions. We believe this is related to the higher 1044 number of parameters that appear in the model, which need to 1045 correspond to the large number of model parameters that need 1046 to be adjusted to more closely match the action intentions of 1047 most people in different scenarios. The prediction accuracy 1048 of our proposed method does not reach 90%, and the decisive 1049 actions need to be improved in terms of safety, efficiency, and 1050 satisfaction. 4. However, the IF set can only represent affiliation, dis-1052 affiliation, and hesitation and do not represent neutral and 1053 rejected cognitions, so the model does not achieve more than 1054 90% accuracy in predicting human action intentions. In our 1055 future work, we will optimize the model using picture fuzzy 1056 sets [25], [26] and similar improvement methods to improve 1057 the prediction model's performance. 1058 5. The simulation experiment is done in a laboratory 1059 environment using a computer game and cannot represent 1060 the choice of human responses in a natural setting. In a 1061 real industrial scenario, workers will be influenced by the 1062 impact of the work environment, task efficiency require-1063 ments, cobot actions, and other factors, which may produce 1064 different requirements for the cobot's action decisions which 1065 is the future research direction of HRC. 1067 We design a cobot action decision-making method based on 1068 the IF set and game theory. The decision-making method 1069 integrates the three factors of human action intention, safety, 1070 and efficiency and provides an effective strategy for the cobot 1071 action decision. This method models human action intentions 1072 through CPT and IF set to reason about the human action 1073 intention in situations where there is no optimal trajectory for 1074 proximity HRC to achieve collision avoidance. The optimal 1075 action of the cobot is also calculated based on the IF set of 1076 two Nash equilibrium solutions for the human-robot scenario, 1077 similar to the static chicken game. Through experiments, 1078 it was found that the present method can achieve an accuracy 1079 of 85.22% in predicting human action intention. At the same 1080 time, the participants' satisfaction with the cobot's decision 1081 can reach a satisfactory level. Unlike the study in the liter-1082 ature [14], our study focuses more on situations where the 1083 human-cobot distance and speed are very close, with a high 1084 risk of collision. Also, different from the study in the literature 1085 [12], we consider the human as a boundedly rational agent 1086 and simultaneously consider the three action intentions of 1087 passing through, yielding, and hesitating. It is possible to 1088 predict human action intentions better and provide a reference 1089 for the cobot's action decision-making. VII. CONCLUSION for cobot actions when both humans and cobots compete for 1092 the shared space. We analyzed the bounded rational way of human cognition for efficiency and comfort and established the IF set of humans for efficiency and comfort, respectively. prehensively analyzing three factors: human action intention, safety, and efficiency. a spatial conflict arises between a human and a robot in	7,504	2022-01-01T00:00:00.000	[ "Computer Science" ]
Family GPS Tracking for Android In smart phone field, the security, tracking lost and kidnapping prevention are one of the main areas in current days. Today family members safety is more important when they are outside home. So, there is a need for a tracking system for kids where the parents can monitor their kids at anytime from anywhere. These security goals are achieved by the Android System, Global System for Mobile (GSM), Global Positioning System (GPS) and Short Message Service (SMS) technologies. The proposed application must be installed in all family member's smart phones, and these smart phones will use GPS services without need for internet connection, where the GPS is used to locate the child specifically and to track the child whenever he changes his place. The location obtained by only depending on GSM by sending SMS which contain a link of Google map shows the information of the position. The application also has the ability to trigger a help SMS to his parents when the child is in a dangerous situation by using his headphone, and this is considering a help message in real time. Introduction In today's world, over 80% of people in the world having smart phones [8], the year 2018 is the first year where the use of smart phones will exceed the use of phones without operating system. According to the International Center for Missing and Exploited Children, about 8 million children are reported missing each year. This horrific number includes kidnappings in wars and even kidnappings by relatives and 90% of these children are found. But this does not eliminate the fact that these children who are absent from their families may be subject to persecution, harassment and human trafficking cases, some of whom lose their effect permanently. In addition to kidnapping there is also disappearance, where children ages 14 years and 17 years are missed or ran away from home [1]. In addition, if parents are planning to have a trip, they are always worried about their kids well fare. This concern may have an impact on the parents to have their children outside the house for any reason like going to school, playing with friends or even family trip. Moreover, parents can have their children out of sight and thus they are afraid of the possibility of having their children kidnapped. Consequently, the proposed application is supposed to be used by parents with the purpose of helping to locate absent or unavailable kids. It is a well recognized fact that today's children have their own smart phones accompanying them almost all of the time. This can be of a great advantage in such a worrisome situation of a missing child. In this work, Global Positioning System (GPS) is utilized in conjunction with one of the fundamental services of a mobile which is SMS, the GPS uses longitude and latitude to track the location. This project has two sides, the application at the parents' side will permit parents to try to locate the child and then receive a reply about the position of the child on the Google map. On the child's smart phone side, information is collected to locate the smart phone. An application will be executed at the backend. The interplay between the parents and child's smart phones are executed through Short Message Service (SMS). A great and unique advantage of this SMS dependent application is that it can still be functioning in the absence of internet connection and even in the absence of General Packet Radio Service (GPRS) supporting software on the target smart phones. The system operates by locating the child's smart phone when the later sends a help SMS. Along with location the child may use Headphone to send a notification or help message to his parents that he is in danger. Previous Works Many studies have been done by researchers for this kind of applications, and they are different in the manner, tools and techniques. In Katore T.D. and others [4], their application "allows the parents to get their child's location on Google Maps and also their browsing history. The main responsibility of parent device is to send a request location SMS to child's device to get location of child. While the responsibility of child's device is to reply the GPS position to the parent's device upon request". Sucitra M. [8], his application "allows the parent to get their child's location on their mobile as notification. The system consists of two sides, child side and parent side. The child's device main duty is to sense the signals of an alert from the child and send a notification to the parent device. On the other hand, the parent's device main responsibility is to receive the notification from the child's device". Vanjire S. and others [10] "propose a developing Android Application which is based on Location Based Services (LBS) & provides different location-based services like nearest friend locator, family member location finder". Shyam N. and others [7], "proposed system takes the advantage of the location services provided by GSM and GPS module kit, Radio Frequency Identification (RFID). Information such as GPS coordinates and time are gathered and sent to the parent's phone that's preregistered on the module kit. The communication between the parent and the child module kit is done using SMS". Contributions This paper describes location discovering and tracking system which aims at checking the child's welfare through his smart phone through achieving the following three objectives:  This application represents an alarm system to the family that operates by sending an alert SMS to the parents when their children or family members send help SMS start with code "Location".  The application also has the advantage of automatically sending an SMS to parents if the child takes off his head phone.  The location data (Google Map) can be shared online. Global Positioning System (GPS) "The GPS is a space based on satellite navigation system that provides location and time information in all weather conditions, anywhere on or near the Earth where there is an unobstructed line of sight to four or more GPS satellites" [10]. The domains for the GPRS system are present in the united states and it is continually supplying data for a spectrum of purposes that range from the very sensitive military and national security issues to the simple driving and tourist guidance. The only one prerequisite for using this facility which is having a GPS access. [10] GPS can be utilized by ordinary users as a navigation system. In practice, the GPS receiver containing devices uses the triangulation technique for positioning objects on the ground and for this purpose they derive data from at least the nearest three satellites. The outcome comes in the form of a spotted geographic location longitude and latitude. The programmers of software applications are racing to produce user friendly applications utilizing these available parameters to provide useful data for walking or driving instructions with high accuracy rate approaching figures as small as 10 meters. Being spotted by the GPRS can take some time especially when the target is in an overcrowded or moving situation as for example when driving a car in a crowded large city center. Initiation delay of the GPRS also depends on the starting method of its receiver. The starting time required for a GPS lock on a target is usually relied on how the GPS receiver starts. The preexisting data available for the receiver favors which satellite to look for according to the receiver's last position and accessibility of the satellites in the sky. This takes more time than a hot start but not as long as a cold start. Thus, it is more or less an opportunistic process to lock onto the signal of the nearest available satellite and is much the same like "polling" which is a time-consuming method when compared to have a fixed target satellite, see figure (1) [6]. Figure 1: Architecture of GPS System. [6] The innovative "assisted GPS Technology" was later introduced to overcome some of the delay problem on the expense of some memory consumption in the smart phone. This works by installing ephemeris when wireless connection is available, and this installed data can then use the cell phone towers to triangulate the user's location in a relatively short period. [10][5] [3] The structure GPS system is represented by three segments: the space segment, the control segment, and the user segment. The space segment is composed of a nominal constellation of 24 operating satellites that transfer one-way signals that give the current real time GPS satellite position. The control Segment traces the GPS satellites, uploads updated navigational data, and keeps health and status of the satellite constellation. The user segment is represented by the user's GPS receiver devise which analyses data from the space and control segments to specify the target's threedimensional real-time position. [1][5] Android: Mobile Operating System Android is a mobile operating system which uses one common approach to application development. Android software designers have to develop their own applications using this open source system. users can then install these applications and use them on any android operated device, namely smart phones. [3] The main justification for choosing Android Operating System while designing this work was its huge popularity. It was announced that its market share has currently reached 48.8 and the rocketing number of android smart phone users is 91.4 in USA alone. [4] Another advantage of android system is its Application Programming Interface (API) access to the Google maps through which the required locations can be plotted on the map by the user. [10] Google Maps Google Maps supply a map of an open API (Google Maps API). Developers can easily incorporate the Google Maps service into their application. [3] The GPS supplies the smart phone devices with longitude and latitude coordinate which are then plotted on the map as symbols. Various extra properties were added for better utilization of these software's such as magnification and exploration of the surrounding areas by simply running the fingers over the mobile monitor. [5] Through the known geographic location, this application helps the parents to follow a child's mobile device. Taking benefits of this technology by using available tools and free services like Google maps and GPS, we can establish a very reliable location tracking system. The basis of this program is GPS. To develop this application Android Studio will be used. Android Studio is an enhancement tool that enables programmers to develop programs for several mobile phones. As a reference map Google Maps are used. Short Message Service (Sms) The contact between the parents and the child applications is done through SMS. SMS provides the system exceptional characteristics. It will permit the system to function in the absence of the availability of internet connection therefore permits the application to be worked on mobiles that don't support GPRS, 2G or 3G internet connectivity. The system sends the position of child's mobile to the parent's one when the parent likes to make sure on the child. [2][1] The GPS uses longitude and latitude to track the location of SMS which is used to communicate both with child side and parent side applications. [6] The system is able to send the child's mobile exact location to the parents' mobile when parent demand to check the child's location or the child sends notification for help by SMS start with code "Location". Application Development: This application is designed for both parents and children. The parents and children both should have mobiles that support GPS and SMS utilities. The system to be developed is based on a simple basis, that is, the use of SMS for interact between the family members, parents and child. It is configured in an easy manner so that it will contain few components and less user interaction, thus it is made more user-friendly. The design of Family GPS application to be carried out is depicted in figure (2), which comprises of two sides. • Parent side: This side conducts as a server for the system despite not being a real server. This alleged parent server can then utilize the SMS and available Google maps to spot the child's location. The prerequisites for this operation are availability of telephony and internet services. In the parent side application, it begins with examining the existence of the child's mobile number saved in the contacts. If the number is not available, the application notifies the user for insertion and the application will store the number in the mobile. Next, if child sent a "Location" message, the application in the parent side requests the child's position by sending an SMS to the mobile of the child. The child's mobile would answer with the current position coordinates of the child. Finally, the application will handle the coordinates and display it in the Google map of the parent's mobile. So, the parent can share the Google map with any other like police (which is the third objective), and the parent can also use GPS to determine his distance from his child. Figure (3) shows the Interface of the parents' side application and showing location of child by Google map. The child side is considered here as a client side for the system. The child side is similarly an Android mobile but is carried by the child at the time of location request. Hereby, location coordinates are derived from available GPS or network and are sent back to the parent through SMS. The difference here from the parent side is that child should also be in the range of the telephone towers regardless of the availability of internet services. On the child side, mobile and location services must be activated and are operating on the child side for the system to begin working. After the location services check, another function is run to examine if the parent's mobile number have been entered. If the number has not been entered then notify the user to enter the numbers of the parents and save them in the favorite memory. Notice that this task must only be done by parents so it is secured by a password that must be only known to the parents. Also, as we mention earlier as a second objective, in case that the child can't write SMS, he has the ability to notify his parent by takeoff head phone from his phone, and the application will automatically send the Location SMS. Conclusion And Future Work In conclusion, the proposed Family GPS application uses GPS, Android mobile and SMS in order to locating missing or lost children. The children effectively alert their parents about their position by sending SMS start with code "Location" or the SMS will be sent automatically by taking off the head phone from mobile. This location and position of children information can be obtained online from Google map. The application is cost efficient and does not need any additional device, it only needs smart phones. As a future work, the suggested system can be applied to notify parents via SMS when the children enter or leave the school, enabling the school administration and parents to keep track of the bus online, or specifying different safety zones if the child get out of them, the application will notify the parents via SMS. Acknowledgement The authors would like to acknowledge and thank University of Mosul, College of Computer Sciences & Mathematics for supporting the research work.	3,609.2	2020-05-01T00:00:00.000	[ "Computer Science" ]
A Stochastic Flows Approach for Asset Allocation with Hidden Economic Environment An optimal asset allocation problem for a quite general class of utility functions is discussed in a simple two-state Markovian regimeswitching model, where the appreciation rate of a risky share changes over time according to the state of a hidden economy. As usual, standard fi tering theory is used to transform a fin ncial model with hidden information into one with complete information, where a martingale approach is applied to discuss the optimal asset allocation problem. Using a martingale representation coupled with stochastic fl ws of diffeomorphisms for the filtering equation, the integrand in the martingale representation is identifie which gives rise to an optimal portfolio strategy under some differentiability conditions. Introduction Th optimal asset allocation problem has long been an important topic in fin ncial economics.From the practical perspective, the problem may be of importance to investment managers in banks, fin ncial institutions, hedge funds, insurance companies, and pension funds.The problem may also be of interest to individual investors and policyholders of defin d contribution pension funds.A scientifi approach to the optimal asset allocation problem was pioneered by Markowitz [1].In the Markowitz paradigm, a single-period model was considered and the problem was formulated as a mean-variance optimization problem.Merton [2,3] pioneered the study of the optimal asset allocation problem based on the maximization of an expected utility in a continuous-time economy.Using dynamic programming, Merton derived a Hamilton-Jacobi-Bellman (HJB) equation governing the value function of the problem under some diff rentiability conditions.For a particular class of utility functions, say a power utility, Merton obtained a closed-form expression for an optimal portfolio strategy which is known as the Merton ratio. Pliska [4], Karatzas et al. [5], and Cox and Huang [6] pioneered an alternative approach to the optimal asset allocation problem in continuous time.This approach is known as the martingale approach.Th key idea of the martingale approach is not unlike that of a risk-neutral valuation of a contingent claim.Firstly, an equivalent martingale measure under which the discounted optimal wealth process is a martingale is determined.Then the integrand in a martingale representation of the discounted optimal wealth process is used to identify an optimal portfolio process.For discussions on the martingale approach, one may refer to, for example, Cvitanic and Karatzas [7], Karatzas and Shreve [8], Elliott and Kopp [9], and Pham [10].Gerber and Shiu [11] considered an approach based on a tool used in actuarial science, namely, the Esscher transform, to discuss the optimal asset allocation problem.Thi approach is related to the martingale approach; see the discussion by Boyle [12]. Recently, the optimal asset allocation problem in Markovian regime-switching models has received some attention in the literature.Th rationale of considering the problem in Markovian regime-switching models is to incorporate the impact of structural changes in economic conditions on price dynamics and investment decision making.Some works on the optimal asset allocation problem in Markovian regimeswitching models are in, for example, Zhou and Yin [13], Yin and Zhou [14], Sass and Haussmann [15], Baeuerle and Rieder [16], Jang et al. [17], Nagai and Runggaldier [18], Sotomayor and Cadenillas [19], Zhang et al. [20], Elliott International Journal of Stochastic Analysis and Siu [21], Elliott et al. [22], Korn et al. [23], Siu [24][25][26], Shen and Siu [27], and others.Both situations where the modulating Markov chain is observable and where it is hidden were considered.Different approaches to stochastic optimal control such as the HJB dynamic programming approach, the martingale approach (coupled with Malliavin calculus), and the backward stochastic diff rential equation approach were adopted. In this paper we study an optimal asset allocation problem for a quite general class of utility functions in a simple twostate Markovian regime-switching model.We suppose that the appreciation rate of a risky share changes over time according to the state of a hidden economy.Th evolution of the two-state hidden economy over time is assumed to be governed by a continuous-time, two-state, hidden Markov chain, where the two states may be interpreted as "Expansion" and "Recession." As usual, the optimal asset allocation problem is discussed in two steps.Firstly, standard filtering theory is adopted to turn an economy with hidden information into one with complete information.The latter is called a "filtered" economy and is complete.Th t is, in the "filtered" economy, there is a unique equivalent martingale measure.A martingale approach is then used to discuss the optimal asset allocation problem in the "fi tered" economy.Using a martingale representation coupled with stochastic flows of diff omorphisms for the filtering equation, the integrand in the martingale representation is identifi d which gives rise to an optimal portfolio strategy under some differentiability conditions.A partial differential equation for the optimal wealth conditioning on the values of the underlying state variables is also obtained.Thi may be called an optimal wealth function which characterises a functional relationship between the current optimal wealth and the current values of the underlying state variables.The approach adopted in this paper can be used in a general -state case.However, it seems that, in the two-state case, the results of the filters and the partial differential equation for the optimal wealth are neater than those arising from the -state case.Indeed, the two-state case may not be without practical relevance.Taylor [28,29] provided some discussions on the practical relevance of using a two-state Markovian regime-switching process for modeling fin ncial returns. Th rest of the paper is structured as follows.Th next section presents the model dynamics in the original economy with hidden information and in the "filtered" economy.In Section 3, the optimal asset allocation problem in the "filtered" economy is presented and the martingale approach is used for solving the problem.The use of stochastic flows to identify the integrand in the martingale representation in the filtered market is then presented in Section 4. The final section gives a summary and suggests some possible topics for further research. Model Dynamics and Filtering Th modelling and filtering frameworks presented here are standard in the literature on optimal asset allocation in a hidden Markovian regime-switching model; see, for example, Siu [24][25][26], Korn et al. [23], Wei et al. [30], and Elliott and Siu [21], amongst others. A continuous-time economy with two primitive securities, namely, a bond and a risky share, is considered.As usual, uncertainty in the economy is described by a complete probability space (Ω, F, P), where P is a real-world probability measure.The time parameter set of the economy is given by a fin te time horizon T := [0, ], where < ∞.Th evolution of the state of a hidden economy over time is modeled by a continuous-time, two-state, hidden Markov chain {X() \| ∈ T} on (Ω, F, P).The state space of the chain is taken to be {e 1 , e 2 }, where e 1 = (1, 0) ∈ R 2 and e 2 = (0, 1) ∈ R 2 .y is the transpose of a vector or a matrix y.The state space {e 1 , e 2 } is called the canonical state space of the chain and was adopted in Elliott et al. [31].Th states "e 1 " and "e 2 " may be interpreted as "Expansion" and "Recession" of an economy, respectively.As usual, the probability laws of the chain are specifie by its intensity matrix, or rate matrix, which is defin d as where 1 , 2 > 0. A two-state Markovian regime-switching model for asset price dynamics may be justifie both theoretically and empirically (see Taylor [28,32]).Indeed, Taylor [29] pointed out that a two-state Markov chain is sufficient to distinguish a good economy from an economy experiencing distresses. Let be the constant continuously compounded rate of interest of the bond, where > 0. Then the evolution of the bond price over time is governed by For each ∈ T, let () be the appreciation rate of the risky share at time .Again it is supposed that () is modulated by the chain as where := ( 1 , 2 ) ∈ R 2 ; 1 and 2 are the appreciation rates of the risky share when the economy is in an expansion and when it is in a recession, respectively, where 1 > 2 ; ⟨⋅, ⋅⟩ is the scalar product in R 2 ; since the chain {X() \| ∈ T} is hidden, the appreciation rate () defin d as above is unobservable. Let be the constant volatility of the risky share, where > 0, and {() \| ∈ T} is a standard Brownian motion on (Ω, F, P).Then we suppose that under P the evolution of the price of the risky share over time is governed by Note that if one considers a general situation where the volatility is modulated by the hidden Markov chain as well, one may need to take into account some potential issues.Firstly, standard filtering theory may not be conveniently used to turn the economy with hidden states into one with observable states.Secondly, since is completely identifi d by the predictable quadratic variation of the price process of the risky share, it should be observable.Thirdly, an exact, finitedimensional filtering equation for the hidden Markov chain may be difficult, if not impossible, to derive if the volatility is also modulated by the hidden Markov chain.One may refer to, for example, Guo [33], Gerber and Shiu [34], Siu [24], and Elliott and Siu [21], for related discussions.Note that the price process {() \| ∈ T} of the risky share is observable.However, both the drift process {() \| ∈ T} and the Brownian motion {() \| ∈ T} are unobservable. In what follows, we will adopt standard filtering theory to turn the economy with hidden states into one with observable states. Firstly the information structure is specifie .Let F := {F() \| ∈ T} be the right-continuous and P-complete filtration generated by the price process {() \| ∈ T} of the risky share.Thi describes the flow of observable market information.Let F X := {F X () \| ∈ T} be the rightcontinuous, P-complete filtration generated by the chain X.For each ∈ T, let where For any integrable, G-adapted, process, {() \| ∈ T}, let { φ() \| ∈ T} be its F-optional projection under P.Then, for each ∈ T, Here is the expectation operator under P.It is known that the optional projection takes into account the measurability in (, ) ∈ T × Ω. Defin , for each ∈ T, Thi is the conditional, or posterior, probability that the hidden economy is in an expansion at time given the observable information up to time .We suppose that (0) = 0 , where 0 is a given constant taking a value in (0, 1).Define the process { Ŵ() \| ∈ T} by putting Then it was shown in Lipster and Shiryaev [35] that { Ŵ() \| ∈ T} is an (F, P)-standard Brownian motion.Thi is called the innovations process.The the following lemma was due to, for example, Lipster and Shiryaev [35] and Elliott [36] (see Chapter 18 therein). Lemma 1. Let := ( 1 − 2 )/.Then, under P, the conditional probability process {() \| ∈ T} is governed by the following stochastic differential equation: Under P, the price process {() \| ∈ T} of the risky share can be expressed in terms of { Ŵ() \| ∈ T} as follows: Then under P, This is used as the price process of the risky share in a "filtered" economy with complete observations.It is clear that the filtered economy is complete. Martingale Approach for Asset Allocation The aim of this section is to adapt the martingale approach to optimal asset allocation in the "filtered" economy described in the last section.The martingale approach to optimal asset allocation was pioneered by Pliska [4], Karatzas et al. [5], and Cox and Huang [6].Th mathematical basis of this approach is the martingale method for stochastic optimal control which was pioneered by Rishel [37], Duncan and Varaiya [38,39], and Davis [40].Th martingale approach has been used to discuss optimal asset allocation problems in some filtered financial models (see, e.g., Sass and Haussmann [15], Korn et al. [23], and Siu [24] and the relevant references therein).In this section, the classical convex dual arguments in, for example, Karatzas and Shreve [8], Pham [10], and Cvitanic and Karatzas [7], will be used.Unlike the Lagrange multiplier arguments, the classical convex dual arguments do not require the change of order of differentiation and integration (The author would like to thank the referee for pointing out this.).Th developments here are standard and follow those in Cvitanic and Karatzas [7] (Section 7 therein).Recall that, in the "filtered" model, the price process of the risky share under P is given by Furthermore, under P, the filtering equation is given by where (0) = 0 ∈ (0, 1). International Journal of Stochastic Analysis Let {() \| ∈ T} be a portfolio process, where () is the amount of money invested in the risky share held at time .Suppose that {() \| ∈ T} is self-fin ncing.Then under P, the wealth process { () \| ∈ T} corresponding to the self-fin ncing portfolio process {() \| ∈ T} is governed by To simplify the notation, the superscript is suppressed and write () for () unless otherwise stated.Th initial wealth Then as usual, an admissible portfolio process is defin d as follows. The for each portfolio process ∈ A 1 (V 0 , 0 ), initial wealth V 0 ∈ (0, ∞), and initial information state 0 ∈ (0, 1), the performance functional (; V 0 , 0 ) is defin d by Th objective of an economic agent is to select ∈ A 1 (V 0 , 0 ) so as to maximize (; V 0 , 0 ), that is, to solve the following optimization problem: where Φ(V 0 , 0 ) is the value of the optimization problem.Let { θ() \| ∈ T} be a bounded, F-predictable process defin d by Consider the F-adapted process {Λ() \| ∈ T} defin d by putting This is an Karatzas and Shreve [8], Page 92 therein).Then where Q is the expectation operator under Q. Lemma 3. Th random variable defined in (29) satisfies and, for all Th following theorem is a standard result and is a particular case of Proposition 7.3 in Cvitanic and Karatzas [7].Note that the form of an optimal portfolio process is identifi d in the proof of the theorem, so we present the proof following that in Cvitanic and Karatzas [7]. Th orem 4. Let be a positive, F()-measurable, integrable random variable such that Then there exists a portfolio process such that ∈ A(V 0 , 0 ) and V 0 , () = , P-a.s. From Corollary 5, it is clear that the optimal terminal wealth is given by defin d in (29) and the corresponding admissible optimal portfolio process is given by * , where * is given in the proof of Theorem 4; that is, Consequently to determine the optimal portfolio process { * () \| ∈ T}, the integrand {() \| ∈ T} in martingale representation ( 35) must be determined.In the next section the integrand {() \| ∈ T} will be identifi d using the concept of stochastic flows of diffeomorphisms. Stochastic Flows and Optimal Portfolio In this section the stochastic flows of diff omorphisms for the filtering equation will be first discussed.Then under some mild differentiability conditions the integrand in the martingale representation for the discounted optimal wealth process is identifi d which, in turn, gives rise to an expression for the optimal portfolio.The concept of stochastic flows of diff omorphisms has been used for option pricing and hedging; see, for example, Colwell et al. [41], Colwell and Elliott [42], Elliott and Kopp [9], and Elliott et al. [43], amongst others. Th following lemma gives an expression for the derivative M , ()/.( Th result follows by noting that (48) has a unique solution. Then we have the following lemma. By the unique decomposition of a special semimartingale, the predictable term of finite variation in (64) must be indistinguishable from the zero process.Consequently, The Therefore, (59) follows by noting that * (, , ) = (, , )/ − and by differentiation. Again by the unique decomposition of a special semimartingale, the integrand processes of the stochastic integrals in (64) and (65) must be indistinguishable.Consequently, Diff rentiating with respect to inside the expectation and using a version of Bayes' rule give Furthermore, Diff rentiating with respect to inside the expectation and using a version of Bayes' rule and Lemma 6 give Consequently, Expression (61) for the optimal portfolio then follows by noting that Remark 9.The partial differential equation for the optimal wealth function arising from the two-state situation is neater than that from a general -state case (Th author would like to thank the referee for pointing out this.).The former may also have some more numerical advantages than the latter.Thi may represent an advantage of considering the two-state situation. Remark 10.It is known that the martingale approach to optimal asset allocation and the risk-neutral pricing and hedging of a contingent claim are related to each other.The partial differential equation for the optimal wealth function may provide some further insights into the link between them.In particular, if the optimal terminal wealth is considered the terminal payoff of the claim, the optimal wealth function may be considered the value function of a self-fin ncing replicating portfolio of the claim and hence the price function of the claim under the no-arbitrage principle. In this case, the partial differential equation for the optimal wealth function may be interpreted as the partial differential equation for the price function of the claim. Remark 11.There are some advantages of using the stochastic flows approach comparing to the Malliavin calculus approach and the Hamilton-Jacobi-Bellman (HJB) dynamic programming approach.Compared with the Malliavin calculus approach, the stochastic flows approach does not involve the use of infin te-dimensional calculus and may be more intuitively appealing than the Malliavin calculus approach.This advantage of the stochastic flows approach in a general context has also been discussed in, for example, Elliott and Kohlmann [47,48], amongst others.Furthermore, the stochastic flows approach gives rise to a partial differential equation for the optimal wealth function, whereas this result may not be obtained using the Malliavin calculus approach.Of course, the Malliavin calculus approach does have some advantages over the stochastic flows approach.One advantage is that the former can be used in the non-Markovian situation while the latter can only be used in the Markovian situation.In this sense, the Malliavin calculus approach is more general than the stochastic fl ws approach.Nevertheless, in the filtered market considered here, the wealth process and the state probability process are jointly Markovian with respect to the observed filtration, so the stochastic flows approach is applicable in the current set up.Compared with the HJB dynamic programming approach, the stochastic fl ws approach provides a way to justify some differentiability conditions.In particular, some differentiability conditions for the optimal wealth function in Theorem 8 may be justified by the stochastic flows of diffeomorphisms, whereas the HJB dynamic programming approach itself may not be used to justify some differentiability conditions required in the verification theorem and viscosity solutions to the HJB equation may be required.Furthermore, the verification theorem arising from the HJB dynamic programming approach requires the uniform integrability condition for the value function, which may be uneasy to verify in some situations. Using the stochastic flows approach, the optimal portfolio process depends on the optimal wealth function instead of the value function.Consequently, the verific tion of the uniform integrability of the value function is not required in the stochastic flows approach.(Th author would like to thank the referee for stimulating the discussion.) Conclusions An approach based on a martingale representation and stochastic flows of diff omorphisms was adopted to discuss an optimal asset allocation problem in a simple two-state Markovian regime-switching market.Thi approach may be able to accommodate quite a general class of utility functions under some differentiability conditions.As usual, standard fi tering theory was first used to transform the original market into one with complete observations, where the latter is complete.When the values of the underlying state variables were given, using stochastic flows of diffeomorphisms for the fi tering equation and the unique decomposition of a special semimartingale, the integrand of a martingale representation for the discounted optimal wealth process was identifi d which gave rise to an optimal portfolio strategy.A partial differential equation for the optimal wealth function was also provided.The results may be generalized to the case of an -state hidden Markov chain though some complicated notation may be involved.Th approach considered here may pave a way for some potentially interesting topics for future research.Firstly, one may consider incorporating portfolio constraints in the optimal asset allocation problem.Secondly, one may extend the current model to a Markovian regime-switching jumpdiffusion model.Thi dly, as relating to the second topic, one may consider the application of the approach to study an optimal investment problem of an insurer.In this case, one needs to take into account insurance liabilities which may be modeled by Markov-modulated compound Poisson processes.Th papers by Elliott and Siu [21] and Siu [26] may provide some clues along this direction.	4,926.6	2015-01-27T00:00:00.000	[ "Economics", "Mathematics" ]
Discovery of l-threonine transaldolases for enhanced biosynthesis of beta-hydroxylated amino acids Beta-hydroxy non-standard amino acids (β-OH-nsAAs) have utility as small molecule drugs, precursors for beta-lactone antibiotics, and building blocks for polypeptides. While the L-threonine transaldolase (TTA), ObiH, is a promising enzyme for β-OH-nsAA biosynthesis, little is known about other natural TTA sequences. We ascertained the specificity of the TTA enzyme class more comprehensively by characterizing 12 candidate TTA gene products across a wide range (20-80%) of sequence identities. We found that addition of a solubility tag substantially enhanced the soluble protein expression level within this difficult-to-express enzyme family. Using an optimized coupled enzyme assay, we identified six TTAs, including one with less than 30% sequence identity to ObiH that exhibits broader substrate scope, two-fold higher L-Threonine (L-Thr) affinity, and five-fold faster initial reaction rates under conditions tested. We harnessed these TTAs for first-time bioproduction of β-OH-nsAAs with handles for bio-orthogonal conjugation from supplemented precursors during aerobic fermentation of engineered Escherichia coli, where we observed that higher affinity of the TTA for L-Thr increased titer. Overall, our work reveals an unexpectedly high level of sequence diversity and broad substrate specificity in an enzyme family whose members play key roles in the biosynthesis of therapeutic natural products that could benefit from chemical diversification. A ryl non-standard amino acids (nsAAs) that contain a hydroxyl-group on the β-carbon are found naturally in many highly effective antimicrobial non-ribosomal peptides (NRPs) such as vancomycin 1 , ribosomally synthesized and post-translationally modified peptides (RiPPs) such as ustiloxin B 2 , and industrially as small molecule antibiotics and therapeutics such as amphenicols 3,4 and droxidopa 5 .Beyond their current natural and industrial uses, aryl beta-hydroxy non-standard amino acids (β-OH-nsAAs) share structural similarity with nsAAs used for genetic code expansion 6 , a technology that has had a profound impact on chemical biology and drug development.Efficient enzymatic synthesis of stereospecific β-OH-nsAAs could pave the way for inexpensive, one-pot production of chemically diverse ribosomal and non-ribosomal peptide products (Fig. 1a).Chemical diversification is valuable for drug development for purposes such as improving cell permeability 7 , maintaining effectiveness 8 , and increasing potency 9 .Further, fermentative, one-pot production of β-OH-nsAAs could enable their integration into more complex products like NRPs, RiPPs, and proteins, which are typically produced through fermentation because of their high requirements for protein synthesis and cofactor regeneration 10 .Until recently, strategies for the biosynthesis of β-OH-nsAAs in cells were limited by restricted substrate specificity or thermodynamic favorability.Many naturally occurring β-OH-nsAAs are produced within NRP synthase complexes in which the active enzyme performing the betahydroxylation is highly specific, or post-translationally in RiPPs by hydroxylases which are poorly characterized enzymes, limiting the potential for product diversification [11][12][13] .Alternatively, threonine aldolases (TAs) are a well-established enzyme class that exhibits substrate promiscuity and have been engineered to maintain high stereospecificity for β-OH-nsAAs production [14][15][16] .However, TAs naturally favor the decomposition of β-OH-nsAAs and require high concentrations of glycine for efficient product formation, limiting their use in fermentation. A novel enzyme class known as L-threonine transaldolases (TTAs) can perform similar chemistry to TAs with low reversibility, high stereoselectivity, and high yields.TTAs are fold-type I pyridoxal 5'-phosphate (PLP)-dependent enzymes that catalyze the retroaldol cleavage of L-threonine (L-Thr) to form acetaldehyde and a glycyl-quinonoid intermediate that then reacts with an aldehyde acceptor to form a β-OH-nsAA.Interestingly, TTAs have higher sequence similarity to serine hydroxymethyltransferases (SHMTs) which naturally catalyze the formation of serine from glycine 17 .Three types of TTAs have been identified: fluorothreonine transaldolases (FTases) 18 that act on fluoroacetaldehyde acceptors; threonine:uridine 5' aldehyde transaldolases (LipK, AmbH) 19,20 that act on uridine 5' aldehyde acceptors; and L-TTAs that act on aryl aldehyde acceptors.In 2017, the TTA known as ObiH (or ObaG) was discovered as a part of the obafluorin biosynthesis pathway that natively catalyzed the aldol-like condensation of L-Thr and 4-nitrophenylacetaldehyde to produce the corresponding β-OH-nsAA (Fig. 1b) 21,22 .Since its discovery, ObiH and a 99% similar variant, PsLTTA, have been characterized to exhibit activity on over 30 aldehyde substrates as a purified enzyme and in resting cell biocatalysts, with notably little to no activity on aromatic aldehydes that contain strongly electron-donating functional groups [23][24][25][26][27][28] .In these contexts, ObiH was shown to maintain low reversibility and high stereospecificity with a preference for the threo diastereomer, the isomer found in many natural products 1,29,30 .ObiH, and TTAs more broadly, are promising alternatives to produce chemically diverse β-OH-nsAAs.While ObiH expresses well in heterologous hosts like Escherichia coli, it has reported limitations in substrate scope, has a low L-Thr affinity, and has not been studied in fermentative conditions.Further, the aldehyde substrates for ObiH are unstable and potentially toxic in live cell contexts. To address these challenges, we sought to further characterize ObiH, the products of other naturally occurring genes whose translations share similarity with known TTAs, and the ability of TTAs to form β-OH-nsAAs during heterologous expression in cells grown under aerobic conditions.At the outset of our study, ObiH, PsLTTA (a 99% similar homolog) 26 , and a promiscuous FTase (FTaseMA) 31 , were the only TTAs characterized to act on aromatic aldehydes.Furthermore, early studies 26,27 did not report testing of certain potentially useful aldehydes such as those that contain large hydrophobic moieties for cell penetration 7 or handles for bio-orthogonal click chemistry [32][33][34] .Additionally, the reported K M of ObiH for L-Thr (40.2 ± 3.8 mM 22 ) would suggest that the reaction would not proceed well in fermentative conditions without supplementation of L-Thr since natural E. coli L-Thr concentrations are low (normally <200 μM 35 ).Interestingly, LipK and FTaseMA were reported to have lower L-Thr K M (29.5 mM 19 and 1.2 mM 31 , respectively), but both are reported to have poor soluble expression in E. coli.Together, these observations offer promise for identifying a natural TTA that accepts a broad aldehyde substrate scope, has a high L-Thr affinity, and is active in heterologous host E. coli.Very few TTAs have been identified in nature, and many are likely annotated as hypothetical proteins or SHMTs based on their primary amino acid sequence. In this paper, we addressed each of the challenges associated with engineering in vivo biosynthesis of β-OH-nsAAs in a model heterologous host: low L-Thr affinity, protein solubility in E. coli, and aldehyde substrate stability (Fig. 1c).To enable rapid screening of many aldehydes and enzymes, we first optimized a high-throughput in vitro assay for characterization of TTAs on diverse aldehydes and demonstrated activity of ObiH on aldehydes that contain handles for bio-orthogonal conjugation.To explore the natural TTA sequence space, we then generated a sequence similarity network (SSN) of enzymes with high similarity to ObiH, FTase, and LipK.After appending a solubility tag to many distantly related TTAs, we observed dramatically improved enzyme expression and identified previously unreported TTAs that exhibit higher L-Thr affinity, faster reaction kinetics, and broad substrate scope.Remarkably, one of the best TTAs tested is annotated as a hypothetical protein and shares only 27.2% sequence identity with ObiH.Next, we biosynthesized β-OH-nsAAs by expressing the TTAs in cells that were engineered for aldehyde stabilization, and we coupled the TTAs to a carboxylic acid reductase (CAR) to limit toxic aldehyde accumulation.Finally, we demonstrated the activity of several CARs and a TTA in vitro and in growing cells to produce 4-azido-β-OH-phenylalanine (4-azido-β-OH-Phe), an nsAA with a wellestablished handle for bio-orthogonal conjugation.Our work brings the field closer to achieving a one-pot synthesis of chemically diverse peptides and proteins through the biosynthesis of β-OH-nsAAs in cells growing in aerobic conditions after supplementation with aldehyde or acid precursors. Results Optimizing a high-throughput assay for screening TTA activity on diverse aldehydes.To expand our understanding of the TTA enzyme class, we wanted a high-throughput method for rapid screening of multiple enzymes and candidate aldehyde substrates.We began by analyzing a previously reported coupled enzyme assay (Fig. 2a and S1) based on the addition of an alcohol dehydrogenase (ADH), which consumes NADH to reduce the coproduct acetaldehyde in a manner that can be monitored at 340 nm 19,25,36 .Unfortunately, this coupled assay for TTA activity suffers from false positives and confounding variables which we sought to address.First, the commercially available ADH from Saccharomyces cerevisiae (ScADH) exhibits activity on many aromatic aldehydes which were candidate substrates for ObiH (Fig. S2a).We briefly investigated other ADHs from E. coli to attempt to identify an alternative that might limit this undesired activity while remaining active on the desired acetaldehyde coproduct, but we did not identify a better ADH (Fig. S3).To address the false positives observed from ADH activity on the aldehyde acceptor, we optimized the concentrations of the ADH and aldehyde used in the reaction, and we introduced a control in which only the ADH and substrate were present ("no TTA").Second, the characterized TTAs are known to catalyze the decomposition of L-Thr in the absence of an aldehyde substrate, which is an undesired reaction that also generates an acetaldehyde co-product and thus another false positive 37 (Fig. S2b). To account for the background production of acetaldehyde by the TTA with L-Thr, we introduced a control in which the reaction contained TTA and ADH but lacked an aldehyde substrate ("L-Thr").Another limitation of the TTA-ADH coupled assay is that many of the aromatic aldehyde candidate substrates absorb at the same measurement wavelength which we accounted for by using low aldehyde concentrations (Table S1).With each limitation addressed, we validated the TTA-ADH coupled assay by performing high-performance liquid chromatography (HPLC) analysis, using the chemically synthesized β-OH-nsAA standard for the assumed product from 3, over a time course where we observed that the addition of the ScADH improves reaction rates three-fold (Figs.S4-6).As previously reported by others 25,36 , we were also able to improve β-OH-nsAAs yields when using the ScADH coupled to a co-factor regeneration system (Fig. S7).As the last step of verification, we screened the TTA-ADH coupled assay with ObiH before and after photo-treatment 37 .We observed no differences in reaction rate between the two phototreatment conditions and continued to assay the TTAs without photo-treatment (Fig. S8). Upon assay validation, we sought to rapidly probe the activity of ObiH on diverse aldehydes to expand the potential chemical handles of β-OH-nsAAs.We successfully screened ObiH against 16 unique substrates in a single experiment (Fig. 2b, c).We validated the activity of ObiH on substrates like the native substrate, 4-nitro-phenylacetaldehyde (15), and 2-nitrobenzaldehyde (3), which ObiH has been reported to exhibit high activity on.Our screen included nine substrates not previously tested with ObiH to our knowledge; activity on seven of these substrates was confirmed with new peak formation via HPLC or LC-MS (Figs.S9-S20).These substrates include aldehydes that contain amines, conjugatable handles, or larger hydrophobic groups to improve the chemical diversification of β-OH-nsAA products.Our results supported the known general trend 23,26 that aldehydes containing electron-withdrawing ring substituents are the preferred substrates of ObiH.As expected, the amine-aldehydes were very poor substrates for ObiH, which we hypothesize is because of the strong electron-donating potential of amines.Despite the observed trend that ObiH does not accept aldehydes containing strongly electron-donating ring substituents, we did observe that there was some activity on aldehydes with moderate electron-donating potential like 4-methoxy-benzaldehyde (9), 4-biphenylcarboxaldehyde (10), and 2-napthalaldehyde (12).Activity on larger, hydrophobic substrates is promising because these substrates can be used to modulate cell permeability for peptides.Additionally, we observed the activity of ObiH on terephthalaldehyde (7) and 4-boronobenzaldehyde (13) which both contain groups that can serve as bioconjugatable handles.With these results, we hypothesized that the TTA-ADH coupled assay can provide a broad and deep initial lens into the functional characterization of this under-explored enzyme class when used under appropriate conditions and with important controls. Bioprospecting for diverse, putative TTAs.We used bioprospecting as an approach to advance our understanding of the TTA enzyme class and potentially discover a TTA capable of overcoming the limitations of ObiH such as its low affinity for L-Thr.Using a protein sequence similarity network (SSN) that was generated with over 800 sequences produced from a BLASTp search of ObiH, LipK, and FTase, we selected 12 additional putative TTAs (Fig. 3a).We selected five putative TTAs from the same cluster as ObiH, all exhibiting >50% sequence identity to ObiH, in addition to seven randomly selected putative TTAs from clusters with 20%-30% sequence identity to ObiH 38 (Fig. 3b and S21).For one enzyme from the ObiH cluster, we arbitrarily cloned a variant to contain a 36-residue truncation from the N-terminus (StTTA-Δ36) such that its new N-terminal residue would align with the sequence of ObiH and the other candidate TTAs.RaTTA and SNTTA were selected from the cluster containing LipK, DbTTA from the cluster containing FTase, and TmTTA from the cluster containing sequences annotated as SHMTs.Lastly, three TTAs (NoTTA, PbTTA, and KaTTA) were selected from distinct clusters with no characterized enzymes.The broad range of sequence identity of candidate TTAs from 20 to 80% with respect to ObiH and to each other indicates a broader sampling of the TTA-like sequence space in any one study than past efforts to our knowledge. Upon selecting our list of candidate TTAs, we proceeded to test the heterologous expression of codon-optimized genes in E. coli for purification and in vitro biochemical characterization.Given the reported difficulty of expressing LipK and FTases 19,31 , we were not surprised to observe little to no expression of the TTAs from the clusters containing FTase and LipK; however, we also observed low expression of TTAs from unexplored clusters, and unexpectedly, two from the cluster containing ObiH.Simple nongenetic methods for improving protein expression like changing culture temperature were unsuccessful.Instead, we hypothesized that the appendage of a small solubility tag, the Small Ubiquitinlike Modifier motif (SUMO tag) 39,40 , could improve expression.We observed that the tag dramatically improved the expression of 11 TTAs (Fig. 3c and S22).To create the option of removing the SUMO tag if it were to impact activity, we cloned a TEV protease site 41 between the SUMO tag and each TTA gene.With the addition of the SUMO-tag, we successfully purified nine TTAs for further screening.Interestingly, we only observed the vibrant pink color characteristic of ObiH 22,37 with PiTTA, BuTTA, and s-KaTTA.All other TTAs had a very faint pink color or no coloration at all under the expression conditions we tested. Screening and characterization of bioprospected TTAs.Once we purified the putative TTAs, we screened them for aldol-like condensation activity.We first screened each purified enzyme with the SUMO tag fusion intact using the TTA-ADH coupled assay.Our choice to characterize SUMO-tagged proteins was well justified for three reasons: (1) the predicted structures generated with AlphaFold2 42 suggested the N-terminal region is distal from the active site for all TTAs screened; (2) the ultimate goal was to identify better homologs for expression under fermentative conditions where tag removal would be too complex or resource intensive; (3) we tested one TTA with and without the SUMO tag to verify that the tag did not impact activity (Fig. S23).We then screened each purified enzyme using the TTA-ADH coupled assay with 2-nitro-benzaldehyde, 3, the best-performing substrate from the screen of ObiH that was not a substrate of the ScADH.We observed that five enzymes (PiTTA, CsTTA, BuTTA, s-KaTTA, and PbTTA), had activity comparable to or better than ObiH (Fig. 4a).Given the activity of these distantly related enzymes and their annotation as SHMTs or hypothetical proteins, we wanted to further validate the amino acid substrate specificity of the active enzymes and further screen the inactive TTAs.We performed an in vitro assay over 20 h using 3 as the aldehyde substrate and either L-Thr, Glycine (Gly), or L-Serine (L-Ser) as the candidate amino acid.Since the TTA-ADH coupled assay is specific to L- Thr, we analyzed TTA activity via HPLC with a chemically synthesized β-OH-nsAA standard for the assumed product from 3. We confirmed that the active purified TTAs (PiTTA, CsTTA, BuTTA, s-KaTTA, and PbTTA) only act with L-Thr with no β-OH-nsAA formation using L-Ser or Gly (Fig. S24).Further, this result confirmed that after 20 h, ObiH, PiTTA, CsTTA, BuTTA, s-KaTTA, and PbTTA all approached 100% conversion of the aldehyde to the final β-OH-nsAA product.s-KaTTA and PbTTA produce almost stereochemically pure isomers of the threo β-OH-nsAA with de value of 97% and 98%, respectively, which is better than the de value of 80% for products from ObiH (Fig. S25).Of the inactive enzymes (NoTTA, TmTTA, DbTTA, and StTTA-Δ36), we observed that StTTA-Δ36 was active with the formation of the β-OH-nsAA product from 3 and L-Thr, suggesting it is too slow to detect using the TTA-ADH coupled assay.NoTTA, TmTTA, and DbTTA yielded no product, which leaves the possibility that they could be TTAs that do not accept 3 under the conditions tested or that they may not be TTAs. O b iH P iT T A B sT T A C sT T A B u T T A S tT T A T m T T A R a T T A S N T T A N o T T A K a T T A P b T T A D b T T A To explore the possibility that DbTTA and TmTTA are TTAs active on other related aldehydes, we sought to examine their activity with L-Thr and aldehyde substrates with different ring substituent positions (2), bulkier, hydrophobic chemistry (10), and aldehyde chain length ( 14) using the TTA-ADH coupled assay.Neither of these proteins appeared to have any TTA activity, nor the reported L-Thr decomposition activity (Fig. S26).We did not perform this analysis for NoTTA because we did not observe L-Thr decomposition activity, and this was predictive of inactivity on the additional substrates for both DbTTA and TmTTA. For those enzymes with comparable or faster activity than ObiH, we next sought to determine their affinity for L-Thr, which we obtained by performing the TTA-ADH coupled assay at different L-Thr concentrations and a non-saturating phenylacetaldehyde concentration of 1 mM (Fig. 4b, and S27).Notably, our assay yielded a lower K M for ObiH towards L-Thr, 29.5 mM, than the literature value (40.2 ± 3.8 mM).Two differences between our assays were the substrate, phenylacetaldehyde (14) instead of 4-nitrophenylacetylaldehyde (15), and the assay format, ADH coupling rather than a discontinuous HPLC assay.We used phenylacetaldehyde for the enzyme kinetics assay because it does not interfere with the absorbance at 340 nm, is structurally similar to the previously reported substrates for TTA screening, and is a low enough concentration to avoid observing background ADH activity.While we choose phenylacetaldehyde for this investigation, we believe this analysis could be performed with many different aldehyde substrates and may yield distinct kinetic parameters.Because a live cellular environment would also contain alcohol dehydrogenases for the reduction of acetaldehyde, it is possible that the K M values that we are measuring using the TTA-ADH coupled assay may be more realistic for our envisioned applications.Encouragingly, under these conditions, we observed that s-KaTTA and PbTTA have lower L-Thr K M than ObiH (19.1 mM and 10.9 mM, respectively).Interestingly, many of our TTAs such as PiTTA, CsTTA, BuTTA, and PbTTA have higher measured L-Thr k cat values than ObiH using phenylacetaldehyde as the aldehyde substrate (Fig. 4b).Thus, each of the characterized enzymes is faster or has higher affinity for L-Thr than ObiH does and may prove to be improved alternatives to ObiH depending on the desired application. Given the broad substrate scope of ObiH, we sought to examine a set of aryl substrates that would span the spectrum of electronic properties and include some that ObiH exhibits little to no activity on.By providing a set of seven substrates to all six TTAs, we aspired to help elucidate the landscape of specificity within this family while possibly identifying variants that exhibited higher activity or altered specificity (Fig. 4c).We specifically selected substrates with ring substituents with different electron-withdrawing properties (1, 3, 6, 7, 8), substituent size (12), and aldehyde chain length (15) to compare the activity of the putative TTAs to ObiH.We observed several interesting activities-for example, the TTAs that appeared to have higher k cat values in the ObiH cluster, such as PiTTA and BuTTA, remain relatively selective and are both reported to be a part of biosynthetic gene clusters for obafluorin 43 (Table S2).Additionally, one of the most active TTAs, PbTTA, also maintains high activity on a diverse array of substrates, originates from a different cluster of the SSN as ObiH, and exhibits low sequence identity (27.2% identity).This suggests that the TTA enzyme family may be broader than previously thought, with many more active homologs worthy of characterization for the elucidation of natural products or for applications in biocatalysis and synthetic biology. Comparative sequence analysis for characterized TTAs.To help shed some light on the potential molecular basis for substrate specificity, we performed a comparative sequence analysis of the active TTAs with a focus on known residues implicated in catalysis (H131, D204, K234) or PLP-stabilization (Y55, E107, and R366) in ObiH, as well as two loop regions that are reported to contribute to substrate specificity 37 .We performed a multiple sequence alignment across the enzymes selected and a series of characterized fold-type I PLP-dependent enzymes, including LipK from Streptomyces sp.SANK 60405 19 , FTase from Streptomyces cattleya 18 , and SHMT from Methanocaldococcus jannaschii 44 (Fig. S28).Many of the active TTAs within the ObiH cluster had the same residues at these sites.However, PbTTA and KaTTA appeared to have a modified residue at E107 which is reported to perform hydrogen bonding for PLP stabilization (Fig. 4d).This was not surprising as this residue is not conserved across related PLP-dependent enzymes.Further, we evaluated two loop regions from ObiH between Tyr55 and Pro71 (loop 1) as well as Glu355 and His363 (loop 2) that are reported to contribute to substrate specificity given their role in SHMTs as folate binding regions 45 .While loop 1 appears to be composed of different residues across the TTAs screened, PbTTA has a unique 11 amino acid insertion in the equivalent loop 1.We then aligned the published ObiH crystal structure with an AlphaFold prediction for PbTTA and observed a β-sheet within loop 1 of PbTTA (Fig. 4e).In contrast, loop 1 in ObiH is relatively unstructured and published MD simulations 37 of ObiH suggest loop 1 is highly flexible. Since this enzyme class is recently discovered, we wanted to explore the unique sequence properties of each cluster to determine if there are any distinguishing features across clusters.By examining each cluster one at a time and aligning all sequences within each cluster to ObiH, we identified that catalytic residues (H131, D204, and K234) are conserved across the clusters containing ObiH, LipK, FTase, KaTTA, and PbTTA (Fig. S29).Further, R366 is highly conserved (>90%) for all clusters analyzed.As highlighted for KaTTA and PbTTA, E107 is not conserved.For E107, each cluster appeared to have a different predominant residue in that position.Additionally, given the distinction between loop 1 of ObiH relative to SHMTs and PbTTA, we wanted to explore the sequence context of this loop region for all the clusters containing TTAs.It appears that this region is a defining characteristic for many of these clusters (Fig. S30).Each cluster appears to have on average a different length which may contribute to distinct substrate specificities. In vivo production of β-OH-nsAAs.Our last objective was to explore the biosynthesis of β-OH-nsAAs in metabolically active cells growing in aerobic conditions given our eventual desire to couple these products to ribosomal and non-ribosomal peptide formation.Production of the targeted β-OH-nsAA using cells that are growing during aerobic fermentation would need to meet three requirements: (1) Soluble expression of TTAs; (2) Affinity towards L-Thr at physiologically relevant concentration; (3) Import and stability of aryl aldehyde substrates in the presence of live cells.We hypothesized that the identified TTAs may perform better than ObiH in growing cells because the faster reaction rate of the enzyme could enable aldehyde utilization prior to aldehyde degradation by the cell.In addition, a higher affinity for L-Thr could improve titers achieved in the absence of supplemented L- Thr.Thus, we decided to test the top-performing TTAs in live cells and compare titers for different enzymes, specifically ObiH which has the highest expression, PbTTA which has the lowest L- Thr K M and highest k cat but low expression, and BuTTA which has the second highest catalytic rate with high expression.Using the SUMO-tagged constructs, each enzyme was screened in a 96well plate, fermentative conditions in wild-type E. coli MG1655 with supplementation of either 0 mM, 10 mM, or 100 mM L-Thr and 1 mM 3. We then analyzed titers after 20 h via HPLC analysis using the chemically synthesized β-OH-nsAA standard for the assumed product from 3. PbTTA performed the best with the highest titer of 0.47 ± 0.04 mM β-OH-nsAA with 100 mM L-Thr supplemented as well as the highest titer with physiological levels of L-Thr at 0.09 ± 0.01 mM β-OH-nsAA in growing cells (Fig. 5a, b).Thus, we confirmed the production of the β-OH-nsAA in growing cell cultures; however, we wondered whether we could improve titer by implementing an aldehyde-stabilizing strain. To investigate whether the knockout of genes that encode aldehyde reductases would result in improved yields of β-OH-nsAA, we transformed the plasmid that harbors our TTA expression cassette into another E. coli strain that was engineered to stabilize aromatic aldehydes, the RARE strain 46 .The RARE strain has been shown to stabilize many aryl aldehydes, including 1, 9, and 12, by eliminating potential reduction pathways 46,47 .We then repeated the experiment in the RARE strain and once again found that PbTTA produced the highest titer with 0.61 ± 0.04 mM produced with 100 mM L-Thr and 0.13 ± 0.01 mM produced with natural L-Thr levels (Fig. 5c, d).These 1.3x improvements in titer with the RARE strain suggest that stabilization of the aldehyde can improve β-OH-nsAA titers for certain chemistries, despite observing some reduction of the aldehyde to the corresponding 2-nitro-benzyl alcohol as well as reduction of the nitro-group (Fig. S31).While we did not see as large of an improvement for this chemistry as anticipated, our study suggests that the E. coli RARE strain transformed to express PbTTA is a promising chassis for β-OH-nsAA production during aerobic fermentation. Finally, to partially address the toxicity of supplemented aldehydes in fermentative contexts, we investigated whether we could couple a TTA to a CAR to create a steady and low-level supply of aldehydes biosynthesized from carboxylic acid precursors.We coupled PbTTA to a well-studied CAR from Nocardia iowensis (NiCAR) to produce a β-OH-nsAA from the corresponding acid in aerobically growing RARE.We performed an initial screen with 2 mM 4-formyl benzoic acid (8), a proven substrate for NiCAR 48 but not for PbTTA, which would install a conjugatable aldehyde group onto a potential β-OH-nsAA product.We sampled cultures for HPLC analysis 20 h after the addition of the carboxylic acid precursor and observed a peak corresponding to the β-OH-nsAA (Fig. 5e, f).Additionally, there was greater production of the β-OH-nsAA when starting with the corresponding acid precursor compared to the aldehyde substrate, demonstrating that the addition of the CAR can improve final titers.To our knowledge, we are the first to demonstrate the production of this β-OH-nsAA from either the acid or the aldehyde and we were able to produce it in aerobically growing cells.Additionally, the RARE host maintains the aldehyde e Schematic of biosynthesis of β-OH-nsAA from an acid precursor when the TTA is coupled with a CAR in the RARE strain.f β-OH-nsAA peak area for 4formyl-β-OH-phenylalanine from 4-formyl benzoic acid (8) and terephthalaldehyde (7) within the RARE strain with pACYC-NiCAR and pZE-s-PbTTA for the coupled production and RARE with pACYC-s-PbTTA, otherwise.Peak area is calculated as the area under the curve for the new peak corresponding to the product in the absorbance spectra for the appropriate wavelength from HPLC.All experiments were performed with technical triplicates.Each replicate is represented as its own data point with error bars representing standard deviations. functional handle of the β-OH-nsAA.The addition of a CAR to this cascade limits the impact of aldehyde toxicity and instability on final product titers and provides the opportunity for future β-OH-nsAA production as a de novo pathway from glucose given the natural abundance of carboxylic acids. Pathway development for a bioconjugatable β-OH-nsAA.With the promise of the CAR-TTA coupling, we wanted to investigate the generalizability of this pathway to produce a β-OH-nsAA that has a more common and versatile bio-orthogonal conjugation handle.We chose the 4-azido functionality as our target and explored whether it could be made from a 4-azido-benzoic acid precursor.To our knowledge, this precursor would be a substrate never previously tested with any CAR enzyme and its product would be a substrate never tested with any TTA enzyme.Given the prevalence of the azide group as a bio-orthogonal conjugation handle, we selected 4-azido-benzoic acid as the target substrate to produce the corresponding β-OH-nsAA product (Fig. 6a).We first studied a panel of three CARs with a diverse substrate scope and high soluble expression 48 (Fig. 6b).We observed activity of all the CARs on the acid substrate, so we then coupled the CAR directly to PbTTA in an in vitro assay to identify the β-OH-nsAA (Fig. 6c).The CAR-TTA coupling is valuable because the carboxylic acid precursor is 100-fold less costly to purchase than the aldehyde precursor and the aldehyde is likely to be toxic to cells if supplied at high concentrations.The in vitro coupling also successfully produced a β-OH-nsAA product verified as a new peak on the HPLC (Fig. S32).We did observe similar production across all CAR-TTA pairings despite the distinct activity of the CARs which suggests that PbTTA may be a limiting step in this cascade.Finally, given the potential to produce novel peptide or protein products in cells, we wanted to confirm the activity of this cascade in growing cells, which was successful for all CAR-TTA pairings with MavCAR producing the highest titer determined by product peak area after 20 h (Fig. 6d).To our knowledge, we are the first to produce a β-OH-nsAA that contains an azide functionality from either carboxylic acid or aldehyde precursors, which could be useful for chemical diversification of β-OH-nsAAs, and associated products formed by fermentation using engineered bacteria. Discussion We sought to expand the fundamental understanding of the TTA enzyme class to ultimately develop a platform E. coli strain for fermentative biosynthesis of diverse β-OH-nsAA from supplemented aromatic aldehydes or carboxylic acids.To achieve this, we had to overcome a series of challenges including low protein solubility, low activity on non-ideal substrates, and low L-Thr affinity.We successfully identified a solubility tag that improved the expression of 11 of the selected TTAs.We then expressed, purified, and tested nine previously uncharacterized enzymes at the study's outset.We successfully identified these TTAs through bioprospecting and rapid analysis of diverse enzymes via an in vitro TTA-ADH coupled assay.Of these characterized enzymes, we identified PbTTA, which expresses well in E. coli, can act on a diverse array of substrates, has a higher affinity towards L-Thr than ObiH, and has a higher catalytic rate when using 14 and L- Thr as substrates.We tested this enzyme in a series of fermentative contexts in an aldehyde-stabilizing strain and coupled it with a CAR to produce β-OH-nsAAs in aerobically grown cells.Heterologous expression in model bacteria such as E. coli is a well-documented problem for many TTAs, including LipK, and FTase 19,31 , where ObiH is the exception.The SUMO-tag appeared to improve the solubility of many enzymes that share sequence similarity to ObiH, LipK, and FTase, such that some enzymes that were unable to be expressed initially were expressed and purified.Fortunately, the SUMO-tag did not appear to impact enzyme activity for the enzymes screened, which agrees with predicted structures.Our findings and further computational predictions suggest that an N-terminal SUMO-tag may improve protein expression for similar sequences.Furthermore, our construct design facilitates the removal of the tag if needed without impacting enzyme structure. As a target enzyme for broad amino acid biosynthesis, several studies of PsLTTA and ObiH suggest a trend of limited activity on aldehydes with electron-donating ring substituents and varying activity based on the position of the ring substitution [23][24][25][26][27] .We Fig. 6 Activity of CARs and PbTTA to produce 4-azido-β-OH-phenylalanine.a Reaction scheme for the conversion of 4-azido-benzoic acid to 4-azido-β-OH-phenylalanine.b Initial rate of NADPH depletion measured for three purified CARs when provided the previously unreported candidate substrate of 4azido-benzoic acid.c β-OH-nsAA production measured by peak area for an in vitro coupled assay with the specified CAR and PbTTA.d β-OH-nsAA production measured by peak area in aerobically cultivated cells of the E. coli RARE strain transformed to express each CAR on a pZE vector and pACYC-s-PbTTA.Cultures were supplemented with 4-azido-benzoic acid during the mid-exponential phase and sampled after 20 h of growth.Peak area is calculated as the area under the curve for the new peak corresponding to the product in the absorbance spectra for the appropriate wavelength from HPLC.Experiments were performed in technical triplicate with each replicate represented.Error bars are standard deviations. observed similar trends with ObiH; however, we were able to expand the substrate scope to a variety of other substrates including those with some electron-donating properties like 4-methoxy-benzaldehyde, 9. We identified substrates with amine chemistry that appeared to be substrates for ObiH, offering an opportunity for diversification of the potential β-OH-nsAA products.Other chemistries like 4-formyl-boronic acid, 13, and terephthalaldehyde, 7, can act as bioconjugatable and reactive handles for antibiotic and non-ribosomal peptide diversification, as well as for protein engineering applications.Additionally, we wanted to determine if these trends hold for the TTAs we identified.Using a selection of aldehydes with different electronic properties, we observed that the TTAs within the ObiH cluster (PiTTA, CsTTA, and BuTTA) maintain the trends observed with ObiH.Further, we observed that PbTTA has a broader substrate scope and maintains high activity on most substrates screened, including 4-azido-benzaldehyde produced from CAR coupling. The combination of our SSN, our experiments, and our analysis using biosynthetic gene cluster (BGC) discovery tools 43 has revealed that TTAs may be much more versatile in the biosynthesis of natural or unnatural antibiotics than previously understood.The diversity of enzymes that we observed that had TTA activity suggests that there are likely many more natural enzymes capable of performing these aldol-like condensations.Additionally, the origin of ObiH, LipK, and FTase in natural product synthesis suggests that there may be other natural product syntheses that rely on this chemistry.For example, within the LipK-like enzyme cluster, there are eight published enzymes reported to be a part of several distinct nucleoside antibiotic biosynthetic gene clusters (Fig. S33).Of the enzymes we evaluated in our study, RaTTA and SNTTA are a part of predicted spicamycin and muraymycin BGCs, respectively (Table S2) 43,49 .Even with the addition of the SUMO-tag, we were only able to purify SNTTA and we observed no TTA activity on aromatic aldehydes.KaTTA, one of the active TTAs we identified, is a part of predicted valclavam BGC (Table S2).Upon further analysis, we identified OrfA and an OrfA-like protein described in the literature 50,51 that are in the same cluster as KaTTA.Interestingly, several enzymes tested and identified to have TTA activity are not a part of any known or characterized BGCs (BuTTA, PbTTA, StTTA-Δ36).This could provide an opportunity for further exploration of natural products based on the discovery of enzymes with this activity.BuTTA and PbTTA are two such enzymes that warrant further investigation into their genomic context for the elucidation of potential natural products. Finally, we successfully developed an E. coli strain for β-OH-nsAA production by using an aldehyde-stabilizing strain 46 and by coupling the TTA with a CAR for β-OH-nsAA production from an acid substrate.There are ample opportunities to explore additional aldehyde and acid substrates, develop new pathways from glucose, and improve accessible L-Thr concentrations with metabolic and genome engineering 52 .The production of diverse β-OH-nsAA in fermentative contexts should also enable the formation of complex ribosomally and non-ribosomally translated polypeptides for potential drug discovery.Ultimately, this study brings us a step closer to a platform E. coli strain for the production of diverse β-OH-nsAAs in fermentative contexts. Methods Strains and plasmids.Escherichia coli strains and plasmids used are listed in Table S3.Molecular cloning and vector propagation were performed in DH5α.Polymerase chain reaction (PCR) based DNA replication was performed using KOD XTREME Hot Start Polymerase for plasmid backbones or using KOD Hot Start Polymerase otherwise.Cloning was performed using Gibson Assembly with constructs and oligos for PCR amplification shown in Table S4.Genes were purchased as G-Blocks or gene fragments from Integrated DNA Technologies (IDT) or Twist Bioscience and were optimized for E. coli K12 using the IDT Codon Optimization Tool with sequences shown in Table S5.The following plasmids are available on Addgene with the Addgene ID listed in parentheses: P14 (204629), P15 (204630), P17 (204631), P18 (204632), P24 (204633), and P25 (204634). Overexpression and purification of threonine transaldolases.A strain of E. coli BL21 transformed with a pZE plasmid encoding expression of a TTA with a hexahistidine tag or a hexahistidine-SUMO tag at the N-terminus (P1-P26) was inoculated from frozen stocks and grown overnight in 5 mL LBL containing kanamycin (50 μg/mL).Overnight cultures were used to inoculate 250-400 mL of experimental culture of LBL supplemented with kanamycin (50 μg/mL).The culture was incubated at 37 °C until an OD 600 of 0.5-0.8 was reached while in a shaking incubator at 250 RPM.TTA expression was induced by the addition of anhydrotetracycline (aTc) (0.2 μM) and cultures were incubated shaking at 250 RPM at either 18 °C for 24 h, 30 °C for 5 h then 18 °C for 20 h or 30 °C for 24 h.Cells were centrifuged using an Avanti J-15R refrigerated Beckman Coulter centrifuge at 4 °C at 4000 × g for 15 min.The supernatant was then aspirated and pellets were resuspended in 8 mL of lysis buffer (25 mM HEPES, 10 mM imidazole, 300 mM NaCl, 400 μM PLP, 10% glycerol, pH 7.4) and disrupted via sonication using a QSonica Q125 sonicator with cycles of 5 s at 75% amplitude and 10 s off for 5 min.The lysate was distributed into microcentrifuge tubes and centrifuged for 1 h at 18,213 x g at 4 °C.The protein-containing supernatant was then removed and loaded into a HisTrap Ni-NTA column using an ÄKTA Pure GE FPLC system.Protein was washed with 3 column volumes (CV) at 60 mM imidazole and 4 CV at 90 mM imidazole.TTA was eluted in 250 mM imidazole in 1.5 mL fractions over 6 CV.Samples from selected fractions were denatured in Lamelli SDS reducing sample buffer (62.5 mM Tris-HCl, 1.5% SDS, 8.3% glycerol, 1.5% beta-mercaptoethanol, 0.005% bromophenol blue) for 10 min at 95 °C and subsequently run on a sodium dodecyl-sulfate polyacrylamide gel electrophoresis (SDS-PAGE) gel with a Thermo Scientific PageRuler TM Prestained Plus ladder to identify protein-containing fractions and confirm their size.The TTA-containing fractions were combined and applied to an Amicon column (10 kDa MWCO) and the buffer was diluted 1000x into a 25 mM HEPES, 400 μM PLP, 10% glycerol buffer.This same method was used for purification of the CAR enzymes, E. coli pyrophosphatase, E. coli ADHs, and phosphite dehydrogenase (PTDH). Threonine transaldolase expression testing.To test expression of the threonine transaldolase library, 5 mL cultures of MAJ14-26 and MAJ53-65 were inoculated in 5 mL cultures of LBL containing 50 μg/mL kanamycin and then grown shaking at 250 RPM at 37 °C until mid-exponential phase (OD = 0.5-0.8).At this time, cultures were induced via the addition of 0.2 μM aTc and then grown shaking at 250 RPM at 30 °C for 24 h.After this time, 1 mL of cells was mixed with 0.05 mL of glass beads and then vortexed using a Vortex Genie 2 for 15 min.After this time, the lysate was centrifuged at 18,213 × g at 4 °C for 30 min.Lysate was denatured as described for the overexpression and then subsequently run on an SDS-PAGE gel with Thermo Scientific™ Spectra™ Multicolor Broad Range Protein Ladder and then analyzed via western blot with an HRP-conjugated 6*His, His-Tag Mouse McAB primary antibody.The blot was visualized using an Amersham ECL Prime chemiluminescent detection reagent. In vitro enzyme activity assay-TTA-ADH coupled enzyme assay.High-throughput screening of purified TTAs was performed with a TTA-ADH coupled assay using purified TTA and commercially available alcohol dehydrogenase from S. cerevisiae purchased from Millipore Sigma.Aldehyde stocks were prepared in 50-100 mM solutions in DMSO or acetonitrile.Reaction mixtures were prepared in a 96-well plate with 100 μL of 100 mM phosphate buffer pH 7.5, 0.5 mM NADH, 0.4 mM PLP, 15 mM MgCl 2 , and 100 mM L-Thr with the addition of 0.25 mM to 1 mM aldehyde depending on the background absorbance at 340 nm (Table S1), 10 U ScADH, and 0.25 μM purified TTA unless otherwise specified.Reactions were initiated with the addition of an enzyme.Reaction kinetics were observed for 20-60 min in a SpectraMax i3x microplate reader at 30 °C with 5 s of shaking between reads with the high orbital shake setting.The following controls were included for every assay: reaction mixture without aldehyde, without TTA, and without enzyme (TTA or ADH).Rates were calculated by identifying the linear region at the beginning of the kinetic run and converting the depletion in absorbance to the depletion of mM NADH using an NADH standard curve. In vitro enzyme activity assay-CAR-TTA coupled enzyme assay.In vitro CAR activity assays were performed as previously reported 48 using 2 mM NADPH and 2 mM ATP, 20 mM MgCl 2 , and 0.75 μM CAR and E. coli pyrophosphatase.For in vitro coupling with the CAR and TTA, the same in vitro CAR assay was performed with the addition of 2 μM TTA, 0.4 mM PLP, and 100 mM L-Thr; however, rather than monitoring the reaction with the plate reader, the plate was left shaking at 1000 RPM with an orbital radius of 1.25 mm at 30 °C overnight.The reaction was then quenched after 20 h with 100 μL of 3:1 methanol:2 M HCl.The supernatant was then separated from the protein precipitate using centrifugation and analyzed via HPLC.HPLC analysis.Metabolites of interest were quantified via HPLC using an Agilent 1260 Infinity model equipped with a Zorbax Eclipse Plus-C18 column.To quantify aldehyde and β-OH-nsAAs, an initial mobile phase of solvent A/B = 95/5 was used (solvent A, water + 0.1% TFA; solvent B, acetonitrile + 0.1% TFA) and maintained for 5 min.A gradient elution was performed (A/B) as follows: gradient from 95/5 to 50/50 for 5-12 min, gradient from 50/50 to 0/100 for 12-13 min, and gradient from 0/100 to 95/5 for 13-14 min.A flow rate of 1 mL min -1 was maintained, and absorbance was monitored at 210, 250, and 280 nm. Culture conditions.For screening TTA activity in aerobically growing cells, we inoculated strains transformed with plasmids expressing TTAs into 300 µL volumes of MOPS EZ Rich media in a 96-deep-well plate with appropriate antibiotic added to maintain plasmids (50 μg/mL kanamycin (Kan)).Cultures were incubated at 37 °C with shaking at 1000 RPM and an orbital radius of 1.25 mm until an OD 600 of 0.5-0.8 was reached.OD 600 was measured using a SpectraMax i3x plate reader.At this point, the TTAs were induced with the addition of 0.2 μM aTc for TTA expression.Then, 2 h following induction of the TTAs, 1 mM aldehyde was added to the culture.Cultures were then incubated for 20 h at 30 °C with metabolite concentration measured via supernatant sampling and submission to HPLC. For the CAR-TTA coupled assay, the strains transformed with a plasmid expressing a TTA and a second plasmid expressing a CAR were grown under identical conditions with the addition of 34 μg/mL chloramphenicol (Cm) to maintain the additional plasmid.Further, 0.2 μM aTc and 1 mM IPTG were added to induce protein expression, and 2 mM aldehyde, or acid was added at the time of induction.Following induction, the cultures were grown for 20 h at 30 °C while shaking at 1000 RPM with product concentrations measured via supernatant sampling and submission to HPLC. Creation of Protein Sequence Similarity Network (SSN).Using NCBI BLAST, the 500 most closely related sequences as measured by BLASTp alignment score were obtained from three characterized threonine transaldolases, FTase, LipK, and ObiH.After deleting duplicate sequences 53 , 1195 unique sequences were obtained, which were then submitted to the Enzyme Function Initiative-Enzyme Similarity Tool (EFI-EST) 50 to generate a sequence similarity network (SSN).Sequences exhibiting greater than 95% similarity were grouped into single nodes, resulting in 859 unique nodes, and a minimum alignment score of 85 was selected for node edges.The SSN was visualized and labeled in Cytoscape 54 using the yFiles Organic Layout. Sequence alignment.Multiple sequence alignments were performed using ClustalOmega alignment within JalView 55 using the "dealign" setting and otherwise default settings of one for max guide tree iterations, and one for the number of iterations (combined).The sequence identity matrix was generated using the online interface for the Multiple Sequence Alignment tool from ClustalOmega 38 . Structure prediction.Structures of the putative TTAs were produced using AlphaFold2 CoLab notebook 56 using the provided default settings with no template, the MMseqs2 (UniRef +Environmental) for multi-sequence alignment, unpaired +paired mode, auto for model_type and 3 for num_recycles.We then moved forward with the model ranked the highest.We performed the alignment of chains A and B from the crystal structure of ObiH (PDB ID: 7K34) and the AlphaFold model for PbTTA using the align command in PyMOL with all default settings.The same alignment protocol was implemented for aligning the AlphaFold2 models of putative TTAs with and without the SUMO tag.Reporting summary.Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article. Fig. 1 Fig. 1 Threonine transaldolases are promising enzymes for the biosynthesis of chemically diverse β-OH-nsAA products.a Depiction of potential applications for β-OH-nsAAs including diversified antibiotics, genetic code expansion, and novel non-ribosomal peptides.b Depiction of the natural biosynthetic gene cluster from Pseudomonas fluorescens that is responsible for the biosynthesis of the antibiotic obafluorin.One of the key enzymes in this pathway is ObiH, a threonine transaldolase (TTA) discovered in 2017.c Schematic of what we investigated in this study: (1) ObiH activity on multiple candidate substrates; (2) Bioprospecting for candidate TTAs of lower protein sequence identity than previous efforts; (3) A genetic strategy to improve TTA expression; (4) The biochemical characterization of candidate TTAs in regard to substrate scope and L-Thr affinity; (5) The potential for TTAcatalyzed formation of beta-hydroxylated non-standard amino acids during aerobic fermentation using an engineered chassis for aldehyde stabilization. Fig.2Use of a TTA-ADH coupled assay for screening activity of ObiH on a diverse array of aromatic aldehyde substrates.a Reaction schematic for coupled enzyme reaction that enables reaction monitoring at 340 nm if appropriate conditions and controls are used.Important negative controls are no addition of aldehyde (to account for the rate of L-Thr decomposition) and no addition of ObiH (to account for potential ADH-catalyzed reduction of the aldehyde substrate).b Initial rates of ObiH on aldehyde substrates relative to an L-Thr background measurement and ADH background activity on aldehydes.The horizontal line indicates the L-Thr background decomposition observed in the TTA-ADH coupled assay.Any activity greater than the dotted line and the corresponding ADH activity is considered a successful activity of the TTA on that aldehyde.The experiment was performed in triplicate with each replicate displayed as an individual data point and error bars representing standard deviations.c Chemical structures of the aldehydes investigated in this study.Asterisks indicate substrates never previously screened with TTAs to our knowledge. Fig. 3 Fig. 3 Bioprospecting and expression of putative threonine transaldolases.a A Protein Sequence Similarity Network (SSN) containing 859 sequences related to ObiH, LipK, and FTase with selected putative TTAs highlighted in yellow.Existing enzymes characterized in the literature are highlighted in teal except those found in the largest cluster which contains many SHMTs.b Sequence identity matrix for all selected TTAs in this study.c Western blot of all TTAs with the tagged and untagged TTA constructs demonstrating improved expression of TTAs with a SUMO solubility tag.Proteins that contain an N-terminal SUMO tag followed by a TEV protease cleavage site, and no other changes, are shown in lanes indicated by the 's'. Fig. 4 Fig. 4 Characterization of putative threonine transaldolases.a Screen of all purified TTAs using the TTA-ADH coupled assay on 3. The experiment was performed in triplicate with each replicate as an individual point.Error bars represent standard deviations.b Apparent L-Thr K M and k cat measurements for TTAs that exhibited activity greater than or equal to ObiH calculated using non-linear regression.Parenthetical values represent the 95% confidence interval.c Heatmap showing initial rates for six active TTAs against multiple aromatic aldehyde substrates.d Multi-sequence alignment of the predicted conserved catalytic residues for the six active TTAs.e Superimposed structure and predicted structure illustrating the Tyr55-Pro71 loop region of ObiH compared to the predicted equivalent region for PbTTA.ObiH is represented in green with the loop region highlighted in yellow with the PLP highlighted in gray indicating the region of the active site.The PbTTA is indicated in blue with the corresponding loop region highlighted in pink. Fig. 5 Fig.5Biosynthesis of β-OH-nsAAs in metabolically active cells during aerobic fermentation.a Schematic of β-OH-nsAA biosynthesis with supplemented aldehyde in a wild-type E. coli strain.b β-OH-nsAA titer measured after 20 h for s-ObiH, s-BuTTA, and s-PbTTA with 0, 10, and 100 mM of L-Thr supplemented and 1 mM 3 supplemented.c Schematic of β-OH-nsAA biosynthesis with genomic modifications to improve aldehyde stabilization.d β-OH-nsAA titer measured after 20 h for s-ObiH, s-BuTTA, and s-PbTTA with 0, 10, and 100 mM of L-Thr supplemented and 1 mM 3 supplemented.e Schematic of biosynthesis of β-OH-nsAA from an acid precursor when the TTA is coupled with a CAR in the RARE strain.f β-OH-nsAA peak area for 4formyl-β-OH-phenylalanine from 4-formyl benzoic acid(8) and terephthalaldehyde(7) within the RARE strain with pACYC-NiCAR and pZE-s-PbTTA for the coupled production and RARE with pACYC-s-PbTTA, otherwise.Peak area is calculated as the area under the curve for the new peak corresponding to the product in the absorbance spectra for the appropriate wavelength from HPLC.All experiments were performed with technical triplicates.Each replicate is represented as its own data point with error bars representing standard deviations. Statistics and reproducibility.All experiments were performed in either biological or technical triplicate as reported in the figure legends.Triplicate data is represented in each figure as separate data points and error bars represent standard deviations across triplicate values.Some full experiments were performed in replicates on separate days to confirm reproducibility.The data represented within the manuscript is that of a single day's experiments that is representative of all replicates. 2 Use of a TTA-ADH coupled assay for screening activity of ObiH on a diverse array of aromatic aldehyde substrates.a Reaction schematic for coupled enzyme reaction that enables reaction monitoring at 340 nm if appropriate conditions and controls are used.Important negative controls are no addition of aldehyde (to account for the rate of L-Thr decomposition) and no addition of ObiH (to account for potential ADH-catalyzed reduction of the aldehyde substrate).b Initial rates of ObiH on aldehyde substrates relative to an L-Thr background measurement and ADH background activity on aldehydes.The horizontal line indicates the L-Thr background decomposition observed in the TTA-ADH coupled assay.Any activity greater than the dotted line and the corresponding ADH activity is considered a successful activity of the TTA on that aldehyde.The experiment was performed in triplicate with each replicate displayed as an individual data point and error bars representing standard deviations.c Chemical structures of the aldehydes investigated in this study.Asterisks indicate substrates never previously screened with TTAs to our knowledge.	11,877.6	2023-09-11T00:00:00.000	[ "Chemistry", "Biology" ]
ROS-Unity3D based system for monitoring of an industrial robotic process Planning and monitoring the manufacturing of high quality one-of-a-kind products are challenging tasks. In the implementation of an industrial system, the commissioning phase is typically comprised of a programming phase and an optimization phase. Most of the resources are commonly invested in the optimization of the process. The time and cost of the implementation can be reduced if the monitoring system is not embedded in the industrial process, but kept instead as a decoupled task. In this paper we present a framework to simulate and execute the monitoring task of an industrial process in Unity3D, without interfering with the original system. The monitoring system is made of external additional equipment and is decoupled from the industrial task. The monitoring robot's path is subject to multiple constraints to track the original process without affecting its execution. Moreover, the framework is flexible thanks to the Unity-ROS communication so that the monitoring task can be carried on by any ROS-compatible device. The monitoring system has been applied to a robotic system for heavy, multi-pass TIG welding of voluminous work-pieces. The results of the implementation show that the constraints for monitoring were satisfactory in the 3D environment and capable for real robot application. I. INTRODUCTION For complex industrial processes that produce high quality, one-of-a-kind products, planning is one of the most timeconsuming phases. However, during the execution of the process constant and accurate monitoring is necessary to ensure that what has been planned, simulated and tested is accurately reproduced on the real workpiece. The methods of classical automation are therefore not suitable for this kind of production. Processes are difficult to adapt, and the complex commissioning phase prevents companies to react to these market demands in time. Automation and industrial robotics however allow to automate such complex processes while maintaining a high level of flexibility. While there are many different industrial applications that reflect this scheme, [1], [2], [3], in this paper we will focus on the particular processes of heavy grinding and welding as they have been used as test cases for our implementation. In the industry, for both rigid and flexible automation solutions, the commissioning phase remains one of the most expensive parts of the process [4], [5]. During the commissioning phase the optimization is the most time expensive task, where the process is tuned and calibrated with respect to all the controllable parameters. If the optimization phase was too much time consuming, it would be hardly justified for one-of-a-kind types of production since the investment wouldn't be supported by a high-volume production. This is one of the reasons why it is not feasible to embed the monitoring system in the industrial process itself. Moreover the monitoring process is mainly used to optimize the industrial system, and with an independent monitoring solution it can be re-used on other applications once it fulfils its purpose. The framework described in this paper aims at providing the tools to simulate and test the monitoring strategy with an industrial robot for heavy welding and grinding task, in order to shorten the time of the optimization phase and make the commissioning phase more efficient. Figure 1 shows the current setup for the welding task in our lab. It is also important to notice that the remote monitoring system serves also as a means to provide external assistance on systems which are fully operational. Therefore, after the commissioning phase a system might benefit from external monitoring to evaluate the product quality even though it was not originally designed to include such a system. Nonetheless, the framework and the results presented in this work are not to be seen as constrained to heavy welding and grinding processes. In fact, the monitoring system described in this paper is independent of the robotic hardware and also independent of the industrial process being monitored. Multi-modal feedbacks can now be more easily combined to communicate much more information about the real system and also to provide more immersive experiences when we interact with the digital environment [10]. In fact, the gap that exists between the user's actions in the digital world and their effect on the real system is slowly but steadily narrowing. Concretely, the intention to later integrate in our framework multi-modal interactions for man-machine communication is reflected in the choice of Unity [11] as the component in charge of displaying the digital models and interface with the user. Although not the only possible option, the game engine Unity is supported by an active community and a well grounded ecosystem, besides being among the top choices for AR/VR development [11]. In Unity we design the user interface and the monitoring process in a 3D environment before having the task performed by a real robot. In contrast to a simulation software (e.g. Gazebo [12]) Unity3D is not conceived to accurately reproduce real world scenarios in terms of dynamics and physics, but rather to visualize models in a 3D environment and interact with them just as it is the case in games development. The main advantage thus is in the flexibility to design the user interface and the user's interactions with the virtual world. The second core component is the well known Robot Operating System (ROS), [13] which is used to directly communicate with all the hardware components which are not directly involved in the UI design, such as robots and sensors. ROS allows the framework to interface with any ROS-compatible robot without the need to change the control strategy or the monitoring task. In this paper we present a framework to simulate the monitoring task of an industrial process in Unity3D. The monitoring process is subject to multiple constraints in order not to interfere with the industrial task. Moreover, we discuss the flexibility of the framework that allows for easy deployment of the monitoring task on a real robot. II. RELATED WORK Due to the flexibility we intend to give to our framework, there are many different research contributions that need to be acknowledged and approaches that have to be mentioned. As previously stated, the industrial process of welding treated in this paper should not pose a limitation in the conclusions we want to draw. The framework is to be seen as independent of the particular robotic task being treated here, but we shall mention related work for similar processes in order to provide better context to the need of a monitoring system for the entire process. In contrast to high-volume fully automated production systems where monitoring is mainly used for quality control, in smaller batches production systems there is often the need for continuous monitoring. In [14] Pfeifer et al. describe the advantages of an inspection system along a micro-assembly line. Buschhaus et al. present in [15] monitoring of a robotbased process for the metallization of three dimensional molded interconnected devices. The monitoring is crucial in this processes as it serves as in-line correction method in order ensure high quality results. The implementation of a monitoring task for a robot welding application is a continuation of the concept presented in [16]. In their work, Zimber et al. discuss how an autonomous industrial manipulator (AIMM) can be used for monitoring an industrial process. Ultimately, due to the increasing demand from SMEs of remote support solutions for their industrial robotic systems, a monitoring AIMM can significantly improve remote maintenance and assistance. The separate solution brings significant advantages, since it doesn't interfere with the industrial process. In fact, if additional sensors have to be integrated in the industrial process they have to be included in the design process and their presence can redefine the optimization process of the task itself. If additional equipment or additional support tasks are not included in the design process, it is possible that they cannot be integrated at all without an immense investment of resources. The monitoring robot can be programmed also after an industrial process optimization phase and it doesn't require modifications in the original system. It serves as a less expensive, quicker-tointegrate external equipment. For what regards the ROS-Unity communication and its advantages, there are several papers that explored the potential of such connection and that in general investigated the potential of Unity for designing Human Robot Interface (HRI). The work of Bartneck et al. in [17] is one of the first papers advocating the user friendliness of Unity for the design of human robot interaction. One of the main arguments of the paper is that programming robot behaviours and interactions is easier in Unity due to the presence of a set of tools for animations and visual programming used in game development. However, in their work they decide not to involve any middle-ware solution for robotic hardware and implement all the communication logic and HRI within the Unity environment. Other works ([18] [19]) that followed explored further the possibility of using ROS as middleware solution, while still managing the HRI in Unity or similar software. The research works mentioned here are in the field of telepresence and teleoperation where multi-modal user interactions are essential. Nonetheless we mainly cite such contributions to highlight the flexibility and modularity allowed by the connection between Unity and ROS. In [20] Codd-Downey et al. proposed an architecture Unity-ROS to control a mobile robot in virtual reality. Furthermore, Pan et al. [21] proposed an approach for simulating a robotic welding task in Unity. As previously mentioned, Unity is a game engine and not originally thought as a simulation software and therefore lacking proper tools to include the dynamics of the system and accurate hardware parameters. As the authors point out, such system could be beneficial for educational purposes and training. From a slightly different perspective, we see the Unity environment as where, besides the user interactions, some of the higher level logic is processed and then communicated to the hardware through ROS or other dedicated channels. Concretely, by receiving real-time information about the state of the system we can animate the 3D environment accordingly and apply constraints on the monitoring robot's motion. III. SYSTEM DESCRIPTION The framework can be divided into two parts: • The industrial process (i.e. the robotic welding task) • The monitoring process The system in charge of performing the welding task is based on the work presented by Horvath et al. in [22] and shown in Fig. 2. Furthermore, the welding process data are communicated to ROS and thus made "accessible" to the monitoring process, which is based in Unity and communicating with hardware through ROS. The general architecture of the whole framework is shown in Fig.3, where the connection ROS-Unity is the main bridge between the different subparts of the system. The monitoring process is directly linked to Unity as it is designed and implemented with the game engine. It is also worth mentioning that the monitoring task is linked to multi-modal manmachine communication to reflect the HMI design process that takes place in Unity. In fact, it is possible to integrate different types of feedbacks (tactile, audio, visual) into the same digital environment without the need of additional software. The monitoring task should be only partly automated, in the sense that the user should have the freedom to adjust the view to his/her needs without having to worry not to interfere with the welding process currently ongoing. In this context, multimodal feedbacks can increase the user's comfort when he/she takes control over the monitoring robot. Furthermore, the connection with multi-modal communication also reflects our intent to eventually interface the system with VR/AR equipment to investigate immersive telepresence applications. The 3D environment use for the simulation is created by importing the CAD models of the robot and the workpiece involved in the welding process. Such models are then placed in Unity along with the 3D model of the robot used for the monitoring task, as it is shown in Fig.4. The welding application provides the data about the welding path, which can be displayed into Unity and visualized together with the workpiece. The reference frames of the welding robot, the torch and the workpiece are also communicated by the welding application to Unity through ROS. Once the welding path is available in Unity, the monitoring program calculates the path for the observation taking into account the torch orientation and the welding robot configuration in order to avoid collisions with the system. The default monitoring strategy consists of simply following the welding process while keeping the welding torch and the part of the workpiece being machined inside the camera's field of view. However, sometimes this strategy is not the most desired one by the user, which should be then allowed to tweak and adjust the camera position if necessary. Therefore the Unity scene allows for user commands that modify the camera position and orientation while tracking the ongoing process. It is then possible to observe in the 3D environment the welding robot moving according to the real-time joint values provided by the industrial process, furthermore the user can see through the camera view of the simulated monitoring robot and observe how the process while it's performed. The details of the actual implementation of the framework and the monitoring scene in Unity will be discussed in the following section. IV. IMPLEMENTATION This section treats the implementation of some sub-parts of the system, mainly regarding the communication between all elements. The last part of this section describes how the animation/control of the 3D model is carried out in Unity. A. ROS In the implementation of the system, the communication between Unity, ROS and the welding process is based on the following elements: • C# Rosbridge for Unity-ROS. This script establishes the connection and allows for invoking Ros services • Rosbbridge script for Welding process and ROS. This script allows the welding system software to publish data onto topics. • Ros topics of the welding process. Currently the data published are welding path positions, workpiece reference frame and robot reference frame Regarding this specific welding process the industrial system doesn't allow for external control commands, meaning that Unity can only fetch the real-time information to synchronize the monitoring task but cannot interfere with the ongoing welding operation. B. Joint Reading It is thus important for the monitoring robot to receive at runtime the welding robot configuration, that is its joint values. The welding robot used in our system is a NACHI MC-50, while the robot model used for simulating the monitoring is a NACHI MZ-04 and each robot has 6 DOF. Although the welding robot is ROS-compatible, in our implementation we exploit a different communication channel to receive the robot's encoder values at runtime. In fact, we use a Rasperry-Pi connected to the robot to read the encoder values through UDP communication. The implementation details of such device are not the subject of this paper, but for the sake of clarity Unity receives precise encoder values of the welding robot through UDP communication. Even though the Fig. 4. The Unity environment and the 3D visualization of a welding path. The 3D model of the object has been hidden to better visualize the path. In the detail it is shown the monitoring robot with the visualization of the IK solver constraints. device allows for joint control, as previously mentioned this capability is disabled for Unity since the monitoring process is not allowed to control the welding task. The encoder values received in Unity are then converted into joint angles ([rad]) with double float precision. The conversion is defined by the formula: where i is the joint index, Enc i is the encoder value of joint i received through UDP connection and the parameters Θ of f set i , Enc of f set i and π i are constants obtained from the robot controller beforehand. The update frequency of the encoder values has a limit of 5ms (both for writing and reading), which limits the Unity maximum frame rate to 200 fps. However, such a limit is more than enough for real time application and does not constitutes a bottle neck in our system. In the 3D simulation we display the robotic cell thanks to the CAD models obtained from the industrial process. The digital welding robot is synchronized with the joint values coming from the real robot. We assume that the monitoring robot is equipped with a camera mounted on the end effector and we use the end effector's reference frame for the camera's orientation. C. Animation in the 3D Environment When the welding starts, the monitoring robot starts tracking the welding torch by keeping it in the camera's field of view. The robot is animated via a c# IK solver based on the work presented in [23] and made available as a Unity-plugin. In addition to the field-of-view constraint, the monitoring robot needs to take into account the following constraints: • The distance from the torch must not be lower than a certain threshold; • Avoid collision with the welding robot; • Keep the welding torch in the centre of the camera; Each frame, the encoder values received from the Raspberry-Pi are converted into radians and used to update the position of the 3D model. Then, the monitoring algorithm enforces the constraints on the 3D model of the MZ-04 and then compares the newly calculated joint values with the ones of the previous frame. If there is a difference between the two frames it proceeds by performing the inverse conversion to obtain the corresponding encoder values that the actual robot should reach. Finally, the new encoder values are sent through UDP to the Raspberry-Pi connected to the MZ-04. This process is summarized in Algorithm 1. In Algorithm 1, n dof is the number of joints of the monitoring robot. Moreover, it is worth observing how Step 6 of the algorithm implies that the IK-solver modifies the pose of the 3D model, according to the objectives that are active in that frame. V. RESULTS AND DISCUSSION The system has been evaluated in a simulation conducted entirely in Unity. The model of the welding robot was programmed to move along a test path (see green line in figure Convert θ i from radians to degrees 4: Update the 3D model of the i-th joint 5: end for 6: Enforce the IK-solver constraints 7: for i = 1 to n dof do 8: Obtain the new θ i in degrees 9: Convert θ i from degrees to radians 10: Obtain EN C i with the inverse conversion 11: Store EN C i in the array EN C new 12: end for 13: return EN C new 4, and three main objectives were set on the model of the monitoring robot: welding torch had to remain in focus (lookat constraint); maximum distance between the welding torch and the monitoring robot's end effector; collision avoidance with the welding robot. Every run consisted of the welding robot performing the path once (back and forth), while we observed the behaviour of the monitoring robot. During the monitoring simulation the robot's configuration could occasionally fluctuate due to the multiple objective optimization. In fact, since the optimization algorithm is based on GA, the robot might move from its current configuration to one with a higher fitness. However, a monitoring simulation is considered successful when the main objectives presented in section IV are satisfied. This means that two successful simulations may have slightly different monitoring paths, but they both accomplish collision avoidance while keeping track of the welding torch and the workpiece. Given a specific instance of the objective, we are not interested in the global optimum within the given search space, but rather a sub-optimal solution in a limited time frame (since the search is computed at run-time). The main reason why we considered different solutions acceptable is due to our intention to include also commands given by the user to control the monitoring view. Therefore, since in the future the monitoring path will be modified at runtime by the user's actions, the system must allow for some flexibility in the robot configurations. In the welding task considered in this paper we did not incur situations were one or more of the objectives could not be satisfied. However, it is important to consider such cases to prevent unexpected behaviours from the monitoring robot. In fact, when not all objectives can be satisfied the robot might jump between configuration that optimize different objective that however share similar weights. In order to prevent these fluctuations, we decided to implement an agent in charge of supervising the IK solver at runtime. Concretely, in the event of configurations which do not fulfil one or more constraints this agent will add a special constraint to the optimization function of the IK solver. The additional objective is called "displacement objective" and its sole purpose is to punish all new configurations found which are "distant" from the current one in terms of joint space. It is important to observe that the agent is also ensuring that the objective are satisfied with the same priority with which they have been listed in the previous section. This is achieved by changing the weights at runtime in a fashion that consistently reflects the aforementioned order. Thus, with the assumption that the priorities are kept intact, the displacement objective ensures that robot is not "jumping" to a new configuration which is significantly different from the current one, even if the overall fitness of the solution would improve. The experiments in simulation show that the monitoring robot is capable of tracking the welding torch without specific knowledge of the welding path (the trajectory was only known by the welding robot model). In the bigger framework, it helps proving that such a model-based approach is suitable for remote monitoring of an industrial task. In this work, the monitoring task has been implemented entirely in simulation, checking that the constraints were satisfied in the 3D environment. However, it is possible to implement the very same simulation on a real robot and this will be part of our future works. The intention is to exploit the UDP communication that has been used to synchronize the system with the industrial process, and use it this time for joint control of the monitoring robot. In this context we will conduct tests to assess the capability of the system to decrease the time for troubleshooting compared to a situation where monitoring was absent. VI. CONCLUSION AND FUTURE WORK In this paper we presented a framework for robotic monitoring of an industrial process. The key achievements of this work are the following: • Remote monitoring system for an industrial robotic process. • Flexibility of the system due to ROS-Unity communication. The monitoring can be executed with any ROScompatible hardware. • Non-invasiveness of the remote monitoring. The parameters of the industrial process remain unmodified and the monitoring equipment can be introduced without compromising the welding task. • Compact solution to set up a monitoring strategy. The monitoring robot is controlled in the same framework that provides the camera view. The framework has been used for the planning and evaluation of the monitoring strategy on the welding application. One of the objective is to move toward a shorter set up time thanks to the decoupling from the original process. We are currently running tests in our lab in order to collect more data. The system provides a more flexible compared to an embedded monitoring solution that would have to be designed taking into consideration the welding path and the welding equipment, and that couldn't be re-used on different installations. Finally, we aim at extending our framework for multimodal man-machine communication (4MC) and VR/AR devices for remote monitoring.	5,510.4	2017-12-01T00:00:00.000	[ "Engineering", "Computer Science" ]
Design of Automated Real-Time BCI Application Using EEG Signals This study proposed a design of real time BCI application using EEG recording, pre-processing, feature extraction and classification of EEG signals. Recorded EEG signals are highly contaminated by noises and artifacts that originate from outside of cerebral origin. In this study, pre-processing of EEG signals using wavelet multiresolution analysis and independent component analysis is applied to automatically remove the noises and artifacts. Consequently, features of interest are extracted as descriptive properties of the EEG signals. Finally, classification algorithms using artificial neural network is used to distinguish the state of EEG signals for real time BCI application. Introduction Electroencephalogram (EEG) is the recording of electrical activities of human brain using electrodes attached to the scalp.Conventionally, EEG is used for clinical diagnosis of epilepsy and sleep disorder.In recent decades, EEG have been studied and finding increasing use in Brain Computer Interface (BCI) application.The EEG signals are described in frequency bands of delta (0.5 to 4 Hz), theta (4 to 8 Hz), alpha (8 to 12 Hz) and beta (12 to 32 Hz), each of which is attributed to different aspects of brain activity. In practical settings, EEG signals are often contaminated by noises and artifacts during the recording.There are two types of artifacts that contaminated the EEG signals, namely, biological and environmental artifacts. 1,2iological artifacts are signals arising from non-cerebral origin of the human body, such as cardiac, ocular or muscles activity.On the other hand, environmental artifacts are signals that originate from outside of human body, such as interference from external devices.Artifacts recorded in EEG signals distorted power spectrum and influenced the decision-making process of BCI application.Therefore, a pre-processing step to remove the noises and artifacts of EEG signals is necessary.and signal artifacts. 3,4Wavelet based multiresolution analysis using Discrete Wavelet Transform (DWT) is more effective in removing the target artifacts, while retaining the cerebral activities of interest in EEG signals. 5,6Meanwhile, Independent Component Analysis (ICA) algorithm using blind source separation is able to isolate the target artifacts into a separated Independent Component (IC). 2,7Combinatorial use of Wavelet Multiresolution Analysis (WMA) and ICA are able to isolate and remove noises and artifacts without incurring substantial loss of cerebral activities of interest in EEG signals. 1,8 this study, we designed an automated pre-processing step using WMA and ICA to remove noises and artifacts in EEG signals.Then, feature extraction and classification of EEG signals is applied for real time BCI application. Wavelet Multiresolution Analysis WMA incorporates the steps of DWT and inverse DWT. 8 The DWT is an implementation of wavelet transform using discrete set of wavelet scales and transitions. 6,9WT consists of sequential application of low-and highpass filters to decompose a discrete signal into multiple wavelet components, as shown in Figure 1.Here, x[n] represents a channel of EEG signal passed through a low pass filter, g[n] and a high pass filter, h[n] simultaneously.This process is repeated until each channel of the EEG signal is decomposed into n levels of wavelet details, i.e.D1(t), D2(y), ..., Dn(t) and a mother wavelet of An. On the other hand, inverse DWT is applied in a similar but reversed sequence by combining wavelet details and mother wavelet into a discrete EEG signal. Independent Component Analysis ICA model describes multivariate signals as a mixture of its source components, by assuming that the multivariate signals, x are separable into their statistically independent and non-Gaussian source components, s.The relationship between the signals and its source components is described by the equation x As (1) In equation ( 1), A is the unknown mixing matrix estimated by using the ICA algorithms. 7,10The un-mixing matrix, W is then computed as the inverse of estimated mixing matrix.The source components, s are revealed by using the equation s Wx (1) On the other hand, inverse ICA is accomplished by multiplying the inverse of estimated mixing matrix, W -1 with the source components, s. Artificial Neural Network Artificial Neural Network (ANN) is a computational model based on interconnected adaptive neurons that resemble the biological neural network. 11Each processing neurons operating in parallel receives input from the input layer or its preceding tier and further the processed information to its successive tier or output layer.ANN can be described by the transfer functions of their neurons, the learning rules and the connection formula. 12ANN are adaptive and able to learn by observing and weighting on the importance of input datasets.In supervised machine learning, ANN model trained with sufficient number of training data can be used to make prediction or classification on test data to determine the dataset in which the test data might belong. Design of System This study proposed a design of real time BCI application using the following steps: EEG recording, pre-processing, feature extraction and classification.The design of the system is illustrated in Figure 2. EEG Pre-processing Automated pre-processing step is applied to the recorded EEG signals using the following steps: (1) Wavelet Multiresolution Analysis is applied to decompose each channel by DWT to 8 levels with mother wavelet of db8. 5 Then, only the wavelet components of D3 to D8 corresponding to the frequency range of 0.5 to 32 Hz is retained to remove unwanted noises and artifacts outside the frequency range of interest.(2) Blind source separation using ICA is applied to isolate the artifactual components, in this case, the eye blink artifacts to be removed by wavelet denoise algorithm. 1,13(3) Separated ICs and wavelet components are recombined using inverse ICA and inverse DWT respectively to reconstruct the clean EEG signals as shown in Figure 4. Feature Extraction The pre-processed EEG signals are separated into 1 second epochs and features of interest are extracted by using DWT.Each channel of the EEG signal is decomposed to wavelet details each represent the frequency bands of EEG signals defined as delta (0.5 to 4 Hz), theta (4 to 8 Hz), alpha (8 to 16 Hz) and beta (16 to 32 Hz) bands.The coefficients of the wavelet details are taken as features that characterize the properties of the EEG signals in each corresponding frequency band. Classification Classification of EEG signals is conducted using ANN.The ANN model is trained using both target and nontarget datasets and applied to classify the test data in real time BCI application.In this study, we applied the system to classify target state of "alert with eyes-open" and nontarget state of "relax with eyes-closed". 14By using 1190 epochs of data and 10 layers of hidden neurons, an overall accuracy of 97.6 % calculated using 10 fold cross validation is achieved as shown in Table 1.P -705 Discussion The Wavelet based multiresolution analysis are applied as wavelet decomposition is more effective in preserving the structure of the EEG signals in both time and frequency domains. 5,13Mother wavelet of db8 is selected due to its balanced performance and computational simplicity for real time application. 5ter the pre-processing steps, feature extraction using wavelet decomposition is applied to extract the features of interest.The features of interest are the coefficients of wavelet details each corresponding to frequency bands activities of delta, theta, alpha and beta band.The wavelet coefficients are computed due to its better performance in characterizing the structure of EEG signals.Lastly, supervised machine learning using trained ANN model is applied for classification of EEG signals.We are eager to apply the system in BCI applications in real time setting. Conclusion This study proposed a design of automated real time BCI application using EEG signals.EEG signals are notably noisy and contaminated by artifacts.The proposed pre-processing steps using WMA and ICA effectively removed the noises and artifacts with minimal distortion to the cerebral activities in frequency bands of interest.Then, feature extraction is applied using DWT to extract the coefficient of wavelet details corresponding to each frequency band of interest.Lastly, classification algorithm using ANN are applied to classify the EEG signals in real time BCI application. Fig. 1 . Fig. 1.Block diagram of Discrete Wavelet Transform (DWT) of a discrete signal, x[n].The annotation ↓2 denotes sampling reduction by a factor of 2, i.e. two-fold down-sampling of the signal. Fig. 4 . Fig. 4. Similar segment of EEG signals after pre-processing step is applied.Eye blink artifacts and high frequency noises are removed while underlying cerebral activities within frequency range of interest (0.5 to 32 Hz) is retained. Table 1 . Classification result for "alert with eyesopen" and "relax with eyes-closed" EEG signals. proposed system design function to automatically filter and classify target data of EEG signals in real time BCI application.Recorded raw EEG signals are preprocessed using a combination of WMA and ICA to remove noises and artifacts while retaining cerebral activities in frequency bands of interest.The preprocessing steps are important as EEG signals are highly contaminated by noises and artifacts in a practical setting.	2,027.2	2017-01-19T00:00:00.000	[ "Computer Science" ]
Measurements of the vector boson production with the ATLAS detector Measurements of the Drell-Yan production of W and Z bosons at the LHC provide a benchmark of our understanding of perturbative QCD and probe the proton structure in a unique way. The ATLAS collaboration has performed new high precision measurements at a center-of-mass energy of 7 TeV. The measurements are performed for W, W− and Z bosons integrated and as a function of the boson or lepton rapidity and the Z mass. Unprecedented precision is reached and strong constraints on Parton Distribution Functions, in particular the strange density are found. Z boson cross sections are also measured at center-of-mass energies of 8 TeV and 13 TeV, and cross-section ratios to the top-quark pair production have been derived. This ratio measurement leads to a cancellation of systematic effects and allows for a high precision comparison to the theory predictions. The production of jets in association with vector bosons is a further important process to study perturbative QCD in a multi-scale environment. The ATLAS collaboration has performed new measurements of Z boson plus jets cross sections, differential in several kinematic variables, in proton-proton collision data taken at a center-of-mass energy of 13 TeV. The measurements are compared to state-of-the art theory predictions. They are sensitive to higher-order pQCD effects, probe flavour and mass schemes and can be used to constrain the proton structure. In addition, a new measurement of the splitting scales of the kt jet-clustering algorithm for final states containing a Z boson candidate at a center-of-mass energy of 8 TeV is presented. Introduction The production of W and Z bosons via the Drell-Yan process is widely studied at the LHC experiments with proton-proton collision, including the ATLAS experiment [1].Those processes have a clear signature (especially in the lepton decay channels: Z → ll and W → lν) and are characterized by small background contamination.Due to the high production cross section, it is possible to collect large statistics samples of events containing vector bosons. W and Z boson events, in turn, are common backgrounds in Standard Model measurements, including Higgs boson measurements, and in new physics searches.As a consequence, an improved description of the vector boson production has a positive impact in many ATLAS physics analyses. Nevertheless, studying in details the production of vector bosons is useful to test the Standard Model accuracy and extract its parameters.It is possible to probe the proton structure, costraining Parton Distribution Functions (PDFs), and to test the perturbative QCD theory, comparing experimental results to state-of-the-art predictions. In this review, a selection of seven of the most recent ATLAS results on vector boson physics will be presented: W and Z boson production cross section at 7 TeV [2] and 13 TeV [3]; ratio of t t over Z boson production cross section at 7, 8 and 13 TeV [4]; Z+jets production cross section at 13 TeV [5]; triple differential Z boson cross section at 8 TeV [6]; angular coefficients of Z boson leptons [7] and splitting scales of anti-k t algorithm clustered jets [8], both measured with 8 TeV data. W and Z boson cross section at 7 TeV Several high precision vector boson cross section measurements have been performed on 4.6 fb −1 7 TeV data, both inclusive and differential.Very good agreement is found between the ATLAS results and the Standard Model in terms of lepton universality, with the measurement of the W → eν/W→ µν ratio and the Z → ee/Z→ µµ ratio, both found to be compatible with unity (see Fig. 1).The measurement of the W → lν/Z→ ll ratio is sensitive to the strangeness content of the proton.The experimental precision reached is comparable and sometimes better than the one on theoretical predictions.The ATLAS result is compared with predictions obtained adopting different PDFs; in all cases, the predicted cross section ratio values are higher than those measured in data (see Fig. 2), clearly pointing out the need for improvements in the description of the proton structure. The W boson charge asymmetry A l , measured as a function of the pseudorapidity of the lepton \|η l \|, defined as: is found to be in good agreement with the predictions (see Fig. 3), with an accurancy better than 1% across the whole range considered (\|η\| < 2.5).The Z boson production cross section measured as a function of the leptons rapidity \|y ll \| is found to be in agreement with predictions for values of \|y ll \| > 1, while for \|y ll \| < 1 the predictions are underestimating the data results (see Fig. 3).This discrepancy can be interpreted as enhanced PDF strangeness content.Therefore, the results have been used to further constrain the proton PDF: with a profile fit, the new version of the ATLAS based PDF set has been derived, namely ATLAS-epWZ16.The new PDF shows unsuppressed strangeness as a function of Bjorken-x; in particular, the variable R s , built as: is found to be higher than unity in the x range considered (10 −3 < x < 10 −1 ), with larger model uncertainties in the high and low x regions (see Fig. 4).The ATLAS data have the best sensitivity in the region with Q 2 = 1.9 GeV 2 and x = 0.023, where the R s is measured to be 1.13 ± 0.05 (exp) ± 0.02 (mod) +0.01 −0.06 (par).With this high precision measurements, ATLAS is definitely accessing the proton structure. W and Z boson cross section at 1TeV The W + , W − , W ± and Z boson production cross sections have been measured separately in the lepton decay channels with 81 pb −1 13 TeV data, within the fiducial region defined by the detector acceptance.The ATLAS results are compared to theoretical predictions obtained with different PDFs, showing good agreement with many of them.The precision obtained is of the order of 2% on W boson and 1% on Z boson cross section, excluding the 2.1% uncertainty due to the luminosity determination (see Fig. 5). The cross section ratios benefit from significant cancellation of experimental uncertainties, resulting in higher constraining power.The W + /W − ratio measured by ATLAS is found to be sistematically lower than the predictions, showing better agreement with the CT14 and MMHT14 PDFs.The W/Z ratio is found to be compatible with the different predictions, within the uncertainty.Both resuts are shown in Fig. 6. t t over Z boson cross section at 7, 8 and 13 TeV The t t/Z cross section ratio has been measured at 7, 8 and 13 TeV: results are given directly for each considered √ s and also as double ratios of the two processes at different √ s.The results are corrected to a common phase space and, furthermore, the systematic uncertainties are properly correlated among measurements.The cross section ratio R tt/Z is defined as: where σ Z→ee and σ Z→µµ are the Z boson cross section measured in the electron and the muon channel, respectively. The results are compared to calculations performed at next-to-next-to-leading-order accuracy using recent PDF sets.The final results uncertainty is further reduced thanks to the cancellation of systematic uncertainties in the ratios, with the data results being more precise than predictions, especially in the 7 and 8 TeV cases (see Fig. 7).The six measured cross sections (t t and Z cross sections at 7, 8 and 13 TeV) are used together within a profile fit to further constrain the ATLAS-epWZ12 PDF: significant reductions of the uncertainties are obtained for the gluon distribution function for x values near 0.1 and for the light-quark sea for x < 0.02 (see Fig. 8). Z+jets production cross section at 13 TeV The measurements of the production cross section of a Z boson in association with jets is performed on 3.16 fb −1 13 TeV data.Inclusive and differential cross sections are measured for events containing a Z boson decaying to electrons or muons and produced in association with up to seven jets with p T > 30 GeV and \|y\| < 2.5.The data results are compared to different order predictions: in general, good agreement is found within the uncertainties with the generators which use next-to-leading-order matrix elements and the more recent next-to-next-to-leading-order fixed-order predictions. In particular, leading-order predictions are observed to have a too hard jet p T spectrum compared to data, for values of jet p T > 200 GeV.Regarding the jet multiplicity observable, the models considered are sensibly underestimating or overestimating the data results for high jet multiplicity (number of jets higher than 4).Both results are shown in Fig. 9. Triple differential Z boson cross section at 8 TeV The Z boson differential cross section in the lepton decay channel has been measured with 20.2 fb −1 8 TeV data.The results are presented as a function of the invariant mass m ll , the dilepton rapidity \|y ll \|, and the angular variable cosθ * , between the outgoing lepton and the incoming quark in the Collins-Soper frame. The cross sections are used to determine the Z boson forward-backward asymmetry A FB , defined as: as a function of m ll and \|y ll \|. The forward-backward asymmetry is found to change sign at the Z boson mass and to increase with \|y ll \|, as expected.A FB values are compatible with 0 for m ll close to Z boson mass, while they are positive (negative) for events with m ll lower (higher) than the Z boson mass.The measurements achieve high-precision (1% level, excluding the uncertainty in the integrated luminosity) and are in agreement with predictions, with a slight tendency of the prediction to underestimate the data results, as can be seen in Fig. 10. Drell-Yan lepton angular correlations The angular correlations between lepton pairs has been measured with 20.3 fb −1 8 TeV data, in the invariant mass region close to Z boson mass.The complete set of angular coefficients A 0−7 is extracted by polinomial decomposition of the lepton polar θ and azimuthal φ angular variables, measured in the Z boson Collins-Soper frame. Results are compared to several available next-to-next-to-leading-order predictions.Evidence is found for nonzero A 5,6,7 observables, consistent with expectations.The observable A 0 −A 2 , built as the difference between A 0 and A 2 , is found to be higher than zero, as expected, when introducing higher order corrections.All the predictions are observed to be a factor of 2 lower than the data results (see Fig. 11). Anti-k t jet splitting scales Splitting scales of the jets produced in association with a Z boson, clusterized with the anti-k t algorithm [9], are measured with 20.2 fb −1 8 TeV data.Starting from charged particle tracks as input, jets are clusterized according to the minimum distance criteria: minimum distance between the tracks or between the tracks and the beam axes.After the jet clusterization, the splitting scales √ d k are defined as the square root of d k , where: with d i j defined as: and d ib as: With this definition, the 0-th order splitting scale corresponds to the p T of the leading jet, while the Nth order of the splitting scale can be interpreted as the distance at which a N jet event can be resolved as a N + 1 jet event. Data results are compared to next-and next-to-next-to-leading-order predictions: significant deviations are found between data and state-of-the-art predictions in various regions of the observables (see Fig. 12).Compatible results are found between the electron and muon channel Z boson decay, and between clusterization performed with the anti-k T radius parameter equal to 0.4 and to 1.0. Conclusion High precision ATLAS measurements of vector boson observables have been performed with 7, 8 and 13 TeV data.In several measurements, the experimental precision has been higher than the prediction one: in particular, the uncertainty had been further reduced in the W + /W − , W/Z and t t/Z cross section ratio measurements, due to the factorization of some systematic uncertainties. The results have be exploited to improve the ATLAS-epWZ12 PDF precision and to derive the new version of the ATLAS based PDF, ATLAS-epWZ16, having reduced uncertainty compared to the previous version and showing unsuppressed strangeness content. Several results are found to confirm the needs for higher order corrections.The remaining discrepancies will motivate further work to improve the precision and the modelling of the predictions.ATLAS precision measurements are therefore a powerful mean to improve our understanding of perturbative QCD. Figure 1 . Figure 1.Measurement of the electron-to-muon cross-section ratios for the W and Z bosons.The green shaded ellipse represents the 68% CL for the correlated measurement of W and Z ratios, while the black error bars give the one-dimensional standard deviation.The orange and blue shaded bands represent the combination of the ratios as measured at the LEP and SLAC e + e − colliders.The SM expectation is indicated with an open circle.Taken from Ref. [2]. Figure 2 . Figure 2. Fiducial cross sections times leptonic branching ratios of the W → lν/Z→ ll ratio.The data (solid blue line) are shown with the statistical (yellow band) and the total uncertainties (green band).The uncertainties of the theoretical calculations correspond to the PDF uncertainties only.Taken from Ref. [2]. Figure 3 . Figure 3. Left: lepton charge asymmetry A l production as a function of the lepton pseudorapidity.Right: Z boson differential cross-section measurement as a function of the leptons rapidity \|y ll \|.Predictions computed at NNLO QCD with NLO EW corrections using various PDF sets (open symbols) are compared to the data (full points).The ratio of theoretical predictions to the data is also shown.The predictions are displaced within each bin for better visibility.The theory uncertainty corresponds to the quadratic sum of the PDF uncertainty and the statistical uncertainty of the calculation.Taken from Ref. [2]. Figure 4 . Figure 4. Left: R s ratio of the PDFs from the present ATLAS-epWZ16 determination shown in the region of maximum sensitivity of the ATLAS data, 10 −3 < x < 10 −1 .Uncertainty bands represent the experimental (exp), model (mod) and parameterization (par) components in red, yellow and green, respectively.Right: R s in the region of ATLAS best sensitivity, Q 2 = 1.9 GeV 2 and x = 0.023.Bands: present result and its uncertainty contributions from experimental data, QCD fit, and theoretical uncertainties; Closed symbols with horizontal error bars: predictions from different NNLO PDF sets; Open square: previous ATLAS resuls.Taken from Ref. [2]. Figure 5 .Figure 6 . Figure 5. Ratio of the predicted to measured fiducial cross section for the combined electron and muon channels using various PDFs.The inner (outer) band corresponds to the experimental uncertainty without (with) the luminosity uncertainty.The inner error bar of the predictions represents the PDF uncertainty while the outer error bar includes the sum in quadrature of all other systematic uncertainties.Taken from Ref.[3]. Figure 7 . Figure 7.The tt/Z cross section ratio for √ s = 7 TeV (left) and 8 TeV (right): results are compared to predictions based on different PDF sets.The inner shaded band corresponds to the statistical uncertainty, the middle band to the statistical and experimental systematic uncertainties added in quadrature, while the outer band shows the total uncertainty, including the luminosity uncertainty which almost entirely cancel in these ratios.The theory predictions are given with the corresponding PDF uncertainties shown as inner bars while the outer bars include all other uncertainties added in quadrature.Taken from Ref. [4]. Figure 8 . Figure 8. Impact of the ATLAS Z boson and tt cross-section data on the determination of PDFs.The bands represent the uncertainty for the ATLAS-epWZ12 PDF set and the uncertainty of the profiled ATLAS-epWZ12 PDF set using Z boson and tt data as a function of x for the total light-quark-sea distribution (left) for the gluon density (right).The plots also show the impact of only including the ATLAS tt data set.Taken from Ref.[4]. Figure 9 . Figure 9. Left: measured cross section as a function of the leading jet p T for inclusive Z+ ≥ 1,2,3,4 jet events.Right: measured cross section as a function of the inclusive jet multiplicity for Z+jets events.The data are compared to the predictions from Z + ≥ 1 jet Njetti NNLO (p T spectra only), BLACKHAT+SHERPA, SHERPA 2.2, ALPGEN+PY6, MG5_aMC+PY8 CKKWL, and MG5_aMC+PY8 FxFx.The error bars correspond to the statistical uncertainty, and the hatched bands to the data statistical and systematic uncertainties (including luminosity) added in quadrature.Taken from Ref. [5]. Figure 10 . Figure 10.Fiducial cross section as a function of m ll , the dilepton rapidity \|y ll \| in the invariant mass range close to the Z boson mass.The data are shown as solid markers and the prediction from Powheg is shown as the solid line.The inner error bars represent the data statistical uncertainty and the solid band shows the total experimental uncertainty.The contribution to the uncertainty from the luminosity measurement is excluded.The hatched band represents the statistical and PDF uncertainties in the prediction.Taken from Ref. [6]. Figure 11 . Figure 11.Distributions of the angular coefficients A 0 − A 2 as a function of p Z T .The results are compared to the DYNNLO and POWHEG MINLO predictions (left) and various SHERPA predictions (right).The error bars for the calculations show the total uncertainty for DYNNLO, but only the statistical uncertainties for POWHEG MINLO and SHERPA.Taken from Ref. [7]. Figure 12 . Figure12.Charged-only distributions for the 0-th (left) and the 7-th (right) splitting scales in the electron channel using the jet-radius parameter R=0.4.The size of the error bars reflects the statistical uncertainty, while the combined statistical and systematic uncertainty is indicated by the grey band.Theoretical predictions from SHERPA with NLO multijet merging (MEPS@NLO) and from POWHEG+PYTHIA8 with NNLO matching (NNLOPS) are displayed including error bands for the generator uncertainties.Taken from Ref.[8].	4,123	2018-01-01T00:00:00.000	[ "Physics" ]
Distributional Learning of Lexical Tones: A Comparison of Attended vs. Unattended Listening This study examines whether non-tone language listeners can acquire lexical tone categories distributionally and whether attention in the training phase modulates the effect of distributional learning. Native Australian English listeners were trained on a Thai lexical tone minimal pair and their performance was assessed using a discrimination task before and after training. During Training, participants either heard a Unimodal distribution that would induce a single central category, which should hinder their discrimination of that minimal pair, or a Bimodal distribution that would induce two separate categories that should facilitate their discrimination. The participants either heard the distribution passively (Experiments 1A and 1B) or performed a cover task during training designed to encourage auditory attention to the entire distribution (Experiment 2). In passive listening (Experiments 1A and 1B), results indicated no effect of distributional learning: the Bimodal group did not outperform the Unimodal group in discriminating the Thai tone minimal pairs. Moreover, both Unimodal and Bimodal groups improved above chance on most test aspects from Pretest to Posttest. However, when participants’ auditory attention was encouraged using the cover task (Experiment 2), distributional learning was found: the Bimodal group outperformed the Unimodal group on a novel test syllable minimal pair at Posttest relative to at Pretest. Furthermore, the Bimodal group showed above-chance improvement from Pretest to Posttest on three test aspects, while the Unimodal group only showed above-chance improvement on one test aspect. These results suggest that non-tone language listeners are able to learn lexical tones distributionally but only when auditory attention is encouraged in the acquisition phase. This implies that distributional learning of lexical tones is more readily induced when participants attend carefully during training, presumably because they are better able to compute the relevant statistics of the distribution. Introduction How do we learn the regularities that exist in our highly structured environment? One approach is that we learn by tracking the statistics-ranging from simple frequency counts to complex conditional probabilities-of the regularities in the environment. This form of acquisition is broadly termed statistical learning and evidence shows that it is a domain-general mechanism which can be used to acquire auditory (linguistic and musical) and visual knowledge [1][2][3]. Distributional learning, in which the statistics tracked by the learner are merely the frequency count of the to-be-learned distribution [4], is a specific form of statistical learning which has been proposed to account for how phonetic categories are acquired. For example, Japanese listeners learn from their linguistic environment that there is only one category (unimodal or single peak distribution) along a particular acoustic dimension in Japanese, the Japanese /r/, whereas English listeners learn that there are two categories (bimodal distribution) along the same acoustic dimension, English /r/ and /l/, resulting in the well-known difficulty native Japanese adults face in discriminating English /r/ and /l/ [5]. This form of phonetic category acquisition has been studied empirically with consonants [4,[6][7][8] and vowels [9][10][11][12]. The procedure in a distributional learning experiment is typically as follows: all participants are trained on a continuum spanning a minimal pair (e.g. /d/-/t/). For one group, the frequency of the training tokens along the continuum follows a distribution that should promote a single central category (a unimodal distribution); for another group, the frequency follows a distribution that should induce two separate categories towards the end of the continuum (a bimodal distribution). Discrimination performance on the target minimal pair between the two distribution conditions is compared after training. Distributional learning is said to occur when discrimination of the minimal pair (such as the end tokens of the continuum) by those trained on the bimodal distribution improve significantly more than those trained on the unimodal distribution. It has been proposed that distributional learning underlies perceptual attunement [13,14], in which young infants, who are previously universal speech perceivers, are better at discriminating items that are specific to their environment (and therefore, familiar) and worse at discriminating unfamiliar items. For example, infants are initially able to discriminate virtually all speech sounds in the world but by their first birthday, their perception has become tuned to just those that are relevant in their linguistic environment [15][16][17][18]. However, because distributional learning has mostly only been studied with consonants and vowels [7][8][9]11], it remains to be investigated whether distributional learning underpins perceptual attunement of lexical tones. In lexical tone languages, a change in pitch signals a change in meaning of the lexical item. For example, in Mandarin, the word /ma/ when spoken in a high level tone (/ma55/) means 'mother', whereas when spoken in a dipping tone (/ma214/), it means 'horse' (lexical tones are represented using Chao values [19] in this paper. Chao values represent a relative scale of pitch height and pitch contour from 1 to 5, with 1 being the lowest in pitch and 5 being the highest). Although approximately 70% of the world's languages use tones to signal a change in meaning [20], lexical tone is very much understudied compared to consonants and vowels. In tone languages, the linguistic status of lexical tones is interesting as, on the one hand, their function is similar to segments (consonants and vowels) in that they are phonemic and, just like segments, tones also undergo perceptual attunement by the first year of life [21][22][23]. On the other hand, lexical tones are structurally more like suprasegmentals, such as intonations [22,24], in that fundamental frequency (F0, the physical property of pitch) overlays the speech form in the production of tones (lexical tones over vowels and intonation over words/sentences). If learners do acquire lexical tones distributionally, this would suggest that at least in terms of acquisition, tones are similar to segments (i.e., consonants and vowels). A handful of distributional learning studies have investigated lexical tones either with infants or adults [25,26], Taken together, their results seem to suggest that after 12 months of age, there is no effect of distributional learning of lexical tones. However, this may be due to the choice of stimuli; those previous studies used Mandarin Tone 55 (high level) and Tone 51 (falling) and tested non-tone language participants such as Dutch [25] and Australian English [26] participants. For these participants, Tone 51 may be relatively easy to discriminate as it is acoustically similar to the declarative intonation in Dutch [27] and English [28]. In other words, participants may have relied on their native language intonation to discriminate the Mandarin tone minimal pairs instead of forming lexical tone categories based on the training distribution. Therefore, one of the aims of the present study is to examine whether non-tone language listeners are able to form lexical tone categories distributionally when a difficult minimal pair of lexical tones is used. Typically, distributional learning experiments involve passive listening [7,8] based on the assumption that learners discover the distribution structure by tracking the items that they hear. However, some researchers suggest that statistical learning is more effective when attention is given to the to-be-learned items [29,30]. Furthermore, some suggest that the effect seen in several distributional learning studies that utilised an enhanced bimodal distribution, that is, a bimodal distribution in which the relevant acoustic cue is artificially exaggerated [9,10], is due to a top-down modulation of attention to the acoustic cue that signals the contrast between a minimal pair, rather than a stimulus-driven effect of the distribution per se [31]. In other words, the authors argue that the exaggeration of the acoustic property in an enhanced distribution of a minimal pair resulted in the learners being made aware of the relevant acoustic cue that would then bootstrap them to discriminate the minimal pair. Thus, it appears that two different types of attention may affect statistical/distributional learning: (i) a general attention to the acquisition phase that encourages learners to attend carefully throughout the entire training stimuli set, which would then allow the learners to compute the relevant statistics of the training stimuli; and (ii) a stimuli-specific attention that allows learners to pick up the relevant acoustic properties necessary for discrimination. The former type of attention is reminiscent of those used in electrophysiological studies that have typically found a larger event-related potential (ERP) response when participants actively attended to the auditory stimuli than when they listened to the stimuli passively [32][33][34]. It is argued that attending to the auditory stimuli enables the participants to pay attention to every stimulus in the training set, which is necessary for formulating a robust representation of the standard that is used as a comparison to the deviant stimuli [35]. It is this type of general attention that has not been manipulated in the distributional learning research paradigm, which motivated the second aim of the present study: would learners show better distributional learning when the learners' attention to the listening task is potentiated during the acquisition phase? Accordingly, two types of distributional learning are examined: unattended listening (Experiments 1A and 1B) and attended listening (Experiment 2). In addition, we also wish to investigate the generalisability of distributional learning. Previous research has shown mixed results in terms of whether adults are able to generalise to other consonants with features similar to those on which they were trained [6,7], but it appears that they are able to generalise vowels across different speakers [9][10][11][12]. In this regard, we tested participants with items on which they were trained (Trained) and on similar but unfamiliar items to the participants (Novel) in one of two dimensions: Syllable (i.e., a change in initial consonant: /k h a/ and /na/), or Speaker's Gender (i.e., female and male), or both. It is hypothesised that if adults can learn lexical tones by tracking frequency distributions just as for consonants and vowels, then it is predicted that the Bimodal group will divide the continuum into two separate categories while the Unimodal group will perceive the training continuum as a single category. Consequently, when tested with the end tokens of the training continuum, we predict that not only will the Bimodal group show an improvement in discriminating both Trained and Novel lexical tones at Posttest relative to Pretest, but that the Unimodal group will show no improvement, or even, a decrease in discrimination performance. This should occur because following training, the former should be facilitated by the emergence of two separate categories while the latter should show an interference in performance due to the merging of the continuum to one single category. This twofold outcome of the same mechanism, that is, distributional learning is termed 'the distributional learning effect'. Furthermore, we predict that a larger distributional learning effect will be observed following the attended listening task (Experiment 2) compared to the unattended listening task (Experiments 1A and 1B). Method Participants. Eighty native Australian English listeners who were Psychology undergraduates (68 females; age range 17-44; M age = 21.09, SD age = 5.70) participated. None spoke a tone language and all reported normal hearing. Twenty four participants reported having musical training; however, none had more than two years of musical experience ( 0.5 year = 6; 1 year = 7; 2 years = 11). Ethics Statement. The University of Western Sydney Human Research Ethics Committee approved the study protocol. All participants were recruited from the University of Western Sydney. Participants were given an information sheet and they provided written informed consent prior to participating in the experiment. Written informed consent was obtained directly from the handful of participants who were 17-year-old undergraduates as per the protocol approved by the University of Western Sydney Human Research Ethics Committee (in Australia, it is common for 17 (going on 18) year olds to begin their first year of undergraduate degree). Participants were given course credit for their participation. Test stimuli. Four native Thai speakers (2 females) produced the stimuli, which were two syllables /k h a/ and /na/, each produced with two different tones (Tone 33 and Tone 241) resulting in four different Thai words. The choice of this tone pair is motivated by a previous study that found it is the most difficult for non-tone language listeners to discriminate [36]. Each speaker produced multiple tokens of each sound. Minimal pairs were formed between two tones of the same syllable by the same speaker. Four target minimal pairs were used in this experiment as follows: Female 1 /k h a33/-/k h a241/; Female 2 /na33/-/na241/; Male /k h a33/-k h a241/; and Male 2 /na33/-/na241/ (see Table 1 for duration and F0 values over time). Having two different speakers of the same gender allows us to examine whether participants are able to generalise on a more abstract level rather than normalising to the speaker's pitch range. To ensure that only the pitch contour differed between the four target minimal pairs, we first chose a base waveform for each minimal pair that is comparable in duration and matching for the speaker. Then, we extracted the natural pitch contour from each member class of a minimal pair (Tone 33 and Tone 241) that were equivalent in duration and imposed it on the chosen base waveform for that particular minimal pair. Two further exemplars for each particular minimal pair were then generated by imposing the pitch contour of other natural recording tokens matching for the same word spoken by the same speaker on the same base waveform. Thus, within each minimal pair, there were three different exemplars for each tone, all of which have the same waveform pattern and differed only in their pitch contours. All the stimuli were normalised for amplitude (70dB). The duration of each minimal pair ranged from 493ms to 832ms, but the duration within each minimal pair was equal. The final stimulus set consisted of 24 tokens (2 words x 3 exemplars x 4 speakers), which were used as test stimuli ( Table 1). The test stimuli formed a 2 x 2 factorial: Test Syllable (/k h a/ vs. /na/) x Test Gender (Female vs. Male speaker). Training stimuli. To generate the training stimuli, we formed an 8-step continuum using Exemplar 1 of each minimal pair, with Tone 33 as Token 1 of the continuum and Tone 241 as Token 8. The continuum was created by interpolating the pitch contour of the two end tokens of each minimal pair (see Fig 1). This ensured that the pitch contour morphed from Tone 33 on one end of the continuum (Token 1) to Tone 241 on the other end of the continuum (Token 8), while keeping the waveform pattern consistent among the training tokens within each minimal pair. Both the test stimuli and training stimuli were presented to five native Thai listeners for verification. The test stimuli were identified correctly by the Thai speakers at least 80% of the time, and for each training continuum, there was a decline in percentage of Tone 33 response from Token 1 to Token 8, suggesting that the native listeners were sensitive to the change in pitch contour for the intermediate tokens [37]. Practice stimuli. A 440 Hz sinewave tone and a 440 Hz sawtooth tone, both 800ms in duration, were synthesized using Praat as practice stimuli for the ABX discrimination task. Procedure. Participants were randomly assigned to one of two Distribution Conditions: Bimodal or Unimodal. Within each Distribution Condition, they were again randomly assigned to one of four training minimal pairs: Female /k h a33/-/k h a241/, Female /na33/-/ na241/, Male /k h a33/-/k h a241/, or Male /na33/-/na241/. Thus, there were eight different conditions, each with 10 participants. The experiment was programmed on MATLAB 2012b and it was presented using an Acer TravelMate P653 laptop. The auditory stimuli were presented using a pair of Sennheiser HD650 headphones connected to an Edirol USB Audio Capture UA-25EX audio interface. To familiarise the participants with the format of the ABX discrimination task, four practice trials were presented using the practice stimuli. The participants were told that they had to indicate whether the third sound, X, is similar to the first, A, or the second, B, by pressing the left shift key (A and X are similar) or the right shift key (B and X are similar). They were also informed that they only had 1s to respond in order to maintain vigilance. There were three phases: Pretest, Training and Posttest. At Pretest and at Posttest, the participants were asked to discriminate all four test minimal pairs in an ABX discrimination task. In each trial, A and B were always Exemplar 1 tones, and X was either an Exemplar 2 or Exemplar 3 tone. For instance, a trial may consist of Female /kha33/ 1 -/kha241/ 1 -/kha33/ 2 , where the subscripts indicate the exemplars. The four test minimal pairs were presented eight times each with the order of Exemplar 2 and Exemplar 3 tones as X being counterbalanced, resulting in a total of 32 trials in each test session, the order of which was randomised. There were no replacement trials for slow responses. During the training phase, the distribution of the training continuum was manipulated, depending on the Distribution Condition. As can be seen in Fig 2, the Bimodal participants heard Token 2 and Token 7 most frequently, whereas the Unimodal participants heard Token 4 and Token 5 most frequently. Crucially, the number of times both groups heard Token 1 and Token 8, that is, the A and B test stimuli, was the same. The training continuum was presented 16 times, that is, 256 tokens in total. While distributional learning experiments typically employ 128 training tokens [9,11,12], a pilot study for this experiment with 128 training tokens was conducted and we found no effect of distributional learning. Therefore, we doubled the number of training tokens in order to increase the chances of observing an effect. The order of the training tokens was randomised for each individual and the training phase took approximately five minutes in duration. Once the participants completed the experiment, they were given a language and musical background questionnaire. The entire experiment took approximately 30 minutes to complete. Results Firstly, a 2 (Distribution Condition) x 4 (Training Continuum) x [2 (Session) x 2 (Test Syllable) x 2 (Test Gender)] mixed ANOVA was conducted to determine whether there were any differences across training minimal pairs. Since there was no main effect of training minimal pairs . Note that the pitch contours shown here represent the tone space of the vowel, in which the first 15% and the last 15% of the vowel were excluded to remove possible effects of coarticulation from the preceding consonant and creakiness, respectively. In order to determine whether the participants improved significantly from Pretest to Posttest on particular test dimensions, a set of one-sample t-tests was also conducted on difference scores (i.e., Posttest Scores -Pretest Scores) on Trained and Novel Syllable stimuli (collapsing across Speaker's Gender), as well as Trained and Novel Gender stimuli (collapsing across Syllable) test items for both Unimodal and Bimodal conditions. For both Distribution Conditions, all one-sample t-tests revealed that the scores were significantly different from zero, except for the Trained Gender test items by the Unimodal group after Holm-Bonferroni correction Discussion The current experiment investigated whether non-tone language listeners are able to learn lexical tones distributionally after being trained on a passive listening task. The results showed across Distribution Conditions that generally there were higher discrimination scores at Posttest compared to Pretest. Contrary to previous distributional learning studies [4,6] in which improvement is found for Bimodal but not Unimodal exposure, the participants in both the Bimodal and Unimodal conditions in the present study performed above chance. In other words, while the Bimodal group improved following training as predicted, so did the Unimodal group, which prevents us from conclusively claiming that there is an effect of distributional learning. The discrepancy in results could be due to a difference in the experimental design or a difference in the nature of the stimuli. With respect to design, Maye and Gerken (4,6) employed a training-test phase design while the current study followed a pretest-training-posttest design [9]. Thus, the significant main effect of Session in the present study could simply reflect practice effects for the test stimuli. With respect to stimuli, Maye and Gerken (4,6) used consonants and here lexical tones were used. Further evidence relevant to these alternatives is provided in the following experiments. The results indicate that the gender of speaker at test that was novel (i.e., Novel Gender) was easier to discriminate than the speaker's gender at test on which the participants had been trained (i.e., Trained Gender). More relevant to our hypothesis, the results suggest that despite the improvement from Pretest to Posttest, there was no greater improvement for the Bimodal than for the Unimodal conditions and so, no effect of distributional learning of naturalistic lexical tones. This unexpected finding could be due to the fact that different groups were trained on different minimal pairs, which are not equivalent in initial difficulty. Indeed, a 2 (Test Syllable) x 2 (Test Gender) repeated ANOVA on Pretest accuracy scores revealed a main effect of Test Gender (F(1, 79) Experiment 1B: Unattended Listening on Male /na/ Training Minimal Pair Experiment 1B is similar to Experiment 1A except that more participants were tested (25 in each Distribution Condition) and only one training minimal pair was used: Male /na33/-/ na241/. This choice is motivated by the fact that it is the most difficult to discriminate among the four minimal pairs, with the assumption that participants would benefit more from being trained on a difficult contrast [38]. Method Participants. Participants consisted of 50 native Australian English listeners (39 females), 20 of whom were a subset from Experiment 1A with 30 extra participants recruited for this experiment (15 extra participants in each Distribution Condition). The participants' ages ranged between 17 and 40 years old, with an average age of 21.04 (SD age = 5.52). None spoke a tone language and all reported normal hearing. Twelve participants reported having minimal musical training ( 0.5 year = 4; 1 year = 4; 2 years = 4). Ethics Statement. The University of Western Sydney Human Research Ethics Committee approved the study protocol. All participants were recruited from the University of Western Sydney. Participants were given an information sheet and they provided written informed consent prior to participating in the experiment. Written informed consent was obtained directly from the handful of participants who were 17-year-old undergraduates as per the protocol approved by the University of Western Sydney Human Research Ethics Committee (in Australia, it is common for 17 (going on 18) year olds to begin their first year of undergraduate degree). Participants were given course credit for their participation. Stimuli. The same stimuli as in Experiment 1A were used for the two testing sessions (Pretest and Posttest). Procedure. The procedure was the same as Experiment 1A, with the exception that all the participants were trained on the Male /na33/-/na241/ training minimal pair. Results The same analysis as in Experiment 1A was conducted in Experiment 1B: a 2 x (2 x 2 x 2) Mixed ANOVA with Distribution Condition (Unimodal vs. Bimodal) as a Between Subjects factor; and as Within Subjects factors: Session (Pretest vs. Posttest), Test Syllable (Trained vs. Novel) and Test Gender (Trained vs. Novel). Discussion The same pattern of results was obtained here as in the previous experiment: there was a main effect of Session-which could reflect a practice effect-and a main effect of Test Gender, which shows that participants did better on the Novel Gender (i.e., Female stimuli) items compared to the Trained Gender (i.e., Male stimuli) items. Additionally, here but not in Experiment 1A, there was a significant interaction of Test Gender by Test Syllable, which suggests that Female /na/ test items were easier than Male /na/ test items. Crucially, despite only training the participants on the most difficult minimal pair (Male /na33/-/na241/) and increasing the number of participants in each Distribution Condition, there was no difference between Unimodal and Bimodal distribution conditions at Posttest relative to Pretest. This suggests that there was no distributional learning effect since both groups improved significantly from Pretest to Posttest, which, as discussed in Experiment 1A, may simply reflect a general practice effect. The results are in line with previous research [25,26], in which the participants were tested on a minimal pair of Mandarin lexical tones that was easy to discriminate and which also found no distributional learning of lexical tones. Taken together, the lack of significant distributional learning of lexical tones appears not to be due to the difficulty of the lexical tone minimal pair or to the particular target language. It appears that non-tone language adult learners do not acquire tones distributionally-at least not when the training phase involves passive listening. It may be the case that because the female stimuli were easy to discriminate even at Pretest, it may have bootstrapped the participants to discriminate the other minimal pairs. Another possibility relates to the participants' attention to the training tokens. We suspect that the null results may be due to individual differences in their ability to sustain attention throughout the entire five-minute training phase. Would learners show an effect of distributional learning if they were prompted to listen attentively throughout the training set of stimuli? This is examined in Experiment 2. Experiment 2: Attended Listening In Experiment 2, we repeat Experiment 1B with one crucial difference: modelled on electrophysiological studies [32,39], we added a control task during the training phase, in which pure tones were randomly inserted among the training tokens and participants were instructed to indicate on a paper response sheet when they hear a 'beep'. This demands participants to pay attention to each sound heard during the training phase. Indeed, participants allocate more attentional resources on an attended listening task like this, as indexed by the presence of an ERP response (processing negativity, PN) [40], compared to a passive listening task [39]. Method Participants. Participants were 50 native Australian English listeners (42 females), who had not been tested in Experiment 1A or 1B. Their ages ranged between 17 and 40 years, with an average of 20.82 (SD age = 5.32). None of the participants spoke a tone language and all reported normal hearing. Nineteen participants reported having minimal musical training ( 0.5 year = 2; 1 year = 11; 2 years = 6). Ethics Statement. The University of Western Sydney Human Research Ethics Committee approved the study protocol. All participants were recruited from the University of Western Sydney. Participants were given an information sheet and they provided written informed consent prior to participating in the experiment. Written informed consent was obtained directly from the handful of participants who were 17-year-old undergraduates as per the protocol approved by the University of Western Sydney Human Research Ethics Committee (in Australia, it is common for 17 (going on 18) year olds to begin their first year of undergraduate degree). Participants were given course credit for their participation. Stimuli. The same stimuli from Experiment 1A were used. In addition, the sine wave tone used in the practice task was also used as the beep tone during training. Procedure. The participants were randomly assigned to one of two Distribution conditions: Unimodal or Bimodal. The procedure of Experiment 2 was similar to that of Experiment 1B except that the participants were required to complete an additional vigilance task during training. Participants were provided with a response sheet containing the numbers 1 to 288. They were instructed that they would hear a total of 288 sounds during this phase and that some of those sounds would be beeps occurring randomly throughout the sequence of sounds. The participants were told to follow the sound number being played and circle the sound number every time they hear a beep. A total of 32 beeps occurred interspersed randomly within 256 training tokens. Discussion Experiment 2 investigated whether non-tone language listeners are able to acquire lexical tone categories after listening to a distribution of a lexical tone minimal pair in an attentive task. It was found that the participants generally showed better discrimination performance at Posttest relative to Pretest and on Novel Gender (female) stimuli relative to Trained Gender (male) stimuli. Importantly, this time, there was an indication of successful distribution learning: at Posttest compared to Pretest, the Bimodal participants performed better than the Unimodal participants on Novel Syllable (/k h a/) test items. Unexpectedly, there was no difference between the two Distribution Conditions on Trained Syllable (/na/) test items. However, inspection of the Pretest scores on Trained Syllable test items showed relatively high performance by the Bimodal group on Female /na/ test items compared to the Unimodal group. This may also explain the unexpected main effect of Distribution Condition. Regardless, the results suggest that the Bimodal group showed better discrimination performance after training than the Unimodal group. These results are substantiated by comparing the one-sample t-tests across Experiments 1B and 2. Recall that in Experiment 1B, both the Unimodal and Bimodal conditions showed improvement above chance at Posttest relative to Pretest on all test aspects. However, in Experiment 2, the Unimodal group only showed above chance improvement on one test aspect (Novel Gender) while the Bimodal group showed significant improvement at Posttest on three aspects (Novel Syllable, Trained Gender and Novel Gender). The lack of improvement for the Trained Syllable by the Bimodal group may be again due to the relatively high performance in response to Female /na/ even at Pretest. Nevertheless, the results of Experiment 2 suggest that, by and large, not only did the Bimodal group improve significantly after exposure to the training tokens, the Unimodal group did not improve, which, taken together, indicates an effect of distributional learning in Experiment 2 when both groups' attention was focused (by means of the 'beep' task) on the structure of the distribution encountered. Qualitatively, it appears that the distribution structure had more of an effect on the participants in an attentive task compared to a passive task. Across all three experiments, we found the distributional learning effect only in Experiment 2, in which an attentive task was used. Indeed, a direct comparison of difference scores from Experiments 1B and 2 using a 2 x 2 x (2 x 2) Mixed ANOVA with between-subjects factors Attention (Experiment 1B vs. Experiment 2) and Distribution Condition (Unimodal vs. Bimodal) and within-subjects factors Familiarity (Trained vs. Novel) and Test Aspect (Syllable vs. Gender) revealed a main effect of Familiarity (F(1, 96) General Discussion These studies were designed to add to the current distributional learning literature by examining whether: (i) non-tone language listeners are able to acquire lexical tone categories distributionally; and (ii) attention to the training task modulates the effect of distributional learning effect of lexical tones. The results of this series of experiments suggest that lexical tones may be learned distributionally, as shown by the Bimodal group exhibiting improved discrimination of the endpoints of a lexical tone continuum, while the Unimodal group showed no corresponding improvement, but only when learners are encouraged to pay attention to the tones during the training phase. When learners were trained passively on either a Bimodal or a Unimodal distribution of a lexical tone minimal pair (Experiments 1A and 1B), the two groups showed generally higher performance at Posttest than at Pretest, which is likely due to general practice effects with the test stimuli. On the other hand, when the participants were given a task that encouraged attention to the distribution (Experiment 2), not only was the Bimodal groups' performance on Novel Syllable test items significantly higher than that of Unimodal participants, but the Bimodal group also showed significant improvement from Pretest to Posttest while the Unimodal group showed no such improvement. This suggests that the distributional learning occurs under the conditions in Experiment 2, but not in Experiments 1A and 1B. Furthermore, comparing the difference scores between all three experiments reveals that a suppression of improvement (which is a part of the outcome of distributional learning) is evident when Unimodal participants were trained attentively. In Experiment 2, the Unimodal participants showed no improvement on three of the four test aspects (Trained and Novel Syllables and Trained Gender), while in Experiments 1A and 1B, the Unimodal participants improved significantly on most, if not all, test aspects. This improvement by the Unimodal group in Experiments 1A and 1B and suppression of improvement by the Unimodal group in Experiment 2 is a novel finding, given that no distributional learning studies that employed a pretest-training-posttest design also used a unimodal distribution (e.g. Escudero et al., 2011). The lack of Unimodal suppression on Novel Gender (i.e., female) test aspect in Experiment 2 suggests that the suppression effect may not be strong enough to overcome the relative salience of female stimuli compared to male stimuli. Indeed, the results from Experiments 1A, 1B and 2 suggest that female stimuli were easier to discriminate than male stimuli. It should be noted that this is not due to the fact that there were many more female participants who may be more adept at discriminating stimuli that share a similar pitch range as themselves. The same pattern of results was found amongst the male participants: even at Pretest, female test stimuli were easier to discriminate than male test stimuli (Experiment 1A: t(10) = 2.091, p = .026; Experiment 1B: t(10) = 2.262, p = .047; Experiment 2: t(7) = 7.099, p < .001). The lack of significant difference by the Bimodal group in Experiment 2 relative to the Bimodal group in Experiment 1B may be confounded by the relatively high performance by the former on the Female /na/ test items. Future work could investigate this further by ensuring an equal Pretest performance across distribution groups and experiments. Nonetheless, taken together, these results are in line with previous research suggesting that statistical learning is more effective when attention is given to the to-be-learned stimuli than when stimuli are processed passively [29,30]. Our findings suggest that attention to the training task alone is enough for learners to show an effect of distributional learning without the need to draw the learners' attention to a specific acoustic cue as has been done in studies using enhanced distributions [31]. Since learners allocated more attentional resource to the auditory stimuli in an attended task than an unattended task [39], we argue that learners in an attended task are better able extract the statistics of the distribution. What is not known, however, is whether general attention in the acquisition phase would result in comparable learning to when learners' attention to the specific acoustic cue is manipulated via an enhanced distribution and whether there is any additive effect of both types of attention manipulations. Future studies should address these research questions as they will add to our knowledge of understanding how humans acquire language. Indeed, the simultaneous use of both types of attention is seen in infant-directed speech (IDS): analogous to enhanced distributions in distributional learning research, hyperarticulation of vowels and tones in IDS is proposed to facilitate infants in acquiring phonetic and lexical tone categories by highlighting the acoustic differences in vocalic and tonal contrasts [41][42][43][44], while the use of increased pitch and larger pitch modulations are presumed to capture infants' attention to the speech sounds in general [43,44], similar to the attended task during the training phase in Experiment 2. Still, the question of why attention to the acquisition phase and/or attention to the specific acoustic cue is required for distributional learning of lexical tones and vowels, but not consonants, remains unanswered. It is noteworthy that lexical tones and vowels are perceived less categorically than consonants, and consequently, both lexical tones and vowels overlap and have greater acoustic variability in their production to an extent not evident in consonants [37,[45][46][47]. Therefore, perhaps attention, either in the training phase in general or to the specific acoustic cue, may be necessary during acquisition for speech sounds that tend to be variable in nature. These studies have only considered the acquisition of lexical tone categories (as indexed by a discrimination task on the endpoints of a continuum before and after training) and only by non-tone language speakers. A possible future direction, then, is to investigate whether learners perceive those lexical tones categorically following training by comparing their discrimination at various points on the continuum, that is, the endpoints vs. the within-category tokens. Furthermore, could our results generalise to non-linguistic pitch categories (e.g. musical pitch categories)? If so, then this would suggest that distributional learning may underlie the perceptual attunement of musical systems as well, which would be in line with the Shared Sound Category Learning Mechanism hypothesis [48]. In addition, would there be differential effects for populations who use pitch extensively (such as tone language listeners and musicians)? Given that these groups tend to outperform non-tone language non-musicians in discriminating and learning lexical tones [49][50][51][52], would a brief distributional training on a lexical tone minimal pair still provide an advantage for the 'pitch experts' above and beyond their extensive experience with pitch? Work is currently being undertaken in our laboratory to address these issues. In sum, it was found that, contrary to other studies on distributional learning of lexical tone [25,26], non-tone language listeners are able to acquire lexical tones distributionally, but only when learners actively attend to the training stimuli. Nonetheless, this suggests that lexical tones have the same linguistic status as consonants and vowels, at least in terms of acquisition. The present study also adds to the growing distributional learning literature by providing a direct comparison between unattended and attended training; distributional learning effect is more readily observed when an attended task is used during training than when an unattended task is used, presumably because the learners are better able to extract the relevant statistics of the distribution. In order to fully understand distributional learning as a learning mechanism, further research is required to investigate whether the same mechanism extends from speech to the music domain; whether there are population differences (language background; musical background) in the effects of distributional learning; and the role of attention during acquisition in general and to specific acoustic cues of the auditory stimuli.	8,845.2	2015-07-27T00:00:00.000	[ "Linguistics" ]
A CNN-BASED FLOOD MAPPING APPROACH USING SENTINEL-1 DATA : The adverse effects of flood events have been increasing in the world due to the increasing occurrence frequency and their severity due to urbanization and the population growth. All weather sensors, such as satellite synthetic aperture radars (SAR) enable the extent detection and magnitude analysis of such events under cloudy atmospheric conditions. Sentinel-1 satellite from European Space Agency (ESA) facilitate such studies thanks to the free distribution, the regular data acquisition scheme and the availability of open source software. However, various difficulties in the visual interpretation and processing exist due to the size and the nature of the SAR data. The supervised machine learning algorithms have increasingly been used for automatic flood extent mapping. However, the use of Convolutional Neural Networks (CNNs) for this purpose is relatively new and requires further investigations. In this study, the U-Net architecture for multi-class segmentation of flooded areas and flooded vegetation was employed by using Sentinel-1 SAR data and altitude information as input. The training data was produced by an automatic thresholding approach using OTSU method in Sardoba, Uzbekistan and Sagaing, Myanmar. The results were validated in Ordu, Turkey and in Ca River, Vietnam by visual comparison with previously produced flood maps. The results show that CNNs have great potential in classifying flooded areas and flooded vegetation even when trained in areas with different geographical setting. The F1 scores obtained in the study for flood and flooded vegetation classes were 0.91 and 0.85, respectively. INTRODUCTION Flooding are among of the most common and destructive natural hazards that cause social and economic disruption as well as causing loss of human lives. Besides the floods caused by heavy rainfall; coastal floods and rapidly melting snow and ice in mountainous areas should also be taken into account as hazard triggering factors. In addition, dam flooding can also occur after sudden and heavy rainfall and or due to infrastructure failure. Although different triggering factors exist, in flood events, a quick assessment of the event followed by a quick response is important in many aspects. The flood extent mapping studies utilize spectral information from optical remote sensing data, synthetic aperture radar (SAR) data or a combination of these two together (Shen et al., 2019a). Although the data from optical sensors have been used for flood monitoring for a long while, they have significant limitations in flood assessment studies due to atmospheric conditions, e.g., cloud cover, and their inability to provide data at night (Clement et al., 2018). SAR sensors is a valuable data source to detect and monitor floods as they can provide data in all-weather conditions and also at night (Manavalan, 2017). On the other hand, water-like surfaces such as shadows, speckle effect, and geometric correction found in SAR data can be limiting factors in flood mapping studies. Studies aimed at detecting floods from SAR data began to appear in the literature in the 1980s and have since been developed (Lowry et al., 1981). Thanks to the increase in SAR sensors and advances in remote sensing and computer vision algorithms in recent years, SAR data has been widely used in flood extent mapping and magnitude analysis. In this context, various methods have been used in the literature. These can be briefly listed as visual interpretation (Oberstadler er al., 1999), manual and automatic histogram thresholding (Nakmuenwai et al., 2017), supervised classification (Pulvirenti et al.,2013;Tavus et al., 2019Tavus et al., , 2020Tavus et al., , 2021, automatic segmentation , region growing (Matgen et al., 2011), fuzzy logic (Twele at al., 2016), change detection (Giustarini at al., 2012;Zhao et al., 2019), combination threshold and change detection and interferometric SAR coherence (Chini et al., 2019;Li et al., 2019;Pelich et al., 2021). Recently, there have been significant advancements in the supervised machine learning (ML) algorithms, especially the deep learning (DL) methods and the Convolutional Neural Networks (CNNs) (Jia et al., 2014). Unlike the pixel-based learning approaches, the CNNs can take advantage of the spatial structure of the target segment. Besides, an automatic feature presentation splits the feature space by reducing the uncertainties in the data. Due to these features, CNNs have become a method that has been successfully applied for flood mapping in recent years, as in many other application areas. Gebrehivot et al. (2019) investigated the potential of CNN method to detect floods from high-resolution unmanned aerial vehicle (UAV) images. As a result of the study using VGG-based fully convolutional network (FCN-16s), it was emphasized that it can successfully detect the flooded regions in the images in comparison to the conventional classification methods, such as FCNs and support vector machines (SVMs). Nemni et al. (2020) designed a CNN-based approach for extracting the flooded areas in Sentinel-1 SAR data. In the study, the flood masks were created with classical semiautomatic techniques, manual cleaning and visual inspections; and various CNN architectures were investigated. The methodology significantly reduced the time for producing the flood maps. The CNNs in the study achieved F1 scores of 91% and 92 % over the test dataset. Peng et al. (2019) proposed two different CNNs (PSNet-v1 and PSNet-v2) to predict the similarity between Planet Scope multispectral images with 3 m spatial resolution before and after flooding. Both architectures achieved superior performance with approximately 89% and 95% F1 score in 2017 Hurricane Harvey and 2018 Hurricane Florence, respectively. Similarly, Potnis et al. (2019) proposed an Encoder-Decoder neural network (NN) based on the Efficient Residual Factorized Convnet (ERFNet) for multi-class segmentation for analysing the urban floods from WorldView-2 data with 2 m spatial resolution. The ERFNet architecture proposed in the study provided an average Intersection Over Union (IoU) score of 0.484 and an overall accuracy value of 87%. Thus, it showed promising results in urban flood assessment with the satellite optical images. Rambour et al. (2020) introduced a SEN12-FLOOD dataset containing co-registered Sentinel-1 and Sentinel-2 images for flood detection and used the ResNet-50 network for flood mapping. With a state-of-the-art network (Resnet-50), the accuracy achieved with the SAR data was 75%, while the combination of RGB and SAR data provided 90% overall accuracy. Bonafila et al. (2020) introduced Sen1Floods11 dataset with Sentinel-1, and permanent and flood water. Permanent water and flood water surfaces were segmented using fully convolutional neural networks (FCNNs). The study results indicated that radar data with DL models can outperform the threshold-based algorithms for flood detection. In addition, the training data with automatic labels obtained from the optical images yielded to higher accuracy in comparison to the handlabelled scarce data. Konapala et al. (2021) investigated the potential of combinations data from Sen1Floods11 (Sentinel 1 and Sentinel 2), and Shuttle Radar Topography Mission (SRTM) data for generating accurate flood detection. As a result of the study evaluating the performance of the methodology with K-fold cross-validation using U-Net CNN, a median F1 score of 0.62 was obtained when only radar data were employed. A F1 score of 0.73 was obtained with the use of Sentinel-1 and altitude information together. As a result of the literature review, it was clear that Sentinel-1 data has great potential in flood mapping, but have limitations in comparison to the optical data due to the nature of the flood events. On the other hand, the CNNs have been successfully used in many applications. Here, we applied a modified version of the U-Net architecture for multi-class segmentation to Sentinel-1 and SRTM data with 30 m resolution for accurate flood mapping. At the same time, we focused on further exploration of the potential of SAR data in identifying flood and flooded vegetation areas. In this paper, we present and discuss the initial results of the study. In Section 2, the datasets used here, the pre-processing steps, label/mask generation, and U-Net architecture are explained. Section 3 presents the multi-class segmentation results and their accuracy metrics. Finally, the conclusions of the study and future work are presented and discussed in Section 4. MATERIALS AND METHODS In this section, an overall methodological workflow, the study area and the datasets, the details of the CNN architecture and the validation approach are explained. Overall Methodological Workflow The overall methodological workflow of the study is given Figure 1. The study sites can be named as Ordu, Turkey, Sagaing, Myanmar, Ca River, Vietnam, and Sardoba, Uzbekistan ( Figure 2). The sites were selected based on the availability of test data and the occurrence of recent major flood events. The input data includes pre-and post-event Sentinel-1 (S1) data and the elevation information from SRTM. A number of preprocessing methods were applied to S1 data to obtain the polarization information, to reduce the noise, and to remove systematic errors caused by the terrain. The input features used in the CNN architecture includes thus the S1 polarization data and the SRTM data. For the model training and validation, masks for flood and flooded vegetation classes were produced using a stepwise automatic thresholding approach with OTSU. The flood maps were produced with the CNN model and an accuracy assessment was performed using the test data in Ordu and Sagaing. Further details are explained in the following subsections. Datasets Here, Sentinel-1A C-band Interferometric Wide (IW) swath mode and Level 1 ground range detected (GRD) products were utilized. The products have vertical (V) and horizontal (H) polarization (i.e., VV+VH) information with a ground sampling distance (GSD) of 10 m. Datasets for each area were obtained from the ESA Copernicus Programme (Copernicus, 2020). The S1 data used in the study were chosen based on the acquisition dates considering the flood occurrence (before and after flood). The characteristics of the study data and the ground conditions, such as wet or dry, are summarized in Table 1. Feature Preparation Workflow The input features involved in the CNN architecture include the VV and VH polarization and the SRTM digital elevation model (DEM). The multi-class segmentation approach classifies the pixel as non-flood, flood, and the flooded vegetation. In order to determine the flooded (FL) area and flooded vegetation (FV) classes and thus to form the mask pixels to be utilized in the CNN architecture, the processing steps given in Figure 3 were applied. The mask data to be used in the model training phase were produced with the approach listed below. This approach is basically based on the data preparation stages that are part of the work carried out by Nemni et al (2020). The main difference here is that the threshold values determined for the classes are obtained automatically from the Multi-OTSU threshold algorithm instead of manual detection (Liao et al., 2001). In order to generate labelled mask data for model training, the following steps were applied to the data from Sardoba and Sagaing regions denoted as DS 1-4 in Table 1, respectively.  Before using the S1 images, some necessary preprocessing steps, such as radiometric correction, image speckle filtering, and orthorectification, were applied. Details on these processes can be found in Tavus et al. (2021).  The OTSU threshold method was applied to each image (pre-& post-event VV and VH) in order to determine the flood pixels.  Based on the flood-induced change in the field, difference VV and VH images were produced by taking the differences between the thresholded pre-& post-event VV and VH data. At this stage, VV and VH flood masks were obtained with the values as 1: representing the flood and 0: representing the background. Afterwards, the pixels labelled as 1 in both of the masks were recorded as a flood mask.  In order to determine the FV pixels, FL pixels were extracted from the difference VV and VH images with applying flood mask produced in the previous step.  The FV pixels were produced by applying the OTSU threshold method to difference images, which do not contain flood pixels anymore.  As in the generation of FL pixels, the final FV mask was produced by taking the overlapping pixels of the VV and VH vegetation pixels at this stage. In this mask, 0 represents the background while 2 label represents the flooded vegetation.  Majority filters applied to FL and FV masks and combined as a single mask. Finally, opening followed by closing morphological filters were also applied to the data in order to remove the elements that could not be removed by the majority filter, such as holes, noise, and borders remaining in the combined mask. In Figure 4, FL, FV, and merged masks of Sardoba study area are given. Input data for all regions were created by stacking pre-& postevent VV and VH, which were produced with the preprocessing in the beginning of mask production stage; and the SRTM DEM data correspond to the area and then compress to 8-bit. Finally, the data preparation process was completed by arranging the 5band input data as 2562565 and the mask data representing 3 classes as 2562561 input size. As a result of this process, a total of 1086 images were generated and randomly allocated to the train, validation and test datasets, with have sample percentages of 72%, 18%, and 10%, respectively (Table 2). CNN Architecture for Pixel-based Classification In this study, a modified version of U-Net architecture was used for the class segmentation task. The modifications were applied by using ResNet-50 model in the encoder part, removing upsampling layers, and replacing them with the transposed convolution layers in the decoder part (He et al., 2015). The input images included 5 channels, and no pre-trained weights were available ( Figure 5). Therefore, all layers were initialized with the Glorot uniform initializer (Hanin and Rolnick, 2018). Table 3 shows the model configurations. ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume V-3-2022 XXIV ISPRS Congress (2022 edition), 6-11 June 2022, Nice, France RESULTS AND DISCUSSIONS Here, the statistical results obtained from the CNN predictor are presented and map results from Sardoba, Ordu and Ca River are discussed. The model trained with the data of Sardoba and Sagaing regions has been tested in the Ordu and Vietnam areas. CNN Model Accuracy The combination of Categorical Cross Entropy and Dice index was used as loss function and Adam method used as an optimizer. F1-score was used as accuracy metric for the performance evaluation. Table 4 shows the model results obtained from model training and validation samples. Table 5 shows the F1-Score of each class calculated using the model predictions obtained from test data. The results presented in Table 4 show that the test and validation accuracies are similar. However, the test results in Table 5 presented for the individual classes are better. The FL class could be predicted with a higher accuracy than the FV class. The FL class prediction performance is comparable with the results of the recent CNN-based flood mapping studies in the literature. Table 5. F1-Score of each class in the test data. Figure 6 shows test data parts used as ground truth and the model predictions in Sardoba site. It must be emphasized that the prediction results were obtained from the test split, which was 10% of all samples. Ordu, Turkey Test Site Results The flood maps of Ordu and Ca River (DS 5-8) test sites were produced with the model trained in Sardoba and Sagaing (DS 1-4) regions. The Ordu test site was analyzed for the subareas of Terme and the Yesilirmak River region of Samsun Province. Both provinces are located in the northern part of Turkey, in the Black Sea Region. The study area is complex due to rugged topography and mixed land cover with inland water bodies (i.e., streams and rivers), urban settlements, open terrain, and agricultural and dense forest areas. In the research conducted by Kocaman et al. (2020), Ordu flood map was generated from Sentinel-1 and Sentinel-2 data with Random Forest (RF) classification algorithm. In Figure 7, the FL and FV pixels produced from OTSU threshold algorithm, CNN and RF are shown together with zoom-in views. Please note that the FV and the FL classes were merged in one class in the RF results and marked as flooded area in Figure 7. Figure 7 shows the results around the Yesilirmak River. As can be seen from the RF result, the river is densely surrounded with agricultural and forest lands. As a result of the comparison of the CNN and RF results, an essential outcome of the model was that the pixels around the river were also labelled as FV ( Figure 7), which shows the high prediction performance of the proposed method. The results of Terme, another subarea of Ordu site, are shown in Figure 8. In this subarea, although the OTSU method could not produce useful outputs, the CNN results were similar to the RF results, which employed optical data as input feature as well. Again, the FV and the FL classes were represented with a single class (flood area) in the RF results in Figure 8. Ca River, Vietnam Test Site Results Another test was applied on the data of the flood event that occurred in Vietnam on September 6, 2019. This area represents a flood event that reaches much larger extents in comparison to the Ordu test area and spread to a relatively smooth topography. As there is no previous study or external reference data for this area, the CNN results were visually compared with data from the Flood Mapping Tool (FMT) published by Hamid Mahmood (2022) (Figure 9). CONCLUSIONS AND FUTURE WORK In the present study, a CNN architecture was proposed for the mapping of flooded areas (FL) and flooded vegetation (FV) from Sentinel-1 data and SRTM DEM. Four different test sites, i.e., Ordu, Turkey, Ca River, Vietnam, Sardoba, Uzbekistan, Sagaing, Myanmar, with major flood events were utilized for this purpose. While the model training dataset was produced from the Sardoba and Sagaing test sites, further evaluations were carried out by using external references in Ordu and Ca River. The training dataset was split as train (72%), validation (18%) and test (10%) samples. The U-Net architecture on ResNet-50 backbone was implemented for the multi-class segmentation. The results show that the F1 scores obtained from the test samples were 0.91 for the FL and 0.85 for the FV classes. The present study is the first one for detecting FV class with a CNNbased classifier. The visual assessments carried out in Ordu and Ca River also show high quality output of the method. The results from a subarea of Ordu, the Terme, also show that SAR data have potential for the detection of floods in the urban area. On the other hand, the Ordu site has rugged topography, which indicates that the use of SRTM DEM as input feature can also be recommended for accurate flood mapping in such areas. As future work, it is planned to improve the results with higher resolution SAR data and further tuning of the proposed methodology. As an example, data augmentation techniques for SAR data can be employed in order to investigate the influence of such techniques on small datasets. Fine tuning can also be applied to the CNN model trained in this study to assess its performance on different flood areas in the world. In addition, modifying the CNN architecture to utilize features from both SAR and optical sensors may improve the overall results. Furthermore, different deep learning architectures for segmentation such as LinkNet, PSPNet, etc. can also be combined with different encoders, (e.g., SE-ResNeXt50, seresnet34) in order to assess impact of the backbones and architectures on the results by using the same dataset. Finally, the proposed approach can be applied to more datasets at different geographical locations having diverse characteristics for further validation.	4,552.8	2022-05-17T00:00:00.000	[ "Environmental Science", "Computer Science", "Engineering" ]
Loss-cone instability modulation due to a magnetohydrodynamic sausage mode oscillation in the solar corona Solar flares often involve the acceleration of particles to relativistic energies and the generation of high-intensity bursts of radio emission. In some cases, the radio bursts can show periodic or quasiperiodic intensity pulsations. However, precisely how these pulsations are generated is still subject to debate. Prominent theories employ mechanisms such as periodic magnetic reconnection, magnetohydrodynamic (MHD) oscillations, or some combination of both. Here we report on high-cadence (0.25 s) radio imaging of a 228 MHz radio source pulsating with a period of 2.3 s during a solar flare on 2014-April-18. The pulsating source is due to an MHD sausage mode oscillation periodically triggering electron acceleration in the corona. The periodic electron acceleration results in the modulation of a loss-cone instability, ultimately resulting in pulsating plasma emission. The results show that a complex combination of MHD oscillations and plasma instability modulation can lead to pulsating radio emission in astrophysical environments. The author present some very interesting observations of a solar radio pulsation event that exhibits a very fast (2 s) period and interpret it in terms of a loss-cone instability that is coupled to an MHD sausage mode oscillation. The authors perform a wavelet analysis which convincingly shows a periodicity during most time of the flare. Although similar events have been reported earlier, what is unique in this study is that the authors have high-resolution EUV images from AIA/SDO and the tool of magnetic modeling at hand, which helps them to identify an X-point geometry that plays a keyrole in the interpretation. The data quality is very good and the Figures are of highest quality and very informative. Having said all these positive aspects, I think that the interpretation could be substantially improved, and therefore I suggest that the authors do some more in-depth work before the manuscript is considered for publication. Essentially there are at least two interpretations: (A) MHD sausage mode, and (B) nonlinear limit cycles. The authors should discuss these two models in more detail, otherwise we do not learn anything in this paper. The first model could be possible if: (A1) the periodicity is very regular (as expected for a standing wave in an eigen-mode), (A2) if the modulated radio flux amplitude is sinusoidal with symmetric pulse shapes (as it is expected for a MHD sausage mode in the linear regime), and (A3) if the sausage mode period (P_sausage > 2.62 a / v_A ) is consistent with the observables a and v_a. The loop diameters a could be measured from the AIA images, and the Alfven velocity v_A can be obtained from the magnetic field (using the NLFFF method at the oscillating location) and the electron densities n_e (using a differential emission measure analysis, and possibly a hydrostatic density model). The possible range of the observables (a, n_e) should be determined and a mean value and standard deviation of the predicted sausage mode period P_sausage could be derived, including measurement uncertainties. The second model (B) could also be modeled in more detail, which is completely absent in the present version. The authors mention a Lotka-Volterra equation system with a limit cycle for the energy exchange between plasma particle and wave distribution. In such a model, the wave growth (of the observed radio flux) has an e-folding growth time t_g and a decay time t_d for empting the loss-cone from particles, while the limit cycle period has the mathematical solution of P = square root ( t_g * t_d), which is just the geometric mean of the growth and decay time. It is also straightforward to calculate the modulated wave flux profile W(t) from the Lotka-Volterra equation and to show whether it can produce the non-symmetric pulse shapes, which look like sawtooth profiles as shown in Fig. 4. One can also use the NLFFF magnetic field model to calculate the mirror ratio of the loss-cone in the X-point region of the magnetic field. Together with some simplified transport model for empting the loss cone, one can derive the decay time t_d. If the authors interpret the radio flux in terms of electron cyclotron maser emission, they could use the theoretically calculated growth times for the various X, O, and W modes. Once the authors have quantified the two models by using all the constraints they have from the (excellent) observations, they should be able to present a decisive discussion which model fits the data better. There are a number of possible criteria which the authors have not used. For instance, sausage (or kink-mode) mode oscillations usually exhibit very periodic but strongly damped time profiles, while limit cycles can be periodic over much longer time intervals, with occasional intermittent glitches (see Froment et al. 2015 for instance). In addition, are the saw-tooth pulses more consistent with sausage modes or with limit cycles ? Some details: The text could be shortened in a number of places. For instance the first half of the abstract says nothing new, the abstract could just start at: "Here we report on ..." Introduction, "... but this has never before been achieved ..." This is an overstatement Introduction: "... modern challenge ..." Not only in modern times ! Results: The acronyms or instruments Orfees and NDA are never explained Many other acronyms are not defined or at the wrong places: PFSS, NLFFF, The term "thorns" should probably be "sawtooth pattern" The term "pulse crash" probably means "pulse decay phase" Discussion: "This is not a region where we would expect a loss-cone instability". Really ? A magnetic field of B > 39 G is cited. Is this the average field strength along a loop? The corona is 3-dimensional, a single 0-dimensional value is useless, unless it is explained over what volume this value has been averaged. Reviewer #2 (Remarks to the Author): This paper reports an intriguing observation of radio oscillations that occurred during a solar flare. The radio source is attributed to emission produced by a loss-cone instability modulated by an MHD sausage mode oscillation in a compact loop near a magnetic null point. It is a very interesting observation but I am not convinced that the authors have demonstrated the proposed physical scenario. The scenario is complex and involves several steps. A large flare and flux rope eruption occurred in AR12036. The radio oscillations were excited ~400 Mm to the east of the flare in AR12035. • The authors argue that the EUV wave produced by the flare excited MHD oscillations in a compact, hot magnetic loop located near a magnetic null point. o It is not clear from Supplementary Fig. 1 that the EUV wave (shown at 13:03 UT) in fact excited MHD oscillations at the time in question (1254)(1255)(1256)(1257)(1258). The non-pulsating radio emission begins in this source at ~1245. The role of the EUV wave as an exciter needs to be shown more carefully. o While the authors identify an EUV loop near the location of the oscillating radio source that could be the oscillator, they cannot show that the loop is in fact oscillating because the SDO cadence is too slow. • The authors argue that loop oscillations periodically modulate magnetic reconnection and electron energization near the null point. This is asserted by the authors but cannot be shown. o If the loop is implicated in reconnection, shouldn't it show some signs of being "activated" in radio or EUV emissions because fast electrons have access to the oscillating loop? Here, too, the timing and sequence of events is important. • The authors claim that electrons are injected into an adjacent coronal loop that acts as a magnetic trap, presumably as part of the reconnection process -but it is left unstated. o Into which magnetic loops might the electrons be injected based on the location of the null and the NLFF extrapolation? What is the mirror ratio of these loops? What is the height of the pulsating radio source? • Next, a loss cone electron distribution results in the trap that is unstable to the production of Langmuir waves. o Perhaps, but why would the loss cone be set up near the oscillating loop instead of the conjugate magnetic footpoint? Is the adjacent loop/trap itself compact? The authors raise the idea that there are electron beam signatures in the spectrum (not obvious to this reader). What is the role of electron beams in their scenario? By claiming spectral signatures of electron beams, the beams are themselves producing type III-like signatures? And loss-cone driven type IV? Do they attribute loss-cone quenching to the purported electron beams? If so, why are the claimed beamlike signatures in phase with the pulse peaks? • The authors claim that the Langmuir wave are then converted to radio waves that escape to the observer. o The authors argue that the emission is plasma radiation. The Hewitt et al (1985) reference they provide is a formal study of the mechanism in an unmagnetized plasma (no polarization). The degree of polarization from harmonic plasma emission is expected to be weak (Melrose 1980), not the 70% degree of circular polarization reported here. The evidence is by no means "conclusive" that the suggested mechanism is operative. If the emission is harmonic plasma radiation the density in the source is 1.5x10^8/cm^3. Is that commensurate with the plasma environment at the apparent height of the radio source? o The authors go on to discuss the spectral index of the radio pulses. The relevance of references to work about loss-cone driven electron cyclotron maser emission was not clear to me. In summary, each of the steps described in the overall scenario is plausible at some level but by no means inevitable. The observational evidence supporting each step is more suggestive than conclusive and, in some cases, completely absent. The overall scenario is speculative. Given the fact that remote pulsations occurred as part of a large flare, erupting flux rope, and fast halo CME, aren't other means of energizing and distributing electrons over a wide volume, and exciting MHD or other periodic phenomena, equally plausible? Reviewer #3 (Remarks to the Author): The manuscript presents a detailed study of quasi-periodically pulsating radio emission generated in the solar corona, supplemented by the study of the magnetic field and plasma geometry in the source region. The phenomenon of quasi-periodic pulsations is now at the forefront of the ongoing research effort in solar and stellar physics, which makes the results presented in the paper especially timely. The main novel element of the paper is the use of the combination of spatially-resolving observations in radio and EUV, and the state-of-the-art theory. It allowed to authors to trace the whole physical chain of the induction of the pulsations, and use it for a convincing interpretation of the observed phenomenon. In my opinion, the paper is a very solid and timely contribution to the research field of solar radio physics and plasma physics, and should be published as soon as possible. I would suggest some minor revision of the text. Abstract: "In rare cases, ..." -I would suggest to modify this statement, as quasi-periodic pulsations of the radio emission intensity are a rather frequent phenomenon. p. 2: "... known as a Lotka-Volterra system" -better, "modelled by ..." " Solar radio pulsations usually display periodicities of ~0.3-4 seconds" -those pulsations have much longer periods too. I'd suggest to rephrase this statement. p. 17, the estimation of the sausage mode oscillation period. I would remove the factor of 2.62, as recent works (including the cited in this estimations) demonstrated that this factor appears only in the case of a loop with a step-function perpendicular profile of the fast speed, while for smoother profiles this factor is different. Hence, a more correct estimation would be P less than a/C_F, where C_F is the fast speed. But, this modification does not affect the result and conclusions. Reviewer #1 (Remarks to the Author): Comment 1: Review of manuscript "Modulation of a loss-cone instability due to a MHD sausage mode oscillation in the solar corona" by Carley et al., submitted to Nature Communications. The author present some very interesting observations of a solar radio pulsation event that exhibits a very fast (2 s) period and interpret it in terms of a loss-cone instability that is coupled to an MHD sausage mode oscillation. The authors perform a wavelet analysis which convincingly shows a periodicity during most time of the flare. Although similar events have been reported earlier, what is unique in this study is that the authors have high-resolution EUV images from AIA/SDO and the tool of magnetic modeling at hand, which helps them to identify an X-point geometry that plays a keyrole in the interpretation. The data quality is very good and the Figures are of highest quality and very informative. Having said all these positive aspects, I think that the interpretation could be substantially improved, and therefore I suggest that the authors do some more in-depth work before the manuscript is considered for publication. Response: We would like to thank the referee for their useful comments and suggestions on our article. We have significantly improved the interpretation with a more comprehensive analysis of the physical mechanisms at play in the event. The second half of the paper is rewritten, and we have added four new Figures, namely Figure 6 and 8 in the main article text and Supplementary Figures 4 and 5. Details responses to the referee's comments are given below. We believe the new article more accurately demonstrates the proposed physical scenario and addresses all of the referee's concerns. All changes to the article text are in blue. Comment 2: Essentially there are at least two interpretations: (A) MHD sausage mode, and (B) nonlinear limit cycles. The authors should discuss these two models in more detail, otherwise we do not learn anything in this paper. The first model could be possible if: (A1) the periodicity is very regular (as expected for a standing wave in an eigen-mode), (A2) if the modulated radio flux amplitude is sinusoidal with symmetric pulse shapes (as it is expected for a MHD sausage mode in the linear regime), and (A3) if the sausage mode period (P_sausage > 2.62 a / v_A ) is consistent with the observables a and v_a. The loop diameters a could be measured from the AIA images, and the Alfven velocity v_A can be obtained from the magnetic field (using the NLFFF method at the oscillating location) and the electron densities n_e (using a differential emission measure analysis, and possibly a hydrostatic density model). The possible range of the observables (a, n_e) should be determined and a mean value and standard deviation of the predicted sausage mode period P_sausage could be derived, including measurement uncertainties. Response: We have performed a much more comprehensive analysis of the MHD sausage mode oscillation concept. The new analysis is outlined in the new 'MHD sausage mode oscillations' section and now includes a new Figure 6 in the main article. We have followed the referee's recommendations and have performed a much more accurate analysis of the loop geometry, as well as a differential emission measure (DEM) analysis to obtain density. We have also exploited the 3D information provided by the NLFFF extrapolation to estimate a more accurate estimate of the magnetic field in the region of the pulsations (at the small loop). This comprehensive analysis now shows the loop physical characteristics (width, density, temperature and magnetic field) can produce a sausage mode oscillation of period range of P < 1.3 -1.6 seconds. This matches quite well the observed period 2.3+/-1.0 seconds of the pulsations and means the sausage mode oscillation idea is feasible. While a range of estimates for the loop properties is possible given the range of parameters of width, temperature, density and magnetic field, a true estimate of uncertainty is not possible due to the lack of uncertainty analysis from the NLFFF magnetic magnetic field values (Wiegelmann et al. 2012). That said, given the relationship of P to the magnetic field (via Alfvén speed), their fractional uncertainties are approximately equal e.g, delta_P=delta_B, so a 20% uncertainty in B will lead to ~20% uncertainty in P. Hence P is not highly sensitive to the B uncertainty. Comment 3: The second model (B) could also be modeled in more detail, which is completely absent in the present version. The authors mention a Lotka-Volterra equation system with a limit cycle for the energy exchange between plasma particle and wave distribution. In such a model, the wave growth (of the observed radio flux) has an e-folding growth time t_g and a decay time t_d for empting the loss-cone from particles, while the limit cycle period has the mathematical solution of P = square root ( t_g * t_d), which is just the geometric mean of the growth and decay time. It is also straightforward to calculate the modulated wave flux profile W(t) from the Lotka-Volterra equation and to show whether it can produce the non-symmetric pulse shapes, which look like sawtooth profiles as shown in Fig. 4. Response: While the pulse profiles may be fit using solutions of the differential equations of a Lotka-Volterra system, we find that the physical characteristics of the pulse do not match with theoretical predictions of such a system. Several authors have shown that the relationship between the pulse period t and pulse flux F for an LV system should be F ~ t 2 (Aschwanden and Benz 1988, Fleishman et al. 1994. Using a peak finding algorithm, we find that no such relationship exists for the pulsation data, with F and t being uncorrelated, see Supplementary Figure 4. Furthermore, the LV system is likely to result in a more quasiperioidic pulsation and larger variability in the flux. The pulsations presented in this event are more steady and slowly varying. This does not specifically rule out such a mechanism being at play, but it makes it less likely than the MHD oscillations scenario. The above arguments are now stated in a new Supplementary Material section entitled 'Likelihood of a Lotka-Volterra system'. Comment 4: One can also use the NLFFF magnetic field model to calculate the mirror ratio of the loss-cone in the X-point region of the magnetic field. Together with some simplified transport model for empting the loss cone, one can derive the decay time t_d. If the authors interpret the radio flux in terms of electron cyclotron maser emission, they could use the theoretically calculated growth times for the various X, O, and W modes. Response: Using some of the magnetic field estimates from the NLFFF we have included a more complete discussion on wave growth and particle decay time, and the consequences for loss-cone mirror ratios and distances travelled by the particles to reach the loss-cone, as outlined in the new Supplementary Material section 'Viability of the sudden reduction mechanism'. This discussion, along with Figure 6c, now gives a better idea of the physical conditions involved in creating the loss cone. As for the mode of radiation, several authors have shown the circumstances under which we may observe X, O, and Z-modes in response to a loss-cone instability. The X and O modes are generally only employed in situations where ωe/Ωe <1 (the limiting value varies slightly depending on the energetic electron distribution involved), where Ωe is electron cyclotron frequency and ωe is electron plasma frequency. Such conditions are only like to be met low in the corona, where the magnetic field is strong and emissions reach millimetric wavelengths. The Z-mode however has high growth rate in regions of ωe/Ωe >1. Based on this ratio and the more accurate density and magnetic field analysis, we provide further theoretical basis for the interpretation of the emission as being from the Z-mode. This is outlined in the 'Emission mechanism and loss-cone instability modulation' section of the main article and also in the new 'Dominant ECM modes' section of Supplementary Material. The referee mentions the possibility of the W-mode (assumed the referee means Whistler mode from the ECM). Indeed this is a possibility but there is not much evidence for the existence of Whistlers in the explanation of narrow-band pulsating radio sources. Their growth rate is high for ratios ωe/Ωe >1. In this instance, the magnetic field would need to be 81 G in order to produce a whistler of frequency Ωe. From the NLFFF extrapolation, this is likely to occur deeper in the corona below <10 Mm where this ratio is unlikely to be fulfilled. Comment 5: Once the authors have quantified the two models by using all the constraints they have from the (excellent) observations, they should be able to present a decisive discussion which model fits the data better. There are a number of possible criteria which the authors have not used. For instance, sausage (or kink-mode) mode oscillations usually exhibit very periodic but strongly damped time profiles, while limit cycles can be periodic over much longer time intervals, with occasional intermittent glitches (see Froment et al. 2015 for instance). In addition, are the saw-tooth pulses more consistent with sausage modes or with limit cycles ? Response: The two new sections 'MHD sausage mode oscillations' and 'Likelihood of a Lotka-Volterra system' now outline the feasibility of the two physical mechanisms and which of the two is more likely. We find the MHD sausage oscillation mechanism to be more likely in this instance. As we discuss in the conclusions, if the sausage mode oscillation were to directly modulate the emission (via modulation of the plasma properties), we would expect more symmetric pulse profiles. However, because the mechanism is indirect (sausage-mode, electron beam formation, loss-cone quenching), this does not necessitate a symmetric pulse profile. We have also remade Figure 4 such that the 'reverse drift bursts' (previously called 'thorns') are now seen more clearly. We have also analysed these RD bursts, showing they have the same characteristics to previous observations of reverse drifters that occur from electron beam propagation (Aschwanden et al. 1995). This provides further evidence of the idea that electron acceleration was involved in the pulsations, meaning the modulation of radio emission was indirect. This points are outlined in the discussion section. As for the duration of the pulsation event (2.5 minutes), it is unclear why these pulsations last so long (showed no strong damping). This could be due to the prolonged disturbance of the region supplied by the passage of the EUV wave, as in McLean et al. (1971) for example. Passage of an EUV wave in bringing about oscillatory dynamics of a null point has been recently by highlighted by Thurgood et al. (2017). However the exact involvement of the EUV wave in bringing about such activity in this event remains unclear. These points are highlighted in the 'EUV wave kinematics section'. Comment 11: The term "thorns" should probably be "sawtooth pattern" Response: The sawtooth pattern is applicable to a single frequency. However, the pattern we indicate is formed over multiple frequencies. We have replaced the word 'thorns' with the more general 'reverse drift bursts', which is a more generic term to describe signatures which quickly drift to higher frequencies. We have also analysed these bursts showing they have the characteristics of previously observed reverse drift bursts that are signatures of electron beam propagation (Aschwanden et al. 1995). Comment 12: The term "pulse crash" probably means "pulse decay phase": Response: Agreed, we have replaced 'crash' with 'decay phase' Comment 13: Discussion: "This is not a region where we would expect a loss-cone instability". Really ? Response: The discussion has been re-written and this statement removed. Response: With the more accurate determination of magnetic field in the NLFFF extrapolation ( Figure 6) we have now quoted the volume in which we average the magnetic field. The volume is now shown as a blue box in Figure 6c of the article. Reviewer #2 (Remarks to the Author): Comment 1: This paper reports an intriguing observation of radio oscillations that occurred during a solar flare. The radio source is attributed to emission produced by a loss-cone instability modulated by an MHD sausage mode oscillation in a compact loop near a magnetic null point. It is a very interesting observation but I am not convinced that the authors have demonstrated the proposed physical scenario. Response: We would like to thank the referee for their useful comments and suggestions on our article. We have significantly improved the interpretation with a more comprehensive analysis of the physical scenario at play in the event. The second half of the paper is rewritten, and we have added four new Figures, namely Figure 6 and 8 in the main article text and Supplementary Figures 4 and 5. Details responses to the referee's comments are given below. We believe the new article more accurately demonstrates the proposed physical scenario and addresses all of the referee's concerns. All changes to the article text are in blue. Comment 2: The scenario is complex and involves several steps. A large flare and flux rope eruption occurred in AR12036. The radio oscillations were excited ~400 Mm to the east of the flare in AR12035. • The authors argue that the EUV wave produced by the flare excited MHD oscillations in a compact, hot magnetic loop located near a magnetic null point. o It is not clear from Supplementary Fig. 1 (1254)(1255)(1256)(1257)(1258). The non-pulsating radio emission begins in this source at ~1245. The role of the EUV wave as an exciter needs to be shown more carefully. Response: Agreed. We have provided a more careful kinematic analysis that shows the expansion of an EUV disturbance along great circles from the erupting active region, see Figure 8. The disturbance reaches the location of the pulsating radio source at the time the pulsations begin. This provides an indication that the the EUV disturbance was involved in the perturbation of the region that caused the pulsations. Comment 3: o While the authors identify an EUV loop near the location of the oscillating radio source that could be the oscillator, they cannot show that the loop is in fact oscillating because the SDO cadence is too slow. Response: Agreed. It is generally not possible to image such fast pulsations in EUV imaging. Despite this, we have performed a much more comprehensive analysis of the MHD sausage mode oscillation concept. The new analysis is outlined in the 'MHD sausage mode oscillations' section and now includes a new Figure 6 in the main article. We have performed a much more accurate analysis of the loop geometry, as well as a DEM analysis to obtain density. We have also exploited the 3D information provided by the NLFFF extrapolation to estimate a more accurate estimate of the magnetic field in the region of the pulsations (at the small loop). This comprehensive analysis now shows us that the physical location of the pulsating radio source in 3D dimensions was next to a loop whose physical characteristics (width, density, temperature and magnetic field) can produce a sausage mode oscillation of period range of P <1.3 -1.6 seconds. This matches quite well the observed period 2.3+/-1.0 seconds of the pulsations and means the sausage mode oscillation idea is feasible. We have also tested a second primary method by which pulsations could be generated, namely the Lotka-Volterra system. Several authors have shown that the relationship between the pulse period t and pulse flux F for an LV system should be F ~ t 2 . Using a peak finding algorithm, we find that no such relationship exists for the pulsation data, with F and t being uncorrelated. Furthermore, the LV system is likely to result in a more quasiperioidic pulsation and larger variability in the flux. The pulsations presented in this event are more steady and slowly varying. This does not specifically rule out such a mechanism being at play, but it makes it less likely than the MHD oscillations scenario. Comment 4: • The authors argue that loop oscillations periodically modulate magnetic reconnection and electron energization near the null point. This is asserted by the authors but cannot be shown. Response: Agreed. It cannot specifically be shown that electron acceleration was via magnetic reconnection. We have reserved mention of such a mechanism to the discussion section, where we state it is remains a possible scenario due to electron acceleration being located at a magnetic null. That said, lack of evidence of magnetic reconnection does not change the general conclusion of our paper that there is loss-cone modulation via an MHD oscillation. Comment 5: o If the loop is implicated in reconnection, shouldn't it show some signs of being "activated" in radio or EUV emissions because fast electrons have access to the oscillating loop? Here, too, the timing and sequence of events is important. Response: While the loop showing some motion or brightening in the EUV images would be good evidence for its involvement, there does not necessarily need to be a signature of the loop activation. For example, electrons accelerated in the null-point region may not have access to the small loop. If no energetic particles enter the small loop it would not show significant signs of heating/flaring. Indeed the loop shows little sign of activation throughout the event, apart from one leg being slightly hotter than the other (see Figure 6d). The slightly hotter leg may be a signature of heating at its base due to chromospheric evaporation, but such a conclusion can be only tentative. Nonetheless, we believe the analysis we have provided shows the MHD oscillation modulation of emission is a feasible one. Comment 6: • The authors claim that electrons are injected into an adjacent coronal loop that acts as a magnetic trap, presumably as part of the reconnection process -but it is left unstated. Response: We have provided a more thorough analysis and discussion on this idea including an estimate of where the electrons are accelerated (Figure 6c). This now gives a clearer picture of the environment in which the acceleration took place. Specifically identifying the trap location is not possible due to the limited spatial resolution of the radio imaging. However, there are multiple locations within the magnetic environment of the pulsations where a magnetic trap and loss-cone could form e.g., on the adjacent magnetic footpoints (around Q) or at any of the other magnetic footpoints around the null point. This is discussed further in the Discussion & Conclusion section. Comment 7: o Into which magnetic loops might the electrons be injected based on the location of the null and the NLFF extrapolation? What is the mirror ratio of these loops? What is the height of the pulsating radio source? Response: As stated in the previous comments, it is generally not possible to pin-point the exact injection or reflection point of the electrons due to the limited spatial resolution of the radio imaging. Despite this, the analysis performed in Figure 6 gives a much clearer indication of the environment of the pulsating radio source in three dimensions. Based on the analysis of the radiation being from Langmuir waves due to Z-mode ECM emission (or from stream instability plasma emission in the case of the reverse-drift bursts), we have also provided an estimate of the height of the radio source as being 13 Mm for fundamental emission, or 55 Mm for harmonic (indicated in Figure 6c). This places it within the vicinity of the small loop. Based on the magnetic field estimates in this vicinity, particle injection and plasma wave growth times, we have outlined a discussion in the Supplementary Material section 'Feasibility of the sudden reduction mechanism' which discusses particle injection, plasma wave growth times, mirror ratios and estimated heights of the radiation. The idea in this instance is electron acceleration at the null point, with these electrons propagating into the adjacent footpoint regions, which could adjacent footpoint at Q. This idea is very similar to the studies of Benz et al. (1976), Aschwanden et al. (1993Aschwanden et al. ( , 1995. The reverse drift bursts are likely the signature of initial electron acceleration (via plasma emission) while the injection of these electrons into the trap is evidenced by sudden reduction of loss cone emission. It is not possible to estimate a size for this trap, as it is likely smaller than the radio source sizes. The most information we have is that the trap at Q is smaller than the radio source size of 200 arcseconds at this location. That said, absence of trap size estimate does not affect our calculations and conclusions. We outlined a discussion of particle injection into a trap in Discussion & Conclusions section. Comment 9: The authors raise the idea that there are electron beam signatures in the spectrum (not obvious to this reader). What is the role of electron beams in their scenario? By claiming spectral signatures of electron beams, the beams are themselves producing type III-like signatures? And loss-cone driven type IV? Do they attribute loss-cone quenching to the purported electron beams? If so, why are the claimed beam-like signatures in phase with the pulse peaks? Response: We have reprocessed the zoomed dynamic spectrum in Figure 4 so that the features are more clearly seen (now labelled simply as 'reverse drift bursts' or RD). We have also provided new analysis and references to show they have the same drift rate (220 MHz/s) and morphology as the reverse drift bursts studied previously, similar to Aschwanden et al. (1995). Their exciter speed is calculated to be 0.2 c using the hydrostatic density model derived in the 'MHD sausage mode oscillations' section. In light of this new evidence, it is now more clear that these features are from electron beams propagating into greater densities (down a magnetic loop) in the corona. Their emission mechanism is likely to be standard plasma emission generated from Langmuir waves driven unstable from an electron beam (as in the standard quasi-linear theory for type IIIs). Once these electron beams are injected into an adjacent trap, they result in sudden reduction of the losscone emission that is taking place in the trap. The reduction in emission is a combination of the time taken for electrons to be injected into the trap and the time taken for the decay of plasma waves via collisions -this idea is largely the same as the Benz et al. (1976) study. Therefore, it is possible for the signature of the beam and sudden reduction to be out of phase due to one following the other in quick succession. We have provided a more detailed explanation of this in 'Emission mechanism and loss-cone instability modulation', 'Discussion & Conclusions' and a calculation of the feasibility of the idea in the Supplementary Material 'Viability of the sudden reduction mechanism' section. Comment 10: • The authors claim that the Langmuir wave are then converted to radio waves that escape to the observer. o The authors argue that the emission is plasma radiation. The Hewitt et al (1985) reference they provide is a formal study of the mechanism in an unmagnetized plasma (no polarization). The degree of polarization from harmonic plasma emission is expected to be weak (Melrose 1980), not the 70% degree of circular polarization reported here. The evidence is by no means "conclusive" that the suggested mechanism is operative. If the emission is harmonic plasma radiation the density in the source is 1.5x10^8/cm^3. Is that commensurate with the plasma environment at the apparent height of the radio source? Response: While the Hewitt et al. (1985) references mentions small levels of polarisation, there are several other studies which mention it could be quite large e.g., Melrose et al. (1978) showing that second harmonic plasma radiation from Langmuir waves generated in a lose-cone distribution may be X polarised depending on the direction of Langmuir waves, viewing angle, and loss-cone angle. Using their Figure 2 and 3 and using ωe/Ωe=1.6 and loss-cone angle of 30 degrees from our event we find polarisation can be X mode polarised from ~50-100%, depending on viewing angle. We have now quoted these references in 'Emission mechanism and loss-cone instability modulation'. In order to conclude the mechanism is operative we use several characteristics including the source location, spectral indices, flux, narrow bandwidth, polarisation and ratio of plasma to gyro frequency. The extreme spectral indices in particular are good evidence in favour of plasma emission from the ECM Z-mode mechanism. As for the height of the source, we have used the plasma frequency and its harmonic to estimate density then a hydrostatic equilibrium to estimate height of such a density. Assumption of fundamental emission places the radio source height 13 Mm, while assumption of harmonic emission places it at 55 Mm, both of which place the radio source within the vicinity of the small loop (see Figure 6c). Comment 11: o The authors go on to discuss the spectral index of the radio pulses. The relevance of references to work about loss-cone driven electron cyclotron maser emission was not clear to me. Response: In Benz et al. (1976aBenz et al. ( , 1976b) the authors discuss the phenomenon of some narrowband radio bursts having very large spectral slopes of >10. They discuss the challenges in the theoretical explanation of such extreme spectral characteristics and specifically rule out synchrotron radiation. Synchrotron spectral slopes are usually an order of magnitude lower than this, even in cases of significant Razin suppression. The authors then discuss the most likely theoretical scenario to produce spectral indices of this magnitude, concluding that such a spectrum is from 'plasma waves in transverse direction to the magnetic field…excited by trapped particles due to the presence of a loss cone' i.e., the ECM Z-mode. The event in this study is another rare example of a radio burst having extreme spectral slopes and the evidence we provided corroborates and supports the previous interpretations of this emission being from electrostatic waves generated by a loss cone instability. We have reworded the text in 'Emissions mechanism and loss-cone instability modulation' so the reference to previous studies is now more clear. Comment 12: In summary, each of the steps described in the overall scenario is plausible at some level but by no means inevitable. The observational evidence supporting each step is more suggestive than conclusive and, in some cases, completely absent. The overall scenario is speculative. Given the fact that remote pulsations occurred as part of a large flare, erupting flux rope, and fast halo CME, aren't other means of energizing and distributing electrons over a wide volume, and exciting MHD or other periodic phenomena, equally plausible? Response: We have provide more complete and robust evidence on the interpretation of the physical scenario we have postulated, namely the operation of an MHD wave in the modulation of the radio emission. We have also included an assessment of another primary mechanism (the Lotka-Volterra system), showing that this mechanism is less likely. It is possible that the erupting CME may have been involved in triggering the pulsations (as is evidenced from the arrival of the EUV disturbance), but it is unlikely that the the radio source belonged to the CME itself. Carley et al. (2016) Figure 10 has shown that the legs of the CME (where one might expect a loss-cone, for example) were rooted in the flaring active region and not connected to the pulsing source. ------------------------------------------------ Reviewer #3 (Remarks to the Author): The manuscript presents a detailed study of quasi-periodically pulsating radio emission generated in the solar corona, supplemented by the study of the magnetic field and plasma geometry in the source region. The phenomenon of quasi-periodic pulsations is now at the forefront of the ongoing research effort in solar and stellar physics, which makes the results presented in the paper especially timely. The main novel element of the paper is the use of the combination of spatiallyresolving observations in radio and EUV, and the state-of-the-art theory. It allowed to authors to trace the whole physical chain of the induction of the pulsations, and use it for a convincing interpretation of the observed phenomenon. In my opinion, the paper is a very solid and timely contribution to the research field of solar radio physics and plasma physics, and should be published as soon as possible. I would suggest some minor revision of the text. Response: We would like to thank the referee for their useful comments and suggestions on our article. We have significantly improved the interpretation with a more comprehensive analysis of the physical mechanisms at play in the event. The second half of the paper is rewritten, and we have added four new Figures, namely Figure 6 and 8 in the main article text and Supplementary Figures 4 and 5. Responses to the referee's comments are given below. All changes to the article text are in blue. Comment 1: Abstract: "In rare cases, ..." -I would suggest to modify this statement, as quasi-periodic pulsations of the radio emission intensity are a rather frequent phenomenon. Response: Agreed, text corrected to 'some cases'. Comment 3: " Solar radio pulsations usually display periodicities of ~0.3-4 seconds" -those pulsations have much longer periods too. I'd suggest to rephrase this statement. Response: Agreed, text corrected to more accurately quote the possible ranges of radio pulsations. Now says 'periodicities from seconds to minutes' Comment 4: p. 17, the estimation of the sausage mode oscillation period. I would remove the factor of 2.62, as recent works (including the cited in this estimations) demonstrated that this factor appears only in the case of a loop with a step-function perpendicular profile of the fast speed, while for smoother profiles this factor is different. Hence, a more correct estimation would be P less than a/C_F, where C_F is the fast speed. But, this modification does not affect the result and conclusions. Response: Agreed, we have altered the calculation of the pulse period and now use a/C_F. This calculation is also significantly more accurate, given the new analysis of the region density and magnetic field, see Figure 6. The more accurate analysis still support our conclusion that the pulsations was due to an MHD sausage mode oscillation. Reviewers' comments: Reviewer #1 (Remarks to the Author): The authors have the manuscript sufficiently improved. I recommend publication. Reviewer #2 (Remarks to the Author): I thank the authors for being responsive to referee input, which was extensive. While the revision addressed many points raised by the referees, it also raises new questions. In addition, for this referee, the primary objection to the work remains: that the interpretation of the observations is more complex than the data can justify and relies on too many assumptions and guesses to be convincing. I fully acknowledge that this is an interesting and important observation but, in my opinion, the authors over-interpret the data. I cannot recommend the paper for publication in its present form. I urge them to confine their attention to a clear presentation of the data and a fairly brief and critical assessment of what the data might imply about energy transport, release, and modulation. Additional comments: Intro: p. 2 -"This energy release results in the acceleration of particles to relativistic energies and the generation of high intensity bursts of radio emission." Well, EM radiation across the spectrum including, notably, hard X-ray emission. p. 2 -"The radio emission is thought to be generated by a plasma instability..." Perhaps at this particular wavelength. At frequencies above 1-2 GHz plasma emission rarely plays any role. p. 2-"Direct high cadence radio imaging of the behaviour of electron beams during a pulsation event has the ability to confirm one (or some combination) of these mechanisms." Has the "potential" to confirm one or more mechanism. General comment: it is puzzling to me that the authors do not make use of RHESSI observations. (They do use FERMI data in their earlier paper on this event.) There is coverage during the time in question. It would be useful to establish whether there is HXR emission in the non-flaring active region or whether the HXR emission in the flaring active region has any connection to the distant pulsating radio source. The same applies to ancillary radio data from, for example, Ondrejov. Results: p. 6-"possible cyclotron mechanism or loss-cone instability" The cyclotron mechanism refers to a resonance; a loss-cone provides free energy that could drive either plasma radiation or cyclotron maser radiation. I don't understand the argument that the type III "most likely" injected energetic electrons into the quiet AR. They type III appears to be directed out into the corona and perhaps the IPM. Is there evidence that it is bidirectional? The rise in radio flux in the quiet AR appears to be well underway by the time the type III occurs. p. 7&8-The case for reverse drift components -and for downward electron beams -is not very convincing and may be more apparent than real. Were the spectral data corrected for the instrument bandpass? p. 9-Why do the authors claim that electrons accumulate at the loop footpoint? Downward-moving electrons will either precipitate or mirror back up to greater heights. Referring to Fig. 5a, the systematic change in the source centroid with an apparent speed of ~700 km/s is striking. Why is the motion in all bands confined to the observed direction? How does the detailed timing relate to the EUV wave, which has a speed that is only 310 km/s? p. 10&11: If the variation in the position of source maxima is apparent (p.10), it is not necessary to quote a speed of 0.1c (p.11). The authors have embraced the presence of electron beams with very little observational evidence. I am also puzzled by the statement: "This shows that while there is pulsing energisation of electrons at P, the electrons accumulate at the adjacent footpoints close to R and Q ...". Energisation due to what? p. 11-Referring to Fig. 6 the authors claim to identify a magnetic loop in which a sausage mode MHD oscillation could be supported. In panel (a), how does the (systematic) source motion shown in Fig. 5 relate to the NLFFF extrapolation? Is the NLFFF extrapolation consistent with the presence of the loop identified in EUV and shown in Fig. 6c? How do fast electrons gain access to the loop? p. 13-The authors estimate the density in the loop using a DEM analysis of the AIA data. Is this necessary? They are trying to establish "whether sausage modes could be responsible for the radio pulsations". If the radio emission is due to fundamental plasma radiation the density in the source is ~6x10^8/cm^3, a factor of a few less than they estimate based on the DEM. How does the DEMbased estimate relate to the assumption of a density of 10^9 at the base of the loop? p. 14-Using an average extrapolated magnetic field strength seems hazardous. Do the authors have any way to determine and propagate uncertainties in estimated quantities here? Bottom: The authors argue (in the supplementary material) that the pulsations are not well described by an L-V system, yet they comment on p. 9 that the "light-curve of the pulsations shows a sawtooth like pattern ... (this sawtooth characteristic is unusual, as [MHD] pulsations usually have a symmetric rise and decay phase)." This discrepancy between the observations and the MHD interpretation is not adequately addressed. p. 15-"Observations of such large spectral slopes are rare". Insofar as narrowband emissions are often seen, this statement is not accurate. The "only emission that can explain such spectra..."? The references given are generic and do not really show that ECM amplification of Z-mode is the only possible mechanism. p. 16-The authors claim that the observed pulsations are highly X-mode polarized. On the other hand, they claim that the reverse-drift emissions are beam-driven fundamental plasma radiation, which should be O-mode polarized. Is such a difference in polarization observed? p. 17-"roughly coincides" -please be more precise and show, for example, in Fig. 1 of the supplementary material the start time and end time during which the EUV wave could be interacting with the quiet AR. The authors again acknowledge the expectation that an MHD sausage mode would yield a symmetric pulse profile. They then comment that "In this instance given the involvement of electron acceleration and the asymmetric pulse profiles, the modulation may be indirect." What does this mean? How does it yield the observed sawtooth pattern? The point made above that the ECM source and the electron beam emission should be oppositely polarized is relevant here, too. Reviewer #3 (Remarks to the Author): The authors managed to revise the paper according to my comments, and I am happy to recommend the acceptance. Reviewer #2 (Remarks to the Author): I thank the authors for being responsive to referee input, which was extensive. While the revision addressed many points raised by the referees, it also raises new questions. In addition, for this referee, the primary objection to the work remains: that the interpretation of the observations is more complex than the data can justify and relies on too many assumptions and guesses to be convincing. I fully acknowledge that this is an interesting and important observation but, in my opinion, the authors over-interpret the data. I cannot recommend the paper for publication in its present form. I urge them to confine their attention to a clear presentation of the data and a fairly brief and critical assessment of what the data might imply about energy transport, release, and modulation. Response 1: We would like to thank the referee for further detailed comments on our revised article. To address the referee's remaining concerns we have re-written parts of the article (indicated in blue text) and Supplementary Figures 2 and 8. We have also shorted the text and made the discussion section more concise. We are confident that this additional analysis addresses the referee's remaining concerns and is fully supportive of our hypothesis. Additional comments: Intro: p. 2 -"This energy release results in the acceleration of particles to relativistic energies and the generation of high intensity bursts of radio emission." Well, EM radiation across the spectrum including, notably, hard X-ray emission. Response 2: Agreed. Text in the introduction has been changed to account for this. p. 2 -"The radio emission is thought to be generated by a plasma instability..." Perhaps at this particular wavelength. At frequencies above 1-2 GHz plasma emission rarely plays any role. Response 3: Agreed. Changed to "The radio emission at metric wavelengths". p. 2-"Direct high cadence radio imaging of the behaviour of electron beams during a pulsation event has the ability to confirm one (or some combination) of these mechanisms." Has the "potential" to confirm one or more mechanism. Response 4: Agreed. Changed to 'potential'. General comment: it is puzzling to me that the authors do not make use of RHESSI observations. (They do use FERMI data in their earlier paper on this event.) There is coverage during the time in question. It would be useful to establish whether there is HXR emission in the non-flaring active region or whether the HXR emission in the flaring active region has any connection to the distant pulsating radio source. The same applies to ancillary radio data from, for example, Ondrejov. Response 5: Agreed. We have now included a new Supplementary Figure 8 showing the pulsations compared to FERMI GBM fluxes at energies 26-50 keV and 50-100 keV. While individual pulsations are not possible to distinguish in the HXR flux (it is primarily noise at such timescales), there is a long timescale trend in the HXR flux from 12:55-12:58 UT that the radio pulsation flux follows quite closely. Furthermore, we have also included a new analysis of microwave data from RSTN San-Vito, see Supplementary Figure 8. This shows that metric radio emission from the quiescent active region follows closely both HXR and microwave, with the trends being particularly apparent in panel e. Overall, this provides extra evidence in favour of the idea that the pulsation source is due to electrons accelerated to lower altitudes (reverse drift bursts), with the microwave being from gyrosynchrotron emission from the electrons during the precipitation, and the HXR being from the the impact of electrons in the low corona/chromosphere where they emit via the usual thick-target model. During the pulsation time, the microwave and HXR light curve shows no correspondence with the Flare AR radio source (see panel e), so it is unlikely the electrons originate from the flare itself. As for RHESSI light-curves, it shows much the same behaviour as FERMI GBM, but at lower time resolution so we have not included it here. The Ondrejov data, along with Orfées, in Figure 1 below shows some sporadic emission from 500 to ~1200 MHz at the time of the pulsations. This may be an indicator of plasma emission due to energetic electrons associated with the pulsations. However, this emission also takes place at the same time as the 'Flare Continuum B' emission from the Flare AR, as described in Carley et al. (2016) for this same event. We cannot specifically associate the Ondrejov bursty emission to energetic electrons accelerated during the pulsations. Results: p. 6-"possible cyclotron mechanism or loss-cone instability" The cyclotron mechanism refers to a resonance; a loss-cone provides free energy that could drive either plasma radiation or cyclotron maser radiation. Response 6: We have altered the end of this sentence to simply state the emission mechanism may be related to the magnetic field, making no further assumption on the type of resonance or emission mechanism at this point in the article. I don't understand the argument that the type III "most likely" injected energetic electrons into the quiet AR. They type III appears to be directed out into the corona and perhaps the IPM. Is there evidence that it is bidirectional? The rise in radio flux in the quiet AR appears to be well underway by the time the type III occurs. Response 7: Agreed. We have changed the interpretation here to include the slow rise before the type III. The slow rise is an indicator of an increase in activity, while the type III is at least part of this initial activity. Supplementary Figure 8 now shows an analysis of the relationship between metric radio, HXR and microwave. The initial type III labelled as peak 'p1' in panel c is seen in all three, indicating that the electron acceleration is indeed bi-directional, i.e., the microwave and HXR peak indicates electron precipitation and impact in the chromosphere at the same time electrons escape into the corona to produce a type III burst. As for the rise in radio activity before the type III, this begins at ~12:45 UT and may be related to the rise in activity in the flare AR. For example in Supplementary Figure 8c there is rise in HXR and microwave activity. This is accompanied by a rise in activity both in the Quiet and Flare AR around 12:45 observed in 228 MHz. It is possible that some level of sympathetic flaring took place between these two active regions e.g., due to the eruption and expansion of the Flare AR, the Quiet AR becomes perturbed and increases in activity. In a new Supplementary Figure 2 we plot the integrated flux over the Quiet AR pulsation region from AIA in all EUV channels. There is a slow rise from ~12:45 UT in the hotter filters (131, 94 Angstroms), while the cooler channels decrease in flux -overall this indicates that the emission measure weighted temperature of the plasma is decreasing, however there is some level of activity in the 'hotter' channels that indicated a heating. The rise in EUV flux along with the rise in microwaves, HXR and metric radio is a good indicator that there is some electron acceleration occurring within the Quiet AR in the lead up to the pulsation activity. It might be suggested that electrons accelerated in the flare AR had access to the pulsations site, however there is no correlation between radio activity from the Flare AR and Quiet AR during the pulsations (Supp Figure 8e). Text has been added to Page 6/7 and the Discussion section of the main article and the Supplementary material to account for these points. p. 7&8-The case for reverse drift components -and for downward electron beams -is not very convincing and may be more apparent than real. Were the spectral data corrected for the instrument bandpass? Response 8: The reverse drift components are weak, but they have the same spectral characteristics as reverse drift bursts, including a drift rate that matches previous observations and a speed (using the density model derived from the DEM estimates) that matches well with the speed predicted for electrons generating such reverse drifts. This is now combined with the close correlation between the HXR and microwave observations as described above, providing more evidence that the pulsations are indeed from continuously accelerated electrons. Yes, the instrument is calibrated to provide SFUs and the dynamic spectra are background subtracted. This ensures the spectrogram is 'flattened' for antenna and instrument response and the features we see are real. p. 9-Why do the authors claim that electrons accumulate at the loop footpoint? Downward-moving electrons will either precipitate or mirror back up to greater heights. Response 9: Agreed. Accumulate is perhaps the wrong word to use. We have changed the wording here from accumulate to 'concentration' to indicate that some electrons are located around Q. Fig. 5a, the systematic change in the source centroid with an apparent speed of ~700 km/s is striking. Why is the motion in all bands confined to the observed direction? How does the detailed timing relate to the EUV wave, which has a speed that is only 310 km/s? Response 10: Referring to We have now revised a more detailed analysis of the source motion. Due to these new results we have moved this discussion to the Supplementary Material in the section 'Apparent motion of the pulsating radio source'. This includes Supplementary Figure 5 that shows all source maxima motion (including large-scale drift eastward and fine timescale back-forth motion during pulsations) is due to flux variation between radio sources at P and Q e.g., the figure shows that a successive flux increase over several minutes at P leads to and apparent source drift eastwards, while the superimposed pulsing of the source at P leads to the back-forth motion evident in Supplementary Movie 1. We have now removed any mention of actual source motion or speed as this new analysis shows it is merely an apparent drift due to alternating source brightness at Q and P. p. 10&11: If the variation in the position of source maxima is apparent (p.10), it is not necessary to quote a speed of 0.1c (p.11). Response 11: Agreed. As mention in previous comment, and now shown in Supplementary Figure 5, all source motion is apparent, so we have removed any mention of source speed to avoid confusion. The authors have embraced the presence of electron beams with very little observational evidence. I am also puzzled by the statement: "This shows that while there is pulsing energisation of electrons at P, the electrons accumulate at the adjacent footpoints close to R and Q ...". Energisation due to what? Response 12: Given the new evidence of the pulsations overall trend being well correlated with HXR and microwave data (over long-timescales), we now have a stronger indication that the activity involved electron acceleration. This combined with the fact we also see reverse drift bursts is good evidence to suggest the presence of electron beams. As for the energisation, it is unclear as to what specifically accelerates these electrons. We have been careful not to over-interpret the data and merely mention that because we observe evidence for electron beams at a magnetic null-point point in the corona, magnetic reconnection is a viable mechanism. For clarity, a new Figure 8 in the main article now indicates the proposed mechanism, with positions of oscillating loop, site of reconnection, and P and Q indicated. As mentioned above and now shown in the new Supplementary Figure 5, the systematic motion is an apparent one. There is no observable overall motion of the site of energetic electrons towards the east. It is difficult to resolve loops on such size-scales in the NLFFF extrapolation. Despite this loop not being resolved specifically, it is present in the EUV data and clearly has magnetic footpoints in the magnetogram, shown in Figure 2 above. We have taken an average of the magnetic field strength in a box that approximates the size of this loop (see Response 15 below for interpretation of uncertainties involved in this extrapolation). From Figure 8 of the main article, access to the surrounding environment (footpoints at Q or the loop itself) could be due to to reconnection at the null point, with field lines around Q or the loop being involved in this reconnection. Again we stress that reconnection is not specifically observed, being mentioned only in the discussion section of the paper as a possibility. Lack of observation of reconnection however does not change the overall results of the paper. p. 13-The authors estimate the density in the loop using a DEM analysis of the AIA data. Is this necessary? They are trying to establish "whether sausage modes could be responsible for the radio pulsations". If the radio emission is due to fundamental plasma radiation the density in the source is ~6x10^8/cm^3, a factor of a few less than they estimate based on the DEM. How does the DEM-based estimate relate to the assumption of a density of 10^9 at the base of the loop? Response 14: The density estimate was necessary to derive the Alfvén speed, which then gives an estimate of the sausage mode period that such a loop would generate. The base density estimate was based on the order of magnitude of the loop density e.g., we assume that the base density of the corona surrounding/outside the loop region is of a similar order of magnitude to the small loop. From this base density we estimate that 6e8 cm -3 is at ~15 Mm altitude (slightly above the loop). Using a base density of 2e9 cm -3 (small loop average density) the source would be placed at 37 Mm, still within the vicinity of the loop. Considering the immediate environment of the loop has no greater density than the loop itself (the loop represents a slightly over-dense structure in this environment) this height represents a maximum altitude of 228 MHz based on a hydrostatic equilibrium model. p. 14-Using an average extrapolated magnetic field strength seems hazardous. Do the authors have any way to determine and propagate uncertainties in estimated quantities here? Response 15: While a range of estimates for the loop properties is possible given the range of parameters of width, temperature, density and magnetic field, a true estimate of uncertainty is not possible due to the lack of uncertainty analysis from the NLFFF magnetic magnetic field values (Wiegelmann et al. 2012). That said, given the relationship of period P to the Alfvén speed v, sound speed c and loop width a, a propagation of uncertainty shows that the P fractional uncertainty is given by Now, given the relationship of Alfvén speed v, magnetic field B and density n, and the sound speed c dependency on temperature T, the fractional uncertainties of these quantities are With delta_n/n=0.07, deltaT/T=0.2 and delta_a/a=0.1, the relationship between the P and B fractional uncertainty is shown in Figure 3 below. This shows that delta_P/P~delta_B/B, hence P is not highly sensitive to the B uncertainty. We have included this in the Methods section of the paper. ΔP The authors argue (in the supplementary material) that the pulsations are not well described by an L-V system, yet they comment on p. 9 that the "light-curve of the pulsations shows a sawtooth like pattern ... (this sawtooth characteristic is unusual, as [MHD] pulsations usually have a symmetric rise and decay phase)." This discrepancy between the observations and the MHD interpretation is not adequately addressed. Response 16: In the Discussions & Conclusions we gave a reason for the pulse asymmetry and the relation to an MHD wave mechanism. This involved the MHD wave being only involved in the electron acceleration and indirectly involved in radio emission modulation. We have clarified the text in the Discussion & Conclusion section to make this point clearer. p. 15-"Observations of such large spectral slopes are rare". Insofar as narrowband emissions are often seen, this statement is not accurate. The "only emission that can explain such spectra..."? The references given are generic and do not really show that ECM amplification of Z-mode is the only possible mechanism. Response 17: Narrow band emissions are often seen but to our knowledge the measurement of spectral slopes of 21 and -13 are rare in the literature. We have removed "the only emission that can explain…". We now say that such strong spectral slopes can be explained theoretically using plasma emission from a loss-cone instability (Z-mode). This theoretical explanation is supplemented by an ensemble of other pieces of observational evidence (source position at magnetic footpoints, association with B-field, polarisation, and ratio of plasma to cyclotron frequency) allowing us to conclude that this emission is due to the Z-mode of loss-cone instability. p. 16-The authors claim that the observed pulsations are highly X-mode polarized. On the other hand, they claim that the reverse-drift emissions are beam-driven fundamental plasma radiation, which should be O-mode polarized. Is such a difference in polarization observed? Response 18: Indeed fundamental plasma from type III radio burst is expected to be polarised in the sense of the O-mode. In this case, any emission associated with the pulsations is in X-mode. As explained in Melrose et al. (1978) and Dulk & Suzuki (1980) the polarisation of harmonic plasma emission can be O or X-mode depending on viewing angle. For viewing angles >90 degrees, the sense of polarisation can reverse, meaning harmonic plasma emission can be in the sense of the X-mode. Since these are reverse drift bursts and assumed to be downward propagating electrons on a negative magnetic field, the viewing angle for the emission is expected to be >90 degrees and in the opposite sense to what we would expect from a type III. Hence it is theoretically possible for the emission to be plasma emission and X-mode in the case reverse drift bursts. p. 17-"roughly coincides" -please be more precise and show, for example, in Fig. 1 of the supplementary material the start time and end time during which the EUV wave could be interacting with the quiet AR. Response 19: The approximation here was due to the fact the EUV wave is faint and its front is not very well defined in space. From Figure 7 of the main article, the EUV front makes contact with the quiet AR (~100 Mm) quite early in the event, around 12:45 UT. The could explain the rise in emission at this time observed in radio and HXR in Supplementary Figure 8 and in EUV in Supplementary Figure 2. It first interacts with the pulsation region at ~12:54 UT, when the pulsations begin. One of the last interactions the wave has with the Quiet AR is at 13:03 UT, seen in Supplementary Figure 1c, after which the radio emission diminishes. We have included these points in new text in the 'EUV wave kinematics' section. p. 18&19-the words "most likely" or "likely" are over-used because the various suggestions have not been compellingly demonstrated. Response 21: Pages 18 & 19 are reserved for a discussion section of the paper where we postulate the most likely interpretation of the observations and analysis we have presented. We have been careful here not to make definitive statements. For example, we have shown that there is good evidence to suggest that an MHD wave was involved in these pulsations, and there is evidence to the contrary for the LV system. Use of the word 'likely' in this instance means there is more supportive evidence for the MHD oscillation mechanism. We have changed the language on page 18 & 19 in this regard The authors again acknowledge the expectation that an MHD sausage mode would yield a symmetric pulse profile. They then comment that "In this instance given the involvement of electron acceleration and the asymmetric pulse profiles, the modulation may be indirect." What does this mean? How does it yield the observed sawtooth pattern? Response 22: We have now added further detail to the Discussion & Conclusion section and a new Figure 8 to clarify this point. We also mentioned on page 19 the sawtooth rise-phase being due to build up of loss-cone emission (and sometimes direct observation of the reverse drift bursts), while the sawtooth decay phase is due to loss-cone emission quenching via particle injection. This section also makes reference to the Supplementary Material where we discuss the theoretical details and plausibility of this scenario in the 'Viability of the sudden reduction mechanism' section. The point made above that the ECM source and the electron beam emission should be oppositely polarized is relevant here, too. Response 23: As in the Response 18, for reverse drift bursts the sense of polarisation for plasma emission can be X-mode in the case of reverse drift bursts. I am afraid the authors and I are not really converging. Below are a few additional comments on the various components of their scenario. If I've mischaracterized or misunderstood aspects of the scenario after three readings, it is indicative of how hard it might be for a typical reader to follow. I leave it to the editor on how they wish to proceed. The authors suggest the following scenario to account for the observed radio imaging (NRH) and spectroscopic (Orfées) observations: • An MHD sausage mode oscillation is excited in a coronal loop by the passage of an EUV wave driven out from a large flare in an adjacent active region o I think the authors have fairly convincingly shown that the passage of the EUV wave is associated with the onset of the pulsations • The oscillation leads to the periodic acceleration of electrons near a magnetic null point where intense 228 MHz emission is produced o The weak pulsations in source Q could just be a sidelobe response to the strong pulsations in P (p. 7) o In Fig. 5, panels a-c, where are P and Q relative to the candidate pulsating loop? o The authors claim that the intense, pulsating 228 MHz emission is fundamental plasma radiation (p. 12). What produces (and confines) the underlying spectrum of plasma waves? What is the degree and sense of circular polarization of the pulsations? Same as it is for the emission near 210 MHz (Fig. 6)? • The accelerated electrons stream down from P toward the foot points of adjacent loops at site Q; the downward streaming electrons produce reverse-drift type III radio emission o Are the reverse-drift type IIIs fundamental plasma radiation? What is their polarization degree and sense? o Is there any evidence for classical type IIIs drifting from 228 MHz to lower frequencies in the spectrum? What is the emission <200 MHz in Fig. 7 in the Supp. Materials? o Looking at the Fig. 4, lower panel, it appears that there is a radio signature correlated with the type IIIs at frequencies up to 327 MHz. Can't the 327 MHz data be filtered (remove the background with a running boxcar or similar) in order to isolate the location, timing, and polarization of the purported reverse-drift type IIIs at 327 MHz source relative to P and Q? o I asked the authors why they hadn't used FERMI and RHESSI HXR observations. They now make use of HXR and microwave light curve data and claim that it is well correlated with the observed meter-wavelength phenomena. But what about RHESSI maps? It would be extremely interesting and important to see where an associated HXR source is located relative to P and Q. • At site Q, a loss cone is set up that is unstable to the production of Z-mode waves that are subsequently converted to electromagnetic waves -the emission is imaged at 228, 298, and 327 MHz o I find the section beginning on p. 13 and continuing (emission mechanism and loss-cone instability modulation) to be very confusing. The claim is that a "secondary mechanism" must be in play to account for the strong narrow-band emission that peaks at ~210 MHz and the authors suggest a loss-cone instability. Isn't that source P? What sets up a loss cone there? o Bottom of p. 14: the ratio of the electron plasma frequency to the electron gyrofrequency is given as 1.6. The statement is made that the ratio is 2 for harmonic emission. Not 3.2? Harmonic plasma radiation is near 2 times the plasma frequency. o The authors suggest that z-mode waves are amplified by the loss-cone and then coalesce, producing x-mode harmonic radiation. They also point out the fact that the 228-327 MHz emission seems to be associated with loop footpoints at source Q. Are the authors claiming loss-cone emission at both points P and Q and that it is harmonic plasma radiation? It is also hard to see how this is consistent with statements about the 228 GHz source on p. 12 since the harmonic emission at source Q would have a lower density than the fundamental emission at source P. o It is hard to see the relevance of several of the references cited in support of the above claims (e.g., the regimes considered by the references relevant to polarization [51-53] don't seem to be relevant to conditions in the source. • The loss cone is periodically quenched by the periodic introduction of fresh electrons streaming down from the pulsating acceleration site P o It seems the authors argue that the loss-cone-driven emission is relatively broadband (100 MHz) in Q. Either the source is extended (plasma radiation from a range of sites) or it is intrinsically broadband, perhaps due to conditions described by Winglee and Dulk (1986). o Why is the emission <200 MHz also subject to sudden reductions? For several reasons, the source at Q cannot be a sidelobe of P. Firstly, Q exists before P is present. Sidelobe sources cannot exist without the source of the primary beam. Secondly, side lobe source flux variation tends to be in-phase with the primary source. There are times during the pulsations when Q is directly out of phase with P e.g., one source often exists without the other. This means they are two independent, real sources. o In Fig. 5, panels a-c, where are P and Q relative to the candidate pulsating loop? Response 3: We have indicated where P and Q are on panel a. This makes it clear that the proposed pulsating loop is directly in the vicinity of P. o The authors claim that the intense, pulsating 228 MHz emission is fundamental plasma radiation (p. 12). What produces (and confines) the underlying spectrum of plasma waves? What is the degree and sense of circular polarization of the pulsations? Same as it is for the emission near 210 MHz (Fig. 6)? Response 4: For reverse drift bursts, the production of plasma emission is via the bump-on-tail instability, followed by induced scattering of Langmuir waves e.g, the standard plasma emission mechanism. As for the loss-cone emission, energetic electrons with an initial power law distribution in trapped coronal loops will eventually develop into a loss-cone distribution that becomes unstable to the production of Langmuir waves. The instability can last on the order of minutes (Benz & Tarnstrom 1976) and result in the production of radio emission via Langmuir wave scattering. In this sense, energetic electrons initially trapped in coronal loops produce the loss-cone emission. The fresh/repeated injection of new electrons would temporarily quench this loss cone. Benz & Tarnstrom (1976) also discuss that the loss cone instability re-establishes itself once the injected electrons have escaped the trap (if they have a fast enough v\|\|). Some of the newly injected electrons may also mirror (those with a smaller v\|\|) and add to the original loss-cone distribution which was already established. This is another means by which the original loss cone distribution could be sustained on the order of minutes From Supplementary Figure 4, the pulsations at 228MHz are negative Stokes V of up to 60-70%. We cannot observe sources at 210 MHz (outside any NRH observing bands), but it is likely they are the same as 228 MHz e.g., the 228 MHz NRH source and the 210 MHz Orfées light curve are identical, so they are from the same source. Unfortunately Orfées are currently uncalibrated in Stokes V for this event. • The accelerated electrons stream down from P toward the foot points of adjacent loops at site Q; the downward streaming electrons produce reverse-drift type III radio emission o Are the reverse-drift type IIIs fundamental plasma radiation? What is their polarization degree and sense? Response 5: It is not possible to definitively say whether they are fundamental or harmonic. A common assumption for plasma radiation is that it is harmonic, given that the emission has escaped the corona. We observe the reverse drifters at the time of the pulsations, which in NRH images are negative Stokes -V polarised. It is possible that the reverse drift bursts are themselves polarized, but given the fact we cannot observe them at all frequencies in their bandwidth using NRH, we cannot say for sure what their degree of polarisation is. Melrose et al. (1978) and Willes & Melrose (1996) describe that emission from the coalescence of Langmuir waves will be X-mode polarised if the Langmuir waves are isotropic. Since the reverse drifters propagate away from the observer, one would expect their associated Langmuir waves to be isotropic (because we observe the emission), and hence X-mode. This assertion applies only to harmonic emission. o Is there any evidence for classical type IIIs drifting from 228 MHz to lower frequencies in the spectrum? What is the emission <200 MHz in Fig. 7 in the Supp. Materials? Response 6: There is evidence in the form of the initial type III at 12:56:10 UT, which can be seen in Supplementary Figure 8 in the Orfées and NDA dynamic spectrum, which has a corresponding signature in the HXR and microwave data. However, there is no evidence of forward drift bursts related to the reverse drifters. The emission <200 MHz may be part of the entire continuum of emission produced during the ECM Z-mode. o Looking at the Fig. 4, lower panel, it appears that there is a radio signature correlated with the type IIIs at frequencies up to 327 MHz. Can't the 327 MHz data be filtered (remove the background with a running boxcar or similar) in order to isolate the location, timing, and polarization of the purported reverse-drift type IIIs at 327 MHz source relative to P and Q? Response 7: Indeed part of the 327 MHz emission in panel d may be from the reverse drifters. It is extremely weak in Orfées and difficult to isolate. The spectrum as it exists has been background subtracted, smoothed and intensity scaled. After much experimentation this is the best enhancement of the dynamic spectrum we can achieve to observe the reverse drifters. Due to the weakness of these bursts at 327 MHz it is difficult to isolate them in time and compare them to the source in the images. This is made more difficult by the fact that the 327 MHz source in the images does not show pronounced pulsations, so establishing a temporal relationship with the reverse drifters may not be possible. That said, 327 MHz is also negative Stokes V polarised, so is related to whichever mechanism is responsible for the lower 228 MHz pulsating source (Zmode ECM). o I asked the authors why they hadn't used FERMI and RHESSI HXR observations. They now make use of HXR and microwave light curve data and claim that it is well correlated with the observed meter-wavelength phenomena. But what about RHESSI maps? It would be extremely interesting and important to see where an associated HXR source is located relative to P and Q. Response 8: RHESSI is an indirect imager and has a limited dynamic range for imaging. Moreover, apart from snapshot imaging (requiring a high count-rate) "regular" imaging cannot be achieved at a time resolution less than 4s. Furthermore, at the time of the radio pulsations (sources P and Q) strong X -ray emission (as well as radio) is still observed in the "flaring" active region. In these conditions RHESSI maps cannot show X-ray emissions associated with P and Q, unless they are stronger than the emissions coming from the flaring AR. • At site Q, a loss cone is set up that is unstable to the production of Z-mode waves that are subsequently converted to electromagnetic waves -the emission is imaged at 228, 298, and 327 MHz o I find the section beginning on p. 13 and continuing (emission mechanism and loss-cone instability modulation) to be very confusing. The claim is that a "secondary mechanism" must be in play to account for the strong narrow-band emission that peaks at ~210 MHz and the authors suggest a loss-cone instability. Isn't that source P? What sets up a loss cone there? Response 9: We have rewritten and simplified this section so it is easier to follow. Many of the details about dominant ECM modes and ratio of plasma to gyro frequencies have been moved into the Supplementary Material.	19,009.2	2019-05-23T00:00:00.000	[ "Physics" ]
Revealing the unexplored fungal communities in deep groundwater of crystalline bedrock fracture zones in Olkiluoto, Finland The diversity and functional role of fungi, one of the ecologically most important groups of eukaryotic microorganisms, remains largely unknown in deep biosphere environments. In this study we investigated fungal communities in packer-isolated bedrock fractures in Olkiluoto, Finland at depths ranging from 296 to 798 m below surface level. DNA- and cDNA-based high-throughput amplicon sequencing analysis of the fungal internal transcribed spacer (ITS) gene markers was used to examine the total fungal diversity and to identify the active members in deep fracture zones at different depths. Results showed that fungi were present in fracture zones at all depths and fungal diversity was higher than expected. Most of the observed fungal sequences belonged to the phylum Ascomycota. Phyla Basidiomycota and Chytridiomycota were only represented as a minor part of the fungal community. Dominating fungal classes in the deep bedrock aquifers were Sordariomycetes, Eurotiomycetes, and Dothideomycetes from the Ascomycota phylum and classes Microbotryomycetes and Tremellomycetes from the Basidiomycota phylum, which are the most frequently detected fungal taxa reported also from deep sea environments. In addition some fungal sequences represented potentially novel fungal species. Active fungi were detected in most of the fracture zones, which proves that fungi are able to maintain cellular activity in these oligotrophic conditions. Possible roles of fungi and their origin in deep bedrock groundwater can only be speculated in the light of current knowledge but some species may be specifically adapted to deep subsurface environment and may play important roles in the utilization and recycling of nutrients and thus sustaining the deep subsurface microbial community. Introduction Fungi are mainly decomposers that play a major role in the biodegradation of plant materials in terrestrial ecosystems. In deep biosphere environments, however, fungal diversity, and their role in ecosystem functioning remains largely unknown. According to the small number of studies conducted thus far viable fungi have been detected in different sub-seafloor and subterranean environments, such as groundwater aquifers, continental sedimentary and hard rocks, and deep subseafloor sediments (Sinclair and Ghiorse, 1989;Madsen and Ghiorse, 1993;Fredrickson and Onstott, 1996;Palumbo et al., 1996;Raghukumar and Raghukumar, 1998;Ludvigsen et al., 1999). Based on a recent review by Nagano and Nagahama (2012) deep sea extreme environments harbor diverse fungal communities. These fungi represent mainly Ascomycota phyla with Eurotiomycetes, Dothideomycetes, Sordariomycetes, and Saccharomycetes being the most abundant fungal classes but also fungi belonging to Basidiomycota and Chytridiomycota have been detected with culture-independed methods. The first viable fungi isolated from deep continental hard rock environments originated from deep crystalline bedrock aquifers in Äspö, Sweden (Pedersen, 1987) and later several yeast species were detected also with DNA-based methods (Pedersen et al., 1996). Physiological properties of fungi isolated with traditional cultivation-based methods from deep crystalline bedrock fractures indicate that they are adapted to and capable of growing in the subterranean environment (Ekendahl et al., 2003). Identification of facultative anaerobic or strictly anaerobic fungi from deep sea environments indicates that anaerobic conditions are not a limiting factor for fungal growth (Cathrine and Raghukumar, 2009;Jebaraj et al., 2010;Raghukumar, 2012). Remains of bacterial biofilms in these environments also suggest that the biofilms may have supported the nutritious demands of the fungal cells in otherwise extremely oligotrophic environments (Gadd, 2006). Fungi generally prefer mono-or polysaccharides as carbon and energy sources, which would have been provided by the bacterial biofilms. In addition, fungi may be involved in the formation of humic aggregates and carbon contribution by fungal biomass as well as production of extracellular enzymes involved in the cycling of nutrients, as suggested by Raghukumar et al. (2010). Cultivation-based techniques reveal only a small part of the fungal communities in any environment and with these methods activity of fungi in deep subsurface environments cannot be determined. Novel sequencing technologies would have great potential for obtaining new information on the diversity and ecological role of fungi in the deep geosphere. However, in contrast to the characterization of the bacterial and archaeal communities of deep subsurface habitats, the fungal communities in deep crystalline bedrock fractures have not been characterized by modern culture-independent methods, such as high throughput amplicon sequencing. In this study DNA-and cDNA-based high-throughput amplicon sequencing analysis of the fungal internal transcribed spacer (ITS) gene markers was used to examine the total fungal diversity and to identify the active members of the fungal communities in deep bedrock fracture zones at different depths in Olkiluoto, Finland. The results reveal previously unexplored fungal communities in deep groundwater of crystalline rock fracture zones. Site Description and Sampling Olkiluoto is an island situated in the western coast of Finland. The bedrock of Olkiluoto belongs to the Fennoscandian Shield and consists mostly of Precambrian highly deformed and metamorphosed migmatitic mica gneisses. The characteristics of the site have been described in more detail by Pitkänen et al. (2004), Posiva (2013), Nyyssönen et al. (2012) and Bomberg et al. (2015). In brief, the groundwater in Olkiluoto is anaerobic and saline and salinity increases with depth from 0.1 g L −1 at ground level to 100 g L −1 at 900 m. The temperature of the groundwater varies from ca. 7 • C at 50 m to 20 • C at 1000 m and pH is slightly alkaline in all fracture zones. Sulfate is enriched in the upper 300 m and beneath this depth zone, only traces of sulfate are observed. The concentration of methane increases with depth from 300 m. Altogether, deep groundwater samples from 17 different boreholes at depths ranging from 296 to 798 m were collected between December 14th, 2009 and August 21st, 2013 from the Olkiluoto island in Finland (Table 1). Sampling was done as described in Bomberg et al. (2015). In short, the samples were collected from multi-packered boreholes as well as from open boreholes where the sampling section was packered-off in order to seal off a specific water-conducting fracture zone from the rest of the borehole. This isolated fracture zone was purged by pumping out the water collected between the packers and allowing water from the isolated fracture zone to run into the packered off section of the borehole. In order to assure that sample water was coming only from fracture zones, the packersealed fracture zones were pumped for at least 4 weeks before sampling. The conductivity and pH of the pumped water was followed, and when the values settled, it was assumed that the water represents the endemic fracture zone water. Microbial biomass for nucleic acid analyses was concentrated from 500 to 1000 mL samples by filtration on cellulose acetate filters (0.2 µm pore size, Corning) by vacuum suction in the anaerobic chamber. The filters were immediately cut out from the filtration funnels with sterile scalpels and frozen on dry ice. Geochemical Analyses of the Groundwater Conductivity, pH, total dissolved solids (TDS), alkalinity, total organic, and dissolved inorganic carbon and different cations and anions were analyzed from the sampled groundwater. Analysis methods are described before (Posiva, 2013;Bomberg et al., 2015). All analyses were conducted by Posiva Oy (Olkiluoto, Finland). Nucleic Acid Isolation Total DNA was isolated directly from the frozen cellulose-acetate filters. The filters were cut to pieces with sterile scalpels in a laminar flow hood, and the DNA was extracted and purified with the PowerSoil DNA extraction kit (MoBio Laboratories, Inc., Solana Beach, CA). The isolation was performed according to the instructions of the manufacturer. The isolated and purified DNA was then stored frozen at −80 • C until use. Total RNA was isolated directly from the frozen cellulose-acetate filter with the PowerWater RNA isolation kit (MoBio Laboratories, Inc., Solana Beach, CA). The filters were thawed on ice and care was taken to minimize the time of thawing. The intact filters were inserted into the bead tubes with flame-sterilized forceps and the RNA extraction was performed according to the manufacturer's instructions. Negative DNA and RNA isolation controls were also included. DNA contamination of the RNA extracts was checked by PCR with bacterial 16S rRNA gene specific primers 8F (Edwards et al., 1989) and P2 (Muyzer et al., 1993). If no PCR product was obtained, sample was assumed uncontaminated and the RNA extract was submitted to cDNA synthesis. If a PCR product was obtained, the RNA extract was first treated with DNase (Promega, Madison, WI) according to the manufacturer's instructions before cDNA synthesis. Aliquots of 11.5 µL of RNA was incubated together with 250 ng random hexamers (Promega, Madison, WI) and 0.83 mM final concentration dNTP (Thermo Fisher Scientific, Vantaa, Finland) at 65 • C for 5 min and cooled on ice for 1 min. The cDNA was synthesized with the Superscript III kit (Invitrogen), by adding 4 µL 5 × First strand buffer, 40 U DTT and 200 U Superscript III to the cooled reactions. To protect the RNA template from degradation, 40 U recombinant RNase inhibitor, RNaseOut (Promega, Madison, WI), was used. The reactions were incubated at 25 • C for 5 min, at 50 • C for 1 h and at 70 • C for 15 min. Two to four parallel reactions per sample as well as no template controls were performed. The parallel reactions were subsequently pooled. RT-PCR was also performed on the negative RNA extraction controls as well as negative reagent RT-PCR controls. Amplification Library Preparation The amplification libraries for 454 high throughput sequencing were prepared by PCR from the DNA and cDNA samples. Fungal ITS fragments were amplified in a two-step PCR. First, a 420-825 bp long fragment was amplified with primers ITS1F and ITS4 (White et al., 1990;Gardes and Bruns, 1993). Length of the ITS region varies between species (Manter and Vivanco, 2007). The product of this PCR was used as template in a secondary PCR with tagged primers ITS1F and ITS2 (Buée et al., 2009) generating a ca. 400 bp product. First step of PCR amplification was performed in 10 µL and second step in 50 µL reactions containing 1x KAPA Fidelity buffer (Kapa Biosystems, Cape Town, South Africa) (2 mM MgCl 2 ), 0.3 mM final concentration of dNTP, 6 pmol of each primer in 10 µL reaction and 25 pmol in 50 µL reaction, 1 unit of KAPA Hifi polyeraze enzyme (Kapa Biosystems, Cape Town, South Africa) and 1 µL of template. The PCR program for both PCR steps consisted of an initial denaturation step at 98 • C for 5 min, 39 cycles of 20 s at 98 • C, 50 s at 50 • C, and 30 s at 72 • C. A final elongation step of 5 min was performed at 72 • C. In addition negative reagent PCR controls with only PCR-grade water as template were performed to rule out possible contamination. PCR products were confirmed in 1 × SYBR safe-stained 1% agarose gel electrophoresis. The pyrotag libraries were sent for sequencing to Beckman Coulter Genomics (Danvers, MA, USA) where amplicon libraries were purified and smallest and largest fragments were removed based on fragment analysis. Pyrotaq libraries were run on a Genome Sequencer FLX 454 System according to manufacturer's protocol (454 Life Sciences/Roche Applied Biosystems, Branford, CT, USA). Sequence Processing and Analysis The sequence reads obtained from the 454 high-throughput sequencing were partly processed with in-house pipeline (Salavirta et al., in press). First, sequences were subjected to quality control using the MOTHUR software version v.1.31.2 (Schloss et al., 2009). During this step, adapters, barcodes, and primers were removed from the sequence reads, and the quality of base-calls was assessed in order to remove erroneous reads from the data set. Subsequently, chimeric sequence reads were removed from the data set with the USEARCH algorithm version 5.2.236 (Edgar, 2010) by de novo detection and through similarity searches against the 97% representative OTU set of the UNITE reference database (Kõljalg et al., 2013). Groups of similar sequences, i.e., Operational Taxonomic Units (OTUs), were selected from the chimera-filtered sequence data set following open-reference OTU-picking protocol of QIIME v. 1.7.0 (Caporaso et al., 2010) against the 97% identity UNITE database OTU sets (Kõljalg et al., 2013). OTU clustering was performed with UCLUST v. 1.2.22q (Edgar, 2010) and the seed sequences were selected as the representative OTU sequences. Only few fungal sequences were amplified from negative control. This is possibly due to two-step PCR and these OTUs based on these sequences were removed from the entire sequence data set. All reads that failed to hit the UNITE reference database with a minimum of 60% identity threshold were discarded as sequencing error. Next, singleton OTUs, i.e., OTUs that were represented by a single sequence, were filtered from the data set. Finally, taxonomy from domain to specieslevel was assigned to OTUs via representative OTU sequences with BLASTN with a maximum E-value of 0.001 (Altschul et al., 1990). Alpha diversity indexes chao1 (Chao, 1984) and Shannon diversity index (Shannon, 1948) were calculated from normalized sequence data where sequence data was subsampled to 1500 sequences to adjust for sequencing coverage. Heatmaps of the fungal communities were generated in the R environment (R Development Core Team, 2008) utilizing the reshape2 (Wickham, 2007), grid (Murrell, 2005), and ggplot2 (Wickham, 2009) packages. Statistical Analysis Non-metric multidimensional scaling analyses (NMDS) for comparing the similarity of the fungal communities at class level between the different samples and the effect of chemical parameters on the microbial communities was performed using the PAleontological STatistics (PAST) program (Hammer et al., 2001). Non-Euclidean Bray-Curtis distance matrix was generated with PAST and correlation coefficient values of the matrix was calculated with 1000 permutations with R. In addition Pearson's correlation between total (DNA fraction) and active (RNA fraction) fungal communities was calculated at genus level with compare_taxa_summaries.py command within QIIME. Venn diagrams of each sample were calculated with MOTHUR showing shared OTUs between DNA and RNA fraction. Accession Numbers The fungal ITS gene region sequences have been submitted to the European Nucleotide Archive (ENA, https://www.ebi.ac.uk/ena/) under accession numbers ERS706390-ERS706426. Fungal Diversity and Community Structure in Different Bore Holes Fungal sequences were detected in the DNA fraction of all of the 19 analyzed fracture water samples and in 18 samples of the RNA fraction. In total 378,831 quality-filtered fungal ITS sequences were obtained from the different fracture zones. The number of obtained sequences ranged from 306 to 24,616 in different samples with a median of 10,941 sequences per sample (Table S1). When comparing the Chao1 OTU richness estimate values to true detected OTU numbers, 22-100% of estimated fungal OTUs were obtained from the subsampled sequence data meaning that sequencing depth was sufficient enough to fully characterize the fungal communities in most of the samples. Altogether, 965 fungal OTUs ranging from 33 OTUs at 328 m in OL-KR6 to 163 OTUs in OL-KR9 at 423 m in DNA fraction and 7 OTUs in OL-KR3 at 303 m to 69 OTUs at 330 m and 405 m in RNA fraction were detected in the total sequence data (Table S1). Fungal diversity based on subsampled OTU richness in the DNA samples originating from the fracture waters peaked at 347 m in OL-KR23 (79 OTUs) and was lowest at 328 m in OL-KR6 (19 OTUs) (Figure 1). In RNA fraction highest OTU richness was detected at 390 m in OL-KR46. However, no clear connection between sampling depth and fungal OTU numbers was detected. In the DNA fraction highest Shannon diversity index (H ′ = 4.3), which indicates the abundance and evenness of the species present, was obtained at 423 m depth in sample OL-KR9 and lowest at 330 m depth in sample collected from OL-KR6 (H ′ = 1.0). In the active fungal community highest diversity was observed at 510 m in OL-KR9 (H ′ = 3.5). No fungi were detected in OL-KR44 at 693 m in the active fungal community. Most of the observed fungal sequences belonged to the phylum Ascomycota (63.9%). Phylum Basidiomycota was represented by 8.9% of all the sequences. Other fungal phyla detected were Chytridiomycota, Glomeromycota, and Zygomycota, which were represented as a minor (0.1-2.1%) part of the whole fungal community. Altogether approximately 25% of the fungal sequences obtained were identified as fungal according to UNITE database but a more specific classification remained unknown. This can be due to insufficient representation of fungal sequences in the sequence databases or these species have not been characterized before and could be considered as novel. Structure of the fungal communities varied between different fracture zones and sampling times (Figure 2). Sequences affiliating with Ascomycota dominated fungal community in both DNA and RNA fraction in most of the fracture zones. However, Basidiomycota was the dominating phylum in OL-KR2 at 559 m (58%) in the total fungal community and in OL-KR9 at 423 m (98.5%) and OL-KR29 at 798 m in the active fungal communities. In addition, Chytridiomycota was the dominating phylum in OL-KR44 at 693 m (66%) in the total fungal community. Sordariomycetes from the Ascomycota phylum was the major active fungal class (47-100% of all the sequences) in most of the boreholes and also the dominating fungal class in the total fungal community in OL-KR13 at 296 m, OL-KR3 at 303 m, OL-KR25 at 330 m, OL-KR3 at 340 m, OL-KR49 at 415 m and OL-KR29 at 798 m (39-65%) (Figure 2). OTUs belonging to Sordariomycetes were mostly related to Nectria genus and minority to Fusarium, Pochonia, Pseudallescheria, and unidentified Hypocreales groups ( Figure S1). Fungal class Eurotiomycetes was identified as dominating fungal class in total fungal community of OL-KR20 at 323 m (84%) and OL-KR6 at 330m (2013) (95%) and in active fungal community of OL-KR25 at 330 m (94%). The majority of the sequences belonging to the class Eurotiomycetes were members of the Penicillium genus. In addition sequences most closely related to genus Aspergillus were found from OL-KR46 at 390 m and OL-KR2 at 559 m as a minor group. Other detected Ascomycota fungal classes in Olkiluoto groundwater samples were Dothideomycetes that dominated the active community in OL-KR6 at 328 m (2010) (40,5%), Leotimycetes that dominated the active fungal community in OL-KR49 at 415 m (47%) and total identified fungal community in OL-KR6 at 328 m (2010) (26%) and OL-KR1 at 572 m (23%) and Saccharomycetes in OL-KR25 at 330 m (19%) and OL-KR6 at 328 m (2010) (5%) total community (Figure 2). In addition sequences belonging to class Orbilliomycetes was only found in OL-KR13 at 296 m (2012), but they contributed only as a minor (1.2%) part of the fungal community. In OL-KR2 at 559 m in the total fungal community and OL-KR29 at 798 m in the active community where Basidiomycota was the dominating identified phylum most of the sequences were closely related to class Microbotryomycetes (37-39%) and more closely to Sporodiobolales order and Sporobolomyces and Rhodotorula genera ( Figure S2). In addition in OL-KR2 at 559 m Tremellomycetes and more specifically Cryptococcus-like yeast sequences were detected in total community and Malasseziales order in active community. Other identified Basidiomycota classes in Olkiluoto boreholes were Agaricomycetes in OL-KR13 at 296 m and in OL-KR49 at 415 m in the active community (Figure 2). Pucciniomycetes was detected in OL-KR46 at 390 m as a minor part of the total community (0.2%). The Chytridiomycota phylum dominated the total fungal community in OL-KR44 at 693 m (66%) and was also present in OL-KR29 at 798 m. Chytridiomycota sequences from these boreholes were most similar with order Rhizophydiales. No Chytridiomycota sequences were detected in the active fungal community. Statistical Analysis of Fungal Diversity and Correlation to Geochemistry Non-metric multidimensional scaling analysis of fungal communities and environmental parameters grouped the samples into four clusters in both DNA and RNA fractures (Figure 3 and Table S2). The deepest samples (693 and 798 m) clustered together in the DNA fraction that indicates that depth has an influence in the fungal community structure and fungal communities in the deepest communities are most similar. In addition, at greater depth higher salinity also affect the communities and slightly lower pH was observed to significantly correlate with fungal community structure (p < 0.05). At depths FIGURE 2 \| Heatmap of fungal taxonomy in deep groundwater of crystalline bedrock fracture zones in Olkiluoto, Finland. Taxonomic classification of the fungal sequence reads obtained by high throughput sequencing of the total (DNA) and active (RNA) fungal communities presented at the class-level. The samples are arranged by true vertical depth from the surface down. d, detected; but relative abundance less than 0.1%. from 296 to 340 m carbon availability (DIC, HCO 3 ) affected the fungal communities in both DNA and RNA fractions and these samples grouped together. In those fracture zones concentrations of DIC and HCO 3 were higher than in deeper fracture zones ranging from 4.1 to 33 mgC L −1 of DIC and 25 to 134 mg L −1 of HCO 3 ( Table 1 and Table S3). Amount of total organic carbon was highest at 693 m in OL-KR44 where concentration of NPOC was 110 mgC L −1 . However, in the NMDS analysis organic carbon concentration did not significantly affect the fungal community structure (p > 0.1). Ammonium, nitrate, nitrite, magnesium, and sulfate concentrations were associated with the changes in fungal community profiles especially in OL-KR46 at 372 and 390 m and OL-KR6 at 328 and 330 m in both RNA and DNA fraction that grouped together (Figure 3). Especially nitrite and nitrate had a significant effect on the fungal communities in the DNA fraction (p < 0.001), but concentrations are very low. The similarities of fungal communities between DNA and RNA fractions were assessed at genus level by Pearson's one-sided t-distribution tests, with the hypothesis that positive correlation would be detected between total and active communities. In the total dataset Pearson's correlation was 0.33 (± 0.3 95% CI, p < 0.001) between the total and active fungal communities. When we compared specific samples, statistically significant correlation (p < 0.05) was found in 10 of the 18 fracture zones (Table S4). Statistically significant Pearson's correlations varied between 0.20 and 0.99 with highest correlation found at 372 m in OL-KR46 and weakest at 303 m OL-KR3. Moderate or strong correlation was found in seven of the fracture zones (0.32-0.99). The number of shared OTUs between DNA and RNA fraction was 0-23% ( Figure S3). Discussion Terrestrial deep subsurface mycology is still an unexplored research field as the major research done until now has been focusing on the diversity and functional studies of bacteria and archaea. To our knowledge this is the first study where fungal communities in deep groundwater of crystalline bedrock fracture zones were studied using high-throughput amplicon sequencing. Our results show that diverse and active fungal communities exist in the deep subsurface in Olkiluoto, Finland. Actually, in most of the studied fracture zones the fungal diversity was higher than what has been detected so far in deep sea environment. Deep sea studies have reported up to 43 fungal OTUs with >99% sequence similarity by using fungal ITS region cloning and sequencing (Lai et al., 2007;Nagano et al., 2010;Singh et al., 2012) and Orsi et al. (2013a) detected up to 26 fungal OTUs with 454 pyrosequencing of the eukaryotic 18S rRNA region, whereas we detected up to 163 OTUs in different fracture zones and depths. In our study fungal ITS1 region was chosen over ribosomal genes because it is highly variable and can separate fungi even at species level (Lindahl et al., 2013). ITS2 is considered generally less variable in length than ITS1 and is somewhat better represented in sequence databases. However, ITS1 and ITS2 share many properties, and similar results can be obtained with these two marker genes (Bazzicalupo et al., 2013). With cultivation-based methods five Rhodotorula and Cryptococcus yeast species and 17 molds have been detected in Fennoscandian rock aquifers in Äspö, Sweden (Ekendahl et al., 2003). Fungi belonging to the Ascomycota phylum were the most abundant in Olkiluoto fracture zones and this is in good agreement with findings from deep sea environments (Nagano and Nagahama, 2012). The deepest fracture zones with higher salinity and temperature and lower pH, were the only ones where fungi belonging to the Chytridiomycota phylum were detected. In addition, the number of basidiomycete species also increased with depth (Figure 2). Altogether 25% of the fungal sequences remained unidentified and some of these OTUs could be potentially novel species that have not been characterized before. Fungal diversity in the deep crystalline fracture water in Olkiluoto was surprisingly high. Unexpectedly no fungal OTUs were detected in the RNA fraction at 693 m depth, where the highest concentration of total organic carbon that fungi could easily use in their metabolism was detected. Although no clear connection between fungal diversity and some of the geochemical parameters were detected, NMDS analysis showed that salinity, which increases with depth, had an influence on the fungal community structure in deeper fracture zones and also nitrogen compounds, sulfate, and inorganic carbon were associated with the changes in the fungal communities at more shallow depths. Fungi are involved in many biogeochemical cycles such as nitrogen and sulfur cycles and fungi are for example able to solubilize minerals, dissolute, and precipitate metal ions, degrade silicates and dissolve rock phosphates in oxygen-limited environments (Gadd, 2006;Sterflinger, 2010). Some filamentous fungi and yeast species are able to oxidase sulfur and sulfur compounds and release sulfate to environment (Wainwright and Grayston, 1989;Reitner et al., 2006;Sterflinger, 2010). Thus, fungi could provide sulfate to sulfate-reducing bacteria and could potentially be involved in the sulfur cycle in subsurface environment. Although fungi and bacteria are competing for the same low amounts of nutrients in subsurface oligotrophic conditions, they can also benefit from each other. For example, Fournier et al. (1998) found that the yeast Rhodotorula rubra has a stimulating effect on the growth of the iron sulfide-oxidizing bacterium Thiobacillus ferrooxidans. Similar cooperation between fungi and bacteria could potentially occur also in Olkiluoto deep fracture zones. Members of the Sordariomycetes were the most commonly observed fungi from Olkiluoto fracture waters. These fungi are ubiquitous and cosmopolitan and function in virtually all ecosystems (Zhang et al., 2006). The group includes pathogens, endophytes of plants, mycoparasites and saprobes involved in decomposition and nutrient cycling but their role in deep biosphere ecosystems is not studied. Sordariomycetes are together with Eurotiomycetes, Saccharomycetes and Dothideomycetes one of the most frequently detected fungal taxa in deep sea environments where living conditions resemble the ones of subterranean deep fracture zones (Nagano and Nagahama, 2012). However, phylotypes within the class Sordariomycetes are few and unique to the studied deep sea areas and their role and functions are still unknown. The fungal sequences obtained from Olkiluoto fracture waters belonging to the Sordariomycetes class were closely related to members of the Nectriaceae family and Nectria and Fusarium genera. The family Nectriceae includes, e.g., facultative anaerobic microscopic fungi capable of using nitrate or nitrite as alternative terminal electron acceptor in their respiration in the absence of oxygen (Kurakov et al., 2008). Fungal species belonging to Nectria have been detected in deep sea sediments (Singh et al., 2012). Some species belonging to genus Fusarium that are capable of denitrification have been found in deep sea environments, especially from oxygen-depleted regions (Jebaraj et al., 2010). NMDS analysis showed correlation between fungal community in OL-KR6 at 330 m, where the most Penicillium -sequences were detected, and ammonium, nitrate and nitrite concentrations. This indicates that these species might be involved in nitrogen cycle also in the deep fracture zones in Olkiluoto. Strong correlation between fungal diversity and nitrate has been found also from deep marine sediments (Orsi et al., 2013b) Penicillium and Aspergillus species are common in outdoor air and terrestrial environments but they are also frequently detected in deep sea environments (Nagano and Nagahama, 2012;Raghukumar, 2012). Deep sea species differed from terrestrial species by their physiological properties and that they were adapted to an aqueous environment (Raghukumar and Raghukumar, 1998;Damare et al., 2006;. Salt-tolerant Penicillium and Aspergillus species have also been identified from oxygen-deficient environments (Raghukumar, 2012) and from anaerobic marine sediments where they were reported to play on important role in the denitrification process (Jebaraj et al., 2010). This suggests a possible versatile role of fungi in major ecological processes in extreme nutrient-poor environments, such as Olkiluoto deep fracture zone fluid. Aureobasidium and Cladosporium genera found especially in OL-KR6 at 328 m from the RNA-fraction and OL-KR13 at 296 m from the DNA-fraction are reported in many deep sea environment studies . Common characteristics of these fungal groups are resistance or adaptation to high osmotic pressure that is essential for survival in extreme conditions such as the ones in deep subsurface environments. The most common classes of the Basidiomycota phylum detected in Olkiluoto groundwater were Microbotryomycetes, Tremellomycetes, and order Malasseziales. These fungal groups are common in deep sea environments (Nagano and Nagahama, 2012) and RNA transcripts from active species within these phyla have been detected in deep marine sediments (Orsi et al., 2013a). Yeast genera within class Microbotryomycetes detected from Olkiluoto were Rhodotorula and Sporobolomycetes related species. The Rhodotorula strain isolated from Fennoscandian rock aquifers in Äspö was able to grow in a wide range of NaCl concentrations (0-100 g L −1 ) and pH interval of 4-10 at temperatures ranging from 4 to 30 • C (Ekendahl et al., 2003). Salinity, pH and temperature in Olkiluoto are also within these limits. This indicates that these yeast species are adapted to conditions in the deep subsurface environment. In Olkiluoto Sporobolomycetes related species were also detected in the RNA-fraction, which proves that these species were active in these conditions. Cryptococcus-like yeasts from class Tremellomycetes that were identified from Olkiluoto have also been detected from other deep igneous rock aquifers sites on the Fennoscandian shield (Ekendahl et al., 2003) and deep sea environments like deep sea methane seeps (Takishita et al., 2006(Takishita et al., , 2007. Cryptococcus-like yeasts were identified from the active community in OL-KR2 at 559 m where also high methane concentration (386 ml L −1 ) was detected that fungi could potentially use in their metabolism. Yeast genera within the order Malasseziales observed in Olkiluoto fracture zone water were closely related to cultured Malassezia sp. LCP-2008 and uncultured Malassezia from deep sea sediments (Singh et al., 2012). Phylotypes belonging to Malassezia sp. have also been recovered from methane hydrate-bearing deep sea sediments (Lai et al., 2007). These yeast species could potentially be methylotrophic and could play a crucial role in converting methane into more accessible carbon and energy substrates for the use of the microbial community (Lai et al., 2007;Raghukumar et al., 2010). In addition, fungi have been found to be involved in methane release in a coal mine (Beckmann et al., 2011). In the coal mine, weathering of coal, and timber were initiated by fungi and in the lower, oxygen depleted regions fungi were observed to perform incomplete oxidation of coal and wood substrates and release reduced carbon substrates, which can be channeled into methanogenesis. Chytridiomycota that dominated in DNA-fraction of OL-KR44 at 693 m were most similar to the order Rhizophydiales that also have been found in oxygen-deficient marine environments (Raghukumar, 2012). Chytridiomycota are the earliest diverging lineage of fungi and produces zoospores, which indicates adaptation to aquatic environments (Nagano and Nagahama, 2012;Raghukumar, 2012). Chytridiomycota was not detected in RNA fraction. However, statistically significant correlation of fungal taxonomy profiles between total and active communities was found in majority of the fracture zones, which suggests that fungal communities were similar in both DNA and RNA fraction. In three of these samples only weak correlation and in eight of the fracture zones no significance correlation was found, which indicates that total and active communities in these fracture zones were different. In half of samples with no significant correlation insufficient sampling depth of the RNA fraction may have affected the result. In OL-KR6 at 328 m, OL-KR9 at 423 m, OL-KR49 at 415 m and OL-KR25 at 330 m total and active communities appear to be truly different, suggesting that different species are active in these fracture zones compared to total community. The universal distribution of the many fungal species detected also in deep subsurface environments raises the question about possible contamination. It is known that controlling contamination during drilling in hard rock is more difficult than for example sedimentary rock and life dwelling in fracture zones is exposed to drill water during drilling. Origin of the fungi in Olkiluoto fracture zones is unknown and possible runoff from the surface or contamination during drilling cannot be ruled out based on this study. On the other hand, fungi can be authentic members of the microbial community in the different fracture zones. In this study water from the drill hole was purged for a long time after drilling and by using packers only water originating from the specific fracture zone was collected. Hydrogeochemical characteristics of the fracture fluids also indicates that the chemical parameters were stabilized confirming that water from specific fracture zone was collected. Most importantly, active fungi were found from Olkiluoto groundwater and this indicates that fungi have adapted to deep biosphere conditions and are able to maintain cellular activity. The metabolic activities of fungi in deep terrestrial environment remain still unknown. However, the first fungal metabolic transcriptomics study from sub-seafloor environment confirms the previous suggestions of living fungi and active fungal metabolism in the deep marine biosphere (Orsi et al., 2013b). The authors showed that 5% of the obtained transcripts were involved in carbohydrate, amino acid, and lipid metabolism suggesting that fungi have a role in organic carbon cycling in sub-seafloor sediment. Fungal expression of transcripts encoding hydrolases involved in protein, carbohydrate, and lipid degradation suggests that they degrade a variety of organic substrates. Fungal dissimilatory nitrate reductase (nar) transcripts involved in energy production were found, which indicate that fungi are involved in nitrogen cycle, probably reducing nitrate and nitrite resulting from nitrate reduction performed by bacteria. In our study surprisingly high diversity of active fungi were detected for the first time in deep groundwater of crystalline rock fractures. Unlike bacterial 16S rRNA gene, the RNA fraction of fungal ITS is only present in the cell when the genomic copy is being actively transcribed and thus is a true evidence that fungi are active in the deep fracture waters (Blazewicz et al., 2013). The most interesting question now is what are these fungi doing in the deep fracture zones in Olkiluoto and in deep terrestrial environment in general and what is their role in the whole microbial community? Metatranscriptomic studies could be the answer and next step in understanding the functionality of the fungal communities in deep subterranean environments. Interesting was that the amount of organic carbon did not correlate with fungal diversity and activity in deep fracture zones suggesting that fungi may have some other functions in deep subterranean environments than degradation of organic materials.	8,080.8	2015-06-09T00:00:00.000	[ "Biology", "Environmental Science" ]
Electrochemical Study of the AISI 409 Ferritic Stainless Steel : Passive Film Stability and Pitting Nucleation and Growth The aim of the present work was to study the passive film stability and pitting corrosion behavior of the AISI 409 stainless steel. The electrochemical tests were carried out in 0.1 M NaCl solution at room temperature. The general electrochemical behavior was assessed using electrochemical impedance spectroscopy (EIS) measurements whereas the semiconducting properties of the passive film were evaluated by the Mott-Schottky approach. Pitting corrosion was investigated using potentiodynamic and potentiostatic polarization tests. Surface morphology was examined using confocal laser scanning microscopy and scanning electron microscopy (SEM). Energy dispersive X-ray spectroscopy (EDS) analyses were carried out to identify the composition of precipitates that could act as preferential sites for the onset of pitting corrosion. The results showed that the passive film presents n-type semiconductive behavior. Grain boundaries played an important role as pitting initiation sites for the AISI 409 stainless steel. Introduction AISI 409 is a ferritic stainless steel commonly used in automotive exhaust systems as well as farm equipment, structural supports and transformers 1 .A new trend for the ferritic stainless steels market is related to civil engineering applications such as the transport of drinking water 2,3 .The engineering applications of the AISI 409 stainless steel have been established due to its relatively low cost and good corrosion resistance at high temperatures.This material is a titanium-stabilized grade with low carbon content 4 .The presence of titanium and the low carbon content are typical features for decreasing the susceptibility to intergranular corrosion during welding operations or high temperature applications such as for components of automotive exhaust systems.Titanium has greater affinity for carbon than chromium, preferentially forming precipitates with this element instead of chromium carbide precipitates.Thus, the traditional sensitization mechanism of stainless steels by the precipitation of chromium-rich carbides would be prevented 5,6 .However, titanium-rich precipitation can lead to the formation of galvanic couples in the microstructure of ASI 409 stainless steels due to the different activity between the ferrite matrix and the precipitates.In this respect, localized corrosion attack would be triggered by the presence of titanium-rich precipitates 6 .TiC, Ti(C,N) and TiN precipitates have been reported to play a role in the onset of pitting corrosion of the AISI 409 stainless steel 7 . In spite of the critical role of pitting corrosion to the safe operation of stainless steel components [8][9][10] , few studies were devoted to investigate this phenomenon on the AISI 409 grade.Balusamy et al. 11 studied the effect of grain size and microstrain induced by surface nanocrystallization after surface mechanical attrition treatment (SMAT) for different times and using different ball sizes on the corrosion behavior of AISI 409 stainless steel samples.They observed that the corrosion behavior depends on the ability of SMAT to promote passivity.When the treatment conditions (ball size and treatment time) enabled the formation of a stable passive film, corrosion resistance was increased with respect to the untreated surface due to nanocrystallization.However, for more aggressive conditions (especially for high ball sizes) corrosion was facilitated due to an increase in defect density.Ha et al. 12 evaluated the effect of non-metallic inclusions on the pitting corrosion behavior of 409L stainless steels refined by the argon oxygen decarburization (AOD) and vacuum oxygen decarburization (VOD) processes.They observed localized corrosion around oxides (Ti, Ca) suggesting that these inclusions may act as active sites for pit nucleation for samples refined by the AOD process.Titanium nitrides were found in samples prepared by the VOD process and they appeared to be immune to localized corrosion in 0.5 M NaCl at 25 ºC.Other authors studied that welding operations decreased the pitting corrosion resistance of the AISI 409 stainless steel but the investigation was focused on intergranular corrosion rather than on the pitting corrosion mechanism.Notwithstanding, the presence of titanium-rich precipitates was confirmed and the onset of localized attack was affected by the type of precipitates formed during heating and subsequent cooling from the welding temperature 13 . Although the reports mentioned above bring useful information regarding the pitting corrosion behavior of the AISI 409 stainless steel, a systematic approach correlating microstructural features with the stability of the passive film, localized corrosion mechanism and pitting morphology of this material is not found in the literature.The aim of the present work was to fill this gap by investigating the passive film resistance and pitting corrosion behavior of AISI 409 stainless steel specimens using potentisotatic and potentiodynamic polarization tests, electrochemical impedance spectroscopy measurements and evaluating the semiconducting properties of the passive film by the Mott-Schottky approach.The microstructure of the alloy was studied using scanning electron microscopy (SEM) coupled with energy dispersive X-ray spectroscopy (EDS). Material and sample preparation The material used in this work was a 2.0 mm thick coldrolled AISI 409 ferritic stainless steel sheet.The steel was tested in the as-received condition.Its nominal chemical composition is shown in Table 1. The steel sheet was cut into pieces of approximately 1.0 cm 2 for the EIS, Mott-Schottky and potentiodynamic tests.For potentiostatic tests small-sized specimens were prepared with an area of approximately 0.1 cm 2 .This procedure was employed to prevent overlap of current transients.Next, the working electrodes were prepared by connecting the stainless steel pieces to a copper wire by means of a colloidal liquid conductive silver paste.The non-working surface areas were sealed with nail polish to prevent crevice corrosion.Next, the specimens were embedded in cold-curing epoxy resin.The exposed surface of each electrode was finished by wet-grinding with a series of emery papers from 200 to 2500 grit.Lastly, the working electrodes were polished with diamond past with 6 µm. Electrochemical tests All electrochemical tests were carried out using a potentiostat/galvanostat Autolab M101.The experimental set-up consisted of a conventional three-electrode cell with a platinum wire as the counter-electrode, Ag/AgCl as reference and the AISI 409 samples as working electrodes.The tests were performed in 0.1 M NaCl solution at room temperature. Three different sets of experiments were performed.The first one is comprised of an evaluation of the general electrochemical response of the electrode surface using electrochemical impedance spectroscopy (EIS) followed by analysis of the semiconducting properties of the passive film by the Mott-Schottky approach.Initially, the open circuit potential was monitored for 24 h in order to ensure a steady state condition.Next, EIS measurements were performed at the open circuit potential (OCP) in the frequency range from 100 kHz to 10 mHz with an amplitude of the perturbation signal of ±10 mV (rms) and an acquisition rate of 10 points per decade.Right after the EIS measurements, Mott-Schottky plots were acquired at a fixed frequency of 1 kHz.The potential was scanned from +0.3 V Ag/AgCl versus the OCP back to the OCP in the cathodic direction with a step of 25 mV.The results were used to characterize the passive film on the surface of the AISI 409 stainless steel electrodes. The pitting corrosion behavior was characterized by potentiodynamic and potentiostatic polarization tests.Thus, a second set of experiments was comprised of an initial monitoring of the OCP for 24 h.Next, the samples were potentiodynamically polarized from the open circuit potential up to +1.0 V Ag/AgCl .The sweep rate was 1 mV.s -1 . Another set of specimens was potentiostatically polarized at the anodic potentials of +100 mV and +200 mV with respect to the OCP for 1800 s.The OCP was initially monitored for 24 h before potentiostatic polarization.This procedure was employed to detect metastable pitting.All the electrochemical tests were conducted in triplicate. Microstructural characterization and pit morphology The microstructure of the AISI 409 stainless steel was observed using optical microscopy (Zeiss Axio Cam ICc 5) and scanning electron microscopy (Leica/Leo 440 I and Jeol/ JSM-6010).Elemental composition of precipitates within the ferrite matrix was determined using energy dispersive X-ray spectroscopy (EDS) analysis.Pit morphology was examined using confocal laser scanning microscopy (CLSM) (LEXT OLS4100 Olympus).X-ray diffraction (XRD) was used to confirm the ferritic structure of the samples and to check for the presence of any additional phases.The analysis was carried out in a Rigaku Multiflex diffractometer in the Bragg-Brentano θ-2θ geometry, using Cu-kα radiation.The angular range was scanned from 35º to 120º in 2θ with a step size of 0.5º and 25 s of acquisition time per step. Microstructural Characterization Figure 1a shows an optical micrograph of the AISI 409 stainless steel used in the present work.The sample was etched in a aqueous solution containing FeCl 3 and HCl.The ferrite grains are clearly visible.The ferritic structure was confirmed by XRD as shown in Fig. 1b.Only Fe-α peaks were observed (JCPDS 6-0696).Notwithstanding, the presence of precipitates was detected, as suggested by the optical micrograph of a non-etched sample shown in Fig. 1c.Titanium-rich precipitates have been identified in Ti-stabilized ferritic stainless steels.The presence of TiN, TiC and Ti(C,N) has been reported 7,14,15 .In this respect, the microstructure of the AISI 409 stainless steel was further characterized by SEM/EDS analysis in order to identify the precipitates observed in Figs.1(a) and 1(c). SEM micrographs in the backscattered electrons (BSE) mode of the AISI 409 stainless steel are shown in Fig. 2. Imaging in the SEM-BSE mode allows distinguishing between microstructural features with atomic number contrast 16 , thus ensuing the identification of small precipitates within a metallic matrix. The general aspect of the sample can be observed in Fig. 2a.Several small and dark features can be observed within the grey ferrite matrix.A more detailed view of the surface is presented in Fig. 2b.The elemental compositions of four selected precipitates were determined using the EDS detector coupled to the SEM instrument.These precipitates are numbered in Fig. 2b.The ferrite matrix is indicated as well.The EDS spectra of these regions are shown in Fig. 3. The spectrum referred to the ferrite matrix is mainly characterized by the main peaks of iron and chromium, as expected.Titanium peaks were observed for the precipitates 1, 2, 3 and 4. The main difference between the precipitates can be ascribed to the presence of carbon and nitrogen.While nitrogen was identified in precipitates 1 and 2 it was absent in the precipitates 3 and 4. According to Michelic et al. 7 the morphology is important to confirm the composition of precipitates in titanium-stabilized ferritic stainless steel.In this respect, rectangular precipitates are associated with pure TiN whereas, nitrogen-containing Ti-rich precipitates with irregular morphology are associated with Ti(C,N) precipitates.The absence of nitrogen in precipitates 3 and 4 suggest that they consist of TiC.Very small TiC can also be present in titanium-stabilized ferritic stainless steels as indicated by Kim et al. 15 .Thus, titanium-rich precipitates were unequivocally identified by SEM/EDS analysis.The corrosion behavior of ferritic stainless steels can be affected by these precipitates as confirmed by the intergranular corrosion studies published by Kim et al. 15 .In the next sections, we investigate the pitting corrosion behavior of the AISI 409 stainless steel and its relationship with the microstrucutural features described in the present section. EIS measurements EIS measurements were performed to characterize the general corrosion behavior of the AISI 409 stainless steel with respect to the stability of the oxide film in the electrolyte.The tests were performed in a 0.1 M NaCl solution at room temperature after an initial 24 h-period of OCP monitoring.The variation of the OCP with time is shown in Fig. 4.After an initial instability period, the OCP continuously decreased up to 67 ks of immersion.Then, a steady-state condition was reached up to the end of the test. The EIS plots are shown in Fig. 5.The Nyquist plot is characterized by a capacitive loop that is little flattened in the low frequency domain (Fig. 5a).This behavior is typical of passive metals such as stainless steels 17,18 .Bode plots are shown in Fig. 5b (phase angle and impedance modulus).The Bode phase angle plot presents a wide plateau extending from 10 2 to 10 0 Hz with the maximum phase angle reaching approximately -80º.A pure capacitive behavior is associated with a phase angle of -90º 19 , indicating an electrode interface that is capable of accumulating electrical charges, avoiding migration of aggressive species such as O 2-and Cl -from the solution across the interface 20 .The closer the phase angle is to -90º, the more perfect is the capacitive response of the electrode surface 21 .Hence, the AISI 409 stainless steel studied in the present work presented an EIS response close to that of a perfect capacitive behavior, suggesting it presents high corrosion resistance 22 .Furthermore, the impedance modulus at low frequencies is high and the plot is characterized by a slope of -1 which is typically observed for capacitive surfaces 23 .The phase angles slightly decreased below 10 0 Hz, indicating high surface stability with respect to the onset of charge transfer reactions and loss of corrosion resistance 24 . Electrical equivalent circuits (EECs) are often used to simulate the experimental EIS data, giving a physical interpretation for the electrochemical response of the different interfaces of the electrode in the electrolyte 25,26 .The data were modeled using a one-time constant EEC (Fig. 6).This EEC was adopted to describe the interface between electrolyte and surface oxides of uncoated stainless steels [27][28][29] .The fitted data are shown along with the experimental data in Fig. 5.In this model, constant phase elements (CPEs) are considered instead of pure capacitors to account for the heterogeneities of the electrode surface 30 .The impedance of a CPE (Z CPE ) is defined in equation ( 1) where j 2 = -1 is the imaginary number, ω is the angular frequency, Q is the magnitude of the CPE (related to its capacitance) and n is the exponent of the CPE is related to the roughness of the surface 31 .The value of n denotes the deviation from the pure capacitive behavior.For an ideal capacitor n = 1, for diffusion-controlled processes n = 0.5 32 . ( The physical meaning of each element is described as follows: R 1 is the electrolyte resistance, Q 1 and R 2 are the double layer capacitance and charge transfer resistance, respectively.The values of the circuit parameters are displayed in Table 2. Good fitting quality was achieved as shown in Fig. 5. 13.1 6.79 0.94 665 The thickness of the passive film can be estimated using the EIS fitted data.The effective capacitance (C) can be calculated from the value of Q according to equations ( 2) and (3) 33 .In these equations, α is the CPE exponent, Z j (f) is the imaginary impedance value for a given frequency (f), Re is the electrolyte resistance, and Rt is the charge transfer resistance.The values of Re and Rt were obtained from Table 2 as given the by fitting the experimental EIS data with the EEC shown in Fig. 6.The value of α was graphically determined according to the procedure described by Orazem et al. 33 by plotting the modulus of the imaginary part of impedance (\|Zj\|) versus the applied frequency (f) as shown in Fig. 7.This value was 0.95 and corresponds to the slope of the linear part of the \|Zj\| vs. f plot in the middle to low frequency range that is related to the response of the passive layer.The value of Q eff was graphically determined from the linear part of the plot of Q versus the applied frequency f, as indicated in Fig. 8.The effective capacitance (C eff ) was, therefore, calculated using equation (3).Next, the thickness of the passive film (L) can be calculated from equation ( 4) 34 where ε 0 is the vacuum permittivity (8.85.10 -14 F.cm -1 ) and ε is the dielectric constant of the passive film which can be assumed as 15.6 for stainless steels 35,36 .Using these values the thickness of the passive film was determined as 14.5 nm which is higher than other reported values for the passive film in stainless steels in low-chloride containing solutions (0.5 nm to 6 nm) [37][38][39][40] .The high charge transfer resistance (R 2 ) reported in Tab. 2 can be due to this relatively thick passive film. ( Mott-Schottky analysis The passive film on the AISI 409 stainless steel was further characterized by investigating its semiconducting properties using the Mott-Schottky approach.The semiconducting behavior is due to the presence of intrinsic point defects in the passive film, leading to an extrinsic semiconductor behavior.Cation interstitials, cations vacancies and oxygen vacancies are the main defects of passive films on stainless steels.The Mott-Schottky approach is often used to study the semiconductor properties of the passive films on these materials 41,42 .The n-type behavior is associated with electron donors and is related with the presence of oxygen vacancies and cation interstitials whereas the p-type behavior is associated with electron acceptors and is related to the presence of cation vacancies 43 .The Mott-Schottky relationships can be used to determine the type of semiconducting behavior and also the doping densities, according to equations ( 5) and ( 6) which are valid for n-type and p-type behaviors, respectively 44,45 .In these equations N D and N A are the donor and acceptor densities, e is the elementary charge, k is the Boltzmann constant, T is the absolute temperature, C is the capacitance, E is the applied potential, E FB is the flat band potential, ε is the dielectric constant of the passive film (15.6 for stainless steels) and ε 0 is the vacuum permittivity (8.85.10 -14 F.cm -1 ).The Mott-Schottky plot is obtained by plotting 1/C 2 against E. A linear relationship with positive slope denotes the n-type behavior whereas a negative slope accounts for the p-type behavior 46,47 . (5) (6) The Mott-Schottky plot of the AISI 409 stainless steel obtained after immersion for 24 h in 0.1 NaCl solution at room temperature is shown in Fig. 9. Only one linear region was obtained throughout the applied potential range which displays a positive slope.This behavior is associated with n-type semiconducting character of the passive film, according to the Mott-Schottky theory.Considering the same potential range employed for the Mott-Schottky plots obtained in the present work, similar results have been reported for the passive films formed on stainless steels [48][49][50] .The positive slope indicates n-type semiconductor and is associated with the Fe-rich outer layer of the passive film formed on stainless steels where oxygen vacancies and cation interstitials are the major dopants 51 .The donor (N D ) density can be calculated from the linear part . . . C e N E E e kT 1 2 of the Mott-Schottky plot, according to equation ( 5).The value of N D was determined as 2.25.10 19 cm -3 .This value is of the same order of magnitude as other reported values for passive films formed on stainless steels 52 and indicate that oxygen vacancies and cation interstitials are the prominent dopants on the passive layer of the AISI 409 stainless steel.The point defect model (PDM) 53,54 can be used to draw some conclusions about the passive film on the AISI 409 stainless steels.As stated in the PDM, oxygen vacancies on the surface of the passive film can react with chloride ions.In this respect, chloride ions are adsorbed into oxygen vacancies, generating cation/oxygen vacancy pairs.As a result, additional chloride ions can adsorb at the film/solution interface and, therefore, give rise to more cation vacancies.This process is autocatalytic.Passive film breakdown can occur as a result of excessive cation vacancies condensed during this process.In this respect, high donor densities in the passive film are associated with low resistance to pitting corrosion [55][56][57] .The results obtained in the present work point to a predominance of donors in the passive film formed on the AISI 409 stainless steel specimens that can lead to pitting susceptibility.The pitting corrosion behavior of the steel is characterized in the next sections. Potentiostatic polarization Current transients were monitored for 1800 s at +100 mV Ag/AgCl and +200 mV Ag/AgCl above the OCP after 24 h of immersion in 0.1 M NaCl solution at room temperature.The results are shown in Fig. 10.Current spikes in the potentiostatic polarization curve are indicative of pit nucleation and metastable pit growth 58 .A gradual increase of current density followed by sudden drop is typical of metastable events, indicating growth and repassivation of unstable pits 59 .This behavior was observed after 620 s for the sample tested at +100 mV (Fig. 10a).The maximum peak of current density was at approximately 0.0005 μA.cm -2 By increasing the applied potential to +200 mV the maximum current density related to the metastable events reached higher values (0.025 μA.cm -2 ) (Fig. 10b).A more quantitative approach was employed to study the metastable events.The radius of the metastable pits (r pit ) was estimated using Faraday's second law, according to equation (7).In this equation the pits are assumed to be hemispherical, M w is the mean atomic weight of the alloy (56 g.mol -1 ), n is the valence (2.16), F is Faraday's constant, ρ is the density of the alloy (7.8 g.cm -3 ), I peak is the current density at the peak, I bg is the current density at the background and t is the time of the event.Following this methodology an average value of 41 nm was found for the metastable pits formed at +100 mV and 151 nm for those formed at +200 mV.These results show the susceptibility of the AISI 409 stainless steel to pitting corrosion and can be related to the n-type semiconducting behavior of the passive film as determined from the Mott-Schottky plot in Fig. 9. 3. Following Ningshen et al. 60 the passive current density was determined at the middle of the passive range.These values are compatible with other reported values for ferritic and austenitic stainless steels in NaCl aqueous solutions 13,61 .Notwithstanding, this result confirm the susceptibility of the AISI 409 to pitting corrosion.Furthermore, the values of the parameters shown in Table 3 can be considered relatively low when compared to optimized stainless steel surfaces 62 . Pit nucleation sites In order to identify preferential pit nucleation sites samples of the AISI 409 stainless steel were examined after the potentiodynamic polarization test using CLSM.Figure 12 shown two selected CLSM images of the AISI 409 stainless steel after potentiodynamic polarization in 0.1 M NaCl solution at room temperature.Figure 12a shows a region of dissolution around a gold-colored precipitate.This type of precipitate is reported to be TiN for Ti-stabilized ferritic stainless steels 7 .Figure 12b shows a small pit formed in the boundary of a grey-colored precipitate.These precipitates were identified by SEM/EDS analysis as described in section 3.1 and are mainly composed of titanium. Further characterization of the pit site was carried out using SEM/EDS analysis.Figure 13 shows stables pits formed after potentiostatic polarization in 0.1 M NaCl solution at room temperature.Some grain boundaries are visible.Pit formed mainly at these sites.EDS elemental mapping of selected pits was performed in order to search for compositional features that could be related to the more active corrosion response of these sites.The area used for elemental mapping is shown in the SEM micrograph displayed in Fig. 14a.It shows several small dark regions spread through the observed area and one pit located at grain boundaries intersection between neighbor grains.EDS mapping for Ti, Cr and Fe are displayed in Figs.14b, 14c and 14d, respectively.Cr and Fe signals arise from the AISI 409 matrix, being homogeneously distributed throughout the whole area, excpet within the pit cavity and in some small dark spots.The dark spots are Ti-rich areas as indicated in Fig. 14b.It is seen that pit formation occurred preferentially at grain boundaries intersections and is not necessarily associated with Ti-rich areas.Notwithstanding, Ti-rich areas are found inside the grains and also at the grain boundaries.In this respect, the high energy grain boundary sites were the preferential regions for the onset of pitting corrosion and this effect could be enhanced by precipitation of cathodic Ti-rich precipitates. Conclusions The pitting corrosion of AISI 409 stainless steel was studied.The microstructure of the alloy consists of a ferrite matrix and titanium-rich precipitates.The passive film on the surface of the alloy presents n-type semiconductive character that favors adsorption of chloride ions according to the point defect model.The susceptibility to pitting corrosion was confirmed by potentiostatic and potentiodynamic tests.Metastable pitting was studied at the anodic potentials of +100 mV and +200 mV above the open circuit potential.Metastable pit radius was estimated as 41 nm and 151 nm after the tests conducted at 100 mV and 200 mV vs OCP, respectively.Pit nucleation sites were mainly grain boundaries intersections.Ti-rich particles were eventually found at the pit nucleation sites.However, pitting corrosion was not detect around Ti-rich precipitates when they were located inside the ferrite grains. Acknowledgements Authors are thankful to the Brazilian agency CAPES for the financial support.Dr. Nelson Batista de Lima (IPEN/ CNEN-SP) is kindly acknowledged for the XRD analysis. Figure 2 . Figure 2. SEM micrographs of the AISI 409 stainless steel: a) General view; b) Detailed view showing precipitates with different morphologies. Figure 3 .Figure 4 . Figure 3. EDS spectra of matrix and precipitates of the AISI 409 stainless steel sample.The spectra are referred to the regions identified in Fig. 2b. Figure 5 . Figure 5. EIS results for the AISI 409 stainless steel after immersion for 24 h in 0.1 M NaCl solution at room temperature: a) Nyquist, b) Bode plots. Figure 6 .Table 2 . Figure 6.EEC used to simulate the experimental EIS data. Figure 7 . Figure 7. Determination of the α parameter, following the procedure described by Orazem et al.31 Figure 8 . Figure 8. Determination of Q eff , following the procedure described by Orazem et al. 31 . Figure 9 . Figure 9. Mott-Schottky plot of the AISI 409 stainless steel after immersion for 24 h in 0.1 M NaCl solution at room temperature. Figure 10 . Figure 10.Current-time curves for the AISI 409 stainless steel immersed for 24 h in 0.1 M NaCl solution at room temperature at two anodic potentials: a) +100 mV Ag/AgCl vs. OCP; b) +200 mV Ag/AgCl vs. OCP). 4 Fig. 11 . Fig.11.In addition, the passive current density (I pass ) could be determined as well as the passive range (ΔE = E b -E corr , where E corr is the corrosion potential).The average values of these parameters are shown in Table3.Following Ningshen et al.60 the passive current density was determined at the middle of the passive range.These values are compatible with other reported values for ferritic and austenitic stainless steels in NaCl aqueous solutions13,61 .Notwithstanding, this result confirm the susceptibility of the AISI 409 to pitting corrosion.Furthermore, the values of the parameters shown in Table3can be considered relatively low when compared to optimized stainless steel surfaces62 . 12 Figure 11 . Figure 11.Tafel plot for the AISI 409 stainless steel after 24 h of immersion in 0.1 M NaCl solution at room temperature. Figure 12 . Figure 12.CLSM images of the AISI 409 stainless steel after potentiodynamic polarization in 0.1 M NaCl solution at room temperature: a) Dissolution around a gold-colored precipitate; b) pit formed in the boundary of a grey-colored precipitate. Figure 13 . Figure 13.SEM micrograph of the AISI 409 stainless steel after potentiostatic polarization in 0.1 M NaCl solution at room temperature. Figure 14 . Figure 14.a) SEM micrograph of the AISI 409 stainless steel after potentiostatic polarization, showing a pit at grain boundaries intersections and dark small Ti-rich precipitates; EDS elemental mapping for: b) Ti; c) Cr and d) Fe. Table 1 . Nominal chemical composition of the AISI 409 stainless steel. Table 3 . Parameters obtained from the potentiodynamic polarization test.	6,393.8	2017-09-18T00:00:00.000	[ "Materials Science" ]
Polarization Super-Resolution Imaging Method Based on Deep Compressed Sensing The division of focal plane (DoFP) polarization imaging sensors, which can simultaneously acquire the target’s two-dimensional spatial information and polarization information, improves the detection resolution and recognition capability by capturing the difference in polarization characteristics between the target and the background. In this paper, we propose a novel polarization imaging method based on deep compressed sensing (DCS) by adding digital micromirror devices (DMD) to an optical system and simulating the polarization transmission model of the optical system to reconstruct high-resolution images under low sampling rate conditions. By building a simulated dataset, training a polarization super-resolution imaging network, and showing excellent reconstructions on real shooting scenes, compared to current algorithms, our model has a higher peak signal-to-noise ratio (PSNR), which validates the feasibility of our approach. Introduction Polarization is one of the important physical properties of light. Different targets show different polarization characteristics due to differences in morphology, texture, water content, and dielectric coefficient of targets. Polarization imaging technology integrates spatial, spectral, and polarization information of the target object, which cannot only improve the information for acquiring the target object but also enhance the ability to detect and identify the target object [1]. At the same time, polarization imaging is a new type of photoelectric detection system. By using polarization imaging to measure the objective reflection of the degree of polarization and polarization angle information, the dimension of imaging information is increased, and the polarization information and the light intensity distribution in two-dimensional space are obtained. Currently, polarization imaging techniques have been widely used in the fields of target detection [2], image defogging [3], biomedicine [4], and remote sensing [5]. In recent years, a division of focal plane (DoFP) polarization sensor has been developed, which integrates an array of micro-polarizers of four polarization states into the image element of a focal plane sensor to collect polarization information of four polarization states simultaneously. This sensor can realize polarization imaging in real-time and ensure that each measurement is performed under the same illumination or radiation conditions. Compared with the time-sharing polarization device, it has the advantages of good realtime, small volume, small mass, compact structure, and high integration. However, the DoFP polarization sensor inevitably causes the problem of half the resolution, so improving the resolution while keeping fast polarization imaging remains a major challenge. In order to solve the polarization resolution reduction problem, various methods have been proposed, such as interpolation and deep learning methods. The interpolation algorithms include bicubic interpolation, Newton polynomial interpolation [6], etc. Deep learning-based algorithms include bicubic interpolation, Newton polynomial interpolation [6], e learning-based reconstruction methods include the polarization image super-re reconstruction network PDCNN [7], the end-to-end output of intensity images, larization degree images, and polarization angle images of Fork -Net [8] and oth ods. These methods mainly focus on the super-resolution reconstruction of polar ages and do not consider the affection of the optical system. With the develop computational imaging technology, computational imaging methods that incorpo Digital Micromirror Device (DMD) in optical systems to compress and encode th path and subsequently reconstruct high-resolution images are widely used in micr medium wave infrared imaging [9], holographic imaging [10], and spectral imag Our technique relies on the scene information being sampled and encoded computer-controlled DMD device, then imaged by the DoFP detector, and finall structed by the depth learning network. The polarization aberration model of the system designed was established based on the polarization transmission theory, training data with polarization aberration were obtained to train the super-resolu construction network. The experimental platform was built to verify the feasibili proposed network in real application scenarios, and the high-resolution polariza ages were successfully reconstructed with a low sampling rate. Block Compressed Polarization Imaging The block-compressed polarization imaging method is an application of th compressed imaging (BCI) theory [6] on the field of polarization imaging, combin the imaging principle of DoFP polarization imaging to obtain low sampling rate an resolution images of polarization information. The DoFP micro-polarizer sensor is composed of a CMOS photodiode sensor and a layer of polarizer added on th the photodiode, as shown in Figure 1. Every four adjacent pixels is used as a co tional unit, and each pixel takes the polarized light at only one of four orientation 45, 90, and 135 degrees [6]. The light intensity of different polarization transmission angles I 0 • , I 45 • , I 90 • and I 135 • is obtained by the four image elements. Since the circular polarization component of the natural target is small, the Stokes vector S can be given by The degree of linear polarization (DoLP)and the angle of polarization (AoP) are given by However, this imaging method inevitably brings a reduction in the resolution, with the polarized image resolution reduced to half of the resolution of the original detector. DMD devices are widely used in compressive imaging because of the high resolution, high reflectivity, and quick response [7]. DMD is an essentially reflective digital semiconductor light modulator consisting of millions of micromirrors on a semiconductor silicon substrate [11]. Each micromirror is controlled by electrostatics independently to determine the angle of deflection of the light. BCI imaging methods can use low-resolution detectors paired with high-resolution DMDs to achieve high-resolution imaging results at a lower cost on the detectors, especially in the medium wave infrared band. Based on our previous optical design results for super-resolution imaging systems [8], a DMD-based polarization super-resolution imaging optical system is designed by combining the above-mentioned Micro-polarizer array DoFP detector. The imaging optical system mainly consists of four parts: telescope objective, DMD, projection objective, and DoFP polarization detector. The target scene is imaged by the telescope objective in the intermediate image plane. Then, the intermediate image is modulated by the DMD and finally imaged on the FPA by the projection objective. The main system parameters are shown in Table 1. The structure of the optical system is shown in Figure 2a, and the modulation transfer function (MTF) of the optical system is shown in Figure 2b. It can be seen that the optical system has good imaging quality with MTF values close to the diffraction limit, basically avoiding the imaging quality degradation in the super-resolution reconstruction introduced by the wave aberration. Firstly, the target scene is sub-regionally processed by the DMD, each sub-region of the scene is projected onto the DMD corresponding sub-region and encoded simultaneously, and then the encoded light intensity information is captured by the focal plane detector, where each pixel of the detector corresponds to a sub-region of the scene, and then the acquired image information is used to reconstruct of the target scene image by the reconstruction algorithm. Analysis of the Effect of Polarization Aberration on Imaging Polarization aberration is used to characterize the change in amplitude, phase, and polarization state of the light after the light traveling through an optical system [12]. The off-axis optical system produces polarization aberration, which affects the polarization imaging quality [13]. To characterize the polarization aberration, a 3 × 3 coherency matrix was used to perform ray tracing of the DMD-based polarization optical system. We transform the local 3D eigen-vibration coordinates into the 3D global coordinates as shown in Figure 3. . Schematic diagram of the eigen-vibration coordinates (K(in, q), P(in, q), S(in, q)), interface coordinates (K(out, q), P(out, q), S(out, q)), and global coordinates (X,Y,Z). Firstly, the target scene is sub-regionally processed by the DMD, each sub-region of the scene is projected onto the DMD corresponding sub-region and encoded simultaneously, and then the encoded light intensity information is captured by the focal plane detector, where each pixel of the detector corresponds to a sub-region of the scene, and then the acquired image information is used to reconstruct of the target scene image by the reconstruction algorithm. Analysis of the Effect of Polarization Aberration on Imaging Polarization aberration is used to characterize the change in amplitude, phase, and polarization state of the light after the light traveling through an optical system [12]. The offaxis optical system produces polarization aberration, which affects the polarization imaging quality [13]. To characterize the polarization aberration, a 3 × 3 coherency matrix was used to perform ray tracing of the DMD-based polarization optical system. We transform the local 3D eigen-vibration coordinates into the 3D global coordinates as shown in Figure 3. Firstly, the target scene is sub-regionally processed by the DMD, each sub-region of the scene is projected onto the DMD corresponding sub-region and encoded simultaneously, and then the encoded light intensity information is captured by the focal plane detector, where each pixel of the detector corresponds to a sub-region of the scene, and then the acquired image information is used to reconstruct of the target scene image by the reconstruction algorithm. Analysis of the Effect of Polarization Aberration on Imaging Polarization aberration is used to characterize the change in amplitude, phase, and polarization state of the light after the light traveling through an optical system [12]. The off-axis optical system produces polarization aberration, which affects the polarization imaging quality [13]. To characterize the polarization aberration, a 3 × 3 coherency matrix was used to perform ray tracing of the DMD-based polarization optical system. We transform the local 3D eigen-vibration coordinates into the 3D global coordinates as shown in Figure 3. . Schematic diagram of the eigen-vibration coordinates (K(in, q), P(in, q), S(in, q)), interface coordinates (K(out, q), P(out, q), S(out, q)), and global coordinates (X,Y,Z). . Schematic diagram of the eigen-vibration coordinates (K(in, q), P(in, q), S(in, q)), interface coordinates (K(out, q), P(out, q), S(out, q)), and global coordinates (X,Y,Z). Coherency matrix in the eigen-vibration coordinates The vibrational plane of the vector is called the eigen-vibration plane of the ray. For the linearly polarized light with a light intensity I a , its eigen-vibration plane and propagation direction are described by the three-dimensional incoherence matrix J P3L , as follows: Coherency matrix in global coordinates To describe the polarization characteristics of the light after passing through different optical interfaces in the same coordinates, it is necessary to transform the coordinates of the light being traced and unify it into the global coordinate system. R is the transformation matrix from the eigen coordinate system to the global coordinate system; therefore, the three-dimensional global coherency matrix of the incident ray J P3(in) being traced can be expressed as: Coherency matrix in interface coordinates The three-dimensional global coordinate system is O-XYZ, but the coordinate system of the incident, refracted, and reflected rays is the S-P-K interface coordinate system. In order to describe the polarization transformation of the optical surface in the global coordinate system, it is also necessary to transform the interface coordinate system S-P-K into the global coordinate system. T 3r(q) denotes the Jones matrix of the optical interface q in the 3D interface coordinate system. The above transformation is carried out on each surface of the optical system to obtain the three-dimensional polarization tracking matrix P 3(q) of each surface; The 3D polarization trace matrix of the entire optical system P 3(opt) is obtained by multiplying the trace matrices of all interfaces together. 4. Calibration of polarization parameters for a micro-polarizer array detector The micro-polarizer array detector is used as a polarization detector element. When the extinction ratio is not ideal, the polarization parameters of the micro polarizer array detector should be calibrated, and its influence should be considered in the polarization transfer model of the system. Jones matrix of the micro polarizer array detector T 3(pol) is mainly determined by two parameters, extinction ratio E and maximum transmittance tx: Then the three-dimensional global coherency matrix of the polarizer detector P 3(pol) can be expressed as follows: The light intensity of different polarization transmission angles The 3D polarization trace matrix of the entire imaging system P 3(sys) = P 3(pol) P 3(opt) . The three-dimensional global coherency matrix of the exiting light J P3(out) is obtained as follows. Then, the light intensity I (out) of the incident light is obtained as follows. With the above equation, we can know that the light intensity affected by the polarization error of the optical system has a nonlinear mapping relationship with the incident light intensity, which will be compensated by the depth learning network proposed later. We add the data obtained from the ZEMAX to the equation and the intensity image is simulated with polarization aberration. The method in detail is as follows: First, to describe the polarization characteristics of the beam, it is necessary to unify all the beams into one coordinate system, so we transform the local 3D eigen-vibration coordinates into the 3D global coordinates. Suppose that the propagation vector of a beam of light in the global coordinate system O-XYZ is K = K x , K y , K z T . Its eigen vibration coordinate system is the three-dimensional coordinate system O-X'Y'Z' obtained by rotating the Z-axis of this global coordinate system in the direction of K (the Z' axis is in the same direction as K). The K x , K y , and K z from the data of the optical system designed by the zemax software are introduced into the equation to obtain the rotation angles ω y and ω x to calculate the transformation matrix R. The resulting transformation matrix R from the eigen-vibration coordinate system to the global coordinate system is With this rotation matrix R, the 3D eigen-vibration coordinate matrix can be converted into a 3D global coherency matrix. Then, we also need to consider the effect of the material-reflecting interface of the DMD on the beam polarization state, and we obtain the reflectivity coefficients r s(q) ,r p(q) of the s and p components on the interface q from the ZEMAX simulation data. The T 3r(q) is expressed as Finally, by adding the obtained transformation matrix R and T 3r(q) to Equations (5) and (6), the polarization intensity images of the four polarization transmission angles after passing through the optical system can be calculated by Equation (11). Compressed Coding Method Due to the unique structure of the polarizer array detector, we have improved on the traditional block-compressed coding method. In the process of the conventional block compression coding, each sub-region of the DMD projected to a single pixel of the detector is coded identically. To perceive polarization states, DoFP cameras integrate a micro-polarizer array with four polarization direction (0 • , 45 • , 90 • , 135 • ) units into the super-pixels of the focal plane array sensor [14]. Due to the unique structure of the polarization detector, we project every 4 × 4 micromirrors on the DMD into a block of 2 × 2 pixels on the detector, where each 1 pixel corresponds to a different DMD encoding, making different encoding patterns for pixels with different polarization transmission angles so that the reconstructed image can be obtained under a lower sampling rate. The conventional micro-polarizer array DoFP camera with 512 × 512 pixels can obtain four intensity images with different polarization transmission angles of 256 × 256 pixels, and the polarization characteristic image with 256 × 256 pixels can be obtained. Our method uses a micro-polarizer DoFP polarization detector with 512 × 512 pixels and a DMD with 1024 × 1024 micromirrors. Multiple intensity images with 256 × 256 pixels are obtained by multiple coded sampling, and the polarization characteristic image with 512 × 512 pixels is gained by the reconstruction network. In this way, the resolution of the polarization characteristic image is doubled. At each time when the DMD pattern is changed, the light intensity is encoded, and a light intensity image is acquired through the 2 × 2 polarization detector to complete a sampling. Compressed data with different sampling rates can be achieved by controlling the number of detector exposures. Our DMD control system can load 8-bit grayscale patterns by controlling the micromirror dithering. So, we implement the 8-bit quantized versions of random Gaussian measurement matrices. The 4 × 4 scene is multiplied by the DMD encoding as shown in Figure 4. When the number of samples is N = 16, the intensity information pixels collected are 16 × 256 × 256, which is the same amount of information collected by a detector with pixels 1024 × 1024, without compressed sampling, and a 512 × 512 image of polarization information can be reconstructed. Therefore, we designed 16 different 4 × 4 compressed sampling matrixes. We tried compressed sampling imaging at N = 12, 8, 4 and 1, i.e., compressed sensing rates CS = 0.75, 0.5, 0.25, and 0.0625, respectively. Network of This Paper Unlike the traditional image compression sensing optimal reconstruction algorithms, deep learning need to learn prior knowledge from massive data. It uses a depth neural network to establish the mapping relationship between input and output. The parameters of the network are trained and optimized by a large amount of data. The sequence images collected by the detector in the compression imaging system are input into the trained network, and the reconstructed images can be directly output. Inspired by ReconNet and ForkNet, we designed a fully connected convolutional super-resolution reconstruction network based on DMD compressed sensing imaging with the polarization aberration correction. Multiple sampled polarized low-resolution images after compression coding are differentially amplified as network inputs, and high-resolu- Network of This Paper Unlike the traditional image compression sensing optimal reconstruction algorithms, deep learning need to learn prior knowledge from massive data. It uses a depth neural network to establish the mapping relationship between input and output. The parameters of the network are trained and optimized by a large amount of data. The sequence images collected by the detector in the compression imaging system are input into the trained network, and the reconstructed images can be directly output. Inspired by ReconNet and ForkNet, we designed a fully connected convolutional super-resolution reconstruction network based on DMD compressed sensing imaging with the polarization aberration correction. Multiple sampled polarized low-resolution images after compression coding are differentially amplified as network inputs, and highresolution S0, DoLP, and AoP images are used as the network outputs. There is no need to train each polarization transmission angle image separately to obtain the final required key information with a simple network. At the same time, our network effectively avoids the block effect problem of DMD compressed imaging, eliminating the need for additional block effect elimination calculations. Our network has three functions: firstly, it learns the nonlinear mapping relationship between the four angular polarization images and the polarization characteristic images; secondly, it realizes the compressed sampling superresolution imaging; finally, it compensates the polarization reflection aberration introduced by the DMD and eliminates the block effect. In summary, our goal is to directly reconstruct the high-resolution S0, DoLP, and AoP images required for polarization detection at a lower sampling rate using the proposed convolutional network in combination with DMD compressed imaging techniques. The architecture of our method is shown in Figure 5. The DoFP polarization detector acquires multiple frames of the compressed coded image, and the size of the frames data is N × 256 × 256. Then, the frames are interpolated into N × 512 × 512 by bicubic interpolation. The three-dimensional data cube composed of these images is the input to the network. The traditional two-dimensional block network is improved into a three-dimensional network. The polarization imaging principle is combined with the compressed sensing coding method to expand the two-dimensional block information into three-dimensional information. The whole image is used as the network input, which can effectively reduce the block effect brought by the traditional block reconstruction algorithm in blocks and obtain better reconstruction results. Our network consists of a multilayer convolutional neural network with the convolutional kernel size shown in Figure 5. The step size of the convolution is set to 1, the activation function is ReLu, and we use appropriate zero padding to keep the feature mapping size constant in all layers. Our network consists of a multilayer convolutional neural network with the convolutional kernel size shown in Figure 5. The step size of the convolution is set to 1, the activation function is ReLu, and we use appropriate zero padding to keep the feature mapping size constant in all layers. The first layer of the convolutional network has a convolutional kernel size of 1 × 1 and generates 32 feature maps, which mainly extract features in each block. Inspired by Forknet, the second to fourth convolutional layers, respectively, use kernels of size (5 × 5, 3 × 3, 3 × 3) with filters (96, 48, 32), which are used to create a mapping of compressed data to low-resolution features. The previous settings for the fifth to seventh convolutional layers are repeated to create a mapping of low-resolution features to high-resolution features. The last layer, which is also the output of the network, directly outputs the 3D matrix, including the S0, DoLP, and AoP images representing the polarization characteristics. The loss function is the mean square error loss function MSE Loss in the PyTorch model. Training Data To train and evaluate the neural network proposed above, we built a dataset containing 100 sets of polarized images. First, we use a high-resolution micro polarizer array camera (2048 × 2448 pixels) to capture different scenes and intercept 1024 × 1024 pixels in the center of the field of view to obtain four 512 × 512 pixel intensity images with polarization transmission angles of 0 • , 45 • , 90 • , and 135 • , and s0, AoP, and DoLP images are calculated as the high-resolution ground truth images for the network input. Then, according to the light intensity tracing formula above, the high-resolution ground truth image and the optical system parameters we designed, the projected image with polarization aberration after DMD reflection is simulated. Finally, the projected image is dot-multiplied with the compressed sampling matrix to simulate the compression coding process and obtain multiple low-resolution polarization images acquired by the DoFP detector as the lowresolution image for the network input. In total, 80 sets were used for training data, 10 sets were used as validation data, and the remaining 10 sets were used for testing. To increase the amount of data, we chunked the images in each group. We split the intensity image of size 512 × 512 into smaller pieces of size 64 × 64 and generated more than 5120 patches. In addition, we used flips and rotations (0 • , 90 • , 180 • , and 270 • ) for data expansion. There are 40,960 patches in total used for training. Figure 6 shows the loss curve on both training and validation sets. multiple low-resolution polarization images acquired by the DoFP detector as the lowresolution image for the network input. In total, 80 sets were used for training data, 10 sets were used as validation data, and the remaining 10 sets were used for testing. To increase the amount of data, we chunked the images in each group. We split the intensity image of size 512 × 512 into smaller pieces of size 64 × 64 and generated more than 5120 patches. In addition, we used flips and rotations (0°, 90°, 180°, and 270°) for data expansion. There are 40,960 patches in total used for training. Figure 6 shows the loss curve on both training and validation sets. Figure 6. Loss curve on both training and validation sets. Simulation Firstly, the effect of polarization aberration on the intensity image is simulated by the formula in Section 2, and the images with polarization aberration are generated, as shown in Figure 7. We can see that the polarization aberration makes the light intensity value of each polarization transmission angle change significantly. Simulation Firstly, the effect of polarization aberration on the intensity image is simulated by th formula in Section 2, and the images with polarization aberration are generated, as show in Figure 7. We can see that the polarization aberration makes the light intensity value each polarization transmission angle change significantly. Table 2. It can be seen that the reconstruction network can still reconstruct a high-quality polarization image even at a lower compression rate equal to 0.0625. The reconstructed images with different compression rates of polarization are shown in Figure 8. To evaluate the reconstruction effect of the algorithm, we compared our algorithm with the compressed sensing algorithm OMP and the deep learning network ReconNet. The reconstruction process is as follows: the inputs of the three algorithms are all low-resolution intensity images of each polarization transmission angle and contain the polarization aberration; OMP and ReconNet are used to reconstruct the high-resolution images of each polarization transmission angle, respectively, whose S0, AoP, and DoLP are calculated by the high-resolution images of each polarization transmission angle after equations (1)(2)(3). Our algorithm outputs the polarization data S0, AoP, and DoLP directly without additional calculations. The ground truth images are high-resolution images without polarization aberration. The PSNR value is used to evaluate the reconstruction effect, and the comparison of the reconstruction effect of different algorithms is shown in Figure 9. The dataset established in 2.5 is used to train the network proposed in this paper. We trained the reconstruction network at N = (1,2,4,8,12) and compression rates cs = (0.0625, 0.125, 0.25, 0.5, and 0.75), respectively. The PSNR values of the directly reconstructed polarized images are shown in Table 2. It can be seen that the reconstruction network can still reconstruct a high-quality polarization image even at a lower compression rate equal to 0.0625. The reconstructed images with different compression rates of polarization are shown in Figure 8. of each polarization transmission angle, respectively, whose S0, AoP, and DoLP are calculated by the high-resolution images of each polarization transmission angle after equations (1-3). Our algorithm outputs the polarization data S0, AoP, and DoLP directly without additional calculations. The ground truth images are high-resolution images without polarization aberration. The PSNR value is used to evaluate the reconstruction effect, and the comparison of the reconstruction effect of different algorithms is shown in Figure 9. Table 3 shows the PSNR values of reconstructed images for different algorithms at a 0.25 sampling rate. It can be seen that the traditional compression sensing algorithm OMP cannot achieve polarization image reconstruction well; ReconNet inevitably brings a mosaic effect, which requires additional correction. In addition, it needs to reconstruct the high-resolution images of four polarization transmission angles first, and then calculate the polarization information image through the formula, which is more complex, and the polarization aberration cannot be compensated. Our algorithm can reconstruct the best image quality at a 0.25 sampling rate in these methods. Experiment We want to use this work on the high-resolution polarization imaging in real scenes with the compressed-encoded polarization imaging device we designed. Therefore, we applied the trained network to our designed experimental imaging platform. We have previously analyzed the alignment error of the DMD compression imaging system and pointed out that within the normal tolerance range, the alignment error has no significant impact on the imaging effect [8]. The core components of the imaging system are a DMD from TI, two dual telecentric projection lenses, and a micro-polarizer array camera from LUCID. In the process of real scene data acquisition, the image will also be affected by a variety of environmental uncertainties (noise, vibration, etc.), which can also lead to an inaccurate imaging model. We have done two experiments separately. The first experiment verifies that the DMD reflection system will cause polarization aberration and shows the impact on polarization intensity image; the second experiment uses the built experimental system to acquire real scenes. Through the depth learning network we propose, we can reconstruct the obtained image by super-resolution. Polarization Aberration Testing of DMD Devices To verify the influence of DMD on the polarization imaging system, we compared the polarization intensity images acquired by the optical system with and without DMD devices. The integrating sphere is used as a uniform light source, a visible light camera with the same pixel size as the micro polarizer camera is used as a detector, and two rotary polarizers are used as the polarizer and polarizer. First, a rotating polarizer 1 is placed in front of the integrating sphere light source as the polarizer, and a rotating polarizer 2 is placed in front of the visible light camera as the polarizer analyzer, which is set to 0 degrees. Then, the integrating sphere light source is turned on, the polarizer is rotated, and the light intensity values collected by the detector are recorded from 0 degrees to 180 degrees, respectively. Finally, the DMD is added to the imaging system, and all micro mirrors on the DMD are loaded with modulated images with a value of 1, that is, all light incident on the DMD is not coded, but is reflected. Similarly, the polarizer is rotated to record the light intensity reflected by DMD. The two groups of polarization intensity images obtained in the two cases are compared, as shown in Figure 10, and it can be seen that the addition of the DMD has an impact on the polarization light intensity, which needs to be compensated. Real Scene Polarization Imaging Experiment To verify the effectiveness of our reconstructed network in real scenarios, we built a two-armed experimental device, as shown in Figure 11. First, the scene is projected on the DMD through the converging lens and the Bi-telecentric objective lens 1. The DMD has a resolution of 1920 × 1080 micromirror elements, each with a size of 10.8 µm, and the micromirrors can be switched at high speed in the ±12 • direction to achieve 8-bit grayscale modulation by pulse width modulation. Then, the reflection direction of each micromirror of the DMD is controlled to encode and modulate the scene. Since the DMD micromirrors are flipped diagonally, we rotate the DMD by 45 degrees with an angle of 24 • between the two double telecentric lenses for the convenience of mounting. Finally, the encoded scene image is captured by the micro polarization camera after passing through the double telecentric objective 2. The polarization camera is chosen with the Lucid Phoenix camera with IMX250 CMOS, which is also rotated by 45 • . We selected a 1024 × 1024 pixel size area on the DMD, corresponding to a 512 × 512 pixel size area on the detector. The whole device is placed on an optical air floating platform to keep it stable. Real Scene Polarization Imaging Experiment To verify the effectiveness of our reconstructed network in real scenarios, we built a two-armed experimental device, as shown in Figure 11. First, the scene is projected on the DMD through the converging lens and the Bi-telecentric objective lens 1. The DMD has a resolution of 1920 × 1080 micromirror elements, each with a size of 10.8 µm, and the micromirrors can be switched at high speed in the ±12° direction to achieve 8-bit grayscale modulation by pulse width modulation. Then, the reflection direction of each micromirror of the DMD is controlled to encode and modulate the scene. Since the DMD micromirrors are flipped diagonally, we rotate the DMD by 45 degrees with an angle of 24° between the two double telecentric lenses for the convenience of mounting. Finally, the encoded scene image is captured by the micro polarization camera after passing through the double telecentric objective 2. The polarization camera is chosen with the Lucid Phoenix camera with IMX250 CMOS, which is also rotated by 45°. We selected a 1024 × 1024 pixel size area on the DMD, corresponding to a 512 × 512 pixel size area on the detector. The whole device is placed on an optical air floating platform to keep it stable. To verify the effectiveness of our reconstructed network in real scenar two-armed experimental device, as shown in Figure 11. First, the scene is pr DMD through the converging lens and the Bi-telecentric objective lens 1. Th resolution of 1920 × 1080 micromirror elements, each with a size of 10.8 µm cromirrors can be switched at high speed in the ±12° direction to achieve 8 modulation by pulse width modulation. Then, the reflection direction of eac of the DMD is controlled to encode and modulate the scene. Since the DMD are flipped diagonally, we rotate the DMD by 45 degrees with an angle of 24 two double telecentric lenses for the convenience of mounting. Finally, the e image is captured by the micro polarization camera after passing through t ecentric objective 2. The polarization camera is chosen with the Lucid Ph with IMX250 CMOS, which is also rotated by 45°. We selected a 1024 × 1024 on the DMD, corresponding to a 512 × 512 pixel size area on the detector. The is placed on an optical air floating platform to keep it stable. We photographed the toy vehicle with the experimental device, and the PSNR values of the reconstructed images at different sampling rates are shown in Figure 12. At the sampling rate of 0.0625, the reconstructed images are shown in Figure 13. We photographed the toy vehicle with the experimental device, and the PSNR values of the reconstructed images at different sampling rates are shown in Figure 12. At the sampling rate of 0.0625, the reconstructed images are shown in Figure 13. Although there is still a large gap between the experimental dataset and the real scene, we still reconstructed a good image quality, which indicates that our algorithm has certain robustness and can be improved even more by increasing the training data of real shooting in the future. Considering the improvement of some specific application object recognition results Discussion For polarization imaging, we have established a depth learning network to realize image compression sensing super-resolution reconstruction while compensating polarization aberration. In the reconstruction process, our network establishes the mapping relationship from the low-resolution intensity images with polarization transmission angles to the high-resolution polarization information images, which simplifies the process of first reconstructing the high-resolution intensity images from the original low-resolution images and then calculating the polarization information images by the formula, and also effectively avoids the error superposition brought by multiple super-resolution reconstructions. Compared with traditional methods, the deep compressed sensing algorithm we use not only has a better reconstruction effect and faster reconstruction speed but also has the function of correcting polarization aberration without additional formula and calculation to obtain polarization information images. The addition of the DMD enables the encoding process in compressed sensing imaging and makes it possible to reconstruct high-resolution images with low sampling rate images. The original polarization intensity image will be affected by polarization aberration caused by DMD, and the images with different polarization transmission angles will be overlapped on the detector. In the experiment, it is found that by using this overlapping, we can reconstruct the polarization information image by a single image with one polarization transmission angle, which is different from the traditional method of reconstructing a polarization information image by the images of four polarization transmission angles. Therefore, the polarization information image with a resolution of 512 × 512 can be reconstructed from a 256 × 256 low-resolution image of a single polarization transmission angle, which means that the sampling rate is 0.0625 (1/16). At present, the resolution of our method is limited by the DMD resolution. In the visible light band, our imaging method does not significantly improve the resolution. However, in the infrared polarization detection dimension, the resolution can be increased to 2-4 times the original image through appropriate training data sets. Because the high-resolution infrared detector is very expensive, our imaging method can significantly reduce the system's cost. We designed a DMD polarization compression imaging optical system using Zemax Although there is still a large gap between the experimental dataset and the real scene, we still reconstructed a good image quality, which indicates that our algorithm has certain robustness and can be improved even more by increasing the training data of real shooting in the future. Considering the improvement of some specific application object recognition results with our method, a nylon false leaf is put in the middle of the real leaves, and the polarization information image of it is obtained by our method. It can be seen from Figure 14 that in the AoP image, the grain of the false leaf is obviously different from that of real leaves. From the visual angle, it can be seen that polarization imaging is helpful for the recognition of artificial targets in the natural background. In the future, we will further expand the range of the experiment and give a quantitative analysis of the improvement of different target recognition probabilities. Discussion For polarization imaging, we have established a depth learning network to realize image compression sensing super-resolution reconstruction while compensating polarization aberration. In the reconstruction process, our network establishes the mapping relationship from the low-resolution intensity images with polarization transmission angles to the high-resolution polarization information images, which simplifies the process of first reconstructing the high-resolution intensity images from the original low-resolution images and then calculating the polarization information images by the formula, and also effectively avoids the error superposition brought by multiple super-resolution reconstructions. Compared with traditional methods, the deep compressed sensing algorithm we use not only has a better reconstruction effect and faster reconstruction speed but also has the function of correcting polarization aberration without additional formula and calculation to obtain polarization information images. The addition of the DMD enables the encoding process in compressed sensing imaging and makes it possible to reconstruct high-resolution images with low sampling rate images. The original polarization intensity image will be affected by polarization aberration caused by DMD, and the images with different polarization transmission angles will be overlapped on the detector. In the experiment, it is found that by using this overlapping, we can reconstruct the polarization information image by a single image with one polarization transmission angle, which is different from the traditional method of reconstructing a polarization information image by the images of four polarization transmission angles. Therefore, the polarization information image with a resolution of 512 × 512 can be reconstructed from a 256 × 256 low-resolution image of a single polarization transmission angle, which means that the sampling rate is 0.0625 (1/16). At present, the resolution of our method is limited by the DMD resolution. In the visible light band, our imaging method does not significantly improve the resolution. However, in the infrared polarization detection dimension, the resolution can be increased to 2-4 times the original image through appropriate training data sets. Because the highresolution infrared detector is very expensive, our imaging method can significantly reduce the system's cost. We designed a DMD polarization compression imaging optical system using Zemax optical design software and analyzed the influence of the imaging system on the aberration of polarization imaging. The local 3D eigen-vibration coordinate system is rotated and transformed into the 3D global coordinate system, and the influence of the DMD reflection interface on the polarization state is analyzed to obtain a simulated image with polarization aberration of the optical system, which can eliminate the influence of the instrument polarization aberration on the imaging and improve the imaging contrast. The training data is added with simulated polarization aberration, compressed and sampled to obtain low-resolution images. When reconstructing the image after real DMD reflection imaging with the network trained from the existing dataset, the reconstruction effect has been better than other traditional reconstruction algorithms, but it still needs to be improved. Therefore, for future work, we look forward to using DMD devices to capture large amounts of data for specific needs in order to create more accurate databases for training. Block-compressed imaging based on deep compressive sensing can achieve fast compression imaging and improve imaging resolution. Polarization imaging can improve detection contrast and suppress flare on the water surface. We can combine the block compressed imaging based on deep compressive sensing and polarization imaging to achieve high-resolution fast polarization imaging. It can be used for sea surface target recognition and tracking. Conclusions In this paper, we propose a polarization imaging method based on DCS by adding a DMD to the optical system to achieve compressed sensing super-resolution imaging. It takes the encoded intensity image of each polarization transmission angle as input and learns the nonlinear mapping between the encoded image and the polarization characteristics directly. The influence of the optical system on the polarization aberration is also analyzed, and the polarization aberration is corrected by network learning while improving the resolution. The network is compared with existing methods and achieves better results in terms of lower sampling rate, quantitative metrics, and visual effects. In addition, the polarization imaging experiments carried out in the paper verify the potential application of the polarization DCS imaging method in the field of high-resolution polarization imaging.	9,500.4	2022-12-01T00:00:00.000	[ "Engineering", "Physics" ]
Mimansa Principle of Interpretation : Numerous scriptures have been found that have been crucial to understanding the Hindu texts. These texts included the complex procedures for determining the true meaning of terms and expressions found in the Vedas and Puranic texts. The Mimansa is the most significant scripture among all others that contained the guidelines for this kind of interpretation. Hindu civilization and culture developed complex norms of interpretation even in their earliest days. Smrities were interpreted according to the guidelines provided by "Jaimini," the author of the Mimamsat Sutras, which were first intended for srutis. One could refer to Mimansa as the Dharmasasthras' stepping stone. This article deals with the various axioms and scientific nature of interpretation, how it differs from maxwell’s interpretation and how it applied in current scenario. Introduction: Jaimini established the Mimansa Rules of Interpretation, which are our customary guidelines for interpretation.Shabar, Kumarila Bhatta, Prabhakar, and others expounded the Sutras of Jaimini.Our great jurists, such as Vijnaneshwara (author of Mitakshara), Jimutvahana (author of Dayabhaga), Nanda Pandit, and others, would frequently refer to these Mimansa Principles whenever they discovered a contradiction between the numerous Smritis or any ambiguity, incongruity, or casus omissus within.The objective of Mimamsa is to provide guidelines for interpreting the Vedas, which are the oldest texts in Hinduism, as well as a rationale for the philosophical significance of observing Vedic rites.The Mimansa principles were originally developed for religious purposes, but because they were quite reasonable and logical, they later found application in other fields like as law, grammar, logic, and philosophy.Among the six Indian philosophical systems (darshans), the Mimansa system is essential to Vedanta and had a major effect on the creation of Hindu law. Axioms of interpretation: For the interpretation of shastras, six axioms of interpretation have been developed.These are 1.The Sarthakyata axiom, according to which each word and sentence needs to have a meaning.2. The Gauravah doshah, or Laghava axiom, which asserts that the construction that shortens and simplifies the meaning is preferred.3. The Arthaikatva axiom, which asserts that a word or sentence occurring in the same place should not have two meanings.A Vakyabheda is a flaw (dosh) that has this double meaning.4. The Gunapradhan Axiom, which says that a word or sentence that seems to express a subordinate notion should either be changed to reflect the major idea or ignored altogether if it conflicts with it.The saying, "bigger fish eats smaller fish" (matsyanyaya27) can be used as an analogy to demonstrate this idea.To shed insight into this, consider the fact that Yajurveda verses are usually said quietly and Samaveda verses are typically recited loudly.The Gunapradhan axiom was employed to determine that recitals must be made in a softer voice because they are required to be recited as part of the Yajurveda rituals.This helped to settle conflicts in some Yajurveda ceremonies, such as Agnyadhana (Primary), which involves the recitation of the lines of Samaveda (Accessory).This is so that the Accessory can fulfil its responsibility of ensuring that the Primary's goal is achieved, as it was created with the Primary in view.5.The Samanjasya Axiom, which says that every effort should be made to reconcile writings that seem to be at odds with one another.This idea has been used by Jimutvahana to resolve discrepancies between Manu and Yajnavalkya's texts about the succession rights.The Nashtasvadagdha Ratha maxim, which was utilised to resolve the discrepancies between the Manu and Yajnavalkya Smriti texts about self-acquired property and ancestral property, serves as one example of the Samanjasya axiom.It is based on a story in which two men set out on a voyage in separate horse-drawn chariots, and when a fire burst out, one man lost his horse while the other's chariot was destroyed by fire.So, utilising the last horse and chariot, they both completed their voyage together for mutual benefit.i As a result, it is well-established that contradictory provisions should, whenever feasible, be interpreted to complement one another since the court has an obligation to prevent "head-on clashes among the provisions of the statute."6.The Vikalpa axiom, which says that the rule more in accordance with fairness and usage should be accepted at one's discretion if there is a genuine and irreconcilable disagreement between two legal norms of equal force.Consequently, when a regulation is a higher legal standard as according to the Badha principles, one takes precedence over the other when comparing, for example, a Shruti and Smriti. Characteristics of mimansa school: Mimamsa schools are characterized by the following: • Emphasis is placed on the interpretation of Vedic texts such as the Samhita and Brahmana; • They contend that the Vedas contain the ultimate truth and are the source of all knowledge; • While performing rituals may help one attain paradise, understanding the rationale behind Vedic ceremonies is equally necessary; • One must comprehend this rationale in order to perform the rites properly and earn atonement; • A person's actions determine their strengths and weaknesses; • If their good deeds persisted, they would enjoy the pleasures of heaven; • However, they will be immune to the eternal cycle of life and able to break free from the never-ending cycle after they have atoned.• Purva Mimamsa is a Karma-Mimamsa system that studies Vedic teachings through Karma-Kanda ceremonies. • The Mimamsa school emphasises the necessity of performing a Yagya in order to receive material and spiritual advantages.• As a result, the philosophical foundation of the Vedas is provided by the Samhita (and Brahmana) sections.• This worldview placed a strong emphasis on the Vedic ritual aspect, which is performing Vedic procedures to achieve salvation. • The Brahmanas employed this strategy to maintain their authority over the populace and to maintain control over the social system. Scientific nature of Mimamsa principle: The division of concepts into categories and subcategories for simple understanding demonstrates the scientific and systematic character of the Mimansa principle.The Vakya principle, for instance, used the subcategories Adhayaahra and Anusanga to fill in words and expressions that were missing, and Upakarasha and Apakarsh to move clauses within sentences so that they could be understood clearly. Notably, there are contemporary interpretation guidelines that like Maxwell's, which allow for violence in certain scenarios, just as the statute does.According to the Supreme Court of India, "courts can sometimes supply words which have been accidentally omitted" in S.S. Kalra v. Union of India.ii In Tribhuwan Misra v. D.I.O.S the Saamanjasy concept of interpretation was applied in to reconcile two opposing division bench findings.This was done on the basis of the aphorism "lost horse and burned chariot" (Nasarhatasva Dagdhartha Nyaya). In Mahabir Prasad Dwivedi v. State the Anusanga principle of interpretation was applied in-depth to make the statute more democratic and equitable, something that could not have been accomplished with Western principles. In Vinay Khare v. State of U.P. the Allahbad High Court resolved the dispute over candidate selection by using the Laghava principle of interpretation and concentrating on the written exam rather than the inperson interview to reduce the possibility of bias, favouritism, and arbitrariness.The candidates received equal marks overall. Mimansa vs maxwell: There are two different legal interpretation theories: Mimamsa and Maxwell.While Mimamsa principles are a scientific system of interpretation that was developed in India from very early times, Maxwell's principles of interpretation are primarily used in Western law courts.Interpreting the law so that it can be successfully applied to a specific scenario before him is one of a judge's main responsibilities.The foundation of Maxwell's rules of interpretation is the notion that a statute's language should be interpreted normally and that the legislature's aim should be inferred from the words used. On the other side, the Mimamsa principles are more thorough and systematic.While Maxwell's concepts are limited to the interpretation of statutory law, they can also be applied to the interpretation of judgements.iii The Mimamsa principles are superior to Maxwell's principles of interpretation in two ways: 1. they are more comprehensive and methodical, and 2. they can be applied to the interpretation of judgements as well as statutes, whereas Maxwell's principles are limited to statutory law interpretation.Adhyahara is the term for casus omissus in Mimamsa.The adhyahara concept allows us to amend a legal document.Nonetheless, Maxwell's lack of further explanation and reference of the subcategories falling under the broad category of casus omissus illustrates the superiority of the Mimamsa principles over his concepts in this specific field.The usefulness of Mimamsa principles is not diminished by the fact that they occasionally produce diverse outcomes.Different outcomes are also produced using Maxwell's concepts.This merely serves to highlight the need for care when applying interpretation principles.Interpretation principles make good servants but bad masters.Just because something is foreign doesn't mean it has to be rejected.Westerners have a lot to teach us that is beneficial. Use of mimansa in current legal system: The meaning of the legal provisions has been investigated through the use of the Mimangsa Rules of Interpretation.After citing a "Shloka," the Supreme Court took one of these principles into practise.In the case of UP Bhoodan Yagna Samiti, UP V. Braj Kishore, the Supreme Court of India made the following observation: "In this country, we have a heritage of rich literature, it is interesting to note that literature of interpretation also is very well known."Many Shlokas that have been recognised for hundreds of years have articulated the fundamentals of interpretation.In Beni Prasad v. Hardai Bibi, Sir John Edge, the Chief Justice of the Allahabad High Court at the time, made reference to the Mimamsa concept.Similar to this, the Gunapradhan Axiom of the Mimamsa principle had been applied in Amit Plastic Industry, Ghaziabad v. Divisional Level Committee, Meerut to interpret section 419 of the UP Sales Tax Act.In the cases of M/s Ispat Industries Ltd vs. Commissioner of Customs and M/s Craft Interiors Pvt.Ltd vs. Commissioner of Central Excise, the Supreme Court recognised the significance of the Mimansa Rules of Interpretation. Conclusion: In summary, the Mimansa Principles provide a customary framework for interpreting legal texts, especially when it comes to Hindu law.The Mimansa Principles were initially developed for the purpose of understanding religious texts, but they are now seen as sufficiently reasonable and scientific to be used to the interpretation of contemporary laws and rulings.With a flexibility and reason lack from Western principles of interpretation, the Mimansa Principles offer a distinctive method for statutory interpretation.They are characterised as scientific and rational, with the goal of improving the democracy, equity, and reason of the law.The Indian legal system has recognised and applied the Mimansa Principles, proving their applicability in contemporary statute interpretation.When used by judges, they can be an effective tool for reshaping the law to make it more democratic, fair, and logical.The Mimansa Principles have a historical basis in religious practises, but they have progressively found application in other domains like as philosophy and law, demonstrating their flexibility and relevance in modern legal interpretation.In general, the Mimansa Principles give an alternative perspective to Western legal principles by providing distinctive and historically grounded method of interpreting legal texts within the framework of Hindu law.	2,552.2	2023-11-25T00:00:00.000	[ "History", "Philosophy" ]
Inhaled Placental Mesenchymal Stromal Cell Secretome from Two- and Three-Dimensional Cell Cultures Promotes Survival and Regeneration in Acute Lung Injury Model in Mice Acute lung injury (ALI) and acute respiratory distress syndrome (ARDS) is a common clinical problem, leading to significant morbidity and mortality, and no effective pharmacotherapy exists. The problem of ARDS causing mortality became more apparent during the COVID-19 pandemic. Biotherapeutic products containing multipotent mesenchymal stromal cell (MMSC) secretome may provide a new therapeutic paradigm for human healthcare due to their immunomodulating and regenerative abilities. The content and regenerative capacity of the secretome depends on cell origin and type of cultivation (two- or three-dimensional (2D/3D)). In this study, we investigated the proteomic profile of the secretome from 2D- and 3D-cultured placental MMSC and lung fibroblasts (LFBs) and the effect of inhalation of freeze-dried secretome on survival, lung inflammation, lung tissue regeneration, fibrin deposition in a lethal ALI model in mice. We found that three inhaled administrations of freeze-dried secretome from 2D- and 3D-cultured placental MMSC and LFB protected mice from death, restored the histological structure of damaged lungs, and decreased fibrin deposition. At the same time, 3D MMSC secretome exhibited a more pronounced trend in lung recovery than 2D MMSC and LFB-derived secretome in some measures. Taking together, these studies show that inhalation of cell secretome may also be considered as a potential therapy for the management of ARDS in patients suffering from severe pneumonia, including severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2), however, their effectiveness requires further investigation. Introduction Acute lung injury (ALI) is a common condition characterized by upregulation of inflammatory mediators. Dyspnea, severe hypoxemia, and pulmonary edema follow ALI, producing significant mortality rates [1]. Acute respiratory distress syndrome (ARDS), the clinical manifestation of ALI, is a major cause of acute respiratory failure, with a 33-48% mortality rate in critically ill patients [2]. ARDS causes extensive surfactant dysfunction and depletion [3], causing pulmonary structural unit collapse, limited gas exchange, and severely reduced blood oxygen levels, leading to fatal hypoxia [2]. Despite extensive ARDS research, mortality in adult patients remains high [4]. and severely reduced blood oxygen levels, leading to fatal hypoxia [2]. Despite extensive ARDS research, mortality in adult patients remains high [4]. Pandemic respiratory viruses lead to ARDS development more frequently than seasonal viruses [5]. A recent worldwide outbreak of pneumonia caused by severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) led to a high percent of ARDS, or even death, among adult patients [6]. However, there are still no approved medicines for ARDS, thus, the development of effective treatment strategies or agents is highly desired. Among these, multipotent mesenchymal stromal cell (MMSC) therapies are considered as a promising approach for the treatment of ARDS in different respiratory viral infection models. MMSCs offer new approaches for numerous pathologies [7], including those related to lung diseases [6,[8][9][10]. MMSCs possess anti-inflammatory, immunomodulatory, regenerative, proangiogenic, antifibrotic, and antimicrobial properties, and clinical trials are currently underway [2,7,11]. The therapeutic effects of MMSCs are mainly mediated by paracrine activity, which involves the release of bioactive substances collectively known as the secretome [7,11]. The secretome contains extracellular vesicles (EVs), growth factors, cytokines, and chemokines, and is considered a promising cell-free therapeutic agent. Due to clinical data indicating their safety, MMSCs are the preferred source of the secretome and derived EVs [11][12][13][14][15][16][17]. MMSC-derived secretome can be sterilized by filtration and produced as an offthe-shelf product, while MMSCs themselves cannot. Moreover, MMSC-derived secretome are free from the safety issues associated with cell-based therapy [6,7,18,19]. Studies of the secretome and derived exosomes as treatment for lung diseases have mainly used systemic administration via intravenous injection or direct intratracheal administration [11]. However, inhalation, enabled by the small size of proteins and EVs, is a less invasive administration route. Furthermore, due to the large pulmonary surface area (approximately 35-100 m 2 ), direct pulmonary delivery provides rapid and prolonged effects [17]. The regenerative capacity of the secretome depends on many factors. For instance, transitioning from monolayer culture to nonadhesive, three-dimensional (3D) culture can dramatically change the secretome contents and therapeutic potential [20,21]. Therefore, we aimed to develop lyophilized biomimetic formulations containing secretome from two-dimensional (2D) and three-dimensional (3D) cultures of placental MMSCs and to assess their therapeutic potential against ALI in vivo. We also used the secretome from 2D-and 3D-cultured lung fibroblasts (LFBs) as an active comparator, which effectively attenuates bleomycin-induced lung fibrosis [22]. The flow chart of the study is shown in Figure 1. 2D and 3D Cell Cultures Placenta-derived MMSCs that were previously isolated and characterized were used in this study [23]. The MMSCs were isolated from term human placenta obtained after Caesarian section. The enzymatic approach allowed for rapid cell isolation-they readily adhered to culture plastic on the next day after tissue digestion and reached 100% confluent monolayer within several days. At passage 3, they exhibited high proliferative activity, as numerous doubling cells were observed in culture, and were characterized by a spindle-shaped morphology ( Figure 2A). Flow cytometry of passage 3 cells revealed negative endothelial and blood cell marker expression (CD14, CD31, CD45, CD34, CD11b, and CD19 < 1%) ( Figure 2E-G) and high expression of surface markers characteristic of MMSCs population (CD44, CD29, CD90, CD105, and CD73 > 85%) ( Figure 2H-L). After confirming the MMSC phenotype of the culture, cells were used for the obtainment of conditioned media. Protein Composition of Cell Secretome To understand the molecular mechanism mediating the regenerative and protective properties of our secretome-based formulations, we estimated proteomic composition by liquid chromatography-tandem mass spectrometry (LS/MS/MS). In secretomes from 2D and 3D MMSC cultures, 281 and 286 proteins were identified, respectively. Only 234 proteins were shared by the 2D-and 3D-cultured MMSC secretomes ( Figure 3A). Of these shared proteins, 23.4% and 43% were annotated as cytoplasmic proteins for 2D-and 3D-cultured MMSC secretomes, respectively ( Figure 4A,B). Furthermore, 23.1% and 9.5% of the shared proteins were annotated as "extracellular region" or "secreted" for 2D-and 3D-cultured MMSC secretomes, respectively ( Figure 4A,B). Annotation of the 47 proteins exclusive to the 2D-cultured MMSC secretome indicated their participation in 22 pathways, including growth factor signaling pathways, the endothelium signaling pathway, the hypoxia response, immune cell activation, angiogenesis, and cytoskeletal regulation ( Figure 3C). Annotation of the 52 proteins exclusive to the 3D-cultured MMSC secretome indicated their involvement in only 13 pathways, including FGF, TGF-β, PDGF, the IFN-γ signaling pathway, angiogenesis, and cytoskeletal regulation ( Figure 3D). Molecular functions for proteins exclusive for 3D-cultured MMSC secretome were more diverse compared with 2D MMSC culture ( Figure 4E,F). The biological process analysis results for MMSC secretome proteins were similar ( Figure 5A,B). 2D-and 3D-cultured LFB secretomes contained 319 and 306 proteins, respectively, and shared 236 proteins ( Figure 3B). Of the shared proteins, 40% and 24.3% were annotated as cytoplasmic proteins for 2D-and 3D-cultured LFB secretomes, respectively, and 22% and 11.4% were annotated as "extracellular region" or "secreted" for 2D and 3D, respectively ( Figure 4C,D). These observations align with data from a previous study of the lung spheroid cell secretome [22]. Annotation of the 83 proteins exclusive to the 2D-cultured LFB secretome indicated their participation in 33 pathways, including growth factor signaling pathways, the morphogenic Wnt signaling pathway, the endothelium signaling pathway, the hypoxia response, immune cell activation, inflammation, blood coagulation, angiogenesis, and cytoskeletal regulation ( Figure 3E). The 70 proteins exclusive to the 3D-cultured LFB secretome were involved in only nine pathways, including Wnt, Notch, and TGF-β signaling ( Figure 3F). Molecular functions and biological processes for proteins exclusive to 3D-cultured LFB secretome were close to secretome proteins from 2D LFB culture ( Figures 4G,H and 5C,D). ALI Model The survival rates of all six groups were monitored daily. The highest mortality rate (40%) was observed in the nontreated group with ALI (ALI + NaCl). All treatment groups and the control intact group exhibited 0% mortality throughout the experiment ( Figure 6A). merous doubling cells were observed in culture, and were characterized by a spindleshaped morphology (Figure 2A). Flow cytometry of passage 3 cells revealed negative endothelial and blood cell marker expression (CD14, CD31, CD45, CD34, CD11b, and CD19 < 1%) ( Figure 2E-G) and high expression of surface markers characteristic of MMSCs population (CD44, CD29, CD90, CD105, and CD73 > 85%) ( Figure 2H-L). After confirming the MMSC phenotype of the culture, cells were used for the obtainment of conditioned media. ALI Model The survival rates of all six groups were monitored daily. The highest mortality rate (40%) was observed in the nontreated group with ALI (ALI + NaCl). All treatment groups and the control intact group exhibited 0% mortality throughout the experiment ( Figure 6A). On day 8, the lung weight in the ALI control group (ALI + NaCl) and in groups treated with formulation from LFB-2D culture was significantly larger than in the control group. This change was likely a characteristic of the stage of ALI development. The lung weights were restored almost to their initial values in the groups that received formulations from LFB-3D culture and MMSC cultures (MMSC-2D, MMSC-3D) ( Figure 6B). Histological Examination NaCl solution (0.9%) did not affect the lung structure, which featured rounded, wellspread alveoli with thin walls and clear lumens ( Figure 6C, blue arrows). The alveolar epithelium was characterized by distinct nuclei and oxyphilic cytoplasm. Bronchioles and On day 8, the lung weight in the ALI control group (ALI + NaCl) and in groups treated with formulation from LFB-2D culture was significantly larger than in the control group. This change was likely a characteristic of the stage of ALI development. The lung weights were restored almost to their initial values in the groups that received formulations from LFB-3D culture and MMSC cultures (MMSC-2D, MMSC-3D) ( Figure 6B). Histological Examination NaCl solution (0.9%) did not affect the lung structure, which featured rounded, wellspread alveoli with thin walls and clear lumens ( Figure 6C, blue arrows). The alveolar epithelium was characterized by distinct nuclei and oxyphilic cytoplasm. Bronchioles and bronchi were lined with cubic and cylindrical epithelium, and lumens were clear ( Figure 6C, green arrow). ALI development on day 8 was accompanied by moderate edema of the loose connective tissue surrounding the bronchi and accompanying vessels. Diffuse infiltration of connective tissue around bronchi with inflammatory elements ( Figure 6D, blue arrows) and areas of desquamation of the bronchial mucosa prismatic epithelium were observed ( Figure 6D, green arrow). Mucus production by goblet cells was enhanced ( Figure 6D, green arrow), and the lung experienced incomplete collapse, producing alveoli lumens with irregular shapes and sizes ( Figure 6E, blue arrow). Interalveolar septa were thickened due to edema of the alveolar septa stroma ( Figure 6E, green arrow). Part of the alveoli contained exudate with an admixture of erythrocytes, fibrin, and desquamated alveolar epithelium ( Figure 6D, orange arrow). We also observed a plethora of blood clots in some vessels of the vascular bed ( Figure 6E, orange arrow). Exudate score and interstitial edema and hyaline membrane formation score are shown in Figure 6J,K. 2D-LFB-Sec inhalation resulted in no peribronchial edema; bronchi and small bronchioles had open, clear lumens. However, most alveoli exhibited incomplete expansion ( Figure 6F, blue arrows) and partial wall thickening ( Figure 6F, green arrows). The exudative score was significantly lower compared to the control ALI group ( Figure 6J). The interstitial edema and hyaline membrane formation score did not differ from the ALI control group ( Figure 6K). 3D-LFB-Sec also significantly reduced ALI severity, as observed on day 8. There was no interstitial edema around the bronchi and associated vessels; bronchi, small bronchioles, and some of the alveoli featured an open, clear lumen ( Figure 6G, blue arrows). However, some areas also exhibited reduced lumen size and uneven alveoli contours ( Figure 6G, green arrows). The exudative score and the interstitial edema and hyaline membrane formation score were significantly lower compared to the control ALI group ( Figure 6J,K). 2D-MSC-Sec treatment relieved ALI on day 8 after LPS administration. An absence of peribronchial edema was observed, and the bronchi and small bronchioles had open, clear lumens ( Figure 6H, blue arrow). The majority of the alveoli were inflated with clear lumens and thin walls ( Figure 6H, green arrows); only isolated, small areas had unevenly thickened alveolar walls ( Figure 6H, orange arrow). The exudative score and the interstitial edema and hyaline membrane formation score were significantly lower compared to the control ALI group and were lower compared to groups treated with secretome from LFB cultures (and Figure 6J,K). 3D-MSC-Sec inhalation resulted in no interstitial edema around bronchi and associated vessels, and bronchi and small bronchioles featured clear, open lumens ( Figure 6I, blue arrows). The alveoli were mainly straightened with clear lumens and thin walls ( Figure 6I, green arrows). Only small areas had unevenly thickened alveolar walls and incompletely expanded alveoli ( Figure 6I, orange arrow). The assessment of exudation and edema has shown that 3D-MSC-Sec treatment is associated with the lowest values of scoring compared to the control ALI group (adjusted p < 0.0001) (and Figure 6J,K). Additional representative histological areas are presented in the Supplementary Materials ( Figure S1). Pulmonary Coagulation and Fibrinogen/Fibrin Deposition Immunohistochemical staining of lung tissue sections revealed fibrinogen deposits in all studied groups, including healthy control animals ( Figure 7A). The percentage of fibrinogen-positive regions in the control group lung tissue was 6.66%. The highest percentage of fibrinogen expression (28.62%), with pronounced edema of the connective tissue and thickening of the interalveolar septa, was observed in the nontreated ALI control group. Repeated inhalation of all investigated formulations resulted in 7-11% fibrin/fibrinogen expression by day 8. This expression was significantly lower than that observed in the control ALI group, but significantly higher than that in the control group ( Figure 7B). No differences were observed between the MMSC-2D and MMSC-3D groups. Discussion The current consensus is that the MSC secretome is their main mode of action. Preclinical studies of lung diseases have demonstrated that the MMSC secretome can effectively treat chronic obstructive pulmonary disease, ARDS, idiopathic pulmonary fibrosis, Discussion The current consensus is that the MSC secretome is their main mode of action. Preclinical studies of lung diseases have demonstrated that the MMSC secretome can effectively treat chronic obstructive pulmonary disease, ARDS, idiopathic pulmonary fibrosis, and severe pneumonia [14,22,24]. In most studies, MMSCs derived from bone marrow, Wharton's jelly of the umbilical cord, and adipose tissue have been used as the source of the secretome or EVs. Placenta is also an available postpartum source of MMSCs that involves no ethical controversies regarding its use for secretome production in clinical practice [25]. Profiles of proteomes secreted by MMSCs from different sources and cultured under the same conditions share similar functional features, however, proteomes of postpartum MMSCs from the umbilical cord and placenta have been predicted to have greater therapeutic potential [26]. Although soluble factors secreted by viable MMSCs into the culture medium offer an attractive cell-free therapeutic alternative, to our knowledge, their regenerative potential has not yet been fully investigated. Therapeutic properties of secretome/exosomes were reported to be modulated by factors related to their culture condition such as cell source, oxygen tension (normoxia vs. hypoxia), growth factor composition, and physical microenvironment [27]. Most studies produced conditioned media (CM) in monolayer culture, but several studies used different 3D cultures, in particular, cell spheroids. Spheroid cultures need a spatial handling but yield more cells compared to conventional monolayer cultures, and thus more secreted factors [28]. Three-dimensional cell culture enables better recapitulation of the in vivo environment and was proven to be beneficial for MMSC culture in terms of enhancing their therapeutic potential [27]. In addition, cells located at the center of the spheroid may be in a relative hypoxic condition compared to cells on the surface, which leads to increase of certain pro-regenerative paracrine signaling molecules and immunomodulatory factors [28,29]. The therapeutic effects of MMSC from 3D culture were also translated to their secretome [28,30]. We have developed an original cell spheroid culturing protocol in our lab. To provide an appropriate nursing ecosystem for MMSC, the fetal calf serum (FCS) and basic fibroblast growth factor (bFGF) were added to the basic growth medium. We have not used a serum-free culture medium and serum starvation because they may cause changes in characteristics of cells and the production yield and properties of exosomes. Specifically, serum deprivation may induce cell death (in MMSC) [31][32][33]. It was shown that when serum concentrations are reduced from 10% to 1%, the size distribution, total quantity, and protein composition of in vitro-derived EVs were different [34]. Depletion of fetal serum from EVs may also negatively effects on secretome content. It is reported that in response to the various depletion process, EV-depleted fetal serum may influence parent cell phenotype and possibly their qualitative and quantitative production of EVs [31,35]. To minimize the influence of fetal serum EVs and growth factors, we conditioned cell cultures over 72 h. Our internal lab data showed that cells actively consume bFGF from the full growth medium, decreasing their concentration from 20 ng/mL to 334 pg/mL after 72 h for 3D MMSC cultures (unpublished data) and 331 pg/mL for 2D cultures [23]. These data align with other studies showing that cells actively uptake soluble serum components and EVs for their growth and survival [36]. Therefore, we chose not to use unconditioned growth medium as a control, but compare the therapeutic potential of CM from 3D-cultured MMSC obtained during spheroid formation and compactization stage and in parallel from monolayer MMSC culture to CM from 2D-and 3D-cultured LFB as a control. The culture conditions were equivalent for all cells in our study. The administration route may influence the biodistribution of secretome products and their final therapeutic effect. Monsel et al. showed that microvesicles (MVs) released by human bone marrow-derived MSCs reduces the severity of endotoxin-induced ALI in mice by transferring keratinocyte growth factor (KGF) mRNA to the injured alveolus. However, the MVs were administered intravenously in this study [24]. Another study showed that inhalation of lung spheroid cell secretome and exosomes promotes lung repair in pulmonary fibrosis more effectively than inhalation of 2D-cultured MSC secretome and exosomes. Interestingly, recent studies have indicated that exosomes reproduce only part of the regenerative potency of the full secretome from which they are isolated [22]. Here, we demonstrated that inhalation of lyophilized secretome-based formulations improved survival in a lethal endotoxin-induced ALI mouse model. We compared formulations containing secretome derived from either 2D and 3D placental MMSC cultures or 2D and 3D LFB cultures, in terms of their regenerative capacity. ALI significantly damaged the lung structure. Inhalation of CM from 2D and 3D LFB cultures resulted in partial lung structure restoration; peribronchial edema was not observed in either group, and the bronchi and small bronchioles had clear, open lumens. However, there were signs of incomplete alveoli expansion and slight wall thickening in both LFB groups. The interstitial edema and hyaline membrane formation score were significantly higher in LFB-2D group. Administration of 2D and 3D MMSC formulations relieved ALI on day 8 after LPS inhalation. Experimental groups treated with MMSC secretome-based formulations exhibited almost full restoration of the lung architecture to a healthy level after three inhalations. Only small areas with unevenly thickened alveolar walls and incompletely straightened alveoli were noted. These findings align with those of previous studies in rats that investigated the effects of systemically administered MSCs and MSC exosomes [24,37]. These studies have shown that secretome-based products can confer a level of protection and regeneration similar or superior to that conferred by the cells themselves. Secretome from stem cells contains signaling molecules and recyclable materials for survival, proliferation, and differentiation of recipient cells. It was shown that MMSC secretome provides several stimulatory and inhibitory bioactive factors at variable concentrations that might sustain physiological kinetics in the local microenvironment [38]. Potential mechanisms underlying the beneficial effects of MMSCs include the following: an increase in human KGF protein levels in injured alveoli that promotes alveolar epithelial cell repair; an immunomodulatory effect on monocytes and alveolar microphages that suppresses cytokine-induced lung injury and lung protein permeability; enhanced alveolar epithelial type 2 cell metabolism via delivery of key metabolic enzymes (glyceraldehyde 3-phosphatase dehydrogenase, pyruvate kinase); a transfer of mitochondria, increasing the bacterial clearance; restoration of vascular endothelial permeability via release of VEGF and HGF; and protection of endothelial cells against apoptosis [2,24,39]. However, aggregation and 3D spatial organization of MMSCs into multicellular spheroids reportedly promotes their anti-inflammatory properties through increased anti-inflammatory factor secretion [20,21,[40][41][42]. Our data have also indicated that 3D-MSC-Sec inhalation has a more pronounced restorative trend in damaged lung structure compared to other formulations, according to the exudative score and the interstitial edema and hyaline membrane formation score. At the same time, no significant differences were found between groups treated with the secretomes from 2D MMSC, 3D MMSC, and 3D LFB cultures. These findings can be partially explained by differences in the composition of the secretome obtained from 3D cell cultures, which potentially may be more effective in restoring disturbed tissue homeostasis. Differences in the contents of exosome RNA cargo in the secretome may also explain differences in treatment capacity [22]. The complexity of cell secretome content indicates that more studies are needed to fully understand the mechanisms of their biotherapeutic activity. Our study also assessed the effect of secretome inhalation on fibrin deposition in the lungs. Nearly all patients with ALI demonstrate abnormalities in alveolar fibrin turnover. These abnormalities vary from subtle changes in molecular coagulation and fibrinolysis markers to more evident fibrin deposition in smaller airways. Fibrin has a key role in host defense, and fibrin depositions activate neutrophils and fibroblasts and decrease alveolar fluid clearance, thereby inactivating surfactant and favoring alveolar collapse, increasing pulmonary dead space, and causing additional endothelial injury [43]. In our study, all secretome-treated groups showed significantly less fibrin deposition than the ALI control group on day 8. However, these results need further clarification, as they may be related to residual heparin content in CM after 72 h of conditioning. This work had several limitations. First, the results of the mouse ALI models may not be easily translated into studies of complex and heterogeneous human populations. Our models also relied on a single, high-dose administration of LPS to healthy animals; however, ALI is usually associated with pre-existing risk factors and develops over a longer period. The lung injury improvement was evaluated solely by histological examinations in this study. It would be better to demonstrate lung injury improvement by measuring cytokines in lung tissue and bronchoalveolar lavage fluid or by immunohistochemical assessment of other markers (recovery of alveolar epithelial cells I and II, the activity of antioxidant enzymes, pulmonary surfactant recovery). The frequency of agent administration may have been another limitation. We used only one dose of cell secretome, an arbitrary choice guided by previous experience. More frequent or continuous administration may be more effective. Future studies will include a dose-response assessment to determine the optimal dosage regimen. In our study, treatment was delivered via inhalation using a nebulizer. An investigation into different administration routes to determine the optimal delivery method and frequency of administration would help maximize benefits. The last important limitation is the presence of residual proteins and EVs from FCS in the studied CM, which could potentially distort and even reduce the therapeutic efficacy of the MMSC secretome [44]. However, we used identical cell culture conditions, so we assume that the contribution of FCS-derived EVs was the same for all cell cultures. Prolonged conditioning leads to most of the EVs in the secretome originating from MMSCs. This enrichment of the CM with EVs derived from MMSCs should potentiate its therapeutic potential. This is consistent with recent studies showing that the regenerative potential of EVs from fetal serum is significantly lower than vesicles from MMSCs [45]. Interestingly, EVs collected from the MMSC cultured in fetal serum-containing medium had greater wound healing and angiogenic effects for endothelial cells compared with FBS-derived EVs [44]. Based on the literature, the MMSC secretome examination may benefit from the use of chemically defined, serum-free, and xeno-free medium that is not only optimized for cell growth and viability for a variety of cell types, but is also free of exogenous contaminating fetal serum-derived EVs and extracellular protein/RNA species [31]. Lastly, we presented here the comparison of the proteomic composition of MMSC and LFB secretomes from 2D and 3D cultures to confirm their differences. It is obvious that the secretome composition may influence biotherapeutic properties and potential for application. In addition to proteins mRNA, miRNA, and lipid content may play a significant role. We acknowledge that it would be beneficial to determine which proteins and RNA are the important functional components determining the therapeutic effects of the secretome. Future experimental directions may include MSC preconditioning [46], cell bioengineering to modulate the secretome profile [47,48], and developing subsequent purification and standardization methods for scaling up secretome production [11]. In summary, we demonstrated that inhaled formulations based on lyophilized CM derived from 2D and 3D placental MMSC and LFB cultures improved survival in mice with LPS-induced ALI. Cell-free secretome-based formulations from 2D and 3D cell cultures were effective and safe in the lethal ALI model, suggesting a possible noninvasive alternative to stem cell-based regenerative therapy. Lyophilization preserves proteins and vesicles and prolongs storage without special requirements, which means that secretome-based products can be stored as an off-the-shelf powder [49]. Further experimental and clinical studies are needed to elucidate the fundamental mechanisms that mediate the protective effects of MMSC secretome from 2D and 3D cell cultures in the settings of ALI. Chemicals and Reagents The following reagents were purchased for the study. Animals Sixty 8-week-old male C57BL/6 mice were purchased from the Federal State Budgetary Institution of Science's Scientific Center for Biomedical Technologies of the Federal Medical and Biological Agency (Andreevka, Russia). Animals were individually housed in stainless steel cages in an air-conditioned room (22 ± 1 • C, 55 ± 5% humidity) with a 12 h/12 h light/dark cycle and ad libitum access to food and water. All experiments were conducted with an effort to minimize suffering and the number of animals used. 2D MMSC and LFB Cell Culturing and Secretome Collection Placenta-derived MMSCs that were previously isolated and characterized were used in this study [23]. Placental tissue transported to the laboratory was first washed in Hank's solution containing 100 U/mL gentamycin for 12 h and was then mechanically cut into fragments of no more than 3 mm. Individual cells were isolated from the placental tissue pieces by incubation for 30 min in 0.15% collagenase type II solution (Sigma, Germany) at 37 • C with constant stirring. Hank's solution was added to the isolated cells to reduce the collagenase activity. The resulting suspension was filtered through a 100 µm pore filter (Becton Dickinson, San Jose, CA, USA) and centrifuged at 300× g for 10 min. The cell pellet was resuspended in full growth medium and plated in Petri dishes at the density of 3 × 10 5 cells/cm 2 under standard conditions (37 • C in 5% CO 2 ). The full growth medium consisted of basal DMEM-F12 with 2 mM glutamine supplemented with 40 U/mL gentamicin, 1% 100× ITS-G, 20 ng/mL bFGF, 15 U/mL heparin, and 10% FCS. The medium was changed 2-3 times per week, and visual inspection of the culture was performed under a Primovert phase-contrast microscope (Zeiss, Jena, Germany). The cell cultures were passaged at a 1:3 ratio upon reaching 80-85% confluence using Versene and 0.25% trypsin solution. For the LFB cells, the culture conditions were equivalent. At passage three, the placental MMSCs and LFBs were cultured for 72 h without changing the medium. After 72 h, 500 µL of the resulting CM was collected individually from every dish for further content analysis. The remainder of the CM was pooled together for each cell culture in 50 mL tubes (10 mL of CM from 2 × 10 6 cells), filtered through 0.22 µm filter to remove any cells and cell debris. The filtered secretome was frozen and stored at −80 • C. Cells at passage 3 were used for characterization and 3D culturing. 3D MMSC and LFB Cell Culturing and Secretome Collection Spheroids were obtained and cultured in nonadhesive multi-well plates prepared from a 2% agarose solution (A-6013, Sigma-Aldrich, Germany) on DMEM/F12 basal medium supplemented with 100 U/mL gentamicin using silicone 3D PetriDish ® molds (Microtissues™, USA). Each agarose plate was placed in the well of a 12-well culture plate (Corning-Costar, Cambridge, MA, USA). Placental MMSCs and LFBs at the third passage were treated with Versene and 0.25% trypsin solutions (BioLoT). The cell suspension was transferred to 15 mL tubes and centrifuged (7 min, 400× g). The resulting pellet was resuspended in the complete growth medium to a concentration of 3.3 × 10 6 cells/mL. Next, 150 µL of that suspension was transferred to nonadherent agarose plates. After an hour, 2 mL of complete growth or induction medium was added to the wells. Spheroids from both placental MMSCs and LFB were cultured for 72 h without changing the media. After 72 h, 500 µL of the resulting CM were collected separately from every agarose plate for further content analysis. The remainder of the CM was pooled together for each cell culture in 50 mL tubes, filtered through 0.22 µm filter to remove any cells and cell debris. The filtered secretome was frozen and stored at −80 • C. Every microplate containing 256 spheroids or 0.5 × 10 6 cells allowed us to collect 2.5 mL of CM, which was equivalent to the proportions described for monolayer cultures. Sample Preparation One hundred microlitres of each sample was transferred into clean tubes and 6 µL of 85% phosphoric acid (up to 5% final concentration; Sigma, Germany) was added and mixed. Then, 800 µL of methanol (J.T.Baker) was added and the samples were mixed again. The resulting suspension was centrifuged at 10,000× g at 15 • C for 10 min (Centrifuge 5424R, Eppendorf, Germany). The precipitate was reconstituted in 50 µL of a denaturing solution consisting of 5 M urea (Sigma, Germany), 1% sodium deoxycholate (Sigma, Italy), 300 mM sodium chloride (Fluka-Honeywell), 10% acetonitrile (Carlo Erbo), 100 mM triethylammonium bicarbonate (pH 8.2-8.5) (Sigma, Switzerland), and up to 10 mM freshly added neutralized TCEP (Sigma, St. Louis, MO, USA). The reconstituted denatured protein was incubated at 45 • C for 30 min with constant vigorous stirring at 1200 rpm (Thermo Mixer, Eppendorf, Germany). Then, 6 µL of 2% stabilized 4-vinylpyridine (Aldrich, UK) in 30% isopropanol (Fisher Chemical) was added to a final concentration of 0.2%. The alkylation reaction was incubated in the dark at 20 • C for 20 min. The sample volume was then increased to 500 µL by adding 384 µL of 75 mM triethylammonium bicarbonate (pH 8.2) and thoroughly mixed. To each sample, 400 ng of trypsin was added from a 100 ng/µL stock (Promega) in 30 mM acetic acid (Carlo Erba) and incubated for 3 h at 40 • C with intermittent stirring (stirring at 1700 rpm for 90 s every 10 min). Then, an additional aliquot of 400 ng trypsin (100 ng/µL) in 30 mM acetic acid was added and the reaction was incubated for 2 h at 42 • C with intermittent stirring as above. At the end of the incubation, 10 µL of absolute formic acid was added to precipitate the reduced deoxycholic acid. The resulting suspension was centrifuged at 12,000× g and 5 • C for 10 min. To remove the residual deoxycholic acid, an equal volume of ethyl acetate (Carlo Erba) was added to 500 µL of the supernatant and vigorously stirred for 3 min at room temperature. Then, the mixture was centrifuged at 10,000× g for 5 min at 20 • C and then incubated at −20 • C for 10-15 min. The samples were removed from the freezer, the surface organic layer was decanted, and 150 µL of acetonitrile (Carlo Erba) was added to the lower aqueous layer containing the peptides. The mixture was centrifuged at 13,000× g for 10 min at 20 • C, and the supernatant was collected and dried under vacuum at 30 • C for 60-70 min with chamber ventilation every 15 min (Concentrator Plus, Eppendorf, Germany). The resulting dry residue was reconstituted in 20 µL of 0.5% formic acid (Sigma, Germany). High-Performance Liquid Chromatography-Mass Spectrometry (HPLC-MS) The analysis was performed using a Xevo™ G2-XS QToF quadrupole time-of-flight mass spectrometer (Waters, UK) coupled with an Acquity™ HPLC H Class Plus chromatography system (Waters). The analysis was carried out in positive electrospray ionization mode with increased sensitivity and a normal dynamic range of measurement. The emitter voltage was 3 kV, the drying gas rate was 680 L/min, the focusing gas rate was 50 L/min, the temperature of the ionization source was 150 • C, and the temperature of the desolator was 350 • C. The voltage across the focusing cone was 67 V with a bias of up to 130 V. The ions were recorded in hybrid data-independent acquisition (DIA) MSe-SONAR mode. Specifically, an initial DIA MS scan was performed from 100-1500 m/z, followed by a SONAR scan with mass isolation with a quadrupole from 400-1100 m/z and an isolation peak width of 22 Da. The time for one complete scan cycle was set at 0.418 s. Fragmentation was carried out in two-phase mode: phase 1-low-energy collision-induced dissociation (CID) fragmentation with argon at 6 eV; phase 2-high-energy ranked CID fragmentation with argon from 15-40 eV. During the analysis, active mass correction (m/z = 556.27) with a low activation energy (9 eV) was performed using a leucine enkephalin standard (50 pg/mL in 50% acetonitrile with 0.1% formic acid). The standard was injected into the ionization source for 30 ms at 5 µL/min every 45 s and isolation within 200 mDa. Chromatographic separation was performed on an Acquity™ UPLC BEHC18 column (1.7 µm particle size, 2.1 × 50 mm column size; Waters) at a flow rate of 0.2-0.3 mL/min and constant temperature of 40 • C. Gradient elution was used, consisting of mobile phase A (aqueous solution of 0.1% formic acid and 0.03% acetic acid) and mobile phase B (solution of 0.1% formic acid and 0.03% acetic acid in acetonitrile) with the following elution scheme: 0-1.5 min, 3% B; 1.5-26.5 min, increased to 19% B; 26.5-42 min, increased to 32% B; 42-43.5 min, increased to 97% B; held in isocratic mode until 47.5 min; decreased to 3% B until 49 min; and held in isocratic mode until 53 min. The flow rate from 43.5-47.5 min was 0.3 mL/min; at all other times, the flow rate was 0.2 mL/min. Data Analysis Raw data files were handled using the PLGS (Protein Lynx Global Server, version 3.0.3, Waters, the UK) search engine. Protein search was performed against human amino acid sequences database obtained from the UniProt KB (release May 2021) as a FASTA file with automatically generated reversed concatenated decoy sequences to estimate a false positive rate. Precursor ion mass tolerance was set at 20 ppm (±10 ppm tolerance window) and fragment ion mass tolerance was set to 8 mDa (±4 mDa tolerance window). S-pyridilethylation was included as a fixed modification, whereas oxidized methionine (oxM), deamidated glutamine(dQ), and asparagine (dN) were variable modifications. The minimal peptide length was fixed to eight amino acids, with only one allowed internal missed cleavage. The false discovery rate (FDR) at 1% was determined for peptide and protein identification by accumulating the reverse database hits. The listed secreted proteins differing in the samples were classified using PANTHER (Protein Analysis THrough Evolutionary Relationships, htpp://pantherdb.org) (21 June 2021) to study molecular functions, cellular components, and pathways. For subcellular classification, the protein subcellular localization predictor mGOASVM (human) (21 June 2021) was used [52]. Preparation of Lyophilized Secretome Based Formulations First, 50 mL of collected secretome was mixed with cryoprotective agent (maltose), passed through filter with a pore size of 0.22 µm, aliquoted (10 mL of secretome per bottle), and frozen at −80 • C for 3 h. Then samples were lyophilized using a Martin Christ Alpha 1-2 LDplus machine (Martin Christ, Osterode am Harz, Germany) for 40 h. At the end of sublimation, the samples were sealed under a vacuum and rolled up. ALI Model A lethal ALI model in mice was induced in 50 male C57BL/6 mice by inhalation of 10 mg/kg LPS. Another 10 mice were used as intact control. Experimental groups were the following: 1. The experimental animals were subjected to inhalation of the secretome formulations (2D-LFB-Sec, 3D-LFB-Sec, 2D-MSC-Sec, and 3D-MSC-Sec) dissolved in 0.9% NaCl solution (1 mL of reconstituted formulation per mouse) at 30 min, 24 h, and 48 h after LPS administration. Ten control animals not exposed to LPS were subjected to inhalation of 0.9% NaCl. For both LPS and drug inhalation, a dynamic method of exposure was used when the concentration of the substance was maintained at a relatively constant level, ensuring the necessary air exchange. To ensure this, we used an original inhalation (exposure) system that allows us to accommodate up to 10 small laboratory animals simultaneously. The system is a sealed exposure chamber equipped with a nebulizer with an aerosol sensor, inlet, and discharge pipes with special degassing traps. The mortality of animals in experimental groups was monitored daily. After the 8 days of treatments, the survived animals were euthanized by placing them in a CO 2 chamber. The animals were then sent for dissection and pathomorphological examination. Histology On the 8th day of the study, the surviving animals from all groups were sacrificed, followed by lungs necropsy for macroscopic description, determination of lung weight (g) and histological assessment of lung tissue. Lungs were fixed in 10% neutral formalin, dehydrated in alcohols of ascending concentrations, and embedded in paraffin. Paraffin sections 5 µm thick were obtained using an SM 2000R microtome (Leica, Germany), stained with hematoxylin and eosin (H&E), and examined using a DM1000 microscope (Leica, Germany). Analysis of Fibrin/Fibrinogen Deposition Quantitative analysis of fibrin/fibrinogen expression in lungs was carried out using immunohistochemical staining of tissue sections with polyclonal rabbit antibodies against fibrinogen (Abcam, ab64238). The percentage ratio of fibrinogen-positive areas to the total area of the lung tissue was determined by the following equation: % fibrinogen = S fibrinogen / S lungs * 100% where % fibrinogen was the percentage of fibrinogen expression regions, S fibrinogen is the area of the fibrinogen-positive regions in the image, S lungs is the area of the lung tissue in the image. For immunohistochemical staining, 5 µm paraffin sections of lung tissue were deparaffinized using xylene, 96 and 70% alcohol and boiled for 20 min in a universal antigen retrieval reagent (Abcam, ab208572) in a microwave oven. A hydrogen peroxide blocking solution (Abcam, ab64218) was then applied to the entire tissue section surface for 10 min. Protein blocking solution (Abcam, ab64226) was used for 5 min at room temperature to reduce nonspecific background staining. Sections were stained with primary rabbit polyclonal antibodies against fibrinogen (Abcam, ab34269) in a solution of 1% BSA in PBS overnight at 4 • C. Rabbit-specific IHC polymer detection kit HRP/DAB (Abcam, ab209101), consisting of amplifying and detecting solutions, was used for signal amplification. Tissue sections were then stained with a DAB chromogen (Abcam, ab64238) for 2 min, and the cell nuclei were stained with hematoxylin (Abcam, ab220365) for 5 min at room temperature. After every step, the sections were washed with TBS solution (Abcam, ab64204) 3-4 times. The preparations were finally mounted under cover glasses using mounting medium for IHC (Abcam, ab64230). Light images were registered at Axio Scope A1 microscope (Zeiss, Germany) using 20× dry objective and ZEN 3.0 software (Zeiss, Germany). Quantitative analysis of fibrinogen expression in the obtained images was performed using the ImageJ Fiji software (version 1.2; WS Rasband, National Institute of Health, Bethesda, MD) based on the previously described protocol [53]. Briefly, images were split into separate channels for hematoxylin and DAB using the "Color deconvolution" command. The intensity limit was determined for the DAB channel to cut off the background noise without affecting the positive signal. Then, the percentage of positive areas in the image converted into pixels ("Area fraction") was measured. The tissue ROI was selected on the initial image using the "Color Threshold" function, with the further calculation of its area in pixels. Statistical Analysis Results are expressed as the mean and standard deviation (SD) or standard error of the mean (SEM). The data were checked for normal distribution using the Anderson-Darling test. Nonparametric tests were used to compare differences among small samples (n ≤ 10), the Mann-Whitney U test was used to compare differences between two independent samples (control vs. experimental), and the Kruskal-Wallis test with Dunn's post hoc analysis was used to compare differences among more than two samples. The reliability of immunohistochemical data was assessed using Brown-Forsythe and Welch one-way analysis of variance (ANOVA) and the Games-Howell test (for groups with n > 50). Differences between groups were considered statistically significant at p < 0.05. All statistical analyses were performed using GraphPad Prism software version 9.2.0 for Windows (La Jolla, CA, USA). In the figure legends, n refers to the number of samples or mice.	9,587.2	2022-03-22T00:00:00.000	[ "Biology", "Medicine" ]
Preparation and NMR spectra of four isomeric diformyl[2.2]paracyclophanes (cyclophanes 66) Four isomeric dialdehydes 4, readily available from cycloaddition of propiolic aldehyde (2) to 1,2,4,5-hexatetraene (1), were separated by chromatography and recrystallization, and were characterized by their spectroscopic data. The individual isomers can now be easily identified from their 1H NMR spectra even if only one of them is present. Introduction Previously, we reported that the [2 + 4] cycloaddition of 1,2,4,5-hexatetraene (1) to propiolic aldehyde (2) produced a mixture of four [2.2]paracyclophane dialdehydes 4. This result is in agreement with the generation of the p-xylylene intermediate 3 in the first step of the reaction, which can dimerize by four different modes (Scheme 1) [2]. Depending on the amounts and the quality of the solutions containing the tetraene 1 and the dienophile 2, more than 60 g of the dialdehyde mixture 4 can be obtained in a single run, corresponding to maximum yields of 46%. Both starting compounds are very reactive and can decompose, even under refrigeration. Although we have separated the four isomers and used them many times for the preparation of numerous [2.2]paracyclophane derivatives (inter alia annelated derivatives [2], metal complexes [3][4][5], diethynyl derivatives [6,7] and preparation of ligands for chiral reagents [8,9]), we have neither described the separation of these versatile starting materials nor explained their spectroscopic and analytical data in full. With this short communication, we would finally like to remedy this omission, in particular since these compounds are also beginning to attract the attention of other research groups [6,7,[10][11][12][13]. Results and Discussion Separation of the diformyl[2.2]paracyclophanes 4 We have separated the isomer mixture 4 by different methods. The easiest way is by middle pressure liquid chromatography (MPLC), which readily affords all isomers in gram amounts. To obtain larger amounts of specific isomers, e.g. the pseudo-para compound, it might be advisable to use the combination of column chromatography and MPLC or recrystallization described in the experimental section. In alcoholic solvents, pseudo-para-4 is the least soluble of all isomers and may hence be easily separated. NMR spectra of the diformyl[2.2]paracyclophanes 4 As the 1 H and 13 C NMR spectra of [2.2]paracyclophane-4carbaldehyde have previously been fully assigned and, hence, the influence of the substituent upon the 1 H and 13 C NMR chemical shifts of all positions of [2.2]paracyclophane are known [14], the assignment of the dicarbaldehyde isomers could, in principle, be derived from the comparison of their experimental chemical shifts with those calculated by assuming substituent chemical shift (SCS) additivity. However, there is no certainty that such an assumption is justified, as it is difficult to judge whether steric and/or electronic substituent interactions could cause significant deviations from SCS additivity. Furthermore, the SCSs exerted by the formyl group upon the nuclei of the distant ring are relatively small, \|Δδ H \| < 0.11 ppm, \|Δδ C \| < 0.88 ppm. We therefore decided to seek independent experimental proof for the four isomeric structures. For this purpose, consideration of their symmetry is helpful. The ps-gem and the ps-meta isomer both have a symmetry element (plane and two-fold axis, respectively), which causes each CH 2 CH 2 bridge to have only two different 1 H chemical shifts. The protons of the bridge close to the formyl substituents form an AA´XX´ spin system, see Scheme 2. These protons are labelled H a or H s depending on whether they are anti or syn with respect to the 4-CHO group. The most characteristic and most easily recognized feature of an AA´XX´ spectrum is the distance N, defined as \|J(AX)+J(AX´)\|. In the ps-gem isomer, N equals \|J gem +J trans \| with J gem ≈ -13 Hz and J trans ≈ +4 Hz [14] . Hence, N is expected to be near 9 Hz. In the ps-meta isomer, N equals \|J gem +J cis \| with J gem as above and J cis ≈ +10 Hz, hence N ≈ 3 Hz. Spectra a and b in Figure 1 have N = 9.3 and 3.5 Hz, respectively, and therefore correspond to the ps-gem and the ps-meta isomer in this order. The ps-ortho and the ps-para isomer have a 2-fold axis and a centre of symmetry, respectively, which render the two bridges in each compound equivalent. So each of these isomers gives rise to a single AKRX spin system for their bridges, see spectra c and d in Figure 1. The distinction of the ps-ortho from the ps-para isomer is made possible with the help of an NOE experiment. In the isomer belonging to the spectrum shown in Figure 1c, the multiplet at δ = 3.11 ppm experiences a strong NOE when the resonance of Scheme 2: Spin systems of the CH 2 CH 2 protons in the isomeric dialdehydes 4. Protons at C-9 and C-10 in ps-gem-and ps-meta-4 form YY´ZZ´ spin systems, which need not be considered here. Protons at C-9 and C-10 in ps-ortho-and ps-para-4 form AKRX spin systems equivalent to those of the C-1 and C-2 protons. 5-H (ortho to the 4-CHO group) is saturated, and vice versa. Its coupling constants show that the 3.11 ppm multiplet belongs to 9-H s (cis to the deshielded proton 10-H s , which itself is syn to the second formyl group, see Scheme 3). This proves the ps-ortho arrangement of the two aldehyde groups. Then, by default, the spectrum in Figure 1d is that of ps-para-4. The 1 H and 13 C NMR data of the isomers are given in Table 1 and Table 2. The question of SCS additivity posed above can now be answered as follows. In the 1 H NMR spectra, additivity is satisfied with, in general, \|Σ SCS obs -Σ SCS calc \| ≤ 0.06 ppm, except for 2-H a and 2-H s in ps-meta-4 where one finds 0.11 and 0.17 ppm, respectively. In the 13 C NMR spectra, SCS additivity is less strict, namely \|Σ SCS obs -Σ SCS calc \| ≤ 0.7 ppm, except for C-2, C-4 and C-5 in ps-gem-4 with 1.2, 1.1 and 0.9 ppm, respectively. We also find that SCS additivity was distinctly better in the previously studied analogous bis(dimethoxycarbonyl) [2.2]paracyclophanes [15] compared to the present dialdehydes. This may be attributed to the larger magnitudes of the SCS of the aldehyde compared to those of the methyl ester functional group. Conclusion Although the present method of preparing the dialdehydes 4 involves extensive separation work (chromatography and recrystallization), it offers certain advantages compared to alter-native (more specific) routes [10][11][12][13]. It is much shorter than many of the alternatives, which involve the rather expensive starting material, [2.2]paracyclophane. This method also provides four different substitution patterns in one step, two of which are interesting chiral derivatives (pseudo-ortho-and pseudo-meta-4). This leads directly to derivatives with a preparatively very useful functional group. The four individual isomers can now be easily identified, even if only one isomer is present, because each one shows a characteristic multiplet for the proton 2-H s at δ H = 4.1-4.2 ppm: i.e., ps-gem-4 and ps-meta-4 give an AA´XX´ pattern with N = 9.3 and 3.5 Hz, respectively; to a good approximation, ps-ortho-4 has a dd pattern with J = 13 and 10 Hz; and ps-para-4 displays a ddd pattern with the two inner lines coinciding (rel. intensities 1:1:1:2:1:1:1, J = 13, 10.5 and 2.5 Hz). Another characteristic is the small chemical shift difference δ 5-H −δ 7-H of 0.08 ppm for ps-ortho-4, whereas this difference is 0.24-0.41 ppm for all other isomers. Experimental Separation of the isomers 4 After the cycloaddition has been carried out [3], the precipitate formed was removed by filtration. The mother liquor was concentrated to half its volume and the solution stored in the refrigerator overnight. The newly formed crop was filtered off and united with the first precipitate. The combined solid reaction products were treated with chloroform in a Soxhlet extractor to remove any polymeric material produced. The solvent was removed and the residual solid could be directly separated by MPLC chromatography (silica gel) with dichloromethane/ethyl acetate (98:2, v/v) to provide the analytically pure isomers 4 (product ratio: pseudo-gem-: pseudoortho-: pseudo-meta-: pseudo-para-4: 1.8 : 1.1 : 1.0 : 2.8); (R f = 0.47, 0.55, 0.60, 0.63). To reduce the separation effort, the above product mixture can be enriched first by standard silica gel chromatography with dichloromethane/ethyl acetate (98:2, v/v). This procedure provided three fractions: a fast moving one consisting of pseudo-paraand pseudo-meta-4, a middle fraction containing all four isomers, and a third fraction consisting of the pseudo-orthoand the pseudo-geminal isomers. If necessary, this process can be repeated, thus allowing separation of the four dialdehydes into two fractions each containing two isomers. These may then be subjected to MPLC chromatography or repeated normal pressure column chromatography, allowing, for example, the separation of pure pseudo-ortho-4. Alternatively, the pseudo-meta/para-mixture could be purified by recrystallization from methanol, in which the pseudo-para isomer is only poorly soluble. On refluxing this mixture in methanol, followed by separation of the precipitate, practically pure pseudopara compound was obtained. To remove the last traces of isomeric impurities from pseudo-para-4, it was recrystallized from n-butanol. Spectroscopic properties The NMR spectra were measured at Table 2.	2,121	2010-09-29T00:00:00.000	[ "Chemistry" ]
A Neighbor-Based Probabilistic Broadcast Protocol for Data Dissemination in Mobile IoT Networks The recent trend of implementing Internet of Things (IoT) applications is to transmit sensing data to a powerful data center and try to discover the valuable knowledge behind ‘‘Big Data’’ by various intelligent but resource-consuming algorithms. However, from the discussion with some industrial companies, it is understood that disseminating real-time sensing data to their nearby network-edge applications directly would produce a more economical design and lower service latency for some important smart city applications. Therefore, this paper proposes an efﬁcient broadcast protocol to disseminate data in mobile IoT networks. The proposed protocol exploits the neighbor knowledge of mobile nodes to determine a rebroadcast delay that prioritizes different packet broadcasts according to their proﬁts. An adaptive connectivity factor is also introduced to make the proposed protocol adaptive to the node density of different network parts. By combining the neighbor knowledge of nodes and adaptive connectivity factor, a reasonable probability is calculated to determine whether a packet should be rebroadcasted to other nodes or be discarded to prevent redundant packet broadcast. Extensive simulation results have validated that this protocol can improve the success ratio of packet delivery by 13% ∼ 28% with a similar end-to-end transmission delay and network overhead of the most state-of-art approaches. I. INTRODUCTION Japan is one of the advanced countries in the world and it is facing to various representative social issues in healthcare, nursing-care, disaster prevention, energy saving and so on.The concept of smart city that utilizes Internet-of-Things (IoT) technologies to strengthen social infrastructures opens a new door for innovative solutions to the aforementioned issues and also creates a big commercial market.By observing this trend, the National Institute of Information and Communications Technology (NICT) of Japan is cooperating with some industrial companies to deploy a large-scale wireless IoT platform in the Sumida ward of Tokyo [1].In this IoT platform, hundreds of different social ''Things'' like vending machines, beverage delivery trucks, and taxis are equipped with Wi-SUN [2] communication units that have a transmission range of several hundred meters in the real field.Thus, these wireless-enabled ''Things'' comprise large scale mobile wireless networks that cover a service region of 10 km 2 , i.e., more than 70% area of the Sumida ward [1].This kind of networks consists of social ''Things'' is named mobile IoT The recent trend of implementing IoT systems is to gather sensing data greedily, and transmit them to a powerful data center for analysis [3].Since the amount of such data is typically too huge to be perceived, acquired, managed, and processed by traditional software/hardware tools within a tolerable time, they are named ''Big Data'' [4].Various intelligent but resource-consuming approaches have been proposed by researchers to discover the valuable knowledge behind this kind of ''Big Data'' [5]- [8].However, from the discussion with our industrial partners, it is understood that many kinds of sensing data are valuable only to human acting (living) in a local area and only when they are fresh.Figure 2 indicates an example scenario of alerting rushing children to nearby vehicle drivers.Assume that some sensing device is carried by a child (e.g., a sensor embedded in school-bag) and it transmits an alert signal when detecting the child is rushing in street.By disseminating this real-time signal in the local mobile IoT networks, nearby drivers can be notified to slow down their vehicles and avoid traffic accident.Thus, instead of simply uploading data to a data center, disseminating realtime sensing data to their nearby edge applications in local IoT networks directly would produce a more economical design and lower service latency for some important smart city applications. Broadcasting is a suitable strategy to meet the previous demand.In a naive protocol named simple flooding (SF) [9], every wireless node simply rebroadcasts a newly received data packet and this operation can deliver the packet to all nodes in the networks (assuming that there is no separation).Therefore, this strategy is not only easy to be implemented in resource-constrained wireless nodes like vending machines and smart meters, but also adaptive to the dynamic network topology of mobile nodes like vehicles.However, it has been realized that SF protocol suffers an issue of ''broadcast storm'' [10], i.e., the massive redundant broadcasts of packets incur many collisions and interference in wireless channel which lead to excessive packet loss.Many improved protocols have been proposed by researchers to overcome the ''broadcast storm'' issue.Williams and Camp categorized these protocols into three classes: ''probability-based', ''areabased'', and ''neighbor-based'' protocols [9], and Kim et al. further indicated that neighbor-based protocols perform best among them [11]. Since limiting the number of redundant packet broadcast is the key to alleviate the ''broadcast storm'' issue, and the neighbor-based strategy performs better than the areabased and probability-based ones, this paper proposes a neighbor-based probabilistic broadcast (NPB) protocol to facilitate the data dissemination in mobile IoT networks.The main contributions of this paper are: (1) A novel scheme for determining the rebroadcast delay of a packet is proposed.The node which has more neighbors that have not received a data packet (named uncovered neighbors in this paper) triggers a shorter delay to rebroadcast the packet.This delay scheme not only decreases the probability for nodes to rebroadcast redundant packets, but also alleviates the collisions and interference in wireless channel that lead to excessive packet loss.(2) A novel scheme for determining the rebroadcast probability of a packet is also proposed.This scheme considers the statistics of uncovered neighbors and network connectivity to determine the probability for nodes to rebroadcast a packet.The resulting probability is composed of two parts: a.Additional coverage ratio, that is the ratio of a transmitter's neighbors that will be newly covered by one rebroadcast to its total amount of neighbors; b.Adaptive connectivity factor, that considers the local density of nodes in different parts of networks to maintain the connectivity of packet delivery.The rest of this paper is organized as follows: Section II reviews previous works.Section III describes the proposed NPB protocol in detail.Section IV shows the evaluation results of NPB protocol and its comparisons with other stateof-art approaches.Section IV briefly discusses the deployment of protocol.Finally, conclusions are drawn and future work is mentioned in the last section. II. RELATED WORK Broadcasting is an effective strategy to disseminate data, but its overhead can be quite large, especially in highly dynamic mobile networks [12].Tseng et al. studied conventional broadcast protocols analytically and experimentally [10], and showed that SF protocol incurs large network overhead and causes many problems such as redundant broadcasts and channel collisions.Keshavarz-Haddad et al. proved that solving the optimal broadcast strategy that minimizes the number of packet broadcast is equivalent to find out the minimum connected dominating set (MCDS) of network topology [13].However, this optimal solution is little used in real situations, mainly because of two reasons: (1) Solving the MCDS of a given network topology is NP-hard [13] that indicates its complexity increases exponentially with the size of networks; and (2) It is hard to maintain an accurate MCDS of mobile wireless networks since their topologies constantly change according to the movements of network nodes.As a result, many researchers proposed their efficient heuristic algorithms to approximate the optimal broadcast behavior in mobile wireless networks. Haas et al. proposed a gossip-based broadcast protocol [14], in which each node rebroadcasts a newly received packet with a pre-configured probability.Their protocol can save up to 35% overhead compared with SF protocol.However, when the network density is high or the network traffic load is heavy, the improvement of their protocol is quite limited [12].Kim et al. proposed a probabilistic broadcast protocol based on the coverage area and neighbor confirmation of wireless nodes [11].Their scheme uses the coverage area of nodes to set rebroadcast probability, and uses the neighbor confirmation of nodes to guarantee the reachability of data packets.A considerable number of works such as [15]- [17], were proposed under a counter-based strategy.In these protocols, a counter C is initiated to record the total number of the same data packet received by a wireless node.This C is compared to a pre-configured threshold.If C reaches or exceeds the threshold, the probability for the node to rebroadcast such packet is zero, else its value approaches to one.Although these protocols can reduce network overhead, they also increase the end-toend delay of packet delivery significantly.A location-based strategy was suggested in [18] and [19].In these works, data packets are rebroadcasted based on the locations of wireless nodes, mainly by using GPS.As a result, many nodes can be excluded from the rebroadcast process that in turn enhances network performance.In contrast, using GPS not only increases the cost of wireless nodes, but also prevents indoor nodes from adopting these protocols. Two recent effective broadcast protocols that inspire our work are the neighbor coverage-based probabilistic rebroadcast (NCPR) protocol proposed by Zhang et al. [20] and the neighbor-based dynamic connectivity factor routing (DCFR) protocol proposed by Ejmaa et al. [21].NCPR protocol proposes a scheme for calculating the rebroadcast delay of a data packet, and allows a node which has more common neighbors with the previous transmission node to rebroadcast the packet first.After the delay expires, it decides whether a packet should be rebroadcasted or discarded based on the node's uncovered neighbors.DCFR protocol can be treated as a revision of NCPR protocol that eliminates the need of configuring global network parameters and improves the success ratio of packet delivery by 2% ∼ 4% in most situations.Compared with the previous two protocols, our proposed NPB protocol adopts a reverse strategy to determine the rebroadcast delay of a packet.That is, the priority of a node to rebroadcast a packet decreases, when it has more common neighbors with the previous transmission node.As illustrated in Sect.IV, this novel delay scheme and its accompanied scheme for determining the rebroadcast probability of a packet enable NPB protocol to improve the success ratio of packet delivery by 13% ∼ 28%, while retaining a similar end-to-end packet transmission delay and network overhead of NCPR and DCFR protocols.Ruiz and Bouvry provided a comprehensive survey of this research topic [22]. III. NEIGHBOR-BASED PROBABILISTIC BROADCAST PROTOCOL As described in Sect.I, the key to optimize a broadcast protocol is to limit the number of redundant packet broadcast.Figure 3 compares the performances of SF protocol and an optimal broadcast protocol that minimizes redundant packet transmission.By using SF protocol, when the node n 1 broadcasts a packet, its neighbors n 2 , n 3 , n 4 , and n 5 will receive this packet.Since this is a new packet to n 2 , n 3 , n 4 , and n 5 , they will rebroadcast this packet.Similarly, n 6 and n 7 will receive the packet and rebroadcast it again.Therefore, a total of 7 packets are broadcasted by 7 nodes in the network.However, Figure 3(b) illustrates that only two times of packet broadcast from n 1 and n 4 can disseminate this packet to all nodes in the network.It is easy to validate that n 1 and n 4 constitute the MCDS of the network topology shown in Figure 3, and this observation is consistent to the conclusion made in [13].Due to the reasons explained in Sect.II, it is not smart to directly apply this optimal solution to dynamic mobile networks.Therefore, this section proposes a novel NPB protocol to implement efficient data broadcast in the mobile IoT networks discussed in this paper. As scrutinized in the following parts of this section, NPB protocol uses the number of uncovered neighbors to determine the rebroadcast delay of a packet, and uses the additional coverage ratio of a packet and a metric named adaptive connectivity factor to determine its rebroadcast probability.The same to other neighbor-based broadcast protocols, NPB protocol requires a data packet to carry the neighbor list of its transmitter.Since this mechanism is not the main focus of this paper, NPB protocol adopts the same efficient mechanism used by NCPR [20] and DCFR [21] protocols to reduce the overhead of neighbor data transmission.The details of this mechanism are available in [20]. A. REBROADCAST DELAY When a node n i receives a data packet p k (the suffix k indicates the sequence number of packet) from its previous transmitter n p , it can use the neighbor list of n p contained in p k to estimate how many of its neighbors have not been covered by the p k broadcasted by n p .To quantify this, the uncovered neighbor set of n i for p k is given by where N(n i ) and N(n p ) are the neighbor sets of nodes n i and n p .Due to the characteristic of broadcasting a packet, the node n i may receive duplicate p k from its neighbors other than n p .Therefore, n i can continuously update U k (n i ) with the duplicate p k from other neighbors.It is obvious that when U k (n i ) becomes empty, there is no need for n i to rebroadcast p k since all of its neighbors have received p k already.This introduces a tradeoff on the rebroadcast delay of p k .When rebroadcast delay is long, there is a long period for n i to eavesdrop duplicate packets and empty its U k (n i ) to prevent redundant packet rebroadcast, while this also increases the end-to-end delay of data dissemination.Thus, NPB protocol determines the rebroadcast delay of p k as where \| • \| is the size of a set, d max is the upper limit of rebroadcast delay, and a(n i ) is the average number of n i 's neighbors.Due to the mobility of n i , a(n i ) is a dynamic variable and the moving window average strategy described in [23] is used to calculate it: where a t (n i ) indicates the average number of n i 's neighbors at the time point t, and β ∈ [0, 1] is a convergence factor.It can easily be proved that, as time passes, the moving window average strategy ensures the convergence of a(n i ), and this proof is independent of β [23].Indeed, the value of β represents a trade-off, i.e., a larger β reduces the error of a(n i ) but increases the time to reach its steady state, and vice versa. The scheme presented by Eq. ( 2) uses a truncated linear function to determine the rebroadcast delay of p k .When n i has a relatively large number of uncovered neighbors (i.e., \|U k (n i )\| ≥ a(n i )), it will rebroadcast p k without any delay.Otherwise, the rebroadcast delay of p k linearly increases to d max (when \|U k (n i )\| = 1) with the decrease of its uncovered neighbors.This scheme can be understood in the following way: When n i receives p k but \|U k (n i )\| is small, the profit for n i to rebroadcast p k is small since it can only deliver p k to a few additional nodes.From another point of view, since \|U k (n i )\| is small, there is a higher possibility to empty U k (n i ) by letting n i eavesdrop duplicate p k for a longer time.Thus, this scheme assigns a longer delay to n i when U k (n i ) is small, and vice versa.Taking the network topology depicted in Figure 3(b) as an example, and assumes that a(n 4 ) = a(n 5 ) at the time of packet transmission.When n 4 and n 5 receive p k from n 1 simultaneously, n 4 will rebroadcast p k earlier than n 5 , because it has two uncovered neighbors (i.e., n 6 and n 7 ), while n 5 has only one (i.e., n 7 ).After eavesdropping the p k rebroadcasted by n 4 , n 5 realizes that there is no need to broadcast p k again because all of its neighbors have received p k already, i.e., \|U k (n 5 )\| = 0. Finally, it is interesting to note that this scheme for determining the rebroadcast delay of a packet is reverse to the one adopted by NCPR [20] and DCFR [21] protocols.Both NCPR and DCFR protocols let a node which has more common neighbors with the previous transmission node to rebroadcast a data packet first.According to Eq. ( 1), more common neighbors (i.e., a larger \|N (n i ) ∩ N (n p )\|) usually indicates a smaller uncovered neighbor set (i.e., a smaller \|U k (n i )\|).According to Eq. ( 2), a smaller \|U k (n i )\| leads to a longer rebroadcast delay in NPB protocol.As illustrated in Sect.IV, this completely different strategy leads to better performances of data dissemination in different network settings. B. REBROADCAST PROBABILITY A node that has a longer rebroadcast delay may eavesdrop duplicate data packets from other nodes that have shorter ones.For example, when the node n i receives a duplicate p k from its neighbor n j , it knows that how many of its neighbors have been additionally covered by this packet.Thus, n i can update its uncovered neighbor set according to the neighbor list of n j , by using Eq. ( 1).After updating U k (n i ), n i discards this duplicate p k since it has been broadcasted already.It should be noted that NPB protocol does not try to update the rebroadcast delay of p k according to the updated (i.e., smaller) U k (n i ).This is because of two reasons: (1) It prevents the continuous increase of 1-hop rebroadcast delay when n i receives duplicate packet every time; and (2) It prevents the frequent computation of protocol timer and keeps NPB protocol simple and stupid, i.e., the general principle of industrial development.When the timer of rebroadcast delay expires, n i obtains the final U k (n i ) that contains its neighbors which have not received p k yet.This final U k (n i ) is used to determine the probability for n i to rebroadcast p k . Let the additional coverage ratio of n i to rebroadcast p k denote the ratio of n i 's neighbors that will newly receive p k to the total number of n i 's neighbors, With the increase of C k (n i ), more nodes will benefit from n i s rebroadcast.Thus, the probability for n i to rebroadcast p k should be increased accordingly.Besides, Xue and Kumar have derived that when every node connects to more than 5.1774 * log(N node ) of its neighbors, the probability of a network being connected is approaching one with the increase of N node [24], where N node denotes the total number of nodes in the network.This means, in order to disseminate p k to all nodes in the network, at least 5.1774 * log(N node ) neighbors of n i need to receive and rebroadcast this packet, i.e., \|N (n i )− U k (n i )\| > 5.1774 * log(N node ).Thus, a metric named adaptive connectivity factor is defined as The value of N node used in this equation can either be configured by a system administrator in advance as Zhang et al. suggested in NCPR protocol [20], or be estimated by the distributed algorithm proposed by Ejmaa et al. in DCFR protocol [21].From Eq. ( 5), it can be observed that when the number of n i 's neighbors that have received p k is larger than 5.1774 * log(N node ), A k (n i ) is less than one.This means that n i moves in the dense area of network, and p k can be disseminated to the other parts of network without the rebroadcast from n i .Otherwise, it means that n i moves in the sparse area of network, and it prefers rebroadcasting p k to maintain the connectivity of packet dissemination. The probability for n i to rebroadcast p k is given by the multiplication of its additional coverage ratio and adaptive connectivity factor, In Eq. ( 6), C k (n i ) indicates how many next-hop nodes will benefit from the rebroadcast of p k from n i , but it does not consider the connectivity of packet dissemination.A k (n i ) compensates this drawback by increasing the rebroadcast probability of n i when it moves in the sparse area of network, while decreasing this probability when it moves in the dense area of network.As illustrated by the evaluation results in Sect.IV, the combination of these two factors makes NPB protocol adaptive to different network settings.It should be noted that although the value of P k (n i ) in Eq. ( 6) may become greater than 1 in some cases, this will not affect the behavior of NPB protocol.It only indicates that the local node density of network around n i is so low that n i should rebroadcast this packet to maintain the connectivity of packet dissemination. the overall computational complexity of NPB protocol is linear to the of nodes in the networks, i.e., O(N node ).The pseudo-codes of NPB protocol are listed in Algorithm 1. IV. PERFORMANCE EVALUATIONS This section presents the evaluation results of the proposed NPB protocol.In the following evaluations, NPB protocol was compared with NCPR [20], DCFR [21], and SF [9] protocols by using the ns-3 network simulator.NCPR and DCFR are two state-of-art broadcast protocols introduced in recent literatures, while SF is a conventional one and can be treated as the benchmark for other protocols. Mobile nodes were uniformly distributed in a rectangular area at the beginning of simulations.The distributed coordination function (DCF) of IEEE 802.11 standard was used as MAC layer protocol, and a radio channel model of 1Mbps was used to verify the performances of different protocols in resource-constrained IoT environment.The transmission range of every node was 250 meters, and they moved at a speed of 1 m/s according to the random waypoint mobility model [25].The convergence factor β in Eq. (3) for calculating the average number of neighboring nodes was set to 0.8.According to the preliminary simulation results, the performance of NPB protocol is not sensitive to β when it is greater than 0.5.Thus, the impact of β is not further discussed in the remaining parts of this section.The max delay (d max ) used by NPB, NCPR and DCFR protocols to determine their rebroadcast delay was 128 ms, which is equivalent to the upper limit of the random back-off time in many IEEE 802.11 implementations.In every trial of simulation, each mobile node generated 200 data packets with a default interval of 1 second and disseminated them to other nodes by using different broadcast protocols.Table 1 summarizes the default parameters of simulations, and the following results are an average of 100 experimental trials. The previous mentioned four broadcast protocols were compared based on three criteria: Average packet delivery ratio: the average ratio of the data packets that are successfully received by nodes to the total amount of data packets that are generated in simulation. Average end-to-end delay: the average delay of successfully received data packets from the time they are originally generated in simulation. Network overhead: the average amount of data transmitted by nodes in simulation.To preserve the fairness of comparison, the total amount of data transmission is measured instead of the number of transmitted data packets.This is because neighbor-based broadcast protocols like NPB, NCPR, and DCFR require a data packet to carry the neighbor information of its transmitter.4 presents the performances of different broadcast protocols with varied node density.When the number of nodes in the area was small, there was little collision and interference in wireless channel.SF protocol achieved the best packet delivery ratio since it broadcasted more redundant packets than other improved protocols, and these redundant packets improved the possibility of successful packet delivery.However, with the increase of node density, many redundant packets produced by SF protocol led to severe channel collisions and interference that not only degraded the success ratio of packet delivery but also increased the endto-end delay of packet delivery drastically.Compared with the state-of-art NCPR and DCFR protocols, NPB protocol improved the success ratio of packet delivery by 13% ∼ 25%, and retained a similar level of network overhead.It proves that NPB protocol is more effective on managing different packet broadcasts, e.g., it prioritizes a broadcast that can deliver a packet to more uncovered neighbors, and it increases broadcast probability when a node moves in the sparse area of networks.Figure 4(b) shows that NPB protocol resulted in a shorter end-to-end delay than NCPR and DCFR protocols when the node density of network was low, and a little longer delay when the node density of network was high.Recall that both NCPR and DCFR protocols allow a node which has more common neighbors with the previous transmission node to rebroadcast a data packet shortly.With the increase of node density, the common neighbors between receiving and transmission nodes increase, and their schemes tend to reduce the delay of rebroadcasting a packet.As described in Sect.III, NPB protocol adopts a reverse scheme against NCPR and DCFR protocols, and it tends to increase the endto-end delay of packet delivery when node density increases.Nevertheless, since NPB protocol can adaptively decrease the broadcast probability of nodes in dense networks, the end-toend delays of these three protocols were at the same level in any setting of node density. Figure 5 presents the performances of different broadcast protocols with varied packet size.The performance of SF protocol degraded drastically with the increase of packet size.This was because nodes need a longer time to broadcast a bigger packet, and many redundant packets produced by SF protocol severely congested wireless channel.NPB protocol achieved the best packet delivery ratio among four protocols.Compared with NCPR and DCFR protocols, it improved the packet delivery ratio by 16% ∼ 22%, with a very little higher and steady network overhead.It should be noted that the two metrics, i.e., packet delivery ratio and network overhead, must be jointly considered.Purely focusing on reducing network overhead is meaningless.For example, it is easy to achieve a very low network overhead if every node never rebroadcasts any data packet, while this strategy leads to an unacceptable packet delivery ratio.Similar to the simulation results of varied node density, there was no obvious difference among the end-to-end delays of these three protocols.It is clear that this simulation on increasing packet size is equivalent to the simulation on decreasing the link bandwidth of mobile nodes, since both of them increase the time to broadcast a packet. Figure 6 presents the performances of different protocols with varied packet transmission interval.With the increase of packet transmission interval, there were less data packets need to be processed in a time unit and this also alleviated the collisions and interference in wireless channel.Similar to the previous discussions under a low node density, SF protocol achieved a better packet delivery ratio than other improved protocols under the situation of a long interval, since its redundant packet broadcasts improved the possibility of successful packet delivery.The packet delivery ratio of NPB protocol was near to that of SF protocol.Compared with NCPR and DCFR protocols, NPB protocol improved the success ratio of packet delivery by 17% ∼ 28%, and retained a similar level of end-to-end delay and network overhead. Figure 7 presents the performances of different protocols with varied speed of mobile nodes.When the speed of mobile nodes increases, the neighboring node list maintained by NPB, DCFR, and NCPR protocols changes more frequently and becomes less accurate.Since these three protocols make their rebroadcast decisions based on the neighbor list of packet transmitter, their performances degraded with the increase of mobility speed.Nevertheless, NPB protocol achieved the best performance among all protocols even when nodes moved at a speed of 10 m/s, i.e., a usual upper limit of vehicle speed on urban roads.Compared with NCPR and DCFR protocols, it improved the success ratio of packet delivery by 18% ∼ 27% under a similar level of end-to-end delay and network overhead. V. DISCUSSIONS ON PROTOCOL DEPLOYMENT Figure 8 shows an overview of our under constructed IoT platform in the Sumida ward of Tokyo.At the current stage of demonstration experiment, 100 vending machines and 65 vehicles including beverage delivery trucks and taxis are equipped with wireless routers that have a coverage radius of 250 ∼ 300 meters in the real field.Every few second, these wireless nodes gather various sensing data from the sensors deployed in their surrounding environments, and encapsulate them into a small data packet of 100 bytes.These data packets are disseminated in the mobile IoT networks by SF protocol and processed by different smart city applications deployed at the edges of networks.As proved by the previous simulation results, SF protocol is not scalable to the increase of network nodes in this platform.Thus, there is a strong intention for us to replace it with the proposed NPB protocol in near future.The ultimate objective of this project is to construct an IoT platform covers a service region of 10 km 2 .Since it may be not necessary to disseminate local sensing data in such a wide area, the time-to-live (TTL) value of data packets can be used to constrain their dissemination scope within several hundred or 1 ∼ 2 Finally, to make this platform compatible with existing ''Big Data'' systems, a few gateway routers may be deployed in networks to gather disseminated sensing data and deliver them to a remote data center for further analysis.This kind of hybrid mobile IoT networks is also economical to industrial companies because it is very expensive to maintain an internet access for every wireless node separately. VI. CONCLUSIONS AND FUTURE WORK This paper proposed a neighbor-based probabilistic broadcast protocol to facilitate the data dissemination in mobile IoT networks.The proposed protocol uses the number of uncovered neighbors to determine the rebroadcast delay of a data packet, and utilizes the additional coverage ratio of the packet and a metric named adaptive connectivity factor to determine its rebroadcast probability.Extensive simulation results have validated that the proposed protocol successfully improves packet delivery ratio by 13% ∼ 28% with a similar endto-end delay and network overhead of the most state-of-art approaches.For future work, there is a strong intention for us to implement the proposed protocol in an under constructed IoT platform in Tokyo. FIGURE 2 . FIGURE 2. An application scenario of traffic safety. FIGURE 3 . FIGURE 3. The comparison between two broadcast protocols.(a) Simple flooding protocol.(b) The optimal broadcast protocol. FIGURE 4 . FIGURE 4. Node density.(a) The packet delivery ratios of different protocols.(b) The end-to-end delays of different protocols.(c) The network overheads of different protocol Figure Figure 4 presents the performances of different broadcast protocols with varied node density.When the number of nodes in the area was small, there was little collision and interference in wireless channel.SF protocol achieved the best packet delivery ratio since it broadcasted more redundant packets than other improved protocols, and these redundant packets improved the possibility of successful packet FIGURE 5 . FIGURE 5. Packet size.(a) The packet delivery ratios of different protocols.(b) The end-to-end delays of different protocols.(c) The network overheads of different protocols. FIGURE 6 . FIGURE 6. Packet transmission interval.(a) The packet delivery ratios of different protocols.(b) The end-to-end delays of different protocols.(c) The network overheads of different protocols. FIGURE 7 . FIGURE 7. Speed of nodes.(a) The packet delivery ratios of different protocols.(b) The end-to-end delays of different protocols.(c) The network overheads of different protocols. FIGURE 8 . FIGURE 8.The under constructed IoT platform in Tokyo.(a) Wireless router (box version and tablet version).(b) Vending machine distribution in the Sumida ward of Tokyo.(c) Wireless-enabled vending machine.(d) Wireless-enabled vehicle. Finally, since network nodes with NPB protocol only make rebroadcast decisions according to their local neighbor information and Eqs.(1 -6) are linear or truncated linear functions, TABLE 1 . The default parameters of simulations.	7,414.4	2020-01-02T00:00:00.000	[ "Computer Science", "Engineering", "Environmental Science" ]
A REGENERATION PROOF OF THE CENTRAL LIMIT THEOREM FOR UNIFORMLY ERGODIC MARKOV CHAINS Central limit theorems for functionals of general state space Markov chains are of crucial importance in sensible implementation of Markov chain Monte Carlo algorithms as well as of vital theoretical interest. Diﬀerent approaches to proving this type of results under diverse assumptions led to a large variety of CTL versions. However due to the recent development of the regeneration theory of Markov chains, many classical CLTs can be reproved using this intuitive probabilistic approach, avoiding technicalities of original proofs. In this paper we provide a characterization of CLTs for ergodic Markov chains via regeneration and then use the result to solve the open problem posed in [17]. We then discuss the diﬀerence between one-step and multiple-step small set condition. Introduction Let (X n ) n 0 be a time homogeneous, ergodic Markov chain on a measurable space (X , B(X )), with transition kernel P and a unique stationary measure π on X . We remark that the ergodicity means that lim n→∞ P n (x, ·) − π tv = 0, for all x ∈ X , where · tv denotes the total variation distance. The process (X n ) n 0 may start from any initial distribution π 0 . Let g be a real valued Borel function on X , square integrable against the stationary measure π. We denote byḡ its centered version, namelyḡ = g − gdπ and for simplicity S n := n−1 i=0ḡ (X i ). We say that a √ n−CLT holds for (X n ) n 0 and g if where σ 2 g < ∞. First we aim to provide a general result, namely Theorem 4.1, that gives a necessary and sufficient condition for √ n-CLTs for ergodic chains (which is a generalization of the well known Theorem 17.3.6 [11]). Assume for a moment that there exists a true atom α ∈ B(X ), i.e. such a set α that π(α) > 0 and there exists a probability measure ν on B(X ), such that P (x, A) = ν(A) for all x ∈ α. Let τ α be the first hitting time for α. In this simplistic case we can rephrase our Theorem 4.1 as follows: Furthermore we have the following formula for the variance σ 2 g = π(α)E α τα k=1ḡ (X k ) 2 . Central limit theorems of this type are crucial for assessing the quality of Markov chain Monte Carlo estimation (see [10] and [5]) and are also of independent theoretical interest. Thus a large body of work on CLTs for functionals of Markov chains exists and a variety of results have been established under different assumptions and with different approaches (see [9] for a review). We discuss briefly the relation between two classical CLT formulations for geometrically ergodic and uniformly ergodic Markov chains. We say that a Markov chain (X n ) n 0 with transition kernel P and stationary distribution π is • geometrically ergodic, if P n (x, ·)−π(·) tv M (x)ρ n , for some ρ < 1 and M (x) < ∞ π-a.e., • uniformly ergodic, if P n (x, ·) − π(·) tv M ρ n , for some ρ < 1 and M < ∞. Recently the following CLT provided by [8] has been reproved in [17] using the intuitive regeneration approach and avoiding technicalities of the original proof (however see Section 6 for a commentary). Roberts and Rosenthal posed an open problem, whether the following CLT version for uniformly ergodic Markov chains due to [4] can also be reproved using direct regeneration arguments. Theorem 1.4. If a Markov chain (X n ) n 0 with stationary distribution π is uniformly ergodic, then a √ n−CLT holds for (X n ) n 0 and g whenever π(g 2 ) < ∞. Moreover σ 2 g := Xḡ The aim of this paper is to prove Theorem 4.1 and show how to derive from this general framework the regeneration proof of Theorem 1.4. The outline of the paper is as follows. In Section 2 we describe the regeneration construction, then in Section 3 we provide some preliminary results which may also be of independent interest. In Section 4 we detail the proof of Theorem 4.1, and derive Theorem 1.4 as a corollary in Section 5. Section 6 comprises a discussion of some difficulties of the regeneration approach. Small Sets and the Split Chain We remark that ergodicity as defined by (1) is equivalent to Harris recurrence and aperiodicity (see Proposition 6.3 in [13]). One of the main feature of Harris recurrent chains is that they are ψ−irreducible and admit the regeneration construction, discovered independently in [12] and [1], and which is now a well established technique. In particular such chains satisfy Definition 2.1 (Minorization Condition). For some ε > 0, some C ∈ B + (X ) := {A ∈ B(X ) : ψ(A) > 0} and some probability measure ν m with ν m (C) = 1 we have for all x ∈ C, The minorization condition (4) enables constructing the split chain for (X n ) n 0 which is the central object of the approach (see Section 17.3 of [11] for a detailed description). The minorization condition allows to write P m as a mixture of two distributions: where . Now let (X nm , Y n ) n 0 be the split chain of the m−skeleton i.e. let the random variable Y n ∈ {0, 1} be the level of the split m−skeleton at time nm. The split chain (X nm , Y n ) n 0 is a Markov chain that obeys the following transition ruleP.P and Y n can be interpreted as a coin toss indicating whether X (n+1)m given X nm = x should be drawn from ν m (·) -with probability εI C (x) -or from R(x, ·) -with probability 1 − εI C (x). One obtains the split chain (X k , Y n ) k 0,n 0 of the initial Markov chain (X n ) n 0 by defining appropriate conditional probabilities. To this end let X nm Note that the marginal distribution of (X k ) k 0 in the split chain is that of the underlying Markov chain with transition kernel P. andP From the Bayes rule we obtaiň and the crucial observation due to Meyn and Tweedie, emphasized here as Lemma 2.2 follows. whereas hitting times τα(n) are defined as follows: We define also Tools and Preliminary Results In this section we analyze the sequence s i (ḡ), i 0. The basic result we often refer to is Theorem 17.3.1 in [11], which states that (s i ) i 0 is a sequence of 1-dependent, identically distributed r.v.'s withĚs i = 0. In our approach we use the following decomposition: A look into the proof of Lemma 3.3 later in this section clarifies that s i and s i are well defined. Proof. First note that s i is a function of {X (σα(i)+1)m , X (σα(i)+1)m+1 , . . . } and that Y σα(i) = 1, hence by Lemma 2.2 s 0 , s 1 , s 2 , . . . are identically distributed. Now focus on s i , s i+k and Y σα(i+k) for some k 1. Obviously Y σα(i+k) = 1. Moreover s i is a function of the pre−σα(i + k)m process and s i+k is a function of the post−(σα(i + k) + 1)m process. Thus s i and s i+k are independent again by Lemma 2.2 and for A i , A i+k , Borel subsets of R, we havě By the same pre-and post-process reasoning we obtain for A i1 , . . . , A i l Borel subsets of R thať , and the proof is complete by induction. Now we turn to prove the following lemma, which generalizes the conclusions drawn in [7] for uniformly ergodic Markov chains. Lemma 3.2. Let the Markov chain (X n ) n 0 be recurrent (and (X nm ) n 0 be recurrent) and let the minorization condition (4) hold with π(C) > 0. Then where π C (·) is a probability measure proportional to π truncated to C, that is Proof. The first equation in (13) is a straightforward consequence of the split chain construction. To prove the second one we use Theorem 10.0.1 of [11] for the split m−skeleton with A =α. Thus τ A = τα(1) andπ := π * is the invariant measure for the split m−skeleton. Let C ⊇ B ∈ B(X ), and compute This implies proportionality and the proof is complete. and is a function of the random variable By µ i (·) denote the distribution of (14) on X m . We will show that µ i does not depend on i. From (8), (11) and the Bayes rule, for x ∈ C, we obtaiň Lemma 3.2 together with (15) yieldš Note that νm(dy) P m (x,dy) is just a Radon-Nykodym derivative and thus (16) is a well defined measure on X m+1 , say µ(·). It remains to notice, that µ i (A) = µ(A × X ) for any Borel A ⊂ X m . Thus µ i , i 0 are identical and hence s i , i 0 have the same distribution. Due to Lemma 2.2 we obtain that s i , i 0 are 1-dependent. To proveĚ π * 0 s 2 i < ∞, we first note that νm(dy) P m (x,dy) 1/ε and also π C (·) 1 π(C) π(·). Hence where µ chain is defined by π(dx)P (x, dx 1 ) . . . P (x m−2 , dx m−1 ). Thus We need a result which gives the connection between stochastic boundedness and the existence of the second moment of s i . We state it in a general form. Theorem 3.4. Let (X n ) n 0 be a sequence of independent identically distributed random variables and S n = n−1 k=0 X k . Suppose that (τ n ) is a sequence of positive, integer valued r.v.'s such that τ n /n → a ∈ (0, ∞) in probability when n → ∞ and the sequence (n −1/2 S τn ) is stochastically bounded. Then EX 2 0 < ∞ and EX 0 = 0. The proof of Theorem 3.4 is based on the following lemmas. Proof. Let (X ′ i ) be an independent copy of (X i ) and S ′ k = n i=1 X ′ i . Moreover let (ε i ) be a sequence of independent symmetric ±1 r.v.'s, independent of (X i ) and (X ′ i ). For any reals (a i ) we get by the Paley-Zygmund inequality, Hence for sufficiently large n by the Weak LLN. Thus Corollary 3.7. Let c 2 < Var(X 1 ), then for sufficiently large n, Proof. Let t 0 be as in Lemma 3.5 for δ = 1/16, then Hence by Lemma 3.5 we obtain t 0 c √ n/4 for large n. for sufficiently large n. Since (n −1/2 S τn ) is stochastically bounded, we immediately obtain Var(X 1 ) < ∞. If EX 1 = 0 then 1 √ n S τn = S τn τ n τ n n √ n → ∞ in probability when n → ∞. A Characterization of √ n-CLTs In this section we provide a generalization of Theorem 17.3.6 of [11]. We obtain an if and only if condition for the √ n-CLT in terms of finiteness of the second moment of a centered excursion fromα. Furthermore we have the following formula for variance Proof. For n 0 define l n := max{k 1 : m(σα(k) + 1) n} and for completeness l n := 0 if m(σα(0) + 1) n. First we are going to show that Thus we have to verify that the initial and final terms of the sum do not matter. First observe that by the Harris recurrence property of the chain σα(0) < ∞,P π * 0 -a.s. and hence lim n→∞Pπ * 0 (mσα(0) n) = 0 andP π * 0 (σα(0) < ∞) = 1. This yields The second point is to provide a similar argument for the tail terms and to show that For ε > 0 we havě where we use thatα is an atom for the split chain, we deduce form the Lebesgue majorized convergence theorem that (20) holds. Obviously (19) and (20) yield (18). We turn to prove that the condition (17) is sufficient for the CLT to hold. We will show that random numbers l n can be replaced by their non-random equivalents. Namely we apply the LLN (Theorem 17.3.2 in [11])) to ensure that lim n→∞ l n n = lim Let n * := ⌊π(α)nm −1 ⌋, n := ⌈(1 − ε)π(α)nm −1 ⌉, n := ⌊(1 + ε)π(α)nm −1 ⌋. Due to the LLN we know that for any ε > 0, there exists n 0 such that for all n n 0 we havě P π * 0 (n l n n) 1 − ε. Consequently Since (s j ) j 0 are 1-dependent, M k := k j=0 s j is not necessarily a martingale. Thus to apply the classical Kolmogorov inequality we define M 0 k = ∞ j=0 s 2j I {2j≤k} and M 1 k = ∞ j=0 s 1+2j I {1+2j≤k} , which are clearly square-integrable martingales (due to (17)). Hencě where C is a universal constant. In the same way we show thatP(max n * +1 l n \|M l −M n * +1 \| > β √ n) Cεβ −2Ě ν * m (s 2 0 ), consequently, since ε is arbitrary, we obtain The last step is to provide an argument for the CLT for 1-dependent, identically distributed random variables. Namely, we have to prove that Observe that (19), (20), (24) and (25) imply Theorem 4.1. We fix k 2 and define ξ j := s kj+1 (ḡ) + ... + s kj+k−1 (ḡ), consequently ξ j are i.i.d. random variables and Obviously the last term converges to 0 in probability. Denoting , and σ 2 s :=Ě ν * m (s 0 (ḡ)) 2 . we use the classical CLT for i.i.d. random variables to see that converges to N (0, σ 2 g ), with k → ∞. Since the weak convergence is metrizable we deduce from (26), (27) and (28) that (25) holds. The remaining part is to prove that (17) is also necessary for the CLT to hold. Note that if n k=0ḡ (X k )/ √ n verifies the CLT then ln−1 j=0 s j is stochastically bounded by (18). We use the decomposition s i = s i + s i , i 0 introduced in Section 3. By Lemma 3.3 we know that s j is a sequence of 1-dependent random variables with the same distribution and finite second moment. Thus from the first part of the proof we deduce that ln−1 j=0 s j / √ n verifies a CLT and thus is stochastically bounded. Consequently the remaining sequence ln−1 j=0 s j / √ n also must be stochastically bounded. Lemma 3.1 states that (s j ) j 0 is a sequence of i.i.d. r.v.'s, henceĚ[s 2 j ] < ∞ by Theorem 3.4. Also l n /n →π(α)m −1 by (21). Applying the inequality (a + b) 2 2(a 2 + b 2 ) we obtaiň Remark 4.2. Note that in the case of m = 1 we haves i ≡ 0 and for Theorem 4.1 to hold, it is enough to assume π\|g\| < ∞ instead of π(g 2 ) < ∞. In the case of m > 1, assuming only π\|g\| < ∞ and (17) implies the √ n-CLT, but the proof of the converse statement fails, and in fact the converse statement does not hold (one can easily provide an appropriate counterexample). Uniform Ergodicity In view of Theorem 4.1 providing a regeneration proof of Theorem 1.4 amounts to establishing conditions (17) and checking the formula for the asymptotic variance. To this end we need some additional facts about small sets for uniformly ergodic Markov chains. Theorem 5.1. If (X n ) n 0 , a Markov chain on (X , B(X )) with stationary distribution π is uniformly ergodic, then X is ν m −small for some ν m . Hence for uniformly ergodic chains (4) holds for all x ∈ X . Theorem 5.1 is well known in literature, in particular it results from Theorems 5.2.1 and 5.2.4 in [11] with their ψ = π. Theorem 5.1 implies that for uniformly ergodic Markov chains (5) can be rewritten as The following mixture representation of π will turn out very useful. Lemma 5.2. If (X n ) n 0 is an ergodic Markov chain with transition kernel P and (29) holds, then Remark 5.3. This can be easily extended to the more general setting than this of uniformly ergodic chains, namely let P m (x, Related decompositions under various assumptions can be found e.g. in [14], [7] and [3] and are closely related to perfect sampling algorithms, such as coupling form the past (CFTP) introduced in [15]. Proof. First check that the measure in question is a probability measure. Proof. (i) is a direct consequence of (30). To see (ii) note that Y n is a coin toss independent of {Y 0 , . . . , Y n−1 } and X nm , this allows for π * instead of π on the RHS of (ii). Moreover the evolution of {X nm+1 , X nm+2 , . . . ; Y n+1 , Y n+2 , . . . } depends only (and explicitly by (8) and (9)) on X nm and Y n . Now use (i). Our object of interest is Next we use Corollary 5.4 and then the inequality 2ab a 2 + b 2 to bound the term A in (31). Since π(C) = 1, we have σ 2 g = εm −1 (I + J). Next we use Lemma 2.2 andĚ π * Z 0 (ḡ) = 0 to drop indicators and since for f : X → R, alsoĚ π * f = E π f, we have Now, since all the integrals are taken with respect to the stationary measure, we can for a moment assume that the chain runs in stationarity from −∞ rather than starts at time 0 with X 0 ∼ π. Thus 6 The difference between m = 1 and m = 1 Assume the small set condition (4) holds and consider the split chain defined by (8) and (9). The following tours {X (σ(n)+1)m , X (σ(n)+1)m+1 , . . . , X (σ(n+1)+1)m−1 }, n = 0, 1, . . . that start whenever X k ∼ ν m are of crucial importance to the regeneration theory and are eagerly analyzed by researchers. In virtually every paper on the subject there is a claim these objects are independent identically distributed random variables. This claim is usually considered obvious and no proof is provided. However this is not true if m > 1. Let ν 4 (d) = ν 4 (e) = 1/2 and ε = 1/8. Clearly P 4 (x, ·) εν 4 (·) for every x ∈ X , hence we established (4) with C = X . Note that for this simplistic example each tour can start with d or e. However if it starts with d or e the previous tour must have ended with b or c respectively. This makes them dependent. Similar examples with general state space X and C = X can be easily provided. Hence Theorem 4.1 is critical to providing regeneration proofs of CLTs and standard arguments that involve i.i.d. random variables are not valid.	4,680.2	2008-01-24T00:00:00.000	[ "Computer Science", "Mathematics" ]
Infection-induced extracellular vesicles evoke neuronal transcriptional and epigenetic changes Infection with the protozoan Toxoplasma gondii induces changes in neurotransmission, neuroinflammation, and behavior, yet it remains elusive how these changes come about. In this study we investigated how norepinephrine levels are altered by infection. TINEV (Toxoplasma-induced neuronal extracellular vesicles) isolated from infected noradrenergic cells down-regulated dopamine ß-hydroxylase (DBH) gene expression in human and rodent cells. Here we report that intracerebral injection of TINEVs into the brain is sufficient to induce DBH down-regulation and distrupt catecholaminergic signalling. Further, TINEV treatment induced hypermethylation upstream of the DBH gene. An antisense lncRNA to DBH was found in purified TINEV preparations. Paracrine signalling to induce transcriptional gene silencing and DNA methylation may be a common mode to regulate neurologic function. Strategy for identifying paracrine signalling in DBH down-regulation. Initially, it was considered that several different factors could explain the disproportionate decrease in DBH expression (and hence NE) relative to the small percent of parasitised cells during chronic infection (e.g. neuroimmune responses). As decreased NE and DBH down-regulation (relative to other genes) has been observed in vitro with noradrenergic cell lines, mechanisms other than the host immune system are involved although host cell immunity remained a possibility. A nuclear run-on assay was performed to assess whether the DBH down-regulation was at the transcriptional or post-transcriptional level. De novo transcription in nuclei isolated from infected cell cultures was measured by immunocapture of incorporated biotin-UTP. Lower amounts of nascent DBH mRNA (relative to standards) were found in host cell nuclei from infected than uninfected noradrenergic cell cultures (21 ± 1.6fold, p = 0.00062) ( Fig. 2A). Similarly, de novo transcription of DBH was downregulated in infected human noradrenergic cells (21 ± 1.7-fold, p = 0.032) (Fig. 2B). Hence, infection induced transcriptional gene silencing (TGS) of DBH in human and rat noradrenergic cells. We then investigated the spreading of DBH down-regulation by examining uninfected cells in cultures to determine whether cells exposed to infected cells were also suppressed in NE as this could help explain the large change in expression observed. A transwell system was used that permits uninfected cells to be exposed to T. gondii-infected cell products. This method was chosen because it differentiates between diffusible signals and parasites injecting components into cells without invasion, as was observed with a Cre/loxP assay in infected mouse brains 36 . DBH expression was measured in uninfected rat noradrenergic cells in the bottom reservoir of the transwell system with the top reservoir containing an infected culture (Fig. 2C). DBH expression in the cells exposed to infected cultures was found to be down-regulated (7.9 ± 2.8-fold, p = 0.02) suggesting that a transmissible factor was released from infected cells. This was subsequently identified as EVs. In contrast, noradrenergic cells that were exposed to cultures containing heat-killed T. gondii ('mock-infected') were unchanged in DBH expression (Fig. 2D). Exposure of a human neuronal cell line to infected cells in the transwell system induced a larger decrease in DBH expression (37± 16-fold, p = 0.0038) than the rat cell line (Fig. 2E). The observed DBH down-regulation is likely to be a minimal baseline as it remains possible that vesicles may stick to the transwell membrane and that EV passage is restricted. Parasite restriction to the upper reservoir of the 0. 4 www.nature.com/scientificreports/ www.nature.com/scientificreports/ reservoirs and monitoring propagation (data not shown). Transwells were set up with infected fibroblasts in the top reservoir and uninfected noradrenergic cells in the bottom reservoir to assess whether the down-regulation was cell-type specific (Fig. 2F). The noradrenergic cells exposed to T. gondii-infected fibroblast cultures were unchanged in DBH gene expression. A further indication that EVs were the permeable effector responsible for the DBH down-regulation was finding that the insoluble components, separated from soluble factors by ultracentrifugation, contained the DBH down-regulating activity in preliminary tests (data not shown). This provided the rationale for EV isolation and testing. Epigenetic changes associated with DBH down-regulation during infection. As our findings indicated that TGS was responsible for DBH expression changes, the epigenetic state of the DBH gene was investigated 37 . Methylation Sensitive Restriction Enzyme qPCR (MSRE-qPCR) was used to monitor DNA methylation levels in the DBH gene's upstream region where the majority of CpGs are clustered (Fig. 3A). Methylation in the DBH upstream region rose from 16 ± 4.6% to 66 ± 3.8% in infected cultures of noradrenergic cells during the course of the infection ( Fig. 3B; p = 0.00072). As the noradrenergic cells are sensitive to pH changes (ie. neurotransmitter synthesis and synaptic transmission affected) 38,39 , alkaline-shocked tachyzoites were used Cells are orange, parasites red, signalling factors in vesicles (green) and soluble (yellow). (D) DBH expression measured using RT-qPCR (relative to GAPDH) in PC12 cells from the bottom layer with either uninfected control, T. gondii infected, or mock-infected cells in the upper transwell layer. Mock-infected cells were incubated with heat-killed T. gondii tachyzoites n = 3, p < 0.05. One-way ANOVA p = 0.017, Tukey's post hoc test uninfected and mock-infected vs infected p = 0.020 and p = 0.033, respectively. (E) Noradrenergic BE(2)-M17 cell expression of DBH mRNA cultured in transwells and treated as in (C); n = 3, p = 0.0038. (F) Infected fibroblasts in the upper layer of the transwell system. DBH expression levels were measured in exposed noradrenergic cells. DBH gene expression was not significantly altered in cells exposed to infected fibroblasts (p = 0.84); ± SEM shown, n = 3, Student's t test. 40 . This procedure elevated expression of bradyzoite markers BAG1 and SAG4 (Fig. S5). DBH methylation was also increased in infected human noradrenergic cultures (2.8-fold; range 2.3-4.8-fold; p = 0.0011) with a time-dependent increase in methylation in the region profiled (Fig. 3C). As an indicator of EV involvement in the epigenetic changes, cells were treated with an inhibitor of EV biogenesis in the transwell system to restrict parasites and parasite-infected cells from contact with the uninfected cell culture layer. The cells exposed to infected cultures had > 50% higher methylation of the DBH promoter (Fig. 3D). These cultures were treated with GW4869, a sphingomyelinase inhibitor that disrupts vesicle budding required for endosomal formation. Addition on GW4869 to the transwell system abrogated the DBH hypermethylation observed in uninfected noradrenergic cells exposed to infected cultures. Treatment with GW4869, a neutral sphingomyelinase (N-SMase) inhibitor, disrupted the DBH promoter methylation by permeable factors in the transwell system. DBH promoter methylation in the lower layer exposed cells was quantified by MSRE qPCR as described above. ± SEM shown, n = 3, Student's t test, p = 0.0093. www.nature.com/scientificreports/ In order to examine the epigenetic effects of T. gondii on DBH in neurons, ex vivo experiments were performed with infection of organotypic brain slices of the prefrontal cortex, nucleus accumbens and ventral tegmental areas and the TGS in the neuron population measured, as the percentage of infected neurons during chronic infection is 0.002-0.14% 29,41 . Methylation of the DBH upstream region rose from 29 ± 2.7% in the brain tissue slices to 74 ± 4.6% in the infected slices (p = 0.000051) (Fig. 4A). DNA methylation levels were then analysed in the brains of chronically-infected mice. Neurons were purified from other brain cell types by FACS and DBH methylation was measured. The DBH gene in infected animals was 53 ± 7.7% methylated in neurons compared to 6.3 ± 2.0% in uninfected mice (p = 0.000045, Fig. 4B). For comparison, levels of total genomic DNA methylation were measured (Fig. 4C). No change in global DNA methylation was observed in noradrenergic cells or neurone Slice cultures were either uninfected (light grey, n = 6) or infected with wild-type induced-bradyzoite T. gondii at MOI 5 (dark grey, n = 3) and MOI 7 (black, n = 3). Infection was monitored for 5 days by light microscopy; biological repeats contain 3 slices from the same rat p = 0.04, *p = 0.000051 shown. One-way ANOVA, p = 0.0005 Tukey's post hoc test MOI 7 vs uninfected p = 0.0004. (B) Methylation at the DBH promoter region in neurons from uninfected (grey, n = 9) and chronically-infected (black, n = 11) mice. Neurons were enriched from harvested brains by FACS using NeuN antibody and DBH methylation quantified by MSRE-qPCR. p = 0.0004. (C) Total genomic DNA methylation measured by ELISA of both uninfected and infected rat catecholaminergic PC12 cells (n = 3, p = 0.52), human neuronal BE(2)-M17 cell (n = 5, p = 0.59) and neuronal nuclei enriched by FACS from mouse brain tissue for uninfected (n = 6) and chronically-infected mice (n = 3, p = 0.15). No significant change in methylation was identified by Student's t tests, ± SEM shown. (D) Expression of DBH mRNA measured RT-qPCR relative to GAPDH following treatments. Uninfected (grey) and T. gondii-infected (black) noradrenergic PC12 cells treated with trichostatin A (TSA), 5-azacytidine (5-AC) and RG108; n = 5, p < 0.01. (E) As in (D) with noradrenergic human BE(2)-M17 cells treated with TSA, 5-AC and RG108; n = 5, p < 0.05. n = 5. Student's t test, ± SEM shown. (F) 5′ DBH hypermethylation was measured by MSRE-qPCR (as above). Uninfected (grey) and T. gondii-infected (black) PC12 cells were treated with trichostatin A (TSA), 5-azacytidine (5-AC) and RG108; n = 5, p < 0.01. (G) Uninfected (grey) and infected (black) BE(2)-M17 cells treated as in (F) with TSA, 5-AC and RG108; n = 5, p < 0.05. Student's t test, ± SEM shown. www.nature.com/scientificreports/ with T. gondii infection, as has previously been found 42 . Hypermethylation of CpG residues upstream of the DBH gene were also found by NGS genomic bisulfite sequencing of infected cultures (Fig. S6). The mechanism responsible for the parasite-induced DNA methylation and chromatin remodelling associated with the silencing was explored by investigating the role of DNA methyltransferase (DNMT) and histone deacetylation in this process. Infected noradrenergic cells were treated with the DNMT inhibitors RG108 and 5-azacytidine (5-AC). Both compounds disrupted the parasite-induced DBH down-regulation and DNA hypermethylation in rat noradrenergic cells, relative to marker (Fig. 4D, F). The DNMT inhibitors similarly blocked the TGS and epigenetic changes in human neuronal cells (Fig. 4E, G). This was not due to parasite sensitivity to the inhibitors since the inhibitors were non-toxic to parasites at concentrations tested (data not shown) and T. gondii lacks 5-methylcytosine 43 . This implies that the DBH silencing involves DNMT. In contrast, the histone deacetylase inhibitor trichostatin A did not abrogate the DBH down-regulation or hypermethylation in infected cultures ( Fig. 4D-G). Hence the mechanism may not involve histone deacetylation or activation was not captured within the experimental timeframe of 5 days and histone deacetylase is active at a different time in the epigenetic modification pathway. Scientific Reports To directly compare the epigenetic changes in infected versus uninfected (but infection-exposed) cells in the same culture, populations were enriched for cells containing GFP-expressing T. gondii (GFP+) and GFP− cells. DBH hypermethylation was found in both populations of cells (Fig. 5A, B; ANOVA test, p = 0.0089). Indeed, the GFP(-) cells exhibited DNA methylation levels at least equal to the GFP+ cells. This further supports the supposition that the silencing is spread and direct infection is not required for hypermethylation of the DBH gene in a cell. In order to observe the induction of DBH hypermethylation in uninfected cells exposed to infected cultures, the methylation status of the DBH upstream region in noradrenergic cells was measured by MSRE-sqPCR. Methylation of the DBH 5' region was 34 ± 4.8% greater (p = 0.0013) in rat cells exposed to infected cells compared to exposure to uninfected cultures (Fig. 5C). Human noradrenergic cells exposed to infected cells also had increased methylation (49 ± 1.9%, p = 0.0003) of the DBH upstream region (Fig. 5D). Hence, both transcriptional down-regulation and promoter hypermethylation are induced in human and rat noradrenergic cells exposed to T. gondii cell cultures. Based on the above findings, the isolated TINEVs ability to induce epigenetic changes was next investigated. DBH gene promoter methylation, examined using the MSRE-qPCR assay, increased from 16 ± 3.9% to 81 ± 7.9% over the course of the experiment. In a time course of exposure to TINEVs, methylation was not significantly changed after 24 or 48 h of exposure but hypermethylation was observed at 96 h of incubation ( Fig. 5E; ANOVA p = 0.0008). Hence DBH transcriptional down-regulation and chromatin remodelling are induced by EVs from infected cells. Toxoplasma-induced neuronal EVs contain an antisense DBH lncRNA. We investigated the RNA cargo for association with the TINEV-induced TGS and epigenetic modifications. Preliminarily, the sensitivity of the TINEVs to ultraviolet (UV) irradiation was tested to assess whether an RNA component could be involved 44 . TINEVs treated with UV radiation no longer induced hypermethylation of the DBH promoter (Fig. S7). Although, it is possible that the UV radiation inactivated proteins. With the specificity of the differential gene expression and the role of non-coding RNAs (ncRNA) in regulating gene expression, we explored the potential presence of a long non-coding RNA (lncRNA). LncRNAs have roles in neuronal gene expression such as several highly specific antisense transcripts [45][46][47][48] . EVs have been found to contain miRNAs and lncRNAs, although their functional significance is still unclear. With the specificity of the DBH down-regulation in T. gondii infection and its magnitude with TINEV treatment, the identification of a long non-coding RNA (lncRNA) was investigated 45,49 . A panel of primers walking upstream stepwise from the DBH coding region were used to screen for an antisense RNA 37 . These identified a lncRNA in infected cultures (Fig. S8). RNA purified from TINEV preparations were found to contain the DBH antisense lncRNA ( Fig. 6A-C, Fig. S9, S10). Although the lncRNA location internally as EV cargo is unconfirmed, it is unclear of the functional significance of the location 50 . The lncRNA is complementary to the DBH upstream region containing cis-regulatory elements and crosses the transcription start site. As a positive control for samples, miR-21 miRNA served to confirm RNA quality purified from the TINEVs 51 . Intriguingly, the timeframe observed for the DBH hypermethylation in this study (Fig. 5E) is similar to dynamic DNA methylation changes observed in cells treated with promoter targeted antisense RNA 52 . The observations above represent the first study to show a specific and functionally-relevant lncRNA to be transmitted from one neuronal cell to another regulating neurotransmission 53 . The DBH antisense lncRNA was found to contain the conserved U1 snRNP recognition site that is a consensus sequence identified in chromatin-bound lncRNAs 54 . Further, predicted secondary structure analysis places the U1 snRNP recognition site in a loop region in minimum free energy secondary structures of DBH antisense lncRNA (Fig. 6D). In the published study, the recognition site needs to be in a loop region to allow binding of U1 snRNP to the lncRNA to promote chromatin interaction. Future experiments will further investigate this novel induced lncRNA. Discussion Intracellular infection can manipulate gene expression not only within the host cell but signalling to other cells. In this study it was found that T. gondii infection resulted in the release of EVs from noradrenergic cells that specifically induced gene silencing and DNA hypermethylation in vitro and in the brain. Many cells of the CNS release EVs including astrocytes, glia, oligodendrocytes, Schwann cells and neurons [55][56][57] . EV secretion is emerging as an important mechanism of cellular communication within the CNS 58 www.nature.com/scientificreports/ cultures and delivered into the brain induced DBH down-regulation in neurons. EVs are continuously released in the brain limiting our ability to detect and isolate those induced by the 0.002-0.14% of cells that are infected during chronic T. gondii infection and restricted by the current lack of biomarkers for TINEVs. The findings presented here provide a mechanism that can resolve the enigma of an observable decrease in NE when only a small percentage of brain cells are infected. We propose a model of T. gondii inducing TINEV release with host lncRNAs that modulate the host environment 59 . Studies of extracellular vesicles secreted from pathogen-infected cells, including the apicomplexan parasites, have reported altering the host immune response [60][61][62][63][64] . For example, cells infected with Epstein-Barr virus release EVs containing viral miRNAs that can downregulate cytokine expression in uninfected monocyte-derived dendritic cells. Human miRNAs complexed with Argonaute 2 are released in EVs from Plasmodium-infected red blood cells and induce pro-inflammatory cytokines in macrophages and endothelial cells. The findings reported here, in contrast, demonstrate a role for TINEVs in altering neurophysiology. TINEVs spreading to uninfected neurons carried host mammalian lncRNA, regulated gene expression, and modified chromatin. Intriguingly, NE modulates CNS cytokine release and hence the resultant NE decrease from DBH down-regulation will also www.nature.com/scientificreports/ affect the neuroimmune response. Host hypomethylation of the mammalian arginine vasopressin promoter in the amygdala of T. gondii-infected rodents has also been observed 42 . EV-miRNAs have been reported that alter DNA methylation in surrounding cells 63 . Here, we identified a natural mammalian antisense lncRNA in TINEV preparations. The lncRNA is complimentary to the DBH 5'UTR and contains a consensus sequence for chromatin localization. It remains possible that DBH down-regulation is imposed by cell-autonomous immunity in noradrenergic cells by another component of TINEVs that has yet to be defined. Chromatin localisation of the lncRNA may occur via U1 snRNP binding at the transcriptionally active DBH gene and interacting with RNA polymerase II to disrupt transcription and facilitate DNA methylation. In this study, diffusible products of infected fibroblasts including EVs did not induce DBH down-regulation (Fig. 4F). In the future it will be interesting to assess the intercellular crosstalk from infected neurons to different brain cell types and to examine how T. gondii induces host cells to express and package selective lncRNAs. Evidence is rapidly accumulating to support the role of EVs in the regulation of neurogenesis, neuronal connectivity and neuroimmune communication 44,65 . The findings in this study describe a novel form of neurotransmission regulation that may contribute to maintaining chronic infection. We identified a natural antisense lncRNA in TINEV preparations that specifically altered neurotransmitter signalling. In a recent study, reduced www.nature.com/scientificreports/ CNS noradrenergic signalling during chronic T. gondii infection correlated with noradrenergic-linked behaviour changes 11,66 . Chronic infection with bradyzoite-infected neurons permits continuous delivery of TINEVs into the brain. In this study, TINEVs delivered by intracranial injection twice daily for 3-days down-regulated DBH. As behaviour changes are measured in rodents during the 4th to 6th week of infection, a sustained supply of TINEVs (or the antisense lncRNA) is necessary for comparative behaviour tests. Also, reduced DBH has been found to elevate dopamine release from LC neurons (such as those modulating learning and memory) 7,67 which could explain increased dopamine signalling that has been observed with T. gondii infection 68 . The neuroimmune system may also be affected by EVs. Changes in NE levels can alter cytokine responses via adrenergic receptors on astrocytes and microglia. Indeed, treatment with an adrenergic receptor agonist reversed the increased locomotion of infected mice 3,69 . Finding that EVs can induce TGS and modify DNA methylation neurons may represent a mechanism, beyond parasitic machinery, in innate neurophysiological function and enhance our understanding of the role of transcriptional regulation and dynamic DNA methylation in memory and learning. Infection of PC12 and BE(2)-M17 cells. Cells were cultured in multi-well plates at a density of 5 × 10 4 cells/ml. When stated, transwell plates (Nunc™ cell culture inserts, Thermo Scientific) were used, with cells plated at 2.5 × 10 4 cells/ml. Following 24 h of growth on 6-well plates, cells were infected with induced Prugniaud parasites in upper-wells, maintaining a multiplicity of infection (MOI) of 1. Cells were harvested immediately following infection (day 0) and after 3 and 5 days of infection for downstream processing. Mock-infection was performed using heat-killed T. gondii parasites subjected to 80 °C for 10 min. The cultures were monitored daily by light microscopy. Treatment of cells with purified TINEVs. EVs were extracted and purified from cell medium of uninfected and infected cells in exosome-depleted media, as described below. PC12 cells plated on 6-well plates and grown to a density of 2 × 10 4 cells/ml were treated with EVs at a 10:1 culture media concentration equivalent (ie. 1 ml of PC12 cell culture treated with EVs purified from 10 ml of culture). Cells were treated every 24 h. DNA and RNA were harvested 24 h after the last treatment. Rodent experiments. Nine-week-old wild-type Sprague Dawley male rats weighing 245 ± 15 g were obtained from Charles River Laboratories (Margate, UK) and were used in line with the United Kingdom Animals (Scientific Procedures) Act 1986 and ethical standards set by the University of Leeds Ethical Review Committee and in accordance with the ARRIVE guidelines. Animals were housed individually and maintained on a 12-h light-dark cycle with access to food and water ad libitum. Rats were monitored for health and fitness parameters of weight and food intake as shown in the supplemental data and the one that did not recover after surgery (lost more than 10% of their body weight) was not used in the experiments. Animals were randomly www.nature.com/scientificreports/ assigned from a total of 11 rats; 10 were used in this study. Animal care was blinded with investigators aware of the treatments. Animals were treated twice daily for 3 days. The outcome was measured by expression of DBH and compared by Student's t-test. Six male Lister Hood rats (Harlan Ltd, UK), 9-weeks old and weighing circa 300 g were infected and monitored as previously described 11 . Rats were housed in individual cages and maintained on a standard light-dark cycle with access to chow diet and water ad libitum. Animals were infected by intraperitoneal (IP) injection with tachyzoites. Throughout the experiments animals were monitored for illness or weight loss (more than 10%) and sacrificed 5-6 months post-infection. Brains were snap-frozen for cryosectioning and RNA extraction. (BALB/cAnNCrl × C57BL/6NCrl) F1 generation male mice (29 mice, no exclusions) were housed up to five to a cage and maintained on a standard light-dark cycle with access to chow diet and water ad libitum. Throughout the experiments animals were monitored for illness or weight loss (more than 10%) and sacrificed 4-6 weeks after infection in the University of Leeds research animal facility with procedures approved by the University Animal Care and Use Committee and following Home Office, HSE, regulations and guidelines for Animals (Scientific Procedures) Act 1986 published in 2014 and in accordance with ARRIVE guidelines with considerations of the replacement, reduction, and refinement in the use of animals for research. The mouse infections, brain dissection and frozen cryostat sectioning was performed as described 11 . Organotypic brain slice cultures. Sprague-Dawley rat pups (one litter) from females housed individually with ad libitum chow diet and water at postnatal days 3-4 were anesthetized as approved, following the Kyoto University Animal Research Committee approved protocols in the Kaneko Laboratory at Kyoto University. Sectioning was performed as described 41 . Brain was removed and separated into two hemispheres. Coronal sections (350 μm thickness) were prepared under sterile conditions. Slices were dissected to include the ventral tegmental area (VTA). These slice cultures were placed on 30 mm insert membranes (Millicell 0.4 μm; Millipore). Slice culture medium contained RPMI supplemented with 10% horse serum (GIBCO, USA), 5% FBS and additional 6.5 mg/ml glucose and 2 mM l-glutamine. The brain slice cultures were cultured at 37 °C in a 5% CO 2 for 15 days after dissection prior to use in experiments. Extracellular vesicle intracerebral treatment. Intracerebral injection into adult rats followed previously described experimental procedures 73 . TINEVs extracted and purified from rat catecholaminergic PC12 cells as described below were used to treat rats. Rats were stereotactically (David Kopf Instruments) implanted with indwelling bilateral cannula targeting the locus coeruleus. After one week of recovery, implanted rats received a 2 μl infusion twice a day for 3 days with TINEVs (4 μg protein) isolated from infected PC12 cells. EVs from uninfected PC12 cells served as a control. On the fourth day rats were anaesthetized and received an injection 3 μl bromophenol blue through each side of the bilateral DVC cannula (to verify their placement) prior to harvesting brain tissue. Brain tissue was dissected into prefrontal cortex, midbrain, and hindbrain regions for processing for RNA. Fluorescence-assisted cell sorting (FACS) of mouse neurons. Mouse brain samples were processed as described 74 . Briefly, brains were suspended in lysis buffer (0.32 M sucrose; 0.1 mM ethylenediaminetetraacetic acid; 5 mM CaCl 2 ; 3 mM Mg(Acetate) 2 ; 10 mM Tris-HCl pH 8; 1 mM dithiothreitol; and 0.1% Triton X-100) and processed using a 15 ml glass dounce homogenizer (Wheaton, UK) in 5-10 strokes on ice. Samples were centrifuged at 100,000xg with a two-step sucrose gradient to purify nuclei. Nuclei were stained with primary antibody NeuN (Abcam, EPR12763 1:250) and secondary antibody Alexa-488 (Abcam, ab150077, 1:1000). All nuclei were also stained with Hoechst 33342 (SIGMA, USA). Samples were sorted using the Becton Dickinson FACSAria II system. Extracellular vesicle characterization. Extracellular vesicle purification. TINEVs were isolated from media by ultracentrifugation as described 75,76 . For all experiments, EVs were isolated in parallel from uninfected cells. Briefly, one day prior to harvest, complete media was replaced with media containing exosome-depleted FBS (Systems Bioscience, CA). Cell cultures of uninfected and infected cells (five days following infection) were harvested and centrifuged at 3000×g for 10 min at 4 °C. The supernatant was isolated and centrifuged at 160,000×g for 2 h in a Type 60 Ti fixed angle rotor at 4 °C. The pellet was resuspended in one ml of 90% sucrose. Eleven layers of one ml of sucrose 70%-10% w/v were then added and centrifuged at 70,000×g for 16 h at 4 °C in a Type 60 Ti fixed angle rotor. Supernatant was collected in two ml fractions. PBS (8 ml) was added to the fraction 1.1-1.2 g/ml sucrose with a density corresponding to that of EVs and centrifuged at 160,000xg for 70 min at 4 °C to recover purified EVs. Methods and data on purification and characterization were input to the EV-TRACK knowledgebase (EV-TRACK ID: EV220363) (https:// evtra ck. org/). UV treatment. EVs were isolated as described. Prior to UV treatment, DBH down-regulation by TINEVs was verified following incubation of noradrenergic cells for 24 h. TINEVs were exposed or mock exposed to UV light at 302 nm for 5 min at 4 °C using a transilluminator (Syngene, Cambridge). Dot blotting. Dot blotting was performed using the Exo-Check Exosome Antibody Array (Systems Bioscience, CA) as per manufacturer's instructions. Briefly, 500 μg protein equivalent of purified EVs isolated from PC12 culture were lysed and ligated to horseradish peroxidase (HRP) overnight at 4 °C. Ligated protein was then incubated with the antibody membrane for two hours at room temperature. After three washes SuperSignal West Femto Chemiluminescent Substrate kit (Thermo Scientific, UK) was used as per manufacturer's instruc- Transmission electron microscopy (TEM). EVs were purified as described above. Electron microscopy was performed as described 77 . Freshly isolated EVs were resuspended in cold Dulbecco Phosphate-Buffered Saline (DPBS) containing 2% paraformaldehyde. EVs were mounted on copper grids, fixed with 1% glutaraldehyde in cold DPBS for 5 min at room temperature, negative stain was performed with uranyl-oxalate solution at pH 7 for 5 min, and embedded with methyl cellulose-UA for 10 min at 4 °C. Excess cellulose was removed, and samples were dried for permanent preservation. Electron microscopy was performed in the University of Leeds facility using a Titan Krios 2 electron microscope. Analysis and extracellular vesicle diameter were measured using ImageJ software. Molecular techniques. DNA extraction. Classic phenol-chloroform method extraction was performed as described in 78 Nuclear run-on. Nuclear run-on was performed as described 79 Methylation sensitive restriction enzyme (MSRE) quantitative PCR. DNA was extracted from cell cultures and isolated as previously described. DNA concentrations were confirmed using a Nanodrop spectrophotometer. Samples were divided into reference and test samples. Test samples were digested with the highfidelity methylation sensitive restriction enzymes (MSRE) HpaII, MaeII and SmaI (TaiI) in the Tango buffer (Thermo Fisher Scientific, USA). These MSREs were chosen based on the restriction enzymes sites present in the DBH upstream region. Reference samples were mock digested. Digestion was performed by heating to 25 °C for 30 min, 37° for 30 min and 65° for 30 min. RT-qPCR was then performed. Percentage methylation calculated based on the difference in Cq values between test and reference samples. Digestion was confirmed with methylated and unmethylated human DNA standards provided with the OneStep qMethyl kit (Zymo) as per manufacturer's instructions. Additionally, primers and MSRE digestion was confirmed with MSRE-PCR. qPCR was performed using SYBR®Green Real-Time PCR Master Mix (Thermo Fisher). Directional RT-PCR for DBH lncRNA screening. Three single-stranded forward (i.e. sense) directional primers were designed to detect antisense RNA and are labelled RT-Primer 1, RT-Primer 2, and RT-Primer 3 (Extended Data Fig. 6A). The RT-designed primers are located (1) in the first exon, (2) near the transcription start site, and (3) − 378 bp upstream of the DBH coding sequence. Four pairs of PCR primers were then designed to amplify products from the RT primer synthesized-template to screen for the presence of antisense lncRNA (Extended Data Fig. 6A). A list of primers used is shown, Supplementary Table 2. Total RNA was extracted from T. gondii infected-PC12 cells that were harvested on day five post-infection using the Direct-zol RNA MiniPrep Plus kit (Zymo Research, USA) according to the manufacturer's instructions. DNase I treatment was implemented as suggested by the manufacturer. In addition, to ensure that all trace amounts of DNA contamination were removed, the eluted RNA samples were treated again with DNase enzyme using TURBO DNA-free™ kit (Invitrogen, USA) according to the manufacturer's instructions as in earlier studies 37 www.nature.com/scientificreports/ with RT-primer 1 to 3 with a Maxima H Minus First Strand cDNA synthesis kit (Thermo Scientific, USA) following the manufacturer's instructions. For each of the samples, there were several controls including priming the cDNA synthesis with random hexamer primer, a negative control reaction to assess the gDNA contamination in the RNA sample containing all components for RT-PCR except the Maxima H Minus enzyme mix, and no template control to assess reagent contamination. The mixtures were incubated as follows: 25 °C for 10 min, 50 °C for 30 min, 65 °C for 20 min and then the reaction was terminated by heating at 85 °C for 5 min. The reaction products served as templates for PCR with GoTaq ® G2 Hot Start Master Mix (Promega, USA), 300 nM forward primer, 300 nM reverse primer, and 2 μl template DNA. Thermal cycling was 3 min at 95 °C, followed by 30 cycles of 95 °C for 30 s, 57 °C for 30 s, 72 °C for 20 s and final termination at 72 °C for 5 min in a thermocycler (Applied Biosystems, USA). All PCR products were resolved and visualized by 1.5% to 2% w/v agarose gel electrophoresis. For DNA sequencing, the specific band was gel-excised and gel-extracted using QIAQuick Gel Extraction Kit (Qiagen, Germany). The eluted DNA concentrations were then measured by the NanoDrop spectrophotometer (Thermo Fisher Scientific, UK). The PCR products were subcloned into a TOPO TA cloning vector (PCRTM4-TOPO-Invitrogen TA cloning kit, Thermo Fisher Scientific, UK) and transformed into XL-10 Gold ultracompetent cells (Agilent Technologies, USA). Two clones bearing inserts of each sample were sent for Sanger Sequencing to Genewiz, UK. Global methylation. Global methylation was measured using a colourimetric, ELISA adapted method, Availability of data and materials All data generated and analysed during this study are included in this published article [and its supplementary information files]. The raw datasets from the current study are available from the corresponding author on request.	7,375.2	2020-11-27T00:00:00.000	[ "Biology" ]
Contact-engineered reconfigurable two-dimensional Schottky junction field-effect transistor with low leakage currents Two-dimensional (2D) materials have been considered promising candidates for future low power-dissipation and reconfigurable integrated circuit applications. However, 2D transistors with intrinsic ambipolar transport polarity are usually affected by large off-state leakage currents and small on/off ratios. Here, we report the realization of a reconfigurable Schottky junction field-effect transistor (SJFET) in an asymmetric van der Waals contact geometry, showing a balanced and switchable n- and p-unipolarity with the Ids on/off ratio kept >106. Meanwhile, the static leakage power consumption was suppressed to 10−5 nW. The SJFET worked as a reversible Schottky rectifier with an ideality factor of ~1.0 and a tuned rectifying ratio from 3 × 106 to 2.5 × 10−6. This empowered the SJFET with a reconfigurable photovoltaic performance in which the sign of the open-circuit voltage and photo-responsivity were substantially switched. This polarity-reversible SJFET paves an alternative way to develop reconfigurable 2D devices for low-power-consumption photovoltaic logic circuits. Emerging reconfigurable technologies based on two-dimensional (2D) materials are expected to empower next-generation electronic devices with lower power consumption and higher expressive capability [1][2][3][4][5] . Different from conventional complementary metaloxide semiconductor (CMOS) technologies, 2D-materials-based transistors show the dopant-free polarity control that the conduction mechanism reversibly switches between n-type and p-type operation modes under gate-voltage sweeps [6][7][8] . These transistors are regarded as Schottky junction field-effect transistors (SJFETs), in which the carrier injection is determined by the Schottky barrier (SB) formed at the metal/semiconductor interfaces (MSIs) 9,10 . However, this ambipolar behavior renders the SJFET hard to be switched off properly 11 . Especially in downscaling of 2D SJFET, the strong electrostatic control can shrink the off-state SB, resulting in a larger leakage current and a smaller current on/off ratio, which hampers its application towards the large-scale integration and low static power consumption 12,13 . Contact engineering as an effective modulation method has been proposed to suppress ambipolarity under gate-voltage sweeps and achieve an ultra-low off-state current in the SJFET [14][15][16][17] . The ambipolarto-unipolar conversion of transport polarity control in 2D SJFET has been realized by introducing dual gates under the source and drain, respectively, where the injection and conduction of carriers can be individually modulated 4,18 . But the multi-gate layout in integrated circuits impeded device scaling due to gate-length limitations. The insertion of an atomically thin insulator layer such as hexagonal boron nitride or the air gap in MSI was also effective in improving the interface quality and suppressing the drain leakage 19,20 . The tunneling conduction mechanism, however, inevitably decreases the on-state current density. An easier method to build unipolar SJFET with an uncompromised on-/off-state current ratio is to use asymmetric source/drain electrodes with different work functions 21,22 . However, the strong Fermi-level pinning induced by the metal-induced gap states (MIGs) and trap-induced gap states between deposited metal/ 2D semiconductor interfaces usually generates an unpredictable SB height and brings the uncertainty to control the unipolarity of SJFETs 21 . Emerging layered semimetals/metals provide a state-of-the-art approach for building the MSI to achieve ambipolar-to-unipolar conversion in van der Waals (vdWs) layered SJFETs. The bond-free integration is unaffected by lattice mismatch, or defects induced in metal deposition process to avoid trap states in the MSI 7 , making the SB height more controllable. Furthermore, semimetals with a near-zero density of states at the Fermi level have been verified to avoid MIGs and achieve the ideal MSIs 23 . Graphene as a typical Dirac semimetal, whose Fermi energy can be effectively tuned by electrostatic gating has the potential to build the reconfigurable barrier transistor 24 . Beyond graphene, the family of transition metal dichalcogenide (TMD) also furnishes a rich variety of semimetals, e.g., 1T′-MoTe 2 25,26 , 1T′-WTe 2 27,28 , 1T ′-PtSe 2 29 , and 2H-NbSe 2 30 . These semimetals possess a broad range of work functions and are expected to create predictable and high-quality all vdWs Schottky junctions 31,32 . Here, we reported on a runtime reconfigurable WSe 2 SJFET with epitaxially-grown WTe 2 and mechanically exfoliated multi-layer graphene (MGr) contacts, in which the WTe 2 contact effectively suppressed the carrier injection to realize the ambipolar-to-unipolar polarity conversion controlled by the single bottom gate. Because carrier injection only allowed tunneling from the MGr contact, the WSe 2 SJFET shows an alternative carrier polarity between n-type and ptype, by applying positive and negative source-drain voltage V ds , respectively. The SJFET under both p-type and n-type unipolarity conditions suppressed the leakage currents to 2 × 10 −10 μA/μm, while the controllable I ds on/off ratios with a maximum of 10 6 retained. The static power consumption induced by the leakage off-state I ds current was suppressed to 10 −5 nW. The SJFET also worked as an electrically gate-tunable Schottky rectifier with a near-unity ideality factor of~1.0 and a high rectifying ratio of 3×10 6 . The asymmetrically contacted SJFET showed a reconfigurable photovoltaic performance with the open-circuit voltage V oc substantially tuned from 0.29 V to −0.47 V and the self-powered photoresponsivity markedly tuned from 61.7 to −12.7 mA/W. Both the negatively and positively gate-biased asymmetric photodiodes showed high filling factors with a maximum of 0.68, indicating larger shunt resistance and smaller leakage. As a facile design method, the WTe 2 /MGr contact strategy is also applicable to other 2D materials such as WTe 2 /MoS 2 gate-tunable n-type Schottky diode, to boost 2D reconfigurable SJFETs in applications towards lowstatic-power-consumption and run-time reversible photovoltaic electronics. Results and discusssion The conversion from ambipolarity to reconfigurable unipolarity The ambipolarity of the SJFET is attributed to the energy level alignments and evolution at both the source and drain Schottky contacts. Figure 1a shows the schematic of the SJFET with intrinsic ambipolar transport polarity, with its band alignment evolution and the corresponding ambipolar transfer curves schematically shown in Fig. 1b. According to the general theory of the Schottky barrier based on 2D semiconductors 9 , there are two main types of carrier injection mechanisms: (1) thermionic emission (TE) when the gate voltage V g is smaller than flat band voltage V FB , which is given by Eqs. (1) and (2): where is the φ B barrier height, ψ s is the surface potential, γ is the inverse band movement factor and calculated by the semiconductor capacitance C S , the interface trap capacitance C it , and the oxide Fig. 1 \| The transition from ambipolarity to unipolarity behavior of the Schottky junction field-effect transistor (SJFET). a Schematic of global gated SJFET with symmetric drain (D) and source (S) contacts. b Schematic of ambipolar transport behavior and corresponding band structure diagram. Both n-and p-branch on-state currents are attributed to the tunneling current due to the barrier reduction and thinning. TE represents thermionic emission. DT represents directing tunneling. c Offset contact geometry enabling p-type and n-type transport behavior and the corresponding band structure diagram. d Suppression of leakage currents by utilizing WTe 2 bottom contacts. The red and blue circles represent the hole and electron, respectively. The arrows represent the carrier injection. V g and V ds are the gate voltage and drain-source voltage, respectively. I ds is the drain-source current. V t is the threshold voltage. capacitance C ox . (2) thermally assisted Fowler-Nordheim tunneling (FNT) and direct tunneling (DT) when the V g exceeds the V FB , which is given by Eqs. (3) and (4): where is the f (E) is the Fermi-Dirac distribution of the contact metal, M 2D (E) is the number of 2D conduction modes in the semiconducting channel, and the T WKB (E) is the SB transmission probability, and the λ SB is the characteristic length. As shown in Fig. 1b, when a positive V g > V FB is applied, the ψ s increases and the energy-band diagram is bent upwards to enable an electron tunneling current I tunnel from source to drain, which significantly contributes to the on-state current I on on the n-branch. Reversely, when a negative V g < V FB is applied, the ψ s decreases, and the downward-bent band induced hole injection from the drain through direct tunneling, which contributes to the I on on the p-branch. By setting V g at zero, the DT current is suppressed, and only the lower TE current contributes to the off-state current due to the larger width of the contact barrier. Hence, when V g exceeds the V FB , both n-and p-I on are dominated by the I tunnel through the narrowed barrier width. To achieve the ambipolar-to-unipolar transition, we proposed an offset contact strategy to suppress the carrier injection from the source by utilizing the shielding effect of the bottom electrode to avoid electrostatically ψ s shifting (Fig. 1c). Firstly, we designed an SJFET with the MGr offset contacts, as shown in Supplementary Fig. 1a, in which the source and drain were contacted at the bottom and the top surfaces of the WSe 2 channel, respectively. Supplementary Fig. 1b-c show the transfer curves of the WSe 2 FET at various V ds with the offset contacts. When V ds < 0, the WSe 2 transistor showed a p-type transport polarity with an on-state I ds of 0.18 μA/μm. Reversely, the WSe 2 transistor showed an n-type polarity when V ds > 0. The output curves at negative and positive V g also indicated that the back-to-back junctions were asymmetrically modulated, as shown in Supplementary Fig. 1d. However, the carrier injection from the source side was only suppressed slightly, as shown in Supplementary Fig. 1e-g, which induced a higher off-state I ds of 0.4 × 10 −2 μA/μm and a lower on/off ratio (~10) for the MGr-contacted WSe 2 SJFET. This weak ambipolar-to-unipolar transition was hard to be simply explained by the air-gap-induced barrier widening 20 . The contact interaction between the WSe 2 and the bottom contact also played an important role to repress the leakage current, which will be discussed in Part III. According to our previous report based on the MGr/MoS 2 /WTe 2 vertical junction 33 , WTe 2 possessing an appropriate work function and weak interlayer interaction with WSe 2 is expected to enlarge the tunneling width of SB at WTe 2 /WSe 2 interface. Therefore, we can optimize the bottom contact material by utilizing WTe 2 to suppress the leakage currents and fulfill the ambipolar-to-unipolar transition by only allowing carrier injection from the top contact side, as shown in Fig. 1d. Preparation and electrical characterization of WTe 2 contacts To obtain a high-quality WTe 2 bottom electrode, we proposed a onestep epitaxial growth method to prepare large-size WTe 2 . Figure 2a shows the photograph of epitaxially-grown WTe 2 in which thicker MoTe 2 flakes were synthesized first and worked as growth seeds to provide nucleation sites and reduce the nucleation energy barrier of WTe 2 . Compared to direct WTe 2 growth, the epitaxial growth of WTe 2 showed a large-size and high-quality surface and clear W(Mo)Te 2 interface, advantageously reducing the unexpected Fermi level pinning and controlling the contact barrier precisely. More information about the sample growth and characterization is in the Methods section and Supplementary Note 1. Figure 2b-d show the Raman intensity mapping indicating the heterostructure properties and excellent uniformity of MoTe 2 and epitaxial WTe 2 . The surface quality of the WTe 2 was probed by atomic force microscope (AFM), as shown in Fig. 2e. Both the thin WTe 2 at the edge and the thick MoTe 2 possessed smooth surfaces, which were expected to work as the vdWs Schottky contact with a clean and desired interface. As a promising candidate to build the semimetal-semiconductor junction, the potential and resistance of epitaxially-grown WTe 2 were further investigated. Figure 2f shows the surface-potential image of WTe 2 measured by Kelvin probe force microscopy (KPFM). The edge WTe 2 possessed a higher surface potential compared to the thick MoTe 2 region and the difference in potential between MoTe 2 and WTe 2 was 79 meV. The work function (WF) of WTe 2 was~5.17 eV by using Au film (WF Au = 5.1 eV) as the reference (Fig. 2g). Figure 2h and i show the I d -V d curves of W(Mo)Te 2 with various thicknesses characterized by 2-terminal and 4-terminal methods, respectively. The measurements are detailed in Supplementary Fig. 5. Since the 4-terminal resistance (R 4T ) eliminated the contact resistance, it was used to assess the intrinsic electrical resistance of the WTe 2 (The details were discussed in Supplementary Note 2). The WTe 2 with a thickness t of~5 nm showed the highest resistance of about R 4T = 9.41 kΩ. For the sample with thickness between t = 5-20 nm, the resistance decreased to 1.95 kΩ and the bulk sample (t > 20 nm) possessed the lowest conductivity of 0.58 kΩ (Fig. 2i). Compared to the R 4T , the 2-terminal resistance (R 2T ) showed a more pronounced change with thickness, indicating the contact resistance 2R contact = R 2T -R 4T between the transferred Au film and W(Mo)Te 2 increased with the decreased thickness, as shown in the inset of Fig. 2j. Further, the current density (V ds = 0.1 V) of WTe 2 devices at different temperatures was measured by the 2-terminal method (Supplementary Figs. 6-7). The current density of WTe 2 with a small thickness of~5 nm was positively correlated with the temperature. As the thickness increased, the WTe 2 exhibited a weak temperature dependence. However, the WTe 2 with a thickness larger than 20 nm showed a negative temperature coefficient of current density, as shown in Supplementary Fig. 7b-d. This metal-semiconductor transition was consistent with exfoliated W(Mo)Te 2 flakes in previous reports 34 , which may be attributed to contact resistance and the surface absorption of the hydroxyl group in ambient conditions. The transfer curves of WTe 2 with varied thicknesses also support this transition (Supplementary Fig. 7e-g). The bulk sample exhibited a constant source-drain current I ds when the gate-voltage V g swept, but the I ds measured in the thin WTe 2 were modulated by gate-voltage steadily, showing a weak p-type characteristic, especially in the lowtemperature range. Carrier injection capability comparison between MGr and WTe 2 To compare the carrier injection capability of different contact materials and geometries, we built the WSe 2 -SJFET using symmetric top and bottom contacts with the exfoliated MGr and epitaxially grown WTe 2 . All devices were fabricated by the dry-transfer method to avoid the formation of defect-induced states. Figure 3a, b show the schematics and transfer curves of the top-contacted device at various V ds . Compared to the FET with top WTe 2 contacts, the FET with MGr top electrodes showed symmetric transfer characteristics with higher currents, indicating the MGr possessed higher carrier injection efficiency for both electrons and holes, which was the reason why the MGr used as the top contact. We also measured the transfer curves of the FET with other bulk or layered metal contacts, but most of these contacts showed an asymmetric carrier injection efficiency and resulted in a stronger p-branch in the I ds -V g curves, as shown in Supplementary Fig. 8. Figure 3c, d show the transfer curves of the bottomcontacted FET at varied V ds . The FET with MGr bottom contacts still showed a high and symmetric on-state I ds . Reversely, I on of the WTe 2 -Article https://doi.org/10.1038/s41467-023-39705-w contacted WSe 2 device was only~10 −9 μA/μm at V ds = 1 V, which was six orders of magnitude lower than that of the MGr-contacted FET, the same behavior as shown in the output characteristics ( Supplementary Fig. 9b, c). The results suggested that the WTe 2 bottom contacts exhibited a lower carrier injection efficiency due to the self-shielding effect of bottom contact 14 , weak interfacial interaction, and vdWs-gapinduced tunneling barrier at the WTe 2 /WSe 2 interface. To explain the contact-geometry-induced suppression of carrier injection, we show the schematic of the current flow pathways in a typical MS surface contact geometry in Fig. 3e−i, where the current flow from metal (A) to channel (D) passes through two regions including the vdWs gap at MS interface (B) and the WSe 2 overlapped with metal (C). We further reduce the surface contact region of WSe 2 FETs into a resistor network under the diffusive approximation, and the contact R C is expressed in the transmission line model 35 : where ρ SC is the sheet resistance of the 2D semiconductor beneath the contact, r c is the specific resistivity of the MS interface, L c is the contact length, respectively. For the top contact geometry, modulated by the global bottom-gate, ρ SC was decreased as the amplitude of V g increased, which reduced the R C and improved the on-state currents. However, for the bottom contact geometry, ρ SC was hardly tuned by the bottom gate due to the shielding effect of the electrode, resulting in a large contact resistance and a poor on-state current density. The shielding effect of the bottom electrode was also verified by simulation using the COMSOL Multiphysics package, as shown in Supplementary Fig. 10, in which both the electric field and carrier density of the WSe 2 atop the bottom contact were hardly to be modulated by V g . Besides, the vdWs gap between the channel and the vertical side wall of the bottom contact ( Supplementary Fig. 10b-o also led to the large and nonadjustable contact resistance due to the large interface resistance r c 36 , which was discussed in the previous report 20 . The same tendency could be derived from the schematic energy-band diagrams of the MS structures. As shown in Fig. 3h, the width of the n-type (p-type) Schottky barrier was narrowed as the V g increased (decreased) in the WTe 2 top contact geometry. In contrast, it is difficult to be modulated in the WTe 2 bottom contact geometry (Fig. 3f). Apart from the contact geometry, the contact materials were also important. Supplementary Fig. 11 shows the potential difference at the WSe 2 /WTe 2 , and WSe 2 /MGr interfaces measured by KPFM. Compared to the potential difference between WSe 2 and MGr, there existed a smaller potential difference of 37 meV between WTe 2 and WSe 2 , indicating that their WFs were horizontally aligned; therefore, WTe 2 had a small charge transfer doping to the WSe 2 and avoided the ψ s shift of the contacted WSe 2 . Meanwhile, for the MGr bottom contact, the electrical contact was dominated by the edge interface of the MGr electrode (Fig. 3j), which was thinner than that of WTe 2 contacts and difficult to suppress the tunneling injecting current. To verify that, the flat-band barrier heights (SHB) of the MGr bottom contact are calculated by 2D thermionic emission mode, as shown in Eq. (6) 9 : where A is the junction area and A* is the effective Richardson-Boltzmann constant. The obtained Φ B as a function of V g is shown in Supplementary Fig. 9, which indicated that the I on of both p-and n-branch was based on the tunneling mechanism due to the gate-thinned barrier. The few-layered WTe 2 also showed a weak interfacial interaction with the orbital overlapping to WSe 2 , compared to the Au film which possessed a similar WF to WTe 2 . We measured the potential difference of WSe 2 on Au and WTe 2 substrate using the WSe 2 on SiO 2 wafer as the reference (Supplementary Fig. 12). The results showed a positive potential difference (~300 meV) between WSe 2 on Au film and the WSe 2 on WTe 2 flake, indicating an unexpectedly strong doping effect of Au film due to the interfacial state, such as metal-induced gap states (MIGS), defect states, and the interface dipoles 37 . Meanwhile, we also compared the intensity and shape-variation of the Raman characteristic peaks of WSe 2 on Au and WTe 2 flake (Supplementary Fig. 13). For out-of-plane vibrational A 1g mode affected by the electrostatic environment change, its full width at half maximum (FWHM) was enlarged as WSe 2 overlapped on Au film compared that of WSe 2 overlapped on WTe 2 ( Supplementary Fig. 13c-f), indicating the strong charger transfer doping effect on Au film 38,39 . The detailed comparison is discussed in Supplementary Note 3. Reconfigurable unipolar WSe 2 SJFET with asymmetric contact To suppress the ambipolar behavior while the high on-state performance retained, we fabricated a WSe 2 SJFET with bottom-contacted WTe 2 and top-contacted MGr electrode, as the drain and source contacts, respectively, as shown in Fig. 4a. The optical images of the devices are shown in Supplementary Fig. 9a and the thicknesses of the MGr, WSe 2 , and WTe 2 were 13.2 nm, 6.2 nm, and 11 nm, respectively. The transistor characteristics were dominated by both source-drain polarity and control. When V ds was positively biased, as shown in Fig. 4b, the WSe 2 SJFET showed an n-type characteristic, and the onstate current (I on ) increased to 6 × 10 −3 μA/μm as V ds increased to 1 V. Meanwhile, the off-state current was suppressed to~10 −10 μA/ μm at V ds = 1 V and V g = −60 V, hence, a maximum on/off ratio higher than 10 6 was achieved. When V ds was negatively biased shown in Fig. 4c, the WSe 2 FET displayed a p-type characteristic with I on of 1.1 × 10 −2 μA/μm at V ds = −1V. Similar to the n-type one, V t increased as the amplitude of V ds decreased, but the leakage current still remained to be below 1.9 × 10 −9 μA/μm at V ds = 1 V and V g = −50 V. Note that the threshold voltage (V t ) shifted with increasing amplitude of V ds , because the strong drain electric field penetrated into the channel region and thinned the barrier, resulting in compromised gate-control capability, which was named by the drain-induced barrier lowing (DIBL) effect 40 . The source-drain current I ds on/off ratio at different V ds was summarized in Fig. 4d, showing that the on/off ratio of p-type SJFET was tuned continuously more than 10 6 and all off-state currents were suppressed at 2.6 × 10 −8 μA/μm. Meanwhile the on-state I ds of SJFET in the offset geometry were not decreased in comparison with that of WSe 2 FET with the symmetric top contact geometry, as shown in Supplementary Fig. 14. The off-state leakage power consumption was calculated by P static = V ds × I ds . When V ds = 1 V, the P static of the n-type FET was 1.8 × 10 −5 nW at V g = −50 V with a high on/off ratio, although the p-type FET shows a higher P static of 2.7 × 10 −4 nW at V g = 50 V. To verify the necessity of the offset contact geometry, we also measured the asymmetric FET in the top contact and bottom geometries, both of which could not simultaneously achieve the ambipolarity to unipolarity conversion (low leakage current) and a high I ds on/off ratio (Supplementary Fig. 15). Meanwhile, we also replaced the bottom electrode with the Au film, as shown in Supplementary Fig. 16. The bottom-Au-contacted FET showed poor reconfigurability, verifying the WTe 2 indeed played an important role in the polarity control. In addition, the effective two-terminal field-effect mobility (μ eff ) for electron and hole on varied V ds were also calculated by Eq. (7): where V g is the applied back gate voltage and C i is the capacitance of the SiO 2 dielectric layer (~11.5 nF/cm 2 ). Figure 4d shows that the μ eff of electron for the WSe 2 FET was almost twice the μ eff of hole for the device with the MGr contacts. Both μ eff of electron and hole were strongly influenced by V ds since the calculated effective μ FET were limited by the contact barrier. To qualify the V ds -induced switching of transport polarity, we measured the barrier heights of the asymmetric SJFET at variable temperatures ( Supplementary Fig. 17). The Φ B-n and Φ B-p of top-MGr contact were obtained from the slope of a linear fit to ln (I ds /T 1.5 ) as a function of 1/k B T, by employing the 2D thermionic emission Eq. (6). Figure 4e, f show that the Φ B-n at V ds = 1 V was extracted at V FB = 17 V to be 79 meV and the Φ B-p at V ds = −1V was extracted at V FB = −3V to be 142 meV, although the p-branch I on was slightly higher than the n-branch I on . The calculated results indicated that both p-and n-I on were mainly attributed to the tunneling currents, hence the barrier width instead of height determined the onstate current density. Figure 4g shows the reconfigurable rectifying behavior of the SJFET with different V g in which the rectifying direction was switched by V g . The maximum rectifying ratio reached 3×10 6 in the positive rectifying direction and 2.5 × 10 5 in the negative rectifying direction, whereas the rectifying ratio of the all-MGr-contacted WSe 2 Schottky diode was only about 10 at V ds = ±1 V. To further evaluate the rectifying performance of the gate-tunable WSe 2 Schottky junction diode, an ideality factor (n) was estimated at a small forward bias (here is 0.02-0.35 V) by fitting to Schottky diode Eq. (8). where I ds , I s , V ds , and V T denote the drain current, reverse leakage current, drain voltage, and thermal voltage, respectively. As the gate voltage swept from positive to negative in Fig. 4h, the ideality factor n derived from the parameters of the fitting equation was nearly fixed on 1 with negligible variation, indicating a near-ideal diode attribute in the reconfiguration process. Figure 4i summarizes the reported ideality factors and off-state current of the 2D SJFET, indicating the high quality of the asymmetric contacted SJFET and the application potential towards lower static power dissipation. The reconfigurable rectifying operation was based on unpinned energy level at the MGr/WSe 2 interface and the strong carrier-injection suppression capability of WTe 2 . As shown in Supplementary Fig. 18, when V g > 0 at V ds > 0, the gate-electric field induced strong electron accumulation and reduced the ψ s of the MGr-contacted WSe 2 . Hence the SB width was thinned to promote the electron injection from MGr through the DT (Supplementary Fig. 18b). In contrast, when V g < 0, the width of barrier at WTe 2 /WSe 2 interface remained constant due to the shielding effect, which reduced the off-state hole current leakage ( Supplementary Fig. 18c). Reversed carrier injection process happened at V ds < 0 ( Supplementary Fig. 18d-f), only holes were allowed to be injected from the MGr side when V g < 0. The WTe 2 contacting strategy can also be applied to fabricate the reconfigurable MoS 2 SJFET (Supplementary Fig. 19). The SJFET with Au/WTe 2 contacts showed a gatetunable rectifying characteristic with rectification ratios ranging from 1 to 10 5 . Compared to similar transport curves of Au/MGr contacted SJFET at V ds = ±1 V ( Supplementary Fig. 19d and e), the transport behavior of Au/WTe 2 contacted device was determined by the sign of V ds , because the electron injection from WTe 2 was inhibited, as shown in Supplementary Fig. 19c. Gate-tunable photo-response of the SJFET Because the SJFET is regarded as equivalent to two back-to-back Schottky junctions at the asymmetric contact interfaces, the photoresponse was tuned by both the V ds and the V g . We used two devices to investigate the photo-response and the optical image are shown in Supplementary Fig. 10a, b. The data in Fig. 5a-c were derived from sample 2#. We first investigated the photocurrent I p and photoresponsivity at positive and negative I ds when V g = 0 ( Supplementary Fig. 20), which showed a nearly linear increase with the laser power intensity. More details of power-dependent photo-response are shown in Supplementary Note 4. More importantly, the SJFET also showed a potential as a self-powered photodetector due to its tunable photovoltaic performance. Supplementary Fig. 20c shows the powerdependent temporal short-circuit current I sc at V g = 0. The I sc was slightly lower than the photocurrents at the same power density, but the photovoltaic response had a smaller dark current and a lower power consumption since V ds was not required. The photovoltaic responses of the WSe 2 FET were further tuned by V g . Figure 5a, b shows the output curves on varying positive and negative V g under the same laser power series (P in = 3.5 mW/cm 2 , and a wavelength of 635 nm). With V g positively increasing, I sc and opencircuit V oc gradually entered the saturated region. The negative V g modulated behavior was similar. The gate-tunable V oc and I sc were summarized in Fig. 5c, showing that the V oc was tuned from 0.29 V to −0.47 V and the self-powered responsivity R sc was tuned from −12.7 mA/W to 61.7 mA/W. Figure 5d, e shows that regardless of the positive and negative gate bias, I sc and output electrical power density showed an exponential increase with light power intensity, and V oc also monotonically increased (Fig. 5f). Hence the power conversion efficiency η PV calculated by η PV = P out /P photo was almost fixed at 0.37% at V g = 60 V and 0.15% at V g = −60 V, respectively, although the incident power increased by two orders of magnitude. Although the output electrical power density and V oc were effectively modulated through the gate control under varying laser power density, I sc showed a weak gate-tunable capability, in which the corresponding self-powered responsivity reached 30 mA/W at V g > 10 V and 1 mA/W at V g < −10 V, respectively. The filling factor (FF) qualifies how closely a photovoltaic device acts as an ideal source. Figure 5g, h show the gate-modulated FF of the SJFET at varying incident power density due to the change of V oc . As the amplitude of V g increased, the SJFET yielded an increased FF, reaching 0.60 at V g = −60 V and 0.68 at V g = 60 V. Supplementary Table 1 shows a photovoltaic performance comparison among the asymmetric contacted WSe 2 SJFET and the previously reported photovoltaic devices, which implies the high photovoltaic performance of WSe 2 SJFET at both positive and negative V g . The reversible photovoltaic performance rendered the asymmetric SJFET to work as the self-powered logic inverter at an ambient light level, as shown in Fig. 5i and Supplementary Fig. 20d, with V g as the input signal and V oc as the output signal. Even at low illuminance level (P in = 0.1 mW/cm 2 ), the logic inverter still showed the obvious V oc switch from −0.2 V to 0.28 V, which further decreased the static power dissipation in integrated circuits due to null V ds applied on the SJFET. Conclusions In conclusion, we proposed a contact-engineered SJFET with the reconfigurable polarity and low leakage current, achieved by employing the asymmetrically vdWs semimetal contacts in which the carriers were only injected from the MGr contact and the injection was suppressed at the epitaxially-grown WTe 2 bottom contact. The asymmetrically contacted WSe 2 SJFET in the offset geometry showed the conversion between ambipolarity and unipolarity and the alternative carrier polarity was determined by the drain bias. Meanwhile, the leakage currents were effectively suppressed to 2 × 10 −9 μA/μm and the device showed a controllable I ds on/off ratio with a maximum of 10 6 . The off-state leakage power consumption was reduced to 10 −5 nW (ntype) and 10 −4 nW (p-type) at V ds = ± 1 V. Also, the WSe 2 SJFET also exhibited a reversible rectifying behavior with a maximum rectifying ratio of 3 × 10 6 and an ideality factor of 1. Advantageously from the electrically gate-tuned contact barrier, the drain-engineered SJFET exhibited a runtime reversible photovoltaic performance in which the sign of the photo-responsivity was substantially tuned and the V oc was switched markedly between the −0.47 V and 0.29 V. Furthermore, based on the photovoltage-reversible properties of the photodiode, a logic optoelectronic device was designed to realize the switch between positive situation to negative situation by manipulating the gate voltage. This contact engineering strategy is generally applicable to other 2D materials such as the electrically gate-tunable n-type MoS 2 Schottky diode. The modulation of carrier injection in 2D materials also provides an alternative route to reduce the logic-circuit complexity and promises innovation for the future applications of computational sensors and optical communications. Note: during revision of this manuscript, we became aware of a related work 20 . Methods One-step epitaxial growth of WTe 2. The molten-salt-assisted thermal chemical vapor deposition (CVD) method was used to synthesize WTe 2 . A mixture of 20 mg hydrate (NH 4 ) 6 Mo 7 O 24 ·4H 2 O, (NH 4 ) 10 W 12 O 41 ·xH 2 O (Sigma-Aldrich) and sodium cholate (Sigma-Aldrich) in a mass ratio of 5:5:1 and the SiO 2 substrate was placed in the middle of the heating zone, with a Te lump placed 1 cm away from the substrate. Throughout the growth process, a carrier gas mixture of H 2 / Ar at a flow rate of 10/100 sccm was utilized. The temperature of the heating zone gradually increased to 760-860°C and held for 3-5 min. By using these two mixed hydrates as the precursor, the MoTe 2 /WTe 2 semimetal heterostructures were epitaxially synthesized in a one-step method in which the thicker MoTe 2 flakes were synthesized firstly, then the WTe 2 epitaxially were grown along the edges of MoTe 2 . As the reaction time increased, the interspaces of MoTe 2 frameworks were covered with the polycrystalline WTe 2 to form a continuous MoTe 2 /WTe 2 film, as shown in Supplementary Fig. 2e. More information about the sample growth is detailed in Supplementary Note 1. Material characterization. AFM (Bruker, Dimension Icon) in the tapping mode TUNA mode were employed to measure the thickness of device, while the contact potentials of the different areas were measured via the Kelvin probe force microscopy. Micro-Raman investigation was performed using HORIBA LabRAM HR Evolution system with 532 nm laser excitation (the laser spot was ∼1 μm). The morphology and chemical composition distribution of WTe 2 /MoTe 2 were analyzed by SEM, and XPS (Thermo Fisher Scientific, K-Alpha+). The crystal structure of 2D flakes was characterized by the TEM (FEI Tecnai F200 systems) operated at 80 kV. The TEM sample was prepared using PMMA-supported wet-transfer method. Electrical characterization. The SJFET devices were tested in a Cascade probe station under high vacuum conditions. The electrical measurement was performed through the Keithley 4200 semiconductor characterization system. Electrical conductivity measurements were taken from 340 K to 80 K with a cooling rate of 2 K/min. The dwell time at each test temperature was 10 min. The 635 nm lasers were used for light illumination and controlled by the Thorlabs ITC c Gate modulation of the responsivity (R sc ) at V ds = 0 V and open-circuit voltage V oc , respectively. d, e Output electrical powers at V g > 0 and V g < 0 as a function of drain-source voltage, respectively. P in is the incidentlight intensity. The dashed line indicates the increasing trend of V oc with P in . f Output electrical power at V g > 0 and V g < 0 as a function of incident-light density. The inset shows V oc vs incident power density. the incident power density. η PV is the power conversion efficiency. The standard deviations were used as error bars. g, h Fill factor (FF) at V g > 0 and V g < 0 as a function of the incident power density. The FF increased with the increasing amplitude of V g . i Logic inverter based on the gate switchable photovoltaic performance. The white light power intensity was ranged from 0.1 mW/cm 2 to 30 mW/cm 2 . 4001. The power density was 30 mW/cm 2 . During the measurements, the devices were positioned at the center of the light spot. Data availability Relevant data supporting the key findings of this study are available within the article and the Supplementary Information file. All raw data generated during the current study are available from the corresponding authors upon request.	8,594.6	2023-07-17T00:00:00.000	[ "Engineering", "Materials Science", "Physics" ]
Teaching Specialty Courses Using Educational Applications: Five Teaching Scenarios from Technical Vocational Education The main objectives of this study are to examine the teaching of specialty courses using educational applications and to assess technology-enhanced learning. More specifically, a unit of analysis of 25 teachers from five different subjects developed five teaching scenarios that utilize specific educational applications. Emphasis was placed on the targeted integration of these applications into the learning process and their framing with participatory and learner-centered educational techniques. According to the teachers participating in the research, the most important criteria for selecting an educational application are the relevance of the application to the teaching unit, the structure and content of the application, the ease of use, the possibility of assessing progress with quizzes and interactive tests, and the free access and use of the application by students. Concerning the benefits of using the apps, these focus on differentiating traditional teaching and providing new learning incentives, enhancing student interest, developing digital skills, enhancing collaborative learning, and lifting time and space constraints. The originality of this research lies in the use of specific educational applications for the development of teaching scenarios for specialty courses. Introduction The way a subject is taught is of utmost importance and influences greatly the course of a learning process.By using appropriate teaching methodology and participatory and experiential educational techniques, a teacher can inspire a positive learning climate, activate student participation, and ensure the achievement of the teaching objectives (Papadakis, Kalogiannakis & Zaranis, 2018). Teaching scenarios can also contribute to this direction, as predefined plans that describe in detail the structure and content of an educational process.The teaching scenarios' purpose is to support the teacher in planning and implementing the teaching, considering the lesson's objectives, students' needs, and available resources.More specifically, teaching scenarios, depending on the particular learning situation, can contribute to the planning and organization of teaching by enhancing the implementation of innovative approaches (Burden & Kearney, 2016;Cederquist & Golüke, 2016). Moreover, in technical vocational education and training, teaching must keep pace with the latest technological and scientific developments and be updated according to the situation.This is the only way to achieve the best possible learning outcomes, and educational applications are an essential part of this, as they can qualitatively enhance the teaching experience, provide modern learning opportunities, stimulate student interest, and make a lesson more experiential (Bottentuit Junior, 2020;Christensen & Knezek, 2018). Educational applications are now a common feature of smartphones and tablets.Their use in the learning process is becoming more and more common, especially since the onset of the pandemic (Schmidthaler et al. 2023).Moreover, most students use smartphones and tablets extensively and are highly proficient in using various applications available through these devices (Dias & Victor, 2017).At the same time, many teachers of different disciplines are adopting the use of educational applications. Portable Devices Mobile devices offer a wealth of possibilities for learning.Smartphones, digital music players, pocket computers and personal digital assistants are just a few examples of devices that can be used to facilitate users through sophisticated ways of interaction and communication.The extensive variety of mobile devices and corresponding applications used in the educational setting has transformed the way students enrich their knowledge.As Cui and Wang (2008) point out, smartphones will play a more important role in education.They are powerful tools and feature almost every function of personal computers.Their main advantage is the ability to access the Internet while their availability depends on commercial prices. The fact that mobile learning is becoming commonplace in both secondary and higher education is not an exaggeration (Dias & Victor, 2017;Matzavela & Alepis, 2021).Several studies show that integrating mobile devices into the educational process is important.Mobile learning offers students the opportunity to transcend the boundaries of traditional classrooms, improving the efficiency of the educational process and enhancing their autonomy.In this context, a plethora of English-language applications for smartphones, iPads, and other devices have been developed in recent years (Teodorescu, 2015).Peters (2007) states that "the communication and data transfer capabilities offered by mobile technologies (m-technologies) can significantly reduce the dependence on fixed work and study locations and have the potential to revolutionize the way we work and learn".However, a mobile-connected society creates new challenges in terms of education delivery. All previous categories of mobile devices use different operating systems.Operating systems are the systems that control and coordinate data input, output, and processes in a device.They are the core software.The two most prevalent operating systems in the smartphone and tablet market are iOS and Android.They are followed by RIM's BlackBerry OS, Microsoft's Windows Mobile, Mozilla's Firefox OS, Canonical Ltd's Ubuntu Phone, Samsung's Bada, Hewlett-Packard's WebOS and Nokia's Symbian. Some Technical Characteristics of Portable Devices Screen: Modern mobile devices, mainly smartphones and tablets, have a variety of technical features. Screen Resolution: The number of pixels displayed.The pixel density (ppi), i.e. pixels per inch, is calculated based on the size of the screen. Video Streaming Capability: transfer video data over a network or the Internet. Other Additional Features Portable devices are equipped with a radio receiver, various connection ports, a built-in microphone and speakers, and a battery.Many devices are dust and water-resistant to an extent. Advantages of Portable Devices Mobile devices offer significant benefits (Matzavela & Alepis, 2021).Students can improve their learning outcomes when they have access to schedules, grades, assignments and discussion groups via mobile devices.These devices offer instant access, collaboration and constant presence, making them ideal for modern learning (Dias & Victor, 2017). Access: Mobile devices provide instant and efficient access to information regardless of time and place (Geddes, 2004).The cost of these devices is lower than desktop or laptop computers, making access feasible for more people. Environment: Meaningful learning requires a connection to authentic activities where students can safely experiment and learn from their mistakes (Schank & Cleary, 1995). Cooperation: Mobile devices facilitate collaboration between students regardless of physical location, in real-time, including text and multimedia (Geddes, 2004). Presence: Physical presence or the use of communication applications is critical to the learning process (Geddes, 2004).Klopfer, Squire & Jenkins (2002) definitively identify certain properties of mobile devices that offer unparalleled educational advantages: portability, social interaction, environmental sensitivity, and personalization.Studies unquestionably demonstrate that users value mobile devices for motivation, ease of use, improved writing, understanding of technology, and flexibility. Disadvantages of Portable Devices Mobile devices have significant advantages, but they also have disadvantages that can limit the learning process.The main difficulties include usability and application incompatibility issues (Kukulska & Traxler, 2005).Software is often not fully compatible with hardware, and users report cost, privacy, and security issues (Ally, 2009). According to various researchers, the most common problems include: Battery and data storage issues (Stošić & Bogdanovic, 2013) Device reliability and sound quality Wi-Fi signal loss and connection problems Increased costs of Internet use (Thornton, 2002) Device and application incompatibility Small screens with low readability and limited keyboard Limited storage space In conclusion, mobile devices offer significant advantages and limitations that affect the full exploitation of their potential. The Advantages of Using Educational Applications in the Learning Process Educational applications capture students' interest by ensuring their active participation in the learning process.They provide interactive learning environments with exercises, quizzes and playful activities.They also provide access to a variety of educational materials, such as images, videos and specialized texts, enriching traditional teaching.In some cases, where textbooks are old and the information they contain needs to be updated, educational apps can act as a supplement by updating the content of the curriculum (Christensen & Knezek, 2018).Furthermore, the scoring and feedback system of many applications should not be overlooked, as it directly evaluates students' performance, helping them recognize their potential and enhancing their confidence and autonomy.It enhances collaborative learning experiences and supports informal and distance learning.Mobile phones make all activities easier, better, more fun, useful, and more interesting than traditional learning.Students feel more comfortable (Brown, 2005).Recording progress with scores and grades encourages student interest and active participation.Educational applications have a variety of assessment quizzes, with varying degrees of difficulty, which are tailored to the learning needs and cognitive level of each student. Furthermore, the didactic use of educational applications familiarises students with the use of technology and improves their digital skills, thus preparing them for the challenges of the new digital age (Triantafyllou, 2019).Through educational digital environments, students practice the knowledge and skills they will need to use modern technology in their later professional and social lives (Criollo, Guerrero-Arias, Jaramillo-Alcázar & Luján-Mora, 2021).The findings of Attewell & Webster's (2005) and Attewell's (2005) studies demonstrated that learning using mobile devices significantly enhances students' literacy and digital literacy skills.This allows them to identify their existing competencies.Learning through mobile devices fosters independent and collaborative learning, empowering students to identify their needs for support.It also effectively addresses resistance to using technology, engages reluctant learners, and boosts their self-esteem and confidence.Educational applications remove the limitations of space and time.Students can access knowledge almost anywhere and anytime, providing learning opportunities outside the classroom.This form of learning is ubiquitous and pervasive.It is embedded in the lives of young learners, making use of time that would otherwise be unused (such as when waiting or traveling).It is portable, allowing learning anywhere and anytime, thus offering flexibility and autonomy to learners.It ensures instant communication and allows access to learning even from remote and isolated areas. Finally, Yousuf (2007) concluded that mobile technologies, such as smartphones, have enormous potential for flexible communication, which can be used strategically to support and enhance students' memory. While there are many advantages to mobile learning, it is important to acknowledge the disadvantages compared to traditional learning.Examples include the lack of human contact, the fact that learners are easily distracted by tablet applications such as games, and the pedagogical use of mobile audio has practical limitations as it requires headphones, or else bystanders could be disturbed. Method There is a clear need for more extensive case studies that link the teaching of specific subjects with the use of specific applications.This has been overlooked in recent years, despite several scientific research projects being conducted internationally on the use of educational applications in teaching.This fact was an additional motivation for the present study, for which qualitative and quantitative research was used with the method of purposive sampling.This method has clear advantages, as utilizes the experience and network of acquaintances of the researchers, selects cases typical of the topic under study, and saves time and costs during the implementation of the research (Campbell et al., 2020;Kelly, Bourgeault & Dingwall, 2010).However, purposive sampling has some weaknesses.The subjective judgment of the researchers is inevitable, and the results of the survey may not always be generalizable and representative (Robson, 2010). For the needs of the present study, five different subjects were selected, which are taught in secondary schools of technical vocational education in Greece: Agricultural Sciences, Information and Communication Technologies, Economics, Engineering, and Health Sciences.A purposive sample of five teachers of corresponding specialties, studying at the Higher School of Pedagogical and Technological Education (ASPETE), was selected for each of the above subjects.These teachers were asked to plan the teaching of a unit using a specific educational application as part of the Specialty Course Teaching, they were attending.Thus, a total of five working groups were formed, each consisting of five teachers.The groups were initially asked to select a specific lesson and a teaching unit in their area of expertise.Then they designed a 45-minute lesson for the teaching unit they had chosen, selecting an educational application to be used in the whole learning process.At the same time, they were required to use participatory teaching techniques and activate the students' potential to the maximum. Once the teaching scenarios had been developed by the teachers' working groups, a qualitative research project was initiated with five semi-structured interviews conducted with one teacher from each group.These interviews were designed to capture the participants' views on individual dimensions of the use of educational applications and to prepare the ground for the formulation of a short but comprehensive questionnaire that would follow in the later phase of the quantitative research.The interviews revealed some key findings, which informed the design of the questionnaire.It was important to keep the questionnaire simple, short and comprehensive to ensure it met the study objectives. Results The following are the five teaching scenarios for the five different subjects, as formulated by the teachers' working groups under the guidance of the teacher of the course in Teaching Specialty Subjects. 1ο Teaching scenario -Agricultural Sciences Lesson: Arboriculture, Teaching unit: Citrus trees Teaching objectives: Upon completion of this unit, students will be able to: a) list the tree species belonging to the Citrus family, b) identify the basic morphological and cultural characteristics of Citrus, c) list the basic cultural care of Citrus, and d) appreciate the importance of Citrus as a branch of Horticulture Educational application: PictureThis (Plant Identifier) Lesson design: at the start of the lesson, the students are asked to brainstorm words and concepts associated with the term "citrus".The teacher writes these on the board and then presents an educational video on the main trees belonging to the citrus family.The students are then divided into three working groups and each group is assigned a tree belonging to the citrus family, e.g.orange, mandarin and lemon.They record the basic growing characteristics.The student groups then enter the educational application PictureThis from the tablets in the agriculture lab and record the basic cultural care of the trees in the areas of fertilization, plant protection and irrigation by reading the information and viewing the corresponding pictures in the application.Then, each group presents their findings to the whole class, giving the whole group of students a complete picture of the basic trees belonging to the citrus family.Finally, a short educational video is shown, highlighting the importance of citrus trees as a branch of modern arboriculture. Educational application: ICT Coach Lesson Design: The lesson is conducted in the school's computer laboratory.The teacher presents the hardware part of computers, including the main processing units, input devices such as keyboards, mice and scanners, and output devices such as monitors, speakers and printers.The students are then divided into two working groups.One group is tasked with recording information about the particular features and information about the operation of the input devices, while the other group records information about the output devices.After ten minutes, each group presents the information they recorded about their respective devices to the whole class.Then, with the teacher's guidance, the teams use the tablets in the IT Lab and enter the ICT Coach training application, where they check the correctness of the data they recorded, compared with the data recorded in a corresponding section of the application.Each student then takes a quiz with questions within the application to test their knowledge gained from the teaching of the specific module. 3ο Teaching scenario -Economic Sciences Lesson: Principles of Economic Theory, Teaching Unit: Market Forms. Teaching objectives: Upon completion of this unit, students will be able to: a) list the four market forms; b) identify the basic characteristics of each market form; c) formulate examples of firms and sellers operating in the four market forms; d) realize the importance of market forms in the modern economic environment. Educational application Learn Economics -Tutorial Guide Lesson design: Students are divided into four groups and each group is given a worksheet with one of the four market forms (monopoly, oligopoly, full competition, monopolistic competition).With the help of a slide and after an explanatory discussion, each group must identify the individual characteristics of each market form.Each group then presents to the whole class the market form in which they have identified its characteristics, followed by a discussion.The teacher then asks the students to open their tablets, go to the Learn Economics -Tutorial Guide app and, open the specific tutorial section in the app, and compare the data written there with the data they recorded earlier in their groups.Then, students must complete the relevant quiz available in the app to check their knowledge and receive feedback. 4ο Teaching scenario -Mechanical engineering Lesson: Automotive Systems, Teaching unit: Types of car tires Teaching objectives: Upon completion of this unit, students will be able to: a) describe the structure and properties of tires, b) list the materials used in the construction of tires, c) identify the basic types of tires, and d) realize the importance of tires for the driving safety of a car. Educational application: Tyre Size Calculator Lesson design: The lesson is conducted in the School's Mechanical Engineering Laboratory.In the beginning, a short educational video is shown, in which different types of cars are described, each of which has a different type of tire.This is followed by a class discussion in which the students are asked to describe the tires they have just seen in the video.Then a second educational video is shown which briefly shows how tyres are made and the different materials that make them up.The teacher then divides the pupils into five working groups and assigns each group a type of tire category: winter tires, summer tires, all-weather tires, SUV tires and run-flat tires.On the tablets available in the Mechanical Engineering Lab, the groups open the Tire Size Calculator app and look up the specific characteristics of their assigned tire type with the guidance of the teacher.They then record these characteristics on a worksheet and present them to the whole class.The teacher then presents a PowerPoint file in which the main types of tires are grouped in a table with relevant photographs.Finally, there is a discussion on the importance of tires in driving safety, and a short educational video is shown, showing road accidents that were avoided due to appropriate tires.5ο Teaching scenario -Health Sciences Lesson: First Aid, Teaching unit: Trauma Teaching objectives: Upon completion of this unit, students will be able to: a) define the concept of trauma, b) distinguish between the different types of traumas, c) effectively treat trauma, and d) realize the importance of proper management of trauma. Educational application: First Aid -IFRC Lesson design: The educational technique of discussion is used initially, with the teacher asking students to recall incidents of injury to either themselves or their loved ones.Based on what is said in the discussion, the different types of injuries are recorded, and a PowerPoint file is shown.Students are then divided into working groups of 4-5 people, with each person trying to list the main characteristics of each type of trauma and the ways of dealing with it.The groups then present the results of their work to the whole class, followed by a short educational video by the teacher on trauma.Then, the First Aid -IFRC educational application is used, which the teacher asks the students to download on their tablets in the nursing laboratory of the school unit.The student work groups are asked to enter the application, in the "Trauma" section, and compare the data and information recorded there with the data they had previously recorded themselves.Finally, students must complete the quizzes available in the educational application.This is a fun way to test the knowledge they have gained from the teaching of this module and receive relevant feedback. After the working groups of 25 teachers completed the above teaching scenarios, they were asked to answer a short questionnaire.This was constructed for the needs of this study.Among the 25 participants in this case study, 14 were female and 11 were male (Table 1).In terms of age, 10 of them were between 31-40 years old, 8 between 41-50, 6 between 21-30, and only one between 51-60.Furthermore, 16 of them had postgraduate degrees, while 9 did not.The next step was to identify the key criteria for selecting an educational application.We took the findings from the interviews conducted during the qualitative research phase and relevant literature into account when asking participants to rate the importance of each criterion on a five-point scale (5 = very important, 4 = important, 3 = moderately important, 2 = slightly important, 1 = not at all important).Table 2 clearly shows that the five most important selection criteria are the relevance of the app to the curriculum unit (4.98), the structure and content of the app (4.91), ease of use (4.76), the ability to assess progress (with quizzes and interactive tests) (4.52), and free access and use of the app by students (4.12).Participants were then asked to rate the importance of the benefits of using educational applications in teaching on a five-point scale (5 = very important, 4 = important, 3 = moderately important, 2 = somewhat important, 1 = not at all important).The findings of the relevant literature (McQuiggan, Kosturko, McQuiggan, Sabourin, 2015) and the qualitative research both identified five specific strengths (Table 3). Discussion Educational applications are becoming increasingly popular in the learning process.This is particularly evident since the onset of the COVID-19 pandemic (Menon, 2022).In light of the rapid technological developments taking place in the digital reality, traditional teaching is being enriched with modern educational applications.These applications can make teaching more engaging, interactive, and participatory (McQuiggan, Kosturko, McQuiggan & Sabourin, 2015).Also, educational applications can be integrated into teaching scenarios that, depending on the circumstances, can upgrade the quality of teaching, especially in specialty courses (Burden & Kearney, 2016;Cederquist & Golüke, 2016). In modern times, most schools have the necessary equipment, and teachers can enrich their teaching with appropriate educational applications.As more and more educational applications are created nowadays, it is up to the teacher to decide how to make the most effective use of an educational application, depending on the level of education, the cognitive level of the student population, the type of lesson and the individual teaching techniques.In any case, educational applications do not replace the role of the teacher but function as complementary and auxiliary tools in a learning process.Teaching remains human-centered, and it is the individual teacher's responsibility to decide how to effectively integrate an educational application into the course flow. Moreover, the concerns that arise from the use of smartphones and tablets in the learning process must also be considered (McQuiggan, Kosturko, McQuiggan & Sabourin, 2015).Mobile devices present unique challenges for students and teachers alike.The relative inexperience of teachers and school settings, coupled with the specific characteristics of mobile devices, can lead to distractions in the learning process and instances of students becoming involved in distracted situations.In some cases, there may be inappropriate or malicious use of these devices by individual students.Furthermore, the potential misuse of these devices can result in health issues for students and teachers, such as headaches or visual problems.Therefore, teachers must provide increased supervision and counseling in a supportive context to ensure the safety and well-being of everyone involved. Educational applications in the learning process are a particularly interesting issue and an important field of study in the field of educational research.Therefore, it is crucial to research the individual dimensions of the issue, which concern all those involved in the field of education: teachers, students, parents, education executives, developers of educational applications, and educational policymakers.Mobile applications (apps) are the ideal solution for learners who need to update their information and skills while learning, without the constraints of time and place (Yang, 2022).This case study adds to the existing body of knowledge and provides a foundation for future research in the field of educational app utilization in teaching specialty courses. 2ο Teaching scenario -Information Technologies & ICT (Information Communications Technology) Lesson: Computer Science, Teaching Unit: The hardware part of the computer Teaching objectives: Upon completion of this unit, students will be able to: a) define the hardware part of a computer; b) list the various input and output devices of a computer; c) demonstrate the components of the mainframe and peripheral units; and d) appreciate the individual functions and utility of the hardware part of a computer. Table 1 . The profile of the participating teachers Table 2 . Criteria for selecting an educational application Table 3 . Advantages of using educational applications	5,551	2024-06-25T00:00:00.000	[ "Education", "Computer Science" ]
The low lying scalar resonances in the $D^0$ decays into $K^0_s$ and $f_0(500)$, $f_0(980)$, $a_0(980)$ The $D^0$ decay into $K^0_s$ and a scalar resonance, $f_0(500)$, $f_0(980)$, $a_0(980)$, is studied obtaining the scalar resonances from final state interaction of a pair of mesons produced in a first step in the $D^0$ decay into $K^0_s$ and the pair of pseudoscalar mesons. This weak decay is very appropriate for this kind of study because it allows to produce the three resonances in the same decay in a process that is Cabibbo allowed, hence the rates obtained are large compared to those of $\bar{B}^0$ decays into $J/\psi$ and a scalar meson that have at least one Cabibbo suppressed vertex. Concretely the $a_0(980)$ production is Cabibbo allowed here, while it cannot be seen in the $\bar{B}^0_s$ decay into $J/\psi a_0(980)$ and is doubly Cabibbo suppressed in the $\bar{B}^0$ decay into $J/\psi a_0(980)$ and has not been identified there. The fact that the three resonances can be seen in the same reaction, because there is no isospin conservation in the weak decays, offers a unique opportunity to test the ideas of the chiral unitary approach where these resonances are produced from the interaction of pairs of pseudoscalar mesons. I. INTRODUCTION The rates for D 0 decay into K 0 s and a scalar resonance, f 0 (980), a 0 (980) are measured by the CLEO collaboration in Ref. [1] and Ref. [2] respectively and the rates are relatively large. The f 0 (980) is seen through its decay into π + π − and the a 0 (980) through the π 0 η channel. Related references on the issue can be seen in the PDG [3]. Theoretical work on these decays is scarce and is mostly devoted to issues related to CP violation or D 0 −D * 0 mixing. In Ref. [4] a thorough study is done of the D 0 → K 0 s π + π − reaction and the amplitude is parametrized in terms of form factors, resonance parameters and different couplings, amounting to a set of 33 free parameters, which are fitted to the Belle [5] and BaBar [6] data. The purpose is to have a good amplitude that can be used to determine the D 0 − D * 0 mixing parameters and the Cabibbo-Kobayashi-Maskawa (CKM) angle γ. The aim of the present work is different, we only evaluate the part of the D 0 → K 0 s π + π − amplitude corresponding to a K 0 s and two pions propagating in s-wave, which will show the f 0 (500) and f 0 (980) resonances. In addition we study the D 0 → K 0 s π 0 η amplitude, where the a 0 (980) resonance shows up, and relate it to the former one. However, we show that, by using basic symmetries and the chiral unitary approach to deal with the meson meson interaction in coupled channels, one is able to determine the shapes of the different amplitudes and the relative weight to each other with no free parameters. Hence genuine predictions for the shapes of these amplitudes and the relative weights of f 0 (500), f 0 (980) and a 0 (980) can be made and compared with experiment. The chiral unitary approach for meson meson interaction makes use of the Bethe Salpeter (BS) equation in coupled channels. One takes all possible meson meson channels that couple within SU(3) to certain given quantum numbers and the BS equation guaranties exact unitary. The kernel (potential) for the BS equation is taken from the chiral Lagrangians [7,8] and there is freedom for only some regularization scale in the meson meson loops, which is fitted to the meson meson scattering data. A good agreement with experimental data is obtained up to 1.2 GeV [9][10][11][12][13][14]. One of the consequences of this approach is that the resonances f 0 (500), f 0 (980), a 0 (980) and κ(800) are automatically generated from these potentials and the use of the BS equations. In this way these resonances qualify as dynamically generated states, some kind of composite, or molecular, meson meson states, in the same way as the deuteron qualifies as a bound state of a proton and a neutron and not a more exotic object [15]. The approach not only provides the meson meson amplitudes but has been tested successfully in virtually any reaction where any of the former resonances is produced. The latest test was the study of the B 0 and B 0 s decays into J/ψf 0 (500) and J/ψf 0 (980) which was done in Ref. [16] (a list of different reactions where the former resonances are produced can also be found there), where a natural explanation was given of the observed facts that theB 0 s decays into J/ψf 0 (980), while no signal is seen for J/ψf 0 (500), and theB 0 decays into J/ψf 0 (500) and only a small fraction is seen for the J/ψf 0 (980). The D 0 decay into K 0 s and a scalar resonance, f 0 (500), f 0 (980), a 0 (980) is a privileged case to test the nature of these resonances. Indeed, as we shall see, the three processes are Cabibbo allowed and the rates of production are big compared to those of theB 0 decays into J/ψ and one of these reso-nances, where necessarily one of the vertices, the V cb , is Cabibbo suppressed [17][18][19]. On the other hand, the a 0 (980) has not been reported inB 0 ,B 0 s decays. As one can see in Ref. [16,19], in the decay ofB 0 s into J/ψ one gets an extra ss pair that has I = 0 and does not allow the a 0 (980) production upon hadronization. On the other hand in the B 0 decay into J/ψ one gets an extra dd pair that could lead to the a 0 (980) upon hadronization, but the process is doubly Cabibbo suppressed. It is found there that a signal is seen for the f 0 (500) production and only a small fraction is reported for f 0 (980) production [18]. One should expect also a minor rate for a 0 (980) production in this case and, in fact, this mode of decay is not reported. In the present case the a 0 (980) production is allowed and the rates are large [2]. The fact that we have now weak interactions that allow for isospin violation permit that both the f 0 (980) and a 0 (980) resonances are produced in the same reaction. This is a novelty with respect to strong interactions that are isospin conserving. The present weak decay presents then a new challenge since one can determine the relative weight of production of each one of these resonances in the same reaction, a new situation with respect to what one has in strong interaction reactions. II. FORMALISM The process for D 0 → K 0 s R proceeds at the elementary quark level as depicted in Fig. 1 (A). The process is Cabibbo allowed, the sd pair produces theK 0 , which will convert to the observed K 0 s through time evolution with the weak interaction. The remaining uū pair gets hadronized adding an extraqq with the quantum mumbers of the vacuum,ūu +dd +ss. This topology is the same as for theB s → J/ψss (substituting the sd by cc) [19], that upon hadronization of the ss pair leads to the production of the f 0 (980) [16], which couples mostly to the hadronized KK components. The hadronization is implemented in an easy way following the work of Ref. [20]. One Hence the uū component of Fig. 1 (B) can be written as, uū(ūu +dd +ss) = (M · M) 11 . (3) Next, we rewrite the qq matrix M in terms of meson components, and we have M corresponding to the matrix φ [21][22][23] This matrix corresponds to the ordinary one used in chiral perturbation theory [7] with the addition of 1 √ 3 diag(η 1 , η 1 , η 1 ) where η 1 is a singlet of SU(3), taking into account the standard mixing between η and η ′ . The term 1 √ 3 diag(η 1 , η 1 , η 1 ) is omitted in the chiral Lagrangians because the [φ, ∂ µ φ] structure of the Lagrangians renders this term inoperative. In Ref. [16] the ordinary φ matrix of chiral perturbation theory was also used. Here we consider the full φ matrix of Eq. (4) since we are concerned with physical η plus π 0 production. Hence upon hadronization of the uū component we shall have where we have omitted the η ′ term because of its large mass. This means that upon hadronization of the uū component we have D 0 →K 0 P P , where P P are the different meson meson components of Eq. (5). This is only the first step, because now these mesons will interact among themselves delivering the desired meson pair component at the end: π + π − for the case of the f 0 (500) and f 0 (980), and π 0 η for the case of the a 0 (980). The multiple scattering of the mesons is readily taken into account as shown diagrammatically in Fig. 2. Analytically we shall have 2: Diagrammatic representation of π + π − and π 0 η production. (a) direct π + π − production, (b) π + π − production through primary production of a P P pair and rescattering, (c) primary π 0 η production, (d) π 0 η produced through rescattering. and where V P is a production vertex, containing the dynamics which is common to all the terms. G is the loop function of two mesons [9] and t ij are the transition scattering matrices between pairs of pseudoscalars [9]. The f 0 (500), f 0 (980), and a 0 (980) are produced in s-wave where π 0 π 0 , π + π − have isospin I = 0, hence these terms do not contribute to π 0 η production (I = 1) in Eq. (7). Note that in Eq. (6) we introduce the factor 1 2 extra for the identity of the particles for π 0 π 0 and ηη. The t matrix is obtained as where V ij are the transition potentials evaluated in Refs. [9,24]. Explicit expressions for I = 0 are given in Ref. [16]. We have the I = 1 case new here and we present the matrix elements below with f the pion decay constant, f = 93 MeV, and s is invariant mass square of the mesonmeson system. The loop function G [9] is regularized by means of a cut off. When the ηη channel is explicitly taken into account the cut off needed is smaller than in Ref. [9] and we follow [16] where it was taken equal to q max = 600 MeV. Finally, the mass distribution for the de-cay is given by 1 where pK0 is theK 0 momentum in the global CM frame (D 0 at rest) andp π is the pion momentum in the π + π − rest frame, and similarly for the case of the π 0 η production. Before closing this section we should mention that in a three hadron final state one must look for the interaction of three particles, for which one must in principle deal with Faddeev equations [25]. Most of the applications of Faddeev equations are done for three baryon systems but calculations for three mesons are becoming available [26]. However, for the purpose of the present work it is instructive to follow the idea in Ref. [27] for the analogous D + → K − π + π + reaction. In this work two body unitarity is imposed on the two body systems and diagrams related to three body unitarity are evaluated perturbatively. They are found relevant close to threshold but fade away rapidly of higher energies. What we have done is in this line and we have unitarized the π + π − , π 0 η (and coupled channels pairs) but theK 0 has been left as a spectator. In principle we should also look at the interaction ofK 0 π − which can lead to the κ resonance [10], yet the topology of Fig. 2 (a) does not favor s-wave interaction ofK 0 π − . And furthermore, the κ can also come from a different topology of the diagrams than those considered in Fig. 2 (a) for instance producing a π + meson from the 1 The decay amplitude tD0 →K 0 π + π − depends on the invariant mass, M inv = √ s, of the meson-meson system. c quark via direct conversions of W into π + (see section IV, Fig. 5 (A)). This is why the κ is better seen in the D + → K − π + π + reaction, as discussed in Ref. [27]. We do not consider the πK interaction leading to the κ, with the argument that the κ, being a very broad resonance in the πK invariant mass, only contributes a smooth background below the π + π − , or π 0 η invariant mass distribution when one looks for the f 0 (980) or a 0 (980) signals and is taken into account in experimental analysis of these two latter resonances. In this sense, the diagram of Fig. 1 chosen and the interaction that we have considered is also what corresponds to the K 0 s [π + π − ] s , M 2 amplitude of Ref. [4], the one that considers the s-wave interaction of the pions and the f 0 (500) and f 0 (980) resonances, or the a 0 (980) when we consider in addition the K 0 s [π 0 η] s amplitude. III. RESULTS In Fig. 3, we show the results of our calculation. We have taken the cut off q qmax = 600 MeV as in Ref. [16]. We superpose the two mass distributions dΓ/dM inv for π + π − (solid line) and π 0 η (dashed line). The scale is arbitrary, since it corresponds to taking V p = 1000 in Eqs. (6) and (7), but it is the same for the two distributions, which allows us to compare f 0 (980) with a 0 (980) production. As we discussed before, it is a benefit of the weak interactions that we can see simultaneously both the I = 0 f 0 (980) and I = 1 a 0 (980) productions in the same D 0 →K 0 R decay. When it comes to compare with the experiment we can see that the f 0 (980) signal is quite narrow and it is easy to extract its contribution to the branching ratios by assuming a smooth background (shown in Fig. 3 by the dotted line) below the f 0 (980) peak as a continuation of the f 0 (500) broad structure at lower energies. For the case of the π 0 η distribution we get a clear peak that we associate to the a 0 (980) resonance, remarkably similar in shape to the one found in the ex- The π + π − (solid line) and π 0 η (dashed line) invariant mass distributions for the D 0 → K 0 π + π − decay and D 0 →K 0 π 0 η decay, respectively. A smooth background is plotted below the a 0 (980) and f 0 (980) peaks. periment [2]. Yet it is obvious that not all the strength seen in Fig. 3 can be attributed to the a 0 (980) resonance. One should recall that the chiral unitary approach provides amplitudes, in this case the π 0 η amplitude, but the amplitudes provide poles that one associates to resonances but also background contributions, and this is the case of the π 0 η distribution. In order to get a "a 0 (980)" contribution we subtract a smooth background that we depict by a open dotted line in the figure. By doing that we have a remaining "resonant" shape with an apparent width of 80 MeV, which is in the middle of the 50 − 100 MeV of the PDG [3]. Integrating the area below these structures we obtain where we have added a 20% theoretical error due to uncertainties in the extraction of the background. Experimentally we find from the PDG and the Refs. [1,2], Γ(D 0 →K 0 a 0 (980), a 0 (980) → π 0 η) = (6.5 ± 2.0) × 10 −3 , The ratio that one obtains from there is The agreement found between Eq. (18) and Eq. (21) is good, within errors. This is, hence, a prediction that we can do parameter free. As we mentioned, the explicit consideration of the ηη channel in the meson meson interaction, required to use a cut off q max = 600 MeV [16] to agree with experimental amplitudes, smaller than in Ref. [9] where this channel was omitted. We use the same cut off here. Yet, we want to show explicitly that the ratio obtained does not get spoiled even if a wide range of cut offs are used. In Fig. 4, we show the results for five different, higher values of q max . The magnitude of the a 0 (980) production grows a bit with q max , with the prescription taken above, but the strength of the f 0 (980) production also grows as a consequence of an increase in the width. One can also see that the peak of the f 0 (980) moves to lower energies, what puts constraints on q max , but we see that, even within this broad range of values of q max , the ratio of Eq. (18) remains within the errors of this equation and is a solid prediction. It should not go unnoticed that we also predict a sizeable fraction of the decay width into D 0 →K 0 f 0 (500), with a strength several times bigger than for the f 0 (980). The π + π − distributions is qualitatively similar to that obtained in Ref. [16] for theB 0 → J/ψπ + π − decay, although the strength of the f 0 (500) with respect to the f 0 (980) is relatively bigger in this latter decay than in the present case (almost 50% bigger). TheB 0 → J/ψf 0 (500), f 0 (500) → π + π − decay mode, The π + π − (black curves) and π 0 η (blue curves) invariant mass distributions with different cut off q max for the D 0 → K 0 π + π − decay and D 0 →K 0 π 0 η decay, respectively. together with the f 0 (980) one have been identified in Ref. [28] through a partial wave analysis, and the rates obtained are comparable with the findings of Ref. [16]. Such a partial wave analysis is not available from the work of Ref. [1], where the analysis was done assuming a resonant state and a stable meson, including many contributions, but not the K 0 s f 0 (500). Yet, a discussion is done at the end of the paper [1] in which the background seen is attributed to the f 0 (500). With this assumption they get a mass and width of the f 0 (500) compatible with other experiments. Further analyses in the line of [28] would be most welcome to separate this important contributions to the D 0 → K 0 s π + π − decay. IV. FURTHER CONSIDERATIONS Our results are based on the dominance of the quark diagrams of Fig. 1. In the weak decay of mesons the diagrams are classified in six different topologies [29,30]: external emission, internal emission, W -exchange, Wannihilation, horizontal W -loop and vertical W -loop. As shown in Ref. [31], only the internal emission graph (Fig. 1 of the present work) and W -exchange 2 contribute to the D 0 →K 0 f 0 (980) and D 0 →K 0 a 0 (980) decays. In Ref. [4] the D 0 →K 0 π + π − decay is studied. Hence, only the D 0 → K 0 s f 0 (980) decay can be addressed, which is accounted for by proper form factors and taken into account by means of the M 2 (K 0 s [π + π − ] s ) amplitude, which contains the tree level internal emission, and W -exchange (also called annihilation mechanism). In order to establish connection with the work of Ref. [4], let us draw the external emission and W -exchange diagrams pertinent to the D 0 →K 0 π + π − decay, as shown in Fig. 5. It is also instructive to recall the basic non-leptonic Hamiltonian at the quark level responsible for this transition [32][33][34] This Hamiltonian transforms as an isospin I = 1 operator. Consequently the decay amplitude of D 0 → Kππ is (23) where the two meson system M 1 M 2 (π + π − here) can have I = 0, 1, 2. This is the case in the diagram of Fig. 5 (A) where the cū, π + intermediate state can have I = 1/2, 3/2, which allows the π + π − system to have I = 0, 1, 2 in the finalK 0 π + π − state. However, the diagram of Fig. 5 (A) will not contribute to our resonance production which requires the π + π − S wave loop, as seen in Fig. 2, due to the vector structure of Eq. (22) in the csW + vertex of Fig. 5 (A). This is also the case in the phenomenological analysis of Ref. [31]. Then, in the remaining mechanisms of Fig. 2 and Fig. 5 (B) the π + π − can only be in I = 0 or 1. In our study we have isolated the S wave of the pions in order to get the f 0 (500), f 0 (980) resonances, and the a 0 (980) in the case of π 0 η. Certainly, the operator of Eq. (22) allows other angular momenta, and indeed experimentally ρ meson and other mesons can be obtained, but the experimental analysis of Refs. [1,2] with partial wave analysis separate the contributions of f 0 (980) and a 0 (980) production, which allows us to compare directly with these data without the need to look into other channels. Also, although in principle the amplitudes depend on two independent Mandelstam variables as seen in Ref. [4], the fact that we do not consider theK 0 π − interaction (leading to the κ), which would just provide a background in the π + π − mass distribution for the reasons discussed at the end of section II, makes our amplitude dependent upon the invariant mass of π + π − or π 0 η. Concerning the W -exchange diagrams, which we have ignored in our approach, we would like to argue in favor of its relative smallness with two arguments: firstly, in Fig. 1 (A) we can see that theū quark of the D 0 is a spectator. We thus have a one body operator at the D 0 quark level. However, in the W -exchange one involves the two quarks of D 0 and the amplitude squared involves the probability to find two quarks, smaller than that of finding one quark. This situation is typical in nuclear reactions, where the Wexchange would have its equivalent in the exchange currents [35]. The second argument is that in the W -exchange diagram of Fig. 5 (B) there is a double hadronization compared to the single hadronization of Fig. 1 (B). The hadronization reverts into a decreased rate for two meson production compared to the single meson of the original qq, which we can estimate in about one order of magnitude from the experimental rate [3,36] (see Ref. [16] for details), In the literature there is much discussions about the relevance of the W -exchange mechanism. In Ref. [31] an empirical analysis is done based on giving a weight to the different topological mechanisms, and the Wexchange mechanism (evaluated under the assumption that the f 0 and a 0 resonances are qq or tetraquark states) appears of the same order of the internal conversion, with opposite sign, that makes the C − E combination in a 0 production bigger than the C + E combination in f 0 production. 3 However, in the same paper, a factorization approach is followed (see section V of Ref. [31]) in which the W -exchange contribution is claimed to be suppressed and is neglected in that approach. The present work neglects the Wexchange mechanism and produces a large a 0 (980) production relative to f 0 (980) due to the mechanism of final state interaction. We 3 The C and E are the contributions of the internal conversion and W -exchange, and C − E and C + E the combinations found in Ref. [31] for a 0 and f 0 production, respectively. should note that in both cases, the intermediate production of KK states, and further rescattering to give π + π − or π 0 η in the final states, is a novelty of our approach compared to other approaches and an essential ingredient in the results due to the strong coupling of the f 0 (980) and a 0 (980) resonances to KK. The dominance of the internal emissions in this kind of processes is also supported in other works [19,28,[37][38][39]. In Ref. [4] a detailed discussion is made of results in different works. The W -exchange mechanism in Ref. [4] depends on two unknown form factors which are fitted to the data and a phase which is unknown. From a fit to the data, a minimal strength of about 20% is obtained for the W -exchange mechanism, suggesting that the contribution could be bigger. It is clear that this issue is still open but the relative smallness of the W -exchange mechanism has many arguments in favor, and our study, producing a big ratio of a 0 (980) versus f 0 (980) production due to final state interaction in coupled channels, neglecting the Wexchange mechanism, provides extra support for its smallness. Note that this a 0 /f 0 large ratio was the main reason of the relatively large weight of the W -exchange mechanism in the fit of Ref. [31]. Studies along the lines of Ref. [4] for D 0 →K 0 π 0 η would help bring extra light into this issue. V. SUMMARY AND CONCLUSIONS We have studied the decay of the D 0 decay into K 0 s and a scalar resonance, f 0 (500), f 0 (980), a 0 (980). For this purpose we have identified the weak mechanism that allows the formation of aK 0 , that will act as a spectator, and a pair of mesons, KK, ππ, π 0 η, ηη, etc., that upon interaction will give rise to the f 0 (500), f 0 (980), a 0 (980) resonances. The first step is the production of aK 0 s and a pair of qq, which upon hadronization leads to these pairs of mesons. The hadronization is done in an easy way, by looking at the flavor content in meson meson of the hadronized qq pair. This is sufficient in the present case where we only aim at determining the shape of the invariant mass distributions and the relative weight of the different production modes, but not absolute rates. Once the weight of the differentK 0 -meson-meson components has been determined we then allow these meson-meson components to interact, using for it the chiral unitary approach, and they give rise to the f 0 (500), f 0 (980), a 0 (980) resonances. They are seen in the π + π − invariant mass distributions [f 0 (500), f 0 (980)] and the π 0 η distribution [a 0 (980)], and we not only get the poles of these resonances but also realistic mass distributions that can be compared with experiment. We found the shape of the π 0 η distribution rather similar to the one found in the experiment, and we obtained a ratio of the branching ratios for a 0 (980) and f 0 (980) production in good agreement with experiment, all of it accomplished without any free parameter, meaning that the parameters of the theory have been determined before hand in the study of the meson meson interaction. We emphasized the fact that it is the nature of the weak interactions, that allows for isospin violations, what made possible the production of the a 0 (980) and f 0 (980) resonances in the same decay. This is a most welcome feature that has allowed to test simultaneously the production of the two resonances in the same reaction offering new test for the chiral unitary approach than allowed in strong interaction reactions, providing yet one more example of support for the dynamically generated nature of the low lying scalar mesons.	6,851.6	2014-09-01T00:00:00.000	[ "Physics" ]
Automatic planning of MR-guided transcranial focused ultrasound treatment for essential tremor Introduction Transcranial focused ultrasound therapy (tcFUS) offers precise thermal ablation for treating Parkinson's disease and essential tremor. However, the manual fine-tuning of fiber tracking and segmentation required for accurate treatment planning is time-consuming and demands expert knowledge of complex neuroimaging tools. This raises the question of whether a fully automated pipeline is feasible or if manual intervention remains necessary. Methods We investigate the dependence on fiber tractography algorithms, segmentation approaches, and degrees of automation, specifically for essential tremor therapy planning. For that purpose, we compare an automatic pipeline with a manual approach that requires the manual definition of the target point and is based on FMRIB software library (FSL) and other open-source tools. Results Our findings demonstrate the high feasibility of automatic fiber tracking and the automated determination of standard treatment coordinates. Employing an automatic fiber tracking approach and deep learning (DL)–supported standard coordinate calculation, we achieve anatomically meaningful results comparable to a manually performed FSL-based pipeline. Individual cases may still exhibit variations, often stemming from differences in region of interest (ROI) segmentation. Notably, the DL-based approach outperforms registration-based methods in producing accurate segmentations. Precise ROI segmentation proves crucial, surpassing the importance of fine-tuning parameters or selecting algorithms. Correct thalamus and red nucleus segmentation play vital roles in ensuring accurate pathway computation. Conclusion This study highlights the potential for automation in fiber tracking algorithms for tcFUS therapy, but acknowledges the ongoing need for expert verification and integration of anatomical expertise in treatment planning. . Introduction Transcranial MR-guided focused ultrasound therapy (tcMRgFUS) offers precise thermal ablation of brain tissue.It effectively treats movement disorders like Parkinson's disease and essential tremor (ET) in a minimally invasive manner (Elias et al., 2016;Bond et al., 2017).The specific target area for tremor treatment is the ventral intermediate nucleus (VIM) of the thalamus, which acts as a synaptic "hub" mainly for the contralateral (but to a certain extent also the ipsilateral) cerebellothalamic tracts (CTTs) and the pallidothalamic tracts (PTTs; Gallay et al., 2008).The VIM cannot be localized in standard clinical magnetic resonance imaging (MRI); thus, attempts have been made to develop better thalamic segmentation procedures: either based on functional connectivity, specific magnetic resonance (MR) sequences aiming at improved contrast within the thalamus or using diffusion tensor imaging (DTI)based tractography to provide increased accuracy to localize the VIM (Tourdias et al., 2014;Fasano et al., 2016;Akram et al., 2018;Dallapiazza et al., 2019;Shepherd et al., 2020;Purrer et al., 2021).The therapeutic ablation target, located at the inferior/lateral border of the thalamus, is quite small and necessitates a precision of approximately 0.5 mm.It is crucial to avoid major fiber tracts such as the corticospinal tract (CST) and the medial lemniscus (ML), which are in close proximity to the VIM and the target bundles (the CTT and the PTT), and must not be included in the ablation.Although diffusion tractography is prone to poor image resolution, high noise in the signal and distortion artifacts in sequences (Maier-Hein et al., 2017), fiber tracking algorithms can be utilized to better understand target areas and risk zones in the brain that require careful delineation (Fasano et al., 2016).This information greatly assists in the planning process, both prior to and during the procedure, by establishing safety margins and accurately calibrating the ablation positions according to the patient's unique anatomy. We propose that it is essential to reconstruct the target bundles and the fiber bundles that require protection with utmost accuracy, aligning them with individual morphological images instead of relying solely on atlas-based coordinates.This becomes even more significant due to the ongoing debate regarding whether the most efficient structure to be targeted lies within the VIM or the respective bundles. The problem of qualitative or quantitative evaluation of fiber tracking algorithms is well known.The aim of this article is neither to show that one fiber tracking algorithm is better than the other nor to establish new methods or metrics for comparing different methods.Rather we aim to answer the question whether a full automatic pipeline is possible at all or whether a manual, fine-tuned fiber tracking and segmentation is needed, which is time-consuming and requires expert knowledge of complex open-source neuroimaging tools.To this end, we examine whether fiber tracking results are highly dependent on the algorithm, pipeline or degree of automation used, or whether they are quite robust and, in combination with an automated calculation of the standard treatment coordinate (Jameel et al., 2022), can provide a robust planning basis for essential tremor therapy (ETT). . Materials and methods TcMRgFUS therapy at the University Hospital Bonn is performed within the neurosurgical operating area.The therapy system employed consists of the INSIGHTEC Exablate Neuro system with a helmet-like phased array ultrasound transducer containing 1024 elements operating at a frequency of 660kHz, in conjunction with a 3T MRI scanner (Discovery MR750w, GE Healthcare, Chicago, IL, USA).The patient's head is securely immobilized using a dedicated MR-compatible stereotactic frame affixed to the helmet transducer.T1w overview images are used for co-registration with the preoperative image data.The standard treatment coordinate is manually determined first according to the manufacture suggestions and neurosurgical practice on a manually anterior commissure (AC) posterior commissure (PC) reoriented T1-weighted image as described in the last part of Section 2.5.2.During the treatment, the target region is manually optimized using the coordinates of the tracked fiber bundles on axial, coronal and sagittal images by applying individual, subtherapeutic, lowintensity test US pulses, followed by neurological examination to assess the single sonication effects.The fiber bundles probable locations (relative to the AC/PC and the midline) are used to move the temperature center in the presumed optimal direction (target bundles) and avoid no-go areas (CST, ML).Once a target is identified to be free of side effects when raising the temperature core to 50 • C and gaining (temporarily) tremor control, the region is thermally ablated with higher energy pulses to reach 57 • C to 60 • C. In the following subsections we describe the used data sets, the definition of the used anatomical regions of interest (ROIs) as well as the manual and the automatic planning pipelines.In addition, we describe the metrics we used for the evaluation. . . Data sets The data set used in this study consists of planning, treatment, and follow-up data for 45 patients who underwent tcMRgFUS treatment for ET at University Hospital Bonn.The components of a full patient data set are summarized in Table 1. The full data set was available for 23 patients.For two other patients, one was missing the tracked fibers of the CTT and the other, the ML.For 15 patients, only planning image data (no fiber tracts) and treatment data were available.For the remaining five patients, only planning image data were available. . . tcFUS treatment data The treatment log data was used to estimate the accuracy of the automatically predicted standard coordinate (see Section 2.5.2).We did not consider all ablation points but only those where a maximum average temperature of at least 55 • C was reached, which is considered therapeutic.This can be more than one position per patient.The treatment positions were registered to the T1 planning coordinate system via the registration matrix determined during treatment (automatic registration followed by manual correction of the treating physician). . /fnimg. . . .Anatomical definition of seed, include, and exclude ROIs Table 2 shows the ROIs required for fiber tracking.Although we chose the same anatomical landmarks for the manual and the automatic planning pipelines, only for the automatic pipeline, the precentral gyrus was used as the seed region for both the CTT and the CST.As fiber tracking algorithms are usually non-bijective (start at A and end at B, start at B and do not end at A), it is much easier to track the respective bundle in its entirety with a comparatively much larger seed region and determine the borders of the bundles more precisely.However, the probabilistic approach used in the manual pipeline leads to a large number of different fibers being determined per seed point anyway. . . Manual planning pipeline Aligning the image data along the line connecting the AC and PC not only provides a standardized view but also provides the basis for calculating the standard target coordinate for tcMRgFUS treatment.Therefore, an AC-PC orientation was manually set by reorienting the T1-MPRAGE.These reorientation parameters were also applied to all derived probability maps that represent the fiber tracts (see Section 2.4.2). . . .Preprocessing DTI images were preprocessed using FMRIB software library (FSL) (Smith et al., 2004).First, we corrected for the susceptibility distortions by applying the "topup" toolbox on the reverse phase encoding b = 0 image.This created a field corrected mask that we used to correct for eddy currents and motion using the "eddy" toolbox.Bayesian estimation and crossing fiber modeling along the principal diffusion directions were calculated using the bedpost toolbox in order to perform probabilistic tractography (Hernández et al., 2013).After the standard coordinates for treatment were determined on anatomical T1-MPARGE images, we further preprocessed the images.First, we skull-stripped and non-linearly transformed the individual T1-MPRAGE images of each patient to the Montreal Neurological (MNI) space atlas (Mori et al., 2008;Oishi et al., 2008).We obtained ROIs of different brain structures (see Section 2.3) in MNI 1-mm space in order to use them for fiber tractography.While the ROI of the thalamus was obtained from subcortical segmentation of the T1-MPRAGE of each individual, the other ROIs were non-linearly warped to the individual space to create the tracts in individual anatomical space.We also registered the T1-MPRAGE images to the diffusion space in order to apply ./fnimg. . the transformation matrix to the seed, include, and exclude ROIs for probabilistic tractography. . . .Fiber tracking using FSL Probabilistic tractography was performed using the default parameters of "probtrackx2, " which is part of the FSL diffusion toolbox FDT (Behrens et al., 2007) as exemplified in the study by Ferreira et al. (2021).All fiber tracts were normalized by dividing through total streamline counts reaching from each voxel of seed to target.Furthermore, we applied a threshold to the probability maps to remove false-positive streamlines, details are explained in Section 2.6.Normalized and thresholded tracts were inserted in the statistical parametric mapping pipeline with T1-MPRAGE to localize the tracts. . . Automatic planning pipeline . . . Summary of the automatic pipeline The overall data flow and algorithmic components of the proposed automatic planning pipeline are shown in Figure 1.Details on each step are given in Sections 2.5.2, 2.5.6.In the first step, the T1 image is automatically reformatted to the standardized AC-PC view, from which the geometric standard coordinate for treatment can be derived.For fiber tracking, the T1 image is registered to the diffusion-weighted imaging (DWI) space, where all the following steps are performed.The registration step also includes a deformable atlas registration to the patient data to gain anatomical region labels of the whole brain, which is used for local parameter adaptation during fiber tracking.The DWI data are preprocessed to extract a DTI.The registered T1 image and the DTI image are the input for a DL-based ROI segmentation algorithm that generates seed, include, and exclude ROIs for fiber tracking.These ROIs, the deformed atlas brain regions, and the DTI data are used to perform the final fiber tracking of the target bundles (CTT, PTT) and the no-go areas (CST, ML). . . . DL-based AC-PC alignment and standard target coordinate The AC and the PC (see Section 2.4 for an explanation) were detected using a 3D U-Net (Çiçek et al., 2016).To this end, the landmarks were manually annotated in the ETT cohort using markers.To develop the algorithm, a subset of 30 cases was used for training, 5 cases were used for validation during training, and 10 cases were used for evaluation.For training of the segmentation model, the markers were converted to masks and dilated.The 3D U-Net was then trained to predict the AC and PC marker blobs based on the T1 MPRAGE image.From the predicted masks, markers for the AC and the PC were extracted as the center of gravity of the predicted blob for the AC and the PC.The model was trained using the Dice loss function (Milletari et al., 2016) with learning rate of 10 −4 with a batch size of 5 until convergence of the training loss, while the best-performing parameter state was tracked based on the Jaccard score on the validation data. Using the predicted positions of the AC and the PC, the T1 image was rotated in the axial and sagittal plane so that the AC and the PC were in the same axial plane and on a vertical axis parallel to the voxel grid.To account for large rotations in the coronal plane (around the axis defined by the AC and the PC), the transformed image was registered to an AC-PC-aligned atlas.The rotation within the coronal plane was then applied to the transformed image to generate the final standardized AC-PC view. The standard target coordinate was calculated based on the AC-PC reoriented image: anterior to the PC by 25% of the AC-PC distance, 1 mm superior, 14 mm lateral left/right (depending on the treated side) as typically performed at University Hospital Bonn.Different research groups may use slightly deviating definitions of the standard coordinate and the definition has evolved over time (Jameel et al., 2022). . . . Registration We register all available data sets, including the MNI atlas, that we used fiber tracking and for generating training data to the diffusion-weighted images.T1 data sets are rigidly registered to the diffusion MRI data; thereby, as a distance measure, we used normalized gradient fields.The atlas (for details, see Section 2.5.6) is registered to the diffusion-weighted data in a deformable way: after an initial rigid registration, the atlas is registered in a nonparametric way (distance measure: normalized gradient fields, optimization algorithm L-bfgs) using a multilevel approach. . . . DTI preprocessing We propose to supersample the data to an isotropic target voxel size with a higher order filter.This supersampling does not add any information to the image but allows for a simple trilinear interpolation in the later tracking phase.As proposed by Hahn et al. (2006), we use a Lanczos-3 filter in the spatial domain, which is a good compromise between computational speed and filter accuracy.In addition to spatial resampling, we smoothed the data with a Gaussian filter.Finally, we calculated the diffusion tensors using the well-known Stejskal-Tanner equation. . . . DL-based ROI segmentation All ROIs that we used as seed regions or include/exclude regions for fiber tracking were segmented using DL, as proposed by Hänsch et al. (2022).The segmentation algorithm was developed on a subset of 30 cases for training and 5 cases for validation (the same split as detailed in Section 2.5.2).A total of 24 ROIs (see Section 2.3 for details) is required per patient to generate the target bundle and no-go areas so that the manual annotation of the whole training set would be very timeconsuming.Therefore, the majority of the training segmentations was created automatically via a deformable atlas registration (see Section 2.5.3), which propagates the ROIs that we already computed for the manual pipeline; see Section 2.4.1.For a subset of the ROIs (thalamus, ventricles, pre-and postcentral gyrus), the accuracy of the registered ROIs was found to be unsatisfactory.Therefore, these ROIs were segmented fully manually (thalamus) and using semiautomatic algorithms and tools for ventricles (Hahn et al., 2001) and gyri (Weiler and Hahn, 2015) the remaining ROIs based on the T1 image and the colorcoded direction map derived from DTI; see Hänsch et al. (2022) for details. For ( 1) and ( 2), standard 3D U-Nets (Çiçek et al., 2016; with 3 and 4 resolution levels, respectively) were trained.For (3), an anisotropic U-Net (Chlebus et al., 2022) was trained to allow for a larger receptive field in the axial plane.Moreover, the two input images (the T1 image and the color-coded direction map) were processed in separate convolutional pathways before combining the extracted feature maps at the lowest resolution level in the U-Net architecture.All models were trained using the Dice loss function.For the multi-structure segmentation in (3), with highly varying volumes of structures to be segmented, a weighting scheme was applied that weights each structure with its inverse volume throughout the training set.This way, smaller structures are assigned a higher weight and are less likely to be omitted by the model. . . . Adapted fiber tracking Fiber tracking was performed using a fiber tracking algorithm that locally adapts its parameters to specific regions of the JHU-MNI-ss white matter atlas (Klein et al., 2015).The atlas, also known as the "Eve atlas" (Mori et al., 2008;Oishi et al., 2008), is based on T1-MRI data from 152 healthy volunteers and consists of 176 regions.For each position during fiber tracking, the covering atlas region was calculated, for which optimized parameter sets were defined in advance and used locally to calculate the fiber segments within that region. If the fibers are computed for all seed points, the tracking process is repeated if a minimum number of fibers has not been tracked or if the maximum number of iterations has not been reached.In our experiments, the minimum number of fibers per bundle was set to 100, and the maximal number of iterations has been set to 500, except for the CTT, where up to 2,500 iterations are allowed as it is challenging for the tractography algorithm to find fibers running through the very small red nucleus and the thalamus at the same time. . . Evaluation We registered all data sets including derived image data and coordinates to the T1 space of each patient.Due to the overall small cohort size, we evaluated each measurement on as many cases as possible for which the required data were available. We compared the average distance in mm between actual treatment positions extracted from the log data (registered to T1) and the calculated standard coordinate (a) manually set in SPM (statistical parametric mapping) and (b) automatically computed by our DL approach.These data are available for 40 patients.For comparing the fiber bundles tracked by both approaches, we compare the distances between the fiber bundles (represented by the probability maps) and (a) the center of the lesion (which has been manually segmented by an experienced neurologist using 6 months' postoperative MRI data), (b) the average treatment point, and (c) the calculated standard coordinate.These data are available for 25 patients (one missing FSL-based CTT, one missing FSL-based ML).Note that for a fair comparison, the probability maps delivered by both approaches have to be thresholded in such a way so that the resulting bundles have nearly the same diameter and show only fibers that are defined as highly likely by each method.This can be achieved by scaling each probability map to [0,1] and using the following heuristically determined thresholds: For any probability map derived by the FSL approach, we take voxel values into account only if their value is at least 0.1.Considering the adapted fiber tracking (AFT) approach, probability maps representing the CTT, the PTT, or the CST are thresholded by 0.2, and maps representing the ML are thresholded by 0.5.If using the same threshold for both approaches or at least the same thresholds for same bundles, the fiber bundle tracked by the FSL approach would always be thinner with respect to the diameter.Even when the threshold is close to 0, the bundle is still thinner because FSL is not able to track fibers at the borders of the 3 fiber bundles very well.Additionally, we examine the influence of the clinician's decision whether the treatment points were selected close to the suggested automatic standard coordinate or whether the treatment points were farther away; see Section 3.1.For that purpose, we define the average deviation of K treatment points p(i, j) (corresponding to patient j) from the calculated standard coordinate stdCoord(j) for a single patient j by dev where \|p(i, j)−stdCoord(j)\| denotes the Euclidean distance between p(i, j) and the standard coordinate.Then, the average deviation for a given set of N patients where for M different j, dev(j) fulfills the condition cond is defined as dev(j). (2) As the VIM is located at the lateral border of the thalamus, it is interesting to examine how distances behave depending on the amount of lateral/medial change between calculated standard coordinate and final treatment points.The deviation of the calculated standard coordinate from the AC/PC line is always 14.0 mm as defined.Thus, for patient j the absolute deviation of K treatment points p(i, j) from the calculated standard coordinate with respect to their lateral/medial change is defined as where p x (i, j) denotes the x-component of p i corresponding to patient j.Correspondingly, we define the average deviation for a given set of N patients where for M different j, dev x (j) fulfills the condition cond as . Results The time required to determine both the standard coordinate and the fiber bundles using the FSL-based pipeline amounts to at least 4 h for a single patient (AMD(R) Ryzen threadripper 3960x 24-core processor).The computational time needed for all the necessary processing steps in the AFT pipeline is approximately 6 minutes for a single patient [using the following system: Intel(R) Core(TM) i7-4770K central processing unit (CPU) @ 3.50 GHz, NVIDIA TITAN Xp].With the automatic AFT-based pipeline, the only requirement is to select the location of the DICOM data and the treatment side of the patient.The rest is automated, resulting in the employee's engagement time or waiting time being nearly identical to the required computational time.Out of the approximately 6 minutes, the registration, preprocessing, segmentation, and automatic AC-PC alignment take about 2.5 min (DL segmentation requires an additional 3 min if there is no graphics card in the PC and the computation is performed on the CPU).The remaining 3.5 min are needed for fiber tracking: the CST and the ML can usually be calculated within seconds, while the CTT requires more time.This time could be reduced if the required minimum number of 100 fibers per bundle were reduced to 50 and/or if the maximum number of iterations for CTT tracking were decreased from 2,500.The CTT is computationally intensive because we use the entire precentral gyrus as a seed region, resulting in tracking significantly more fibers per iteration compared to using a small ROI such as the red nucleus (see Figure 2). . . Standard coordinate The mean euclidean distance of the standard coordinate to the treatment positions (>55 • C) across 40 patients is 1.90 ± 0.72 mm (automatic computation of standard coordinate) and 3.15 ± 1.45mm (manually defined standard coordinate).The treatment was therefore on average performed within the range of 2 mm to the automatically computed target location.A study on deep brain stimulation found ablation positions deviating by more than 2 mm (in the AC-PC plane) from the optimal location to be associated with poorer tremor control (Papavassiliou et al., 2004).Therefore, it is recommended also for tcMRgFUS to aim for an ablation position within a 2-mm range of the optimal target location (Focused Ultrasound Foundation, 2019).Our automatically computed standard coordinate fulfills this requirement, if we assume the treated locations to be optimal with respect to tremor control. . . ROI segmentation A prerequisite for the fiber tracking step is an accurate automatic segmentation of fiber tracking ROIs.Tables 3, 4 show the mean and the standard deviation of the Dice score for all individual ROIs on the 10 test cases for DL algorithm development.In the case of registered "reference" segmentations, this must be considered a quality estimate only, as the registration itself may be inaccurate.The highest mean Dice score is achieved for the thalamus, which is an important ROI for all considered tracts.For the cerebral peduncle, the resulting Dice scores are low (<0.5), which may be explained by the very small size of this ROI.Overall, all required ROIs are detected by the automatic segmentation pipeline. . . Fiber tracts When using the FSL pipeline, the PTT was not restricted to the area between the globus pallidus interna and the thalamus but could extend throughout the brain, especially toward the cortex.Consequently, many false-positive fibers were computed, which would result in very small (or zero) distances between the standard coordinate and the PTT.In the AFT pipeline, however, we restricted the pathways of the PTT between the two ROIs, especially the anterior part of the thalamus.This allows for a reconstruction of the PTT that closely matches the descriptions and results demonstrated by Kwon et al. (2021; see Figure 3) and could be utilized for therapy planning.Thus, we decided not to compare the results of the PTT in more detail as it appears to be of little help due to the significant differences of both approaches.For all other bundles, distances between the borders of calculated fiber bundles and the treatment points and the calculated standard coordinate and the center of the lesion for each individual patient are compared in Figures 4, 5.All plots show that both approaches yield very similar results, especially for the no-go bundles of the CST and the ML.It should be noted that the thresholding as described in Section 2.6 shrinks the fiber bundles in such a way that their extent is comparable for both fiber tracking approaches.Thus, even for the target bundle, the CTT, distances greater than zero can occur.For the CTT, some minor deviations between both approaches can be detected; see also Table 5: the average distance between the CTT and the center of the lesion is 1.56 mm (FSL) and 1.68 mm (AFT); between the CTT and the average treatment point, 1.40 mm (FSL) and 1.74 mm (AFT); and between the CTT and the calculated standard coordinate, 2.01 mm (FSL) and 1.42 mm (AFT). As the average lateral/medial deviation of the treatment points from the standard coordinate is 1.00 mm and as the average deviation of the standard coordinate from the treatment points FIGURE Results of tracking the pallidothalamic tract using both approaches (blue/cyan: FSL, red/yellow/green: adapted fiber tracking). is 2 mm, we used those two values for splitting the groups for a more detailed consideration: dev <2 and xdev <1 represent average distances between treatment points and standard coordinates that are closer together than for dev ≥2 and xdev ≥1 (details are provided in Section 2.6).If averaging all 9 measurements (i.e., the results of all 3 bundles times the 3 distances to the center, the treatment points, and the standard coordinate), then the difference between FSL and AFT is 13.6% in the case of considering only dev ≥2 ; it is 11.85% for dev <2 ; the difference between FSL and AFT is 16.04% for xdev ≥1 and 14.57% for xdev <1 .Although the differences are not as significant as expected, this means that if one only considers patients for whom the later treatment points are closer to the calculated standard coordinate, both algorithms achieve even more similar results.In contrast, if the doctor takes a greater influence on the treatment points and defines them farther away from the automatically calculated standard coordinate, the results differ a little bit more.Note that the doctor's modifications are based exclusively on the FSL results.Thus, especially for the CTT target bundle, the values for the FSL algorithm are always a little bit better (=smaller) than those from the AFT algorithm when determining the distance to the treatment points and to the lesion center.The situation is different for the distance to the calculated standard coordinate, where the AFT algorithm consistently shows smaller distances.In this case, if only the distance between the CTT and calculated standard coordinate is considered, the difference between the two methods is greater than 20% for almost all groups (except for dev ≥2 , where the difference is 17.73%). . Discussion One notable difference between both approaches is that the AFT-based approach computes continuous fiber tracts belonging to a specific fiber bundle.In the used FSL pipeline, which utilizes the red nucleus as a seed ROI, it does not generate continuous fibers that pass through all ROIs but, rather, a probability map where appropriate fiber portions need to be identified through clever thresholding. Both the analysis of individual patients (Figures 4, 5) and the analysis of average distances show high similarities between the results.The largest difference occurs in patients where the treatment positions deviated at least 1 mm laterally/medially from the calculated standard position (i.e., xdev ≥1 ).Here, four out of nine measurement results show a difference of more than 20%.In comparison, in all other cases, only 1 or a maximum of 2 out of 9 measurement results show a corresponding difference.However, it should be noted that the FSL-based results were used for optimizing the treatment points, and the treatment points were optimized based on these fiber tracts.Therefore, it was expected that the greatest difference would exist for xdev ≥1 , as this is where the relatively largest modifications were made by the doctors.Even when considering the average distance from the calculated standard coordinate to the center of the target bundle, the CTT (see Table 5), it can be observed that the AFT-based pipeline determined a distance of only 1.42 mm, whereas the FSL-based pipeline determined a distance of 2.01 mm. The analysis of cases with larger deviations often revealed differences in segmented or registered ROIs.For example, the segmentation of the precentral gyrus in the MNI Atlas extends deep into the brain, not just on the cortical surface.This means that in the FSL pipeline, a fiber only needs to touch a voxel in a lower region to be counted.The subsequent path of the fiber is not taken into account by the pipeline.For instance, fibers from the CST or the CTT can easily drift into the postcentral gyrus. Furthermore, precise segmentation is crucial not only for the very small red nucleus (see Figure 6) but also for the accurate segmentation of the thalamus.If the segmentation of the thalamus is slightly too large, the fiber tracking algorithm will quickly drift toward the internal capsule, where there is strong anisotropy, and mistakenly track parts of the pyramidal tracts.Conversely, if the segmentation of the thalamus is too small, the calculated standard coordinate may end up outside the thalamus, and only fibers that do not pass through the VIM will be computed (see Figures 6, 7). By using a DL network, the segmentation can be significantly more robust and accurate.Another advantage of DL is reliability: while in our experiments, there were instances where the postcentral gyrus was incorrectly determined instead of the precentral gyrus during ROI-based segmentation (requiring subsequent manual correction), in all cases of DL-based segmentation, the correct gyrus was segmented.Note that the evaluation was conducted on a subset of data used for developing the DL algorithm as there was a very limited number of complete data sets available.However, it is important to mention that for most ROIs, there was no gold-standard reference segmentation available anyway. As an additional limitation it should be noted that during the treatment process, registration errors can occur in the treatment log data.This registration is typically performed to align the treatment position with the planning data using a registration matrix. Our results have demonstrated that automatic fiber tracking and automatic determination of the standard coordinate are highly feasible.The AFT approach for automatically determining the fiber bundles required for ETT planning as well as the DLsupported calculation of the standard coordinate have shown that anatomically meaningful and plausible results can be achieved.Compared to a different, FSL-based pipeline where steps are manually performed or triggered and where the standard coordinate is manually determined, the results are generally quite similar in terms of measured distances to the target bundle, the standard coordinate, and the achieved lesion.This is particularly remarkable considering that the individual components differ Measurements are shown for both fiber tracking approaches (AFT and FSL) and are also split up into patient groups where distances between the calculated standard coordinates and treatment points are considered.Numbers are in bold if the two algorithms yield results that differ by more than 20%.AFT, adapted fiber tracking; FSL, FMRIB software library; CTT, cerebellothalamic tracct; CST, corticospinal tract; ML, medial lemniscus. FIGURE Representative fiber tracking results for the two pipelines.Green: CTT, blue: ML, red: CST.Upper row: FSL, bottom row: AFT.Treatment position and standard coordinate are marked by a red circle.The visualized thalamus and the red nuclei are identical in both rows for a better comparison of the fiber tracts; however, ROIs for fiber tracking have been di erently determined for the manual and automatic planning pipeline as described in Sections . .and . . .This explains, for example, the fact that only a part of the CTT (tracked by FSL) passes the red nucleus.CTT, cerebellothalamic tract; ML, medial lemniscus; CST, corticospinal tract; FSL, FMRIB software library; AFT, adapted fiber tracking; ROI, region of interest. Frontiers in Neuroimaging frontiersin.orgalgorithmically (preprocessing, fiber tracking algorithm).One reason for that is that the fiber bundles are strictly constrained by a number of narrowly defined ROIs.In addition, for both approaches, diffusion tensors were calculated, which allow for a robust reconstruction of fiber tracts.However, in individual cases, the bundles may still deviate more from each other.Upon closer examination of these cases, it is often found that the were segmented differently.Here, the DLbased approach has a notable advantage over the registration-based approach, as it can typically produce more accurate segmentations. Therefore, precise segmentation of the ROIs is extremely important, even more so than fine-tuning the fiber tracking parameters or choosing the algorithm, particularly if the algorithm relies on these ROIs.The treating physicians should never focus solely on the fiber tracts but always consider the ROIs used, ideally in combination with color-coded DTI images.A correct segmentation of the thalamus plays a crucial role in nearly all pathways, while the correct segmentation of the very small red nucleus is crucial for computing the CTT.Assessing the segmentation of the pre-and postcentral gyrus on axial slice images is not easy, especially if the images are not aligned to the AC/PC line.Checking those segmented gyri on a threedimensional volume rendering (skull-stripped) is significantly easier and reduces errors. A fully automated pipeline seems desirable both in the scientific and clinical context, as it saves the user a lot of time and manual work.The FSL-based pipeline takes approximately 4 hours to complete per patient, while the AFT-based approach takes an average of only 6 minutes per patient.In the clinical context, one would also want to enable manual control and correction of ROIs and fiber bundles to allow optimal treatment planning.DL algorithms that learn the anatomical course of the fiber tracts themselves might provide additional safety and might help prevent continuity errors or drifting into areas with strong anisotropy.In particular, because the patients are neither tumor nor stroke patients, the fiber tracts are more similar, making it easier to create a valid training set.It is obvious that doctors should still verify these tracts using anatomical knowledge and landmarks. to generate training data.In total, three U-Nets were trained: (1) rough brain extraction based on the T1 image and the registered brain outline, (2) ventricles based on the T1 image, and (3) FIGURE FIGURESchematic overview of the automatic planning pipeline to compute the standard coordinate for treatment and fiber tracts.DWI, di usion-weighted imaging; AC, anterior commissure; PC, posterior commissure; DTI, di usion tensor imaging; ROI, region of interest. FIGURE FIGUREPrecise segmentation of the red nucleus constitutes the basis for tracking the correct part of cerebellothalamic tract.M, midbrain; SN, substantia nigra; CP, cerebral peduncle. FIGURE FIGUREDistances (in mm) between core of fiber bundle and (a) average treatment point (plots in left column) and (b) calculated standard coordinate (plots in right column).As the CTT is the target bundle for treatment, distances are clearly smaller for this specific bundle compared to no-go bundles of the CST and the ML.CTT, cerebellothalamic tract; CST, corticospinal tract; ML, medial lemniscus; AFT, adapted fiber tracking; FSL, FMRIB software library. FIGURE FIGUREThree-dimensional visualization corresponding to Figure.Green: CTT, blue: ML, red: CST.Upper row: FSL, bottom row: AFT.The calculated standard coordinate is shown as a bright sphere.The DL-based segmented thalamus is identical in both rows for a better comparison of the fiber tracts.CTT, cerebellothalamic tract; ML, medial lemniscus; CST corticospinal tract; FSL, FMRIB software library; AFT, adapted fiber tracking; DL, deep learning. TABLE Components of a complete patient data set used in this study.data T1 MPRAGE planning image (1x1x1 mm 3 , 3D sequence) acquired on a 3T Philips Medical Systems Achieva scanner DWI planning data (2x2x2 mm 3 , 56 gradient directions, b-value = 1200) acquired on the same scanner fiber tracts of the CTT, PTT, CST, and ML from manual FSL pipeline used for therapy planning at University Hospital Bonn (see Section 2.4) TABLE Anatomical ROIs used for fiber tracking of the CTT, the PTT, the CST, and the ML. TABLE Mean and standard deviation of the Dice score for all computed regions of interest with manually curated reference segmentations on N = test cases of the deep learning algorithm development. TABLE Mean and standard deviation of the Dice score for all computed regions of interest with registered, uncurated reference segmentation on N = test cases of the deep learning algorithm development. TABLE Average distances (in mm) between core of fiber bundle and (a) center of lesion ( months' postoperative), (b) average treatment point, and (c) calculated standard coordinate.	8,635.2	2023-10-26T00:00:00.000	[ "Engineering", "Medicine" ]
Routing in Dense Human Crowds Using Smartphone Movement Data and Optical Aerial Imagery In this paper, we propose a navigation approach for smartphones that enables visitors of major events to avoid crowded areas or narrow streets and to navigate out of dense crowds quickly. Two types of sensor data are integrated. Real-time optical images acquired and transmitted by an airborne camera system are used to compute an estimation of a crowd density map. For this purpose, a patch-based approach with a Gabor filter bank for texture classification in combination with an interest point detector and a smoothing function is applied. Furthermore, the crowd density is estimated based on location and movement speed of in situ smartphone measurements. This information allows for the enhancement of the overall crowd density layer. The composed density information is input to a least-cost routing workflow. Two possible use cases are presented, namely (i) an emergency application and (ii) a basic routing application. A prototypical implementation of the system is conducted as proof of concept. Our approach is capable of increasing the security level for major events. Visitors are able to avoid dense crowds by routing around them, while security and rescue forces are able to find the fastest way into the crowd. OPEN ACCESS ISPRS Int. J. Geo-Inf. 2015, 4 975 Introduction Major events, like music festivals or football games, attract tens of thousands of people.Unfortunately, accidents can happen every time despite high security preparation, and the consequences are often crucial due to the high number of visitors.In recent literature, crowd simulations that can be used for security issues during major events are typically not based on real-time sensed information but rely on empiric or physical heuristics.An overview of current methods and approaches for such crowd simulations (e.g., evacuations) is provided by [1].Compared to that, in situ information has been a crucial part of navigation approaches in the field of robotics [2].A recent study by Hillen et al. is focusing on fusing real-time in situ and remote measurements to create a more realistic estimation of people movement using agent-based modeling [3].Except for extracting only position information from optical remote sensing data as conducted by Hillen et al. [3], studies in the field of crowd monitoring present promising results in estimating crowd density and dynamics [4,5].The derived information can be utilized for a routing approach for major events. However, event organizers and security authorities usually have very limited near real-time information about the location of visitors at the event site, despite the high penetration rate of smartphones in the general public [6].Terrestrial sensors, like security cameras, are often only available at the most important spots and only have a limited field of view.Recent airborne monitoring technology is able to provide additional high-resolution imagery in real time [7].Due to their mobility and large field of view, even large event sites can be captured in a couple of minutes.The increasing ground resolution of the captured images (e.g., 9 cm; see Figure 1) allows for efficient detection of objects like cars and persons [8]. In this paper, real-time aerial images are combined with smartphone movement data and integrated into a routing tool for major events.It is designed to be used by the official authorities (i.e., police or ambulance) as well as the visitors themselves.It can be of great use for open-air music festivals, where large groups could gather spontaneously in less monitored locations, or for events in city centers (e.g., fairs), as they often take place in multiple locations.In the case of music festivals, for example, the motion of the crowd directly in front of a concert stage is hardly predictable during a live performance.Although the number of people in the crowd might be below the area's maximum capacity, the number of persons per square meter (crowd density) can quickly reach a critical level and lead to dangerous situations.If the crowd density rises above a certain threshold, the situation can become life-threatening and the authorities must intervene.But even in less dramatic scenarios, the perception of each person in these situations might be different.A person's physical condition, hydration level, degree of intoxication, or even the weather conditions at an event could all be factors that influence a person's desire to leave a crowded area in the fastest way possible.We assume that the fastest way is synonymous with the route with the lowest crowd density. For these critical and non-critical situations, we propose a routing concept based on fusing 9-cm optical aerial images with movement data from smartphone users in near-real time.The major aim is to provide an up-to-date crowd density map with a least-cost routing functionality for the event visitor as well as for rescue forces and security authorities. Methods In this section, the automatic patch-based crowd density estimation and the extraction of movement trajectories from mobile smartphone sensor data are described.Furthermore, it is shown how these two types of sensor data are combined to derive a cost layer for least-cost routing purposes. Crowd Density from Aerial Images In the following, the process to automatically compute a density map of a crowd from aerial images ("crowd detector") is described.This map is then input to a location-based individual routing application using a raster-based least-cost path calculation. Before focusing on the computation of the crowd density map, we describe the necessary features of the aerial camera system.The routing system should work at large festivals with an event area of several square kilometers.Despite this large area, the density map must cover all crowded places and must be updated regularly.One major requirement for the generation of an up-to-date routing recommendation is the timely delivery of the density map to the server.To achieve this, a processing system with aerial cameras and an integrated data link must transmit the images to the receiving station on the ground.The processing includes a georeferencing and orthorectification step, which is not mandatory for the crowd detection itself, but for the fusion of image and trajectory data (Section 2.3).The authors of [9] describe a system with an on-board orthophoto generation and a bidirectional air-to-ground data transmission.Their system works well with the routing system presented in this paper, although the installation of the processing system could be located on the ground as well. The aforementioned requirements of large coverage and short update intervals of the density map lead to a trade-off between the field of view of the cameras and the spatial resolution.Hence, in this study, we use the method described by Kurz et al. [10], which is able to detect crowds in aerial images with a resolution of approximately 9 cm.At this resolution, a single person appears as a small blob of roughly 5 × 5 pixels.In very crowded scenes these blobs are hardly discernible due to occlusion and changing lighting conditions.In this case, they instead form a heterogeneous texture without any orientation or regular pattern structure (Figure 1a).Moreover, the appearance of these textures distinctly changes depending on background pixels and lighting conditions (Figure 1b). To overcome these problems we use the crowd detection tool chain proposed by Meynberg et al. [10].The approach convolves image patches of a constant size with a Gabor filter bank and uses a concatenation of filter responses as the input feature vector for a support vector machine (SVM).Readers are referred to [10] for development details.In the context of this paper, we use the term "crowd texture" to describe an image region where persons stand very close to each other and form one coherent structure.The term "crowd patch" is defined as a patch containing this crowd texture.We now summarize the main processing steps of their method, which estimates the person density in aerial images for which no further a priori knowledge is required. Detecting Interest Points The purpose of this step is two-fold.First, it detects corners and saves the coordinates as possible locations of a person, which are the basis for the density estimation described in Section 2.1.4,and second, it significantly reduces the search space for the filtering in Section 2.1.2.The search space reduction is motivated by the fact that during a flight campaign a large number of images are being taken, each image having a resolution of around 18 megapixels.With the original images as the input, the outputs of this step are possible pixel locations where a high number of people are located.Because of the computationally expensive feature extraction and classification steps, the search space should be limited to image regions where a high number of people is likely to occur.In 9-cm resolution aerial images, one person appears like a small blob or corner with a size of roughly 5 × 5 pixels.Therefore, a corner detector by Rosten and Drummond [11] is first applied to the whole image, which reduces the locations of possible crowd textures from theoretically all pixel positions to only those positions that are detected as a corner, and hence have the necessary condition to be considered for further processing.In this way, the number of filter operations that are performed according to Section 2.1.2can be reduced by a factor of 1000 (depending on image content) to allow for usage in time-critical scenarios. Finding a Feature Vector Representation for Crowd Image Patches The input of this step is an array of all possible pixel locations of crowd textures and the output is a set of feature vectors that discriminate well between image patches with crowds and image patches without crowds.In the following, we describe how such a feature vector is created.Every image patch is convolved with a bank of Gabor filters.These filters are particularly appropriate for texture representation, first introduced by Manjunath and Ma [12], as they encode both orientation and scale of edges in a low-dimensional feature vector.Image patches with regular building structures (Figure 2a) result in a strong response at certain orientation angles (Figure 2c).On the contrary, a crowd patch (Figure 2b) gives a high response in every direction (Figure 2d), because the persons do not form any regular pattern.In this way, it is possible to construct effective, discriminating feature vectors for the binary classification task that follows.Let ∈ be an image patch at a candidate pixel position (x,y).Its Gabor wavelet transform is thus defined as where , is one Gabor filter function where s and k determine the scale and the orientation angle of the filter, respectively.Next, for each combination , , the mean , and variance , are computed and stacked into the final feature vector ∈ : , , , , with being the number of orientations and being the number of scales of the filter bank. Classification with Non-Linear Support Vector Machine The set of all feature vectors is then passed as a matrix to an SVM with a radial basis function kernel.Its return value is a vector of scores that determines if an image patch has been classified as a crowd patch or not. Crowd Detection: From Binary Classification to Continuous Density Estimation Now the list of possible person locations (Section 2.1.1)and the classification result (Section 2.1.3)are combined.From now on, only those possible person locations that lie within a positively classified image patch are considered and are used to compute a probability density function with a Gaussian kernel over the image domain.In this way, the crowd density estimation can be expressed as an intensity value.This value can be assigned to every pixel of the original image and not only to a finite and very sparse set of detected corners.To this end, the assigned value is not calibrated with a verified crowd density that can hardly be measured in a real-world scenario; however, it is still sufficient to serve as a two-dimensional cost function in this context. Movement Trajectories from Smartphone Data The preliminary step for working with smartphone sensor data is the implementation of an app to record specific internal sensor data via the Android API [13].For this work, the current geographic location of the smartphone user and the corresponding movement direction along with the estimated movement speed are derived from different smartphone sensors like the acceleration sensor, compass, and GPS/GNSS (Global Navigation Satellite System).Based on previous investigations, we claim that the position accuracy of GPS/GNSS is sufficient especially for events in rural areas but also for events in small and medium-sized cities [3].In Figure 3 an exemplary dataset of recorded smartphone movement data (geographic location plus corresponding moving direction) of three different smartphone users is visualized.The movement speed of the user at the specific geographic location is used to re-evaluate the image-based crowd density estimation.We assume that high movement speed is an indicator for a low crowd density whereas a slow movement speed suggests a high crowd density.Rastogi et al. conducted a comprehensive comparison of pedestrian walking speeds based on literature from 1967 to 2007 [14].The values of average speed for adults vary from 1.32 m/s to 1.51 m/s.Furthermore, every study revealed significantly lower walking speeds for elderly people (generally over an age of 65 [15]) with 0.97 m/s to 1.34 m/s.In addition, [16] found that pedestrians in groups tend to be slower (1.54 m/s alone compared to 1.41 m/s in groups for younger pedestrians).Thus, varying circumstances have to be considered regarding the analysis of movement speed.It is especially important for major events that it is determined whether the user is only moving slowly because of a specific reason (e.g., to look for shirts at the merchandising booth or to buy something to drink), or if the user actually has to move slowly because of high crowd density.This might result in the misinterpretation of high crowd density for less crowded regions. The smartphone app transfers the raw movement data into a spatial PostgreSQL/PostGIS database on a web server using a mobile Internet connection.Afterwards, the dataset is integrated in a GeoServer in order to be accessible via Web Map Service (WMS) and Web Feature Service (WFS) interfaces by the Open Geospatial Consortium (OGC).Thus, the data can be integrated in any processing chain or application over the Web.Above that, the data can be directly processed in the database or can be integrated in OGC Web Processing Service (WPS) processes, which is of high importance regarding the following conversion from movement speed to crowd density information. Information Fusion The image-based crowd density estimation and the density information derived from smartphone movement data are afterwards combined to create a cost layer for the least-cost navigation.For this purpose, the geographic information system GRASS GIS [17] is used.GRASS GIS is highly suitable for this work as it is a) raster-based and b) supported by the Python implementation of the OGC WPS standard named pyWPS [18].This allows for future Web-based real-time processing of the aerial image and smartphone data. The density layer derived from the aerial image data serves as a base layer for the raster-based least-cost routing process.For this purpose, the layer is reclassified to a range between 0.43 (no density) and 7.1 (high density).Fruin identified these values of (a) 0.43 people per m 2 (p/m 2 ) and (b) 7.1 p/m 2 as the crowd densities with (a) a normal walking speed and (b) no movement at all [19].Above that, Fruin investigated a reduced walking speed at 2.0 p/m 2 , involuntary contact between people at 3.57 p/m 2 , and potential dangerous crowd forces starting at 5.55 p/m 2 , which are important thresholds for further evaluation of the crowd density information [19].Weidmann investigated the relation between the local speed and the local density of people as well and formalized it as: where is the speed at a certain density, , is the maximum speed at full freedom (1.34 m/s), is an empirically derived fit parameter (1.913 p/m 2 ), is the crowd density in p/m², and is the crowd density at which no movement is possible anymore (5.4 p/m 2 ) [20].Wirz et al. conducted an empirical study based on smartphone measurements to verify this relation [21].Based on Equation (3), the crowd density can be estimated from the smartphone data using 1.913 ln 1.34 0.646929 for each local speed recorded with the smartphone.This smartphone-based density information is imported into GRASS GIS as well and is converted from vector to raster format in order to be combinable with the information derived from the image data.For this, the point (now the respective pixel) of the smartphone measurement is expanded so that it delivers information for the area surrounding the smartphone user as well.Thus, the people density calculated with Equation ( 4) is adopted for the local neighborhood of the test person as well.This is more realistic instead of using only the point measurement for the density enhancement.Otherwise, the density at the pixels next to the measurement would be totally different. Based on the complemented density information, a least-cost layer can be calculated using the GRASS GIS function r.cost.The initial point for this calculation is the current geographic location of the user.The cumulative cost of moving from this point to each cell is calculated and stored in the resulting cost layer.Finally, the least-cost path between an arbitrary point and the current location of the user can be calculated using the function r.drain. Application Scenarios In the following, two test scenarios are outlined in which our real-time routing approach is beneficial.The first scenario describes a tool to escape from emergency situations, whereas the second scenario presents a generic decision support application that can be used in multiple situations.The "give and take" principle is essential for all applications, meaning that both the event attendee and the organizer have to deliver information to receive a result.This concretely means for our scenarios that the event attendee has to send information about their current location and speed via smartphone and the organizer has to provide an aerial imaging system covering the event area.Only if information from both sides is available can the results be provided as desired.This means that on the one hand the user can fetch an overview map of the current crowd distribution and use the least-cost routing, and on the other hand the organizer can guarantee a high security standard and use the routing app for the event's security and rescue forces. The Fastest Way Out of a Crowd One application in which the real-time navigation can be used is during music festivals.The crowd in front of the music stages is often very dense.Combined with extreme weather conditions (e.g., high temperatures) and physical fatigue, this might result in dangerous situations.In such crowds there is no chance to get a full overview of the situation and to find the best way out, especially for persons with a low body height.Due to the lack of orientation, the person might go towards an even denser region within the crowd, not knowing that a free space might be very close by. Our routing approach can be integrated in an emergency app provided by the event organizer.The guests have to provide their current location and speed measured with their smartphone.In exchange for that, they are able to view an overview map with the current crowd distribution in the event area and are able to use the described emergency navigation (Figure 4).Additionally, the security and rescue forces of the event could directly receive the location of the visitor when the app is used.Thus, help can be on-site much earlier.The overview map has to be provided by the organizer by recording aerial imagery during the event (e.g., with helicopters, drones, or similar).It may even be possible for typical SLR cameras to be installed in high positions to cover special regions, like the space in front of the music stages, for example. The Fastest Way towards a Point-of-Interest (POI) The real-time routing approach can be adapted to any event that has appropriate image data available for the event site.In the following, this assumption is stressed for the example of football games.In this particular use case, dense crowds are gathering in short time frames (i.e., before or after the game as well as during the half-time break).A concrete example for this is reported by officials of the Borussia Park in Mönchengladbach (Germany).After the football games, the main way towards the parking spaces is commonly blocked by the police to escort the fans of the opposing team to their buses.In the meantime, many people have to wait while more and more people are streaming out of the stadium towards the parking spaces.The organizers try to avoid complications by opening gates that allow people to go the longer way around the stadium on the other side.In this situation, this way would be much faster compared to waiting in the dense crowd.However, the people that are streaming out of the stadium are often not aware of (i) the blockade by the police and (ii) the option to use an alternative way. Our routing approach could help to ease this situation by informing and navigating some visitors to the alternative route.Figure 5 illustrates the generic workflow in which the user has to provide their current location along with the target of routing.This information is sent to a web server where the actual calculation for the least-cost route is executed.The resulting route is afterwards visualized on the smartphone of the user, who constantly reports (automatically in the background of the app) his location and speed.For this example, it is essential to integrate the smartphone data of as many users as possible to avoid potential jams caused by the system itself.As soon as the alternative route is crowded as well, the system carefully has to decide which direction to choose.If the main way is open again and the crowd dissolves, the navigation system routes people along the typical way.Thus, potential mass panics or at least a tremendous gathering of people can be avoided. In general, real-time navigation based on our routing approach could be used during any major event; for example to reach the nearest refreshment shop during a musical festival or city event.Even navigation through the city streets to a specific parking garage with an emphasis on avoiding large crowds (e.g., in front of stages or booths) can be useful.In any case, the advantages are on both sides, for the event organizers and the guests.The guests utilize the tool to avoid stress, overexcitement, and anger on the one hand, while on the other hand the organizers can ensure security during the event and are able to increase the event's attraction by providing a modern smartphone navigation app.In addition, security and rescue forces are able to utilize the app for their efforts and could reach the location of an emergency earlier. Figure 5. Conceptual design of a smartphone app for least-cost routing during or after a major event (in this case a football game). Prototype Implementation A prototype implementation based on non-real time data is conducted as a generic proof of concept for the integration of the two data sources: remote sensing and smartphone data.test data for the prototype are recorded during the music festival Wacken in 2013 (Figure 6a).One can clearly see the dense crowd standing in front of the stages.First, the crowd density is estimated using a fast multi-core C++ implementation.Consecutively, the smartphone data has to be recorded and transferred to the web server for analysis.This allows for a future real-time implementation of the concept.Finally, the calculation of the least-cost path is performed, combining both data sources in GRASS GIS. Crowd Density Estimation in Aerial Images-Implementation Details The density estimation step requires panchromatic aerial images with a spatial resolution of ca.9-20 cm.Because the fusion with location data requires georeferenced and orthorectified images, a real-time orthorectification module processes every aerial image before the crowd density estimation starts.In this time-critical scenario we choose a fast GPU-supported orthorectification and georeferencing implementation [9], which processes a typical 18-Mpix image in under 200 ms.It uses an interior parameter set determined prior to a flight campaign by a self-calibrating bundle adjustment.The system is equipped with an IGI AEROcontrol GPS/IMU unit [22], which records the external orientation parameters for every image with a sufficient angular accuracy.In this way, no ground control points are needed, which is an important fact in this real-time scenario.Moreover, the projection step uses a digital elevation model of the Shuttle Radar Topographic Mission (SRTM) to derive the 3D world coordinates from the image coordinates.It has a resolution of 25 m and a 16-bit quantization. The actual crowd-density estimation, consisting of the four steps described in Section 2.1, takes the orthorectified image as an input (Figure 6a), computes filter responses of patches at promising locations, and classifies the filter responses into the two classes "crowd" and "no-crowd."Based on supervised learning, the classification step with an SVM needs initial training.To this end, the SVM is trained with roughly 10,000 image patches; each has a size of 64 by 64 pixels.Concretely, for training, one patch is convolved with 24 Gabor filters, consisting of S = 4 scales and K = 6 orientations (see Section 2.1.2for details).Each filter has a width of 48 pixels.The resulting 24 mean values and 24 variances of each filter response are then used as the final 48-dimensional feature vector, representing one image patch and being the input for the training of the SVM.Of course the computationally intense training can be done offline.During the event, under real-time conditions, the filtering is performed in parallel on several patches using C++ with OpenMP.The efficient libSVM library is then able to produce the results almost immediately.Finally, the succeeding Gaussian filtering (Section 2.1.4)produces a grey-value 8-bit image of the same size as the original input image, where an intensity value of zero corresponds to a very low crowd density and a value of 255 corresponds to the highest measured crowd density (Figure 6b). Figure 6c illustrates the calculated density information from low (blue) to high (red) with the orthorectified image in the background.This image serves as the interface with the GRASS GIS system and provides the basis for the least-cost routing. Calculating the Least-Cost Path using GRASS GIS The density layer and the smartphone information are imported into GRASS GIS to perform the least-cost calculation.In a first step, a cumulative cost layer is created based on the current location of the user using r.cost.For testing purposes, a position in front of the stages in a highly crowded area is assumed.The cumulative costs can then be used to navigate to a defined point or to navigate towards a less dense area out of the crowd.For the latter case, a point within a less dense area has to be identified using, for example, a nearest point functionality.If the destination of the routing is known, the least-cost path can be calculated using r.drain.An exemplary result can be seen in the left image of Figure 7. Afterwards, the costs derived from the smartphone data are fused with the cost layer derived from the aerial images.As the aerial information might be obsolete, the smartphone data indicate a high crowd density where the original path (left image in Figure 7) is located.Thus, an updated least-cost path is calculated based on the new cost information (see right image in Figure 7).It can be seen that the route for the user has changed significantly compared to the result shown in the left image because of the added cost information derived from smartphone movement data.Instead of moving away from the dense crowd, which is the typical behavior (e.g., in a stress situation), the new calculation suggests that the user move sideways straight through a medium-density crowd to reach a free space much faster. Conclusions and Outlook In this paper, we proposed a least-cost navigation based on the fusion of real-time aerial image data and smartphone sensor data.The image data are used to estimate an extensive crowd density map.For this purpose, a patch-based approach with a Gabor filter bank for texture classification in combination with an interest point detector and a smoothing function is applied.The GPS/GNSS location information and the current movement speed of a user are gathered with a smartphone app.Thereafter, the in situ crowd density is estimated based on the smartphone information and allows for enhancement of the overall density information.Finally, a least-cost routing is performed based on the composed density information using GRASS GIS. Two possible applications for the integration of our navigation approach are presented.The emergency application can support people who want to quickly escape from a dense crowd (e.g., during a music festival).Above that, the routing approach in general can help in various situations, for example after a football game where certain routes are blocked by the police. To exploit the real benefit of the least-cost approach, it is necessary that it can be conducted in real time.This means that as a first step, the information from both data sources (image and smartphone) has to be accessible in real time.This can be achieved via existing standards of the OGC like the WMS, the WFS, or the Web Coverage Service (WCS).Furthermore, the processing of the data and the calculation of the cost layer as well as the least-cost path estimation have to be available in real time.Using GRASS GIS, all processing steps presented in this paper can be integrated in a web-based infrastructure to enable the actual real-time usage as a navigation application. Overall, it can be stated that the advantages of our approach are twofold, both for the event organizers and the event attendee.The guests can use the tool to avoid stress, overexcitement, or anger, whereas the organizers can raise the security level of the event and increase its attraction at the same time by providing a modern smartphone navigation app.In addition, security and rescue forces are able to utilize the app for their efforts and could reach the location of an emergency faster. Future work will be the real-time implementation of our routing approach by bringing together aerial imagery and smartphone movement data in real time during a major event.In addition, the transition from routing to navigation according to the presented application scenarios has to be the subject of future research. Figure 1 . Figure 1.Example of two 100 × 100 pixel image patches containing human crowds (9-cm resolution).The major challenges are varying lighting conditions, varying backgrounds, and mutual occlusions.(a) High crowd density, many occlusions; (b) low contrast. Figure 2 . Figure 2. Two original images (a) and (b) with their respective response images (c) and (d) after convolving both with a Gabor filter.Subfigure (c) shows the response of the regular structure in the original image (a), while (d) shows the response of the unstructured crowd in the original image (b). Figure 3 . Figure 3. Visualization of movement data from three different smartphone users (green, blue, and red) derived from the internal sensor data via the Android API.The dots symbolize the GPS/GNSS location of the smartphone user.The lines point towards the corresponding movement direction of the user. Figure 4 . Figure 4. Schematic representation of an emergency smartphone app.The crowd density is visualized in the background.The fastest escape route is emphasized with a red arrow. Figure 6 . Figure 6.Image (a) shows a subset of an unprocessed aerial image from a music festival with a dense crowd standing in front of a stage.Image (b) is the computed crowd density layer as a gray value image (black = low density, white = high density).For illustration purposes we created a composite image (c), where the density is laid onto the original image (blue = low density, red = high density). Figure 7 . Figure 7. Resulting least-cost paths (red line) based on the original cost layer derived from the aerial image only (left) and with the additional cost information (circles) derived from smartphone sensor data (right).The background consists of the two cost layers over the corresponding aerial image scene.One main stage of the music festival is located in the lower right of both images.	7,371	2015-06-12T00:00:00.000	[ "Computer Science", "Engineering", "Environmental Science" ]
New Exact Soliton Solutions of the ( 3 + 1 )-Dimensional Conformable Wazwaz–Benjamin–Bona–Mahony Equation via Two Novel Techniques In this work, the ( 3 + 1 )-dimensional Wazwaz–Benjamin–Bona–Mahony equation is formulated in the sense of conformable derivative. Two novel methods of generalized Kudryashov and exp − φ ℵ are investigated to obtain various exact soliton solutions. All algebraic computations are done with the help of the Maple software. Graphical representations are provided in 3D and 2D profiles to show the behavior and dynamics of all obtained solutions at various parameters’ values and conformable orders using Wolfram Mathematica. Introduction Partial differential equations (PDEqs) have attracted a particular interest from researchers in the fields of natural sciences and engineering due to the applicability of these equations in modeling various scientific phenomena in interdisciplinary sciences such as mathematical physics, mechanics, signal and image processing, and chemistry. Most physical systems are not linear; therefore, nonlinear partial differential equations (NLPDEqs), particularly nonlinear evolution equations (NLEEqs) (see [1]), have inspired researchers to investigate the existence of exact solutions for such equations. Finding new exact solutions for NLPDEqs can significantly provide a good interpretation for the physical meaning and dynamics of these equations. Therefore, several research studies have recently been done on developing new methods for solving NLPDEqs exactly. Some of the most notable methods that have been applied to solve some interesting NLPDEqs are Equation (1) was first proposed by Wazwaz [10] by formulating a new three-dimensional modified version of BBMEqs, known as the Wazwaz-Benjamin-Bona-Mahony equation (WBBMEq), via coupling or various generalized contexts or as combination of both of them. Higher dimensional problems with various applications can be described by the WBBMEq with its spatial and temporal variables [8,9]. Therefore, there is a significant need to find exact solutions for WBBMEq to interpret their physical meaning and dynamics. Fractional differential equations (FDiEqs) are generalized forms of integer-order differential equations. FDiEqs have attracted the interests of researchers due to the ability of these equations in modeling various scientific phenomena better than the integer-order ones (refer to [11][12][13]). The behavior of some physical systems can be interpreted better than the integer-order ones due to the nonlocality of fractional derivatives, and many systems have memory effects. There are many definitions of fractional derivatives, but the most common ones are Riemann-Liouville and Caputo fractional derivatives where the properties of linearity are commonly shared among these derivatives, but other properties such as product rule, constant, chain rule, and quotient rule are not satisfied. FDiEqs are considered as a powerful tool in modeling scenarios (see [14][15][16]), but this tool comes with various challenges when dealing with FDiEqs due to the difficulty of obtaining exact or analytical solutions where the solutions can be very complicated or impossible to obtain them for certain cases. As a result, to overcome some of challenges associated with nonlocal fractional derivatives, a new generalized fractional derivative of local-type, named conformable derivative, was proposed by Khalil et al. which is basically a stretch for the usual limit-based definition where all usual derivative's properties are satisfied [17]. Many research studies have discussed the mathematical analysis and applications of conformable derivative such as conformable Laplace's equation [18] and generalized conformable mean value theorems [19]. Seadawy et al. [20] and Bilal et al. [21] investigated various soliton solutions for conformable WBBMEq using the methods of simple ansatz and generalized exponential rational function, respectively. Inspired by all above studies, this work is mainly aimed at obtaining new exact solitary solutions for a version of WBBMEq formulated in the sense of conformable derivative (ComD) with the help of two novel techniques: generalized Kudryashov method and expð−φðℵÞÞ method. The general fractional formulation of WBBMEq can be expressed as follows: where D ζ is the fractional operator of order: ζ ∈ ð0, 1. The exact solutions of Eq. (2) have been investigated in some research works using the ðG ′ /GÞ-expansion method [9], modified simple equation method [8], and Riccati-Bernoulli Sub-ODE method [8]. However, according to the best of our knowledge, none of previous research works has investigated Eq. (2) in the context of ComD via the methods of generalized Kudryashov and expð−φðℵÞÞ. Therefore, all results in our work are new and worthy. This work is divided into the following sections: Essential notions about conformable derivative are presented, and the methodology of our two proposed approaches is discussed in Section 2. In Section 3, the main results of our work are presented. In Section 4, the graphical comparisons of our obtained exact solutions are represented in both 2D and 3D plots for various values of parameters and ζ. The conclusion of our work is provided in Section 5. Fundamental Preliminaries and Methodology Some important notions about conformable derivative are introduced in this section. In addition, the methodology of two different approaches, namely, generalized Kudryashov method and expð−φðℵÞÞ method, is also described, respectively. The conformable derivative can be defined as follows: Definition 1. Given a function Ψ : ½0,∞Þ ⟶ R ∋ ∀t > 0, the conformable derivative of order ζ ∈ ð0, 1 of Ψ can be expressed as follows: Let Ψ be ζ-differentiable in some ð0, ϱÞ, where ϱ > 0, and the limit of D ζ t ðΨðtÞÞ exists as t ⟶ 0 + ; then, from this definition, we obtain the following: The theorem [17] below shows that D ζ t satisfies all usual limit-based derivative's properties as follows: Theorem 2. For ζ ∈ ð0, 1, let functions Ψ and Φ be ζ-differentiable at a point t; then, we have the following: (a) D ζ Journal of Function Spaces Consider the following form of NLEEq, with 4 independent variables: x, y, z, and t, formulated generally in the sense of fractional derivative: where Ψ = Ψðx, y, z, tÞ is an unknown function and T is a polynomial of Ψ and its partial derivatives in which all of the nonlinear terms and highest-order derivatives are included in Eq. (5). First of all, to solve Eq. (5), we use the following traveling wave transformations for ComD: where p, q, γ, and δ are all constants with the condition: p, q, γ, δ ≠ 0, and δ is the wave speed. According to the above transformations, Eq. (5) is reduced to the following ODE. The derivative with respect to ℵ is represented by a prime. Equation (7) should be integrated term by term one or more times. The Generalized Kudryashov Method. From GKuM, the obtained solution for the reduced equation is constructed by a polynomial in ℏðℵÞ as [4,22]: where p k ðk = 0, 1, ⋯, jÞ, b l ðl = 0, 1, ⋯, wÞ are constants which are needed to be determined ∍p J ≠ 0, q W ≠ 0, and L = LðℵÞ is the solution of the following equation: The solution of Eq. (9) can be expressed as follows: where I 1 is an integration constant. According to the homogeneous balance principle (HBPrp), the positive integers: J and W in Eq. (8) can be obtained with the help of Eq. (7). In addition, a polynomial, ℏ, can be determined by the substitution of Eq. (8) into Eq. (7) along with Eq. (9). Now, by equating all the coefficients of polynomial ℏ to 0 in order to construct a system of algebraic equations, this system is solved with the aid of the com-puter software such as Maple and Wolfram Mathematica in order to find the values of p k ðk = 0, 1, ⋯, jÞ, q l ðl = 0, 1, ⋯, wÞ. At the end, all soliton-type solutions of the reduced Eq. (7) can be found by the substitution of these obtained values and Eq. (9) into Eq. (8). 2.2. The expð−φðℵÞÞ Method. From ExpM [3], the obtained solution for the reduced equation is constructed by a polynomial in expð−ΦðℵÞÞ as follows: where p j ðp w ≠ 0Þ are constants which are needed to be found, and φðℵÞ satisfies the following auxiliary ODE: Note that Eq. (12) has distinct solutions which are expressed as follows: Case 1. When χ2 − 4ϑ > 0 and ϑ ≠ 0, the hyperbolic function solutions are expressed as follows: Case 2. When χ2 − 4ϑ < 0 and ϑ ≠ 0, the trigonometric function solutions are expressed as follows: Case 3. When χ2 − 4ϑ > 0, ϑ = 0 and χ ≠ 0, the hyperbolic function solutions are expressed as follows: Case 4. When χ2 − 4ϑ = 0, ϑ ≠ 0 and χ ≠ 0, the rational function solutions are expressed as follows: Case 5. When χ2 − 4ϑ = 0, ϑ = 0 and χ = 0, we have the following: From the above cases, the integration constant is represented by I. By the substitution of Eq. (11) into the reduced Eq. (7) and collecting all terms together that are in the same order of expð−φðℵÞÞ j ðj = 0, 1, 2, ⋯Þ, the polynomial in terms of expð−φðℵÞÞ is verified. Then, by equating all coefficients to 0, a set of algebraic equations is constructed for p j ðj = 0, 1, :: Solutions of the (3 + 1)-Dimensional Conformable WBBMEq Exact soliton solutions of the proposed Eq. (2) are obtained in this section via GKuM and ExpM. By integrating Eq. (18) once with respect to ℵ, we obtain the following: Let us substitute Eq. (19) into Eq. (7) and collect the coefficient of each power of expð−ΦðℵÞÞ j . Now, a set of algebraic equations of p, q, γ, δ, p 0 , p 1 , χ and ϑ is obtained by equating all coefficients to 0. The Graphical Comparisons of Solutions Since the measurements in physics are local, this can make ComD suitable for modeling many physical phenomena, and it is also efficient to work with ComD to obtain exact solutions for NLPDEqs although ComD does not have some of the essential properties to be categorized as a fractional derivative. From the authors' opinion, formulating and solving NLPDEqs in the sense of ComD are always recommended since all local and nonlocal fractional derivatives have both advantages and disadvantages. Therefore, exploring new properties and definitions for all local and nonlocal fractional derivatives are helpful while working on modeling Conclusion Two novel techniques of generalized Kudryashov and expð−φðℵÞÞ have been applied in this work to investigate Journal of Function Spaces exact soliton solutions of the (3 + 1)-dimensional conformable WBBMEq in the sense of ComD. The obtained solutions are new which imply that the studied techniques provide efficient results. All algebraic computations in this work have been done using the Maple software. Graphical representations have been provided for all obtained solutions at various parameters' values and conformable orders with the help of Wolfram Mathematica. All in all, the studied methods can be potentially applied to solve various NLPDEqs that are apparent in many important nonlinear scientific phenomena in physics and engineering. Our results can be further extended in future research works into solving various classes of higher dimensional nonlinear partial differential equations which will provide a major contribution to soliton theory and mathematical physics. Data Availability No data were used to support this study.	2,594.2	2021-07-21T00:00:00.000	[ "Mathematics", "Physics" ]
Through-Wall UWB Radar Based on Sparse Deconvolution with Arctangent Regularization for Locating Human Subjects A common problem in through-wall radar is reflected signals much attenuated by wall and environmental noise. The reflected signal is a convolution product of a wavelet and an unknown object time series. This paper aims to extract the object time series from a noisy receiving signal of through-wall ultrawideband (UWB) radar by sparse deconvolution based on arctangent regularization. Arctangent regularization is one of the suitably nonconvex regularizations that can provide a reliable solution and more accuracy, compared with convex regularizations. An iterative technique for this deconvolution problem is derived by the majorization–minimization (MM) approach so that the problem can be solved efficiently. In the various experiments, sparse deconvolution with the arctangent regularization can identify human positions from the noisy received signals of through- wall UWB radar. Although the proposed method is an odd concept, the interest of this paper is in applying sparse deconvolution, based on arctangent regularization with an S-band UWB radar, to provide a more accurate detection of a human position behind a concrete wall. As a result, this research utilizes a nonconvex arctangent regularization function which is parameterized by a parameter tuner to avoid the nonconvex optimization of the cost function, and, thus, realize the local optimal point [32][33][34][35][36][37]. This advantage will be applied in through-wall UWB radar application, in this paper, to extract human ranges behind a wall from a noisy received signal. The advantages of UWB radars, which work by emitting short pulses of high-frequency electromagnetic wave (EM), are that they can provide high penetration, high-range resolution, less harm to human health, and lower power consumption, compared with continuous-wave radars [16]. UWB radars are thus suitable for through-wall applications and the detection of human subjects behind solid objects [38][39][40][41][42][43][44][45][46]. The S-band frequency range (2)(3)(4) was used in the proposed radar scheme to provide both high spatial resolution and wall penetration [38][39][40][41][42][43][44][45][46][47][48]. In previous works [43,44,48], UWB radars identified human objects behind a wall by detecting respiration rate. However, UWB radar algorithms for respiration and/or heartbeat detection need to capture received signals at least 512 times per minute (60 s), for over 1 cycle vital sign signal, with Nyquist sampling condition. As a result, we are interested in the sparse deconvolution algorithm with arctangent regularization for reconstruing human range from only one received signal, and for faster detection. This method provides high performance for detecting humans by range movement, but it is difficult to distinguish between a standing human and static objects. In real-world applications, humans have motions so that the detectable range is sufficient for obtaining their positions [41]. The organization of this research is as follows: Section 1 is the introduction. Section 2 details the theoretical background of convolution, deconvolution, and sparse deconvolution. Section 3 describes sparse deconvolution with arctangent regularization and the majorization-minimization algorithm. Section 4 deals with the experimental setup and results in the detection of human subjects behind a wall. The concluding remarks are provided in Section 5. Theoretical Background We begin this paper with necessary background knowledge for reconstructing target signals in through-the-wall UWB radar application, which briefly introduces convolution, deconvolution, and sparse deconvolution with convex and nonconvex functions with the MM algorithm. Convolution Model A block diagram of the UWB radar system is experimentally determined in Figure 1. When the transmitter emits, part of the energy is reflected off the wall, and objects detected by receivers are then captured by an oscilloscope [43][44][45]. A recorded radar data is a linear system where a UWB wavelet h(t) is convolved with the reflectivity series x(t). In practice, the radar data (y) can be expressed in a matrix form with environmental noise (see details in [1]). For complexity reduction of matrix inversion, the reflectivity series (x) will be assumed to be the same size of y, where y ∈ R N is the received signal in vector form, x ∈ R N is the reflectivity series (sparse signal) in vector form, w ∈ R N is white Gaussian noise in matrix form, and H ∈ R N×N is a convolution matrix. The convolution matrix H is a Toeplitz structure and determined as H = A −1 B, where A and B are band matrices (sparse matrix) ∈ R N×N [36]. If the matrices A and B are so far from the band matrix, such as N = 2, then also A −1 B, based on the Equation (2), is so far from the exact result. where the matrix elements of A and B consist of a j and b i coefficients. The a j and b i are derived from the Z-transform of the wavelet (Gaussian pulse) h(n), where h(n) = nr n sin(ω 0 n) for analysis of A and B band matrices [36]. The Z-transform of the wavelet is mathematically expressed. Deconvolution Model Deconvolution is estimating x from y (y = Hx + w), but x cannot be directly solved by rearranging in the form of x = H −1 (y−w) or x ≈ H −1 y, because H may be singular and non-square matrices that H has no inverse form . The alternative approach to estimating x is by minimizing the cost function F(x) = \|\|y − Hx\|\| 2 2 . The estimated signal x is derived by optimizing the cost function. The estimated signalx is as consistent with y as possible, according to the square error criterion; however, H T H may not be invertible. To avoid this problem, we must add a regularization term into the cost function; as shown below, it is the solution to the more general problem [20,41]. It is the cost function in the general form to solve the solution x, as well known x = argmin F(x), where \|\|y − Hx\|\| 2 2 is a closed convex function to force x to be consistent with the measured signal y. φ(x) is a regularization term (or penalty function) that can enhance some desired features of x, and λ is the regularization parameter that balances the importance of data fidelity about the estimated signal x. In the case of the low noise received signal y, we can roughly estimate the reflectivity series x by the optimization problem, Equation (10), with the regularization term of small energy (L2 norm) where I is the identity matrix. The notation x 2 2 represents the sum of squares of the vector x, Sparse Deconvolution Model with L1 Norm (Convex Regularization) Given the limitation of L2 norm regularization and the noisy characteristic of received signal y, we can easily estimate x to be a sparse signal (spike) from y by minimizing Equation (10) with a convex regularization term of the L1 norm where x 1 is called L1 norm regularization (convex regularization) represented by the sum of absolute values of vector x, (12) is the sparse signal deconvolution problem in a simple form; this problem cannot be solved in an explicit form, because the cost function is not differentiable. It must be solved by using an iterative numerical algorithm; this research used the majorization-minimization (MM) algorithm, which will be discussed in the next section. Sparse Deconvolution Method This section has been divided into three parts. Section 3.1 and 3.2 propose the arctangent regularization and numerical method for this sparse deconvolution, respectively. The last part presents a flowchart of the proposed method for through-wall UWB radar. Sparse Deconvolution with Arctangent Regularization (Non-Convex Regularization) To improve signal sparsity, the L1 norm (convex) regularization in Equation (12) could be replaced with a nonconvex regularization [20][21][22]. In this paper, the nonconvex arctangent regularization in Equation (13) was used in place of the L1 norm to improve the detection performance of the through-wall UWB radar scheme. Arctangent regularization can provide reliable minimization and a fast solution using the MM algorithm, see Section II-E in [37]. The arctangent function φ atan (x, α) is a parameterized regularization with α > 0 (e.g., logarithmic and first-order ration) and is mathematically expressed as The derivative of φ atan is written as shown below, to prepare in the MM algorithm. Minimizing the Cost Function Using Majorization-Minimization (MM) Approach Sparse deconvolution with L1 norm, Equation (12), and arctangent regularization, Equation (15), cannot be carried out directly because both equations are not differentiable. To estimate the sparse signalx, MM algorithm was utilized to minimize the cost function F(x) by sequentially minimizing the quadratic majorizer G(x), as shown in Figure 3a. This idea is that each G(x) is easier to solve than F(x). The MM approach produces a sequence, x k+1 , each being obtained by minimizing G(x) and converging to the minimizer of F(x), where k is the iteration counter, k = 0, 1, 2, . . . , K. The updated point x k+1 is derived by minimizing the quadratic majorizer G(x) The function G(x) must always be the majorizer (upper bound) of the cost function F(x), G(x) ≥ F(x), ∀x ∈ R N . The tangent point between G(x k ) and F(x k ) is related by In practice, the chosen majorizer G(x) should be relatively easy to minimize. The easy-to-minimize function G(x) is written as where g(x) must be an upper bound for the regularization term λφ(x) in the cost function F(x), as shown in Figure 3b. g(x) = x T Q k x + c(x k ), where c(x k ) is constant vectors independent of x; and Q k is a diagonal matrix, where "diag(.)" is the diagonal matrix operator and the notation './' denotes componentwise division. Therefore, the MM update in Equation (16) by minimizing Equation (18) with respect to x gives Substituting H with A −1 B obtains However, there are two problems, as follows [36]: (1) The update x k+1 (Equation (21)) is mathematically valid, but it may become numerically inaccurate because the entries of Q k go to infinity when the components of x k go to sparse (go to zero). (2) Its inverse matrix is not banded due to (AA T ) −1 , which has a high computational cost; so fast solvers cannot be used here. To address both issues, the matrix inverse lemma was used to alter the non-banded matrix to the banded matrix, given by Substituting Equation (22) into Equation (21) obtains It shows that the diagonal matrix Q −1 k cannot be infinity, even though x k is approaching the sparse signal solution. The iteration counter x k+1 depends on Q −1 k starts with the initial value x 0 = y. where x k ./φ (x k ) for each scalar value of its matrix for L1 norm is given by = \|x\| (25) and x k ./φ (x k ) for arctangent function, according to Equation (14), is given by to avoid non-convex optimization, the parameter tuner α is given by where 'std' is the standard deviation of H T w, and w is the white Gaussian noise [37]. Furthermore, reducing λ leads to an even noisier solution. Increasing λ leads to further attenuation of both the solution and noise. An implementation of the MM algorithm is given in Algorithm 1 to extract a sparse signal from a noisy received signal, where the elements of the diagonal matrix, x k ./φ (x k ), are denoted in Equation (26). (3) and (4). We illustrate an example in Figure 4 that, to extract the object time series from the noisy signal in Figure 2, the reflectivity series x was assumed to have four objects, x = δ (n − 30) + 0.7δ (n − 80) + 0.5δ (n − 100) + 0.3δ (n − 130). Given λ = 2, we compare the normal deconvolution by L2 norm Equation (11) to both sparse deconvolutions by the L1 norm Equation (12) and arctangent Equation (15). In Figure 4, the estimated sparse signal using the normal deconvolution with L2 norm regularization contained a great deal of noise, and was weak, rendering it inapplicable to real-world radar applications. Meanwhile, in Figure 5, we can see that the L1 norm solution is slightly attenuated compared with the arctangent solution, therefore, the sparse deconvolution using the arctangent regularization will be included as an efficient method to increase the detection performance for through-the-wall UWB radar, as described in the next section. Through-Wall UWB Radar with Sparse Deconvolution Based on Arctangent The sparse deconvolution with arctangent regularization is so efficient in extracting the sparse signal from the noisy reflected signal (raw data) that lowpass, bandpass, and smooth filters are no longer required [16][17][18][19][20][21][22]. Figure 6 illustrates the steps for extracting the sparse signal for through-wall UWB radars. In the first step, the noisy received signal was calibrated to remove the unwanted signals: [42][43][44][45]. Figure 7 illustrates the pre-and post-calibration (beyond t = 0) received signal. The calibrated received signal y(t) was mathematically expressed by where τ is the time index of unwanted signals (i.e., antenna coupling, upon-impact wall reflection, and inner-wall reflection), it can be calculated by where t 1 is the first wall reflection derived by "findpeaks" (Matlab command); factor 3 is a constant to allow sufficient time for upon-impact wall reflection and inner-wall reflection; factor 2 is round-trip delay; D wall is the wall thickness; v wall is the speed of the wave in the wall; ε r is the relative permittivity; and the permittivity (ε 0 ) and permeability (µ 0 ) of vacuum are 8.854187 × 10 −12 F/m and 4π × 10 −7 H/m [42]. Note that the singular value decomposition has been widely used to solve such problems, but it takes a long time to process during its matrix inversion. Second, before using the sparse deconvolution, the calibrated signal was down sampled to avoid running out of memory throughout N × N matrix inversion (long calculations), given r down is the down sampling ratio under the Nyquist sampling condition. In this flow chart, the down sampling was done by the integer factor method (e.g., "downsample" in Matlab). Besides, the N × N band matrices (A and B) must also be down sampled by the same r down , and the normalized angular frequency (ω 0 ) was recalculated by where f c is the center frequency, ∆t is the time resolution of the analog-to-digital converter (ADC), and r down is the down sampling ratio. Figure 8 depicts the experimental setup of the through-wall UWB radar scheme, corresponding to the block diagram in Figure 1. Table 1 tabulates the specifications of the experimental equipment and parameters. The experiments were carried out with a concrete wall approximately 20 cm in thickness. The concrete wall was fashioned from three columns of concretes. In Figure 8, the S-band UWB pulse (2-4 GHz) from the UWB source was fed into the power amplifier (PA) and to the Vivaldi Tx antenna. The reflected signal received by the Vivaldi Rx antenna was amplified by the low-noise amplifier (LNA) and sent to the oscilloscope. The oscilloscope captured the received signal and transferred it to a computer via a GPIB port. The data were then discretized by MATLAB 2018a for sparse deconvolution. Calibrating Recevied Signal When the system obtains the received signal from the oscilloscope, as shown below (example at 3 m), it will be calibrated by using Equation (28) to remove the wall and antenna responses. From Figure 9, the duration of the received signals captured by the oscilloscope was 35 ns, with a time resolution (∆t) of approximately 3.125 ps. The received data were discretized into 11,100 data points. According to the flow chart in Figure 6, the received signal was calibrated by Equation (28) with the zero-offset τ = 11 ns by computing Equation (29), with t 1 = 6 ns (first wall reflection), D wall = 10 cm (the wall thickness), ε r = 4.5 (the relative permittivity) [47], and v wall = 1.4132 × 10 8 m/s (the speed of the wave in the concrete wall). This calibration is easy and provides quite accurate evaluation, but requires a lot of parameters to work, which is suitable for the known material and thickness of the wall. In a realistic context, with variable obstacles and, sometimes, no obstacle at all, the antenna and wall coupling problems can quickly be removed by observing the setting time shift of the received signal [48]. Note that the singular value decomposition has been widely used to solve such a problem, but it takes a long time to process during its matrix inversion. Sparse Deconvolution Results Human range was extracted from the calibrated signal by using the sparse deconvolution based on the arctangent regularization, accoding to Algorithm 1. The A and B matrices in Algorithm 1 were calculated from Equations (3) and (4), fixed r = 0.9 and ω 0 = 0.2356 rad/sample, respectively. To avoid nonconvex optimization, the deconvolution parameters λ and α were 0.4 and 0.9/λ, respectively, by evaluating the Equation (27) [37]. The experimental results, as shown in Figure 10, were compared with L1 and L2 regularizations by varying a person's distance at 2, 2.5, and 3 m, as well as two person distances at (2 m, 3 m) and (2.5 m, 3 m). In Figure 10, the human detection performance of the through-wall radar scheme with the ordinary deconvolution (L2 norm) was unsatisfactory, while that of the radar schemes with sparse deconvolution (L1 norm and arctangent regularizations) could effectively locate the human subjects behind the wall. By comparison, the arctangent regularization was significantly higher than that with L1 norm. For the experiments with one participant standing at 2, 2.5, or 3 m (Figure 10a-c), the estimated behind-the-wall distances using sparse deconvolution with arctangent regularization were 2.1, 2.55, and 3.1 m, respectively. For the experiment with two participants standing (2 m, 3 m) and (2.5 m, 3 m) behind the wall (Figure 10d,e), the estimated behindthe-wall distances were (2.11 m, 3.2 m) and (2.55 m, 3.16 m). The distance error comes from the other effects of the wave propagation in the concrete wall. These issues are quite difficult to model, because the walls in real-world applications are inhomogeneous and not purely dielectric material [47,48]. Furthermore, for the experiment with two participants standing (Figure 10d,e), the ghost signal is probably difficult to remove due to the reflection of the electromagnetic wave scattering off the nearby participants. Technically, if both the wavelet H and the input signal x are unknown, this is the blind deconvolution problem. It estimates the wavelet model H of the subsurface layer and the transmitted pulse, which is useful for GPR and seismic data [4,5,[23][24][25][26][27][28]. In previous works [43,44,48], UWB radar algorithms for human detection need to capture the received signals at least 512 times per minute (60 s), for over 1 cycle vital sign signal, with a Nyquist sampling condition. The sparse deconvolution algorithm is used for reconstruing human range from only one received signal, and for faster detection. However, with this method, it is difficult to distinguish between a standing human and static objects. In real-world applications, humans have motions, so the detectable range is sufficient for obtaining their positions [41]; sparse deconvolution could be deployed in various through-obstruction applications with faster detection, especially in hostage rescue operations. Conclusions Through-the-wall UWB radar posits that the unknown object time series x is sparse (range domain), and is solved by the sparse deconvolution based on the arctangent regularization to induce sparsity more strongly than the L1 norm. The cost function of this sparse deconvolution model is also composed of the band matrices A and B, which provide a fast solution by the majorization-minimization (MM) algorithm. Moreover, the S-band UWB radar is intended for locating human subjects behind a wall, with the following step-bystep description: (1) calibration; (2) down-sampling; (3) designing the band matrices. To validate this, the through-wall UWB radar scheme with deconvolution based on L2 norm, L1 norm, and arctangent regularizations was experimentally applied to detect human subjects at different behind-the-wall distances, and experimental results were compared. The results showed that the human detection performance of the radar scheme with L2 norm regularization was poor. On the other hand, the radar scheme with L1 norm and arctangent regularizations could effectively detect the human subjects behind a wall. Nevertheless, the human detection performance of the through-wall UWB radar scheme with arctangent regularization was significantly higher than that with L1 norm regularization. Author Contributions: Conceptualization, A.R.; methodology, A.R. and P.P.; validation, A.R. and P.P.; formal analysis, A.R. and P.P; investigation, A.R. and P.P; writing-original draft preparation, A.R.; writing-review and editing, A.R. and P.P.; and funding acquisition, P.P. All authors have read and agreed to the published version of the manuscript.	4,803.4	2021-04-01T00:00:00.000	[ "Computer Science" ]
Assessing the difficulty of annotating medical data in crowdworking with help of experiments Background As healthcare-related data proliferate, there is need to annotate them expertly for the purposes of personalized medicine. Crowdworking is an alternative to expensive expert labour. Annotation corresponds to diagnosis, so comparing unlabeled records to labeled ones seems more appropriate for crowdworkers without medical expertise. We modeled the comparison of a record to two other records as a triplet annotation task, and we conducted an experiment to investigate to what extend sensor-measured stress, task duration, uncertainty of the annotators and agreement among the annotators could predict annotation correctness. Materials and methods We conducted an annotation experiment on health data from a population-based study. The triplet annotation task was to decide whether an individual was more similar to a healthy one or to one with a given disorder. We used hepatic steatosis as example disorder, and described the individuals with 10 pre-selected characteristics related to this disorder. We recorded task duration, electro-dermal activity as stress indicator, and uncertainty as stated by the experiment participants (n = 29 non-experts and three experts) for 30 triplets. We built an Artificial Similarity-Based Annotator (ASBA) and compared its correctness and uncertainty to that of the experiment participants. Results We found no correlation between correctness and either of stated uncertainty, stress and task duration. Annotator agreement has not been predictive either. Notably, for some tasks, annotators agreed unanimously on an incorrect annotation. When controlling for Triplet ID, we identified significant correlations, indicating that correctness, stress levels and annotation duration depend on the task itself. Average correctness among the experiment participants was slightly lower than achieved by ASBA. Triplet annotation turned to be similarly difficult for experts as for non-experts. Conclusion Our lab experiment indicates that the task of triplet annotation must be prepared cautiously if delegated to crowdworkers. Neither certainty nor agreement among annotators should be assumed to imply correct annotation, because annotators may misjudge difficult tasks as easy and agree on incorrect annotations. Further research is needed to improve visualizations for complex tasks, to judiciously decide how much information to provide, Out-of-the-lab experiments in crowdworker setting are needed to identify appropriate designs of a human-annotation task, and to assess under what circumstances non-human annotation should be preferred. Die Software war schwer zu bedienen. Die Software war gut strukturiert. Die Software war einfach aufgebaut. Part A: Personal data In the following, we will first collect some personal data about you. These data are needed for the statistical analysis of this study. Your person cannot be identified by answering these questions. A1. Please enter your age. A2. Please enter your sex: female -male A3. Please enter your course of study. A4. Please enter your native language. A5. Please enter your country of origin. A6. Are you left-handed or right-handed? A7. How much experience do you have in the following areas? Medicine-none, little, much or very much Data Mining-none, little, much or very much Image processing-none, little, much or very much Part B: Questions about the used graphical representation The following is a list of statements about the graphical representation used previously. Please check off the statements you agree with. Multiple check marks are possible. B1. The following questions refer to the tile-based configuration only. -The graphical representation was easy to understand. -The graphical representation was unnecessarily complex. -The graphical representation was too cluttered. -The graphical representation was clearly arranged. -The size of the graphical representation was pleasant and appropriate. -The individual elements of the graphical representation were pleasantly large. -Terms and designations of the graphical representation were easily understandable. -I could understand the graphical representation only with the help of the experimenter. -I consider the graphical representation to be useful. -It was easy for me to use the graphical representation. -With the help of the graphical representation I was able to achieve my work goal. -I could easily compare the instances with the help of the graphical representation. -I was able to quickly compare instances using the graphical representation. B2. The following questions refer to the parallel-based configuration only. -The graphical representation was easy to understand. -The graphical representation was unnecessarily complex. -The graphical representation was too cluttered. -The graphical representation was clearly arranged. -The size of the graphical representation was pleasant and appropriate. -The individual elements of the graphical representation were pleasantly large. -Terms and designations of the graphical representation were easily understandable. -I could understand the graphical representation only with the help of the experimenter. -I consider the graphical representation to be useful. -It was easy for me to use the graphical representation. -With the help of the graphical representation I was able to achieve my work goal. -I could easily compare the instances with the help of the graphical representation. -I was able to quickly compare instances using the graphical representation. -The software was designed to be understandable. -The software was difficult to use. -The software was well structured. -The software was simply structured. -The individual elements of the software were easily recognizable. -The texts within the software were easily understandable.The size of the texts was too small. -The terms and designations used within the software were easy to understand. -The task was clearly formulated. C2. Was the activity sensor perceived as a disturbance during the experiment? yes or no Part D: Feedback D1. Opportunity for criticism, praise and suggestions about the study and this survey. You have reached the end of the survey. Thank you for completing the questionnaire! I hope you enjoyed the study and I thank you for your time and your effort.	1,295	2021-07-29T00:00:00.000	[ "Computer Science" ]
The Difference Of Javanese And Indonesian Vocabularies In Preschool Age Children Background: Vocabulary is an important component of language aspect in children. Vocabulary mastery is a good predictor to see the language ability of children in further age. Speaking using more than one language gives children a broader experience to access language learning. This research was conducted to see the difference of vocabularies in Javanese and Indonesian languages. Methods: The research was conducted using cross-sectional design. Data collection was conducted by distributing questionnaire to parents. The sample design used was total sampling. Data collection was conducted on April-September 2019. Results: The result of analysis using Mann Whitney test shows that there is a difference of vocabularies between Javanese and Indonesian languages. Variance test shows that Javanese vocabularies are higher in quantity than Indonesian vocabularies. Conclusion: There is a difference of vocabularies between Indonesian and Javanese. The use of dominant language in daily life putatively contributes to the findings of research. Broader exploration should be conducted to see the comparison between first and second languages. ARTICLE HISTORY Received : March 17 th , 2021 Accepted : July 12 th , 2021 INTRODUCTION Vocabulary is the basic component of language. Vocabulary mastery is an important indicator to see whether or not child passes through normal language development. Specifically, vocabulary is used in diagnosing language problem (American Psychiatric Association, 2013;Bishop et al., 2017;Paul & Norbury, 2012;Shipley & McAfee, 2021). Limited vocabulary is the primary indicator to see whether or not child has limited language ability. The prediction of language ability at advance and academic levels can be seen from the history of vocabulary mastery in early childhood (Brignell et al., 2019;Kiliç, 2019;McDaniel et al., 2018). Vocabulary mastery gives children an opportunity to master further language ability. For example, to have sentence mastery or ability, a child should be able to use word appropriately based on the context of sentence (Rowe et al., 2012). Some factors affect vocabulary development, either internally or externally (Pratomo et al., 2016). Parents' role is one of factors affecting the child's vocabulary achievement significantly (Bingham et al., 2017;Richels et al., 2013;Rowea et al., 2016). The administration of language stimulus and model is a process passed through in language learning for preschool students (Pratomo et al., 2018). Parents' language use is a real model in child's language attainment. Language uttered by parents in the form of mother tongue is blueprint of language ability generally. It means that when mother uses Indonesian, child will follow her using Indonesian. Indonesia as a state with plural or multicultural populations has great language wealth. Indonesia has 737 languages used as primary and secondary languages, in addition to Indonesian. One of language types used widely is Javanese (Aji et al., 2019). The use of more than one language has considerable advantage. A child has an opportunity of gaining more vocabularies than those using one language (Pransiska, 2017). It is reported that individuals who can use more than one language has broader language complexity. Although the use of two languages has some advantages, the exploration to see the comparison of vocabulary in two languages cannot be done. The objective of research was to see the comparison of vocabulary size in two different languages. The difference arising will give theoretical and clinical implications particularly in speech therapy treatment. This study will focus on vocabulary in performance prediction especially when clinision treat vocabulary aspect in bilingual children. The purpose of study si to find the differences between bahasa Indonesian and Javaness Vocabulary. MATERIALS AND METHOD This research employed cross-sectional approach. Data collection was conducted in Taman Kanak-kanak (Kindergarten) Sri Juwita Hanum and RA Al Kautsar Mojosongo Surakarta. Research permission letter was obtained from the headmaster with letter numbers 064/SPb/ RA-AK/ VIII/ 2019 and 04/ KB-TK/ SJH/ VIII/ 2019. Data was collected on April-September 2019. The sample of research consisted of 4-6 years old students in TKA and TKB. The sample design used was total sampling. The sample size was 66. The sample is normal children that no has any communication problems. This study used the valid and reliable instrument. Each of word that can be uttered spontaneously is scored 1. Data analysis used descriptive statistic and bivariate statictis. To found deferences beetwen vocabulary, we used Mann Whitney Test. RESULTS The result of descriptive analysis indicates that majority respondents are less than 5 years old. Male dominates the distribution of respondents' sex. Data on the distribution of frequency is presented in table 1. Significance value ≤0.001 indicates that there is a statistical difference of vocabularies between Javanese and Indonesian. To see the comparison of words in Javanese and Indonesia, the mean scores of two languages are compared. The comparison of words in Javanese and Indonesian is presented in Table 3. The table above shows that mean score of Javanese vocabularies is higher than that of Indonesian vocabularies. DISCUSSION The result of data analysis shows that there is a difference of word vocabularies in Javanese and Indonesian. Significance value is less than 0.001. The result of comparison between mean scores indicates that vocabulary score of Javanese is higher than that of Indonesian. The hypothesis that the use of Javanese language is more dominant than that of Indonesian is confirmed. The use of language in house environment is an important key to the children's language learning activity (Bingham et al., 2017). Assumption that ethnocultural variable affects the development of children's vocabularies is confirmed. The development of cross-cultural linguistic aspect has been the material of clinical discussion in language disorder field (Connor, 2008). Domination of one language over others, particularly in mother tongue, is an early model developing more rapidly than second language (Kuo & Lai, 2006;Tager-Flusberg, 2015). Another finding proves that language activities of between parents and children is an important process in children language learning (Pratomo et al., 2016). If parents use Javanese, the process of learning vocabularies the children will receive is in Javanese. Children will process lexical process based on the model they have received (Rescorla et al., 2013). Although the result of research successfully finds the fact concerning the difference of vocabularies in Javanese and Indonesia, this finding cannot be generalized into an assumption that one language is different from another, particularly when comparing first and second language. A broader exploration needs to be done on the difference of cross-cultural vocabularies. The dominant use of language at home cannot be used as basic indicator to see the children's language ability. Multiple factor (multifactorial) analysis is required to see broadly the aspects affecting the development of vocabularies in children (AlHammadi, 2017). Clinical implication of research is that an analysis needs to be conducted on the assessment and the intervention based on ethnocultural study on children. Ethnocultural-based assessment and intervention are known to have more functional outcome (Shipley & McAfee, 2021). The clinical decision making not to use language choice based on clinical perception becomes important. Clinician should map detailed vocabulary including type of words and type of languages used. If the selection of targeted words is compatible to the children's contextual need, the intervention produced will be more functional (Gray & Yang, 2015). CONCLUSION This research successfully revealed the difference of vocabularies between Javanese and Indonesian. The result of analysis indicates that Javanese vocabularies have higher score than Indonesian ones. Javanese is putatively used more dominantly in family and social setting of children. The result of research can be used as a material of evaluation for clinicians to choose and to determine appropriately the targeted vocabularies functionally. The appropriate selection of targeted vocabularies gives an opportunity of achieving higher therapy outcome. This research still has some limitations. The sample size needing expansion is one of recommendations for further exploration. Multivariable analysis is required to see more in-depth the variables correlating with the development of children's vocabularies. Evaluation of vocabularies in direct face-to-face form is required to see actually the vocabulary mastery the children have. ACKNOWLEDGEMENT Thank to all institutions for the participation of this study.	1,857.8	2021-07-12T00:00:00.000	[ "Linguistics", "Education" ]
Forecasting of Chinese Primary Energy Consumption in 2021 with GRU Artificial Neural Network The forecasting of energy consumption in China is a key requirement for achieving national energy security and energy planning. In this study, multi-variable linear regression (MLR) and support vector regression (SVR) were utilized with a gated recurrent unit (GRU) artificial neural network of Chinese energy to establish a forecasting model. The derived model was validated through four economic variables; the gross domestic product (GDP), population, imports, and exports. The performance of various forecasting models was assessed via MAPE and RMSE, and three scenarios were configured based on different sources of variable data. In predicting Chinese energy consumption from 2015 to 2021, results from the established GRU model of the highest predictive accuracy showed that Chinese energy consumption would be likely to fluctuate from 2954.04 Mtoe to 5618.67 Mtoe in 2021. Introduction Energy is a vital resource needed for socio-economic development, and it is increasingly of concern to more and more governments and economic sectors because of its extensive application and the strong dependency on it in the processes of production and consumption [1].In recent years, with the rapid development of the Chinese social economy and increase of population, there has been a rapid upward trend in Chinese energy demand and consumption [2].Chinese main energy sources include hard coal, lignite, hydropower, oil, natural gas, geothermal, solar, wind, nuclear, etc., but efficiency of energy production and utilization is too low.In order to meet domestic energy demand, energy import trade volume is increasing year by year.Therefore, China should develop its own corresponding energy production plan to meet the rising domestic energy demand.To ensure energy security, it is important to predict annual energy consumption for a 5-to 10-year period to establish an appropriate energy plan.Energy consumption forecasting is affected by various aspects of socio-economic factors, among which the gross domestic product (GDP), population, import and export trade and other factors are particularly important [3]. The energy consumption model is usually based on historical consumption data and historical data related to energy consumption, such as the economy, population, climate, etc. [4].At present, energy consumption forecasting in the world has three mainstream research methods: planning models, economic models, and machine learning models.The planning model uses linear and nonlinear programming to find the parameters that fit based on historical data.It was O'Neill who first applied the planning model to predict energy consumption in US [5].Meanwhile, this method has also been applied in coal, oil, natural gas, power demand and other fields [6].The economic model combines energy demand with other microeconomic variables and realizes the prediction of future energy demand through the inherent interaction between economic variables.The choice of economic variables is the key to the predicted accuracy [7].The machine learning model breaks through the constraints of the original mathematical calculation in terms of the accuracy of prediction.It realizes and identifies the relationship between the data characteristics through artificial intelligence and realizes the prediction of future energy consumption through the modeling of a large amount of historical data.It identifies the relationships between the various data features by means of artificial intelligence.Therefore, the machine learning model realizes the prediction of future energy consumption based on the training of a large number of historical data.Moreover, there are many models of machine learning.The application of energy consumption includes Autoregressive Integrated Moving Average (ARIMA) model [8], Artificial Neural Network (ANN) model [9], Ant Colony Optimization (ACO) model [10], Particle Swarm Optimization (PSO) model [11] and so on. In view of the dynamic change of energy consumption, Gated Recurrent Unit (GRU) can effectively solve the problem of error caused by the spatiotemporal evolution of energy consumption.It has gating units that modulate the flow of information inside the unit.Compared with the original machine learning method, GRU belongs to a deep learning method, as it can use the memory units in a network to deal with any data sequence of input.Therefore, the ability to learn time series of GRU is greatly superior [12].The GRU may not only study the time series of long spans but also automatically determine the optimal time lag for prediction.In recent years, GRU has been successfully applied to handwriting recognition, human motion identification and robot control, etc. [13], but it is rarely applied in the field of economic forecasting.In this study, we selected three energy consumption forecasting models: multivariable linear regression (MLR), support vector regression (SVR) and Gated Recurrent Unit (GRU).By comparing these three models, we verified the superiority of the GRU model in the simulation of energy consumption from 2008 to 2015 in China.Then, we designed various scenarios to forecast Chinese primary energy consumption from 2015 to 2021.The results will help government to develop a reasonable energy plan. Energy Consumption Forecast In recent years, scholars from all over the world have studied the prediction of energy supply and consumption in the country and the region [14].Sözen (2006) employed the artificial neural network method to obtain the formula to predict the net consumption of energy.The results showed that the error of the net consumption of energy consumption obtained via artificial neural network method was very small [15].Deka (2016) compared five different forecasting technologies using economic and demographic factors to simulate US energy needs with in-depth discussion [16].Torrini (2016) proposed a fuzzy logic approach to extract rules from input variables and to provide Brazil's long-term annual electricity demand forecast [17].Philip (2012) used ARDL and PAM to measure the short-term and long-term influencing factors of energy consumption in Ghana and forecasted Ghana's energy consumption in 2020 [18].Gokhan (2015) predicted Turkey's primary energy consumption (PEC), which provided a predictive derivative model of population, gross domestic product (GDP) and energy consumption by regression analysis [19].Some scholars combine energy consumption with carbon dioxide emissions to establish a correlation forecasting model.Hasiao (2012) applied the improved nonlinear gray model (Bernoulli) to analyze the characteristics of carbon dioxide emissions, energy consumption and actual output in China and to establish a predictive model of numerical iteration [20].Pani (2010) applied correlation analysis to study the correlation of energy consumption, GDP and carbon emissions [21].Wenying (2015) conducted a bottom-up analysis of energy consumption and carbon dioxide emissions from the Chinese steel industry [22].Jain's (2014) findings suggested that the sensor-based energy prediction model was suitable for multi-family residential buildings [23].Wang (2011) analyzed the impacts of implementing new and expected energy and environmental policies with the Long-range Energy Alternatives Planning (LEAP) modeling tool [24].Blanca Moreno (2016) used the combined model of grey neural network and input-output to predict primary energy consumption in the Spanish economic sector [25].Xie (2015) applied the optimized single variable discrete grey prediction model to predict China's total energy production and consumption, and proposed a new Markov method based on the quadratic programming model to predict the trend of China's energy production and consumption structure [26]. Multiple Linear Regression Multiple Linear Regression (MLR) is an important method in multivariate statistical analysis.It makes it possible to estimate the future regression coefficients and model accuracy without sampling the future system.At present, MLR is widely used in the research of many disciplines.Prakasvudhisarn (2015) predicted the electricity consumption of Thailand using the multiple linear regression and ANN models [27].Abuella (2015) presented a multiple linear regression analysis model for solar power probabilistic forecasting [28].Cleland (2010) applied multiple linear regression to usefully analyze the total energy consumption in the New Zealand food manufacturing industry [29].Amral (2008) investigated the short-term load forecast of the demand of the South Suleai power system with the multiple linear regression method and concluded that the short-term load forecasting multiple linear regression model had been relatively easy to develop and regularly update, and was widely used in commercial computing software [30].In Tuaimah's (2014) research, the multiple linear regression method was used to present a short-term load forecast for Iraq's power system requirements [31].Torkzadeh (2014) applied multiple linear regression & principal component analysis (MLR-PCA) as the approach to predict weekly electrical peak load of Yazd city and concluded that the error of this proposed method was quite small [32].Rahman (2014) presented a method for characterizing river water quality with the analysis of multiple linear regression [33].Mata (2011) showed a comparison between the MLR and ANN models to characterize dam behavior under environmental loads [34].Abushikhah (2011) proposed multivariable linear and non-linear regression, which used an hourly daily load to predict the next year's hourly load, and the results obtained using the proposed method suggested that its performance was close [35]. Support Vector Machine The Support Vector Machine (SVM) is an evolutionary algorithm for data exploration, and is an algorithm with a high prediction accuracy [36].Support vector machines can be used to solve nonlinear programming problems, and can predict time series.At present, support vector machines have been widely used in planning, classification, nonlinear fitting and other fields.Its use is grounded in its superiority for solving nonlinear problems and, it has also applied to forecast energy consumption.Li (2009) applied SVM to predict the air conditioning energy consumption of office buildings.The results showed that the accuracy of SVM model prediction was higher than that of the BP neural network [37].Hou (2009) predicted the air conditioning energy consumption of the (Heating, Ventilating, and Air-Conditioning) HVAC system, and the results showed that the SVM model was more accurate than the (Autoregressive Integrated Moving Average) ARIMA model [38].Jain (2014) used the SVR model to predict energy consumption in New York's multi-tenant buildings.Meanwhile, verifying temporal and spatial changes in particulate concentrations can have an impact on the accuracy of the forecast [23].Wang (2015) tried to apply an instance-weighted variant of the SVM with both 1-norm and 2-norm formats to deal with the class imbalance problem [39].Furthermore, Zhang (2013) studied the application of support vector machine in face recognition [40]. Gated Recurrent Unit The Gated Recurrent Unit (GRU) changes the means of original supervising machine learning and solves the problem by carrying the memory unit of the forgotten mechanism.While the GRU deep learning model has drawn attention of late, its application is currently still relatively rare, and is mainly concentrated in computer-related areas.Le (2017) proposed a Gated Recurrent Unit (GRU) based on the Recurrent Neural Network (RNN) to construct an energy decomposition classifier with deep learning, and applied the method to training the model with the UK DALE dataset.From the experiment, Le concluded that the deep learning method was very effective for non-invasive load monitoring (NILM) [41].Chung (2014) evaluated Recurrent Neural Networks (RNN) with three widely used recurrent units: a traditional tanh unit, a Long Short-Term Memory (LSTM) unit and a Gated Recurrent Unit (GRU).Finally, Chung confirmed the superiority of the Gated Recurrent Unit (GRU) [42].Jozefowicz (2015) compared the GRU and LSTM models and found that the GRU model was able to achieve comparable results to the LSTM model on multiple issues, while the GRU model was easier to train [43].Zhou's experiments (2016) showed that GRU had some advantages in learning recurrent neural networks with stable performance and relatively few parameters [44].Tang (2016) conducted an investigation on recurrent approaches to cope with question detection, and then built different RNN and bidirectional RNN (BRNN) models to extract efficient features based on gated recurrent units (GRU) at segment and utterance levels.Tang concluded that the particular advantage of GRU was that it can determine a proper time scale to extract high-level contextual features [45].Rana's (2016) speech experiments with eight different types of noise showed that the run time of the GRU was reduced by 18.16%, and was comparable to the long term short-term memory of the most popular recurrent neural network [46].Huang (2017) verified the use of GRU-ELC units with the most advanced performance on three standard scene marker datasets.This comprehensive experiment showed that the new GRU-ELC unit facilitated the problem of on-site labeling because it could more effectively encode the longer context dependency in the image than the traditional RNN unit [47]. Multiple Linear Regression Model The Multiple Linear Regression model is a method used to deal with the complex relationship between an output variable and multiple explanatory variables.The purpose of its analysis is to predict the output variables with the value of multiple explanatory variables.The main limitation of the model is that the correlation between the variables changes with time and space [48].Assuming an output variable is y i , and some explanatory variables are x i , then the relationship between the output variable and the explanatory variable can be expressed as: Meanwhile, x i,h is the value of the hth explanatory variable for the year i, b 0 is the constant term of the plan, b h is the parameter of the hth explanatory variable, h is the number of all explanatory variables, y i is the value of the output variable for year i, ŷi is the estimated value of the output variable for year i, e i is the prediction error, where e i can be defined as: Support Vector Regression Model The Support Vector Regression model obtains an approximate function g(x) from G = {(x i , y i )} N i=1 in the historical data sample of the correlated variable, which is already known.The data x is mapped to a high dimension feature space by nonlinear method, and then the linear programming is carried out in this feature space [49]. In Equation ( 4), φ i (x) is the characteristic variable, b and w i as coefficients that can be estimated from the data.In this way, the nonlinear programming of a low-dimensional input space can be deduced into a linear programming of high-dimensional feature space.The coefficient w i can be obtained with the minimum function: In Equation ( 5), λ is a normalized constant, and function \| f (x i ) − y i \| ε can be defined as: The minimum function can also be expressed as follows: in addition, the kernel function explains the scalar product of the D i dimensional feature space: The coefficients α i and α * i can be obtained by the following formula: The constraint is Gated Recurrent Unit Neural Network Model The Gated Recurrent Unit (GRU) neural network model adapts to the problem of dependence on a variety of time scales by setting all kinds of cycle units [43] that modulate the flow of information with the gate unit.Assuming that the input of the model is expressed as x = (x 1 , x 2 , • • • , x T ), the logical calculation process is shown in Figure 1. Assume that the activation function h j t of GRU is a function related to time t, which takes the linear interpolation between the activation function h j t−1 at the previous time point and the candidate activation function h j t , which is : At the same time, the update gate z j t determines whether the unit updates the activation function or maintains the proportion and the number of the existing activation functions.The update gate z j t is as follows: The whole calculation process is to sum the existing state and the state of the update calculation, but the GRU model can't control the range of state updates, but every calculation updates all of the states once. The calculation of the candidate activation function h j t is similar to that of the simple RNN calculation, and its computational function is: Among which r t is the reset gate, ⊗ is the vector product.When the reset door is closed (r j t ≈ 0), the contents of the input sequence can be read while the past state is forgotten.The reset gate r j t is calculated as follows: The tanh function above has been very maturely and widely used in some research [45]. Energies 2017, 10, 1453 6 of 15 Assume that the activation function j t h of GRU is a function related to time t , which takes the linear interpolation between the activation function j t h 1 − at the previous time point and the candidate activation function which is : At the same time, the update gate j t z determines whether the unit updates the activation function or maintains the proportion and the number of the existing activation functions.The update gate j t z is as follows: The whole calculation process is to sum the existing state and the state of the update calculation, but the GRU model can't control the range of state updates, but every calculation updates all of the states once. The calculation of the candidate activation function is similar to that of the simple RNN calculation, and its computational function is: Among which t r is the reset gate, ⊗ is the vector product.When the reset door is closed Data Sources In order to verify the predictive accuracy of the above three models for Chinese primary energy consumption, in accordance with the research of Zong (2009) [50], we chose five variables: gross domestic product (GDP), population, import trade volume, export trade volume and energy consumption.Among these variables, the gross domestic product (GDP), population, import trade volume and export trade volume were regarded as independent variables, while energy consumption was a dependent variable.The data selected was from 1965 to 2015, and the data of the four variables of gross domestic product (GDP), population, import trade volume and export trade volume were derived from the World Development Indicator [51].The Chinese primary energy consumption data was from the "BP World Energy Statistics Yearbook" [52].These data are shown in Table 1.In this paper, the total number of data samples was 51. Analysis of Results MLR and SVR models are deterministically mathematical methods, and stable results can be obtained according to the formulas given above.The GRU model is a deep learning neural network, and further constructs the model structure.The GRU model has three layers, including an input layer, a hidden layer, and an output layer.The input layer consists of four input variables: GDP, population, import, and export.The hidden layer consists of three GRU units with time steps of 1, 4, and 6, and each GRU unit contains 32 cells, and the output layer is the characteristic variable of primary energy consumption.The structure of the model is shown in Figure 2. Analysis of Results MLR and SVR models are deterministically mathematical methods, and stable results can be obtained according to the formulas given above.The GRU model is a deep learning neural network, and further constructs the model structure.The GRU model has three layers, including an input layer, a hidden layer, and an output layer.The input layer consists of four input variables: GDP, population, import, and export.The hidden layer consists of three GRU units with time steps of 1, 4, and 6, and each GRU unit contains 32 cells, and the output layer is the characteristic variable of primary energy consumption.The structure of the model is shown in Figure 2. The training and testing of the GRU model were completed using the Keras kit on the PYTHON platform, in which the optimizer was set as "RMSprop", the loss function was set as "MAPE", the loss rate was set as "0.0001", the epoch was set as 2000.In order to prevent the phenomenon of overfitting, the calculation process was added to the validation part, determining whether or not it appears to be the best model parameters. The main research goal of this paper is to compare the accuracy of the three models of MLR, SVR and GRU for medium term Chinese primary energy consumption forecasting.In order to express the advantages and disadvantages of the three models, the paper takes MAPE (mean absolute percentage error) and RMSE (root mean square error) as the results for error.The two error formulas are as follows: The training and testing of the GRU model were completed using the Keras kit on the PYTHON platform, in which the optimizer was set as "RMSprop", the loss function was set as "MAPE", the loss rate was set as "0.0001", the epoch was set as 2000.In order to prevent the phenomenon of over-fitting, the calculation process was added to the validation part, determining whether or not it appears to be the best model parameters. The main research goal of this paper is to compare the accuracy of the three models of MLR, SVR and GRU for medium term Chinese primary energy consumption forecasting.In order to express the advantages and disadvantages of the three models, the paper takes MAPE (mean absolute percentage error) and RMSE (root mean square error) as the results for error.The two error formulas are as follows: At the same time, the y in Equations ( 14) and ( 15) represents the actual primary energy consumption in China, while ŷ in the model represents the predicted value of Chinese primary energy consumption. The experiments using MLR and SVR can be effectively performed, but when performing the experiment using GRU, a very interesting problem emerges.When using all the training data to do the GRU prediction experiments, the error of the predicted MAPE is 14, which blocks the purpose of improving the accuracy of prediction.However, the error of the predicted MAPE is 5.63 when doing the GRU prediction experiment with the data of the first 8 years (2000)(2001)(2002)(2003)(2004)(2005)(2006)(2007), allowing the emergence of the optimal prediction model parameters.The main reason for this situation is that the data input variables are in a state of annual growth.Recent data provides more information for the forecast results, whereas the earlier data will have a detrimental effect on the forecast.The training and test errors in Table 2 show that the GRU model has a higher accuracy in the prediction with the MAPE and RMSE indicators than that of MLR and SVR model in the comparison of the forecasting errors in Chinese primary energy consumption.Figure 3 shows the comparison between the actual value of primary energy consumption in China from 1965 to 2015 and the predicted values of various models.In summary, the GRU model is the best method of research to predict Chinese primary energy consumption; this model will be used in the prediction of Chinese primary energy consumption in the medium term. At the same time, the y in Equations ( 14) and ( 15) represents the actual primary energy consumption in China, while y ˆ in the model represents the predicted value of Chinese primary energy consumption. The experiments using MLR and SVR can be effectively performed, but when performing the experiment using GRU, a very interesting problem emerges.When using all the training data to do the GRU prediction experiments, the error of the predicted MAPE is 14, which blocks the purpose of improving the accuracy of prediction.However, the error of the predicted MAPE is 5.63 when doing the GRU prediction experiment with the data of the first 8 years (2000-2007), allowing the emergence of the optimal prediction model parameters.The main reason for this situation is that the data input variables are in a state of annual growth.Recent data provides more information for the forecast results, whereas the earlier data will have a detrimental effect on the forecast.The training and test errors in Table 2 show that the GRU model has a higher accuracy in the prediction with the MAPE and RMSE indicators than that of MLR and SVR model in the comparison of the forecasting errors in Chinese primary energy consumption.Figure 3 shows the comparison between the actual value of primary energy consumption in China from 1965 to 2015 and the predicted values of various models.In summary, the GRU model is the best method of research to predict Chinese primary energy consumption; this model will be used in the prediction of Chinese primary energy consumption in the medium term. Chinese Primary Energy Consumption Forecasts Based on Different Scenarios After comparing the model errors, the GRU model was used to predict Chinese primary energy consumption from 2016 to 2021.An attempt was made to reduce the uncertainty of the forecast by setting appropriate scenarios, and by using the different scenarios with suitable forecast data.For Chinese Primary Energy Consumption Forecasts Based on Different Scenarios After comparing the model errors, the GRU model was used to predict Chinese primary energy consumption from 2016 to 2021.An attempt was made to reduce the uncertainty of the forecast by setting appropriate scenarios, and by using the different scenarios with suitable forecast data.For gross domestic product (GDP), the forecast data of Chinese GDP published by the International Monetary Fund (IMF) [53] was employed.According to the World Population Prospects (2015) [54], the Chinese population will reach 1.424 billion by 2030.Using Equation (16) to calculate the annual growth rate of the population, it can be inferred that the Chinese approximate growth rate of the population from 2016 to 2021 will be 0.25%.Since there is no authoritative estimate of import and export trade in the world, the initial growth rate, average growth rate and minimum growth rate can only be calculated based on historical data of growth.Taking into account the potential for the country's ongoing transformation and upgrading of the industry, the lowest growth rate is set at the lowest non-negative growth rate from 1965 to 2015.Due to the uncertainty of the import and export trade volume, the forecast of Chinese primary energy consumption from 2016 to 2021 is best set as four possible scenarios, as is shown in Table 3.According to the calculated method of the above data, Chinese gross domestic product and population estimations from 2016 to 2021 are shown in Table 4.The estimated data on Chinese import and export trade levels at different growth rates is shown in Table 5. In Equation ( 16), CAGR is the annual growth rate, V(t 0 ) is the value of the beginning year, V(t n ) is the value of the ending year, the number of years in the whole phase is n. The data of four independent variables are substituted into the trained GRU energy consumption forecasting model; the results of the four different scenarios are compared in Figure 4.The data of four independent variables are substituted into the trained GRU energy consumption forecasting model; the results of the four different scenarios are compared in Figure 4.Although more scenarios would more accurately assess the behavior of the predictive model in predicting possible Chinese energy consumption, the four scenarios were chosen from what realities have been appropriately assumed and computationally proven to achieve superior results, and hopefully represent the spectrum of possible consequences. In Scenario 1, a negative growth of Chinese primary energy consumption is predicted from 3013.96Mtoe in 2015 to 2970.19 Mtoe in 2021, cutting back 1.45% in 6 years.Calculated according to Equation ( 16), the annual growth rate is −0.24%. In Scenario 2, Chinese primary energy consumption forecast indicates an increase of 86.42% from 3013.96Mtoe in 2006 to 5618.67 Mtoe in 2021.According to Equation ( 16), the annual growth rate is 10.9%.Chinese primary energy consumption increases fastest in this scenario. In Scenario 3, the forecast result of Chinese primary energy consumption suggests an increase of 23.6% from 3013.96Mtoe in 2015 to 3725.2 Mtoe in 2021 and an annual growth rate of 3.6% based on Equation ( 16). Scenario 4, China's primary energy consumption forecast reveals a decrease of 3054.96Mtoe Although more scenarios would more accurately assess the behavior of the predictive model in predicting possible Chinese energy consumption, the four scenarios were chosen from what realities have been appropriately assumed and computationally proven to achieve superior results, and hopefully represent the spectrum of possible consequences.16), the annual growth rate is 10.9%.Chinese primary energy consumption increases fastest in this scenario. In Scenario 3, the forecast result of Chinese primary energy consumption suggests an increase of 23.6% from 3013.96Mtoe in 2015 to 3725.2 Mtoe in 2021 and an annual growth rate of 3.6% based on Equation (16). Scenario 4, China's primary energy consumption forecast reveals a decrease of 3054.96Mtoe from 2015 to 2954.04 Mtoe in 2021, with a decrease of 1.99% over six years.The annual growth rate is -0.33% according to Equation ( 16), and Chinese primary energy consumption witnesses the fastest decline in this scenario.To sum up, the four scenarios predict that Chinese energy consumption in 2021 will fluctuate between 2954.04 Mtoe and 5618.67Mtoe. The growing energy demand requires the government to make the right decisions in terms of energy planning.If energy planning results in incorrect underestimates of energy needs, there will be a shortage of energy supply, resulting in an energy deficit.Due to the strong correlation between energy consumption and greenhouse gas emissions, the prediction of future energy consumption can also affect the Chinese reaction to climate change.Through the accurate prediction of energy consumption, environmental managers can not only determine the major sources of carbon emissions, but can also determine whether all kinds of energy have an impact on climate change.Reliable energy forecasts can ensure the energy security of the country, achieving the sustainable development of energy and economy. Conclusions Chinese primary energy consumption forecasting is a key element of the success of national energy security and energy planning.Based on economic and demographic factors, three kinds of Chinese energy forecasting models-multivariable linear programming, support vector planning and gate recurrent unit-have been established for forecasting the energy consumption in China from 2016 to 2021.Through the results of the study, the following three important findings were obtained: 1. Deep learning is the hotspot of current research, and in the GRU there are internal relations between the four economic variables (gross domestic product (GDP), population, import trade volume, export trade volume) and energy consumption.The four economic variables can be used to forecast the primary energy consumption in China; 2. The GRU model is a model based on long and short memory for learning time series data. Compared with the MLR model and the SVR model, the GRU model is superior for the processing of time series data, and the average absolute percentage error of the predicted result can be as low as 5.63.However, when applying this model, the choice of the amount of training data is a key factor in accurate prediction.In particular, for the prediction of macroeconomic variables, recent data is more important to the final forecast result, due to uncertainties in socio-economic change; and 3. The GRU model is used to forecast energy consumption in China from 2016 to 2021, with a finding that Chinese energy consumption in 2021 will fluctuate between 2954.04 Mtoe and 5618.67Mtoe. The proposed model could be one of the best techniques in deep learning.Although this is the first study that applies the GRU model in the prediction of Chinese primary energy consumption, there are more predictive testing technologies and methods that can be implemented.Two directions in the practice of the forecasting can be further pursued.First, continue to enhance the model structure and parameter settings of the GRU forecast method to increase the accuracy of the final energy consumption forecast; and second, select other economic variables related to energy consumption for the energy consumption forecast. Figure 1 . Figure 1.The computing logical of the Gated Recurrent Unit (GRU). Figure 1 . Figure 1.The computing logical of the Gated Recurrent Unit (GRU). The 51-item data sample was used to divide the test samples from 15% of the total sample.The training samples were mainly used to modify the planning model, and the test samples were mainly used to judge the accuracy of the model.The experiment used 43 data items from 1965 to 2007 as training samples and 8 data items from 2008 to 2015 as test samples. Figure 2 . Figure 2. The structure of GRU neural network. Figure 2 . Figure 2. The structure of GRU neural network. Table 1 . Primary energy demand and indicator data of China. Table 2 . Comparison of forecasting error for various models. Table 2 . Comparison of forecasting error for various models. Table 3 . The forecasting scenarios of Chinese energy consumption (2015-2021).The initial growth rate the growth rate calculated in 2015, the average growth rate is the average growth rate from 1965 to 2015, and the minimum growth rate is the non-negative minimum growth rate from 1965 to 2015. Note: Table 5 . China's import and export estimation (2016-2021).In Scenario 1, a negative growth of Chinese primary energy consumption is predicted from 3013.96Mtoe in 2015 to 2970.19 Mtoe in 2021, cutting back 1.45% in 6 years.Calculated according to Equation (16), the annual growth rate is −0.24%.In Scenario 2, Chinese primary energy consumption forecast indicates an increase of 86.42% from 3013.96Mtoe in 2006 to 5618.67 Mtoe in 2021.According to Equation (	7,530.6	2017-09-21T00:00:00.000	[ "Computer Science", "Economics", "Engineering", "Environmental Science" ]
Cooperative Multi-Agent Interaction and Evaluation Framework Considering Competitive Networks with Dynamic Topology Changes In recent years, the problem of reinforcement learning has become increasingly complex, and the computational demands with respect to such processes have increased. Accordingly, various methods for effective learning have been proposed. With the help of humans, the learning object can learn more accurately and quickly to maximize the reward. However, the rewards calculated by the system and via human intervention (that make up the learning environment) differ and must be used accordingly. In this paper, we propose a framework for learning the problems of competitive network topologies, wherein the environment dynamically changes agent, by computing the rewards via the system and via human evaluation. The proposed method is adaptively updated with the rewards calculated via human evaluation, making it more stable and reducing the penalty incurred while learning. It also ensures learning accuracy, including rewards generated from complex network topology consisting of multiple agents. The proposed framework contributes to fast training process using multi-agent cooperation. By implementing these methods as software programs, this study performs numerical analysis to demonstrate the effectiveness of the adaptive evaluation framework applied to the competitive network problem depicting the dynamic environmental topology changes proposed herein. As per the numerical experiments, the greater is the human intervention, the better is the learning performance with the proposed framework. Introduction Reinforcement learning is concerned with the problem of maximizing the rewards of learning objects that need to be effectively controlled within a defined environment. The more complex the human's behavior and system configuration are, the more difficult the problem is, and the longer it takes for the learning object to learn. These problems can be solved through additional tasks such as pre-learning or preprocessing; however, these tasks are not very effective because preprocessing and pre-learning take a long time to complete and corrupt the learning data. Therefore, to effectively solve complex and difficult reinforcement learning problems, methods for solving problems through intuitive and professional human intervention have been proposed. The benefit of learning with the help of humans is that as the learning objects are intended to resemble humans, their learning goals can be clearly defined, and learning can be completed quickly without pre-learning or preprocessing. However, existing studies on reinforcement learning focusing on learning problems with human help were concerned with the single model problem of a simple environment, static simple network and single agent. As such, this study focuses on the two more competitive network topologies and it consists of multi-agent (e.g., combat or soccer) in a complex manner. To learn effectively, learners can model human evaluations in the form of quantitative real numbers to successfully achieve their goals and apply adaptive policies to humans. Herein, we propose a framework whereby the test results are precisely and strategically calculated. The following section provides a review of existing research and literature related to human-machine evaluations and reinforcement learning. Section 3 proposes the cooperative human-machine evaluation framework and its algorithm. The implementation of the proposed framework and the performance analysis of the experimental example are presented in Section 4. Finally, the concluding remarks and future scope of work are presented in Section 5. Background and Literature Review This study applies reinforcement learning to the effective integration of human-machine evaluations toward a goal and compared previous studies on reinforcement learning on the basis of several categories. The learning object may be a robot, a system, or a game, and reinforcement learning is being investigated in various fields that require complex learning toward goals. There are a number of control strategies and relevant applications. Roman, et al. [1] proposed an adaptive control using fuzzy components to control a town crane. Zhang, et al. [2] applied a semi-global state synchronization method to actuator control under unknown nonuniform input delay. When the learning object is a robot, these robots are generally classified as robot arms, humanoid robots, and industrial robots [3][4][5]. Robots with a high degree of freedom usually require complicated and difficult calculations because they require a goal that can perfectly mimic human behavior [6]. Additionally, robots learn how to behave similarly to humans so that they can also learn to expend a collaborative effort to help with human tasks or purposes [7]. The results from these studies can be applied to large-scale industrial applications that begin with learning and imitating simple movements, achieving goals through collaboration with humans, and applying them to real industries [8,9]. Reinforcement learning is also often used in areas such as games, where the goal of learning is clearly stated. This is because the algorithm of reinforcement learning is rewarded and updated according to the actions taken by the learning object. For example, when the object performs a mission in the game, the reward it obtains depends on its actions, which affects the final result and the target value [10]. Studies have also been conducted to learn systems, besides robots and games. System network and communication are complex and computationally heavy because of the need for a system that can optimize performance and goals. Furthermore, the computational load of networks that optimize paths such as an escape is heavy because it is necessary to derive the learning results by computing problems in real time. Therefore, learning methods that effectively deal with the real-time computation of a problem have been investigated [11][12][13]. Moreover, various methods have been studied regarding the problem of reinforcement learning to deal with values of measuring devices such as sensors. There are methods for learning via effectively estimating, interpolating, and approximating values before preprocessing, such as in the case of sensor values [14,15]. In addition, some studies have designed adaptive algorithms to ensure that learning objects are updated more reliably and effectively as they learn [16][17][18]. Table 1 summarizes the algorithms, contexts, and ideas proposed in previous studies and identifies the learning methods used in each study. These studies classified whether the learning object learned via the designed system itself or with human cooperation and also classified the learning objects into robots (robot arms and humanoids), games, or systems. In relation to the existing studies that have explored mimicking of human behavior, research related to the task of table lifting performed by humans and humanoid robots has been conducted. Humanoid robots move by predicting human movements through prediction-based algorithms, and the reliability of the prediction is arrived at by the motion predictor [19]. A study was conducted on dynamically walking and balancing robots that use reinforcement learning to learn dynamic gait without prior learning. It aimed to solve complex control problems with respect to the robot's motion control by mapping the motion space from discrete to continuous areas. The balanced learning method, which used the movement of the robot arms and legs to move the zero-moment point in the robot sole, can keep the biped robot in a static stable state. It showed that the robot can learn how to improve motion in terms of the walking speed in the proposed way [20]. There have also been studies exploring flipping of handkerchiefs and folding of t-shirts. Emphasis was placed on the learning of the robotic arms, and deep reinforcement learning was investigated to learn complex policies through high-level observations such as typing. Because deep reinforcement learning requires a large number of training samples, a method that improves the sample efficiency and learning stability with fewer samples by combining the characteristics of smooth policy updates with automatic feature extraction of deep neural networks was proposed [21]. Similarly, there has been a study involving learning robots that pick up and classify objects. To be able to interact with dynamic objects in an unstructured environment, robots need manipulation capabilities to handle the confusion, change, and object variability. The robots learn a closed-loop policy that maps depth camera inputs to motion commands and compare different approaches to make the problem easier to deal with, including the reward formation, curriculum learning, and the use of pre-trained policies with reduced work to pre-start tasks. Training the robots with heuristics helped achieve the desired behavior [22]. Collaborative robots are widely used in hybrid assembly tasks involving intelligent manufacturing. Research related to teaching-learning-collaboration models, where collaborative robots can learn through human demonstrations and support human partners in the working environment, has been put forth. This approach allows humans to control the robot using natural language instructions according to their personal work preferences. The robot then learns the assembly demonstrations from human using the maximum entropy inverse reinforcement learning algorithm and updates the task-based knowledge with the optimal assembly strategy. In the collaborative process, the robot can leverage the learned knowledge to actively support people in collaborative assembly work. As such, in the case of humanoid robot learning, an object is trained by imitating or pre-learning human behavior. Hence, pre-learning is essential at all stages of simulation or learning [23]. Robotic applications of reinforcement learning often undermine the autonomy of the learning process to achieve practical training time in real physical systems. To overcome this problem, recently developed deep reinforcement learning algorithms based on off-policy training for deep Q-functions can be extended to complex 3D manipulation tasks, and efficiently implement deep neural network policies to train the actual physical robot [24]. In contrast to the case involving interaction between the self-learning agent and the environment, it is recommended to train an agent manually using an evaluative reinforcement framework to update the rewards of human trainer feedback on the current state. Based on the evaluation of the agent's recent performance, the trainer can offer rewards in any form of representation that can be mapped to scalar values. The agent's goal is to act in consideration of the current state by receiving feedback from the human to choose the action that will receive the most rewards. For this purpose, the agent incorporates the reward function obtained from humans and selects the behavior that is expected to receive the highest reward. By learning a reward model for human feedback, agents can act on their goals even when there is no human feedback and choose a task that is expected to maximize their rewards when human feedback is provided. The agent attempts to maximize the immediate reward, assuming that the human trainer had already considered the impact of each behavior, when receiving feedback. This problem is consistent with supervised learning. Assuming each action is a training sample, for the selected action at time t, the state s t at the current time t and the state s t+1 at the next time t + 1 are considered attributes of the sample, and the reward of the human trainer for that action is considered a label [25]. Delivered by the corrective advice communicated by human framework, a reinforcement learning method models new human feedback based on manually training agents through evaluative reinforcement framework. The author used a binary signal in the action domain of the agent. Further, the reward value was updated by appropriately utilizing past human feedback [26]. Existing studies to effectively learn through human evaluation mainly deal with simple, uncomplicated problems of a single network. In a recent study dealing with the situation where two or more network topologies compete and coexist, an effective algorithm is proposed in which human evaluation is adaptively updated [27]. In this paper, propose an algorithm that strategically updates rewards until a stable situation is reached, dealing with an environment where two or more dynamically changing networks compete with each other and adaptively update human evaluation. Human evacuation frameworks used in emergencies have focused on modeling and simulating emergencies. Recently, reinforcement learning has been investigated for real-time shortest path calculation methods that can be used in emergency situations [28]. This method has both advantages and disadvantages in terms of two approaches: agent-based modeling and equation-based modeling. Because agent-based models are slow but accurate, studies are being conducted to produce fast linear models based on reinforcement learning. In addition, the formula-based modeling method is fast, but the range of erroneous measurements is rather large [29]. In-depth reinforcement learning methods have been studied for mining and processing large amounts of data in a dynamically changing environment [30]. Additionally, studies dealing with the resource allocation problems of a large amount of resources and devices have been conducted using deep neural networks to learn the environment and make decisions regarding the allocation problems according to network conditions, such as service latency and requirements [31]. Previous studies related to effective object learning methods for tasks that require human expertise remain useful to overcome the limitations of preprocessing learning, which requires large amounts of computation and numerous samples. Reinforcement learning has evolved in recent years in terms of its applicability to complex sequential decision-making tasks which are generally modeled using the Markov decision process (MDP). Reinforcement learning is a methodology for deciding the next action after inferring the reward based on the achievement of the goal, which is in turn based on the agent's current state and the interactions between the systems exhibiting the current state [32]. The typical learning method among traditional reinforcement learning algorithms is the Q-learning method which infers the maximized reward value by calculating the Q-function, which is a behavioral value function as shown in Equation (1), at each time period [33]. In (1), s t is the state at time t, a t is the behavior of time t, r t+1 is the reward of time t + 1, and α is the learning rate in the range of (0,1). The closer the α is to 1, the more likely it is to induce learning to place a greater emphasis on the current situation and to adjust the share of rewards for future predicted behavior through a discount rate, γ [34]. Methodology of Cooperative Human-Robot Evaluation This study is concerned with the problem of artificial intelligence soccer game shown in Figure 1. This is an example of a special case that differs from those previously investigated. In this paper, to deal with the environment in which two or more dynamically changing multi-agents in networks compete and coexist with each other, designed an artificial intelligence soccer game similar to the artificial intelligence basketball game covered in [27]. Appl. Sci. 2020, 10, x FOR PEER REVIEW 5 of 16 deep neural networks to learn the environment and make decisions regarding the allocation problems according to network conditions, such as service latency and requirements [31]. Previous studies related to effective object learning methods for tasks that require human expertise remain useful to overcome the limitations of preprocessing learning, which requires large amounts of computation and numerous samples. Reinforcement learning has evolved in recent years in terms of its applicability to complex sequential decision-making tasks which are generally modeled using the Markov decision process (MDP). Reinforcement learning is a methodology for deciding the next action after inferring the reward based on the achievement of the goal, which is in turn based on the agent's current state and the interactions between the systems exhibiting the current state [32]. The typical learning method among traditional reinforcement learning algorithms is the Qlearning method which infers the maximized reward value by calculating the Q-function, which is a behavioral value function as shown in Equation (1), at each time period [33]. In (1), is the state at time , is the behavior of time , is the reward of time + 1, and α is the learning rate in the range of (0,1). The closer the α is to 1, the more likely it is to induce learning to place a greater emphasis on the current situation and to adjust the share of rewards for future predicted behavior through a discount rate, γ [34]. Methodology of Cooperative Human-Robot Evaluation This study is concerned with the problem of artificial intelligence soccer game shown in Figure 1. This is an example of a special case that differs from those previously investigated. In this paper, to deal with the environment in which two or more dynamically changing multi-agents in networks compete and coexist with each other, designed an artificial intelligence soccer game similar to the artificial intelligence basketball game covered in [27]. The learning goal of the soccer game in this study is to pass the ball by avoiding the red team soccer players who are obstructed by the blue team soccer players. In real-life football games, players move dynamically on the field and try to pass the ball successfully to a player from the same team. As such, players in football games described as multi-agent form a network topology as shown in Figure 1, and these network topologies show the two teams competing against each other. Each The learning goal of the soccer game in this study is to pass the ball by avoiding the red team soccer players who are obstructed by the blue team soccer players. In real-life football games, players move dynamically on the field and try to pass the ball successfully to a player from the same team. As such, players in football games described as multi-agent form a network topology as shown in Figure 1, and these network topologies show the two teams competing against each other. Each team is given a network topology at time t, and there is a state as a reference when it is expressed as a reinforcement learning problem considering the situation and environment, and there is another state that follows and competes similarly. To deal with this reinforcement learning problem in this special case, this study defines the Q-Function as Equation (2) and calculates the maximum human evaluation reward value by calculating every time cycle period, t. State s 2 t that coexists with state s 1 t is not an independent state, but it is affected by the same action in the same environment. Therefore, it can be defined as Equation (3). Traditional reinforcement learning problems often involved a single network topology with the learning objects. However, this study covers two or more network topologies such as a dynamically changing competitive network. In a single network topology, the reward policy could be learned through a system reward update calculated from the learning object. Conversely, in this study, to establish a reward policy for a complex network in which two networks are learned in the same environment, the procedure for obtaining rewards through human evaluation and the reward of the system calculated from the learning object are properly applied. Here, a procedure is involved that updates rewards through human intervention, while existing studies [25,26] have mapped human evaluation through a binary process; moreover, this study advances evaluation feedback in a more quantitative and accurate form. The human rating is entered as a real value between 0 and 1, and the reward policy is updated in a much more accurate form. The human evaluation reward, h t , obtained in the process of learning a number of times, can be modelled by a Gaussian distribution as shown in Equation (4), where µ is the evaluation mean, and σ is the distribution of standard deviation. In general, the mean of the Gaussian distribution is estimated h t using Equation (5), where n is the sample size and is the number of quantitative rewards from human evaluation that has been learned for a number of times t. The standard deviation of the Gaussian distribution is estimated using Equation (6).σ In repeated learning, when a correction value evaluated by a human is normalized through the distribution, as described above in Equation (6), h t is calculated and h t is estimated. Subsequently, the reward value obtained through the estimated human evaluation via repeated learning is subjected to a procedure for updating it adaptively, as shown in Equations (7) and (8). Here, Equation (7) can be used to minimize the loss function F h t+1 that defines the difference between the estimate h t+1 during time t + 1 and estimate h t the time t. When h t+1 is updated, the step size η should be set differently for each estimate iteration; hence, η is increased when the variation in the estimated human evaluation reward value is small, and η is reduced when the variation in the reward value through the estimated human evaluation is large. G t stores the sum of squareds of the gradient values of h t updated through estimation in time t. In the case of updating h t , the process proceeds to a size inversely proportional to the root of G t in the existing step size η. The reason for this is that if the estimated human evaluation value h t has changed significantly, it will move less. Conversely, if the estimated human evaluation value h t has changed less, it will shift more. At this time, is a small value of between 10 −4 and 10 −8 , meant for preventing division by zero. As per Equation (8), the reward value, h t , adaptively updated by the human evaluation is compared with the human evaluation reward, h t , obtained from learning in the present iteration. As shown in Equation (9), h t is updated again via adopting the greater of the two values. In Equation (9), the correction value h t updated adaptively by human evaluation should be appropriately calculated with the reward value of the system derived from the update of the learning object to determine the final reward value, h * t . In Equation (10), h * t is calculated as the reward, r t , of the system derived from the updating of the learning object, and the correction value, h t , adaptively updated via human evaluation, and it is calculated using the appropriate policy rate, δ, by the human intervene policy rate. On application of the quantitative evaluation of human being proposed in this study adaptively, the framework that established the system reward value and the appropriately calculated policy through the update of learning object is shown in Figure 2. Appl. Sci. 2020, 10, x FOR PEER REVIEW 7 of 16 step size . The reason for this is that if the estimated human evaluation value ℎ has changed significantly, it will move less. Conversely, if the estimated human evaluation value ℎ has changed less, it will shift more. At this time, ϵ is a small value of between 10 and 10 , meant for preventing division by zero. As per Equation (8), the reward value, ℎ , adaptively updated by the human evaluation is compared with the human evaluation reward, ℎ , obtained from learning in the present iteration. As shown in Equation (9), ℎ is updated again via adopting the greater of the two values. ℎ = max (ℎ , ℎ ) In Equation (9), the correction value ℎ updated adaptively by human evaluation should be appropriately calculated with the reward value of the system derived from the update of the learning object to determine the final reward value, ℎ * . In Equation (10), ℎ * is calculated as the reward, , of the system derived from the updating of the learning object, and the correction value, ℎ , adaptively updated via human evaluation, and it is calculated using the appropriate policy rate, , by the human intervene policy rate. On application of the quantitative evaluation of human being proposed in this study adaptively, the framework that established the system reward value and the appropriately calculated policy through the update of learning object is shown in Figure 2. Table 2 shows the complete algorithm involved in the proposed cooperative human-machine evaluation framework. First, the human intervene rate, , the discount rate, γ, and learning rate, , are defined (lines 1-3). The agent observes the new state (line 6). Then, receive the exact quantitative Table 2 shows the complete algorithm involved in the proposed cooperative human-machine evaluation framework. First, the human intervene rate, δ, the discount rate, γ, and learning rate, α, are defined (lines 1-3). The agent observes the new state (line 6). Then, receive the exact quantitative evaluation h from the human (line 10), the update of the adaptive human evaluation reward (lines [11][12][13][14][15]. Finally, when the reward for the Q-function is updated and determined, the Q-function is calculated (lines 17 and 18). This process is repeated for a given time, T, to proceed with the learning process (lines 4). System Implementation and Experimental Results This section explains in detail the cooperative human-machine evaluation framework introduced in Section 3. The implementation of the proposed framework and the numerical analyses using the proposed software program are presented in this section. The program software is developed using MATLAB© (made by MathWorks, Natick, MA, USA)and C++ language. The implemented software program includes several panels and two graph windows. The panels that make up the program deal with the network topology as shown in Figure 3. They form a network in two dimensions as shown in Figure 3a,b and in three dimensions as shown in Figure 3c and are expressed in a distribution to maximize visibility. using the proposed software program are presented in this section. The program software is developed using MATLAB© (made by MathWorks, Natick, MA, USA)and C++ language. The implemented software program includes several panels and two graph windows. The panels that make up the program deal with the network topology as shown in Figure 3. They form a network in two dimensions as shown in Figure 3a,b and in three dimensions as shown in Figure 3c and are expressed in a distribution to maximize visibility. The graph window and the windows that show the calculated values are shown in Figure 4. The windows have several functions. In a soccer game depicted in dynamically changing competitive network topologies as shown in Figure 4a, the players' status changes and the system's reward is calculated and shown. At the same time, the user monitors two competitive network topologies depicted as a football game and inputs feedback on the behavior of the current state with a real number between 0 and 1. The evaluation feedback from the human is calculated by estimation and adaptive reward evaluation calculation, as shown in Table 2. Afterward, the reward from the system in Figure 4a and the human evaluation in Figure 4b are calculated according to the given level of human intervention, and the reward value according to the behavior of the learning object is updated. Figure 5 shows a program that implements the proposed cooperative human-machine evaluation framework for the reinforcement learning of competitive network topologies. The panels and windows that make up the program are described in detail in Table 3. For instance, in a soccer game, there exist both network topologies: one and its enemy, coexisting in the same environment. At the same time, the user monitors two competitive network topologies depicted as a football game and inputs feedback on the behavior of the current state with a real number between 0 and 1. The evaluation feedback from the human is calculated by estimation and adaptive reward evaluation calculation, as shown in Table 2. Afterward, the reward from the system in Figure 4a and the human evaluation in Figure 4b are calculated according to the given level of human intervention, and the reward value according to the behavior of the learning object is updated. Figure 5 shows a program that implements the proposed cooperative human-machine evaluation framework for the reinforcement learning of competitive network topologies. The panels and windows that make up the program are described in detail in Table 3. For instance, in a soccer game, there exist both network topologies: one and its enemy, coexisting in the same environment. calculated according to the given level of human intervention, and the reward value according to the behavior of the learning object is updated. Figure 5 shows a program that implements the proposed cooperative human-machine evaluation framework for the reinforcement learning of competitive network topologies. The panels and windows that make up the program are described in detail in Table 3. For instance, in a soccer game, there exist both network topologies: one and its enemy, coexisting in the same environment. In this study, a soccer game was supposed, and an experiment was carried out through the example of learning to deliver the ball to allies while avoiding the enemy network that obstructed the flow of the ball. The players of both teams do not stay in a fixed position but have a distribution over a defined range and move dynamically; thereby, changing the network topology. The human evaluator visually confirms the degree of obstruction of the path for the ball to be delivered by the enemy and enters a real number between 0 to 1 accordingly. When a game agent passes a ball In this study, a soccer game was supposed, and an experiment was carried out through the example of learning to deliver the ball to allies while avoiding the enemy network that obstructed the flow of the ball. The players of both teams do not stay in a fixed position but have a distribution over a defined range and move dynamically; thereby, changing the network topology. The human evaluator visually confirms the degree of obstruction of the path for the ball to be delivered by the enemy and enters a real number between 0 to 1 accordingly. When a game agent passes a ball without any intervene, an expert (human) evaluates the pass closed to 1 using the developed system. The human's evaluations are subjective and influenced by the locations of the opponents and other reasons. The evaluation score modifies the reward values using Equations (9) and (10). Then, the game agent changes the passing action to obtain higher reward using the provided algorithm shown in Table 2. In the iterative learning, the human evaluation reward is calculated and updated by the provided calculations of the reward value of the system itself and the human evaluation reward value by the cooperative human-machine evaluation framework. Figure 6 shows the results of learning a soccer game represented by two networks that coexist using the proposed cooperative human-machine evaluation framework. There are two algorithms prepared for comparisons with the proposed cooperative human-machine evaluation framework. Table 3. Implemented Cooperative Human-Machine Evaluation Software Program. Type Function Detailed Function Configurations Panel Application of competitive network (e.g., Soccer Game). -Define an iteration -By defining the defense range and attack range, specify the range of the following actions. Panel Network topology. -Representation of a coexisting network topology only with current statues of both players' group. -Monitoring both players' (e.g., Blue and Read team) network topology. -Representation of a coexisting network topology with current statues and the following future statues. -Representation of probability distributions depicting the following statues. Window Adaptive human evaluation. -Input window for human reward evaluation. -Calculation of adaptive human evaluation. -The network topology learned and evaluated by humans in each iteration -It is updated by the adaptive human evaluation strategy between the estimated values and the current evaluated values in the iteration. Window Reward calculation and updates. -Analyzing integrated reward. -Measurement of system performance. -The rewards obtained for each learning are calculated - The average value is updated, and system performance is visualized. The graph of Figure 6a represented by the symbol "o" signifies the reward of cooperative human-machine evaluation framework in each iteration as calculated by the algorithm of Table 2 and the graph of Figure 6a represented by the symbol "" signify the value of the average value of the reward of a cooperative human-machine evaluation framework. Figure 6b is an algorithm in which the human evaluation is evaluated in binary (of 0 and 1). Moreover, the graphs of Figure 6b represented by the symbol "o" signify the reward of the simple evaluation in binary in each iteration, and the graph of Figure 6b represented by the symbol "" signifies the average reward of evaluated in binary strategy. Figure 6c is an algorithm in which the traditional MDP method without human intervention. The graphs of Figure 6c represented by the symbol "o" means the reward of the MDP method in each iteration and the graph of Figure 6c represented by the symbol "" signify the average reward of evaluated in the MDP method. As seen in Figure 6, the proposed cooperative human-machine evaluation achieved the fastest convergence. This result is considered to be the result of human intervention that made quick decisions with specialized knowledge, unlike the way of updating only the reward of the system composed of existing learning objects. This study proposed a strategic method that aims to adaptively and quickly update the human evaluation scale and converge quickly to the maximum value. Table 4 shows how the proposed cooperative human-machine evaluation framework differs in learning performance according to the degree of human intervention. The results are summarized according to the parameter of the degree of human intervention rate δ. The problem addressed in this experiment is that two networks each have 10 nodes and coexist in the learning environment simultaneously. The problem is solved in three cases by applying the proposed cooperative human-machine evaluation framework. The first case solved the problem only with the system reward of the network topology existing inside the environment without any human intervention. As the training iterations were repeated, the reward value increased, and the system learned to converge to the maximum; however, the performance was poor and needed enough iterations. On the other hand, in the second case, if the policy for updating reward is updated by applying the reward calculated through the system's internal reward and the proposed cooperative human-machine evaluation framework at a 50% rate, convergence tended to occur quickly to the maximum reward value. In the last case, where a 90% rate was applied, the highest Q-function value was obtained at the same number of learning while converging to the maximum reward value as the target. This is shown by comparison in Figure 7. Table 4. Comparisons of three different experiment scenarios using the proposed framework. method in each iteration and the graph of Figure 6c represented by the symbol "" signify the average reward of evaluated in the MDP method. As seen in Figure 6, the proposed cooperative humanmachine evaluation achieved the fastest convergence. This result is considered to be the result of human intervention that made quick decisions with specialized knowledge, unlike the way of updating only the reward of the system composed of existing learning objects. This study proposed a strategic method that aims to adaptively and quickly update the human evaluation scale and converge quickly to the maximum value. Table 4 shows how the proposed cooperative human-machine evaluation framework differs in learning performance according to the degree of human intervention. The results are summarized according to the parameter of the degree of human intervention rate δ. The problem addressed in this experiment is that two networks each have 10 nodes and coexist in the learning environment simultaneously. The problem is solved in three cases by applying the proposed cooperative humanmachine evaluation framework. The first case solved the problem only with the system reward of the network topology existing inside the environment without any human intervention. As the training iterations were repeated, the reward value increased, and the system learned to converge to the maximum; however, the performance was poor and needed enough iterations. On the other hand, in the second case, if the policy for updating reward is updated by applying the reward calculated through the system's internal reward and the proposed cooperative human-machine evaluation framework at a 50% rate, convergence tended to occur quickly to the maximum reward value. In the last case, where a 90% rate was applied, the highest Q-function value was obtained at the same number of learning while converging to the maximum reward value as the target. This is shown by comparison in Figure 7. These results may vary depending on the assumed conditions and parameters. However, the proposed cooperative human-machine evaluation framework is effective in the complex network topologies in the context of human evaluation intervention using reinforcement learning. In addition, a quantitative empirical study was conducted from two perspectives to show that the method proposed in this study is effective. As shown in Figure 8, when learning in three different ways during the same time, the time when the pass success rate of the Artificial Intelligence (AI) soccer game reaches more than 95% was compared. The result of learning the pass of players in the AI soccer game using the method of strategically updating the reward in the form of real number through human evaluation, the method proposed in this study, shows a pass success rate of up to These results may vary depending on the assumed conditions and parameters. However, the proposed cooperative human-machine evaluation framework is effective in the complex network topologies in the context of human evaluation intervention using reinforcement learning. In addition, a quantitative empirical study was conducted from two perspectives to show that the method proposed in this study is effective. As shown in Figure 8, when learning in three different ways during the same time, the time when the pass success rate of the Artificial Intelligence (AI) soccer game reaches more than 95% was compared. The result of learning the pass of players in the AI soccer game using the method of strategically updating the reward in the form of real number through human evaluation, the method proposed in this study, shows a pass success rate of up to 96.6%. The binary number reward method achieved a pass success rate of 86.6% within the same time, and when learning using the Markov Decision Process method, a pass success rate of 62.6% was achieved during the same time. Using the method proposed in this study, strategically updating the reward in the form of a real number through human evaluation, the time required for the pass success rate of players in the AI soccer game to reach 95% was t = 29. In contrast, the binary number reward method took t = 73 to reach the 95% success rate of the pass. When learning using the Markov Decision Process method, the time it takes for the pass success rate to reach 95% is t = 93. These results may vary depending on the assumed conditions and parameters. However, the proposed cooperative human-machine evaluation framework is effective in the complex network topologies in the context of human evaluation intervention using reinforcement learning. In addition, a quantitative empirical study was conducted from two perspectives to show that the method proposed in this study is effective. As shown in Figure 8, when learning in three different ways during the same time, the time when the pass success rate of the Artificial Intelligence (AI) soccer game reaches more than 95% was compared. The result of learning the pass of players in the AI soccer game using the method of strategically updating the reward in the form of real number through human evaluation, the method proposed in this study, shows a pass success rate of up to 96.6%. The binary number reward method achieved a pass success rate of 86.6% within the same time, and when learning using the Markov Decision Process method, a pass success rate of 62.6% was achieved during the same time. Using the method proposed in this study, strategically updating the reward in the form of a real number through human evaluation, the time required for the pass success rate of players in the AI soccer game to reach 95% was t = 29. In contrast, the binary number reward method took t = 73 to reach the 95% success rate of the pass. When learning using the Markov Decision Process method, the time it takes for the pass success rate to reach 95% is t = 93. Conclusions As reinforcement learning is applied to several goal-oriented systems, the difficulty of problems and the complexity of the calculations increase, and various methods have been proposed to solve them. In the process of researching algorithms that solve these problems and lead to good performance, methods of learning are studied by adding accurate and quantitative evaluation through human intervention with expert knowledge and experience in reinforcement learning process. This trend has led to the need for pre-processing or pre-learning to make learning objects resemble human behavior or appearance in reinforcement learning. Existing studies involving human intervention at the same time as learning are relatively scarce and instead relied on prior learning. This study proposed a new method of updating the rewards of the system obtained in the process of learning by learning objects and the rewards derived from the evaluation of humans with expertise on the problem. This framework also proposed an adaptive strategy to update rewards resulting from a stable and effective human evaluation. In addition to that, if the learning object was dealing with a simple and independent form, this study deals with the problem of complex network topology. This type of problem suggests the need for a cooperative human-machine evaluation proposed in this study. The proposed framework was implemented as a software program that supports the cooperative human-machine evaluation framework to demonstrate efficiency. Therefore, the effectiveness is demonstrated by comparing the results in various scenarios. The proposed multi-agent framework can be applied to cooperative tasks between human and machines, such as human-robot interaction, autonomous car driving, and artificial intelligence-based industrial tasks. In particular, it contributes to fast leaning process using multi-agents' evaluations. However, the provided framework is limited from the fact that human's evaluation interface and timings are crucial for faster training processes. In addition, various industrial tasks require different human evaluation methodologies. For this reason, the provided framework and its implementation have to be modified with the objectives of applications. Further studies can consider the problem of determining more effectively the degree of human intervention required to update the combination of rewards via human evaluation and rewards from system learning. In addition, the provided framework can be applied to several real-life applications and scenarios. In a problem similar to the dynamic change competitive network topology discussed in this study, human intervention can be learned more quickly and efficiently.	9,812.6	2020-08-23T00:00:00.000	[ "Computer Science" ]
FEM Modelling of Weld Damage in Continuous Cold Rolling of MIG/MAG Butt-Welded Stainless Steel Strips Weld line fracture of butt-welded strips in stainless steel continuous rolling has been studied by numerical simulation using ForgeNxT®. The simulation plan includes weld line geometry as well as weld metal constitutive model parameters. The damage criterion used is the non-dimensional Latham & Cockroft function near the singular points of the weld line. Results are confronted to qualitative observations of fracture frequency and fracture initiation loci on the rolling line. Not surprisingly, the protrusion height of the weld line with respect to the strip top surface is found to be a major factor of risk. A second one pertains to the nature of the alloys used: due to the mushroom-like weld line cross-section geometry, damage at the top surface junction of the base metal and the weld metal becomes critical when the weld metal is harder than the base metal. Finally, on the rolling line investigated, prior to rolling properly speaking, the strip goes through a scale-breaking unit followed by acid pickling to eliminate oxides; the multiple, reverse plastic bending applied there contributes significantly to ductile damage, completing the explanation of why the fracture initiation locus is always found to be the same on the rolling line. Introduction A few stainless-steel cold rolling mills are designed as continuous, 2-stand tandem mills, requiring butt-welding of successive strips. This introduces weak points and indeed, breaks are occasionally observed. In response, either the rolling stand must be opened to avoid rolling the weld, resulting in more out-of-gauge metal; or a break may occur, and the mill must be stopped. Both result in extra costs and productivity loss, which justifies a mechanical study of weld damage. Breaks always occur at 1 st stand exit and at a specific location, the top surface singular point between the weld line and the base metal of the upstream strip (Fig. 1). Break frequency depends on rolled alloy, ferritic/austenitic. The purpose of this paper is to unveil the reasons behind this behavior. Although continuous rolling of carbon and alloyed steels has been in use for decades, there are very few studies in the open scientific literature on the question of the weld line behavior [1]. More abundant are the patents on electric flash butt-welding or Laser butt-welding of plain carbon steel strips for continuous rolling (e.g. [2][3]). Butt-welding may be installed for the continuity of different stages of the cold rolling plant, continuous pickling-rolling, continuous annealing, hot dip galvanizing. The most critical is however when the weld must be plastically deformed in the rolling mill. Documents [1][2][3] point to the main risk factors: poor alignment of the strip ends due to waviness, excessive height of the weld bead, base metal / weld metal hardness ratio, formation of a weak Heat Affected Zone (HAZ) and of course, depending on the strips to be joined, brittleness of the weld metal. 99.9% reliability is claimed in recent suppliers' brochures [4]. Yet, breaks occur. Process Description Welding is the first operation on the line. Hot rolled coils, 2 to 3 mm thick, are butt-welded together for continuity by Metal Inert Gas / Metal Active Gas arc welding (MIG/MAG). Whatever the strip alloy, austenitic or ferritic, an austenitic steel wire is used as the weld metal, for operational simplicity. Strip extremities are sheared, carefully positioned a few mm apart and molten weld metal is poured in the interval. This gives the shape visible in Fig. 1; the "funnel-like" shape inside the welded strip is due to melting of the strip metal in the top surface area by the excess of weld metal which forms the protruding bead. The exact shape of the molten metal zone is therefore not completely controlled and is reproducible in its general features only. Of course, a HAZ forms, with mechanical properties slightly different from the hot rolled metal, but its effect is left for future work. Figure 1: Cross section of a real weld line after the scale-breaker. Strip metal: hot rolled AISI 304L, weld metal: filler wires ER 310 or ER 316LSi. Following welding and prior to rolling properly speaking, several operations are devoted to oxide removal. It starts with the scale-breaker, followed by two-face shot-blasting which completes oxide fragmentation in order to improve the efficiency of the chemical etching to follow. Rinsing tanks close this part of the line. Of interest here is the scale-breaker, which operates an alternate bending under tension (Fig. 2), resulting in superficial plastic strain. Rollers are idle, the entrainment is provided by the strip tension from the rolling mill. Shot blasting certainly adds further superficial strain and damage but this one should be uniform, so that it will not be dealt with in this paper. Next comes the two-stand rolling mill. Work roll diameter is small (120 mm), the rolling speed is quite low for cold rolling (20 m/min or 0.33 m.s -1 ). Strong strip tensions are applied, the front tension being of the order of twice the back tension. Reductions ~35% per pass are performed under oil lubrication. Only the first stand will be dealt with here as all breaks occur at its exit. The rest of the line is briefly sketched here for completeness only, as it has nothing to do with weld fracture: continuous annealing, final etching (electrochemical, then chemical), skin pass rolling and roller levelling to optimize strip shape and surface state, and finally coiling. 570 Achievements and Trends in Material Forming Fracture characteristics. Fracture initiation locus. Initiation seems to be always from the top surface, upstream side of the weld line, at least in the few tens of cases investigated occasionally along the years. Fracture propagates mostly along the interface (or in the HAZ), sometimes bifurcates into the base metal (Fig. 3). The interface itself seems not to be a particularly weak zone. Fracture frequency. Quasi-systematic fracture after pass 1 has been found in the past for the ferritic grade, which is welded using an austenitic wire (explaining this is one of the purposes of the work). Therefore, in practice, the mill is systematically opened for the weld line transit, so that no recent observation is available. Experimental fracture observations are therefore mostly on austenitic grades. Thinning due to bending during scale-breaking. A peculiar observation at the exit of the scalebreaker is a strip thinning of 3-5% all along the strip and 7-10% over the ~5 mm next to the weld line, mainly on the upstream side. This effect can be referred to bending under tension [5,6]. It has been shown in the course of the present study that it does not result in any significant weakness, contrary to the singular points. Its importance lies in the fact that this is a feature which allows checking strip behavior in the scale-breaker. Design of numerical experiment. Thus, only qualitative information was available, mainly on break frequency and fracture locus: a full quantitative comparison was impossible. A parametric study has therefore been carried out, in which the relative risk of fracture has been estimated using the Latham and Cockroft damage function [7]. The first process parameter is weld line geometry: top and bottom bead protrusion height, molten metal wetting angle. As fracture frequency depends on rolled alloy, the base metal / weld metal hardness ratio has been taken as the second, material parameter. As a first stage, only damage from the first rolling pass has been considered. It has proved sufficient to explain fracture locus in the case of the ferritic steel. As this was not the case for the austenitic grade, a study of the scale-breaker has been added, showing a quite significant plastic strain and damage. Finally, the two stages, scale-breaker and rolling, have been chained, resulting in a correct prediction of the maximum fracture risk locus. Numerical Models ForgeNxT®. All computations were carried out using ForgeNxt® [8], a quasi-static implicit Finite Element Method (FEM) software based on velocity-pressure formulation (v,p). It uses tetrahedron (in 3D) and triangle (in 2D) mini-elements, with linear interpolation of pressure and linear + bubble function interpolation of velocity [9,10]. It offers automatic remeshing. However, neither bending nor rolling result in large distortion; moreover, using remeshing first, some inconsistencies were attributed to the field transfer stage. Remeshing has therefore been disactivated in this study. The real problem shows some 3D features, mainly: • weld line is in fact oblique, oriented 5° from the transverse direction. • The MIG/MAG process uses two torches starting from the drive side and the operators' side and meeting halfway: the bead is generally thicker at strip center. However, a 2D plane strain approximation has been used here for CPU cost reasons; the thicker central weld bead will be dealt with as a parametric variation instead. Key Engineering Materials Vol. 926 571 The thermomechanical coupling has not been activated here. Temperature increase is small (low rolling speed) and negligible impact on mechanical properties has been assumed. This could be questioned for unstable austenitic grades due to the sensitivity of martensitic transformation to temperature, but the investigation of this feature is left for future work. Strip characteristics. Geometry. Strip thickness is 2.5 mm for ferritic grade AISI 441, 2.15 mm for austenitic grade AISI 304L. As weld bead shape forbids symmetry ( Fig. 1 and 4), the whole thickness is meshed. The bead shape and dimensions have been chosen as "averages" of several measured ones -significant variations have been mentioned already. Meshing. The problem contains singularities (interface, corner points) where a fine mesh is needed. High sensitivity to the mesh has been found indeed. Furthermore, bending induces strain and stress gradients which are essential to capture. In the following therefore, a mesh size of 0.1 mm is used most of the time in the vicinity of the weld, as shown in Fig. 4. As for the bimaterial character, the single domain strategy was kept: a global geometry is meshed, then the area of the weld metal is defined by an inserted contour and its own mechanical properties are activated there. This single domain character means that neither sliding not fracture is allowed along the interface: the former is impossible in practice anyway, the latter will not be modelled explicitly, only the risk estimated through a damage function. Mechanical properties. Elastic-plastic behavior is assumed. The stress-strain curves based on tension tests performed on the hot rolled metal are shown in Fig. 5 as full lines and represented by Eq.(1) ( ̅ is the equivalent plastic strain and � the von Mises stress): A key point for the following is that, due to partial martensitic transformation, the strain-hardening capacity of the austenitic grades is much larger than that of the ferritic grades. Achievements and Trends in Material Forming On the weld metal (austenitic), only hardness measurement was possible. The shape of the strainhardening curve was copied from the base metals by using the hardness ratio. Thus, for the austenitic grade, the strip hardness is 215 HV, the weld metal (ER 310) hardness is 195 HV (see Fig. 1). Just multiplying parameter A by the hardness ratio 195 / 215 results in the curve "Hv 195" in Fig. 5a. It is used in the following for the weld metal, but also in the parametric study on the austenitic strip mechanical properties. For the latter purpose, another curve "Hv 245" has been built in order to vary more widely the strip / weld metal hardness ratio, known to be an important factor in weld line breaks. The ferritic strip hardness is 195 HV. It is welded with austenitic ER310 as stated before, which has the same initial hardness but work-hardens much more, reaching a state where the weld metal yield stress is above the strip yield stress. This proved decisive for metal flow and damage risk. Overall, the hardness ratio has been scanned in the interval [0.7-1.4] for the ferritic as well as the austenitic case, in order to include, in both cases, situations with harder strip and others with harder weld metal. Examples of these parametric variations are given in Fig. 5b as dotted lines. Scale-Breaking unit. Geometry. The rollers, rigid and idle, are meshed finely to ensure a good geometrical description. In the simulation, their meshes do not rotate but rollers are given a tangential velocity equal to the line velocity. As there is no slip which could dissipate energy, zero friction is assumed for simplicity. Boundary conditions. The whole scale-breaker is more than 1.5 m long. Modelling it in one stage would require a very long meshed strip coupon leading to unreasonable computing times. This is why it has been split into 3 stages corresponding to the 3 "units" of Fig. 2. Thanks to this, the meshed strip length could be reduced to 800 mm. In each, a back tension is applied on the upstream side, the downstream side is given the line velocity. In each section, a flat strip is positioned between the rollers in the opened position, then the latter are vertically displaced to apply bending before the horizontal pull is applied on the strip. This requires clipping the residual curved length for the next unit. Care has been taken not to disturb the mechanical fields in so doing. Boundary conditions. A strip length of 30 mm has been found sufficient for the rolling stage modelling. It allows reaching steady state before the weld line is rolled and resuming steady state once it has been rolled. The strip is imposed a purely horizontal movement at exit (zero vertical component) due to the constraint applied by the second stand. But the upstream end is left free to move vertically as it will be shown that significant bending occurs when the weld bead first contacts rolls. The required strip tensions are applied in the horizontal direction. Friction has been identified by fitting the measured roll load in the steady state (far from weld line): friction factor is equal to 0.3. Analysis of rolling without prior mechanical history. General sketch of the weld line behavior. The weld line contacts the upper roll first, the upstream part bends downward so that the root of the weld line, below the bottom strip surface, in turn contacts the lower roll (Fig. 6a). At this stage, the bead has rotated counterclockwise together with the neighboring segments of the strips. The horizontal back tension fights against this movement and puts in tension the angle upstream (i.e. left on the figures) on the bottom surface. This is a local plastic bending which initiates tensile damage even before this point contacts the roll. The bead forms a temporary obstacle to the penetration as local reduction reaches almost 80%, to be compared to the nominal 35%. But it is entrained, provided sufficiently high friction is ensured. The two gaps just downstream (i.e. right) of the bead, visible in Fig. 6a, form angles of 135° (by design of the root) and 140° ("weld metal wetting angle") which open into flat angles as they progress in the bite without contact; this unbending effect initiates tensile damage at these points (Fig. 6b); beyond closure, they are submitted to compressive stress and further damage is limited, except when crossing steady shear bands which are typical features of strip rolling. Then, the upstream strip metal contacts the lower roll while the weld metal is still in contact with the upper roll, inducing reverse (upward) bending (Fig. 6c). At this stage, the upstream strip is pushed upward while the weld metal of the cap of the bead is pushed downward: this kind of diffuse shear effect initiates damage at the upstream, top surface angle. When the bead has sufficiently progressed into the roll bite for the base metal to form the entry both on the upper and lower roll, the strip becomes flat again (Fig. 6d). These tail movements may depend on the restraint opposed by the weight and stiffness of the nonmodelled, upstream part of the strip. It has been checked that modified upstream boundary conditions, forcing the strip horizontal here also, does not change the results to a significant extent. Note that a non-dimensional Latham and Cockroft function (LC) has been used here [8]: Comparison of different bead heights. As a crucial parameter, bead height has been varied. A height of 2.8 mm has been compared with the nominal 1.4 mm. This corresponds qualitatively to the strip center where the two torches may superimpose their weld metal. Fig. 7 clearly shows that a higher bead very significantly increases LC damage function value. The maximum jumps from 0.38 to quite high 0.81; the increase is mostly on the upper surface, which makes sense -this is where the excess weld metal is thrust into the interval between strips. This confirms the danger of such excess height; indeed, fractures are often found to initiate at strip center. Other geometrical features have been investigated (wetting angles, cap and root shapes) but found of secondary importance compared with bead height. Therefore, next action consisted, starting from the same bead, initially 1.4 mm, in erasing its top to 1/3 of the initial height. Fig. 7c confirms a strong decrease of the damage function everywhere, although again the top surface values vary most. The maximum damage is now on the bottom surface, where the now thicker root weld metal is pushed between the strips. All this suggests a potential remedy. One caveat however: maximum damage is found here at the downstream side of the weld line (i.e. on the right), on the top surface or the bottom surface, whereas experience points to the upstream side (left), top surface. The locus is therefore not predicted correctly. This problem is addressed in the next sections. Effect of the hardness ratio. The austenitic grade is considered again. Playing numerically with the value of A in Eq.(1) makes the weld metal softer / equal in hardness / harder than the base metal. The strain is always large in the center of the weld bead, where the local thickness reduction is largest (Fig. 8, left). For extreme values of Hvstrip/Hvweld, oblique shear bands develop along the interface, in the softer of the two alloys. The right part of Fig. 8 shows how, as a consequence, the most damaged point moves from bottom surface upstream to top surface upstream (i.e. left) as the weld metal gets harder and harder, as its V-shaped part just below the top surface "pinches" the base metal below it. This explains the high frequency of fractures of the ferritic AISI 441 grade. Indeed, it is buttwelded using austenitic ER 310 grade weld metal. In the first few percent of strain, 441 is harder; but beyond ~15% strain, ER 310 weld metal becomes harder. In a 35% thickness reduction pass, the latter situation prevails over 2/3 of the bite length. Indeed, Fig. 9 shows that the maximum damage is found at the correct location, the singular point at the top surface of the upstream strip, near the interface but larger in the strip metal. The reason is again that the underlying strip metal is compressed by the harder weld metal in the triangles above: two shear bands betray this deformation mechanism. This explains why these ferritic strips bonded with austenitic weld metal are more sensitive. Scale-breaker. This explanation does not hold however for austenitic grade rolling where the base metal (hot rolled, fine grains) is always harder than the weld metal (solidification microstructure, large grains). The solution lies in this case in through-process defect hunting, including the scalebreaker. Table 1 and Fig. 10 feature the progression of the strip and weld bead through the 3 stands of the scale-breaker, capturing critical instants where the thick, hence rigid bead contacts the rolls in turn and forces the strip to over-bend compared with the steady-state. This results in (1) thinning of the upstream strip next to the weld line, (2) damage. Achievements and Trends in Material Forming Final values compare well with the near-weld thickness drop measured on the line, 7 to 10% (the latter for ferritic grades). This nourishes confidence in the mechanisms displayed in Fig. 10. In all cases, the most damaged point with the largest plastic strain is top upstream. This is due to the way the protruding weld bead, with its high thickness / high stiffness, contacts the rollers and Key Engineering Materials Vol. 926 climbs over them under a high tension. In unit 1 (Fig. 10a-c), as its root (bottom surface) comes in contact, the bead rotates first clockwise, then counterclockwise over the first roller (rotation of the local normal to the strip with respect to the local normal to the roller). This puts the upstream strip top surface in alternating tension and compression, and vice-versa for the bottom surface. The downstream strip undergoes less strain and stress due to the longer free length. Similar events occur in reverse order when the cap of the weld (top surface) contacts the second roller, with more strain and damage since protrusion is bigger. On exit of Unit 1, strain is larger on the upstream side near the bead, quasi-equal on top and bottom surfaces. In Unit 2 (Fig. 10d-e) and further in the "anti-bow" unit ( Fig. 10f), the top surface experiences larger strain and thinning, resulting in a damage concentration at the top upstream interfacial point. In the steady state, far from the weld line, thinning is of course uniform whereas strain and damage show the traditional plastic bending pattern. Remember that strain has to be plastic to break the brittle scale. Overall, largest damage is of the same order at the one incurred in the 1 st pass rolling process itself. Coupled scale-breaker + 1 st stand rolling computation. Damage through the scale-breaker being larger at the real position where fracture is observed (top upstream), it might explain this preferred locus when rolling damage is added to scale-breaker damage. This has been investigated by chaining the two operations. Fig. 11 compares LC maps in the non-coupled and coupled cases (austenitic + austenitic, standard conditions). Addition of strain and damage incurred during the two steps now balances damage bottom upstream and top upstream. Due to CPU cost however, a coarser mesh (x3) had to be used for the coupled case, so that only a qualitative conclusion can be drawn at this stage. Conclusion Starting from field observation that stainless steel strip butt-weldings fracture more or less often in the first rolling stand, an approximate model of the operation has been built and simulated by the ForgeNxt® FEM package. Coarse as it may be, the Latham and Cockroft damage approach has yielded, we think, a plausible scenario explaining among others the different fracture frequencies of ferritic (AISI 441) versus austenitic (AISI 304L) grades and the fact that, in spite of technological differences and different deformation mechanisms, both share the same privileged fracture locus. First, the analysis of the scale-breaking units has shown unexpectedly high plastic strain and damage. This is due to the protrusion of the weld line -a flat strip being bent under tension in the scale-breaker shows plastic strain and thinning but to a non-dangerous extent. The local behavior of the thick and rigid bead, inducing strong over-bending in its vicinity, is the reason why. Of course, this excess of bending strain is all the larger as a thicker bead is formed at butt-welding. For the ferritic grade, the fact that it is butt-welded with the same austenitic weld metal as the 304L strips is essential: early in the roll bite, the weld metal, with its huge strain-hardening rate, becomes harder than the base metal. It then punches it to the point where an additional shear band forms, bringing high damage. Fracture follows as thickness reduction proceeds in the roll bite. Of course, damage in the scale-breaker adds up and makes things worse. All in all, the addition of these two sources of damage explains why ferritic strips butt-weldings broke quasi-systematically. For the austenitic grade, the hardness difference is in favor of the base metal. Damage from rolling exists but remains moderate, except if a large weld metal thickness excess occurs -or a strong misalignment of the upstream and downstream strips, also studied but not shown here. Here, the increase of damage by preliminary scale-breaking plays an essential role. Reducing bending amplitude in the scale-breaker could be the solution as tests on the mill have suggested in the past. The present study has confirmed the importance of risk factors qualitatively pointed out in the scarce technical literature. It has moreover put comparative figures on the risk and shown precise plastic strain and damage mechanisms. On the other hand, several points of interest have been left aside, such as detailed strip and weld metal mechanical and fracture properties, the effect of the HAZ or of plastic heating, all of them potentially bringing the system to critical behavior. This leaves space for future studies and future progress.	5,800.2	2022-07-22T00:00:00.000	[ "Engineering", "Materials Science" ]
Nucleosides and amino acids, isolated from Cordyceps sinensis, protected against cyclophosphamide-induced myelosuppression in mice Abstract The material basis of Cordyceps sinensis (Berk.) Sacc has not yet been well understood and natural C. sinensis resources are very rare. The present study aimed to clarify the substance basis and compare the protective effect of natural and artificially-cultivated C. sinensis against cyclophosphamide (CTX)-induced myelosuppression. Both natural and artificially-cultivated C. sinensis effectively improved CTX-induced decrease of peripheral blood counts and hemopoietic growth factors, pathological changes, and apoptosis of bone marrow. Importantly, artificially-cultivated C. sinensis showed similar capacity compared with natural C. sinensis. Uridine (1), adenosine (2), L-pyroglutamic acid (3), lysinonorleucine (4), 1,3,5-trimethoxybenzene (5), D-mannitol (6), L-pyroglutamic acid methyl ester (7), tryptophan (8), and phenylalanine (9) were isolated from bioactivity-guided fraction and identified to attenuate CTX-induced myelosuppression in mice. In conclusions, nucleosides and amino acids represented the effective chemical components in C. sinensis. Artificial cultivation can be used as an effective substitute for natural C. sinensis. Graphical Abstract Introduction Cordyceps sinensis (Berk.) Sacc. is a worm in winter but a grass in summer. For centuries, C. sinensis has been used as an immunomodulator in China (Qian et al. 2012;Kang et al. 2015). Due to its extreme host range specificity, C. sinensis grows slowly and natural resources are scarce. The overexploitation has made C. sinensis on the brink of extinction. Recently, artificially-cultivated C. sinensis has been successfully developed in China (Li, Liu, et al. 2019). However, the active ingredients of cultivated C. sinensis has not been fully understood, which needs to be further characterised. In present study, the potential effect on CTX-induced myelosuppression was evaluated in mice after treatment of natural and artificially-cultivated C. sinensis. The active components were identified. The results would provide useful information to understand the effect of C. sinensis for improving hematopoietic function and promote the development process of the cultivated substitute to protect the C. sinensis resources on the verge of extinction. Results and discussion 2.1. Natural and artificially-cultivated C. sinensis exhibited virtually identical protective effect against CTX-induced myelosuppression in mice To evaluate the potential effect of natural and artificially-cultivated C. sinensis on CTXinduced myelosuppression, their chemical compositions were firstly analyzed by HPLC, which showed similar chemical composition ( Supplementary Figure 1 A). Both natural and cultivated strains, significantly improved CTX-induced myelosuppression in mice, as evidenced by increased levels of immune organ indexes, peripheral blood counts, EPO, G-CSF and TPO ( Supplementary Figures 2-4). Moreover, C. sinensis increased bone marrow cell number, improved femoral bone morphology and inhibited cell apoptosis ( Supplementary Figures 5-7). These findings suggested the chemoprotective effects of natural and artificially-cultivated C. sinensis were identical, which means artificial cultivation can be used as an effective substitute for natural C. sinensis regarding the effect on CTX-induced myelosuppression. Water extracts of C. sinensis contribute to the protective effect against CTX-induced myelosuppression in mice In order to elucidate the active components, C. sinensis was extracted by different extract solvents and only water extract increased WBC counts and LYM% level (Supplementary Table 1). These results suggested that the active components of C. sinensis responsible for the leukogenic effects might be water-soluble substances with higher polarity. Then, the exact compounds in ethanol extract and water extract were systematically isolated using preparative HPLC and silica gel column chromatography. Finally, compounds 1-9 were obtained and their chemical structures were identified by 1H and 13 C NMR data (Supplementary Figure 9- Table 2). Natural and artificially-cultivated C. sinensis contained similar contents of these compounds with variations of 0.48-1.78 (contents ratio of artificially-cultivated C. sinensis to natural C. sinensis) (Supplementary Table 3). Actually, the main nutrients including soluble proteins, nucleosides, nucleotides, and adenosine between the natural and artificially-cultivated samples were virtually identical, while the toxic heavy metal levels were significantly lower in artificially-cultivated C. sinensis Li, Han, et al. 2019;Zhou et al. 2019;Guo et al. 2020). Then, the effects of adenosine, uridine, mannitol, tryptophan, phenylalanine and pyroglutamic acid were directly evaluated in CTX-induced myelosuppression mice. First, all six compounds abolished the body weight loss induced by CTX but failed to ameliorate the decrease of indexes of thymus and spleen (Supplementary Figure 27). Furthermore, six compounds displayed varied effects on hematopoietic function of bone marrow in CTX-induced mice. It should be noted that the protective potential of a single compound verified in present study was poorer than that of C. sinensis. These results suggested that these compounds ameliorated CTX-induced myelosuppression to different degrees and the leukogenic effect of C. sinensis could be a synergistic action of multiple components, like adenosine, uridine, mannitol, tryptophan, phenylalanine and pyroglutamic acid. Conclusion Both natural and artificially-cultivated C. sinensis exhibited comparable pharmacological activity against CTX-induced myelosuppression in mice. Nucleosides and amino acids represented the effective chemical components in C. sinensis. The findings suggested that artificial cultivation can be used as an effective substitute for natural C. sinensis regarding the effect on CTX-induced myelosuppression. Disclosure statement No potential conflict of interest was reported by the authors. Funding This work was supported by the National Key Research and Development Program of China (2018YFC1705900), Distinguished professor of Liaoning Province (XLYC2002008), Science Foundation of Department of Education of Liaoning Province (LZ2020054).	1,195.6	2022-02-19T00:00:00.000	[ "Biology" ]
Cost-effectiveness of uterine balloon tamponade devices in managing atonic post-partum hemorrhage at public health facilities in India Objective Post-partum hemorrhage (PPH) is the leading direct cause of maternal mortality in India. Uterine balloon tamponade (UBT) is recommended for atonic PPH cases not responding to uterotonics. This study assessed cost-effectiveness of three UBT devices used in Indian public health settings. Methods A decision tree model was built to assess cost-effectiveness of Bakri-UBT and low-cost ESM-UBT alternatives as compared to the recommended standard of care i.e. condom-UBT intervention. A hypothetical annual cohort of women eligible for UBT intervention after experiencing atonic PPH in Indian public health facilities were evaluated for associated costs and outcomes over life-time horizon using a disaggregated societal perspective. Costs by undertaking primary costing and clinical parameters from published literature were used. Incremental cost per Disability Adjusted Life Years (DALY) averted, number of surgeries and maternal deaths with the interventions were estimated. An India specific willingness to pay threshold of INR 24,211 (USD 375) was used to evaluate cost-effectiveness. Detailed sensitivity analysis and expected value of information analysis was undertaken. Results ESM-UBT at base-case Incremental Cost-Effectiveness Ratio (ICER) of INR -2,412 (USD 37) per DALY averted is a cost-saving intervention i.e. is less expensive and more effective as compared to condom-UBT. Probabilistic sensitivity analysis however shows an error probability of 0.36, indicating a degree of uncertainty around model results. Bakri-UBT at an ICER value of INR -126,219 (USD -1,957) per DALY averted incurs higher incremental societal costs and is less effective as compared to condom-UBT. Hence, Bakri-UBT is not cost-effective. Conclusion For atonic PPH management in India, condom-UBT offers better value as compared to Bakri-UBT. Given the limited clinical effectiveness evidence and uncertainty in sensitivity analysis, cost-saving result for ESM-UBT must be considered with caution. Future research may focus on generating high quality comparative clinical evidence for UBT devices to facilitate policy decision making. Introduction Investment in maternal health improves health outcomes and substantially benefits overall socio-economic development of the society [1][2][3]. Though maternal health financing towards life-saving interventions has steadily improved over time, key factors like sustainable and efficient funding, pertinent health policies and a reliable supply of life saving commodities ensuring coverage of health services has still fallen short [4]. Efficient use of available resources by implementing cost-effective interventions can prevent maternal morbidity and mortality. Post-partum hemorrhage (PPH) is defined as maternal blood loss of 500 ml or more within 24 hours after delivery and is responsible for two-thirds of global maternal deaths occurring due to hemorrhage [5,6]. Obstetric hemorrhage is the leading direct cause of maternal deaths worldwide. Skilled healthcare providers, timely referrals, availability of emergency obstetric services and a reliable supply of life-saving commodities can avoid these deaths occurring due to PPH [3,7]. PPH affects nearly 3-6% of all women giving birth in India [8,9]. Failure of uterus to contract after delivery i.e. uterine atony is the commonest cause of PPH [10]. India accounts for one-fifth of all maternal deaths globally with PPH as its leading cause [11,12]. Management in India begins by preventing PPH using active management with uterotonics, cause-specific PPH treatment, resuscitation for shock and patient referral to higher facilities for further treatment. For atonic PPH, uterotonics remain the mainstay treatment in accordance with the World Health Organization (WHO) guidelines [6]. The WHO recommends using Uterine Balloon Tamponade (UBT) intervention for atonic PPH cases that do not respond to first-line treatment after vaginal delivery provided recourse to surgical interventions, access to blood products, skilled/trained health personnel, resources and treatment protocols are in place [13]. The latest WHO guidance has identified that existing evidence for UBT intervention is disparate however has recognized that it is probably feasible and acceptable to women and healthcare providers in contexts where treatment protocols are available and implemented. A recently concluded systematic review of ninety-one studies has also suggested that UBT use is associated with significant reduction in PPH related invasive procedures [14]. The review reported a pooled UBT success rate of 85.9% in controlling PPH. Although UBT intervention was highly successful in controlling severe PPH bleeding, efficacy and effectiveness evidence between experimental and observational studies was found conflicting with quality of evidence identified as a limitation. Two RCT studies included in the systematic review saw no benefit of introducing UBT in refractory PPH management [15,16]. The respective studies however reported significant associated limitations and biases impacting their results. A recent Cochrane review has concluded that evidence from existing RCT studies is insufficient to determine relative effectiveness of mechanical and surgical interventions in treating primary PPH [17]. The latest WHO guidelines acknowledges that UBT recommendation may not be operationalized in a standard and consistent manner across different settings. For the diverse Indian health system with initiatives such as the 'Dakshata' and 'LaQshya Guidelines' that focus on standardization of institutional deliveries across India by adhering to clinical protocols for management and equipping providers for comprehensive maternal health care, UBT intervention for atonic PPH management as a relatively simple technique can be a life-saving intervention [18]. UBT can potentially avoid need for surgery and even act as a temporizing measure while awaiting transfer to the higher center. Multiple UBT devices specifically designed or assembled for use in PPH management are presently available. Indian guidelines recommend using condom-UBT device for management of refractory atonic PPH [11]. Condom-UBT is an improvised device assembled at service delivery point using readily available components like a male condom, Foley catheter, string/ suture and intravenous (IV) infusion set. The cost of assembly components of condom-UBT device or its modifications available in public health facilities of India is very low, estimated at Indian National Rupee (INR) 128 (USD 2). Bakri balloon and Every Second Matters (ESM) UBT, two United States Food and Drug Administration (US-FDA) approved devices specifically designed for PPH management are also being used across a few public health settings in India [19][20][21]. Bakri-UBT comes in a sterile pack, is ready to use and has a drainage outlet to measure ongoing blood loss. At a market price of INR 9,554 (USD 148), Bakri-UBT is commercially available in India but is quite expensive [22]. ESM-UBT, a low-cost alternative with documented mechanical properties and assembly components available in a sterile pack with an instruction manual is relatively inexpensive at INR 397 (less than USD 5), but is commercially not available in India at present [14,23,24]. India provides free treatment for post-natal complications of pregnancy to women delivering at public health facilities under the Janani Shishu Suraksha Karyakaram (JSSK) scheme [25]. Given the ever growing need for resource prioritization, decision making that is evidence informed is becoming increasingly important. In this regard to reduce health opportunity costs, India is generating cost-effectiveness evidence as a tool to prioritize public health interventions. Choosing a specific UBT device for PPH complication management has associated health and economic consequences. Evidence on economic perspective of UBT intervention is limited globally. A recent systematic review analyzing cost-effectiveness of UBT intervention found only two studies that had evaluated its cost-effectiveness [26]. A Kenyan modelling study found ESM-UBT to be highly cost-effective for severe PPH management [27]. Another study from year 2006 evaluating economic efficiency of various interventions in reducing PPH across four developing countries including India observed UBT to be the most-cost effective alternative among curative options for PPH management. The assessment however identified UBT intervention to not be adequately researched and that cost-effectiveness findings were based on clinical effect estimates from two small case series studies [28]. The systematic review concluded that evidence on cost-effectiveness of UBT intervention is limited and not generalizable to different contextual settings. The growing clinical evidence for individual UBT devices has not yet been used to evaluate and compare multiple UBT device alternatives using a cost-effectiveness approach to inform decision-making. This present study aimed to undertake an economic evaluation for UBT intervention by determining the most cost-effective UBT device for atonic PPH management in the Indian context. This study will help policy makers choose a UBT device that offers the highest value. Objective of this study was to assess cost-effectiveness of condom-UBT, ESM-UBT and Bakri-UBT devices in managing atonic PPH across public health facilities of India. Model overview A decision tree type of decision analytic model was built to estimate the expected costs and consequences of using ESM-UBT or Bakri-UBT device as compared to standard care i.e. condom-UBT in atonic PPH management. Globally, Bakri-UBT is one of the most commonly used UBT devices. ESM-UBT is being assessed for introduction in the Indian public health system as a low-cost alternative. As UBT intervention for atonic PPH management is a discrete event occurring over a short duration without any recurring events or health state transitions, a decision tree model was considered appropriate. As recommended in India, disaggregated societal perspective was used to analyze costs and effects of the intervention [29]. This comprised of public health system costs in delivering UBT intervention and out-of-pocket expenditure (OOPE) incurred by households at the time of childbirth. Effects were measured in terms of Disability Adjusted Life Years (DALY) averted and Incremental Net Monetary Benefit (NMB) of the intervention [30]. Given the high burden of PPH in India, nature of the condition, availability of India inclusive disability weights for maternal haemorrhage from the Global Burden of Disease report, DALYs were chosen as the primary outcome measure [31]. Net monetary benefit framework has an advantage in interpreting negative Incremental Cost-Effectiveness Ratio (ICER) values and hence was used in the analysis. The modeled base-case population included a cohort of Indian women with median age 21 years at first child birth accessing public health facilities for atonic PPH over one-year study implementation duration [32]. Specifically, UBT intervention lasts for a short duration with highest reported mean UBT device retention time of 27.5 hours across reviewed studies [33]. Though PPH event lasts for a short time, consequences or outcomes as a result of the condition or its clinical management may likely result in long-term consequences. This includes disabilities due to hysterectomy, associated infertility or maternal death after PPH event. These disabilities may get further exacerbated for a healthcare setting like India wherein health infrastructure or expertise for conservative surgical procedures may not always be readily available across settings. A life-time horizon was thus considered appropriate to account for consequences of using UBT intervention. A standard three percent discount rate was used to account for all primary costs and model outcomes [34,35]. As costs incurred for UBT intervention are expected to be non-recurring and a one-time event, estimated costs in the model were kept undiscounted. To account for health outcomes occurring over life-time horizon as a result of the intervention, model outcomes were discounted. Model structure The model was built using Microsoft Excel 2016 with Microsoft Visual Basic for Application 7.1. The decision tree as described in Fig 1, begins with a cohort of women delivering in public health facilities of India experiencing atonic PPH that does not respond to uterotonics thereby becoming eligible for UBT insertion. A decision at this point is made regarding choice of UBT device used for management. The pathway followed from this decision node depends on services available at the health facility level accessed by women and clinical course of the condition. For the decision tree as described in Fig 1, public health facilities in India were classified into three levels based on services provided for PPH management as per Indian guidelines [11,36,37]. Primary level care for PPH comprises of Primary Health Centre (PHC) facility that is equipped with medical officer and skilled birth attendants to initially stabilize, insert UBT device and refer women to a higher facility. Secondary level care consists of Community Health Centers (CHC) and Sub-District Hospitals (SDH) additionally equipped with Obstetrics-Gynecology (OBGYN) specialist, Operation Theater (OT) and blood transfusion facilities. Tertiary level care for PPH includes additional advanced intervention and Intensive Care Unit (ICU) facility available at District Hospitals (DH) and medical colleges. Patients from secondary level care requiring ICU facilities are expected to be referred. UBT device insertion is considered after uterotonics like intramuscular or IV oxytocin, IV methyl-ergometrine, sublingual misoprostol fails to control atonic PPH bleeding. Resuscitation measures for shock prevention or management are continued. Cases remaining uncontrolled after UBT insertion would undergo surgical intervention like B-Lynch compression suturing, stepwise devascularization surgery or directly a life-saving obstetric hysterectomy depending on woman's clinical condition and availability of resources at facility. A patient uncontrolled after conservative surgical procedures may also eventually require hysterectomy procedure to control bleeding. Surgical facilities are expected to be available at both secondary and tertiary levels. PPH cases with severe blood loss, post-surgery or any other complications may need ICU facility for management available at tertiary level. A woman may die at any point during the course either due to atonic PPH complication or other maternal mortality causes. Certain assumptions were made in decision tree conceptualization. First, it was assumed that all eligible women would receive one common UBT device across all facilities in the Indian public health system. In ESM-UBT scenario for instance, it was assumed that all eligible women would exclusively receive ESM-UBT device across the public health system and so on for the other two devices. Secondly, it is assumed that all women accessing primary level facilities would be referred after UBT insertion and initial stabilization. An effective system for immediate referral of patients from primary level exclusively to secondary level and from secondary level to tertiary care for ICU service is assumed to be in place. Thirdly, conservative surgeries like B-lynch suturing and stepwise devascularization are grouped together as a single surgical procedure in the model. Model inputs 1. UBT device effectiveness. Published literature was collated to assess clinical effectiveness of individual UBT devices evaluated in this study. Effectiveness was defined as control of atonic PPH bleeding after UBT device insertion measured directly or as surgical interventions/ maternal deaths averted to control bleeding after UBT insertion. As ESM-UBT device was the intervention being assessed for introduction, a systematic review was undertaken after registering with PROSPERO 2019 database (Registration number CRD42019122802) [38]. Only three published ESM-UBT case-series studies were identified to be eligible for the systematic review. For condom-UBT and Bakri-UBT, a targeted review was undertaken by searching three electronic databases for published literature between January 2010 to December 2019. Studies evaluating effectiveness of condom-UBT and Bakri-UBT in atonic PPH management, available in English language were included. For targeted review, a total of 30 articles comprising twelve condom-UBT and nineteen Bakri-UBT studies, one of which was a comparative randomized control trial (RCT) study between the two devices were eventually included. Overall eligible reviewed studies for the three UBT devices comprised of one systematic review, three RCTs and twenty-six prospective or retrospective studies largely indicating low quality of evidence across hierarchical structure. Primary outcome considered by the limited available ESM-UBT studies was survival rates in contrast to the other two UBT devices wherein majority studies reported control in PPH bleeding. The mean summary measure of effectiveness for each individual UBT type was computed from effectiveness rates reported by the studies. All condom, ESM-UBT and nearly two-fifth Bakri-UBT studies reviewed for this analysis reported findings from developing country settings. S1 Table reports the identified eligible studies reviewed to ascertain clinical effectiveness of the three UBT devices. Clinical and epidemiological parameters used in the model were obtained from published literature and facility level Health Management Information System (HMIS). Data on public healthcare facility accessed for childbirth was obtained from National Family Health Survey (NFHS-4), India [32]. As specific data for UBT use in PPH across healthcare facilities was unavailable, probability estimates for incidence of PPH (3.6% of deliveries), atonic PPH incidence (80% of PPH cases), control of cases with medical treatment (90% cases), probability of specific surgical intervention, surgery success rate and morbidity data associated with PPH as reported in Table 1 were obtained from literature relevant to the Indian context [8,9,[39][40][41]. Though the comparative RCT (Condom-UBT versus Bakri-UBT) had some parameters that could have been used in model, the study sample size was very small and no statistically significant differences were observed between the two UBT devices. So, studies with large samples specific to healthcare services in the Indian context were used to compute model parameters wherever available. Probability of PPH and all-cause mortality was calculated using data from the Census of India, Sample Registration Survey [42]. A list of all parameters including UBT effectiveness rates, epidemiological and clinical parameters used in the model are reported in Table 1. Costs. A primary bottom-up economic micro-costing study with a convenience sample of one PHC representing primary care, one SDH representing secondary care and one DH and a government medical college hospital representing tertiary level care in the state of Maharashtra was undertaken to asses cost of managing atonic PPH in Indian public health settings for the year 2017-18 [44]. Costs were calculated specifically for condom-UBT intervention i.e. standard of care with similar estimation for ESM-UBT and Bakri-UBT using respective clinical effectiveness and UBT device price. Cost estimates for individual UBT type with subsequent interventions from primary costing was then used in the decision tree. Cost components in the decision tree included unit cost of UBT insertion, devascularization surgery, hysterectomy, inpatient department (IPD) admission and ICU admission for a PPH event at respective healthcare levels. Unit cost of UBT insertion included UBT device price in addition to the economic cost of resource utilization in the form of human resources, area, equipment, drugs and consumables and overhead utilities. UBT insertion costs were calculated respectively for both labor room and operation theatre to get a weighted aggregate facility level unit cost. This was followed by aggregation of DH and medical college costs respectively to get unit cost for primary, secondary and tertiary healthcare levels. Inpatient admission cost for a unit PPH event was estimated by calculating per-day facility cost for an obstetric patient admitted at the health facility for an average of 3.4 days followed by apportioning to 2.98 days as reported for atonic PPH stay [46]. ICU cost for PPH event admission was similarly estimated by calculating perday ICU admission cost of an obstetric patient admitted in the ICU for an average of 3.47 days followed by apportioning to the reported 1.5 days for PPH admission [41,47]. As PPH related referral data was unavailable during analysis, referral cost was obtained from a published Indian study reporting public health system cost of referrals related to institutional deliveries after making inflation adjustment to the year 2017-18 [45,48]. All respective unit costs were applied to the cohort of atonic PPH cases in India eligible for UBT intervention run through the decision tree pathway. Costs in this study are reported in Indian National Rupee (INR) and United States Dollars (USD) currency. A conversion rate of 1 USD = 64.5 INR for the year 2017-18 was used in cost analysis [49]. Certain costing assumptions were made for the decision model. Cost of blood transfusion and fluid resuscitation were not considered separately and are incorporated in unit costs. A one-time annual training of healthcare providers for UBT intervention irrespective of the type of device was assumed and costs were estimated. Disaggregated societal costs for the decision tree was obtained by combining health system costs, an annual training and OOPE incurred by the households. OOPE was obtained from the 75 th round of National Sample Survey 2017-18 [43]. NSS survey encompasses a very large number of rural and urban Indian households and considers both medical and non-medical expenditures such as food, transport, escort expenditures, lodging, etc. incurred by households on account of childbirth. Table 1 presents the input and assumption parameters used in costing analysis. 3. Valuation of consequences. The health consequences of using any of the three UBT devices after atonic PPH event in this study was estimated using DALY outcome measure. UBT intervention contributes to subsequent reduction in need for surgeries, thus reducing maternal morbidity and mortality [14]. Even with UBT intervention, surgeries like devascularization, obstetric hysterectomy and complications related to PPH may still occur depending on patient's clinical response and type of UBT device used. To ascertain such associated outcomes with each UBT type, number of surgeries and maternal deaths with the three UBT devices were also estimated using event probabilities. DALY as an outcome measure summarizes disease burden by combining both time lost through premature death and time lived in a state of less than full health labelled as disability. DALYs were estimated by calculating Years of Life Lost (YLL) and Years Lived with Disability (YLD) subsequent to UBT intervention for PPH condition using the simplified DALY estimation method [50]. For DALY estimation over life-time horizon, YLL or YLD were calculated at each respective terminal node of the decision pathway. At terminal nodes for PPH event, patients are expected to remain alive with associated disability due to PPH event or its consequences or may die either due to bleeding or other causes of maternal mortality. YLL for premature death due to PPH event was calculated using age-specific life-expectancy of the Indian female population. For the considered hypothetical cohort of Indian women aged 21 years dying due to PPH after childbirth, remaining age-specific life expectancy of 53.79 years from abridged life-tables determined YLLs for deaths in the decision tree [51]. For YLD of those remaining alive, patients having bleeding controlled with UBT intervention alone were assigned disability weight of mild maternal hemorrhage (less than 1-liter blood loss) whereas remaining cases undergoing conservative surgery, obstetric hysterectomy or ICU admission for PPH control were assigned the disability weight of severe maternal hemorrhage (greater than 1-liter blood loss) obtained from Global Burden of Disease study [52]. Additionally, those undergoing hysterectomy for PPH control were assigned the disability weight of secondary infertility for their remaining reproductive life-span. As disability weight for infertility due to obstetric hysterectomy after PPH was unavailable, we assumed infertility due to puerperal sepsis from GBD study to be an appropriate estimate. Time duration of disability due to hemorrhage was assumed to last for a postpartum period of six weeks (0.11 years) whereas in cases of hysterectomy, additional disability weights of secondary infertility were assigned for the remaining reproductive life-span up to 46 th year for the study cohort [53,54]. Number of cases at each terminal node was obtained as the product of probability of events at each preceding node leading to the terminal node. Number of cases at each terminal node was then multiplied with remaining expected life-expectancy for YLLs or outcome specific disability weights for specific defined duration of disability to determine YLDs at terminal nodes. Overall DALYs associated with UBT intervention was thus estimated by combining YLL and YLDs. As an example of DALY calculation, for the estimated number of women having atonic PPH bleeding controlled with hysterectomy after failed medical, UBT and conservative surgical interventions, disability weight of severe hemorrhage for duration of 0.11 years and disability weight of hysterectomy associated secondary infertility for a reproductive time-span of 25 years determined YLD for this terminal node. Similarly, for women remaining alive after control of PPH bleeding with UBT intervention alone, disability weight of mild hemorrhage was assigned for a duration of 0.11 years to determine YLD for this node. DALY calculation in this study did not consider age weighting. A discount rate of three percent to account for future DALY outcomes over life-time horizon as recommended by the Indian reference case was considered in base-case analysis [29]. Table 1 presents the input parameters for PPH consequences used in the model. Cost-effectiveness Incremental cost-effectiveness ratio (ICER) i.e. incremental costs per DALY averted for ESM-UBT versus condom-UBT and Bakri-UBT versus condom-UBT comparisons was calculated. A simplified net monetary benefit framework (Incremental NMB = Incremental benefit � Willingness-to-pay (WTP) threshold -Incremental costs) was estimated for the interventions against comparator i.e. standard care. To determine cost-effectiveness, an India specific WTP was obtained from a study that empirically calculated country specific cost per DALY averted threshold for 97 different low-income and middle-income countries [55]. The study calculated threshold value using India specific health expenditure, survival and morbidity burden. An inflation adjusted Indian value of INR 24,211 (USD 375) per DALY averted for year 2017 was chosen as the threshold for this study [56,57]. As sensitivity analysis evaluates interpretation of higher WTP such as that recommended by the WHO, the chosen lower threshold was considered appropriate for base-case analysis [58]. Sensitivity analysis Sensitivity analysis was used to address uncertainties in the model. Probabilistic sensitivity analysis (PSA) was used to evaluate joint parameter uncertainty effect. Evidence base used for clinical effectiveness of UBT devices was of limited strength with only three studies evaluating ESM-UBT intervention. For this reason, distribution of clinical effectiveness parameter in sensitivity analysis for each UBT device was varied by 20 percent limit at lower end and a ceiling 98 percent (highest reported effectiveness for UBT from literature review) for all comparators. Primary cost parameters used in the model were varied along the 95 percent Confidence Interval (CI) limits derived by running 1,000 Monte Carlo simulations for each unit cost estimate obtained. Parameters of each unit cost were assigned distributions and then varied at both ends by assuming a variation of 50 percent for UBT device price to account for expectation of cost variability, a 50 percent upper and 100 percent lower limit respectively for drugs and consumables given the government procurement at a negotiated price and a 25 percent variation for remaining parameters like human resource salaries, area price, equipment (medical/nonmedical), utilities and utilization of resources. For model, remaining cost parameters like referral cost, training and OOPE were varied by a 20 percent assumption at both ends. Disability weights were varied along CI limits provided by the GBD study. Discount rates were varied between 0 to 5 percent as recommended for India [29]. Remaining model parameters were varied by 20 percent on either side of the base-case value. A Beta distribution was assigned to all probabilities, proportions and disability weight parameters whereas a Gamma distribution was used for costs and resources. To determine influence of each individual model parameter on cost-effectiveness results, One-Way Sensitivity Analysis (OWSA) was undertaken using the fore mentioned distributions. Tornado diagram was used to graphically represent the most influential parameters in the model. Monte Carlo method was used to compute 10,000 PSA simulation results for the model. An ICER (DALY) plane and a Cost-Effectiveness Acceptability Curve (CEAC) at different WTP values was used to represent uncertainty using probabilistic sensitivity analysis results. Table 1 presents the base-case input values used in sensitivity analysis with their upper and lower limits. The model was validated using the AdViSHE tool for health economic models and the study was reported using the ISPOR-CHEERS checklist [59,60]. S1 Appendix provides the checklist tools. The study was approved by the Ethics Committee for Clinical Studies at National Institute for Research in Reproductive Health, Mumbai under approval number D/ ICEC/Sci-29/31/2018. State health department administrative approvals and consent from respective health facilities were obtained to undertake primary costing study. Costs of managing atonic PPH with the three UBT devices in Indian public health system along with upper and lower limit range is presented in Table 2. Cost-effectiveness Our model estimates use of condom-UBT in an atonic PPH case to be associated with 0.22 DALYs lost per patient. Using ESM-UBT as compared to condom-UBT incrementally averts 0.03 DALY per patient. Similarly, using Bakri-UBT is associated with 0.296 DALYs i.e. an additional 0.081 DALYs lost per patient using Bakri-UBT as compared to condom-UBT. An Table 2. Costs and outcomes with UBT devices for atonic PPH management in Indian public health system. Characteristics Condom-UBT (Standard of care) Table 3. Incremental costs, consequences and cost-effectiveness of UBT intervention in atonic PPH management. USD 1,957) is incurred and one less DALY is averted by using Bakri-UBT instead of the recommended condom-UBT choice. Therefore, at base-case, Bakri-UBT is not cost-effective as compared to condom-UBT. Table 3 incremental NMB value of INR -12,185 (USD -189) per patient as compared to condom-UBT, suggesting that Bakri-UBT is not cost-effective as compared to condom-UBT. Sensitivity analysis The tornado diagram as shown in Fig 2 depicts Discussion This analysis presents new cost-effectiveness evidence regarding the choice of UBT devices available for management of atonic PPH complication in Indian public health settings. As clinical effectiveness of Bakri-UBT increases, Bakri-UBT is still not cost-effective (Bakri-UBT is more expensive, more effective) at the willingness to pay threshold of INR 24,211. As effectiveness of Bakri-UBT decreases (masked), Bakri-UBT is not cost-effective (Bakri-UBT is more expensive, less effective) as compared to condom-UBT. Similarly, for condom-UBT clinical effectiveness parameter, as condom-UBT effectiveness increases, Bakri-UBT is not cost-effective (Bakri-UBT is more expensive, less effective) as compared to condom-UBT. As condom-UBT effectiveness decreases, Bakri-UBT is still not costeffective (Bakri-UBT is more expensive, more effective) at the given WTP threshold. Globally, studies have assessed cost-effectiveness of various prophylactic or medical treatment options, estimated health system costs and modelled health outcomes for PPH management. A recent Indian study predicted that enhanced PPH care with non-surgical measures such as UBT intervention is cost-effective and life-saving [61]. Similarly, a Kenyan study from the year 2017 found ESM-UBT device to be highly cost-effective for severe PPH management in Kenyan health facility settings. Our study has analyzed and compared the choices of available UBT devices for atonic PPH management in Indian public health facilities using the economic evaluation approach. Our base-case result suggests ESM-UBT for available limited clinical evidence to be associated with marginally lower costs, fewer DALYs, surgeries, deaths and an incremental positive net monetary benefit as compared to condom-UBT. The outcomes although favor ESM-UBT at base-case, are only marginal and absolute differences in societal costs, health outcomes and net benefits are minimal as compared to condom-UBT. The ICER value at base-case suggests ESM-UBT to be a cost-saving strategy. For this result, the three-time higher ESM-UBT device price as compared to condom-UBT is offset by fewer subsequent surgeries, associated deaths and thus a marginally lower societal expenditure with favorable health outcomes. Base-case comparison between Bakri-UBT and condom-UBT in our study suggests Bakri-UBT to be associated with significantly higher societal costs, higher DALYs lost, higher number of deaths, surgeries and a negative incremental monetary benefit as compared to condom-UBT. The 12 -14]) per patient was computed using the 10,000 PSA simulations. Given the decision-making error probability and limited evidence available especially for ESM-UBT context, we undertook an expected value of information analysis. Total expected value of perfect information (EVPI) for population was determined to quantify the monetary need for evidence generation that reduces decision uncertainty. For this calculation, an undiscounted one-time population cohort was used. The estimated value of population EVPI for ESM-UBT versus condom-UBT was INR 14,323,056 (USD 222,063) for an annual cohort. Given the sensitive nature of model results to clinical effectiveness parameter, the EVPI result can be used by decision makers to assess research prioritization for generation of clinical evidence. Robust comparative clinical effectiveness evidence especially for ESM-UBT versus condom-UBT comparison in the Indian context can be considered to improve and determine net incremental associated benefits to the society. For uncertainty across sensitivity analysis results and marginal differences in associated costs and health outcomes, a cost-saving result for ESM-UBT with existing evidence must be considered with caution and adoption of ESM-UBT in the Indian public heath settings for atonic PPH management instead of the currently recommended condom-UBT device must be made after due deliberation. Bakri-UBT device with significantly higher societal costs and lesser favorable health outcomes was dominated by condom-UBT across base-case and PSA with a ceiling 11% probability of Bakri-UBT being cost-effective. The base-case net incremental monetary benefit for Bakri-UBT was INR -12,184 with a mean value of INR -12,299 [95% CI -12,349, -12,249] (USD -191 [95% ). Condom-UBT device remains a more cost-effective alternative as compared to Bakri-UBT device in Indian health settings across analysis. However, in a scenario with Bakri-UBT device compared against no UBT intervention at all for atonic PPH management, although Bakri-UBT device has high associated societal costs, relative health outcomes and incremental cost-effectiveness ratios are still favorable for Bakri-UBT, suggesting UBT intervention to be a favorable alternative as compared to not providing UBT intervention at all for medically uncontrolled cases. This study has its strengths and limitations. To our knowledge, this is the first such study that has presented cost-effectiveness evidence by comparing available UBT devices for atonic PPH management. We have analyzed cost-effectiveness of UBT devices available in the Indian context using India specific primary costs, epidemiologic and clinical effectiveness data to the available extent. A detailed sensitivity analysis was undertaken in this study to address uncertainties and assumptions used in the model. For the reported uncertainty in our results, we have presented the potential monetary value associated with generation of evidence by undertaking further research to reduce existing uncertainties and further improving robustness in decision making. The study limitations include use of clinical effectiveness input parameter based out of limited robust evidence currently available for all UBT devices, especially for ESM-UBT which is based on findings from three case-series studies that largely assess survival outcomes in African health settings. Unavailability of clinical effect estimates from a systematic robust comparison of clinical effectiveness between UBT devices over identical outcome measures is an important limitation of the study results. Secondly, the study uses disability weights that nonspecifically measures disease burden rather than assessing specific health state outcomes. Though QALY estimation for decision making as recommended in India may have been a more accurate measure, we chose to use condition specific DALY outcome instead of using proxy utility weights from other contextual settings for QALY estimation. Thirdly, the model does not consider potential costs and consequences of a complication like anemia as it is a common morbidity prevalent among Indian women throughout pregnancy and post-partum period [62]. However, given the research question at hand, decision model for given intervention was considered adequate to assist decision-makers. We relied on PPH non-specific outof-pocket expenditure for childbirth to estimate disaggregated societal costs. Additionally, indirect costs like wage loss estimation or productivity loss were not included. Availability of these PPH specific costs may further improve result accuracy. Finally, the study bases its health system cost findings from sample facilities of one contextual setting within India. Though sensitivity analysis plausibly varies health system and OOPE costs, the representativeness of these costs for all India settings may still be limited. Conclusion This study demonstrates that condom-UBT device as recommended for atonic PPH management in India offers better value as compared to Bakri-UBT in Indian public health settings. Although ESM-UBT presents as a cost-saving alternative, decision making between low-cost ESM-UBT and condom-UBT needs further evaluation as differences in costs and health outcomes are marginal, there is uncertainty associated with cost-effectiveness results and clinical evidence available for ESM-UBT is limited at present. Going ahead, high-quality comparative clinical effectiveness evidence is needed to reduce decision-making uncertainty around the choice of UBT device. Future research can also focus on generating India specific health utility scores for PPH, qualitative aspects such as ease of use and provider preferences for UBT devices. Our study results can assist policy makers in prioritizing interventions for atonic PPH management in India. Supply of an affordable and acceptable UBT device particularly needs to be complemented by health system preparedness in the form of skilled workforce and robust referral systems to tackle the PPH emergencies. Supporting information S1 Table. Details of studies included in literature review of the three UBT devices. (DOCX) S1 Appendix. Checklists for study validation and reporting. (DOC)	8,268.4	2021-08-18T00:00:00.000	[ "Medicine", "Economics" ]
Relative positions of the polar boundary of the outer electron radiation belt and the equatorial boundary of the auroral oval Finding the position of the polar boundary of the outer electron radiation belt, relative to the position of the auroral oval, is a long-standing problem. Here we analyze it using the data of the METEOR-M1 auroral satellite for the period from 11 November 2009 to 27 March 2010. The geomagnetic conditions during the analyzed period were comparatively quiet. METEOR-M1 has a polar solar-synchronous circular orbit with an altitude of ≈ 832 km, a period of 101.3 min, and an inclination of 98◦ . We analyze flux observations of auroral electrons with energies between 0.03 and 16 keV, and electrons with 5 energies > 100 keV, measured simultaneously by the GGAK-M set of instruments, composed by semiconductors, scintillator detectors, and electrostatic analyzers. We assume that in the absence of geomagnetic storms the polar boundary of the outer radiation belt can be identified as a decrease in the count rate of precipitating energetic electrons to the background level. It was found that this boundary can be located both inside the auroral oval or equatorward of the equatorial boundary of the auroral precipitation. It was also found that for slightly disturbed geomagnetic conditions the polar boundary of the outer radiation 10 belt is almost always located inside the auroral oval. We observe that the difference between the position of the polar boundary of the outer radiation belt and the position of the equatorial boundary of the auroral precipitation depends on the AE and PC indices of geomagnetic activity. The implications of these results in the analysis of the formation of the outer radiation belt are discussed. Introduction The position of the trapping boundary for energetic electrons in the outer radiation belt (ORB) contains information about the topology of the magnetic field lines of the Earth.For a long time this has been analyzed using data from both low-orbiting and high-apogee satellites (Frank et al., 1964;Frank, 1971;Fritz, 1968Fritz, , 1970;;McDiarmid and Burrows, 1968;Vernov et al., 1969;Imhof et al., 1990Imhof et al., , 1991Imhof et al., , 1992Imhof et al., , 1993;;Kanekal et al., 1998).Using the data of high-apogee satellites, Vernov et al. (1969) showed that the polar boundary of the ORB, also known as the trapping boundary, is located near to ∼ 9 R E in the dayside sector and near to ∼ 7-8 R E close to midnight.These results were further supported by Imhof et al. (1993) using data from the CRRES and SCATHA satellites, and covering distances from ∼ 6 to ∼ 8.3 R E (CRRES) and from ∼7 to ∼ 8.5 R E (SCATHA).Results obtained by Fritz (1968Fritz ( , 1970)), Imhof et al. (1997) and Yahnin et al. (1997) show that the isotropic boundary of energetic particles (i.e., the boundary where pitch angles of particles become isotropic) is located equatorward of the trapping boundary.It means that the ORB trapping boundary can be clearly identified using low-orbiting satellite measurements. A good understanding of the relative positions of the trapping boundary and the equatorial edge of the auroral oval is important for the analysis of the structure of magnetospheric plasma domains and the topology of the geomagnetic field.Comparison of the relative position of the trapping boundary and the auroral oval was statistically done using ground-based auroral observations and satellite observations of the trapping boundary.Akasofu (1968) compared the position of Feldstein's auroral oval with the trapping boundary of the 40 keV electrons obtained by Frank et al. (1964) and statistically showed that the trapping boundary is located inside the auroral oval.However, later Feldstein and Starkov (1970) compared the position of the auroral oval with the results of Alouette-2 observations and concluded that the auroral oval is situated just on the polar border of the trapped radiation region of electrons with energy > 35 keV.Rezhenov et al. (1975) analyzed particle fluxes with energies 0.27, 11, 28 and 63 keV, from the COSMOS-424 satellite, and showed that the trapping boundary is located poleward of the region of low-energy electron precipitation.However, this study was done using the data obtained for only 21 orbits, and was not widely known.Feldstein et al. (2014) stressed (p.120 in their paper) that a poleward (high-latitude) boundary of the diffuse auroral belt without any discrete auroral forms "constitutes the equatorward boundary of the auroral oval and at the same time it is the high-latitude boundary of the radiation belt (RB) of electrons with energies from a few tens to hundreds of kiloelectronvolts (STB -stable trapping boundary for radiation belt electrons)". According to the traditional point of view (see, for example, Paschmann et al., 2002), the auroral oval is mapped to the plasma sheet.In this case the trapping boundary should be located equatorward or at the equatorial boundary of the auroral oval.However, Antonova et al. (2014bAntonova et al. ( , 2015) ) and Kirpichev et al. (2016) showed that most of the auroral oval does not map to the plasma sheet.It is mapped to the plasma ring surrounding the Earth at geocentric distances from ∼ 7 R E to the magnetopause, near noon, and to ∼ 10-13 R E near midnight.They suggested that the plasma in the magnetosphere is in magnetostatic equilibrium, and used the value of plasma pressure as a natural tracer of magnetic field lines, comparing the pressure at low latitudes and at the equatorial plane.Antonova et al. (2017) showed that the outer boundary of this ring in the night sector coincides with the external boundary of the ring current.Results obtained by Antonova et al. (2014bAntonova et al. ( , 2015Antonova et al. ( , 2017) ) and Kirpichev et al. (2016) showed that the auroral oval is mapped to the region of quasitrapping, where drift trajectories of energetic electrons with pitch angles smaller than near to 90 • surround the Earth (Delcourt and Sauvaud, 1999;Öztürk and Wolf, 2007;Ukhorskiy et al., 2011;Antonova et al., 2011a) due to the drift shell splitting effect (which is ordinarily named the Shabansky effect).Such mapping suggests that the trapping boundary should be located poleward of the equatorial boundary of the auroral oval. Therefore, it is very important to establish the true location of the trapping boundary with respect to the equatorial auroral oval boundary.This can be done using simultaneous observations of both auroral electron precipitation and fluxes of energetic electrons.It is well known that the location of the auroral oval and the location of the trapping boundary are strongly affected by geomagnetic activity.Therefore, it is necessary to compare these relative locations using simultaneous measurements of the auroral oval and trapping boundary on the same satellite.However, there are some difficulties related to the detection of the trapping boundaries during the periods of low geomagnetic activity (for example during the solar minimum).In these cases, the level of electron fluxes inside the ORB can be rather low, close to the limit of sensitivity of the instrument.Thus, the detected trapping boundary can be located closer equatorward with respect to the true trapping boundary. Despite the significant number of particle measurements carried out by low-orbiting satellites, the relative location of the trapping boundary and the equatorial boundary of the auroral oval, and how they could be affected by geomagnetic activity, still require careful studies.In this work, we use data of the METEOR-M1 satellite to establish the location of the trapping boundary and of the auroral oval for different levels of geomagnetic activity, which were quantified using the AE and PC geomagnetic indices.The paper is organized as follows.First, we describe the METEOR-M1 satellite instrumentation and the data analysis, including important caveats.Then we obtain the position of the trapping boundary of electrons with energies > 100 keV relative to the equatorial boundary of the auroral oval, and how it varies for small and large values of the AE and PC indices of geomagnetic activity.At the end, we shall discuss the role that our results might play in the determination of features of the high-latitude magnetospheric topology. Instrumentation and data analysis We used the data from the METEOR-M1 satellite launched on 17 September 2009 into a polar solar-synchronous circular orbit with an altitude of ≈ 830 km, a period of ∼ 100 min, and an inclination of 98 • .We used the data of the GGAK-M set of instruments, composed of semiconductor and scintillator detectors, and electrostatic analyzers.In particular, it measured energetic electrons with energies from 0.1 to 13 MeV, and low-energy electrons with energies from 0.032 to 16.64 keV (see more details and available data in http://smdc.sinp.msu.ru/index.py?nav=meteor_ m1; last access: October 2017). For automatic detection of the polar boundary of the ORB and the equatorial boundary of the auroral oval we compared the corresponding fluxes with a background reference flux, calculated for each orbit.For energetic particles we calculated the average flux of electrons with energies > 100 keV in the polar cap and its standard deviation.We assumed that the measured flux can be classified as ORB electron flux if the difference between this flux and the background flux was greater than 5 standard deviations during the continuous time interval of at least 1 min duration (the separate singlepoint spikes are not taken into account).The nearest pole-ward point that satisfies the described criterion is selected as the polar boundary of the ORB.These selection criteria show stable results of the ORB detection, but as a rule they define the boundary at the end of the decline of electron intensity from ORB maximum to the background level.This means that electron fluxes lower than the established criteria, and belonging to the ORB, could be missed.This is why it might shift slightly the obtained boundary equatorward with respect to the true boundary, especially in the case of lowintensity ORB crossing (see the introduction).This means that we could underestimate the number of events for which the polar boundary of the ORB is observed inside the auroral oval.Such underestimation changes slightly the results of the statistical analysis.However, it cannot change the answer to the main question: whether the trapping boundary is located inside the oval or coincides with its equatorial boundary. The automatic detection of the polar boundary of ORB, identified as the trapping boundary, might be affected by the sharp local increases in the energetic electron fluxes sometimes observed at the trapping boundary (see Imhof et al., 1990Imhof et al., , 1991Imhof et al., , 1992Imhof et al., , 1993) ) or just poleward of it.Such fluxes are usually much smaller than the maximum fluxes of the ORB precipitating electrons.Nevertheless, they can be observed during a few hours at the same location in a few consecutive polar satellite orbits (Myagkova et al., 2010;Antonova et al., 2011b;Riazantseva et al., 2012), and alter the automatic detection of the boundary.It was one of the reasons to do a visual inspection of all events. To calculate the position of the auroral oval boundary, we use the value of the total energy flux.We produce the spectra approximation from 0.032 till 16.64 keV with energy step dε = 0.01 keV.Energy flux was calculated as the integral characteristic of low-energy electron spectrum Flux ε = 2π j (ε) • εd(ε) (where j (ε) is the flux for the current value of energy ε).We first calculated the average value and standard deviation of the electron energy flux measured at L < 3 R E , where L is the McIlwain parameter.In the next step we considered the fluxes that exceed the background flux by 7 standard deviations.If the obtained boundary was located at L > 3 R E , we repeated this procedure but calculated the average flux and its standard deviation up to the boundary, determined in the first step.Based on the Vorobjev et al. (2013) definition of the auroral oval, we also imposed an additional criterion on the value of the total energy electron flux: it should be greater than 0.2 erg cm −2 s −1 .The results obtained were also confirmed by a visual inspection. We used the AE index (Davis and Sugiura, 1966) that represents the dynamics of the auroral electrojet to identify the intervals of substorm activity.We also used the Polar Cap (PC) index (Troshichev and Andrezen, 1985;Troshichev and Janzhura, 2012), which was created as a proxy of the dawndusk electric field in the polar cap and Region 1 currents of Iijima and Potemra (1976) intensity.We took for the analysis the 1 min values of the AE and PC indices when the spacecraft was at the equatorial boundary of the auroral oval.Tak-ing into account that there are two PC indices, obtained for the Northern Hemisphere (PCN) and Southern Hemisphere (PCS), we used the corresponding PCN (PCS) indices for northern (southern) crossings of the auroral oval. Figure 1 shows an example of two crossings of the auroral oval in the morning and evening MLT sectors on 1 February 2010, when the trapping boundary was located inside the auroral oval.The top panel shows the spectrogram of low-energy electrons, and the bottom panel shows total energy flux, calculated from the electron spectra presented on the top (red solid line) and counts of electrons with energy ≥ 100 keV (green solid line).Dashed red lines in both panels indicate the position of the equatorial boundaries of the auroral oval and dashed green lines show the position of the polar boundaries of ORB.It is clearly seen that the curves of total energy flux and counts of electrons with energy ≥ 100 keV show the position of the trapping boundary poleward of the equatorial boundary of the auroral oval. According to the omniweb database (http://omniweb.gsfc.nasa.gov; last access: November 2017), the solar wind number density (N SW ) and velocity (V SW ), and three components of the interplanetary magnetic field (IMF) for both equatorial borders, were very common: B x ≈ 2 nT, B y ≈ −4 nT, B z ≈ −1 nT, N SW ≈ 6 cm −3 , and V SW ≈ 450 km s −1 .This event took place in the absence of geomagnetic storms (Dst ≈ −7 nT), and during moderate auroral activity (AE ≈ 300 nT and AL ≈ −260 nT).The values of the PC index were also moderate (PCS ≈ 3) (see http://pcindex.org; last access: November 2017).As can be seen, for this event the trapping boundary of energetic electrons, shown by green dashed lines, is located inside the auroral oval.The differences between the latitudes of the equatorial boundary of the oval and the trapping boundary, Lat, are equal to −5.8 • for the dawn and −1.7 • for the dusk boundaries. Figure 2 shows an event of the trapping boundary located outside the auroral oval observed on 17 January 2010.The satellite crossed twice the auroral oval during very quiet geomagnetic conditions (B x ≈ 2 nT, B y ≈ −1 nT, B z ≈ 2 nT, N SW ≈ 6 cm −3 , and V SW ≈ 350 km s −1 , Dst ≈ −2 nT, AE ≈ 15 nT, AL ≈ −15 nT, PCN < 1).The observed difference was comparatively small: Lat = 1 • for the dawn and 3.3 • for the dusk boundaries. Comparison of events shown in Figs. 1 and 2 could bring a conclusion that the relative location of the trapping boundary and the equatorial boundary of the auroral oval might be affected by the shift of the oval to higher latitudes with the decrease in the geomagnetic activity.However, there are many other events observed for low activity for which the trapping boundary was observed inside the oval.One example of such kinds of events is shown in Fig. 3. It took place on 26 January 2010 during quiet geomagnetic conditions (IMF B x ≈ −2 nT, B y ≈ 4 nT, B z ≈ −1.5 nT, N SW ≈ 3.5 cm −3 , and V SW ≈ 370 km s −1 , Dst ≈ −17 nT, AE ≈ 50 nT, AL ≈ −30 nT, PCS ≈ 2).For this event, Lat = −5.1 • for the dawn and −2.2 • for the dusk sectors.The existence of different types of events requires making a statistical analysis to clarify how the geomagnetic conditions could affect the relative location of both boundaries. Statistical analysis We analyzed the data from METEOR-M1, obtained for more than 6200 auroral ovals and the outer boundary of the ORB crossings.For each crossing, we determined the difference between the geomagnetic latitudes of the equatorial boundary of the auroral oval and of the trapping boundary, Lat.The negative difference Lat < 0 means that the trapping To analyze how these differences could be affected by geomagnetic activity, we divided all data into two data sets according to the AE or PC indices.Figure 4 shows the distribution of the latitude differences Lat for AE > 150 nT and AE < 150 nT for the Northern Hemisphere (a) and Southern Hemisphere (b).As can be seen, the number of events for which the trapping boundary is observed inside the auroral oval increases significantly with the increase in geomagnetic activity, quantified through the AE index.For AE > 150 nT the trapping boundary is located inside the auroral oval for the majority of events for both hemispheres, while for AE < 150 the trend is not so clear -the number of events where the trapping boundary is located inside and outside of the auroral oval is nearly the same.However, for both sets there are a comparatively large number of events, for which this difference is comparatively small. Figure 5 shows the distribution of the latitude differences Lat for PC > 1 and < 1 and for the Northern Hemisphere (a) and Southern Hemisphere(b), respectively.Comparing Figs. 4 and 5, we can see that both distributions are very similar, which can be explained by the high correlation between the AE and PC indices obtained by Vennerstrøm et al. (1991).This correlation is related to the formation of ionospheric current systems as a result of the magnetosphereionosphere interactions, and the dominant role of the Region N shows the number of events under described criteria. 1 currents of Iijima and Potemra (1976) in the formation of the PC index (Troshichev and Janzhura, 2012).However, the obtained similarity in the behavior of the boundaries, using the AE and PC indices as separate measures of geomagnetic activity, was not evident at the beginning of this study.This supports the picture obtained by Akasofu (1968) in which the trapping boundary is located inside the auroral oval.We underline that the described effect can be clearly seen only in the case of simultaneous measurements of plasma and energetic electrons onboard the same satellite, which allow us to observe the trapping boundary inside the auroral oval directly during the local measurements.The statistical comparison of boundaries masks this effect, because the scattering of the position of the discussed boundaries in different crossings can be rather large (the standard deviation in the statistical position of the boundaries ≈ ±2 • for the trapping boundaries and ≈ ±3 • for the equatorial boundaries of the auroral oval), whereas the main parts of Lat distributions in Figs. 4 and 5 show the difference between boundaries within the limits ±2 • in the case of low geomagnetic activity.The observed scatterings in positions of the boundaries are in agreement with early established scattering of the auroral oval boundaries (see Vorobjev et al., 2013, and references therein) and the outer ORB boundary (Kanekal et al., 1998;Kalegaev et al., 2018). The analysis of the shifts of the studied boundaries with the increase in geomagnetic activity requires special attention and is far from the main subject of our research.Figure 6 shows the L (McIlwain parameter) distribution of both boundaries for AE < 150 and AE > 150 nT in both hemispheres.It is possible to see the real shift of the equatorial boundary of the auroral oval equatorward with the increase in AE, which is well known due to multiple auroral oval observations.At the same time the position of the trapping boundary practically does not change with the increase in AE.This result is in agreement with Kanekal et al. (1998), in that, in comparison with plasma boundaries, the energetic particle boundaries show a lower degree of correlation with solar wind B z , V B z , and the Kp index of geomagnetic activity. Discussion and conclusions We analyzed the relative position of the trapping boundary and the equatorial boundary of the auroral oval using simultaneous measurements of the energetic electrons with energy > 100 keV and the auroral electrons made at the same METEOR-M1 satellite.Previous comparisons of the relative position of these boundaries were made mostly statistically using data from different satellites.Our analysis shows that the differences in the positions of both boundaries are typically smaller than the statistical scattering in the position of each boundary.This fact explains why previous statistical studies led to different conclusions, and why the use of statistical results about the location of each boundary cannot answer the question about the relative position of the trapping boundary and the equatorial boundary of the auroral oval. Our study shows the trapping boundary is often located inside the auroral oval.The number of such events would be enhanced if instruments of better sensitivity were used.This is because the trapping boundary is defined as the boundary where particle fluxes become lower than a threshold determined by the sensitivity of a detector in the case of a low level of electron flux inside the ORB, so an increase in the sensitivity would move the detected trapping boundary poleward, i.e., deeper inside the auroral oval.The analysis of the latitudinal difference in the position of both boundaries for AE more or less than 150 nT, and for PC more or less than 1, shows that the number of events when the trapping boundary is observed inside the auroral oval significantly increases with both AE and PC indices. The location of the trapping boundary inside the auroral oval agrees with the latest results on the auroral oval mapping discussed by Antonova et al. (2017).They argue that the auroral oval has the form of a comparatively thick ring for all MLTs.Mapping of the plasma sheet to the ionospheric altitudes cannot produce the structure with non-zero thickness near noon.Therefore, it seems natural to map the auroral oval into the plasma ring that surrounds the Earth, as selected by Antonova et al. (2013Antonova et al. ( , 2014a)), and filled with plasma similar to the plasma in the plasma sheet.Results of Antonova et al. (2014bAntonova et al. ( , 2015) ) and Kirpichev et al. (2016) also support such a conclusion and locate the quiet time equatorial boundary of the auroral oval at ∼ 7 R E near midnight and the polar boundary at ∼ 10-13 R E .It is also important to remember that starting from Vernov et al. (1969) this magnetospheric region is classified as the region of quasitrapping for energetic particles.It contains enclosed drift trajectories inside the magnetosphere, and only particles with near to 90 • pitch angles have drift trajectories crossing the magnetopause.The drift trajectories of particles with other pitch angles are locked inside the magnetosphere.Therefore, the registration of the trapping boundary of energetic electrons with nearly zero pitch angles inside the auroral oval seems quite natural. The observation of the trapping boundary of energetic electrons inside the oval can also be important for the solution of the problem of acceleration of electrons in the ORB, taking into account that the injection of a seed population of relativistic electrons during magnetic storms takes place at the equatorial boundary of the auroral oval (see the results and discussion in Antonova and Stepanova, 2015).Electrons of such a seed population must be trapped inside the magnetosphere and further accelerated to relativistic energies during the recovery phase of a storm, forming a new ORB.Our current studies were done for comparatively quiet geomagnetic conditions.They also point out the necessity to keep studying the position of the ORB boundaries, taking into account an overlapping of the part of the auroral oval and the ORB, using a more sophisticated instrument for the measurement of energetic electrons, and to extend this study to the geomagnetic storm time intervals.For our study we used integrated fluxes of the precipitating electrons with the energy > 100 keV.Hence, our results provide the information about an averaged value of polar boundaries, which might vary significantly from the dynamic low-energy seed population (∼ 100 keV) up to the high (ultra-relativistic energies > 1 MeV), taking into account that the seed electron acceleration to higher energies and the radial diffusion contribute to the redistribution of the electron population (see Reeves et al., 2013;Zhao and Li, 2013;Turner et al., 2015aTurner et al., , b, 2017)).It is necessary to add that the recent results including the observations of the Van Allen probes has led to significant advances in the study of the dynamics of the ORB.For example, Ripoll et al. (2015) showed the existence of a rather stable core of the ORB.The energy dependence of the inner boundary of the ORB was carefully analyzed by Reeves et al. (2016); Ripoll et al. (2017), and injection of the seed population at low latitudes was studied by Turner et al. (2015a).Recent studies (Makarevich et al., 2009;Kunduri et al., 2017;Lejosne and Mozer, 2017) are of special interest, showing a strong increase in transverse electric fields in subauroral polarization streams (SAPS), which according to Lejosne et al. (2018) can modify the picture of particle injection in the slot region.However, it will be interesting to continue research of the outer radiation belt considering the results obtained in our paper. In summary, we can conclude that the trapping boundary of electrons with energy > 100 KeV, which coincides with the polar boundary of the ORB, is often located inside the auroral oval.This applies almost always to high geomagnetic activity times and also, though less often, to low geomagnetic activity times.All this might help to re-analyze the relation between the dynamics of radiation belts and auroral phenomena. Author contributions.MOR produced the statistical analysis of measurements, EEA and MVS developed the theoretical justification for the research, BVM and IAR designed the instruments, VOB produced the primary data processing, and NVS took part in statistical analysis. Competing interests.The authors declare that they have no conflict of interest. Figure 1 . Figure 1.An example of the location of the polar boundary of ORB inside the auroral oval at AE > 150 nT.(a) Spectrogram of low-energy electrons; (b) red solid line -total energy flux, calculated from the electron spectra presented on the top; green solid line -counts of electrons with energy ≥ 100 KeV; dashed red lines mark the position of the equatorial boundaries of the auroral oval; dashed green lines -the position of the polar boundaries of ORB. Figure 2 . Figure 2.An example of observation of the polar boundary of ORB outside the auroral oval at AE < 150 nT.The notations are the same as in Fig. 1. Figure 3 . Figure 3.An example of observation of the polar boundary of ORB inside the auroral oval at AE < 150 nT.The notations are the same as in Fig. 1. Figure 4 . Figure 4.The distribution of Lat for AE > 150 nT (red bins) and < 150 nT (blue bins) for the Northern Hemisphere (a) and Southern Hemisphere (b).N shows the number of events under the described criteria. Figure 5 . Figure 5.The distribution of Lat for PC > 1 (red bins) and < 1 (blue bins) for the Northern Hemisphere (a) and Southern Hemisphere (b).N shows the number of events under described criteria. Figure 6 . Figure 6.The distributions of the position of the equatorial boundary of the auroral oval (green bins) and the polar ORB boundary (red bins) from L (where L is the McIlwain parameter) for the Northern Hemisphere (a, b) and Southern Hemisphere (c, d) for AE < 150 nT (a, c) and AE > 150 nT (b, d).	6,476.4	2018-08-21T00:00:00.000	[ "Physics" ]
Particularities of Exergy Analysis in Air Conditioning Systems The paper examines two methods of choosing the initial reference point for exergy of the moist air flow. The first method is characterized by a zero exergy value at the current temperature and humidity of the outdoor air, which are variable both in daily and annual periods. Another is characterized by constant values of parameters of moist indoor air (its temperature and humidity in the room). A comparative study has shown the advantages of the second method: greater stability of values in air processing processes and their convenient presentation on exergy chart. However, it should be noted that this method provides a non-standard technical representation in which the entire energy flow is reduced to zero. The climatic information (temperature and humidity of the outside air) for numerical-analytical experiment has been used from official open sources. The city where the air conditioning system was supposedly installed is Stavropol. The research has been carried out in the summer and cold periods of the year. It is assumed that there are not installed any large heat sources in the room and the main heat sources appeared depending on the period of the year. Additionally, the air conditioning system was equipped with a recirculation line from the serviced room with a mechanical ventilator. The recirculated air entered the mixing chamber in the air conditioning unit which is reducing the consumption of heat energy in the cold season. There are the processes of air conditioning on the Ramzin's h–d-diagram for warm and cold operational modes of air conditioning unit, the air conditioning system under consideration. Introduction As a rule, the analysis of energy parameters of air conditioning systems is carried out using the balance method, taking into account only the first law of thermodynamics. The drawback of this method is that it does not differentiate types of energy and their energy value in relation to the environment. The very concept of the environment is hidden in the analysis, and the strengths and weaknesses of individual processes and sectors of the energy-consuming system are not disclosed. Therefore, the method of thermodynamic analysis using the concept of exergy (a workable part of energy) has, in our opinion, insufficiently employed reserves. The classical development of the exergy method of thermodynamic analysis was commenced in the end of the XIX century (A. Stodola, L. Guye) and continued during the entire twentieth century (Z. Rant, Ya. Shargut, R. Petela, V. M. Brodyansky, V. S. Martynovsky, S. S. Dolinsky and others). The exergy method of thermodynamic analysis is based both on the first and the second laws of thermodynamics and allows to take into account the quality of energy in terms of obtaining a useful effect under specific environment conditions (t and d) [1][2][3][4]. Based on the evaluation of purely physical processes, the exergy method allows to find the most important areas of application of energy-saving engineering [5][6][7][8] and to define new technical requirements for air conditioning systems ensuring their "innovative" development [9]. At the moment, the method under consideration has found application in many branches of industry [10][11][12]. Although the state of the system with the minimum potential difference relative to the environment has traditionally been selected as the reference point for exergy, this choice is complicated in case of air conditioning systems because there is no stability of environment conditions [13][14][15] and the value of exergy increment of air flow is signvariable. Methods of exergy efficiency assessment Efficiency of major units of air conditioning installation. While attempting to assess the efficiency of the air conditioner unit, it is necessary to correlate the exergy difference of the moist air flow (useful effect in this unit or device) with the exergy expenditures, which include exergy differences of heat transfer medium, mass transfer, electrical capacity of pump drives, ventilators or other means for creating differential air pressure in the conditioner units or in the ductworks. Formation of the exergy balance in the mixing chamber can result in the following expression: where Е R -exergy of the moist air flow in the room; G r -mass flow of the recirculated air; Е Oexergy of the moist outdoor air flow; G I -mass flow of the inlet air; Е M -exergy of the air flow mixture. After simple algebraic transformations and with regard to the zero reference level of exergy, introduced earlier The right-hand member expresses the useful effect of the process of air flows mixing, and the left-hand member expresses the expended air exergy. Taking into account the expenditures of the electrical energy, it is possible to formulate the exergetic efficiency factor for the mixing chamber in the winter season in the form of the following expression: where N r -energy expended by the recirculating ventilator; N 5 -amount of energy of the suction ventilator expended on the overcoming of aerodynamic resistance of the air mixing chamber. The notation should be introduced G r /G I = a. Then the exergy efficiency factor for the mixing chamber may be written as follows: where a -ratio between G r and G I . In the summer season (due to the heating of air in the recirculating duct) the expression (4) will be written as follows: The air heater efficiency is the ratio of the air exergy change to the change of heat transfer medium exergy and the expended electrical energy 2,4 , air calorifier heat transfer med where ΔE air -difference between exergy of air flow mixture and air before air washer; ΔE heat transfer medthe change of heat transfer medium exergy; N 2,4expended electrical energy ensuring the operation of this unit. The air washer efficiency is the ratio of the air exergy difference in this unit to the change of water exergy and the expended electrical energy where ΔE air(1) -difference between exergy of air after air washer and air before air washer; ΔE waterthe change of water exergy; N 3 -aggregate expenditures of electrical energy of pumps, ventilators, and other devices ensuring the operation of this unit. Exergy efficiency of the air conditioning system. The efficiency of the air conditioning system in a simplified form can be assessed by the correlation of the change in the inlet air exergy (conditioning process, which is useful, assimilating heat-and-moisture excesses in the room) with the change of exergy immediately after the ingress of air in the air conditioning installation until the exit of air from it. The simplified analytical expression of the efficiency is written as follows: where Е I -exergy of the moist air flow at the entrance to the room. Taking into consideration the zero level of exergy, taken in this paper as E R = 0: As the reference level of exergy, taken by us, is constant, there is no need to recalculate all the values of the flow exergy in the context of the change in the outdoor air parameters, only the E in or ΔЕ 5 should be recounted. With regard to expressions (4), (4'), (5) and (6) the exergy efficiency of the air conditioning system can be represented as a chart (Fig. 1), where values of expended exergy are singled out on the X-axis, and values of useful exergy are singled out on the Y-axis. In the summer season the preheating air heater is turned off, however, it creates the aerodynamic resistance to the movement of the air flow. Thus, 4 N ′ expresses the fraction of the electrical energy of the ventilator, required for the overcoming of the aerodynamic resistance of the turned off air heater. The aggregate expenditures of the electrical energy by the ventilator can be written as follows: where N 1 -expenditure of electrical energy of the ventilator on the overcoming of the aerodynamic resistance of the ductwork (authors decided not to include this expenditure due to its extremely specific calculation); N 2 , N 3 , N 4 , N 5 -expenditures of the electrical energy of the ventilator on the overcoming of the aerodynamic resistance of the reheat air heater, air washer, preheating air heater and mixing chamber. The results from expressions (4), (4'), (5) and (6), as well as Fig. 1 that in order to increase the exergy efficiency of the air conditioning system it is necessary to reduce the amount of the expended electrical energy. It can be achieved by the reduction of aerodynamic resistances of units of the air conditioning system, hydraulic resistances in the air washer, optimization of the ventilator performance. This analysis is most appropriate to be conducted precisely when selecting the zero reference point of exergy in the attended room, since this makes it convenient to graphically and analytically depict the exergy efficiency of the air conditioning system. Results and discussion. Rationale for introduction of the new reference point for exergy Depending on the area of construction, the average daily temperature difference can exceed 10 о C, and the average annual temperature can exceed 50 о C. Relative humidity varies from 20 to 100 %. Such an amplitude of temperature and humidity fluctuations leads to inconvenience in calculating the exergy efficiency of systems and equipment for creating a given airthermal microclimate. It is also difficult to compare effectiveness of the different versions of systems. It has been considered the change in state of the Stavropol climate region. There are the climatic data from official open sources in Fig. 2, 3. In general terms, the effect of changing the parameters of exergy reference point can be shown in the h-S-diagram (Fig. 4). T T T S S e e c T T T S S where e 1 -exergy of the lower limit of the amplitude of fluctuation of temperature or humidity, kJ/kg; е 2 -exergy of the upper limit of the amplitude of fluctuation of temperature or humidity, kJ/kg; с р -heat capacity of moist air at constant pressure, kJ/(kg⋅K); T 3 -temperature of the flow of moist air, K; T 01 -ambient temperature of the lower limit of the amplitude of climate fluctuation (tgα 1 ), K; T 02 -ambient temperature of the upper limit of the amplitude of climate fluctuation (tgα 1 ), K; S 3 -entropy of the flow of moist air, kJ/kg; S 01 -entropy of the environment of the lower limit of the amplitude of fluctuation of climate parameters, kJ/kg; S 02 -entropy of the environment of the upper limit of the amplitude of climate fluctuation, kJ/kg. Thus, it follows from expression (11) that the same flow of moist air has a different exergy value due to a change in parameters of the exergy reference point. We will determine within what limits this change in exergy can be wrong. Exergy of a moist air flow (e air ) is calculated by the formula from the publication [16]: where e air -exergy of a moist air flow; T x -temperature of the flow of moist air, K; T 0 -ambient temperature, K; d x -moisture content of the flow of moist air, kg/kg; c p air -isobaric heat capacity, J/(kg⋅ o C); R air -universal gas constant for dry air, J/(K⋅mol); d 0 -moisture content of the environment, kg/kg; p x -pressure of moist air in the flow, Pa; p 0 -atmospheric pressure of the environment, Pa. Using formula (2), we will calculate the exergy of the flow relative to the extreme values of fluctuation of characteristics of moist air and compare with the exergy relative to the constant indoor parameters (e 0 ). To do this, let us set a flow of moist air with the parameters: t = 30 о С, ϕ = 60 %. When using the exergy analysis in air conditioning systems, it is necessary to choose not only the temperature but also a representative value of humidity in the room which is constant and ensures comfortable conditions for people. The results of calculations with the selected parameters are summarized in Tab. 1. Tab. 1 shows that ratio e/e 0 ranges from 0.48 to 7.17. The above data indicate that it is necessary to carefully justify the choice of the reference point in open systems. According to the authors, the internal air environment of a building with air parameters maintained at an almost constant heat and humidity level (Fig. 5) can be used as the possible definition of the zero level of exergy. Fig. 5 uses the following notation: Q 1 , ∆d 1heat and moisture transport realized by the air conditioning system; Q 2 , ∆d 2 -heat and mass exchange of the room with the environment (heat input, heat loss and moisture penetration during summer or winter season); Q 3 , ∆d 3 -excess heat and moisture inside the room during winter or summer. This constancy of internal parameters is achieved by the operation of heating, ventilation and air conditioning systems. The operation of one of the simplest air conditioning systems [17][18][19] (Fig. 6) for the summer and winter seasons on the basis of the entered exergy reference point can be represented by the diagram shown in Fig. 7. Fig. 6, 7 uses the following notation: O (O′) -parameters of outdoor air during winter (summer) season; M (M ′) -air parameters at the outlet from the mixing chamber during winter (summer) season; C -air parameters at the outlet from the preheating calorifer; W -air parameters at the outlet from the air washer; I -air parameters at the outlet from the reheat calorifer (inlet air); R -air parameters in the room; R′ -air parameters in the recirculation air duct (it is assumed that the process R-R′ goes along d = const, that is, the temperature gradient in the room is not taken into account; since the air intake for recirculation comes from the working area of the room, when accounting for it the process R-R′ goes along ε = const); Representation of the process of heat and moisture treatment of the inlet air in the h-d-diagram is shown in Fig. 8, 9 [20].	3,252.2	2022-02-04T00:00:00.000	[ "Engineering", "Environmental Science" ]
Strong Washout Approximation to Resonant Leptogenesis We show that the effective decay asymmetry for resonant Leptogenesis in the strong washout regime with two sterile neutrinos and a single active flavour can in wide regions of parameter space be approximated by its late-time limit $\varepsilon=X\sin(2\varphi)/(X^2+\sin^2\varphi)$, where $X=8\pi\Delta/(\|Y_1\|^2+\|Y_2\|^2)$, $\Delta=4(M_1-M_2)/(M_1+M_2)$, $\varphi=\arg(Y_2/Y_1)$, and $M_{1,2}$, $Y_{1,2}$ are the masses and Yukawa couplings of the sterile neutrinos. This approximation in particular extends to parametric regions where $\|Y_{1,2}\|^2\gg \Delta$, {\it i.e.} where the width dominates the mass splitting. We generalise the formula for the effective decay asymmetry to the case of several flavours of active leptons and demonstrate how this quantity can be used to calculate the lepton asymmetry for phenomenological scenarios that are in agreement with the observed neutrino oscillations. We establish analytic criteria for the validity of the late-time approximation for the decay asymmetry and compare these with numerical results that are obtained by solving for the mixing and the oscillations of the sterile neutrinos. For phenomenologically viable models with two sterile neutrinos, we find that the flavoured effective late-time decay asymmetry can be applied throughout parameter space. Introduction Resonant enhancement from mass degeneracies is a way of obtaining sizeable chargeparity (CP ) violating effects, that would be strongly suppressed by powers of small couplings otherwise. Depending on the ratio of the mass splitting to the decay rate in a system of mixing particles, it may either be more advantageous to describe the CPviolating effects as a time-dependent phenomenon due to mixing and oscillations of the almost mass-degenerate states, or, further away from the mass degeneracy, in terms of a time-independent effective decay asymmetry [1]. The important role that resonant CP -violation assumes in many systems that can be tested in the laboratory has lead • We show how the non-relativistic approximations and simplifications, that are of relevance in the strong washout regime, follow from the general treatment of Refs. [24,25]. • We define the effective decay asymmetry ε as the lepton asymmetry that results on average from the decay of one out-of-equilibrium sterile neutrino. When compared to the decay asymmetry introduced in Ref. [28], this definition resembles more closely the expressions that are usually employed in Leptogenesis calculations, such that it leads to a simple and straightforward way of obtaining the lepton asymmetry. We present the relevant equations that determine the freeze-out asymmetry as well as example solutions. • We give an expression for the decay asymmetry taking account of active lepton flavours and their possible correlations. We emphasise that flavour effects should be phenomenologically relevant throughout the parameter space. Again, we illustrate the use of this effective asymmetry with numerical examples. • Since it is crucial for resonant Leptogenesis to treat the decay rate Γ of the sterile neutrinos as matrix-valued, the criterion H/Γ ≪ 1 for the applicability of the approximation in terms of an effective decay asymmetry can only be of schematic meaning. For a simplified scenario with one active lepton flavour only, we determine the smallest eigenvalue associated with the linear differential equation that governs the evolution of the sterile neutrino densities and their flavour-off-diagonal correlations. By comparison with the Hubble rate, this eigenvalue can be used in order to assess whether the approximation in terms of the effective decay asymmetry ε is applicable. • For a phenomenological scenario with two sterile neutrinos, that explains the observed oscillations of active neutrinos, we find that the use of the effective late-time decay-asymmetry can be justified for all regions of parameter space. This conclusion is also based on comparing the eigenvalues of the equations that govern the mixing and the oscillations of the sterile neutrinos with the Hubble expansion rate prior to the freeze out of the lepton asymmetry. Relativistic Resonant Leptogenesis We consider the usual see-saw model for neutrino masses that is given by the Lagrangian Here, the N i are the sterile neutrinos, that observe the Majorana condition N c i = N i , where the superscript c stands for charge conjugation. The Higgs doublet is given by φ and ǫ SU (2) is the antisymmetric, SU(2)-invariant tensor with ǫ 12 SU(2) = 1. The Standard Model (SM) lepton doublets are given by ℓ a , where a = e, µ, τ . When considering the single-flavour model, we drop the index a on the fields ℓ as well as the on Yukawa couplings Y . We make use of the freedom of field redefinitions in order to choose the symmetric matrix M to be real and diagonal, and we refer to the diagonal elements as We describe the generation of the comoving lepton charge density q ℓab in terms of a source term S ab and a washout term W as [24,25] The charge density accounts for the gauge multiplicity, hence we include here the factor g w = 2. Moreover, as mentioned in the Introduction, we allow for the possibility of correlations of the SM lepton flavours. The expansion of the Universe is accounted for through the metric in conformal coordinates g µν = a(η)η µν , where η µν is the Minkowski metric, a(η) is the scale factor and η is conformal time. A prime denotes a derivative with respect to η. In Ref. [24], it is shown that the source term for resonant Leptogenesis through the lepton-number violating Majorana mass can be computed by first solving for the flavour correlations of the oscillating sterile neutrinos, similar to the standard calculations for CP -violation in mixing meson systems [1] or to the lepton-number conserving source in the scenarios that are usually referred to as Leptogenesis from neutrino oscillations [25,[30][31][32][33][34][35][36][37]. The result of Ref. [24] is generalised to include flavour correlations in Ref. [25] and then reads where/ Σ A N (k) is the reduced spectral self-energy of the sterile neutrinos as defined in Ref. [25]. The correlations of the sterile neutrinos are described by iδS N ij (k). Besides the indices i, j for the sterile neutrino flavours, this function corresponds to a rank two tensor in terms of Dirac spinors. It satisfies Kadanoff-Baym equations and the solutions can be decomposed as where σ and ρ are Pauli matrices. In the resonant regime \|M i − M j \| ≪M, the different components may be written as [24] g ahij (k) = 2πδ(k 2 − a 2M 2 )2k 0 δf ahij , whereM = (M i + M j )/2. Moreover, the Kadanoff-Baym equations also imply the relations [24] δf 1hij (k) = δf 3hij (k)a In view of the non-relativistic approximation below, the a = 0 component is of particular interest. The function δf 0hij may be interpreted as the distribution function of the sterile neutrinos and of their flavour correlations. Using the decomposition (4) and the relations (6), the source term (3) can be expressed as S ab ≡ +hk , ω(k) = √ k 2 + aM 2 ,k = (\|k\|, k 0 k/\|k\|) and δf * 0hij (k 0 ) = δf 0hij (−k 0 ). The Kadanoff-Baym equations imply that the sterile neutrino distributions and their correlations satisfy [24,28] where f eq is the equilibrium Fermi-Dirac distribution of the sterile neutrinos. One may alternatively derive this equation using a more heuristic approach in terms of a density matrix instead of the two-point function of the sterile neutrinos. The solution may be substituted back into the source term (3) and eventually into the equation for generating the lepton charge-density (2) in order to obtain predictions for the freeze-out asymmetry. Non-Relativistic Approximations Now, we consider a situation, whereM ≫ T (and all sterile neutrinos are assumed to be close together in mass, \|M i − M j \| ≪M ), as it is of relevance in strong washout scenarios around the time of freeze out. The main simplification arises here due to the fact that modes that do not satisfy \|k\| ≪ aM are strongly Maxwell suppressed, such that we may approximate the four momenta as Due to the same reason, we can neglect the thermal contributions to the spectral selfenergy of the sterile neutrinos, such that it takes its vacuum form For the terms involvingΣ A N that appear in Eq. (8), this implies that we can take the approximate forms Then, we integrate that equation with the result where we have defined This is the comoving non-equilibrium number density of sterile neutrinos, δn ± 0hij = δn ± * 0hji , which is of the form of a Hermitian matrix. The comoving equilibrium number density is denoted by n eq . The Majorana nature of the sterile neutrinos implies that δn + 0hij = δn − * 0hij , a property that is directly inherited from the distribution δf 0h (±ω, k) and that is derived in Ref. [24]. Note that in the non-relativistic limit, the solutions for the sterile neutrino densities are helicity independent. The relativistic generalisation that accounts for helicity is worked out in Ref. [24]. In order to substitute these results into the source term (3), we use the relations (6) that imply a vanishing axial density δf 3hij in the non-relativistic limit. Note moreover that the Dirac trace in Eq. (3) selects then contributions from δf 0h only. The result for the flavoured source term in the non-relativistic approximation then is Note that we do not sum over h here and make use of the fact that in the non-relativistic limit, we can approximate n ± 0+ij = n ± 0−ij . Strong Washout Regime In the radiation-dominated Universe, a(η) = a R η. A particularly convenient choice is η = 1/T , what requires a R = m Pl 45/(4g ⋆ π 3 ) ≡ T 2 /H. Moreover, one can then easily define the parameter z =M /T =M η, that is often used in Leptogenesis calculations. We investigate under which circumstances the maximal enhancement of the decay asymmetry can be attained. For this purpose, we solve the Eq. (12) in the form that is obtained when using above parametrisation in terms of z where andΓ = 1/(8π). Since larger entries of Y correspond to larger washout, it is proposed in Ref. [28] to obtain a simplified approximation in the strong washout regime by neglecting the first term of Eq. (15). To put this more precisely, note that out of the first three terms of Eq. (15), which are the homogeneous terms, the second and the third grow with z. Therefore, neglecting the first term corresponds to taking the late-time limit of the solution. If the late time-limit applies before the freeze-out of the lepton asymmetry, that occurs for z = z f , it leads to a valid approximation of the freeze-out asymmetry. The evolution of the lepton asymmetry is governed by the equation where the last equality defines the time-dependent effective decay asymmetry ε ab (z), in consistency with Eq. (21) below. In view of flavour effects, we have written this in terms of the asymmetries ∆ ℓaa = B/3 − q ℓaa that are conserved by SM interactions and where B is the baryon number density. Off-diagonal flavour-correlations can be accounted for by ∆ ℓab = −q ℓab for a = b, if necessary. Moreover, q φ stands for the charge density in Higgs bosons, that is present in general. We have also expressed Eq. (17) in a way that defines the decay asymmetry ε as the the lepton asymmetry that results from one sterile neutrino that initially drops out of equilibrium as a mass eigenstate. Note that the factor of four in front of ε ab arises because of the two helicity eigenstates of to the two sterile neutrinos. In addition, this equation includes the crucial washout term W in its flavoured variant, that is derived in Ref. [21] 1 , see also Refs. [38,39]. In the present context, we are interested in the situation where the sterile neutrinos are non-relativistic, such that the washout matrix can be approximated by Lepton-flavour violating interactions mediated through SM Yukawa-couplings are described by the term Γ fl ℓab , that is defined and explained in Ref. [21]. In the fully flavoured approximation, one assumes that these interaction delete the off-diagonal correlations in q ℓ and ∆. Effectively, one may then just set the off-diagonal elements to zero and ignore Γ fl ℓab . Solving Eq. (15) when neglecting the derivatives acting on δn ± 0h yields for the offdiagonal correlations (i = j) of the sterile neutrinos Comparing with Eqs. (14) and (17), we identify the time-dependent effective decayasymmetry It can be straightforwardly interpreted as the asymmetry yield per sterile neutrino that drops out of equilibrium. This quantity differs from the CP -violating parameter defined in Ref. [28], that quantifies the yield in terms of the out-of-equilibrium neutrinos that are present at a given point in time. The discrepancy is due to the time delay in the transition from diagonal out-of equilibrium densities to off-diagonal correlations due to oscillations. We write the late-time limit of the decay asymmetry (21) by dropping the argument z, . e.g. ε ≡ ε(∞), for which we find when using Eq. (19) Provided the strong washout approximation holds, it is then easy to solve Eq. (15) numerically. In the fully flavoured regime, q ℓab can be reduced to its diagonal components and the flavoured asymmetry can be calculated in straightforward generalisation (see e.g. Refs. [40,41]) of the methods for the single-flavour case [9,42]. The flavoured expression (22) for the decay asymmetry in resonant Leptogenesis is of importance throughout the parameter space. If the sterile neutrino mass is below 10 9 GeV, the usual treatment of flavoured Leptogenesis should apply, i.e. ε ab can be reduced to its diagonal components, because interactions mediated by SM-lepton Yukawa-couplings effectively erase all coherence [43,44]. (See however Ref. [38] for a counterexample, where even Yukawa-suppressed correlations at low temperature are of importance, due to a special flavour alignment.) At higher temperatures, when the asymmetry results from the decay of one sterile neutrino only, it is sufficient to either deal with two (a linear combination of e and µ) or one single flavour (a linear combination of e, µ and τ ) only. Once the decay of more than one neutrino contributes, as it is the case for resonant Leptogenesis, there will be decay asymmetries in different linear combinations [39,45] that in general cannot be aligned simultaneously. It then appears simplest to take the full expression for ε ab , including the off-diagonal correlations, and compute their evolution following Ref. [21] (see also Ref. [38]). Applicability of Approximations The effective decay asymmetry (22) and the equation for the evolution of the lepton asymmetry (17) offer a simple way of accurately calculating the freeze-out asymmetry even in the resonant regime, where approximations based on the mass splitting of the sterile neutrinos being larger than their width are not applicable. In order to describe the parametric range of validity of neglecting derivatives acting on δn ± 0h in Eq. (15) more precisely, we first take the simplifying assumption of a single lepton flavour only. The effective decay asymmetry can then be expressed in the simple form where X is a dimensionless parameter defined as and where ∆ = is the normalised mass difference, y 1,2 = \|Y 1,2 \| and ϕ is the relative phase of the Yukawa couplings, ϕ = arg(Y 2 /Y 1 ). Note that the solutions to Eq. (15) remain unaltered as a function of z, provided we leave the ratiosM : ∆ : Y 2 invariant. Therefore, such a rescaling leaves ε(z) and the late-time solutions unchanged as well. This invariance can also be explicitly observed in the late-time asymmetry (23). The late-time asymmetry (23) can also be constructed from the solutions given in Ref. [28], such that we note agreement with the results of that work. However, our definition for ε differs from the CP -violating parameter proposed in Ref. [28]. Our choice is motivated by the fact that the result (23) quantifies the yield of lepton asymmetry in a transparent manner and that it allows for a straightforward calculation of the final asymmetry, provided the late-time limit is a good approximation at the time of freeze out, what we illustrate in the remainder of this Section. The expression for the late-time decay asymmetry (23) only leads to an accurate approximation for the process of Leptogenesis, provided the solutions to Eq. (15) reach their late-time form, where the derivatives acting on δn ± 0h may be neglected, prior to the freeze-out of the asymmetry. Based on this requirement, we derive a more precise analytical condition that allows to identify the parametric regions where neglecting the derivatives of δn ± 0h is indeed justified. Since δn ± 0h are Hermitian two by two matrices and moreover, n + 0h = n −t 0h , Eq. (15) corresponds to a coupled set of four real differential equations. The smallest eigenvalue 2 in vicinity of the parametric points where ε is close to unity [cf. Eq. (29)] is given by ǫ = ǫ R2 , which is presented explicitly by Eq. (B1), or alternatively by , whereȳ 2 = (y 2 1 + y 2 2 )/2. Notice also that ǫ is invariant when keeping the ratioM : ∆ : Y 2 fixed. This is more easily seen in the democratic case y 1 = y 2 , where the smallest eigenvalue is given by where ϑ is the Heaviside step function and where we have definedǭ = (a R z/M )ȳ 2Γ . Since (dn eq /dz)/n eq = O(1) around freeze out, one should require ǫ ≫ 1 in order to neglect derivatives acting on δn ± 0h . [A condition that amounts to requiring that the slowest eigenmode of Eq. (15) is faster than the Hubble expansion rate.] This also implies that ǫ ≫ǭ/ǫ. The quantityǭ/ǫ therefore is of phenomenological interest, because it indicates how strong the washout must at least be such that we can justify the neglect of the derivatives of δn ± 0h . In order to relate to the parameters that are typically employed in calculations on Leptogenesis, note thatǭ/z =K = (K 1 + K 2 )/2, where the K i = y 2 iMΓ /H\| T =M are the usual washout parameters [9]. In order to satisfy ǫ ≫ 1 at the time of freeze-out, that occurs for z = z f = O(10), it follows that we must requirē We can therefore use the ratioǭ/ǫ in order to infer the minimal washout strength that is necessary for consistently neglecting the derivatives of δn ± 0h . Note that the washout strengthK can also be employed as an expansion parameter for a series approximation that generalises the truncation of the derivative of δn ± 0h in Eq. (15) in a systematic manner. Details of this are worked out in Appendix A. It is interesting to consider the situation where, for a given value of X, the phase ϕ maximises the decay asymmetry (23). This occurs for ϕ = ϕ M , where and where the asymmetry is then given by For X → 0, the decay asymmetry attains its maximum value ε → 1. Curiously, in this case the CP -violating phase tends to be vanishing, ϕ M → 0. The exact limit can however Ε Ε Figure 1: The ratioǭ/ǫ of the diagonal relaxation rate of the sterile neutrinos to the smallest eigenvalue, with ϕ given by Eq. (28). In order for the derivatives of δn 0h to be negligible, the washout strength should satisfy relation (27). not be reached because for such an alignment scenario, it takes infinitely long for the off-diagonal correlations in δn ± 0h to build up. In particular, this does not occur before freeze-out. In the examples below, we observe however that it is possible in practice to obtain asymmetries that are at least close to maximal. For comparison, we also comment the opposite regime, where X ≫ 1 (which may still allow for ∆ ≪ 1). In that case the asymmetry is maximal when ϕ M (X ≫ 1) = π/4. Substituting ϕ = ϕ M and the value of X 2 in terms of ε from relation (29) into Eq. (26), we find This ratio vanishes as the asymmetry ε goes to 1, which reflects the fact that for large asymmetries, it takes a longer time to build the off-diagonal correlations in δn ± 0h , and the washout should be sufficiently strong in order for the late-time decay asymmetry ε to be a good approximation. The ratioǭ/ǫ is presented in Figure 1. As an illustration for how to interpret the quantityǭ/ǫ, in Figure 2, we show how the parameter ε(z) [as defined in Eq. (21)] evolves in the case where it approaches the late-time value ε = 0.98. We choose two washout strengths, where the weaker one violates the criterion (27) while the stronger one marginally complies with it. In order to obtain these results, we assume thermal initial distributions for the sterile neutrinos and begin to integrate at z = 0. We observe indeed that when relation (27) holds, where z f = O(10), a stationary form for ε(z) corresponds to a good approximation. To see the effect on the freeze-out lepton asymmetry, we take both, the late-time value ε and the time-dependent solution ε(z), and solve Eq. (17), where we assume one single flavour (and consequently suppress the flavour indices), set q φ = 0 for simplicity and take q ℓ = −∆ ℓ . We express the result in terms of the ratio of the lepton-number to the entropy density s, Y ℓ = −∆ ℓ /s and use the value for s with 106.75 relativistic degrees of freedom. For both washout strengths, we observe that initially, there is a substantial deviation between the solutions for Y ℓ that are based on the time dependent ε(z) and its late-time limit. While for the larger washout strength, the freeze-out asymmetries agree eventually up to about 40% accuracy, there is a discrepancy of about a factor of five for the smaller washout strength, that does clearly not satisfy relation (27). Next, we again take y 1 = y 2 but impose fixed values of ϕ, in order to allow for a deviation from the relation (28). In Figure 3, the ratiosǭ/ǫ are presented as functions of ϕ for various values of ε. The curves exhibit two branches, because for a given asymmetry ε and phase ϕ, Eq. (23) has two solutions for X. The two branches join at the point where there is only one root. It is easy to show, using Eq. (23), that the condition for a unique root is ε = cos(ϕ), for which X = sin(ϕ). There are two more curves that we display in Figure 3. First, we show the ratios of the eigenvalues when identifying ϕ = ϕ M , what fixes X through Eqs. (28), and with Eq. (30), we obtain Second, we determine the value of ϕ that minimises the eigenvalue ratio, what defines the graph From Figure 3, we observe asymptotic proximity between these two curves (31) and (32), and moreover, one can check that the junction points for the two solutions for X are close to these curves as well. This implies that ϕ = ϕ M corresponds to a preferable choice for obtaining large asymmetries not only because it maximises ε but also because at the same time, it minimisesǭ/ǫ and therefore the required washout strength. Again, we present in Figure 4 the evolution of the parameter ε(z) and the leptonnumber to entropy ratio Y ℓ for two different washout strengths, what exemplifies the use of the criterion (27) for approximating the freeze-out asymmetry using the late-time decay asymmetry ε. We now move from the simplifying single-flavour model to a more realistic scenario, where several flavours are present and where we take account of constraints from neutrino oscillation data. In order to avoid a proliferation of free parameters, we consider the case where there are only two sterile neutrinos or, alternatively, where a third sterile neutrino decouples. It follows that one of the masses m 1,2,3 of the observed light neutrino states vanishes, i.e. m 1 = 0 for a normal mass hierarchy, which is what we assume here. This leads to a simplified form of the Casas-Ibarra parametrisation of the Yukawa couplings [46] where U ν is the PMNS matrix and v = 246 GeV is the vacuum expectation value of the Higgs field. Note that here, Y is a 2 × 3 matrix. For the PMNS matrix and for the light neutrino masses, we take the best-fit parameters from the global analysis of Ref. [47], and for simplicity, we fix the Dirac and the Majorana phase therein to be zero. The parameter ̺ is a complex angle, and its imaginary part acts here in absence of the PMNS phases as the only source of CP -violation. Moreover, this imaginary part largely controls the absolute value of cos ̺ and sin ̺, i.e. large imaginary parts imply a large washout strength. For definiteness, we are considering this setup at temperatures of about 10 8 GeV, where all second-generation but none of the first-generation Yukawa couplings are in equilibrium. The qualitative picture does not change when going to different temperatures, where other spectator fields give rise to O(10%) corrections to the freeze-out asymmetries [48][49][50]. We can then relate Moreover, at temperatures below 10 9 GeV, the off-diagonal correlations of the left-handed leptons are strongly suppressed due to the SM Yukawa interactions, such that we can neglect the off-diagonal elements of Eq. (17) (see however Ref. [38], where due to alignments of the Yukawa couplings Y the off-diagonal correlations remain non-negligible at even smaller temperatures). The eigenvalues of the equation for mixing and oscillating sterile neutrinos (15) in terms of the Casas-Ibarra parametrisation are given in Eq. (B3). As the oscillatory contributions due to the mass splitting enter as an imaginary part and the damping contributions due to the Yukawa couplings as a real part, we can find a lower bound on the magnitude of these eigenvalues by setting ∆ = 0, what leads to a considerable simplification of the expressions: We also present the individual flavoured baryon-minus lepton asymmetries \|Y ℓaa \| = \|∆ ℓaa \|/s obtained from Eq. (17), using the time-dependent decay asymmetry (solid) and the late-time limit (dashed). The quantities ∆ ℓaa , q ℓaa and q φ are related through Eqs. (34). Since the smallest ratio is ǫ CI R2 /ǭ CI > ∼ 1/6 for normal hierarchy, neglecting the derivatives on δn ± 0h in Eq. (15) is by the criterion (27) (assuming z f = O(10)) a good approximation everywhere in the strong washout regime of resonant Leptogenesis for the phenomeno-logical model with two sterile neutrinos. Moreover, as washout is always strong in that scenario, what we show in Appendix C, we can conclude that using the late-time asymmetry (22) is a valid approximation for any point in parameter space. For the phenomenological model specified above, we solve Eq. (17) with the effective decay asymmetry (21) based on the full numerical solution to Eqs. (15). This, we compare with the solution obtained when using the late-time limit for the decay asymmetry (22) for all times prior to freeze-out. Since by above arguments, there should be no points where the freeze-out asymmetries obtained by the two methods differ by substantial amounts, we show in Figure 5 the evolutions of ε aa (z) from Eq. (21) and the values of ε aa from Eq. (22), along with the asymmetries \|Y ℓaa \| = \|∆ ℓaa \|/s obtained using the time-dependent and the effective late-time decay asymmetries for a typical point in parameter space, for which the width dominates the mass splitting, ∆ ≪ (tr[Y Y † ]Γ) 2 . As anticipated from the analysis of the eigenvalues, albeit the different time evolution at early stages, the freeze-out asymmetries agree very well. Conclusions We have studied the applicability of the late-time decay asymmetries ε for sterile neutrinos in their multi-flavoured and single-flavoured forms (22) and (23) to computations of the freeze-out asymmetry in resonant Leptogenesis. This has been done by comparison with the results obtained from the time-dependent decay asymmetry (21) that is based on the solution to the evolution equation (15) for the mixing and oscillating sterile neutrinos. The evolution equation can be straightforwardly derived from its relativistic generalisation, that was first presented in Ref. [24]. Following Ref. [28], the approximations (22) and (23) are obtained by neglecting the time derivative acting on the non-equilibrium number densities and correlations in Eq. (15). In addition to the numerical comparisons, to gain analytical insight, we have derived expressions for the eigenvalues of the equation that governs the mixing of the sterile neutrinos and their deviation from equilibrium. This analysis reveals that ε can reach its maximum value one provided ∆ → 0 and ϕ → 0 simultaneously. In that case however, also the smallest eigenvalue of the equation describing mixing and oscillations tends to zero, indicating that the approximation in terms of the late-time decay asymmetry is not valid in that limit. Nonetheless, the quantitative analysis (by studying the smallest eigenvalue as well as the numerical solution) reveals that the late-time asymmetry can be a good approximation already for moderately strong washout, even when ε is close to one. To quantify this, cf. Figures 1 and 3 in conjunction with the criterion (27). An increase of the washout strength generally leads to a better approximation. While the derivation of the single-flavour decay asymmetry (23) makes use of the approximation proposed in Ref. [28], its definition is different from the CP -violating parameter introduced in that work. We find the form that is suggested here somewhat more transparent, as it corresponds to the asymmetry yield per sterile neutrino that initially drops out of equilibrium through the Hubble expansion. Moreover, with its definition as in the present work, the parameter ε can be employed in the same way the usual vacuum decay asymmetry is used in standard calculations on Leptogenesis [8,9,50]. We have exemplified this point by explicitly calculating the freeze-out lepton asymmetry in a phenomenological see-saw model that is consistent with the neutrino mixing and oscillation data. We can draw the conclusion that the approximation proposed in Ref. [28], which leads to the late-time asymmetries that we derive and study here, is applicable for Leptogenesis calculations in the strong washout regime of the single-flavour model, unless the CP asymmetry and the mass splitting are very small simultaneously, cf. Eqs. (25,26,B1) and relation (27). For the phenomenological model with two sterile neutrinos that is consistent with the oscillations of active neutrinos, we find that the late-time asymmetries always lead to a good approximation for the freeze-out values of the lepton number densities. One potential caveat is that the early-time evolution of ε(z) may strongly affect the asymmetry present within spectator fields, that in turn can have a substantial impact on the freeze-out lepton asymmetry [52]. It should also be noted, while the strong washout approximation always applies for resonant Leptogenesis with two sterile neutrinos, this does not need not to be the case when more of these are present. When the use of the late-time decay asymmetry cannot be justified, one should simply replace it with the time-dependent decay asymmetry (21) that is based on numerical solutions for the mixing and the oscillations of the sterile neutrinos. Methods for obtaining accurate quantitative results for Leptogenesis in the strong washout regime are therefore available throughout parameter space. and Γ by whereK = (K 1 + K 2 )/2 and X can in the single flavour case be identified with the parameter defined in Eq. (A1). To obtain the solutions, a matrix Ξ is defined, similarly to the one in Ref. [24]: The solution (A1) can now be rewritten as: where the matrices U and V diagonalise Ξ and Ξ * . The corresponding eigenvalues Ξ D are: where ∆ K = (K 1 − K 2 )/(2K), which is zero in the democratic case. We define γ and ω as the real and imaginary parts of the above root. The transformation matrix U is then given by: In the case of a symmetric matrix Ξ, if c is chosen such that det(U) = 1, the matrix inverse can be calculated as U −1 = U T , and there is also the relation V = U * . Rewriting Eq. (15) in terms of Ξ and Ξ * , we can easily obtain the eigenmatrices: where D = (y 2 1 − y 2 2 ) 2Γ2 + 4y 2 1 y 2 2Γ 2 cos 2 ϕ + 2i∆(y 2 1 − y 2 2 )Γ − ∆ 2 . In order to compare the magnitude of the individual eigenvalues, we define in addition and in analogy with the single-flavour model the parameter C Washout Strength in Resonant Leptogenesis with Two Sterile Neutrinos As for the equilibration of the sterile neutrinos, we note that which can be inferred by substituting the observed neutrino masses (with m 1 = 0) and mixing angles [47] into Eq. (B5). Using the relations (36) which takes for Im[̺] = 0.87 its minimum value 0.31. Therefore, the e flavour will always equilibrate sufficiently long before freeze out at z f = O(10).	7,803.6	2014-06-16T00:00:00.000	[ "Physics" ]
DecentTree: scalable Neighbour-Joining for the genomic era Abstract Motivation Neighbour-Joining is one of the most widely used distance-based phylogenetic inference methods. However, current implementations do not scale well for datasets with more than 10 000 sequences. Given the increasing pace of generating new sequence data, particularly in outbreaks of emerging diseases, and the already enormous existing databases of sequence data for which Neighbour-Joining is a useful approach, new implementations of existing methods are warranted. Results Here, we present DecentTree, which provides highly optimized and parallel implementations of Neighbour-Joining and several of its variants. DecentTree is designed as a stand-alone application and a header-only library easily integrated with other phylogenetic software (e.g. it is integral in the popular IQ-TREE software). We show that DecentTree shows similar or improved performance over existing software (BIONJ, Quicktree, FastME, and RapidNJ), especially for handling very large alignments. For example, DecentTree is up to 6-fold faster than the fastest existing Neighbour-Joining software (e.g. RapidNJ) when generating a tree of 64 000 SARS-CoV-2 genomes. Availability and implementation DecentTree is open source and freely available at https://github.com/iqtree/decenttree. All code and data used in this analysis are available on Github (https://github.com/asdcid/Comparison-of-neighbour-joining-software). Introduction Neighbour-Joining (NJ) (Saitou and Nei 1987) is perhaps the most widely used distance-based algorithm for inferring phylogenies.Its success results from its simplicity and computational efficiency: NJ takes a fraction of the time required by other popular approaches (e.g.Maximum Likelihood and Bayesian methods) and is known to perform well in terms of speed and accuracy for large alignments with low sequence divergence (Yang and Rannala 2012).Although other approaches such as Maximum Likelihood and Bayesian approaches perform better than NJ in many situations, the properties of NJ make it an attractive method for a range of applications, including generating starting trees for more computationally expensive approaches, generating rapid trees from large alignments such as SARS- CoV-2 (McBroome et al. 2021), ribosomal RNA (Quast et al. 2013), and DNA barcodes (Ratnasingham and Hebert 2007), and NJ is also widely used to provide guide trees for alignment algorithms (Thompson et al. 1994, Edgar 2004).Despite its utility, even the fastest existing implementation of NJ, RapidNJ (Simonsen et al. 2008), does not scale well for alignments containing more than 10 000 sequences, and there are no scalable implementations of NJ algorithms that are written as libraries that can be easily incorporated into other software.DecentTree seeks to address these limitations. Implementation DecentTree is an optimized and parallel Cþþ implementation of NJ and BIONJ (Gascuel 1997).DecentTree uses the Vector Class Library (VCL; https://github.com/vectorclass)and the multithreading OpenMP to parallelize the computations.Moreover, we reimplemented the RapidNJ algorithm to optimize memory access patterns and reduce CPU cache miss.The VCL version of NJ and BIONJ implementation in DecentTree is denoted with '-V' suffix (NJ-V and BIONJ-V), whereas the RapidNJ reimplementation is denoted with '-R' suffix (NJ-R and BIONJ-R).Like other standard implementations of NJ/BIONJ, DecentTree has a worst-case time complexity of O(n 3 ), where n is the number of taxa.But we expect DecentTree to be faster in most cases due to the highly optimized code. DecentTree is designed as both a stand-alone program and a header-only library that can be easily integrated into other phylogenetic software packages.DecentTree uses Cþþ template classes that allow for flexible configuration of the runtimes such as choosing between single or double precision arithmetic.As input, DecentTree accepts either a distance matrix in Phylip format or a multiple sequence alignment in common formats such as Phylip or Fasta.When users provide a sequence alignment, DecentTree computes the Jukes-Cantor pairwise distance matrix (Jukes and Cantor 1969) from the alignment.As output, DecentTree reconstructs a distance-based tree in Newick format.We checked the correctness of our implementations on simulated data with 100 taxa and alignment lengths ranging from 1000 to 100 000 sites (Supplementary Fig. S1) and assessed the code quality of DecentTree using SoftWipe (Zapletal et al. 2021).The overall score of DecentTree was 5.7, slightly higher than the average score of 5.6 for software evaluated by Zapletal et al. (2021). Benchmarking We compared the performance of the four implementations of the NJ and BIONJ algorithms in DecentTree (NJ-R, NJ-V, BIONJ-R, BIONJ-V) to implementations of the same algorithms in four other software implementations which take distance matrices as input [the BIONJ algorithm implemented in the original BIONJ software, the NJ and BIONJ algorithms implemented in FastME v2.1.6.2 (Lefort et al. 2015), and the NJ algorithms implemented in Quicktree v2.5 (Howe et al. 2002) and RapidNJ v2.3.2] and one other implementation which takes a sequence alignment as input [FastTree (Price et al. 2009, Price et al. 2010)].To do this, we analysed the simulated data we used to check DecentTree (see above), and three challenging empirical datasets: a SARS-CoV-2 alignment (COVID19 datasets) (https://github.com/bpt26/parsimony, accessed on 11 May 2021) (McBroome et al. 2021) and two high-quality ribosomal RNA v138.1 datasets from the SILVA database (Quast et al. 2013), the small subunit (SSU_NR99) and the large subunit (LSU_NR99) datasets.To examine performance across a range of dataset sizes, we randomly subsampled seven subsets of 1000, 2000, 4000, 8000, 16 000, 32 000, and 64 000 sequences from each dataset using Seqtk version 1.3-r116-dirty (https://github.com/lh3/seqtk).For the comparisons which use a distance matrix as input we first computed the distance matrix for each subset using DecentTree and then ran all software using these distance matrices as input (command see Supplementary Information).For the programmes that support multiple-threading (DecentTree, FastME, RapidNJ, and FastTree), we benchmarked them with 1 thread and 32 threads.We set the maximum wall-clock time to 12 h and the maximum memory limit to 500 Gb. These settings resulted in 378 analyses for using distance matrices as input: 21 data subsets Â 9 implementations (4 for DecentTree, 2 for FastME, and 1 for BIONJ/Quicktree/ RapidNJ) Â 2 thread counts; and 210 analyses using multiple alignments as input: for 21 data subsets Â 5 implementations (4 for DecentTree and 1 for FastTree) Â 2 thread counts.We recorded the wall-clock time and peak memory usage of each analysis on a server with 256 CPUs of 2.5 GHz and 1 Tb RAM.To compare the resulting trees, we computed their loglikelihoods using IQ-TREE (Nguyen et al., 2015) under the GTRþG model. Results Analyses of simulated data show that all algorithms in DecentTree, BIONJ, FastME, and RapidNJ performed well, but Quicktree and FastTree (using the NJ anlaysis only) performed poorly in terms of the Robinson-Foulds (RF) distance of the estimated trees to the true trees (Supplementary Fig. S9). Analyses of empirical data show that for analyses which started with distance matrices, DecentTree was the only implementation that completed every analysis within 12 h (Supplementary Fig. S2) and was the fastest implementation on the larger subsets of each of the three datasets.On the smaller subsets of each dataset ( 8000 sequences), RapidNJ tended to be the fastest implementation, although the absolute differences in execution time versus the fastest DecentTree algorithm were small (0.5-36 s when RapidNJ was faster, while for the COVID19 dataset with 8000 sequences and 32 threads available DecentTree was 26 s faster) (Fig. 1 and Supplementary Figs S3-S5).DecentTree tended to be the fastest implementation on the larger subsets of each dataset, particularly when using multiple threads, although this does come at the cost of using $1.5Â to $3Â more memory than RapidNJ (Fig. 1 and Supplementary Figs S6 and S7).On the largest subsets we analysed (64 000 sequences), DecentTree was 2.9 and 5.1 h (1.8 and 5.6 fold) faster than RapidNJ for the COVID19 dataset with 1 and 32 threads, respectively (Fig. 1 and Supplementary Table S1).For the LSU_NR99 dataset, DecentTree was 2.2 and 3.5 h (1.9 and 4.2 fold) faster than RapidNJ with 1 and 32 threads, respectively; and DecentTree was the only software able to complete the analysis of the 64 000 sequence SSU_NR99 dataset in under 12 h (RapidNJ quit without producing any output for this dataset, and we were unable to determine why).The performance of the four DecentTree algorithms (NJ-R, NJ-V, BIONJ-R, BIONJ-V) differed modestly, but NJ-R tended to be the fastest, particularly on the larger subsets, while NJ-V tended to be the most memory efficient.Likelihood analysis shows that the different implementations produce trees that can differ somewhat in their fit to the data.For example, the different implementations produce trees that differ by 30-2681 units of log-likelihood on COVID19 datasets.NJ algorithms tended to produce trees with higher likelihoods than BIONJ algorithms for COVID19 and LSU_NR99, but worse trees for SSU_NR99 (Fig. 1 and Supplementary Figs S3-S7).RF distances (Supplementary Figs S8-S14) show that the tree topologies inferred by DecentTree and other implementations are sometimes different from each other, e.g. by 3.43%-23.51%on 8000 sequences datasets.These large differences are driven by the very large proportion of very short branches in these trees (e.g. more than 70% of branches in the 64 000 sequence COVID tree represent <1 substitution.Supplementary Fig. S15).Because the alignments lack information to resolve these branches, most are resolved effectively at random, leading to large normalized RF distances. We compared DecentTree to FastTree when inferring a tree from an alignment (the results above pertain to trees inferred from distance matrices).The results (Supplementary Fig. S16) showed that when using one thread, DecentTree is generally slower than FastTree, e.g. by 1.4-1.9times on 64 000 sequences, but this is reversed with 32 threads (e.g.DecentTree is 2.7-14.1 times faster than FastTree on 64 000 sequences).	2,160.2	2022-04-10T00:00:00.000	[ "Computer Science" ]
Lorentz Invariance and QCD Equation of Motion Relations for Generalized Parton Distributions and the Dynamical Origin of Proton Orbital Angular Momentum We derive new Lorentz Invariance and Equation of Motion Relations between twist-three Generalized Parton Distributions (GPDs) and moments in the parton transverse momentum, $k_T$, of twist-two Generalized Transverse Momentum-Dependent Distributions (GTMDs), as a function of the parton longitudinal momentum fraction $x$. Although GTMDs in principle define the observables for partonic orbital motion, experiments that can unambiguously detect them appear remote at present. The relations presented here provide a solution to this impasse in that, e.g., the orbital angular momentum density is connected to directly measurable twist-three GPDs. Out of 16 possible Equation of Motion relations that can be written in the T-even sector, we focus on three helicity configurations that can be detected analyzing specific spin asymmetries: two correspond to longitudinal proton polarization and are associated with quark orbital angular momentum and spin-orbit correlations; the third, obtained for transverse proton polarization, is a generalization of the relation obeyed by the $g_2$ structure function. We also exhibit an additional relation connecting the off-forward extension of the Sivers function to an off-forward Qiu-Sterman term. I. INTRODUCTION A fundamental way of characterizing the internal structure of the proton is through sum rules that express how global properties of the proton are composed from corresponding quark and gluon quantities. For example, one may ask what portion of a proton's momentum is carried by either quarks or gluons; or one may ask how the spin of the proton is composed from the spins and orbital angular momenta of its quark and gluon constituents. Elucidating this latter question, the so-called proton spin puzzle [1], indeed counts among the prime endeavors of hadronic physics in the last decades. These questions can be cast in field-theoretic language by considering proton matrix elements of the energy momentum tensor, T q,g (q and g denote the quark and gluon sectors), where p and p describe the incoming and outgoing proton states, and A q,g (t), B q,g (t), (t = ∆ 2 = (p − p) 2 , P = (p + p)/2) are the relevant gravitomagnetic form factors parametrizing the proton matrix elements (Refs. [2,3], for reviews see Ref. [4,5]). These basic constructs of the theory can be accessed experimentally owing to the connection, through the operator product expansion (OPE), of the gravitomagnetic form factors to the Mellin moments of specific parton distributions parametrizing both the forward (p = p ) and off-forward (p = p ) quark and gluon correlation functions. One obtains the following sum rules for momentum and angular momentum, respectively, Eq.(4), the angular momentum sum rule, is also known as the Ji sum rule [3]. All of the distributions entering Eqs. (3) and (4) are observable in a wide class of experiments probing the deep inelastic structure of the proton. H q,g (x, ξ, t) and E q,g (x, ξ, t) are the Generalized Parton Distribution (GPD) functions which depend on the longitudinal momentum transfer between the initial and final proton, represented through the skewness parameter ξ, and the four-momentum transfer squared, t, x being the light cone momentum fraction carried by the parton [6,7]. In particular, H q (x, 0, 0) ≡ q(x), H g (x, 0, 0) ≡ g(x), where q(x) and g(x) are the unpolarized quark (antiquark) and gluon distributions, or the Parton Distributions Functions (PDFs). PDFs have been measured in decades of Deep Inelastic Scattering (DIS) experiments, with impressive accuracy and kinematical coverage, confirming to high precision the momentum sum rule, Eq. (3). To verify the angular momentum sum rule it is necessary to extract the GPDs from experiment, in particular, E q,g . Sufficiently accurate values for the GPDs have just fairly recently started to become available from exclusive deeply virtual scattering experiments, namely Deeply Virtual Compton Scattering (DVCS), Deeply Virtual Meson Production (DVMP) and related processes, conducted most recently at Jefferson Lab and COMPASS (see [8] for a recent review). DVCS experimental measurements are necessarily more involved than ones for inclusive scattering, since they require the simultaneous detection of all products of reaction. The extraction of observables, the GPDs, from experiment is also more complex owing to the increased number of kinematical variables they depend on. An additional hurdle is present for the analysis of angular momentum in both identifying and giving a physical interpretation to the components of the sum rule (4): while the momentum sum rule has an immediate dynamical interpretation in terms of the average longitudinal momentum carried by the different parton components, to obtain a dynamically transparent expression for the angular momentum sum rule one has to break it down into its spin and Orbital Angular Momentum (OAM) components, while simultaneously preserving the gauge invariance of the theory. The decomposition can be performed within two different approaches, by Jaffe and Manohar (JM) [1], 1 2 ∆Σ q + L JM q + ∆G + L JM g = 1 2 (5) and by Ji [3], Longitudinal OAM distributions have been identified with parton Wigner distributions weighted by the cross product of position and momentum in the transverse plane, b T ×k T [9,10]. Parton Wigner distributions can be related, through Fourier transformation, to specific Generalized Transverse Momentum-Dependent Parton Distributions (GTMDs), which are off-forward TMDs. The correlation defining OAM corresponds to the GTMD F 14 (we follow the naming scheme of Ref. [11]). In particular, the OAM distribution is described by the x-dependent k 2 T moment of F 14 . The OAM term differs in the JM and Ji approaches with regard to how the gauge invariance of the theory intervenes through the gauge link in the relevant parton correlator [12]. The difference was recently explicated in the quark sector in Refs. [13,14], where it was shown that JM OAM, L JM q , can be written as the sum of Ji's OAM, L Ji q , plus a matrix element including the gluon field. The latter was interpreted in the semi-classical picture of Ref. [14] as having the physical meaning of an integrated torque stemming from the chromodynamic force between the struck quark and the proton remnant interacting in the final state. To summarize, in both Ji's and JM's expressions, OAM is defined through an imbalance in the distribution of the number density of quarks in longitudinally polarized proton states, when the quark's displacement in the transverse plane is simultaneously orthogonal to its intrinsic transverse motion. JM's definition includes a quark re-interaction which could be, in principle, process-dependent. How can these two pictures of the proton's angular momentum coexist, and what are experimental measurements really probing? The work presented here was motivated by the question of defining a way to test these ideas through observables that would enable direct access to OAM in experimental measurements. While J q,g measurements through GPDs are in progress, GTMDs, providing in principle the density distributions for OAM, remain experimentally elusive objects, since they require exclusive measurements of particles in the two distinct hadronic planes disentangling the k T and b T (or ∆ T ) directions [15][16][17]. GTMDs can, however, be evaluated in ab initio calculations [18]. In a previous publication [19], we showed that the x-dependent k 2 T moment of F 14 entering Eq. (6) can be written in terms of a twist-three GPD, E 2T [11], as Here, we present several extensions of this relation, and describe the details of the derivation comprehensively. In particular, we show that a more general relation holds, where A F14 (x) is a term containing the gauge link dependent, or quark-gluon-quark, components of the correlation function. For a straight gauge link, A F14 (x) = 0, thus recovering the result displayed in Eq. (7). These relations are specific generalized Lorentz Invariance Relations (LIR) connecting the x-dependent k 2 T moments of GTMDs and GPDs. Just as in the forward case [20][21][22], generalized LIR are based upon the covariant decomposition of the fully unintegrated quark-quark correlation function in off-forward kinematics: the number of independent functions parametrizing the correlator is less than the total number of GTMDs and GPDs, thus inducing relations among the latter. Several LIRs have been found between forward twist-three PDFs and k T moments of TMDs. The most remarkable example of an LIR is perhaps the relation between the TMD g 1T and the twist-three PDF g T , leading to the Wandzura-Wilczek relation between the helicity distribution g 1 and g T = g 1 + g 2 [23]. In the presence of a gauge link other than the straight one (e.g. a staple link), LIRs acquire an additional term that cannot be encoded in the available GTMD and GPD structures. As we show in the present paper, this term produces a correction to Eq. (7), leading eventually to the Qiu-Sterman type term of Ref. [14]. Furthermore, by combining Eqs. (7,8) with the quark field Equations of Motion (EoM), we can ascribe the difference between the integrated quark total angular momentum, J q , and the spin, S q ≡ (1/2)∆Σ q , in Ji's description to the integral of the Wandzura-Wilczek component of the GPD combination E 2T + H + E. We find that, at the unintegrated level, a quark-gluon-quark term is also present which integrates to zero consistently with Ji's sum rule. Our relation, therefore, allows one to connect the partonic sum rule originating from the dynamical definition of OAM -through the unintegrated correlation function -and the gravitomagnetic form factors which define the energy-momentum tensor (Eq. (4)). On the other hand, having access to relations at the unintegrated level allows us to extend the treatment to the JM case, where we obtain that the quark-gluon-quark contribution does not vanish upon integration. We show it to reproduce the Qiu-Sterman type term in [14]. In principle, 32 individual EoM relations can be constructed, associated with the 8 twist-two GTMDs in the vector and axial-vector sectors, which each feature independent real and imaginary components; an additional doubling of the number of relations is given by contracting the EoMs in the transverse plane either with the transverse momentum k T or with the transverse momentum transfer ∆ T . However, we place a special focus in the present paper on just three further relations besides Eq. (7) [19] that describe spin correlations stemming from a similar operator structure as for OAM, The three additional relations presented here for the first time involve the k 2 T moments of the following GTMDs: G 11 , which was observed to provide information on the longitudinal part of the quark spin-orbit interaction, or the projection of quark OAM along the quark spin [9]; G 12 , which corresponds to a transverse proton spin configuration and generalizes the TMD g 1T leading to the original Wandzura-Wilczek relation [23,24], and, finally, the naive T-odd part of F 12 which corresponds to the off-forward generalization of the Sivers function, f ⊥ 1T [25], which we relate to a generalized Qiu-Sterman term represented by M F12 in Eq. (11). For G 11 , in particular, by using the EoM we find a relation whose integral in x is consistent with the sum rule found in [26] and revisited in [27]. However, our derivation, valid for arbitrary gauge link structure, allows for a new term representing final state interactions. Furthermore, we stress the importance of the term proportional to the quark mass which appears in this relation as being generated from quark transverse spin contributions. This paper is organized as follows. In Section II we define the general framework: the correlation functions, the gauge link structure, the parametrization of the correlation functions which ensues, and the helicity amplitudes; in Section III we give a detailed derivation of the EoM relations, including explicit quark-gluon-quark terms; in Section IV we derive the LIRs for both OAM and spin-orbit correlations. We discuss their Mellin moments to order n = 3; in Section V we discuss the relations for transverse proton spin configurations and their connection to the forward limit; finally, in Section VI we give our conclusions and outlook. II. FORMAL FRAMEWORK AND DEFINITIONS We base our treatment on the complete parametrization of the quark-quark correlation functions in the proton up to twist four given in Ref. [11]. By applying time reversal invariance, charge conjugation, parity and hermiticity one finds that, at twist two, there are three independent PDFs: f 1 , g 1 , in the chiral even sector, and the chiral odd h 1 ; eight GPDs (four chiral even and four chiral odd); eight TMDs, and sixteen GTMDs. At twist three, one has many more functions due to both the presence of additional couplings (scalar, and pseudoscalar), and to the larger number of kinematical terms in the correlation function parametrizations for the vector, axial vector and pseudoscalar couplings. Each one of the PDFs, TMDs, and GPDs corresponds to specific quark-proton helicity amplitude combinations that can be extracted from various hard inclusive, semi-inclusive and deeply virtual exclusive processes, respectively, and that represent specific polarization configurations, or spin correlations, of partons inside the proton. It is important to distinguish between different types of twist-three objects that will be dealt with in this paper. Canonical twist three effects, describing quark-gluon correlations in the nucleon, appear in the OPE as coefficients of the inverse power terms in a large characteristic scale of the process, e.g. O(M/Q), M being a nonperturbative mass scale. A different class of twist three effects, geometrical twist three, arises from the quark field components which are not dynamically independent solutions of the equations of motion, and that can be expressed, through the equations of motion, as composites of the quark and gluon fields. These are also suppressed by inverse powers of Q (the classification of parton distributions given above concerns this type of twist three objects). The order of canonical and geometrical twist does not match beyond order two: contributions with the same power in M/Q, or same dynamical twist, can be written in terms of matrix elements of operators with different canonical twist. The Wandzura-Wilczek (WW) [23] relations between matrix elements of operators of different dynamical and same canonical twist encode this mismatch, as first exemplified for the polarized distribution functions g 1 and g 2 . A complete set of relations between twist two TMDs and twist three PDFs was presented and discussed for various correlation functions in Refs. [20,22]. These relations are based upon the Lorentz invariant decomposition of the fully unintegrated correlation function with the two quark fields located at different space-time positions, and they necessarily involve parton transverse momentum and off-shellness both through the k T -moments of twist-two TMDs (where k T denotes the quark transverse momentum), and the twist-three PDFs. The different kinds of twist three functions were renamed: intrinsic for geometric, dynamic for canonical, i.e., when an extra gluon field operator is directly involved in the definition, and kinematic which are related to k T -moments of TMDs. These distinctions are useful to keep in mind as we extend both the Lorentz Invariance Relations (LIRs) and the Equation of Motion relations (EoMs) to off-forward kinematics involving GTMDs and their kinematic twist-three constructs, intrinsic twist-three GPDs, and off-forward dynamical twist-three terms. Already the construction of the aforementioned relations between TMDs and PDFs, once taken beyond a purely formal level, encounters obstacles rooted in divergences of the k T -integrations connecting TMDs to collinear objects such as the PDFs. These divergences must be separated off to ultimately contribute to the scale evolution of the collinear quantities. Our treatment similarly relates GTMDs to GPDs through k T -integrations, and thus inherits these issues in complete analogy. In the present paper, we do not present any further developments on this topic beyond what is given in the literature on the connection between ordinary TMDs and PDFs. In general, the precise connection of GTMDs to GPDs still requires further specification. The relations we derive can also be read purely at the GTMD level, before identifying k T -integrals of GTMDs with GPDs. In that form, all components of our relations can be regularized on an identical footing, before identifying their collinear limits. To the extent that our relations derive from symmetries (such as Lorentz invariance), any regularization that respects these symmetries can be expected to leave the relations we derive intact. At appropriate places in our treatment, we will indicate points at which modifications of our results must be countenanced owing to issues of regularization; an example is the standard deformation of TMD gauge links off the light cone, associated with the introduction of a Collins-Soper evolution parameter. This procedure applies likewise to a proper definition of GTMDs. We will also refrain from writing explicitly the soft factors that are required [28] to regulate divergences associated with the gauge connections contained in the bilocal operators defining TMDs and GTMDs. A. Kinematics and correlators The completely unintegrated off forward quark-quark correlation function is defined as the matrix element between proton states with momenta and helicities p, Λ and p , Λ , where the gauge link structure U connecting the quark operators at positions −z/2 and z/2 is discussed in detail in the next section, Γ is a Dirac structure, Γ = 1, γ 5 , γ µ , γ µ γ 5 , iσ µν , and the choice of four-momenta is defined with P = (p + p )/2 along the z-axis, ∆ = p − p as in Ref. [11], where the initial and final quark momenta are k − ∆/2 and k + ∆/2, respectively. Four-vectors w µ are represented in terms of light-cone components, w µ ≡ (w + , w − , w T ); ξ = −∆ + /2P + is the skewness parameter, ∆ T ≡ (∆ 1 , ∆ 2 ), k T ≡ (k 1 , k 2 ), and the four-momentum transfer squared is ∆ 2 ≡ t; we displayed the kinematics also specifically for the ξ = 0 case, which is the case on which we will focus in this study. B. Gauge link structures To ensure gauge invariance, the quark bilocal operator (12) requires a gauge link U along a path connecting the quark operator positions −z/2 and z/2. Two important choices of path are a direct straight line and a staple-shaped connection characterized by an additional vector v, cf. Fig. 1. These different choices will give rise to different genuine twist three contributions to the correlators. The appropriate choice of gauge link path depends on the physical context. In the TMD limit, the staple-shaped gauge link is most relevant, since it encodes final/initial state interactions in SIDIS/DY processes. On the other hand, GPDs are defined with a straight gauge link; as discussed in more detail below and displayed in Fig. 1, only under certain circumstances do GTMDs with a staple-shaped gauge link have a proper GPD limit, with the staple link collapsing into a straight gauge link. In general, GTMDs defined from the outset with a straight gauge link play a separate role, and both the straight and staple-shaped gauge link choices will be treated in this work. Two specific motivations for doing so are the following: • In the context of quark orbital angular momentum, as accessed via the GTMD F 14 discussed in detail further below, both the straight and the staple-shaped gauge connections have a definite, distinct physical meaning [14]. On the other hand, v in general is taken to deviate from the x − axis by a plus component v + in order to regulate rapidity divergences occurring if v is taken to point purely in the minus direction; i.e., v = (v + , v − , 0). Note that, in the two-dimensional projection displayed, the x + and the xT axes fall on top of one another; they are nevertheless of course distinct axes. The separation z is Fourier conjugate to the quark momentum k. Integrating over transverse momentum kT sets zT = 0, i.e., the quark operator positions then fall on the x − axis. Nevertheless, for v + = 0, the path then still retains its staple shape. Only in the v + = 0 limit (staple legs become horizontal in figure) does the staple path collapse onto the x − axis upon kT integration, leading to a bona fide GPD limit in which all parts of the staple link cancel, except for a residual straight link directly connecting −z/2 to z/2. One can alternatively define GTMDs with a straight gauge link from the outset; in terms of the vectors defined in the figure, this simply corresponds to the limit v = 0. A straight gauge link enters the definition of Ji quark orbital angular momentum [29], whereas a staple-shaped gauge link generates Jaffe-Manohar quark orbital angular momentum [12]. Note that F 14 is a genuine GTMD quantity, i.e., a quantity which does not have a TMD or GPD limit. • A central aspect of the following treatment are Lorentz invariance relations (LIRs). In the staple link case, these contain twist-three contributions (frequently referred to as "LIR violating terms", though their role is to maintain Lorentz invariance) which do not reduce to GTMDs. To ascertain their concrete physical content in terms of quark-gluon-quark correlations, it is useful to combine the staple-link LIR with the straight-link LIR (in which these contributions are absent) as well as the straight and staple-link equations of motion. The resulting information is not directly available considering the staple link case alone. In the most basic definition of GTMDs, a staple-shaped gauge link with a staple direction vector v on the light cone is chosen [11], such that v has only a minus component, v = (0, v − , 0, 0). On the other hand, the quark operator separation z is of the form z = (0, z − , z T ), with a two-dimensional transverse vector z T . Note that z is Fourier conjugate to the quark momentum k, and GTMDs are defined in terms of k − -integrated correlators, setting z + = 0. Thus, when one forms the GPD limit of GTMDs by integration over the transverse momentum k T , one sets z T = 0, and v and z then lie along one common axis. In that case, the staple legs collapse onto that one common axis, the parts of the staple legs extending beyond the region in between the quark operators cancel, and one is left with a straight gauge link connecting those operators, as is appropriate for GPDs. However, such a light-cone choice of the staple direction v meets with rapidity divergences, which, in the application to TMDs, are commonly regulated by taking v off the light cone into the space-like region [30]. Then, v is of the form v = (v + , v − , 0, 0), and the GPD limit ceases to be straightforward; even after integration over k T , i.e., setting z T = 0, v and z do not lie on a common axis and the staple-shaped gauge link does not collapse onto a simple straight link connecting the quark operators. The k T -integrated quantities formed in this way are not directly GPDs, but differ from GPDs by contributions which formally vanish in the v + → 0 light-cone limit. An alternative possibility of treating this issue arising with staple links is to modify the GTMD definition such that correlators are not rigidly defined with z + = 0, but instead such that the longitudinal part of z is parallel to v for any chosen v, i.e., z L = (z + , z − ) is parallel to v = (v + , v − ). In that case, integration over k T does indeed lead to collapse of the staple link into a straight gauge link, but this straight gauge link now does not lie on the light cone anymore. In effect, in this way one generates quasi-GPDs in the sense discussed by Ji [31]. In the present treatment, both GTMDs defined from the beginning with straight gauge links, as well as GTMDs defined with staple-shaped gauge links will be discussed. For the latter case, the discussion will be confined to the v + = 0 limit; v + = 0 corrections will not be worked out explicitly. However, it should be kept in mind that these corrections may be important in future applications, and places where they arise will be pointed out as appropriate below. C. Parametrization of unintegrated correlation function We consider the parametrization of the completely unintegrated off-forward correlator, W Γ ΛΛ above, in terms of Generalized Parton Correlation Functions (GPCFs) for the vector, γ µ , and axial vector, γ µ γ 5 , operators. As motivated above, we are also interested in the case of a straight gauge link; the parametrization given in [11], by contrast, is constructed for a staple-shaped gauge link, and its form was chosen such that it is not straightforwardly related to the straight-link case. In this respect, it should be noted that there is considerable freedom in constructing GPCF parametrizations. This is due to the fact that not all Lorentz structures one can write down are independent of one another; they are related by Gordon identities and other relations, as laid out in detail in [11]. After exhausting these relations, 16 GPCFs A F i remain to parametrize the staple-link vector correlator, and also 16 GPCFs A G i remain to parametrize the staple-link axial vector correlator. In the straight-link case, to be discussed in more detail below, 8 GPCFs remain in each case. The staple-link parametrizations given in [11] in neither case contain 8 GPCFs relevant for the straight-link case; some of these were instead chosen to be eliminated in favor of terms intrinsically related to a staple-link structure. The vector correlator parametrization of [11] contains only 7 GPCFs relevant for the straight-link case; one additional one therefore has to be reinstated. The axial vector correlator parametrization of [11] contains only 3 GPCFs relevant for the straight-link case, and therefore 5 have to be reinstated. Thus, one cannot simply delete the Lorentz structures containing the staple direction vector v (denoted N in [11], up to a rescaling) from the parametrizations given in [11] and already arrive at a valid straight-link parametrization. Additional terms are needed, as given below. It would be possible to construct staple-link parametrizations differing from the ones in [11], each containing a full set of 8 structures relevant for the straight-link case, and each an additional 8 structures containing the staple direction v, such that deletion of the latter 8 immediately leads to a valid straight-link parametrization. We do not pursue this here to the full extent, but only give the straight-link parametrizations. In the case of the vector correlator, this is rather simple. The construction of the staple-link parametrization in [11] can be followed verbatim even in the straight-link case, merely omitting all structures containing the staple direction vector v, except for the very last step. In that very last step, the single missing straight-link structure, namely, iσ k∆ ∆ µ , is eliminated in favor of a staple-link related structure. In the straight-link case, the staple-link related structure is not available, and therefore the aforementioned straight-link structure must be kept instead. Thus, one has the straight-link vector correlator parametrization 1 where the first 7 terms are identical to the ones given in [11], and the last one, containing the additional invariant amplitude A F 17 , is associated with the aforementioned missing Lorentz structure. The case of the axial vector correlator is more involved, and the complete construction of the straight-link parametrization is given in Appendix D. We arrive at the form which in fact has only one term in common with the staple-link parametrization given in [11], namely, the one associated with the invariant amplitude A G 1 ; we make choices differing from the ones in [11] even within the straightlink sector. All GPCFs in these straight-link parametrizations are functions of k 2 , k · P, k · ∆, ∆ 2 , P · ∆. In staple-link parametrizations, such as the ones given in [11], the GPCFs additionally depend on all scalar products involving the additional vector v characterizing the staple link. It is interesting to note that, for both the vector and axial vector operators, 8 GPCFs enter the parametrization for the straight gauge link case. This is the same as the total number of GPDs (including twist 2, twist 3 and twist 4). This is expected because the GPDs are defined with quarks separated only along the light cone. The number of GTMDs on the other hand is 16. Because the underlying structure functions, the GPCFs, are fewer in number, we expect the GTMDs to be connected to one another. These relations between the GTMDs are known as the Lorentz Invariance Relations and we discuss them in Sec. IV. D. Generalized Transverse Momentum-Dependent Parton Distributions The unintegrated correlator definining the Generalized Transverse Momentum-Dependent Parton Distributions (GTMDs) is given by, Its parametrization in terms of GTMDs, as defined in Ref. [11], reads as follows. 2 For Γ = γ + , γ + γ 5 , iσ i+ γ 5 , one has, For each correlator listed, the second equality follows once P + is taken to be much larger than all other mass scales. On the other hand, for γ i , γ i γ 5 , one has The GTMDs considered here are complex functions of the set of kinematical variables x, ξ, k 2 T , k T · ∆ T , t; in the case of a staple-shaped gauge link, they furthermore depend on the vector v characterizing the staple, Due to Hermiticity and time reversal invariance, we have that the following GTMD components are odd for ξ → −ξ, This influences which k T -moments of these GTMDs can appear in the ξ = 0 case. E. Generalized Parton Distributions The Generalized Parton Distributions (GPDs) are obtained by formally integrating Eq.(18) over the transverse parton momentum, k T , provided that the gauge link has the appropriate form, cf. the discussion in Sec. II B, For γ + , γ + γ 5 , iσ i+ γ 5 one has, whereas for γ i , γ i γ 5 , The gauge connection for GPDs is a straight link, implying that all GPDs are naive T-even. We use the GPD parametrization from Ref. [11]. As in the first parametrization introduced by Ji [3], the letter H signifies that in the forward limit these GPDs correspond to a PDF, while the ones denoted by E are completely new functions; H, E, H, E parametrize the chiral-even quark operators. In the chiral-odd sector, H T , E T , H T , E T describe the tensor quark operators, the subscript T signifying that the quarks flip helicity or are transversely polarized [32]. The matrix structures that enter the twist three vector (γ i ) and axial vector (γ i γ 5 ) cases are identical to the ones occurring at the twist two level in the chiral-odd tensor sector. Hence, the GPDs have similar names: the corresponding twist three GPD, occurring with the same matrix coefficient, is named F 2T if parametrizing the vector case γ i and F 2T if parametrizing the axial vector case γ i γ 5 , with F = H, E, H, E. F. Helicity Structure To elucidate the helicity structure, which is needed to connect to phenomenological applications and which also serves as a heuristic tool in the construction of LIR and EoM relations below, we introduce the quark-proton helicity amplitudes, [7], where at twist two the bilocal quark field operators, define (non flip) transitions between quark ±, ± helicity states. Note that, in this section only, for the purpose of discussing helicity structure, we drop the gauge link in the bilocal operators to simplify notation. The various LIRs and EoM relations that we derive in subsequent sections correspond to different helicity combinations obtained varying the initial and final proton helicity states. We obtain 8 distinct relations from the following combinations, (+, +) ± (−, −), and (+, −) ± (−, +), in the vector and axial vector sector, respectively. In what follows we derive all four spin correlations. The correlation functions in Eqs. (20,22) can be written in terms of the quark-proton helicity amplitudes as, One finds the following expressions for the proton non flip terms, where, because of the constraints in Eqs. (30), the combinations on the rhs of Eqs.(44a, 44c) and Eqs.(44b,44d) are purely real and imaginary, respectively. The distributions in both transverse coordinate and momentum space corresponding to these GTMDs were analyzed in detail in Refs. [9,10]. F 11 describes an unpolarized quark and proton state, and it reduces to the PDF f 1 in the forward, k T integrated, limit; G 14 describes the quark helicity distribution, or g 1 in the forward, k T integrated, limit. F 14 and G 11 do not have GPD or TMD limits. However, in the forward limit, their average over k T weighted by k 2 T gives [10], where Eq. (45) represents the quark OAM along the z axis in a longitudinally polarized proton, while Eq. (46) gives the quark OAM along the z axis for a longitudinally polarized quark, or a spin-orbit term. The proton spin flip terms read, and, At twist three, the bilocal operators can be written as the overlap of a dynamically independent quark field, φ (good component), and a dynamically dependent quark-gluon composite field, χ (bad component) [33], Notice that the * on the lhs symbolizes the helicity of the quark within the quark-gluon composite field, χ (on the rhs), whose helicity is always opposite so that angular momentum is conserved [34]. As a result, one can form twice as many helicity amplitudes as compared to the twist two case [33], Therefore, At the twist-three level, the following are the expressions for the proton helicity non-flip terms, As we show in subsequent sections, Eqs.(52a) and (52d) allow us to identify the twist-three GTMDs that enter the EoM relations for F 14 and G 11 respectively. Writing the GTMDs that enter the proton helicity flip case one has, and, The helicity amplitude structure is preserved when going to either the GPD or the TMD limit. It plays an important role in defining the observables for the various quantities. The GTMDs defined so far are related to GPDs by integrating them over k T and to TMDs by taking the forward limit (∆ → 0). A. Construction of Equation of Motion Relations Equation of motion relations connect different GTMD correlators of the type defined in Eq. (18), in which the quark creation and annihilation operators are located at positions −z/2 and z/2. To construct them, it is useful to consider initially a somewhat more general correlator in which the quark creation and annihilation operators are located at more freely variable positions z in and z out , respectively. Central to the construction is the observation that, taken between physical particle states, matrix elements of operators that vanish according to the classical field equations of motion vanish in the quantum theory 3 [35]. Thus, in view of the classical quark field equations of motion one has the vanishing correlation function where, specifically, Γ = iσ i+ γ 5 = γ + γ i γ 5 − γ i γ + γ 5 with a transverse vector index i = 1, 2, cf. Sec. II D. Note that the / D and ← − / D operators act on the z out and z in arguments, respectively. Furthermore, no derivatives with respect to z + in or z + out appear in the square bracket; these derivatives are accompanied in the Dirac operator by a factor γ + , implying that the terms in question vanish once multiplied by the structure Γ, which contains an additional factor γ + . Thus, introducing the equations of motion as in (56) is consistent with an a priori specification z + in = z + out = 0. Performing an integration by parts with respect to both z out and z in yields Two types of contributions are generated. The second line of (57) contains the terms in which the derivatives act on the gauge links; these terms will ultimately result in quark-gluon-quark correlators. The third line of (57) contains the standard terms in which the derivatives act on the exponential in the Fourier transformation; these terms result in quark-quark correlators. Proceeding by changing integration variables, and translating the matrix elements by −b, one obtains having taken into account the phases generated in the proton states by the translation. Thus, a δ-function which enforces momentum conservation as expected, δ 3 (p − p − ∆), is factored out; it follows that the rest of the expression by itself must already vanish. Proceeding to simplify the Dirac structures (employing, e.g., the identity γ µ γ ρ γ ν = g µρ γ ν + g νρ γ µ − g µν γ ρ − i σµνρ γ σ γ 5 ), one can finally identify from the third line of (59) the GTMD correlators defined in Eq. (18), and one thus arrives at the equation of motion relations which relate the correlation functions for different Dirac structures, γ i γ 5 , γ i , γ + γ 5 , γ + , iσ i+ γ 5 , and in which the genuine/dynamic [22] twist-three terms, copied from the second line of (59), are given by 4 with Γ = iσ i+ γ 5 . In the following, only the case of vanishing skewness, ∆ + = 0, will be considered further. Relations (60a) and (60b) are generalizations to the off-forward case of the EoM relations involving the k Tunintegrated correlator first introduced in [20,36,37]. In particular, Eq.(60b) leads to the relation between the polarized structure functions g 1 and g 2 first obtained in the forward limit using the same method in Refs. [36,37]. However, notice that, at variance with [36,37], because of the symmetrization introduced in Eqs. (56)-(59), the imaginary parts in Eq.(60b) appear only for the non forward terms (terms multiplied by ∆). As will be discussed further below, these relations represent a first step towards deriving a connection between twist-two GTMDs and twist-three GPDs using a procedure alternative to OPE that highlights the sensitivity to the quark intrinsic transverse momentum. In our case, they attain additional significance in that they provide a framework for describing partonic OAM in the proton in terms of specific distributions, thus helping to clarify possible mechanisms that generate it. A prerequisite for understanding what produces OAM in the proton is that one examines the dynamics encoded in the correlator components at the unintegrated level. B. Gauge Link Structure and Intrinsic Twist Three Term The form of the intrinsic twist-three terms given in Section III A is valid for an arbitrary choice of gauge link U. The gauge link depends parametrically on the locations of its endpoints; the derivative operators quantify those dependences. More concrete forms are obtained by considering particular gauge link paths. An important choice is the staple-shaped gauge link path, the geometry of which was already discussed in detail in Sec. II B, with the legs of the staple described by a four-vector v; this contains also the straight gauge link path in the limit v = 0. Given this concrete choice, a more explicit form of the intrinsic twist-three contributions can be derived. To establish notation, consider a staple-shaped gauge link U connecting the space-time points y and y via three straight segments, which each can be parametrized in terms of a real parameter t as As noted above, the four-vector v describes the legs of the staple-shaped path. The parametrization includes the special case v = 0, in which the staple degenerates to a straight link between y and y given by U 2 (0, 1), whereas U 1 = U 3 = 1. In the following, U i given without an argument means U i ≡ U i (0, 1). As shown in Appendix A, with this parametrization, one arrives at the explicit expression in which only field strength terms remain. In complete analogy, one also obtains for the adjoint term, where in each integral, s parametrizes the position of the color field strength insertion along the gauge link connecting the quark positions. These forms are still completely general. In the following, in particular the k T -integral of the genuine twist-three terms will be of interest, in which case, cf. (61a,61b), the transverse separation z T is set to zero and z has only a minus component, z = (0, z − , 0, 0). Specializing furthermore to the case where also v has only a minus component, v = (0, v − , 0, 0), cf. the discussion in Sec.II B, the staple legs collapse onto a common axis. In this case we define U (x, x ) to denote a straight Wilson line connecting the locations x and x , and obtain, upon identifying the endpoints y = −z/2 and y = z/2, ds U (−z/2, z/2 + sv)γ µ F +µ (z/2 + sv)U (z/2 + sv, z/2) Note that, in both expressions, the first line stems from the variation of the Wilson line which connects the ends of the staple legs, whereas the second line stems from the variation of the staple leg attached to the endpoint with respect to which the derivative is taken. The straight gauge link case is obtained by setting v = 0, i.e., only the first lines in (69) and (70) remain. This limit was already given in [19]. Particularly compact expressions are obtained if one further integrates over the longitudinal momentum fraction x, in which case z = 0 altogether, cf. (61a,61b). For z = 0, the first lines in (69) and (70) vanish, i.e., the genuine twist-three terms integrate to zero for a straight gauge link. On the other hand, in the general staple link case, the second lines remain, and give identical contributions up to a relative minus sign. Combining with the Dirac structure Γ and assembling the complete genuine twist-three expressions, one has in the completely integrated limit, Note that the ij in (71) can be absorbed into the dual field strength F +i = − ij F +j , useful for the analysis within instanton models [26], in which F = ±F . On the other hand, compact expressions also for the second Mellin moments result if one specializes to the straight link case. A weighting by a factor x can be generated by taking a derivative with respect to z − , cf. (61a,61b); in the limit z = 0, only the contributions from the derivative acting on either of the z − prefactors in the first lines of (69) and (70) remain. Thus, one arrives at for straight gauge links. Note that one can obtain, e.g., the right-hand side of (74) by evaluating the v − -derivative of (71) at v − = 0, and multiplying by a factor −i/(2P + ). In other words, we uncover a connection between straight-link quark-gluon-quark correlators such as (74) and v − -derivatives of Qiu-Sterman type terms such as (71), where the latter can be accessed using Lattice QCD TMD data [38], such as given in [39][40][41]. C. EoM Relations involving Orbital Angular Momentum Altogether, Eqs. (60a) and (60b) generate 32 individual relations between GTMDs, obtained by inserting the parametrizations (20), (22), (26), (28): each of the two relations is a two-component equation in the transverse plane; furthermore, the resulting 4 individual component relations are complex, i.e., each comprises a relation for the real (T-even) and the imaginary (T-odd) parts of GTMDs. The resulting 8 relations finally each contain 4 possible helicity combinations, as discussed in Section II F, for the proton helicity conserving, Eqs. (44), and the helicity flip, Eqs.(47) combinations, respectively. We refrain from quoting all 32 of these relations. They can be specialized to the ∆ = 0 TMD limit and to the k T -integrated GPD limit. In the TMD limit, a number of known TMD relations [42] is reproduced, including explicit expressions for the genuine twist-3 parts in terms of quark-gluon-quark correlators, encoded in the M i,S Λ Λ and M i,A Λ Λ terms. For the k T -integrated case, we focus on purely transverse momentum transfer, i.e., vanishing skewness, ξ = 0. In this case, there are potentially 8 relations: Of the original 32, 16 are ξ-odd, and of course only the T-even relations are relevant for the GPD limit. Among these 8 relations, we discuss in detail three which involve exclusively k 2 T moments of GTMDs and GPDs. These three are moreover singled out by the fact that they are also accompanied by three corresponding LIRs. In this section, we present, in particular, the EoM relations describing the quark OAM and spin-orbit contributions. These involve F 14 , which is obtained for the helicity configuration (44b) describing an unpolarized quark in a longitudinally polarized proton, and a relation for G 11 , obtained for the helicity configuration (44d), describing a longitudinally polarized quark in an unpolarized proton. These configurations are obtained by taking the helicity combinations (Λ Λ) = (++) ± (−−), in Eqs. (60a) and (60b), respectively. The relations we obtain constitute xdependent identities tying the definitions, respectively, of partonic OAM, L z , and the longitudinal contribution to the spin-orbit coupling L · S, to directly observable twist-three distributions. We present the third EoM relation, which instead involves transverse polarization, in Section V. As we show below, after taking ∆ + = 0 (without loss of generality in the angular momentum sum rule), we obtain the following EoM relations from Eqs. (60a) and (60b), respectively, where we defined Note that, in the forward limit, this reduces to the standard k 2 T -moment, The genuine twist three contributions are defined as 5 where the expressions for M i,S(A) ΛΛ given in Eqs.(61a,61b,71,72), can be interpreted as quantifying the quark-gluonquark interaction experienced by a quark of specific x, in the given helicity configuration. Eqs.(75, 76) are the equation of motion relations involving the OAM and the longitudinal part of the spin-orbit L · S distributions, defined through F 14 and G 11 , in Eqs. (45) and (46), respectively. They are particularly important among the various GTMD EoM relations that we can write because they allow us to define observables other than the GTMDs to measure the OAM distribution in the proton. All of the distributions in the EoM relations are defined according to the scheme of Ref. [11] (see Section III): H and H are twist two GPDs, in the vector and axial vector sector respectively; E 2T is a twist three GPD in the vector sector, H 2T and E 2T are axial vector twist three GPDs. Eq. (75) relates an intrinsic [22] twist three GPD, E 2T , on the lhs [19], to a twist two GPD, H, the k T -moment of the GTMD, F 14 , Eq.(77), and a genuine twist three term, M F14 . It is obtained by contracting Eq. (60a) with ∆ i /∆ 2 T , forming the (Λ Λ) = (++) − (−−) combination of helicity components, and inserting the GTMD parametrizations of the correlators, yielding Integrating over k T and identifying the resulting GPDs [11] gives i.e., one obtains Eq. (75). Recalling the discussion in Sec. II B, in the case of a staple-shaped gauge link, this requires that the legs of the staple properly collapse upon k T -integration such as to produce GPDs with straight gauge link. Eq. (76) was derived in a similar way. It relates the twist-three GPD combination, 2 H 2T (x) + E 2T (x), to the GPD H, the k T -moment of the GTMD, G 11 , which describes the longitudinal part of the parton spin-orbit distribution, and a genuine twist three term. Notice the appearance of a quark mass term proportional to the GPD 2 H T + E T in the chiral odd sector [43]. Contracting Eq. (60b) with i ij ∆ j /∆ 2 T , forming the (Λ Λ) = (++) + (−−) combination of helicity components, cf. Eq. (44d), and inserting the GTMD parametrizations of the correlators yields Integrating over k T and identifying the resulting GPDs gives i.e., Eq. (76). IV. GENERALIZED LORENTZ INVARIANCE RELATIONS The underlying Lorentz structure of the unintegrated correlator, Eqs. (16,17) allows one to find relations between the x-dependent k T -moments of GTMDs and GPDs. As stated before, this is due to the fact that, for the straight gauge link case, the total number of GPCFs is less than the number of GTMDs. Similar relations connecting the various TMDs, in the forward limit, were derived in Refs. [21,36]. These equations are a consequence of the covariant definition of the correlation function, and they are therefore referred to as Lorentz Invariance Relations (LIRs). The following LIRs, which we derive further below, involve the k T -moments of the GTMDs respectively describing the OAM and longitudinal spin-orbit terms which also enter the EoMs derived in Section III, Eqs.(75,76), On the left hand side, we have k 2 T -moments of twist two GTMDs. These GTMDs are unique in that, in the limit t = 0, they carry the physical meaning of parton longitudinal OAM distribution, F 14 , and longitudinal parton spin-orbit distribution, G 11 . On the right hand side, the integral expressions for the intrinsic twist three GPDs E 2T + H + E, and 2 H 2T + E 2T + H, allow us to access both OAM and the longitudinal spin-orbit term directly from deeply virtual exclusive measurements as these GPDs enter as coefficients of specific azimuthal angular modulations of the cross section. Note that these x-dependent relations are valid also for ∆ T = 0. Note also that, at variance with previous work [13,26,31,44], Eq (85) allows us to obtain directly information on the OAM distribution because its form is not integrated in x (it occurs at the k T -integrated level). Eq.(86) is new: it allows us to connect the longitudinal spin-orbit x-distribution, G (1) 11 , to a specific twist three GPD combination, 2 H 2T + E 2T that uniquely appears in off-forward processes. If we were to work with a staple gauge link, the number of GPCFs would increase to 16 for both the vector and the axial vector case. In this scenario, since the number of GTMDs is the same as the number of GPCFs, we do not expect there to be any LIRs connecting exclusively GTMDs (or their GPD limits). Indeed, if we do try to write these relations, we find that extra terms appear that consist of GPCFs that cannot be combined to form either GPDs or GTMDs. These extra terms, which are required in order to properly encode Lorentz invariance in the relations, have been termed LIR breaking terms [24]. For example, (85) is modified to read with, where the 4-vector v = (0, v − , 0, 0) describes the direction of the staple, which here is taken to extend along the light cone. The amplitudes A F i are the ones appearing in the parametrization given in [11], appropriate for a staple link structure, up to a rescaling stemming from the fact that the staple vector v used here and the analogous vector N used in [11] are related by a rescaling. Note that, if one were to take v off the light cone, v = (v + , v − , 0, 0), cf. the discussion in Sec. II B, additional terms would appear in (88) that formally vanish as v + → 0; examples of such terms in the case of TMD LIRs have been given in [24]. Of course, the GPCFs themselves then also depend on v + . In what follows, we work with straight gauge links, where terms such as (88) are absent; in Sec. IV C, we return to the staple link case and obtain a concrete expression for (88) in terms of quark-gluon-quark correlators, by combining the LIR with the corresponding EoM. A. Construction of Lorentz invariance relations The general structure of the unintegrated correlation function was written in terms of all the independent Lorentz structures multiplied by scalar functions, A F i , A G i in Section II C. The correlation function integrated in k − (and k T dependent) was parametrized in terms of GTMDs in Ref. [11] (Section II D). GTMDs can, therefore, be expressed through k − integrals of the scalar functions A F i , A G i . These expressions are given in Appendix C. As was shown in Section II C, the total number of independent functions in the unintegrated corrrelator is 8 for the vector and 8 for the axial vector sectors. The total number of twist two plus twist three GTMDs is 12 vector and 12 axial vector [11]. Since this number exceeds the number of A F i , A G i functions, the GTMDs will be related to one another. This type of relation that is just originating from the parametrization in terms of Lorentz covariant structures is called a LIR. In the following, we describe the procedure used to derive LIRs between the k T -moments of the twist two GTMDs listed in Section II D, Eqs.(44b, 44d), and the twist three GPDs listed in Section II E. It is based on the following integral relation for amplitudes A depending on the integration variable k via Lorentz invariants as A ≡ A(k · P, k 2 , k · ∆), where X [A; x] is a linear combination of amplitudes A in which the coefficients, aside from containing the invariants k · P , k 2 and k · ∆, may have an explicit x-dependence. This is an off-forward extension of relations used previously in the analysis of TMD LIRs [24,36,37,45]; here, the presence of the additional invariant k · ∆ must be properly accounted for. In view of this complication, it is worth laying out the elements of the derivation of (89); this is presented in Appendix B. B. Relating k 2 T moments of GTMDs to GPDs Since a generic GTMD X can be expressed in the form X = dk − X [A; x], as given in Appendix C, one can use (89) to cast the x-derivative of its k T -moment, X (1) , cf. (77), in terms of A amplitudes. In particular, To complete the LIRs, one constructs the appropriate combinations of GPDs which yield the right-hand sides. The relevant combinations, cf. Appendix C, are 6 To construct the appropriate combinations completing the LIRs, we examine the expression for the proton helicity combination associated with the GTMD appearing on the left hand side of (90), (91), and find the twist-three GPDs corresponding to that same helicity structure. The GPCF substructure of the twist-three GPDs need not, in general, completely match the GPCF combination of the x-derivative of the k 2 T moment of the GTMD. One may need to add a twist-two GPD with the appropriate GPCF substructure. In particular, F 14 describes an unpolarized quark in a longitudinally polarized proton, Eq.(44b) for Γ = γ + ; the twist-three GPD with a similar proton helicity combination is E 2T . Comparing their GPCF decompositions, we see that if we add H + E, we arrive at the LIR, Similarly, G 11 describes a longitudinally polarized quark in an unpolarized proton, Eq.(44d) with Γ = γ + γ 5 . The corresponding twist-three combination with the same proton helicity combination is 2 H 2T + E 2T , with their GPCF substructure given above. By adding the GPD H, this gives us, As already noted further above, in the case of a staple link, these Lorentz Invariance Relations acquire LIR violating terms that we introduce as, cf. (87), These relations are a central result of our paper: they give a connection valid point by point in the kinematical variables x and t = −∆ 2 T among the k T moments of GTMDs that define dynamically OAM and longitudinal spinorbit coupling, specific twist three GPDs, and LIR violating terms that can be expressed in terms of genuine twist three contributions; the latter connection will be elucidated using the example of A F14 , cf. (98), in the next section. C. Intrinsic twist three contributions Lorentz invariance relations (LIRs) derived in the presence of a staple-shaped gauge link generally include additional terms beyond those found for straight gauge links, as exemplified by (87),(88) in comparison to (85). Whereas the staple LIR by itself does not yield the concrete physical content of these terms, considering it in the context of the straight-link LIR as well as staple and straight link EoMs provides more detailed insight into their meaning. To illustrate this, it is useful to pursue the case of the LIRs (85) and (87), relevant for the description of quark orbital angular momentum in the nucleon, further. Subtracting the former LIR from the latter yields giving a concrete expression for A F14 in terms of the GTMD F 14 . Note that, here, the discussion given in Sec. II B should be kept in mind: Formulating the LIRs (as well as the EoMs below) in terms of GPDs assumes that, in the staple-link case, the legs of the staples have properly collapsed upon k T -integration such as to produce GPDs with their straight gauge link structures. This requires the staple link vector v to lie on the light cone, v = (0, v − , 0, 0). Corrections to the above relation would arise from several sources if one were to take the staple vector v off the light cone, v = (v + , v − , 0, 0). On the one hand, the cancellation between the straight and staple link GTMD precursors of the GPDs in (85) and (87) would be incomplete; there would be residual terms corresponding to the difference between the two cases (unless one opts for the alternative quasi-GPD scheme also mentioned in Sec. II B). On the other hand, as already noted in connection with eq. (88), additional amplitudes would enter the GPCF expression. Now, the difference of GTMD k T -moments in (100) can also be extracted from the EoMs: subject again to the above caveats, the GPD terms in the EoM (75) are identical for a straight link and a staple link, and subtracting an instance of (75) with a straight link from an instance with a staple link yields Thus, the additional terms in the staple LIR (87) are associated with quark-gluon-quark correlations, Therefore, we see that, comparing the genuine twist-three terms entering the staple link LIR and the staple link EoM, these encode independent information: the EoM contains M F14 alone, whereas the LIR contains the difference of M F14 and M F14 \| v=0 . As was shown in Refs. [12,14], in the forward limit, and integrated over momentum fraction x, the quantity −F (1) 14 corresponds to Jaffe-Manohar quark orbital angular momentum in the staple link case, whereas it corresponds to Ji quark orbital angular momentum in the v = 0 straight link case. Using (71), we obtain a concrete expression for the difference between the two, where it has been used that 2(∆ i /∆ 2 T )f i = (∂/∂∆ i )f i in the limit ∆ T → 0 for a vector function f which vanishes at least linearly in that limit (this is clear if one decomposes f using ∆ T , f i = ∆ i f \|\| + ij ∆ j f ⊥ ); note that the function on which the ∆ T -derivative acts in (103) satisfies this requirement since the left-hand side is regular at ∆ T = 0. In deriving (103), it has furthermore been used that, once one is considering ∆ T -derivatives, the (++) and (−−) helicity combinations contribute equally to quark orbital angular momentum. Eq. (103) can be interpreted in terms of the accumulated torque experienced by the struck quark in a deep inelastic scattering process as a result of final state interactions [14]. The genuine twist-three term A F14 (x) entering the staple link LIR thus rather directly encodes information about this torque, via repeated integration in x. Eq. (103) reproduces 7 the expression for the torque given in [14]. Analogous considerations apply to the staple link version of the other LIR derived in section IV B. For the spin-orbit sum rule, one has and in the completely integrated, forward limit, This term is analogous to Eq. (103), the only difference being in γ + → γ + γ 5 . D. Eliminating GTMD moments from LIR and EoM relations We now merge the information from the LIR, Eqs.(98, 99), and EoM relations Eqs. (75,76) such as to eliminate the GTMD moments. By eliminating F Considering again separately the vector and axial vector cases one has, These relations are valid for either a staple or a straight gauge link structure (with staple vector v on the light cone in the former case), keeping in mind that A F14 ≡ 0 and A G11 ≡ 0 in the straight-link case. Since the GPDs in these relations by definition are identical in the staple and straight link cases, subtracting a straight-link instance of (106) from a staple-link instance again yields the relation (102) between quark-gluon-quark terms (upon differentiation with respect to x), and one likewise obtains the analogous relation for A G11 . A converse way of stating this is that the terms containing M F14 and A F14 always conspire such that only a straight-link quark-gluon-quark contribution remains, even if (106) is formally written for the staple-link case; the same is true for (107). If one disregards the quark-gluon-quark contributions, and the quark mass term in Eq. (107), one obtains generalizations of the relation derived by Wandzura and Wilczek (WW) in Ref. [23], isolating the twist-two components of E 2T and (2 H 2T + E 2T ). We can then re-express Eqs. (106, 107) as, where we defined Here, E LIR 2T and E LIR 2T are the LIR violating terms containing A F14 and A G11 , respectively, E 2T and E 2T are the genuine twist three terms containing M F14 and M G11 , and E m 2T is the quark mass dependent term. E. x 0 , x and x 2 Moments We now consider the x moments for the twist three GPDs entering Eqs. (106,107). Integral relations for twist three GPDs were first obtained in Ref. [26,44] directly from the OPE while in this paper we derive them by integrating the x-dependent expressions found from the LIR and EoM. 8 It is therefore important to check how the two approaches correspond to one another. For the vector case we have, where one can see that the contributions from A F14 and M F14 cancel in the first two expressions integrating by parts; it is assumed that the integrands are sufficiently well behaved at the boundaries for all such integrations. Notice that Eq.(112a) is an extension of the Burkhardt-Cottingham sum rule to the off-forward case. Eq.(112b), taken in the forward limit, is a sum rule for Ji quark angular momentum, as can be seen by identifying the terms, Finally, Eq.(112c) is the only one containing a genuine twist three contribution. It should be noticed that this contribution was surmised to be the same for all helicity configurations in Ref. [26], while here we see that they are distinct terms. In order to gauge the size of the OAM component, one can use data on the twist two GPDs contributing to the WW definition, and simultaneously extract the twist three GPDs. Detailed comparisons between the two sets of measurements will allow us to constrain this quantity. The axial vector moments are given by, Eqs.(115a,115b,115c) are also consistent with those found in Ref. [26] and revisited in Ref. [46]. In particular, Eq.(115a) is an extension of the Burkhardt-Cottingham sum rule to the off-forward case. Similarly to the vector case, the various terms in Eq.(115b) can be rearranged so as to single out the second moment of a twist-three GPD, namely the combination 2 H 2T + E 2T + H, which in the forward limit can be interpreted through the LIR in Eq. (86) as the longitudinal contribution to the parton spin-orbit interaction (L z S z ) q , cf. (46), One then has, in the forward limit, corresponding to the sum rule where the transverse anomalous magnetic moment, κ q T , and the quark number, e q , have been defined. The quark mass-dependent term which appears in Eq.(115b), technically through the equations of motion, is due to transverse angular momentum components that are present for non-zero quark mass. Note that this term is chiral even, being given by the product of two chiral-odd quantities. We thus find the following partitioning of the terms representing total angular momentum, In the chiral limit, only the longitudinal polarization component is available to the quarks, and the correlation (J z S z ) is then quantified correctly by helicity-weighting the correlator yielding J z , cf. (114), which converts H +E into H. No contribution from E appears due to time reversal invariance. In the presence of a non-zero quark mass, this is modified by the transverse anomalous magnetic moment term, which accounts for the fact that also transverse polarization components are available to massive quarks. Note that one does not have to polarize the proton to observe these correlations between quark spin and angular momentum. V. LIR AND EOM RELATIONS INVOLVING TRANSVERSE SPIN CONFIGURATIONS The main results of this paper are given by the EoM relations in Eqs.(75,76), the LIR relations in Eqs.(85,86), and the WW relations in Section IV, which were obtained for longitudinal proton polarization at ξ = 0. Most of the LIRs [22], however, including the original ones [36,37], were originally derived for the proton helicity flip case, or for transversely polarized proton configurations. It is therefore interesting to study the extension to the off-forward case for these helicity configurations. We obtain the following EoM result for the axial-vector GTMD, where the genuine twist-three term has been defined. Our derivation proceeds in analogy to the steps used in the longitudinally polarized case, with a few important differences, as follows. Multiplying the (Λ Λ) = (+−) component of (60b) with (∆ 1 + i∆ 2 ) and the (Λ Λ) = (−+) component of (60b) with (∆ 1 −i∆ 2 ), subtracting these two component equations and contracting with i ij ∆ j /(2M ∆ 2 T ) yields, upon inserting the parametrizations in terms of GTMDs, Notice that this relation reduces in the forward limit to the one for the polarized structure functions, g 1 and g T [24], namely, taking H 2T → g T = g 1 + g 2 , H → g 1 , and H T → h 1 , as well as taking into account (130), (135) and (137), Taking moments of (138) in x, one obtains, after having converted the states from the helicity basis to a spin quantization axis in 1-direction, and having used a rotation by π in the transverse plane to combine terms associated with spin in the ±1-directions. The case of spin in the 2-direction can be treated analogously. One thus obtains the standard Qiu-Sterman form [14] in the forward limit. For ∆ T = 0, M 12 is an off-forward/generalized analogue of the Qiu-Sterman T q (x, x) term. The EoM relations presented so far in either the longitudinal or transverse proton polarization cases allow us to decompose specific twist-three GPDs into a linear combination of a twist-two GPD, a quark-gluon-quark correlation, the k 2 T moment of a twist-two GTMD and a mass term in the axial-vector case. The k 2 T moment of the GTMD can be eliminated using the LIRs. The resulting relations, when integrated over x, are analogous to the relations provided by Kiptily and Polyakov in [26]. Note, however, that not all EoM relations are of this form; in general, also other GTMD moments besides k 2 T moments appear in the EoM relations that we have not discussed in detail in this work. For instance, one finds a relation in which the GTMD G 13 contributes to the EoM weighted by (k T · ∆ T ). Moreover, that EoM relation cannot have a LIR counterpart within the twist two and twist three sectors, since G 13 is the only GTMD in those sectors which contains the invariant amplitude A G 21 , cf. Appendix C. VI. CONCLUSIONS AND OUTLOOK We presented the derivation of a set of relations connecting k 2 T -moments of GTMDs and twist-two as well as twistthree GPDs, known as Lorentz Invariance Relations (LIRs) and Equation of Motion (EoM) relations. LIRs stem from the Lorentz structure of the off-forward correlation function. By examining their gauge link structure, we find that two different types of relations exist: one obtained by considering a staple-shaped gauge link, where an explicit quark-gluon-quark contribution appears, and one for the straight gauge link, where this term is instead absent. On the other hand, the QCD equations of motion yield complementary relations containing explicit quark-gluon-quark contributions that have a different structure than the ones in the LIRs. By inserting the LIRs in the equations of motion we can eliminate the k 2 T -moments of GTMDs, and obtain relations directly between twist-two and twist-three GPDs. In the absence of genuine twist-three terms, these relations represent off-forward generalizations of the original Wandzura-Wilzcek relations connecting twist-two and twist-three PDFs. Within our general scheme of constructing LIRs, we focus particularly on ones involving the k 2 T -moments of the GTMDs F 14 and G 11 , which describe the x-density distributions of the quark OAM, L z , and longitudinal spin-orbit interaction, L z S z . Our detailed study of the k T -dependence of these OAM-related observables provides physical insight that buttresses previous suggestions in the literature, stemming from OPE-based integral relations, that partonic OAM is described by twist-three GPDs. Our results, therefore, represent a step forward in comprehending parton OAM in the proton, on two accounts. On the one hand, the obtained relations are key to accessing information from experiment on the missing piece in the proton's angular momentum budget: we obtain the x-dependent distribution of OAM through the GPDs E 2T , H and E, which can be readily measured from various azimuthal angular modulations in DVCS and related processes. The new x-dependent expressions written in terms of twist-three GPDs including the genuine quark-gluon-quark terms bring, for the first time, partonic OAM within experimental grasp. On the other hand, taking integrals in x, and using the QCD equations of motion, one recovers the sum rule relating the second Mellin-Barnes moment of a specific twist-three GPD combination, here called E 2T + H + E, to the moments of twist-two GPDs yielding the combination J q − S q . Our result is therefore not only consistent with previous findings hinting at a twist-three nature of OAM [13,26,29,44]: it goes beyond these predictions by providing a physical link, missing from earlier work, which explains how OAM is described at twist-three through its connection with the k 2 T -moment of a GTMD. The, perhaps, most distinguishing merit of these new relations lies in that they provide a handle on the dynamical underpinnings of the parton correlations through which OAM is generated. OAM is present because of the transverse motion of partons when they are displaced from the origin. This is described in QCD by a twist-three parton correlation; the correlation is generated by the Lorentz invariant structure of the proton matrix elements appearing in the QCD equations of motion. The LIRs will allow us to directly connect, on the one hand, twist-three GPD measurements of OAM and spin-orbit correlations, and on the other hand, Lattice QCD evaluations of GTMDs. The k 2 T -moment of F 14 has already been accessed in a preliminary Lattice QCD calculation [18]: GTMD k 2 T -moments can be obtained by generalizing the proton matrix elements of quark bilocal operators used to study TMDs, namely, by supplementing the transverse momentum information with transverse position information through the introduction of an additional nonzero momentum transfer. The calculation in Ref. [18] also includes the gauge connection between the quarks in the quark bilocal operators, enabling the evaluation of both the staple gauge link path used in TMD calculations, characterizing Jaffe-Manohar (JM) OAM, and the straight path yielding Ji OAM. Although this exploration was performed at the pion mass m π = 518 MeV, its results suggest a sizable difference between the two definitions. Our findings provide a perspective for accessing experimentally all terms appearing in both the JM and the Ji definitions: Ji OAM is given by the Wandzura-Wilczek component of E 2T , which is described in terms of twist-two GPDs, while JM OAM is given by the sum of these terms and the genuine/intrinsic twist-three contribution, which we identified as an integral over A F14 , technically a Lorentz invariance relation violating term. Such a term may be obtained by a careful analysis of DVCS type experiments (see e.g. an analogous term in the forward case for the axial vector components g 1 (x) and g 2 (x), [24]). Our findings extend to other GTMDs: here, we have treated specifically G 11 , encoding spin-orbit correlations, and G 12 , the off-forward extension of g 1T , leading to a direct measurement of the color force between quarks. Understanding the role of GTMDs and twist-three GPDs in quark OAM has initiated a fruitful interaction between phenomenology, theory and Lattice QCD which we intend to pursue further. In particular, the structure of the underlying QCD matrix element suggests the study of experimental processes containing two hadronic reaction planes, one associated with the hadron momentum transfer, and one associated with the transverse momentum of the hadronized ejected quark. We envisage developing the description of such two-jet processes to underpin future experimental efforts to access quark OAM directly from GTMDs. Investigations of experimental hard scattering processes/observables that measure OAM have started, and the opportunity to measure OAM using deeply virtual multiple coincidence exclusive processes will be soon within reach at the new Jlab upgrade and, even more promisingly, at an upcoming Electron Ion Collider (EIC). Having understood the mechanisms that regulate quark OAM in the proton paves the way for future studies of the gluon sector which will be crucial to understand the spin of hadrons. Consider a staple-shaped gauge link U connecting the space-time points y and y via three straight segments, which each can be parametrized in terms of a real parameter t as The four-vector v describes the legs of the staple-shaped path. The parametrization includes the special case v = 0, in which the staple degenerates to a straight link between y and y given by U 2 (0, 1), whereas U 1 = U 3 = 1. In the following, U i given without an argument means U i ≡ U i (0, 1). For purely longitudinal P and transverse ∆, this allows one to substitute and furthermore implies U σ +− U = 0; moreover, in combination with iσ µν γ 5 = − 1 2 µνρσ σ ρσ it also yields In addition, it is useful to contract the twist-three equations, which carry a transverse vector index, with the two available transverse vectors k T and ∆ T in order to extract the full information from the equations. The following relations result: For the twist-two vector GTMDs as functions of the A F amplitudes, one obtains: For the twist-three vector GTMDs as functions of the A F amplitudes: For the twist-two axial vector GTMDs as functions of the A G amplitudes:	17,844.4	2017-09-18T00:00:00.000	[ "Physics" ]
Investigation of quadratic electro-optic effects and electro-absorption process in GaN/AlGaN spherical quantum dot Quadratic electro-optic effects (QEOEs) and electro-absorption (EA) process in a GaN/AlGaN spherical quantum dot are theoretically investigated. It is found that the magnitude and resonant position of third-order nonlinear optical susceptibility depend on the nanostructure size and aluminum mole fraction. With increase of the well width and barrier potential, quadratic electro-optic effect and electro-absorption process nonlinear susceptibilities are decreased and blueshifted. The results show that the DC Kerr effect in this case is much larger than that in the bulk case. Finally, it is observed that QEOEs and EA susceptibilities decrease and broaden with the decrease of relaxation time. Third-order nonlinear optical processes in ZnS/CdSe core-shell quantum dots are investigated in [18][19][20]. It is shown that the symmetry of the confinement potential breaks due to large applied external electric fields and leads to an important blueshift of the peak positions in the nonlinear optical spectrum. The effect of quantum dot size is also studied, and it is verified that large nonlinear third-order susceptibilities can be achieved by increasing the thickness of the nanocrystal shell. The authors of [21,22] studied the quadratic electro-optic effects (QEOEs) and electro-absorption (EA) process in InGaN/GaN cylinder quantum dots and CdSe-ZnS-CdSe nanoshell structures. They have found that the position of nonlinear susceptibility peak and its amplitude may be tuned by changing the nanostructure configuration. The obtained susceptibilities in these works are around 10 −17 m 2 v 2 and 10 −15 esu, respectively. In reference [23], self-focusing effects in wurtzite InGaN/GaN quantum dots are studied. The results of this paper show that the quantum dot size has an immense effect on the nonlinear optical properties of wurtzite InGaN/GaN quantum dots. Also, with decrease of the quantum dot size, the self-focusing effect increases. In a recent paper [24], we have shown that with the control of GaN/AlGaN spherical quantum dot parameters, different behaviors are obtained. For example, with the increase of well width, third-order susceptibility decreases. The aim of this study is to investigate our proposed GaN/ AlGaN quantum dot nanostructure from quadratic electrooptic effect and electro-absorption process points of view. In this paper, we study third-order nonlinear susceptibility of GaN/AlGaN semiconductor quantum dot based on the effective mass approximation. The numerical results have shown that in the proposed structure, the third-order nonlinear susceptibilities near 2 to 5 orders of magnitudes are increased. The organization of this paper is as follows. In the 'Methods' section, the theoretical model and background are described. The 'Results and discussion' section is devoted to the numerical results and discussion. Summarization of numerical results is given in the last section. Methods In this section, theoretical model and mathematical background of the third-order nonlinear properties of a new GaN/AlGaN quantum dot nanostructure are presented. The geometry of a spherical centered defect quantum dot and potential distribution of this nanostructure are shown in Figure 1. We consider three regions consisting of a spherical well (with radius a), an inner defect shell (with thickness b − a), and an outer barrier (with radius b). The proposed spherical centered defect quantum dot can be performed by adjusting the aluminum mole fraction. In this paper, the potential in the core region is supposed to be zero, and the potential difference between two materials is constant [25]. There are various methods for investigating electronic structures of quantum dot systems [26][27][28]. The effective mass approximation is employed in this study. The timeindependent Schrödinger equation of the electron in spherical coordinate can be written as [29]. where m i * and V i (r) are effective mass and potential distribution in different regions. They are obtained as follows [30]: and where xd and xb are defect and barrier regions of aluminum molar fraction, respectively. The rest mass of electron is denoted by m e , and ΔE c ( ] is the conduction band offset [30]. The bandgap energy of Al (expressed in electron volts) [30,31]. In a spherical coordinate, Schrödinger Equation 1 can be readily solved with the separation of variables. Thus, the wave function can be written as where n is the principal quantum number, and ℓ and m are the angular momentum numbers. Y ℓm (θ, ϕ) is the spherical harmonic function and is the solution of the angular part of the Schrödinger equation. By substituting Equation 4 into Equation 1, the following differential equation is obtained for R nℓ (r): In order to calculate R nℓ (r), the two E < V 01 and E > V 01 cases must be considered. With change of variables and some mathematical rearranging, the following spherical Bessel functions in both cases are obtained: where R r ð Þ ¼ where For the whole determination of eigenenergies and constants that appeared in the wave function, R nℓ (r) should satisfy the following boundary, convergence, and normalization conditions. After determining the eigenvalues and wave functions, the third-order susceptibility for two energy levels, ground and first excited states, the model should be described [32,33]. Thus, the density matrix method [34,35] is used, and the nonlinear third-order susceptibility corresponding to optical mixing between two incident light fields with frequencies ω 1 and ω 2 appears in Equation 11: where q is electron charge, N is carrier density, α fg = 〈ψ f \| r\|ψ g 〉 indicates the dipole transition matrix element, 25 ω o = (E f − E g )/ħ is the resonance frequency between the first excited and ground states (transition frequency), and Γ is the relaxation rate. For the calculation of thirdorder susceptibility of QEOEs, we take ω 1 = 0, ω 2 = −ω in Equation 11. The third-order nonlinear optical susceptibility χ (3) (−ω, 0, 0, ω) is a complex function. The nonlinear quadratic electro-optic effect (DC-Kerr effect) and EA frequency dependence susceptibilities are related to the real and imaginary part of χ (3) (−ω, 0, 0, ω) [20][21][22]. These nonlinear susceptibilities are important characteristics for photoemission or detection applications of quantum dots. Results and discussion In this section, numerical results including the quadratic electro-optic effect and electro-absorption process nonlinear susceptibilities of the proposed spherical quantum dot are explained. In our calculations, some of the material parameters are taken as follows. The number density of carriers is N = 1 × 10 24 m −3 , electrostatic constant is ε = (−0.3x + 10.4)ε o [30,31], and typical relaxation constants are ℏΓ = 0.27556 and 2.7556 meV which correspond to 15-and 1.5-ps relaxation times, respectively. The quadratic electro-optic effect and electro-absorption process susceptibilities as functions of pump photon wavelength at 15-ps relaxation time are illustrated in Figure 2. In these figure, the solid and dashed lines show 15-and 30-Å well widths, respectively. It is clear that with the increase of the well width, both QEOEs and EA susceptibilities decreased and blueshifted. These behaviors can be related to quantum confinement effect. Because of the increase of well width, the centered defect acts as small perturbation. The third-order susceptibility of GaN/AlGaN quantum dot versus pump photon wavelength with different barrier potentials as parameter is shown in Figure 3. The third-order susceptibility is decreased and blueshifted by the increasing barrier potential. These are related to energy levels and dipole transition matrix element behaviors by dot potential. See Figures four and twelve of [24]. So, the resonance wavelength and magnitude of the third-order susceptibility can be managed by the control of well width and confining quantum dot potential. Same as Figure 2, we illustrate the quadratic electro-optic effect and electro-absorption process susceptibilities as functions of pump photon wavelength at 1.5-ps relaxation time in Figure 4. By comparing Figures 2 and 4, it is observed that the QEOEs and EA susceptibilities decrease and broaden with decreasing relaxation time. In Figure 5, we show the effect of confining quantum dot potential on third-order susceptibility. As can be seen with increasing barrier potential, the third-order susceptibility is decreased and blueshifted. Full-width at half maximum (FWHM) of third-order susceptibility in Figure 5 is approximately ten times broader than the FWHM in Figure 3. The effect of relaxation constant (ħΓ) is demonstrated for two well sizes in Figure 6. It can be seen that the peak of the third-order susceptibility is decreased by the increase of the relaxation rate. It is clear from Equation 11 that the third-order susceptibility has an inverse relationship with relaxation constant. Also, the difference between the peak of susceptibilities in a = 15 Å and a = 30 Å is decreased with the increase of relaxation rate. Conclusions In this paper, we have introduced spherical centered defect quantum dot (SCDQD) based on GaN composite nanoparticle to manage electro-optical properties. We have presented that the variation of system parameters can be tuned by the magnitude and wavelength of quadratic electro-optic effects and electro-absorption susceptibilities. For instance, the results show an increase of well width from 15 to 30 Å; the peaks of the both QEOEs and EA susceptibilities are decreased 7:218 Â 10 −12m 2 V 2 À to 1:062 Â 10 −12m 2 V 2 Þ and blueshifted (59.76 to 37.29 μm). With decreasing dot potential, the third-order susceptibility is increased 2:444 Â 10 −12 m 2 v 2 to 7:218 Â 10 −12 m 2 v 2 and red shifted (45.25 to 59.76 μm). The effect of relaxation constant (ħΓ) which is verified by the peak of the third-order susceptibility is decreased by the increasing relaxation rate. These behaviors can be related to the quantum confinement effect and inverse impact of relaxation constant.	2,220.4	2014-03-19T00:00:00.000	[ "Engineering", "Materials Science", "Physics" ]
Pickling Behavior of Duplex Stainless Steel 2205 in Hydrochloric Acid Solution ,e oxide-scale structure and pickling behavior of oxided 2205 duplex stainless steel in the electrolytes containing hydrochloric acid were investigated. ,e oxide scales mainly consist of two layers: the outer layer is dense Fe2O3, and the inner granular is FeCr2O4 spinel. During the pickling process, pittings form around the boundaries of FeCr2O4 particles or interfaces of two kinds of oxides, which results in that the electrolyte can directly react with the chromium-depleted layer along the pittings to produce an “undercut” effect so that the pickling efficiency is improved markedly. ,e pickling mechanism was discussed, and the model was established. Introduction Duplex stainless steel 2205 is one of the most common kinds of DSSs with the volume fraction of each phase above 30%.Due to the proper austenite-ferrite balance, 2205 exhibits exceptional corrosion resistance properties except for excellent strength and impact toughness [1,2] and thus has been widely used in oil and gas exploration, shipping preparation, flue gas desulfurization, desalination, and other industrial fields [3][4][5][6]. Pickling is one of the most important steps in the manufacture of 2205 and can become the limiting factor of production efficiency.Pickling of 2205 is very difficult for four reasons.Firstly, the oxide scales on 2205 are dense and adherent strongly to the underlying metal.Secondly, the removal of the chromium-depleted layer beneath the oxide scales is imperative due to its low corrosion resistance [7][8][9][10].irdly, the alloying element contained in 2205, such as molybdenum and nitrogen, can improve the stability of the oxide scales remarkably [11].Fourthly, the composition, thickness, and protectiveness of the oxide scales formed on the austenitic phase and ferritic phase are not the same due to the different chromium contents in them [12][13][14]. Researches [15] have shown that electrolytes containing hydrochloric acid can efficiently remove the chromiumdepleted layer for hot-rolled 304 due to the anodic brightening mechanism [16].But whether the hydrochloric acid can improve the pickling efficiency of 2205 is still not clear.Moreover, the researches on the pickling of 2205 mainly focus on the electrochemical pickling [17][18][19], and little work has been done on the chemical pickling.is paper mainly discussed the pickling behavior of 2205 in electrolytes containing hydrochloric acid and the evolution of the oxide scale by chemical pickling.Finally, a hydrochloric acid pickling model was built on these results. Experiment Procedure 2205 duplex phase stainless steel (with a chemical composition of 0.018 wt.% C, 1.2 wt.% Mn, 22.6 wt.% Cr, 5.3 wt.% Ni, and balanced Fe) was hot rolled into a plate, following annealing and blasting treatment (called as oxidized 2205).Specimens (30 mm × 30 mm) were machined from the plater.e phase components of the oxide scale were investigated by a Japan Rigaku D/Max-IIIB X-ray diffractometer with Cu Kα1 radiation (λ �1.5405 Å). e accelerating voltage, emission current, and scanning speed were 40 kV, 40 mA, and 0.2 °/s, respectively.e morphologies and microstructures of the specimens were observed using a UK Leica Cambridge S360 scanning electron microscope (SEM).Analytical grade chemicals and distilled water were used to prepare the electrolyte containing 110 g/L HCl.And a little oxidant was added to the electrolyte to advertise overcorrosion.e tests were carried out at 80 °C under the unstirred condition. Corrosion potential during the pickling process was measured by an electrochemical workstation (PARSTAT ® 2273, USA), and a saturated calomel electrode was used as the reference electrode.When the electrodes were introduced into the test electrolyte, the corrosion potential measurement started.Furthermore, some specimens were immersed into the same electrolyte and taken out after the following time intervals: 30 s, 60 s, and 90 s.Afterwards, the specimens were rinsed with distilled water to remove the residual electrolyte and dried to analyze the evolution process of the oxide scale in the pickling electrolyte by SEM. e specimen rinsed for 90 s was then slightly brushed to remove the residual oxides to observe the micromorphology of the matrix. Composition of Surface Oxide Scale. Figure 1 presents the X-ray diffraction pattern of the 2205 surface oxide scale.It clearly shows the typical diffraction peaks of the matrix indicating that the X-rays completely penetrated the oxide layer so that the possibility of undetected oxide phases was minimized.Moreover, the pattern reveals that the oxide scale consists of Fe 2 O 3 , FeCr 2 O 4 spinel, and SiO 2 , which is in accordance with the research of Li et al. [12]. Cross-sectional Morphology and Elements Distribution of Surface Oxide Scale. Figure 2 shows the SEM image of the cross-sectional morphology and the EDS maps showing the distribution of the main elements, which combine with oxygen to form the surface oxide scale.e thickness of the oxide scale is approximately 10 μm.And the scale can be divided into two layers: the outer layer of iron-rich oxidation and the inner layer of chromium-rich oxidation.And also some silicon oxides are mainly enriched at the interface of the chromium oxide and matrix.Combining this result with the X-ray spectra, it can be inferred that the outer layer is Fe 2 O 3 and the inner layer is FeCr 2 O 4 (a kind of spinel). Corrosion Potential of Pickling Process. e corrosion potential of 2205 pickling in the electrolyte containing hydrochloric acid shows a typical characteristic of hydrochloric acid pickling [16] (Figure 3).In the initial stage, the corrosion potential decreases sharply as the electrolyte permeates the interface of the oxide scale and the chromiumdepleted layer.e corrosion potential stays first at a low level after decreasing down and then abruptly increases up to a relatively high value after duration because of the dissolution of the chromium-depleted layer. is reflects an active-topassive transition rather than an anodic brightening [16]. In addition, as shown in Figure 3, the whole pickling process lasted 60 s.However, the pickling process was kept for 90 s to ensure the uniformity of pickling in the immersion test. Evolution Process of Oxide Scales in Pickling Electrolyte. e SEM images of the oxide scales after immersion in the pickling electrolyte for different times and the matrix after pickling are shown in Figure 4. e EDS results show that the outer layer of Fe 2 O 3 is dense and the FeCr 2 O 4 spinel stacking beneath the outer layer is granular.In the whole pickling process, the lumpy Fe 2 O 3 had little changes, but the amount of FeCr 2 O 4 decreased gradually.ere were some cavities at the boundaries of the FeCr 2 O 4 particles or the interfaces of two kinds of oxides (as indicated by the arrows).As the immersion time increased, the number and size of the cavities increased constantly.Up to 90 s, the surface oxide scale detached completely from the matrix, and the residual oxide could be easily removed by a nylon brush.e surface of the matrix after pickling was smooth without local pitting corrosion or other obvious corrosions, which can satisfy the requirements of cold rolling. Discussion Based on the XRD and EDS analysis results, it is known that the oxide scale is mainly composed of Fe 2 O 3 and FeCr 2 O 4 . e reaction in the electrolyte containing hydrochloric acid is as follows: (1) e variations of the standard Gibbs free energy ΔG θ for chemical reactions (1) and (2) at 80 °C are 19.128kJ and −63.122 kJ [20], respectively, suggesting that FeCr 2 O 4 could be dissolved prior to Fe 2 O 3 when immersed in the same reducing acid liquor.is accounts for why Fe 2 O 3 shows little variation with the increase of time during the whole pickling process, while the FeCr 2 O 4 spinel particles reduce with the increase of time. e electrolyte contains a large amount of Cl − , which can be preferentially adsorbed at the regions with higher energy, such as the boundary of the FeCr 2 O 4 spinel and the interface of the two oxides. e adsorption of Cl − promotes the dissolution of the oxides and the formation of cavities.e volume and depth of the cavities increase constantly with the increase of time till reaching the chromium-depleted layer. en, the elements in the chromium-depleted layer react with the electrolyte as follows (taking Fe and Cr for example): (3) e variations of the standard Gibbs free energy ΔG θ at 80 °C are −91.283kJ and −197.861kJ, [20] respectively, 2 Advances in Materials Science and Engineering meaning that the chromium-depleted layer will dissolve prior to the oxides.ese reactions will produce an "undercut" e ect on the oxide scale, and the reaction product H 2 can also degrade the integrality and adhesiveness of the oxide scale.erefore, the oxide can be removed easily at the end of pickling. According to the results above, a pickling mechanism model is built for the oxidized 2205 in hydrochloric acid solution, as shown in Figure 5.After hot rolling and high temperature annealing, the black-oxide scales on the surface of 2205 are integrated and compact (Figure 5(a)).e oxide scales consist of two layers: the outer layer is dense Fe 2 O 3 , and the inner is FeCr 2 O 4 .A thin chromium-depleted layer is formed between the inner oxide layer and matrix because of the formation of the oxide scales.After blasting, the outer Fe 2 O 3 is mechanically ruptured and partially falls o from According to the thermodynamic calculation, the electrolyte will preferentially react with FeCr 2 O 4 .erefore, the FeCr 2 O 4 crystals around the pit nucleus continuously dissolve, and the pits propagate along the grain boundaries to the matrix (Figure 5(d)).e electrolyte replenishes into the pits to maintain the continuous dissolution of the FeCr 2 O 4 crystals.When the pits penetrate the oxide layer, the chromium-depleted layer is exposed to the electrolyte and preferentially reacts to dissolve.At this time, dissolution of the chromium-depleted layer becomes the main reaction of pickling and causes the "undercut" effect (Figure 5(e)).In addition, the hydrogen bubbles (not marked) generated can promote the fluidity of the electrolyte and mechanically damage the oxide scales.With prolongation of the pickling, the chromium-depleted layer is continuously dissolved until the whole oxide scale breaks away from the matrix, which indicates the end of the pickling process. Conclusions In conclusion, the oxide scales formed on the 2205 hotrolled plate after annealing is mainly divided into two layers: the outer layer is the dense Fe 2 O 3 crystal, and the inner is the granular FeCr 2 O 4 spinel.e outer layer is broken after shot blasting treatment.When put in the electrolyte, the potential of the oxided 2205 decreases rapidly to the minimum value for some time and then gradually increases, showing obvious characteristics of hydrochloric acid pickling.During the pickling process, pittings are firstly formed around the boundaries of the FeCr 2 O 4 particles or interfaces of the two kinds of oxides and then the electrolyte penetrates the oxide scales along the pittings to react with the chromiumdepleted layer directly.is reaction produces an "undercut" effect so that the oxide scales are effectively removed. Figure 1 : Figure 1: XRD pattern of the 2205 matrix and the surface oxide scale. Figure 2 :Figure 3 : Figure 2: SEM images of the cross-sectional morphology (a) and EDS maps of Fe (b), Cr (c), and Si (d) for oxidized 2205. Figure 4 : Figure 4: SEM images of the oxide scales after immersion in the pickling electrolyte for 30 s (a), 60 s (b), and 90 s (c), respectively.e matrix after pickling (d) and EDS spectrums corresponding to points A (e) and B (f ) in (a), respectively.	2,676.8	2019-02-25T00:00:00.000	[ "Materials Science" ]
Long-distance remote comparison of ultrastable optical frequencies with 1e-15 instability in fractions of a second We demonstrate a fully optical, long-distance remote comparison of independent ultrastable optical frequencies reaching a short term stability that is superior to any reported remote comparison of optical frequencies. We use two ultrastable lasers, which are separated by a geographical distance of more than 50 km, and compare them via a 73 km long phase-stabilized fiber in a commercial telecommunication network. The remote characterization spans more than one optical octave and reaches a fractional frequency instability between the independent ultrastable laser systems of 3e-15 in 0.1 s. The achieved performance at 100 ms represents an improvement by one order of magnitude to any previously reported remote comparison of optical frequencies and enables future remote dissemination of the stability of 100 mHz linewidth lasers within seconds. Optical atomic clocks have surpassed state-of-the art microwave clocks in terms of accuracy and stability and allow targeting fractional inaccuracies of 1 part in 10 18 [1,2]. This will have a wide range of physical applications, as e.g. the search for variations of fundamental constants or precision tests of general theory of relativity. A crucial element in achieving this performance are ultrastable lasers which currently limit the short term stability of state-of-the-art optical clocks [2]. Direct comparisons of such clocks are difficult, since today, the complex setup of an optical clock does not allow for transportation. Phase-coherent long-range dissemination and remote characterization of ultrastable optical frequencies have therefore become an important tool in frequency metrology triggering many activities worldwide. Early works demonstrated e.g. a remote characterization of an optical frequency standard using a microwave signal transmitted via a 43 km long optical fiber [3]. The most promising method with the highest demonstrated stability directly transmits the ultrastable optical carrier via phase-stabilized fibers [2,[4][5][6] and was applied to remote optical clock comparisons over distances of 3.5 km [2,7]. Regarding the typical long-range distances between optical clocks and the costs, usage of commercial telecommunication fiber networks and transfer at 1.5 µm are recommended. Links with lengths of up to 251 km [6,[8][9][10] were demonstrated, even including simultaneous data traffic [11]. For these experiments, however, the local and remote end were located at the same position. Recently, an optical clock was compared to a microwave frequency standard via a 120 km telecommunication fiber link [12]. In this work we realize for the first time to our knowledge a remote comparison of 73 km-distant ultrastable lasers via a telecommunication fiber network. The demonstrated performance allows to characterize the stability of 100 mHz linewidth lasers within seconds. With our setup we are able to reveal the flicker floor of our ultrastable lasers of 3 × 10 −15 in only 0.1 s averaging time. This represents a real-time application of a longdistance fiber link at the performance-level of state-ofthe-art optical clocks. In combination with sophisticated fiber stabilization systems [10], the results confirm the potential of long-distance telecommunication fiber links for comparing optical clocks at the level of 10 −17 or below in a few minutes. The lasers compared here are separated by more than 50 km geographical distance and are used for the calcium and magnesium optical frequency standard located at the Physikalisch-Technische Bundesanstalt (PTB) in Braunschweig and at the Institute of Quantum Optics (IQ) at the University of Hanover, respectively. In Hanover, currently two ultrastable laser systems are in operation, L1 and L2. Both are diode lasers systems, which are stabilized to two independent ultrastable optical resonators. L2 uses a horizontal resonator (finesse F = 600, 000) mounted near the symmetry plane for reduced vibrational sensitivity. Since beat note measurements between two systems can only provide an upper bound for the combined stability, L1 and L2 are compared to an ultrastable laser (ML) located at PTB, whose stability is transferred via a 73 km long telecommunication fiber to IQ. The experimental setup is schematically depicted in fig. 1. The stability of the ultrastable master laser (ML) at PTB at 657 nm -a diode laser system stabilized to a vibrationally insensitive optical resonator reaching a demonstrated linewidth at Hz level [13,14] and a flicker floor of ≈ 2 × 10 −15 for 0.1 − 20 s [15] -is transferred to a 1542 nm fiber laser (TL) by means of a femtosecond frequency comb [16]. We inject about 5 mW of the light of TL into the 73 km long fiber link to IQ, connected via the local computer center in Hanover (RZ-H). At IQ, the light is amplified in a bidirectional erbium doped fiber amplifier (EDFA), frequency shifted with an acoustooptic modulator (AOM 2) and afterwards partially reflected back to PTB, using a Faraday rotator mirror. At PTB, the back reflected light is used for active cancellation of phase noise of the fiber by servo feedback to AOM 1. The servo bandwidth of the noise cancellation loop is limited to ≈ 700 Hz due to fiber length [8]. The single pass loss in the 73 km fiber is ≈ 23 dB. Information about the stabilization system and the link between PTB and RZ-H can be found in [10,15]. Part of the transmitted light at 1542 nm is coupled out at IQ and used for comparison to L1 and L2 at 914 nm by means of a second frequency comb. Both frequency combs are based on femtosecond erbium fiber lasers which are frequency doubled into the visible. Thus, the phase stable remote comparison via a transfer laser covers more than one optical octave. The remote characterization of the ultrastable lasers is performed in the time and frequency domains. In the time domain, all frequencies are recorded simultaneously by using a multichannel frequency counter with synchronous readout and zero deadtime. This can be operated either as a Π or an overlapping Λ estimator [17]. The frequency of the transfer beat ν 2 − m2 m1 ν 1 between the delivered cw light ν 1 at 1542 nm and the 914 nm cw light ν 2 of the local ultrastable lasers is calculated using the relation by measuring the mode numbers m 1 , m 2 of the frequency comb, the carrier-envelope offset frequency ν CEO , and the beat note frequencies ν B1 , ν B2 between the cw frequencies and the corresponding comb modes. The results of the characterization of the three independent laser systems are depicted in fig. 2. The Allan deviation (ADEV) shown is obtained by operating the counter as a Π estimator. L2 shows a drift rate of several 10 −15 /s and L1 of ≈ 10 −15 /s, which we attribute to temperature fluctuations of the ultrastable resonators. The drift rate of ML, and thus TL, was identified independently at PTB using a hydrogen maser as ≈ 10 −16 /s. With linear drift correction, for τ ≥ 0.05 s TL is identified as the most stable laser, and L2 as more stable than L1 ( fig. 2). From previous characterizations of the PTB-(RZ-H)-PTB fiber link with instabilities of σ y (τ ) = 3 × 10 −15 (τ /s) −1 [10,15], we suspect that the link may limit the laser comparison up to ≈ 1 s. To overcome this limitation, we implemented a data evaluation based on the modified Allan deviation (ModADEV). The ModADEV is obtained (case (c) in [17]) by operating the counter as an overlapping Λ esti- mator (with an internal gate time of 1 ms). While leaving the instability contributions of frequency modulation noise processes unaffected (within 30% error [17]), the contribution of white phase noise -which is typical for residual link fluctuations -falls off as τ −3/2 . Fig. 3 depicts the ADEV and ModADEV for the L2-TL comparison. Also shown is the instability of the 73 km link, estimated [18] from independent round-trip measurements PTB-IQ-PTB. While the ADEV of the link falls off as σ y (τ ) = (2.1 ± 0.5) × 10 −15 (τ /s) −1 , the ModADEV behaves as σ mod y (τ ) = (7 ± 1.8) × 10 −17 (τ /s) −3/2 . Thus, at 1 s measurement time, the link instability contribution is reduced by a factor of 30. For both ADEV and ModADEV, the L2-TL traces coincide with the estimated link instabilities for very short time scales, i.e. no significant additional measurement noise or laser noise could be detected. Also, both traces eventually reach a flicker floor of a few 10 −15 . (The combined thermal noise contributions of the resonators of L2 and ML is calculated to be 1.1 × 10 −15 .) However, in the case of the ModADEV, the performance of the remote laser comparison is already revealed after 100 ms and shows a flicker floor of 3 × 10 −15 . The remote measurement results are close to the performance of state-ofthe-art ultrastable lasers, as represented in fig. 3 by the previous results of the local comparison of ML to the ultrastable interrogation laser L(Yb + ) of the Yb + experiment at PTB [15]. For analysis in the frequency domain, we additionally implemented the operation in eq. (1) (divided by 4) by analog signal processing [16] using a direct digital synthesizer and double-balanced mixers. The spectrum of the remote transfer beat note 1 4 (ν 2 − m2 m1 ν 1 ), at 48.5 THz, between L2 and TL is depicted in fig. 4. It shows a linewidth of 1 Hz, limited by the spectrum analyzer used. By analyzing the spectrum of the remote beat signal, we optimized the laser system L2 in real-time. The beat signal also enables remote phase locking of a laser to the transferred light at the noise level of the stabilized fiber. To conclude, we demonstrated a remote characteriza-tion of ultrastable laser systems on the 10 −15 level on sub-second time scales. This displays the impressive potential of dissemination of ultrastable optical frequencies using long-distance telecommunication fiber networks for frequency metrology. Exciting applications are comparisons of optical clocks, remote high-precison spectroscopy or gravitational wave detectors. This work underlines the feasibility of expanding the distances bridged by the optical fiber links. Thus, the ultimate idea of a national [10] and European wide fiber network becomes a realistic scenario in the near future, which will strongly stimulate the field of fundamental physics and precison frequency metrology. We gratefully acknowledge long-standing advice and support by W. Ertmer. We thank T. Rosenband (NIST, Boulder) for providing us the design for our new ultrastable resonators. We thank U. Sterr and S. Kraft for providing the ultrastable calcium clock laser, and acknowledge financial support by the DLR and ESA. This joint work was made possible by the framework of the Sonderforschungsbereich 407 and the Centre for Quantum Engineering and Space Time Research (QUEST), with financial support from the Deutsche Forschungsgemeinschaft.	2,511.6	2009-08-28T00:00:00.000	[ "Physics" ]
The complete Chloroplast genome of Stachys geobombycis and comparative analysis with related Stachys species Herb genomics, at the forefront of traditional Chinese medicine research, combines genomics with traditional practices, facilitating the scientific validation of ancient remedies. This integration enhances public understanding of traditional Chinese medicine’s efficacy and broadens its scope in modern healthcare. Stachys species encompass annual or perennial herbs or small shrubs, exhibiting simple petiolate or sessile leaves. Despite their wide-ranging applications across various fields, molecular data have been lacking, hindering the precise identification and taxonomic elucidation of Stachys species. To address this gap, we assembled the complete chloroplast (CP) genome of Stachys geobombycis and conducted reannotation and comparative analysis of seven additional species within the Stachys genus. The findings demonstrate that the CP genomes of these species exhibit quadripartite structures, with lengths ranging from 14,523 to 150,599 bp. Overall, the genome structure remains relatively conserved, hosting 131 annotated genes, including 87 protein coding genes, 36 tRNA genes, and 8 rRNA genes. Additionally, 78 to 98 SSRs and long repeat sequences were detected , and notably, 6 highly variable regions were identified as potential molecular markers in the CP genome through sequence alignment. Phylogenetic analysis based on Bayesian inference and maximum likelihood methods strongly supported the phylogenetic position of the genus Stachys as a member of Stachydeae tribe. Overall, this comprehensive bioinformatics study of Stachys CP genomes lays the groundwork for phylogenetic classification, plant identification, genetic engineering, evolutionary studies, and breeding research concerning medicinal plants within the Stachys genus. genus Sideritis montana distributed in Iran found that while some Stachys plants exhibit distinctive pollen morphological features, they cannot be completely differentiated based on pollen morphology alone.Hence, relying solely on morphological and cytological analysis is insufficient for such a complex genus 10 .The development of sequencing technologies and the expansion of molecular databases have rendered them powerful tools for exploring the differentiation and interspecific relationships of Stachys species.In previous studies, molecular evidence such as ISSR 10,11 , RAPD 10 , and DNA fragments including the nrITS region 7,12,13 and cpDNA fragments 9,14 has been utilized to reconstruct the phylogeny of Stachydeae.For instance, Salmaki et al. 7 conducted nuclear (ribosomal ITS) and plastid (trnL intron, trnL-trnF spacer, rps16 intron) DNA sequence analysis of 143 species in the Stachydeae tribe, and found that both nuclear and plastid DNA data supported the monophyly of the Stachydeae tribe.Phylogenetic studies of Stachys plants based on ribosomal and plastid DNA data 15,16 demonstrated it as an incomplete clade divided into two distinct lineages.The center of diversity for the first lineage is located in the eastern Mediterranean region and has migrated over time to West Asia, Western Europe, Macaronesia, and sub-Saharan Africa.Meanwhile, the second lineage includes Hawaiian mints, Suzukia, all New World Stachys species, and some Old World species 17 .Berumen et al. used cpDNA regions for plant phylogenetic reconstruction and suggested reducing the number of members in the Stachys coccinea complex to three species, i.e., S. coccinea, Stachys lindenii, and Stachys albotomentosa.Meanwhile, their original ranges, including S. pacifica, Stachys manantlanensis, Stachys torresii, and Stachys jaimehintonii should be retained as varieties of S. coccinea 9 .In addition, using nrITS DNA region sequences, Özal et al. found the phylogenetic relationship between newly discovered Stachys species and their close relatives, and the newly discovered Stachys istanbulensis and its relatives Stachys recta and Stachys atherocalyx formed a branch 17 . Chloroplasts are semiautonomous organelles unique to higher plants and some algae, which are also present in a few protists 18 .The CP genome, known for its sequences short and relative independence from the nuclear genome, holds a crucial position as the second largest genome in the plant kingdom 19 .Within angiosperms, chloroplasts harbor a wealth of genetic information and present distinctive characteristics such as small relative molecular weight, simple structure, moderate evolutionary rate, low mutation rate, genetic stability, low cost, and ease of development of microsatellite sequences 20 .partially compensating for the limitations of mitochondrial and nuclear genomes 21 .Moreover, CP genome research considerably triggers studies on single nucleotide polymorphisms (SNPs), phylogenetics 22 , and DNA barcoding 23 , while facilitating investigations into the geographical origins of domesticated crops [24][25][26] .In recent years, CP genomes have found extensive application in classification studies at the genus and even family levels for various plants 27,28 .Chloroplast genomes have been widely applied in taxonomic studies at the genus and even family levels of various plants.For instance, based on the sequences of 79 chloroplast protein-coding genes, Zhao et al. 29 selected 175 species from 79 genera in the Lamiaceae family, proposing a new classification system with 12 subfamilies and 22 tribes within Lamiaceae.Li et al. 30 conducted the first examination of the structural patterns of Pholidota plastomes, providing novel insights into the phylogenetic relationships within Pholidota and its related genera through comprehensive genome data analysis.These studies have demonstrated that utilizing chloroplast genomes can effectively enhance phylogenetic resolution.However, only brief reports on the chloroplast genomes of Stachys species, specifically Stachys sieboldii 31 and Stachys japonica 32 have been published, leaving room for in depth analysis of the chloroplast genomes. Stachys geobombycis is a perennial herbaceous plant of the Lamiaceae family in the Stachys genus, mainly found in various provinces in southern China.The main edible part of S. geobombycis is its underground tuber, which has beneficial effects such as clearing heat and detoxifying, promoting blood circulation and removing stasis, dispelling wind and dampness, nourishing Qi and blood, as well as promoting health and beauty.It is often used as a medicinal resource and cooking ingredient. 33.Currently, studies on S. geobombycis have been rarely reported, with the focus placed on chemical composition and pharmacological properties [34][35][36] .It should be noted that some morphological characteristics of the Stachys genus exhibit minimal differentiation, and phenotypic traits show instability 1 .For example, based on phenotypic traits alone, it is difficult to distinguish S. geobombycis from Stachys sieboldii and Stachys affinis, which also rely on underground tubers for food.This poses significant challenges to the agricultural production of S. geobombycis.Besides, this characteristic hinders rapid and precise classification based solely on morphological attributes, posing challenges to species identification within the Stachys genus 37 .The rapid development of molecular biology and genomics provides valuable genetic information for systematic evolution and species identification in the study of plant chloroplast genomes 38 .However, the chloroplast genomes of the Stachys genus have been relatively underrepresented, and there is a lack of comprehensive and collaborative research on chloroplast genome datasets 31,32,39 .To this end, the complete sequence of the chloroplast genome of S. geobombycis was hereby collected, and compared with the chloroplast genomes of 7 closely related species through comparative genomics.The present study was carried out for the following purposes: (1) to compare the characteristics of the chloroplast genome of Stachys genus and detect differences among 7 species; (2) to identify repeat sequences, simple sequence repeats and genetically variable regions, and select divergence hotspots as candidate DNA markers; (3) to explore their IR expansion and contraction, and estimate genes selective pressure and codon usage; (4) to reconstruct phylogenetic relationships of Stachys species based on the cp genome alignments, and verify their phylogenetic position within Lamioideae.Overall, this study is expected to provide theoretical basis for the genetic breeding and phylogenetic research of Stachys plants. Characterization of the CP genome structure of Stachys species The CP genome of S. geobombycis was submitted to the GenBank database with the accession number OR327475 maintained by the National Center for Biotechnology Information (NCBI).The total length of the chloroplast (CP) genome for S. geobombycis is 150,567 base pairs (bp), and it has been sequenced with an average coverage depth of 1612.19x(Supplementary Fig. 1).It possessed a unique quadripartite structure comprising an LSC (large single copy), an SSC (small single copy), and a pair of IRs (inverted repeats) measuring 81,692 bp, 17,567 bp, and 25,654 bp, respectively (Fig. 1).In the CP genome of S. geobombycis, there were a total of 131 predicted functional genes.Among these, there were 110 unique genes, which could be further classified into different groups, including 8 rRNA genes, 36 tRNA genes, and 87 protein coding genes.Besides, the protein coding genes were divided into 4 groups based on their functions.The first group consisted of 45 photosynthetic genes, the second group included 63 self replication expression related genes, the third group contained 6 other genes, and the fourth group comprised 7 genes of unknown function (Supplementary Table 1).Additionally, 13 in the genome contained intron sequences.Among these, the genes clpP and ycf3 contained two introns each, while the remaining genes had only one intron each (rps16, atpF, rpoC1, petB, petD, rpl2, ndhB, ndhA, ycf1, ndhB, and rpl2). The CP genomes of the 8 studied Stachys species exhibited a characteristic circular double chain structure, varying from 149,523 to 150,599 bp in size (Table 1).The Stachys plastomes displayed the conventional quadripartite architecture, consisting of a LSC region (81,156-81,743 bp) and a SSC region (17,057-17,977 bp) separated by two IR regions (25,250-25,666 bp).The total GC content of all 8 CP genomes was similar, ranging from 38.36 to 38.53%.Variation in the number of genes among individuals was observed in some species.For instance, S. japonica and S. coccinea were reported to possess 123 and 133 genes, respectively 32 .To mitigate the influence of refer-ence genomes and annotation software, the plastid genomes of 7 Stachys species obtained from NCBI were re-annotated utilizing the PGA (Plastid Genome Annotator) program 40 and Geneious v11.0.336 41 , with S. Codon usage analysis In order to investigate codon usage patterns and nucleotide composition in the 8 Stachys plastomes, amino acid frequency, codon usage number, and the relative synonymous codon usage (RSCU) were analyzed and summarized 42 .The results showed that all 58 homologous protein-coding genes in these species consisted of 64 codons, and encoded 20 amino acids, including three stop codons (UAG (), UAA (), UGA (*)) (Supplementary Table 2).The number of encoded codons varied between 21,330 and 22,935 across the species.While the overall count of codons exhibited minimal variation, the types of codons and amino acids remained consistent.Besides, the RSCU value was used to measure the association between the observed frequency and the anticipated frequency of a particular codon.Out of the 64 codons, excluding the three stop codons and the unbiased methionine (Met) and threonine (Thr) (RSCU = 1), 31 codons displayed a preference with RSCU values exceeding 1, indicating a higher priority for these codons.Among them, the AUU codon for Leucine (Leu) had the highest frequency as indicated by an average RSCU value of 1.89.The remaining 31 analyzed codons showed relatively low bias, with RSCU values less than 1 (Fig. 2).The codons in the eight CP genomes of Stachys species exhibited a preference for A/T bases and A/T-ending codons, as evidenced by the GC and GC3s content being below 0.5.Besides, analysis of codon adaptation index values and an effective number of codon values revealed a minor tendency toward biased codon usage in the Stachys species.The frequency of optimal codons was relatively low.Furthermore, the hydrophobicity and aromaticity of the protein, as measured by Gravy and Aromo respectively, had a minimal effect on the observed bias in codon usage. Identification of repeat elements Among 3A).Meanwhile, the length of long repeats differed across the 8 sequenced CP genomes, with the majority falling within the 30-49 bp range (Fig. 3B-D). The highest number of SSRs was identified in S. chamissonis (51), followed by S. coccinea (48) (Fig. 4A), while the smallest number of SSRs, 152, was identified in S. palustris.The most frequently observed SSR type was mononucleotide repeats, ranging from 20 to 31.All species exhibited Mono-, di-, tri-, and tetra-nucleotide repeats, while penta-nucleotide repeats were only observed in S. coccinea, and hexa-nucleotide repeats were present in S. chamissonis, S. palustris, and S. sylvatica species.Furthermore, as shown in Fig. 4B, the SSR distribution was mostly located in the LSC region (51.28-68.75%),followed by the SSC (13.36-28.21%)and IR regions (15.68-20.51%). IR contraction and expansion in the Stachys CP genomes The Comparative genomic analysis The multiple genome alignment method of chloroplast genomes detected only one locally collinear block (LCB) among the 8 sequences (Supplementary Fig. 2).The types, quantities, and arrangement of all genes remained highly consistent within the genus.The chloroplast genomes of this genus were completely collinear, without any occurrence of rearrangement or recombination events, further indicating a high level of conservation.In order to elucidate the sequence differences among Stachys plants, the analysis was conducted with the S. affinis CP genome as reference in the mVista software (Fig. 6), and the results indicated no significant alterations such as large fragment inversions, duplications, or other structural changes of the 8 plastomes.The genomes exhibited a high level of collinearity, indicating a consistent evolutionary conservation at the genomic level.Besides, sequence differences were higher in the LSC and SSC regions compared to the IR region.Consistent with previous studies on angiosperm CP genomes, the coding regions exhibited a higher degree of conservation compared to the non-coding regions.Subsequently, multiple sequence alignments of the complete CP genomes were conducted, and the nucleotide diversity within a 600bp window was then calculated for all 8 CP genomes, The observed values ranged from 0 to 0.07746 (Fig. 7), and the analysis showed the highest difference level of the SSC region compared to other regions (Average pi = 0.007222681).The LSC region had a range of PI values from 0 to 0.04554, with an average of 0.005641219, while the IR region had the lowest Pi values, ranging from 0 to 0.02134, with an average of 0.001354375, indicating the IR region as the most conserved region, which was consistent with the results shown in Fig. 6.In addition, 5 hotspot regions (Pi > 0.02, average = 0.037), including trnH-GUG -psbA, trnK-UUU -rps16, trnG-UCC -trnR-UCU, trnN-GUU -trnR-ACG, and rps12-trnV-GAC, were identified.These sequences could serve as potential markers for further phylogenetic reconstruction and species identification in Stachys species. Phylogenetic analysis In order to ascertain the evolutionary position of Stachys within the Lamioideae subfamily, the complete CP genome sequences of 32 other sequenced CP genomes were aligned by multiple sequence alignments.Phylogenetic reconstruction was carried out using the ML (maximum likelihood) method and BI (Bayesian inference) analysis based on 40 complete CP genomes of Lamiaceae (Fig. 8 and Supplementary Fig. 3).Both methods produced nearly identical tree topologies with high support values of Bayesian posterior probability (PP) and maximum likelihood bootstrap support (BS) in each branch.Almost all nodes on the phylogenetic tree received strong support (PP/BS = 1/100), even though some clades represented a limited number of species.In the phylogenetic tree, Pogostemoneae was identified as the earliest diverging branch, followed by Gomphostemmatae, Colquhounieae, Synandreae, Betoniceae, Galeopseae, Stachydae, Aparlomideae, Phlomideae, Leonureae, Marrubieae, Leucadeae, and Lamieae.All Stachys samples were nested within Stachydeae, with S. geobombycis being closely related to S. japonica and S. affinis.S. byzantina from Western Asia was located in the basal clade of Stachydeae, while species S. chamissonis and S. coccinea from North America were in the basal clade of other Stachydeae species.Besides, the remaining Stachys species from East Asia were grouped together, forming a well supported branch, which was consistent with previous reports 7, 29 . Structure characteristics of the CP genome in the Stachys species The chloroplast (CP) genome of plants is a valuable resource for studying intra and inter species evolution and developing molecular markers 18 .In this particular investigation, CP genome sequencing and annotation of S. geobombycis officinale were conducted, and its features were compared with those of 7 other Stachys species. The findings demonstrate that the Stachys CP genome, like other angiosperms, consisted of a circular double stranded DNA molecule with a conserved quadripartite structure 38,43 .This structure included a large single-copy region (LSC), a small single copy region (SSC), and two inverted repeat regions (IRs).Notably, the Stachys CP genome exhibited significant conservation and similarity to previously reported plastomes within the Lamiaceae family 44,45 .In terms of size, the Stachys CP genome ranged from 14,523 to 150,599 bp, displaying a difference of 1077 bp across different genomes, which indicated the relatively stable and conserved nature of the Stachys CP genome.The major variation in genome size could be attributed to the varying length of the LSC (558 bp), suggesting that changes in LSC length primarily drive the variation in genome length.Overall, the GC content of the 8 Stachys samples ranged from 38.36 to 38.53%.Generally, higher GC content contributed to the stability and complexity of genome sequences 46 .Interestingly, the present study revealed a lower GC content in the LSC and SSC regions of the CP genome than in the IR region, which was a characteristic feature observed in angiosperms.This discrepancy was primarily attributed to the presence of four high GC content rRNA genes in the IR region 47 .Furthermore, the gene composition, protein coding genes, tRNA, and rRNA in the Stachys CP genome exhibited high similarity.This conservation of plastomes was consistent with previous reports in various angiosperms, such as Malvaceae and Araceae, where identical gene content and order were observed [48][49][50] . Similarly, a high degree of conservation in the CP genomes of Stachys species was hereby.Various molecular mechanisms, including maternal inheritance, rarity of plastid fusion, and active repair mechanisms, were found to contribute to the maintenance of CP genome conservation in Stachys 51 , resulting in the typically conservative nature of Stachys plastomes. Analysis of repetitive sequences and codon bias In terms of evolutionary rate and pattern, the synonymous codon usage bias (SCUB) in plant CP genomes differs from that of mitochondrial and nuclear genomes.In addition to being influenced by DNA sequence mutation pressure and natural selection affecting gene translation 52 , SCUB in plant chloroplast and mitochondrial genomes is also associated with other factors such as tRNA abundance, strand specific mutation bias, gene expression levels, and gene length [53][54][55][56][57] .These influencing factors have been widely used to explain variation in codon usage among species and within genomes 58 .In this study, specific codons were found to be more frequently used than other synonymous codons in the nucleotide sequences of protein coding genes in the Stachys species CP genomes, consistent with previous reports.Besides, no distinct species specific features were observed in codon usage levels among the 8 Stachys species.Relative Synonymous Codon Usage (RSCU) is often used to reflect codon bias.The GC content of the CP genomes is a result of the balance between mutation pressure and adaptation, which is one of the most common influences in the formation of codon usage bias 59 .Although synonymous mutations of the third base of a codon do not change the amino acid type, it is still considered an important feature in determining the amino acid type.Therefore, GC3 is often used as a significant indicator of codon preference 60 .Herein, the Stachys species CP genomes, all optimal synonymous codons (RSCU>1), except UUG and UCC, had A or U at the end, leading to a preference for A/T bases throughout the genome.Chloroplast transformation has made significant strides in various domains, including crop salt tolerance, drought resistance, and herbicide resistance in recent years.In conducting chloroplast gene breeding research, it is crucial to consider the stability of chloroplast genes and their genetic diversity 54 .RSCU can affect gene expression by regulating the accuracy and efficiency of translation, with stronger RSCU values associated with higher gene [61][62][63] .In chloroplast gene expression vector design, optimizing codons according to their bias can boost the expression levels of inserted genes in the CP genome.Additionally, known codon usage patterns can help predict the expression and function of unknown genes 64 .The identified codons with RSCU>1 can serve as efficient indicators for detecting the expression levels of hypothetical genes or open reading frames.They are also valuable for designing primers and introducing point mutations in agricultural breeding research.SSRs can be found in various regions of both prokaryotic and eukaryotic organisms, including both coding and noncoding regions 65 .Microsatellite sequences are favored markers in plant genetics and breeding research due to their variability, simplicity in utilization, detectability, and repeatability.They have been extensively employed in studies related to biodiversity and population genetic 66 .In this study, a total of 358 SSRs were detected in the CP genomes of 8 Stachys species, and the majority of SSRs were found in the LSC region, which might be correlated with the length of the LSC region.Additionally, they were predominantly composed of A/T bases, consistent with the AT richness observed in the entire CP genome 19 .Moreover, repeat sequences are essential for studying insertions, deletions, and replacements, and they are highly abundant in the chloroplasts of Stachys members 67 .In this study, a total of 512 long repeat sequences, encompassing 4 types, were identified.Overall, the SSRs and long repeat sequences identified in this study provided useful information for further research on molecular marker development, population genetics, evolution, breeding, species identification, and conservation studies in the Stachys species 68 . Comparative genomics and highly variable regions analysis In angiosperms, it is common for the Inverted Repeat (IR) regions in the CP genome to undergo expansion and contraction.This phenomenon often results in size variations, gene duplications or deletions, and the generation of pseudogenes 69 .Abnormal expansions of the IR region, transferring a large number of genes from the Small Single Copy (SSC) region to the IR region, have been observed in some taxa, such as Paphiopedilum 70 , Bidens 71 , and Pilea 56 .In this study, significant similarities were found in the expansion and contraction of the IR regions in Stachys, with highly consistent distribution and positioning of genotypes in these regions.The IR boundaries were relatively stable, consistent with previous reports in Lamiaceae 49 .The movement of the IR/SSC boundary in Stachys always leads to an increase in the length of the IR region.Overall, the conservation of the IR region in Stachys may contribute to its overall length and structural stability. DHighly variable regions with informative sites served as DNA barcodes, enabling the construction of phylogenetic trees and identification of closely related species, and expediting the discovery of previously unidentified organisms in nature 72 .Due to the insufficiency of classical DNA barcodes (rbcL, matK, psbA-trnH, and ITS2) for species identification and phylogenetics in Stachys, additional highly variable regions at the genus level as potential markers of Stachys should be explored for future identification studies.Based on mVISTA and nucleotide diversity analysis, 5 highly variable regions, including trnH-GUG-psbA, trnK-UUU-rps16, trnG-UCC-trnR-UCU, trnN-GUU-trnR-ACG, and rps12-trnV-GAC, were hereby identified, had been validated as reliable markers in previous studies.For example, Fan et al. 20 found that the amplified fragment of trnH-GUG-psbA could effectively distinguish plants within the Papaver genus.Yang et al. 73 used three pairs of primers amplifying variable DNA sequences located in the psbA-trnK, psbB-psbH, and trnR-trnN regions, and their analysis using Maximum Parsimony showed consistent classification and phylogenetic results, making them useful tools for plant species identification and phylogenetic research.Overall, these candidate barcode regions could provide rich molecular marker development information. Phylogenetic analysis Powerful molecular phylogenetics is the basis for establishing stable classifications and provides a solid framework for understanding diverse patterns, historical Biogeography, and trait evolution 74 .The Lamiaceae family, ranked as the sixth largest among angiosperms, serves as a significant reservoir of essential oils, timber, ornamental plants, culinary herbs, and medicinal herbs.This diversity makes it a crucial subject of study in fields such as ecology, ethnobotany, and floristics 75 .In recent studies, Stachys has been placed in the tribe Stachydeae within the subfamily Lamioideae, and within Stachydeae, 12 genera and approximately 470 species have been recognized 7 .Stachydeae is the largest and most challenging tribe in the subfamily Lamioideae in terms of classification 16 , which has also been the focus of some previous molecular phylogenetic studies 16,76,77 .These phylogenetic studies are mostly constructed based on gene fragments of Stachydeae or multiple gene fragments of a species to build phylogenetic trees.However, due to the limited number of informative sites, they fail to fully explain the phylogenetic relationships and systematic position of Stachydeae plants 78 .Mounting evidence has indicated the suitability of CP genome sequences for inferring phylogenetic relationships across various taxonomic levels 79 .Using complete CP genomes, many deep level phylogenetic questions have been resolved, such as the determination of the earliest diverging lineages of angiosperms 68,80,81 or the phylogenetic relationships among Ferula species 82 .This approach can better elucidate the complex evolutionary relationships among angiosperms.At the same time, CP genome datasets can also address shallow level phylogenetic questions. In our study, we constructed a phylogenetic tree of 40 Lamioideae plants based on sequence data.The results showed well supported nodes in the phylogenetic tree, and Stachy species were not a monophyletic group.This outcome aligned with the overall findings of chloroplast genome-based phylogenetic studies in Lamiaceae 29 .However, previous phylogenetic studies based on chloroplast genomes suggested S. sylvatica as a basal branch of the Stachydea system, while S. byzantina from West Asia was hereby found to be located at the basal branch of Stachydea.This result was consistent with the study by Xue et Sample collection and DNA extraction Fresh leaves of S. geobombycis were collected from Guangning County, Guangdong Province, China ( 21 1.Total DNA was extracted from dried fresh leaves of the samples using a plant DNA extraction kit manufactured by Tiangen Biotech Co., Ltd, and the integrity of the extracted DNA was assessed through 1% agarose gel electrophoresis.Subsequently, the samples were sent to BGI Genomics for further analysis, where the purity and concentration of the total DNA were determined using the NanoDrop 2000 spectrophotometer by Thermo Scientific, USA. Library construction and De novo Genome sequencing MGISEQ-2000 sequencing platform was used to construct a library with an insertion fragment of 500 bp, and paired end sequencing was performed to obtain 150 bp sequences at both ends of each read.Following sequencing, the filtering software SOAPnuke v2.0 33 83 (https:// github.com/ The Beijing Genomics Institute (BGI)-flexlab/ SOAPnuke), developed by BGI, was employed for filtering with specific parameters: (1) Adapter trimming: Reads exhibit a 25% or higher match to an adapter sequence were entirely discarded; (2) Low quality filtering: Reads with bases having a quality value below 20 that account for 30% or more of the total read were eliminated; (3) N removal: Reads containing 1% or more N bases with respect to the entire read were removed; (4) Acquisition of clean reads.The resulting data were stored in FASTQ format for subsequent assembly and annotation Assembly and annotation of the CP genome The assembly of the CP genome was conducted using NOVOPlasty v2.7.2 software; the size of k-mers was 39 84 .The gene annotation for the CP genome of S. geobombycis, and the downloaded complete CP genomes from NCBI were performed using the default parameters of the Plastid Genome Annotator (PGA) program 40 .Additional manual refinements were carried out using Geneious v11.0.3 41 .Upon the completion of the annotation process, the data were submitted to the NCBI database (https:// www.ncbi.nlm.nih.gov/ genba nk/), and the online tool OGDRAW-DRAW Organelle Genome Maps (https:// chlor obox.mpimp golm.mpg.de/ OGDraw.html) was employed visualization of the chloroplast structure. Analysis of contraction and expansion of IR boundaries in CP genomes The synonymous codon usage in the CP genomes of the three mentioned plants was hereby compared using CodonW v1.genes located in repetitive regions, and pseudogenes were excluded.Meanwhile, the Shuffle-LAGAN mode of the online tool mVISTA (https:// genome.lbl.gov/ vista/ mvista/ submit.shtml) 85 was utilized to perform a comparative analysis of the CP genomes of the 8 plants.MISA software 86 was used for SSR analysis of the CP genomes, involving parameters including mononucleotide SSRs (repeat unit of 10), dinucleotide SSRs (repeat unit of 6), trinucleotide SSRs (repeat unit of 5), and tetra, penta, and hexanucleotide SSRs (repeat unit of 4).Furthermore, REPuter (https:// bibis erv.cebit ec.unibi elefe ld.de/ sessi onTim eout.jsf) 87 , an online software, was employed for long repetitive sequence analysis of the CP genomes of the 8 plants, and parameters including Hamming Distance of 3 and a minimum repeat unit of 30 base pairs. Codon preference and repetitive sequence analysis of the CP genome In this study, Geneious v11.0.3 software 41 was used to determine the lengths of the IRa/IRb, LSC, and SSC regions, as well as the boundary genes, in the CP genomes of the Stachys species.To visualize and compare the IR boundaries in the CP genomes of 8 Stachys species, Adobe Illustrator software was employed for creating comparison maps.For the detection of intra species variations, the mVISTA software was utilized to compare the CP genomes within the Stachys species.Meanwhile, Mauve software 110 was applied for the analysis of the homology and collinearity of the CP genome sequences.For calculating nucleotide diversity values, DnaSP v6.0 88 specifically for the CP genomes of Stachys species was used.Finally, the MAFFT v7.487 software 89 was chosen to perform sequence alignments for all CP genome sequences. Phylogenetic analysis To investigate the placement of Stachys within the Lamioideae subfamily and explore relationships between different Stachys species, multiple alignments were conducted using complete CP genome sequences from 40 samples representing the Lamioideae subfamily.This comprehensive dataset included representatives from 18 different genera.Besides, the analysis involved the use of several outgroups, including Lancea hirsuta, Lancea tibetica, Rehmannia solanifolia, Triaenophora shennongjiaensis, Wightia speciosissima, and Mazus pumilus.The Phylogenetic trees of complete chloroplast genomes were constructed using the maximum likelihood (ML) method and the Bayesian inference (BI) method.ML analysis was conducted by IQ-TREE (version 2.1.3)in Phylosuite 90,91 software with a GTR + F + I substitution model and 1000 bootstrap replicates.The Bayesian inference (BI) tree was implemented in MrBayes in Phylosuite 91 and ran for two million generations in total. Based on the Markov chain Monte Carlo (MCMC) algorithm, the best fitting GTR + F + I substitution model was determined with sampling after every 1000 generations, and the running was stopped once the value of the average standard deviation of split frequencies was less than 0.01 Finally, less than 25% of the aging samples were discarded and a consistent tree was constructed based on the remaining samples. Figure 1 . Figure 1.The circular map of S. geobombycis CP genome.The second circle displays the gradient GC content distribution of the genome, where the outermost circle represents the zero level.Gene names and their codon usage bias are in the outer layer of the map.The GC content specific to each gene is visually represented by shaded areas relative to their proportions.The inner genes are shown with arrows indicating their clockwise transcription directions, while the outer genes are indicated with arrows in an anticlockwise direction. genotypes of the IR-LSC and IR-SSC boundaries were almost identical, and the lengths of the IRs in the 8 Stachys cp genomes were relatively conserved (25,250-25,655 bp) (Fig. 5), involving no obvious amplification or contraction events.At the LSC-IRB boundary, a fragment of rps19 gene was detected, with no difference in the 8 CP genomes.At the IRB/SSC boundary, the pseudogene ycf1 was situated 5 bp from the left side of LSC-IRB, with the ndhF gene extending across the LSC region for 28-29 bp, overlapping with the ycf1 pseudogene.At the SSC/IRA boundary, the ycf1 gene was consistently found in all 8 cp genomes and spanned 1092-1093 bp across the IRA boundary.The trnH gene was located 1417-1418 bp from the SSC/IRA boundary within the IRA region.The rpl2 and trnH genes were positioned between the IRa-LSC boundary, with rpl2 located on the left side of the boundary for approximately 93-94 bp, and trnH on the right side, spanning 0-1 bp. Figure 2 .Figure 3 . Figure 2. Codon content for the 20 amino acids and stop codons of CDS of the Stachys species CP genome. Figure 4 . Figure 4. Analysis of the number and type of SSRs in 8 Stachys species plastomes.SSRs with different types are visually represented using different colors.A depicts the analysis of the number and type of SSRs in 8 Stachys species plastomes; B shows the distribution of different types of SSRs in the chloroplast genome. Figure 5 . Figure 5.Comparison of chloroplast genome structure in 8 Stachys species. Figure 6 .Figure 7 . Figure 6.Visualization of alignment of 8 chloroplast genomes using S. affinis as a reference.The vertical scale depicts a range of 50-100% and represents the percentage of similarity. Figure 8 . Figure 8. Phylogenetic relationships among 46 plant species based on CP genome.The Phylogeny is inferred by employing the BI method and concatenating the sequences of protein-coding genes from all the species.The support for each branch is indicated by green circles, while posterior probability values represent the level of support. https://doi.org/10.1038/s41598-024-59132-1 We hereby declare that Stachys is not a plant species covered by the IUCN Policy Statement on Research Involving Species at Risk of Extinction and the Convention on the Trade in Endangered Species of Wild Fauna and Flora.The botanical collection work involved in this research has obtained the necessary permits and approvals from relevant local institutions, and strict compliance with applicable laws and guidelines has been ensured.Moreover, we have minimized the impact on the environment and ecosystems during the collection process and made every effort to maintain the survival and reproductive capacity of the Stachys plants.https://doi.org/10.1038/s41598-024-59132-1 the 8 analyzed CP genomes, a total of 512 long repeats were detected, comprising 135 forward repeats, 198 tandem repeats, 17 reverse repeats, and 162 palindromic repeats.The analysis revealed a varying number of repeated sequences in the 11Stachys CP genomes, ranging from 58 in Stachys chamissonis to 75 in Stachys byzantine.Tandem repeats were the most common type among these repeats, accounting for 33.3-45.43% of the repeats and varying from 22 (Stachys affinis, S. Table 1 . Comparison of CP genome features of eight Stachys species. FeatureS.geobombycis S. affinis S. byzantina S. chamissonis S. coccinea S. japonica S. sylvatica S. palustris Vol.:(0123456789) Scientific Reports \| (2024) 14:8523 \| https://doi.org/10.1038/s41598-024-59132-1 37., who utilized ITS + ETS + 5S-NTS for phylogenetic research on Stachys species37.Furthermore, S. chamissonis and S. coccinea from North America clustered together with other Stenogyne, Phyllostegia, and Haplostachys plants, while the remaining Stachys plants from East Asia formed a clade.This suggested that geographical isolation might have a greater impact on the interspecific relationships within Stachys.Overall, the research results offer important implications for the assessment of genetic diversity and systematic phylogenetic studies of Stachys in the future.However, this study still failed to fully elucidate the relationships between genera.Additionally, the phylogenetic research was solely based on the chloroplast genome.In this case, the nuclear genome of plants should be further analyzed to comprehensively understand the phylogeny of Stachydeae and even Lamiaceae species, and future studies should also have more genera included.Nevertheless, the phylogenetic research still provides valuable resources for the classification, systematic phylogeny, and evolutionary history of Stachys.The present research was primarily conducted to assemble the S. geobombycis genome and to reannotate the entire CP genome of Stachys species.Efforts were made to investigate the characteristics of the cp genome and explore the phylogenetic relationships among Stachys plants.The comparative analysis revealed conserved genome size, gene structure, and organization across the Stachys species. Futhermore, long repetitive sequences, SSRs, and regions with high variability were identified in the Stachys species.Overall, findings serve as a foundation for analyzing genetic diversity, developing mo-lecular markers, and addressing classification and identification challenges within this genus.Additionally, by examining the genetic relationships within Stachys species, this research offers comprehensive insights into phylogenetic connections, sheds light on its evolutionary history, and facilitates further research in this field. • 38 ′ 49.6 ′′ N, 39 • 41 ′ 49.3 ′′ E).This plant is currently preserved at the Guangdong Crop Germplasm Resources Nursery (http:// gdsee dbank.cn/ catal og/ guild/).The germplasm resource number is 20224412265.Please contact<EMAIL_ADDRESS>for free access.The corresponding GenBank accession number in NCBI is OR327475.Related CP genome sequences were retrieved from the NCBI database.Detailed information about the experimental materials is presented in Table	8,260	2024-04-12T00:00:00.000	[ "Biology", "Environmental Science" ]
Poleward eddy-induced warm water transport across a shelf break off Totten Ice Shelf, East Antarctica Ice mass loss in the Wilkes Land sector of East Antarctica and the Amundsen and Bellingshausen Sea sectors of West Antarctica has contributed to a rise in sea levels over several decades. The massive continental ice behind the Totten Ice Shelf, equivalent to a few meters of sea-level rise, is grounded well below sea level and therefore, potentially vulnerable to oceanic heat. Here, we present analyses of comprehensive hydrographic observations at the continental slope and shelf break regions off Totten Ice Shelf. We provide robust evidence that the relatively warm Circumpolar Deep Water that originates at intermediate depths in the Antarctic Circumpolar Current is transported efficiently towards the shelf break by multiple cyclonic eddies. We propose that these semi-permanent cyclonic circulations play a critical role in transporting the available ocean heat towards Totten Ice Shelf, and melting it from underneath, thus eventually influencing the global climate. Circumpolar Deep Water is efficiently transported towards the shelf break off Totten Ice Shelf, East Antarctica by cyclonic eddies, suggest comprehensive hydrographic observations. A recent study has demonstrated that the mass loss of the Antarctic Ice Sheet was not only in the Amundsen and the Bellingshausen Sea sectors of West Antarctica but also in the Wilkes Land sector of East Antarctica 1 . A study noted that the Wilkes Land sector of East Antarctica has been a significant contributor to sea-level rise over the last 40 years. In East Antarctica, the largest ice discharge occurs through the Totten Glacier from the Aurora Subglacial Basin in Wilkes Land, grounded below sea level 2 (Fig. 1a). This basin is characterized as a region potentially vulnerable to ocean thermal forcing in East Antarctica because of its marine-based ice condition 3,4 . The expansive ice in the Aurora Subglacial Basin behind the Totten Glacier would be equivalent to a >3.5 m rise in sea level 5,6 . Compared to other East Antarctic ice shelves, the highest area-averaged basal melt rate is estimated for the Totten Ice Shelf (hereafter TIS, 10.5 ± 0.7 m yr −1 ) 7 . In the Southern Ocean, the strongest source of oceanic thermal forcing is the warm Circumpolar Deep Water (CDW), originating from intermediate depths in the offshore Antarctic Circumpolar Current (ACC) region 8 . Increased mass loss from the West Antarctic Ice Sheet is principally driven by an increase in basal melting of the Amundsen and the Bellingshausen Sea ice shelves by the strengthening of warm CDW inflows 9,10 . CDW inflow across the shelf break is a crucial process controlling the mass balance for ice shelves/sheets and therefore, it eventually influences the Antarctic as well as global climate. While the continental shelf in East Antarctica is typically occupied by cold waters 11 , recent hydrographic observations revealed warm modified CDW (mCDW) inflows at ice fronts of the TIS (~−0.4°C) 12 and Shirase Glacier Tongue (>0°C) 13 , consistent with satellite-derived high basal melt rates for these regions 7 . These are considered as firm observational evidence of the warm ocean-cryosphere interaction beneath the floating ice shelf (and ice tongue), which is atypical in East Antarctica. The limited historical hydrographic observations around the shelf break off TIS [14][15][16][17] suggested that bathymetric features control the across-slope transport of offshore-origin mCDW and depression with a bottom depth deeper than 500 m (Fig. 1b, c) is required for "bathymetric-controlled" mCDW transport. Further, integrated data from Ice Argo floats, hydrographic profiling by seals, and numerical modeling showed the seasonality in the pathway and water properties of mCDW inflows into the depression 17 . However, we do not thoroughly understand mCDW transport from offshore ACC to the continental slope of the focal region, which is strongly associated with the available oceanic heat flux for continental ice melting. Although poleward transport of offshore-origin warm CDW plays a critical role in regulating the available ocean heat for melting continental ice, specific knowledge of ocean heat transport from offshore toward the TIS cavity is lacking. Most recently, Mizobata et al. 18 have provided important suggestions for poleward CDW transport toward TIS along the Sabrina Coast. Using satellite altimeter data, they constructed a spatiotemporally high-resolution dataset for dynamic ocean topography (DOT) over the Southern Ocean, north of shelf break (details in the "Methods" section). The DOT data unveiled a complete picture of an eddy train consisting of four quasi-stational cyclonic eddies with a spatial scale of~100-200 km (Fig. 2); this was partially observed by a previous study 19 . Hydrographic profiles and mooring data revealed that the barotropic current structure in the Vincennes Eddy off Knox and Budd Coasts transports offshore warm water poleward and cold Antarctic Bottom Water equatorward at the eddy's eastern and western limbs, respectively. However, poleward CDW transport induced by cyclonic eddies is only verified for the Vincennes Eddy west of TIS 18 transport by the Poinsett and West-Sabrina Eddies off the Sabrina Coast (Fig. 2), which may be responsible for the ocean heat transport that would be associated with the melting of TIS from below. In other words, estimating the stability of the large marine-based ice sheet behind TIS requires the determination of eddy contributions to poleward ocean heat transport based on in-situ observations. To explore the availability of ocean heat transport for TIS melting, we carried out closely spaced hydrographic observations at the continental slope and the shelf break regions (see the "Methods" section; Fig. 1). Combined with the satellite-derived ocean circulation data (see the "Methods" section; Fig. 2), we demonstrated that quasi-stational cyclonic eddies found off the Sabrina Coast play a critical role in poleward CDW transport, which is one of the presumable upstream determinants of the ocean heat flux into the TIS sub-ice cavity. Results CDW distribution at continental slope and shelf break regions. To describe water mass distribution, we defined hydrographic sections (Fig. 1b, c; described in the "Methods" section). Subsurface water with warm temperature (>1°C), high salinity (>34.7), and low dissolved oxygen (DO) content (<210 μmol kg −1 ), termed as mCDW, is observed at depths~300-500 dbar along the upper continental slope off TIS (113-121°E, zonal Section 1, Fig. 3). The mCDW spatial distribution is worthy of special mention considering remarkably warm cores (1.2-1.4°C) at 350-400 dbar that are spatially scattered rather than continuous, wherein the mCDW is notably thick in the vertical extent (Stations [hereafter, Sta.] 502, 507, 512, and 522, Fig. 3a). The mCDW cores observed in February 2019 were at much higher temperatures than the warmer intrusion of mCDW previously observed from March to June on the upper slope (warmer than 0°C, but mostly cooler than 1°C) 17 . A positional correspondence between eddies and warm cores is defined by the that particularly warm waters are consistently observed around the eastern sides of the cyclonic Poinsett (Sta. 507) and West-Sabrina (Sta. 522) Eddies (Figs. 2b, 3a). We further note that warm cores are also located on the western side of the Poinsett (Sta. 502) and West-Sabrina (Sta. 512) Eddies (Figs. 2b, 3a). It is worth highlighting that similar correspondence is determined for the East-Sabrina Eddy, demonstrated by the expendable conductivity-temperature-depth (XCTD) observations across the southern part of this eddy (Supplementary Fig. 1). The in-situ observational pieces of evidence for Poinsett, West-Sabrina, and East-Sabrina Eddies (this study), as well as Vincennes Eddy 18 , firmly represent a robust correspondence relationship between the locations of eddies and warm cores. Furthermore, semi-permanent features of the Poinsett and West-Sabrina Eddies and corresponding warm water transport at the eddy's eastern/western edges are confirmed based on the hydrographic observations from icebreaker Shirase in December 2019 ( Supplementary Fig. 2) and February 2020 ( Supplementary Fig. 3). Poleward CDW transport induced by eddies. Subsequently, we focus on the possibility of the poleward transport of offshore warm waters induced by an eddy. Figure 4 shows meridional water mass distributions at the western limb of Poinsett Eddy (Section 2) and eastern limb of West-Sabrina Eddy (Section 3), respectively, corresponding to northward and southward flowing regions (Fig. 2). Although these sections do not belong to the same eddy, the distinct difference in meridional extensions of offshore warm water is well-demonstrated by isotherms of 1.0-1.6°C (white contours in Fig. 4a, d) at the western and eastern limbs of the cyclonic eddies. The warmer water is present much closer to the upper continental slope at the eastern limb (~1°C, Fig. 4d) than at the western limb (0.6-0.8°C, Fig. 4a), suggesting that a southward current of the cyclonic eddy is associated with a warmer mCDW. To verify the role of eddies in CDW transport, we now consider the absolute velocity (see the "Methods" section; Fig. 3d) estimated from a combination of the CTD-derived geostrophic velocity and the DOT-derived surface velocity (Fig. 2b). The almost barotropic structures, comprising alternate northward and southward currents, are evident along the continental slope, representing an eddy-like feature. As suggested from the spatial pattern of warm cores (Fig. 3a), both southward and northward currents bring the warm cores along the eastern and western sides of eddies, respectively (Fig. 3d), providing strong evidence to demonstrate the meridional mCDW transport by eddies. Below observed around the western half of West-Sabrina Eddy are located at the southernmost (Section 4, Fig. 5a, b) and intermediate (Section 5, Fig. 5c, d) parts of the meridional sections (locations of warm cores shown by arrows in Fig. 5a, c), although warmer (with lower DO) waters of CDW-origin are generally found in the offshore region. This demonstrates an entirely different aspect of the meridional warm core distribution from that observed at the eastern limbs ( Fig. 5e-h). Inflows of warm mCDW have been previously observed from the shelf break to depression regions with depths of 400 m or more, in the summer of 1996 (BROKE) 14 , late winter of 2001 (SIPEX) 15 , autumn of 2015 (NBP2015 Cruise) 16 , and throughout the year (Ice Argo Float) 17 . The results from historical observations as well as those from this study indicate that mCDW inflows onto the shelf break are not intermittent but continuously occur along with seasonal variability. Hydrographic profiles and mooring data have confirmed that the velocity structure of Vincennes Eddy (Fig. 2a) is almost barotropic 18 . In addition, our estimate for the absolute velocity field across Poinsett to West-Sabrina Eddies showcases the nearly barotropic structure of the meridional velocity component (Fig. 3d). Accordingly, it is reasonable to consider that the DOT-derived surface velocities (vectors in Fig. 2) partially represent the ocean circulation in the intermediate depths of the mCDW layer. The apparent difference in the meridional distribution of warm mCDW at the eastern and western limbs of the cyclonic eddies (Figs. [3][4][5] indicates the efficient poleward mCDW transport by Poinsett and West-Sabrina Eddies. Based on the above observational findings, we determine a unique plausible circulation pattern of offshore warm mCDW, comprising inflows at the eastern limb of the cyclonic eddies and return flows at the western limbs of a part of mCDW once transported onto the shelf break. In other words, the findings strongly suggest that not all ocean heat transported across the shelf break is available for the basal melting of TIS on the Sabrina Coast. Discussion Combined with the satellite-derived DOT dataset covering sea ice areas 18 , the new closely spaced hydrographic profiles have demonstrated that multiple cyclonic eddies off the Sabrina Coast contribute to efficient poleward transport of offshore-origin warm mCDW. It should be noted that the mesoscale eddies (typically with a spatial scale of the order of few tens of kilometers), including a form of anticyclonic warm-core eddies, also contribute to the poleward heat transport across/from the ACC region [20][21][22] , but the DOT dataset's spatiotemporal resolution (monthly mean with 0.2°× 0.2°grid spacing) is insufficient for capturing such mesoscale processes. The bathymetric depression just south of the shelf break has a depth of 500 m and a width of~190 km (Fig. 1), enabling southward access of mCDW onto the continental shelf (i.e., the bathymetric-controlled mCDW intrusions) 16,17 . Note that the bathymetric features of the depression based on the International Bathymetric Chart of the Southern Ocean (IBCSO) data is oversimplified than realistic demonstrated by the multibeam bathymetric data 23 . Furthermore, results from Ice Argo Float, seal-mounted CTD, and numerical modeling posited seasonal variations in inflow paths and water properties of mCDW 17 . In autumn, warmer mCDW flows into the depression from the east following the westward Antarctic Slope Current (ASC), whereas in other seasons, the opposite occurs following an eastward undercurrent carrying relatively cold mCDW 17 . Further, the mCDW inflow paths from east or west of the depression can be consistently linked to the mCDW circulation from the ACC to the upper slope region via eddy-induced transport by the quasi-stational Poinsett and West-Sabrina Eddies (Figs. 2-5). Thus, the warm mCDW circulation from offshore ACC (source) to the shelf break appears akin to an intricate "highway junction" with some fueling stations provided by multiple eddies off TIS on the Sabrina Coast of Wilkes Land, East Antarctica. Here, we estimate eddy-induced heat transport by Poinsett and West-Sabrina Eddies (details mentioned in the "Methods" section). Poleward heat transports of 11.7 TW by Poinsett Eddy and 2.6 TW by West-Sabrina Eddy are estimated; these are greater than the required heat transport of~0.7 TW for basal melting beneath TIS (63.2 Gt yr −1 ) 7 . These estimates suggest that semi-permanent offshore cyclonic eddies transport sufficient oceanic heat for the TIS basal mass loss toward the shelf break. However, caution should be exercised in interpreting this estimation. The near-seafloor mCDW temperature is 0.5-1°C at the shelf break (Sta. 517-518, Fig. 5e; Sta. 523-524, Fig. 5g), whereas it is −0.5 to −0.4°C at the TIS ice front 12 . This notable decrease in water temperature strongly suggests the involvement of water-mass alteration processes, such as mixing with overlying cold Winter Water or heat loss by deep convection associated with sea-ice production in Dalton Polynya, during on-shelf circulation across the shelf break to the TIS ice front. Further, certain amount of the mCDW, once transported across the shelf break, might return offshore (Fig. 5a-d) without being transported beneath the TIS. This suggests that eddyinduced poleward ocean heat delivery across the shelf break is redistributed by the on-shelf water-mass alternation and circulation, determining the amount of the subsequent ocean heat transport beneath TIS. Observational and numerical model studies have pointed out that "eddies" play crucial roles in the cross-slope exchange associated with heat transport toward ice shelves [24][25][26] . For example, in the Eastern Weddell Sea, an eddy-driven overturning is enhanced, especially under weak alongshore winds, resulting in warm water transport across the shelf break toward the Fimbul Ice Shelf 24 . However, the spatial scale of such eddy overturning is much smaller than that of cyclonic eddies off the Sabrina Coast (Fig. 2). Interestingly, our closely spaced profiles also show a discontinuous distribution of warm water near the seafloor at the eastern sides of Poinsett and West-Sabrina Eddies (Section 6 and the southern part of Section 3, Fig. 5e-h), suggesting an eddy-driven overturning as in the Eastern Weddell Sea. This further suggests that larger cyclonic eddies carry mCDW from the offshore ACC to the upper slope region, and a part of these might subsequently cross the shelf break with the assistance of smaller eddies. The direct contribution of cyclonic eddies (including smaller eddy overturning) to the crossslope transport of warm water is still unclear, specifically the extent of merging of mCDW with the westward ASC or eastward undercurrent after it arrives at the upper continental slope. A highresolution mooring array across the upper slope to the shelf break regions is desired to quantify the contributions to the mCDW transport from various-scale eddies, slope jets (ASC and undercurrent), and bathymetric features (and their interactions). Even at almost the same locations (Sta. 507 and 522 at the eastern limbs of Poinsett and West-Sabrina Eddies, respectively), we observe warmer mCDW (>1-1.2°C) than those that were previously found on the western (at 115-116°E and~65°S) and eastern (at 119-120°E and~65.5°S) upper slopes north of the depression 17 ( Fig. 2), and eddy overturning 24 . Moreover, the presence of warm cores is highly localized along the continental slope (e.g., the horizontal scale of the warm core centered at Sta. 507 is quite small, Fig. 3a). In other words, this might lead to misinterpretation of a "visible" difference in the mCDW properties and distribution if the spatial resolution of the profiles is coarse. Thus, the locality of warm cores (Fig. 3) indicates the necessity of closely spaced hydrographic observations to capture the poleward eddyinduced transport of warm mCDW that would control the rate of basal melting beneath the ice shelves along the Sabrina Coast. At present, a reliable DOT dataset for the south of the shelf break has not been constructed due to the low availability of radar signals from the sea surface. Therefore, the satellite-derived DOT cannot provide any information regarding on-shelf mCDW circulation toward the TIS cavity. Comprehensive hydrographic and bathymetric surveys across the continental shelf are indispensable for elucidating the mCDW circulation regulating the ocean heat flux for TIS basal melting. Further, identifying mCDW modification processes over the Sabrina Coast's continental shelf, such as sea-ice production accompanied by seasonal deepening of the overlying cold water layer that would modify the mCDW thickness in the Dalton Polynya (~120°E), east of TIS, is also important to quantitatively evaluate the variability of available oceanic heat flux beneath the TIS cavity. Further accumulation of observational findings from the offshore ACC to the TIS ice front regions are useful in validating and improving the numerical modeling framework. This also contributes to improving the accuracy of future prediction of changes in the East Antarctic ice sheet mass balance, which contains a large mass of continental ice 3,4 , and the associated sea level rise. The Amundsen and the Bellingshausen Sea ice shelves in West Antarctica are most susceptible to the ocean heat flux that results from CDW inflows 8,28,29 as ACC transporting warm CDW approaches the shelf break. Instead, this study posits that offshore cyclonic eddies play a critical role in efficient poleward transport of warm water toward TIS on the Sabrina Coast as in the case of the cyclonic Weddell Gyre that transports warm water toward the Shirase Glacier Tongue with a high basal melt rate 7,13 . This strongly suggests that the presence of such cyclonic circulations is a characteristic of the offshore setting for creating warm on-shelf ocean conditions and associated high basal melt rates in the East Antarctic continental margin, typically occupied by cold waters 11 , and eventually influence the Antarctic and global climate. This study proposes that poleward eddy-induced transport of offshore-origin warm water plays a critical role in regulating the available ocean heat for melting East Antarctic continental ice from underneath. Methods Comprehensive in-situ hydrographic observations off Totten Ice Shelf on the Sabrina Coast. In December 2018 and January and February 2019, we carried out comprehensive hydrographic observations off TIS on the Sabrina Coast in East Antarctica (Fig. 1), under the research project led by the Fisheries Agency of Japan-"multidisciplinary ecosystem survey in the eastern Indian sector of the Antarctic (CCAMLR Division 58.4.1) with a focus on Antarctic Krill" 30 . We successfully obtained high-accuracy vertical profiles of temperature, salinity, and DO from the surface to~10 m above the bottom using the SBE 9plus conductivity-temperature-depth profiler (CTD, Sea-Bird Electronics, Washington State, USA) and RINKO DO profiler (JFE Advantech, Hyogo, Japan). The observation data satisfy the requirement of Global Ocean Shipbased Hydrographic Investigations Program (GO-SHIP) 31 standard accuracies (0.002°C for temperature, 0.002 for salinity, and 1% for DO). From the shelf break to the upper slope regions, we obtained high spatial resolution profiles (Sta. 501-529 at 29 stations), compared with historical ship-based observations 16 (white inverted triangles in Fig. 1c). The closely spaced hydrographic profiles allow for detailed investigation of images of distribution and poleward transport of offshore-origin CDW that would contribute to the melting of TIS. In addition, two meridional sections along 112°15′E and 120°1 9′E are the revisit observation sections of Baseline Research on Oceanography, Krill and the Environment (BROKE) in 1996 14 . Satellite-derived dynamic ocean topography (DOT) 18 . We utilize a monthly mean dataset of 0.2°× 0.2°DOT over the Southern Ocean (except for regions shallower than~1000 m; Fig. 2), derived from satellite altimeter data 18 (from January 2011 to April 2019; for this study, the data period is now extended to February 2020, Supplementary Figs. 2 and 3), to explore the characteristics of ocean circulation, especially eddy fields, associated with poleward CDW transport. This DOT dataset is spatially seamless even in sea-ice covered regions and hence, we can estimate DOT-derived geostrophic velocity at the ocean surface irrespective of the presence of sea ice (vectors in Fig. 2). Using the DOT dataset, Mizobata et al. 18 revealed an "eddy train" consisting of four cyclonic eddies-Vincennes Eddy, Poinsett Eddy, West-Sabrina Eddy, and East-Sabrina Eddy (Fig. 2a). Spatial scales and mean positions of the four cyclonic eddies are 240 × 220 km centered at 109°E and 63.8°S for Vincennes Eddy, 195 × 170 km centered at 115°E, and 64.5°S for Poinsett Eddy, 140 × 220 km centered at 119.5°E and 64.5°S for West-Sabrina Eddy, and 140 × 100 km centered at 124°E and 64.8°S for East-Sabrina Eddy. Section 1 zonally covers two eddies-Poinsett Eddy and West-Sabrina Eddy, and Sections 2 and 3 correspond to the western (i.e., northward flowing region) and eastern (i.e., southward flowing region) limbs of the Poinsett and West-Sabrina Eddies, respectively (yellow circles, Fig. 2). Mizobata et al. 18 conducted hydrographic observations to capture the detailed structure and characteristics of the Vincennes Eddy, while our new observations exhaustively cover the two cyclonic eddies that would relate to the heat transport available for the melting of TIS subice cavity (yellow circles, Fig. 2). Calculation of absolute velocity from hydrographic and satellite-derived DOT data. The DOT is the sum of the baroclinic (steric) and the barotropic (including tidal component, i.e., bottom pressure term) components. As a result, DOT-derived velocity includes both baroclinic and barotropic components, but it can only be obtained at the sea surface. Using CTD measurements, the vertical profile of geostrophic velocity is obtained; however, only the baroclinic component is reflected in the profile. Note that a tidal correction (FES2004) 32 is applied to the sea surface height before obtaining the DOT 18 , and therefore, tides may not alias the absolute velocity calculation. To obtain the vertical profiles of absolute velocity including baroclinic and barotropic components, the entire profile of geostrophic velocity calculated from CTD measurements is shifted such that the surface velocity matches the DOT-derived velocity. In this study, we determined the DOTderived velocity for the location closest to where the geostrophic velocity was calculated (Fig. 2b); the range for determining the DOT-derived velocity was 0.1°. However, since the location between Sta. 522 and Sta. 526 is in the shallow region away from where DOT can be obtained, the search range is changed to 0.2°. Estimation of eddy-induced poleward heat transport. Using along-slope potential temperature (Fig. 3a) and absolute velocity (Fig. 3d) sections, we estimate poleward heat flux at the layer between 400 and 600 dbar of the Poinsett Eddy and West Sabrina Eddy regions, based on the depths of warm water occupation and the seafloor of the shelf break. Ocean heat flux is expressed as follows; R A C P ρðθ À θ ref ÞvdA, where A is the cross-section area, C P is the specific heat of seawater, ρ is the density of seawater, θ is the potential temperature, and v is the cross-section velocity. We set −1.9°C as a reference potential temperature (θ ref ), which is approximately the freezing point of seawater.	5,538.4	2021-08-06T00:00:00.000	[ "Environmental Science", "Geology" ]
Structural, Electrical and Ferroelectric Properties of Lead-free Bi(Fe0.85dy0.15)o3 Electroceramic Compound This paper mainly reports detailed studies of structural, dielectric, impedance and ferroelectric properties of Bi(Fe 0.85 Dy 0.15 )O 3 (termed as BFDO15), fabricated via solid state reaction (SSR) method. Analysis of X-ray diffraction (XRD) data confirm the crystal symmetry changes from rhombohedral to orthorhombic symmetry. The scanning electron micrograph collected through field emission scanning electron microscopy (FE-SEM) shows a regular distribution of grains over the sample surface. The elemental composition of the sample was examined by using energy dispersive X-ray micro-analysis (EDXMA) and this confirms the existence of constituent elements of the sample. The electrical measurement was carried out using a computer-controlled phase sensitive multimeter (PSM) in a frequency range of 1 kHz - 1000 kHz and temperature range of 25 ºC – 500 ºC. Study of dielectric properties shows high dielectric permittivity and small value of dielectric loss in the sample. The frequency dependent impedance and electrical modulus analysis reveals the presence of a semiconducting nature and non-Debye type relaxation process. Analysis of ac-conductivity with respect to frequency obeys the universal Jonscher’s power law. The electric p olarization study shows enhancement in ferroelectric property of the material. Hence, based on the significant enhancement found in the structural, electrical and ferroelectric properties of BFDO15 material, it could be a promising candidate for modern device applications. Introduction A tremendous research interest in multiferroic materials, which are having simultaneously both the ferroelectricity as well as ferromagnetic order in a single phase at higher temperatures, have been experienced since the past few decades. Development of lead-free multiferroic materials added some advantages in our environment concerned and health issues of human beings as they are considered as toxic free materials for electronic device applications including information storage media, actuators, spintronics, transducers, sensors and optoelectronic devices [1][2][3]. Despite the tenacious efforts had been paid for the development of novel single phase multiferroics, bismuth ferrite (BFO) is perhaps the most aware material which have both magnetic and a strong ferroelectric (multiferroic) behavior at/above room temperature. Bismuth ferrite possesses an ABO3 type distorted perovskite rhombohedral symmetry with a space group of R3c below the ferroelectric Curie temperature (Tc = 830 °C), which have the unit cell value of a = b = 5.558 Å and c = 13.87 Å and its rhombohedral angle is 89.3º-89.48º [4,5]. BFO becomes antiferromagnetic with G-type structure below its magnetic Neel temperature (TN = 620 °C). The main reason of the ferroelectricity of BFO is originated from 6s lone pair electrons of Bi 3+ ions while the ferromagnetic/antiferromagnetic property was attributed to partly filled d orbital of Fe 3+ ions [6]. Despite being multiferroicity at/above room temperature, BFO possesses some inherent drawbacks such as structural distortion, high current leakage density, high coercive field, low remnant polarization due to irregular magnetic spin structure which hindered its potential device applications [7,8]. However, researchers had identified the ways to strengthening and improving the multiferroic properties of BFO by substituting other certain components at AB or A/B site. It has learned from the literature review that doping at the A-site using alkaline earth ions is an effective way for enhancing the multiferroic property of BFO [9][10][11][12]. Several groups of researchers had attempted through SSR method, sol-gel technique and solution combustion method at A-site substitution using some rare earth ions including Dy 3+ , Gd 3+ , Sm 3+ , La 3+ and found that enhancement in magnetic properties as well as electrical features of bismuth ferrite [13][14][15][16]. Some investigations reported that substitution by several divalent ions such as Ca 2+ [17], Mn 2+ [18], Ba 2+ [19] and Sr 2+ , Pb 2+ [20] at both sites (AB-site) of BFO had also improved the structural and ferroelectric properties of BFO. Based on the literature survey, most of the Dy doped BFO reported are at A-site doping or co-doping with some other elements while B-site Dy substitution has not yet been widely explored. In this work, we mainly focussed on the impact of B-site Dy substitution on structural, dielectric, electrical (i.e. impedance, modulus, conductivity) and ferroelectric features of BFO having a composition of Bi(Fe0.85Dy0. 15 Toledo digital balance (ML204/A01) was used for weighing the primary ingredients. All the ingredients were mixed homogeneously using mortar and pestle in a dry medium (air) for 3 h and wet (methanol) medium for 3 h in order to obtain a regular mixture sample. The mixed powder has been heated at an optimized temperature and time (750 ºC for 8 h). In order to make a sample pellet, 2% of polyvinyl alcohol (binder) was added to the heated powder. The cylindrical shaped pellets having a diameter of 11.75 mm and thickness 1.8 mm were made by applying iso-static pressure of 5x10 4 Nm -2 . Sintering process of the sample pellets was carried out at an optimized temperature 800 ºC in an electric furnace for 6 h. The high quality alumina crucibles of a cylindrical, boat and tray were utilized during the calcination and sintering process of the material. Characterization of material The basic crystal structure and phase formation of BFDO15 was measured by XRD (D8 Advance, Bruker) by applying CuKα (λ=1.5405 Å) radiation at 2º/min scan speed over Bragg's angle θ (20º ≤ 2θ ≤ 80º). POWDMULT software was utilized for measuring the unit cell parameters, the crystal system and Miller indices of BFDO15. The sample surface morphology has been measured using the FE-SEM (Carl Zeiss) for studying the distribution of grains, grain size, microstructure and density of grains. The elemental study of a sample was carried out by using energy dispersive X-ray spectroscopy (EDXS). The cylindrical shaped pellets were flattened with the help of a fine emery paper in order to get parallel and smooth surface of pellet, and then coated with a high quality silver paste (Alfa Aesar) and dried at 150 ºC for 1 h in order to eliminate the moisture. The dielectric constant, tangent loss and other electrical parameters were measured using a high precision computer-interfaced PSM (N4L, 1735) over a frequency range of 1 kHz -1000 kHz and temperature (25 ºC -500 ºC). A room temperature electric field polarization of the sample was measured using P-E loop tracer (Marine, India). Results and discussion 3.1. Structural study Fig. 1 represents room temperature XRD pattern of sintered Bi(Fe0.85Dy0.15)O3 material. The sharp and narrow X-ray diffraction peaks indicate the proper crystalline nature as well as the development of a pure phase material. However, few peaks which have a very small intensities corresponds to the Bi25FeO40 and Bi2Fe4O9 crystal system [21] (shown as * symbol in the figure) and such a small impurity is a common problem in bismuth-based materials [22]. Measurement of the unit cell dimension was carried out using X-ray data refinement and indexing software known as "POWDMULT" [23]. The unit cell dimensions and crystal system of the sample were estimated according to the best agreement between (a) the observed and Table 1. Sizes of each sample particles (P) were estimated by Scherrer's formula [24]: where λ=0.15407 nm, k is constant = 0.89, θ is maximum peak position and β1/2 is peak width at half maximum. The approximate crystallite size of BFDO15 is 38 nm. the help of a standard software called "ImageJ" and approximated to be 1.039 μm. The microstructural study was carried out using the EDXS as shown in Fig. 2b. This analysis confirmed the presence of bismuth, iron, dysprosium and oxygen (its constituent elements) with their respective atomic ratios and the purity of the sample. Moreover, EDXS analysis also reveals that BFDO15 material was effectively synthesized without impurities. The weight and atomic percentage of the constituent elements of the sample are tabulated in Table 2. capacitance (C\|\| ) were recorded with the help of high accuracy phase sensitive multimeter connected with a computer-controlled temperature furnace. Therefore, in order to know the potential applications and better understanding, it is an essential to study the dielectric properties of the sample. Dielectric constant of the sample is calculated by using the general capacitance formula given below: Microstructural analysis herein, p C = capacitance in parallel mode, t = thickness of the sample pellet, A = area of the sample pellet and 0  = permittivity of free space. for a sintered BFDO15 material. Initially, a temperature independent nature of dielectric permittivity has been detected at low temperature range, however, ɛr slowly increases with an increasing temperature for all the operating frequencies and the first hump was observed at 164 °C, which may be originated due to the magnetic transition in the as-prepared material [25]. Generally, due to the existence of different types of polarization (i.e. ionic, atomic, orientation, space charge) at low frequency region, higher value of ɛr was found at low frequency curves Table 3. then results small value of dielectric permittivity [28]. Moreover, most of the dipoles are not strong enough to follow the increasing frequency at higher frequency region which results decrease of dielectric permittivity at higher frequency region. Therefore, a significant contribution only from electronic polarization is received at higher frequency among the different types of polarizations, which subsequently results small value of dielectric constant [29]. The dispersive spectrums found at lower frequency region suggests the effect of grain boundary which spreads throughout the sample surface [30]. Frequency dependent dielectric property The graphical representation of dielectric loss (tanδ) as function of frequency is shown in Fig. 4b. The nature of the plot may be explained by a two-layer model which was proposed based on Koop's theory by Maxwell and Wagner [31]. This theory states that at low frequency, electrons are very effective as compared with other factors whereas grains are more active as compared with electrons at high frequencies. Due to the high resistive nature at grain boundary, more energy is required for moving charge carriers at low frequency region, which subsequently results high value of dielectric loss. On the other hand, a small amount of energy is used for moving charge carriers at higher frequency region due to the high conductivity and then consequences small value of dielectric loss at higher frequencies. Electrical impedance spectroscopy 3.4.1 Real (Z') and imaginary (Z'') component of impedance study Electrical impedance spectroscopy is one of the most powerful technique for investigating the dielectric and electrical characteristics of advanced ceramic materials. It has been commonly used for studying the grain and grain boundary effects and interface effect in a polycrystalline dielectric and ionic conductors [32,33]. The nature of electrical response as well as characteristics of real and imaginary components of the sample also can be examined using the experimental data. In addition, this technique is utilized for examining dynamics of the movement of ions in solid materials. Generally, an ac-signal is applied to silver coated pellet sample in order to measure the electrical response of the sample specimen, then analyzed and presented in different formats for studying its electrical properties. The real (Z') and imaginary ) can be determined by using the given formula [29]: herein, τ = RC = relaxation time, ω = angular frequency. 1-40 kHz). Then, the impedance spectrums amalgamate regardless to temperature at higher frequency division (40-1000 kHz), which was associated to the release of space charge and electrode effect [34]. The presence of decrement in barrier features with increasing temperature may also be associated with the increase of conductivity at higher frequencies [35]. The dispersive spectrum at low frequency division has also been noticed from the plot, that shows the decreasing nature of Z' with an increasing temperature suggesting the negative temperature coefficient of resistance (NTCR) behavior of the as-prepared material, it is the common feature of a semiconducting materials. The decreasing trend of Z' with a growing temperature as well as frequency is good in line with our other impedance result [29]. process/strength exists in a specific frequency [36]. The electrical relaxation processes of the sample are generally originated due to existence of immobile charge carriers (electron) at lower temperature and vacancies/defects created at higher temperature [37]. We can imagine the spread of relaxation time from the width of peaks as shown in Fig. 4b. According to the ideal Debye type relaxation, the centre of a semicircles should lie on/above the real axis. Nevertheless, it is obvious from the plot that the centre of a semi-circular arcs were falls below the real axis (x-axis), which confirms the presence of non-Debye type of electrical relaxation process in BFDO15 material [38]. Moreover, the graph clearly shows that the grain and grain boundary contribution of relaxation process at lower temperature (˂250 °C) and higher temperature (>275 °C) spectrums, respectively. The behavior of semi-circular arcs reveals the existence of distribution of relaxation time which were originated by irregular distribution of grains in our sample [39]. Electrical modulus spectroscopy The electrical modulus spectroscopy is one of the most useful technique for examining the nature of electrode polarization, electrical conductivity, conduction mechanism, relaxation time and the effect of bulk and grain boundary properties of the as-prepared material [40]. In addition, this method can also be utilized for studying several types of electrical processes which developed in the present studied compound. The real (M') and imaginary part (M'') of modulus can be determined by using the mathematical relation [41] given below: herein, , all the symbols have their usual meanings. peak of M'' shifted towards higher frequency region when the temperature increases and faster movement of charge carriers was observed and these results decreases in a relaxation time [44,45]. The trend behavior of M'' clearly indicates the existence of temperature dependent relaxation process and hopping charge mechanism in the material [46,47]. Moreover, the continuous increasing curve of the M'' spectrum for all the selected temperatures signifies that the ions can successfully jump from one site to the neighbouring site [48]. In addition, the asymmetric broadening of peaks/curves found in the plot recommends non-Debye type of relaxation process in the material. Electrical conductivity study Electrical conductivity analysis of a compound is an important method for determining electrical conduction mechanism which is required for examining transport properties of material. Conduction mechanism of materials is mainly influenced by microstructure of the sample and defects. In perovskite materials, the synthesis process such as calcination and sintering temperature, atmosphere, time, etc. can generate oxygen as well as cation vacancies and other defects [49]. Fig. 7a shows the change of ac-conductivity versus frequency at some temperature sets for Bi(Fe0.85Dy0.15)O3 material. It is noted that the value of ac-conductivity (σac) linearly increases as the frequency increase and tends to merge all the spectra at high frequency site. The frequency dependent electrical conductivity may be divided into two regions: (a) A plateau found at low frequency region and (b) a dispersion at high frequency region. At lower frequency, ac-conductivity seems not depends on frequency by forming a plateau. This frequency independent region is considered as zero-frequency conductivity (σo or σdc), which occurs due to thermally influenced space charge carriers. On the other hand, the frequency dependent conductivity was found at higher frequency region, so this site is considered as acconductivity (σac). Therefore, the σac may be calculated from the dispersive region (higher frequency site) using Jonscher's power law:  is the total conductivity, () dc  is the direct current conductivity of a sample, A is constant depending on temperature and n is the temperature dependent exponential. According to this, when a mobile charge carrier hops from one site to its neighbouring site, it persists in an oscillation state between two possible minimum energy [50]. Generally, at low frequency, the ions jump from their original site to the neighbouring vacant sites, whereas at higher frequencies, some ions move back to their original site through hopping process, which may results increase in conductivity at low frequencies [51]. The exponent n is related to the hopping rate and relaxation time. The hopping process becomes slower as compared with site relaxation, if n ˂ 1, whereas if n > 1, back-hopping conduction process is faster as compared with site relaxation [52]. The plot clearly shows that σac slowly increases as the temperature increase upto certain temperature and then suddenly falls at high temperature. However, it sharply increases as the temperature rises at further higher temperature region. The behavior of σac trend follows Arrhenius conductivity relation:   , where symbols have their usual meanings. In addition, the continuous increases of σac with an increasing temperature confirmed the NTCR behavior of a sample. It has been observed that all the different frequency spectrum were merged at high temperatures which attributed to the recombination of released space charge at higher temperature [53]. Since, the hopping of electron and polaron plays an important role in a conduction mechanism, the value of activation energy, Ea is high at lower frequency. It is also noticed from the plot that the activation energy rapidly increases with an increasing temperature while it decreases with an increasing frequency. As shown in Fig Table 4. Therefore, the density and distributing nature of oxygen vacancy effects the width and height of Schottky barrier and then influenced the resistance of a material which subsequently affected the polarization switching behavior of the sample [54]. Finally, instead of being observed an unsaturated loop, a significant enhancement in the value of remnant polarization has been observed in the present studied sample as compared with pure BFO [55]. Conclusion The electro-ceramic material, Bi(Fe0.90Dy0.10)O3 was fruitfully synthesized via a conventional SSR route. The crystal structure was studied using powder XRD method and found that the structure changes from rhombohedral to orthorhombic symmetry and the average crystallite Figure 1 The XRD pattern of sintered Bi(Fe0.85Dy0.15)O3 material.	4,102	2021-02-08T00:00:00.000	[ "Materials Science", "Physics", "Engineering" ]
Contraction limits of the proton-neutron symplectic model The algebraic approach to nuclear structure physics allows a certain microscopic collective motion algebra to be also interpreted on macroscopic level which is achieved in the limit of large representation quantum numbers. Such limits are referred to as macroscopic or hydrodynamic limits and show how a given microscopic discrete system starts to behave like a continuous fluid. In the present paper, two contraction limits of the recently introduced fully microscopic proton-neutron symplectic model (PNSM) with the Sp(12,R) dynamical symmetry algebra are considered. As a result, two simplified macroscopic models of nuclear collective motion are obtained in simple geometrical terms. The first one is the U(6)-phonon model with the semi-direct product structure [HW(21)]U(6), which is shown to be actually an alternative formulation of the original proton-neutron symplectic model in the familiar IBM-terms. The second model which appears in double contraction limit is the two-rotor model with the ROTp(3)⊗ ROTn(3) ⊃ ROT (3) algebraic structure. The latter, in contrast to the original two-rotor model, is not restricted to the case of two coupled axial rotors. In this way, the second contraction limit of the PNSM, provides the phenomenological two-rotor model with a simple microscopic foundation. Introduction Symmetry is an important concept in physics.In finite many-body systems, it appears as time reversal, parity, and rotational invariance, but also in the form of dynamical symmetries [1][2][3][4][5][6].The standard symmetry approach allows the construction of a Hamiltonian of a system under consideration which is, or nearly so, invariant under a group of symmetry transformations.Group theory then allows one to construct basis states realizing the symmetry and explicit matrix elements for physically interesting transition operators themselves classified by the symmetry.Many properties of atomic nuclei [1,[3][4][5][6] have been investigated using algebraic models, in which one obtains bands of collective states which span irreducible representations of the corresponding dynamical groups. The algebraic approach in nuclear structure physics is also particularly useful because a certain algebra can often be interpreted both at the macroscopic (hydrodynamic) and microscopic levels.There is a formal expansioncontraction group-theoretical procedure [7][8][9] which allows to accomplish such a relation between the microscopic and macroscopic aspects of a given many-particle nuclear system.Contraction limits are obtained at large representations of the microscopic collective algebras under consideration.In this way, at large quantum numbers, one obtains the macroscopic or hydrodynamic limits which show how a discrete microscopic system starts to behave like a continuous hydrodynamic system.That a e-mail<EMAIL_ADDRESS>Permanent address: Institute of Nuclear Research and Nuclear Energy, Bulgarian Academy of Sciences, Sofia, Bulgaria is why such limits are referred to as hydrodynamic or many-particle limits.Through the exploration of macroscopic limits new models appear in simple geometrical terms which reveal the underlying physical dynamics of the original algebraic models. Recently, the fully microscopic proton-neutron symplectic model (PNSM) of nuclear collective motion was introduced by considering the symplectic geometry and possible collective flows in the two-component many-particle nuclear system [10].Further, it was shown that, in its hydrodynamic limit, it reduces to the U(6)-phonon model with the semi-direct product structure [HW(21)]U (6) which unifies both the two-fluid irrotational-flow collective model and a microscopically based U(6) model [11].The latter naturally generalizes the SU(3) model of Elliott [12] for the case of two-component many-particle nuclear system and is related to the valence proton-neutron degrees of freedom.From the hydrodynamic perspective, the U(6)-phonon model therefore includes the irrotational collective flows and their coupling to the intrinsic vortex degrees of freedom.The latter is of vital importance for the appearance of the low-lying collective bands.In this way the extra degrees of freedom contained in this larger U(6) algebraic structure will therefore embrace the basic SU(3) rotor as well as the low-lying vibrational degrees of freedom. The appearance of an U(6) intrinsic structure in both the PNSM and U(6)-phonon model is of significant importance for the microscopic theory of nuclear collective excitations.In this regard, we recall that the popular Interacting Boson Model [13] has clearly demonstrated that simple algebraic ways exist to get collective spectra within U(6)-based scheme.Then, within the framework of both the PNSM and U(6)-phonon model, the full range of lowlying states could be described by microscopically based U( 6) structure along the lines of the IBM, albeit in contrast to the latter, renormalized by their coupling to the giant resonance vibrations.This result could not be overestimated recalling also that in order to obtain the low-lying excited collective bands (e.g., beta band) within the framework of the one-component symplectic model [14] one needs to involve a representation mixing caused by, e.g., pairing, spin-orbit and other symplectic-breaking components of the nuclear interaction (cf.Ref. [15]). In the present contribution, we show that the U( 6)phonon model represents actually an alternative formulation of the PNSM in the familiar IBM-terms.For this purpose, the relevant representation theory of the U( 6)phonon model is shortly considered.It is shown that it coincides with that of the original proton-neutron symplectic model [16].The relation of the U( 6)-phonon model irreps with the shell-model classification of the basis states is considered by reducing the U(6) group to the direct product space of SU p (3) ⊗ SU n (3) irreps, generalizing in this way the Elliott's S U(3) model [12] for the case of twocomponent nuclear system.The physical content of the intrinsic substructure SU p (3) ⊗ SU n (3) ⊃ SU(3) of U( 6), associated with the proton-neutron valence shell degrees of freedom, is exposed further by considering its contraction limit.It is shown that, at large representation quantum numbers, it reduces to that of two coupled rigid rotors (two-rotor model) ROT p (3)⊗ROT n (3) ⊃ ROT (3).The latter, in contrast to the original two-rotor model (TRM) [17], is not restricted to the case of two coupled axial rotors.In this way, the second contraction limit of the PNSM, provides the phenomenological TRM with a simple microscopic foundation. The proton-neutron symplectic model Collective observables of the proton-neutron symplectic model, which span the Sp(12, R) algebra, are given by the following one-body operators [10]: where i, j = 1, 2, 3; α, β = p, n and s = 1, . . ., m = A−1.In Eqs.(1)−(4), x is (α) and p is (α) denote the coordinates and corresponding momenta of the translationally-invariant Jacobi vectors of the m-quasiparticle two-component nuclear system and A is the number of protons and neutrons.By considering the m Jacobi quasiparticles instead of A pro-tons and neutrons, the problem of center-of-mass motion is avoided from the very beginning.Dynamical content of the PNSM is revealed by considering different dynamical groups that can be constructed from the symplectic operators which generate different classical collective motions, including a wide class of both the in-phase (isoscalar) and out-of-phase (isovector) excitations of the proton subsystem with respect to the neutron one, as well as collective excitations of the combined proton-neutron system as a whole.Thus, the Sp(12, R) group provides a quite general framework for the investigation of the nature of classical collective motions in nuclei.The PNSM appears as a hydrodynamic collective model of the proton-neutron nuclear system which include 21 collective irrotational-flow degrees of freedom and an U( 6) intrinsic structure associated with the vortex degrees of freedom [10]. To quantize the model one has to construct the irreducible representations of the Sp(12, R) group, appropriate to the many-particle system.Thus, the model space is spanned by the irreducible representations of the Sp(12, R) algebra.In terms of the harmonic oscillator creation and annihilation operators the many-particle realization of the Sp(12, R) Lie algebra is given by [16]: The number-conserving operators (8) generate the maximal compact subgroup U(6) of Sp (12, R). From the shell-model perspective, the PNSM appears as a natural multi-major-shell extension of the generalized proton-neutron SU(3) scheme which takes into account the core collective excitations of monopole and quadrupole, as well as dipole type associated with the giant vibrational degrees of freedom [16].Each Sp(12, R) irreducible representation is determined by a symplectic bandhead or an intrinsic U( 6) space which can be fixed by the underlying proton-neutron shell-model structure, so the theory becomes completely compatible with the Pauli principle.Moreover, the intrinsic U( 6) structure is of vital importance for the appearance of the low-lying collective bands. Further, at large Sp(12, R) representations, as will see in the next section, the algebraic structure of the protonneutron symplectic model simplifies as a result of the group contraction of its underlying dynamical symmetry algebra. 03012-p.2 3 Contraction of the Sp(12, R) algebra We consider first the sp(2, R) ≈ su(1, 1) subalgebra.The sp(2, R) generators are obtained from the Sp(12, R) ones by summation with respect to both indices i and α [11] They obey the following commutation relations: which differ from the standard angular momentum su(2) commutation relations only by the sign on the right hand side of the last commutator, and are defining for the su(1, 1) algebra, locally isomorphic to sp(2, R) algebra.Note that the operator 2K 0 is the harmonic oscillator Hamiltonian in units ω with eigenvalue σ 0 = (σ 1 + . . .+ σ 6 ) + 6 2 (A − 1).From the commutation relations ( 12) one readily obtains [11] [ where the operator D is defined as D = (2K 0 −σ 0 I)/σ 0 and I is the identity operator.For a harmonic oscillator shellmodel state of excitation energy 2ν ω above the lowestweight state \|σ , the operator D has eigenvalue 2ν/σ 0 , which in the limit σ 0 →∞ (hydrodynamic limit) tends to 0. In practice one needs only the condition 2ν << σ 0 .The U(6) lowest-weight state [σ 1 , . . ., σ 6 ] which defines the Sp(12, R) irreducible representation can be fixed by the underlying proton-neutron shell-model structure.For example, the shell model considerations give the U( 6) irrep [132, 86, 70, 70, 70, 70] for 154 Sm.From the latter one finds that the minimum number of oscillator quanta allowed by the Pauli principle is σ 0 = 957.So, the condition 2ν << σ 0 is satisfied for small ν.Thus, in the limit 2ν << σ 0 , one obtains which can be written in terms of the F and G operators ( 9) and ( 10) From the latter it follows that the operator 1 2 Then the commutation relation ( 15) is satisfied if the following correspondence is made Equation ( 16) together with A 0 ↔ A 0 establish a relation between the sp(2, R) ≈ su(1, 1) ⊂ su p (1, 1) ⊕ su n (1, 1) and u( 1)-phonon algebras.The obtained correspondence is known also as a contraction limit.In a similar manner, one can obtain the correspondence between the remainder of the Sp(12, R) and U( 6)-phonon algebras.Thus, the full correspondence is given by [11]: and their hermitian conjugate counterparts.The set of op- , I} (τ = p, n, δ) generates the hw(21) phonon algebra.The later consists of the IBM-3 building blocks [13] plus the extra degrees of freedom represented by the components of the dipole p-boson.From the above correspondence, however, we see that in contrast to the IBM case in which the s and d bosons represent correlated valence fermion pairs, in the hydrodynamic limit of the Sp(12, R) model, the s and d bosons are in fact superpositions of pairs of harmonic oscillator vector excitations (phonons) (cf.Eqs.( 6)-( 7)) which represent multi-major-shell collective core excitations. The U(6)-phonon and two-rotor models 4.1 The U(6)-phonon model The phonon and u( 6) operators together span a semi-direct sum Lie algebra [hw(21)]u(6) which we call u(6)-phonon algebra.Then, the U(6)-phonon model is a model whose state space carries an unitary irreducible representation of the u(6)-phonon algebra and for which the Hamiltonian and collective observables are simple functions of the u( 6) and phonon operators. If we identify the ±2 ω phonon operators of the hw(21) algebra with those of the two-fluid irrotationalflow model, then it is clear that the PNSM contains the former as a submodel.However, it is also clear from the microscopic realization of the sp(12, R) algebra that the additional u(6) intrinsic degrees of freedom appear which turn out to be crucial for the appearance of low-lying collective bands.From the general linear flow origins of the symplectic model, we know that these extra degrees of freedom are associated with the internal (vortex) motions [10].Among the latter are the vortex spin rotations.In this way, from the hydrodynamic perspective, the U(6)phonon model is an extension of the two-fluid irrotationalflow collective model to include the intrinsic vortex degrees of freedom. NSRT15 An U(6)-phonon irreducible representation is defined by the U(6) highest-weight state \|σ with σ = [σ 1 , . . ., σ 6 ] which is also a phonon vacuum, i.e. satisfies the following equations u\|σ = 0, (18) with σ 0 = σ 1 + . . .+ σ 6 + 6 2 (A − 1), where a, b = 1, . . ., 6 and u represents any annihilation operator of the hw(21) algebra.Note that the u( 6)-phonon algebra is non-compact and hence its unitary irreducible representations are infinite dimensional and have no highest weights.Thus its irreps are defined by the lowest-weight states which are also U( 6) highest-weight states.Such an U( 6) irrep whose highest-weight state is also an U( 6)-phonon lowest-weight state is said to be a lowest-grade U( 6) irrep for that U( 6)phonon irrep.The other states of the basis are then generated by repeated action of the phonon raising operators on the U(6) irreducible states.The phonon operators 6).Then, the U( 6)-phonon basis can be classified according to the chain [11]: and written in the form with U(6) quantum numbers n = [n 1 , . . ., n 6 ] running over the sets of even integers for which n 1 ≥ n 2 ≥ . . .≥ n 6 ≥ 0 and n 1 + . . .+ n 6 = 2N.In (22) the square brackets denote an U(6)-coupling of the polynomial Z (n) (u † ) with the intrinsic U(6) structure \|σ , ρ is a multiplicity index and η denotes a basis for the coupled U( 6) irrep E. The latter can be fixed by considering different subgroup chains which reduce the U(6) to the rotational group SO (3).The shell-model considerations suggest the reduction [16]: The chain (23) naturally generalizes the Elliot SU(3) model [12] by extending the model space to the direct product space SU p (3) ⊗ SU n (3) of proton and neutron subsystems.The SU(3) irreps of the two subsystems are subsequently coupled to the SU(3) irrep of the combined proton-neutron system.The chain (23) corresponds to the following choice of the index η = γ(λ p , μ p )(λ n , μ n ) (λ, μ)KLM labeling the basis states (22).In this way one obtains a representation theory which is precisely the same as that given in Ref. [16]. The phonon structure of the [HW( 21)]U( 6) model therefore enables it to simultaneously include the giant monopole and quadrupole, as well as dipole resonances into nuclear collective dynamics.On the other hand, the low-lying rotational bands are associated primarily with the proton-neutron valence shell degrees of freedom represented by the intrinsic U( 6) structure. If the U( 6)-phonon lowest-weight state \|σ reduces to the phonon vacuum \|0 , then one obtains a 21-dimensional two-fluid irrotational-flow collective model.Thus, by allowing the phonon vacuum to have a non-trivial intrinsic U( 6) structure gives the collective model a much richer structure and a framework for considering the low-lying collective bands in an unified manner.This result is of significant importance for the theory of nuclear collective excitations. As an illustration, the U( 6)-phonon irreducible representation of 154 Sm, determined by the lowest-grade U (6) irrep [132,86,70,70,70,70], is given in Table 1.As we can see from it, the relevant U( 6) irreps comprising the U( 6)-phonon irreducible representation under consideration are not fully symmetric.However, one expects the most symmetric U( 6) irreps E to be dominant in the lowenergy spectra of the heavy deformed even-even nuclei.The classification of the SU(3) basis states for 154 Sm according to the decompositions given by the chain (23) for the U(6)-phonon irreducible representation under consideration, restricted to the most symmetric two-rowed U(6) partitions, is given in Table 2. From the latter the structure of the basis becomes evident.The SU(3) basis states so obtained are precisely those which can be obtained respectively by acting on the intrinsic base space states (λ, μ) by the S U(3) (2, 0) s † τ , d † M,τ and (0, 1) p † M,δ phonon raising operators.We also see repeating of the SU(3) multiplets within different U(6) irreducible representations. The fact that one obtains a nonscalar U( 6)-phonon irreducible representation for a particular nucleus is a very important feature of the present approach, which is a consequence of the two-component composite character of the nuclear systems.As we can see from Table 2, the U(6) intrinsic structure (the first row with 2N = 0) contains many SU(3) multiplets which are appropriate for the description of different low-lying collective bands (ground state, β, γ, etc.) in the spectra of heavy even-even deformed nuclei.In this way the U(6) intrinsic structure of the U(6)-phonon model provide us with a framework for the simultaneous description of low-lying collective states of well deformed nuclei in a manner similar to that of IBM, but in contrast to the latter, with states which are renormalized by their coupling to the giant resonance vibrational degrees of freedom. The coupled two-rotor model The combined su(3) algebra of the proton-neutron system is spanned by the set su(3 The elements of su (3) satisfy the following commutation relations: Now, if we introduce the rescaled mass quadrupole operators q M = Q M √ , where = C 2 [SU(3)] = 4(λ 2 + μ 2 + λμ + 3λ + 3μ) is the eigenvalue of the second order Casimir operator C 2 [S U (3) 3), for the Eq. ( 26) one obtains when →∞.In this way, we have obtained the well known result that, in the limit of large su(3) quantum numbers, the su(3) algebra contracts to that of rigid rotor algebra rot(3) = [R 5 ]so(3) ≡ {q M , L M } [18], where R 5 is the fivedimensional abelian Lie algebra spanned by the commuting mass quadrupole operators and so(3) is the angular momentum algebra.Similarly, one can obtain the contractions of the proton and neutron su(3) algebras.Thus, in double contraction limit, we have obtained the subgroup chain: which contains the algebraic structure of two coupled rigid rotors (two-rotor model).Note, however, that in contrast to the original TRM [17], the present algebraic structure assumes the more general case of two coupled non-axial rotors.This is obvious even on the SU(3) level.We recall that the reduction of the generic U( 6) irreducible representation to the direct product irreps of SU p (3) ⊗ SU n (3) allows irreps of the type (λ p , μ p )⊗(λ n , μ n ) with nonzero values of the quantum numbers λ and μ characterizing the proton and neutron SU(3) irreps.The latter geometrically corresponds to two nonaxial rotors [19].The considered contraction limit gives the phenomenological TRM simple microscopic foundation. From the shell-model perspective, two-rotor model degrees of freedom correspond to the SU p (3) ⊗ SU n (3) ⊃ SU(3) proton-neutron dynamics of the 0 ω space.The hydrodynamic limit (σ 0 → ∞) of the symplectic model shows, from the other side, that the underlying vibrational degrees of freedom are associated with the 2 ω giant resonances of monopole (L=0), quadrupole (L=2) and dipole (L=1) type.Note that the latter is not the giant dipole resonance corresponding to the 1 ω excitation mode, which however can be easily incorporated into the PNSM by making its central extension, i.e. by considering the semidirect structure WSp(12, R) ≡ [HW (6)]Sp(12, R) where the HW( 6) algebra is spanned by the harmonic oscillator raising and lowering operators (5) together with the identity operator I. Then, in double contraction limit, the states of a single Sp(12, R) irrep could be identified with the two-rotor NSRT15 03012-p.5 model states coupled to the 21-dimensional vibrator associated with the giant resonance vibrational degrees of freedom.We note that, from the hydrodynamic perspective, the latter are of irrotational-flow character [10]. Conclusions It was shown that the algebraic approach to nuclear structure physics allows a certain microscopic collective motion algebra to be also interpreted on macroscopic level which is achieved in the limit of large representation quantum numbers.Such limits are referred to as macroscopic or hydrodynamic limits and show how a given microscopic discrete system starts to behave like a continuous fluid. In particular, in the present contribution, two contraction limits of the recently introduced fully microscopic proton-neutron symplectic model with the Sp(12, R) dynamical symmetry algebra were considered.As a result, two simplified macroscopic models of nuclear collective motion are obtained in simple geometrical terms.The first one is the U( 6 At large quantum numbers, the intrinsic substructure SU p (3) ⊗ SU n (3) ⊃ SU(3) of U( 6), associated with the proton-neutron valence shell degrees of freedom, further reduces to ROT p (3) ⊗ ROT n (3) ⊃ ROT (3), i.e. to that of two coupled rigid rotors (two-rotor model).In this way, in double contraction limit, the sp(12, R) algebra reduces to the coupled two-rotor model algebra rot p (3) ⊕ rot n (3) ⊃ rot(3) and a phonon algebra hw(21) of the giant resonance vibrational degrees of freedom.The full range of lowlying collective states could then be described as two-rotor model states, renormalized by their coupling to the giant resonance vibrations. DOI: 10 .1051/ C Owned by the authors, published by EDP Sciences, 201 )-phonon model with a semi-direct product structure [HW(21)]U(6), which is shown to be actually an alternative formulation of the original proton-neutron symplectic model in the familiar IBM-terms.The phonon structure of the [HW(21)]U(6) model enables it to simultaneously include the giant monopole and quadrupole, as well as dipole resonances into nuclear collective dynamics.The relation of the U(6)-phonon model irreps with the shell-model classification of the basis states is further considered by reducing the U(6) group to the direct product space of SU p (3) ⊗ SU n (3) irreps, generalizing in this way the Elliott's SU(3) model for the case of two-component nuclear system.The SU(3) irreps of the two subsystems are subsequently coupled to the SU(3) irrep of the combined proton-neutron system.The U(6) intrinsic structure of the [HW(21)]U(6) model therefore gives a framework for the simultaneous description of the ground state band, as well as the other excited low-lying collective bands.	5,255.6	2016-01-19T00:00:00.000	[ "Physics" ]
Molecular logic of the Zur-regulated zinc deprivation response in Bacillus subtilis Bacteria respond dynamically to the changes in zinc availability. Repression by the Bacillus subtilis transcription factor Zur requires Zn(II), which binds with negative cooperativity to two regulatory sites per dimer to form, sequentially, Zur2:Zn3 and Zur2:Zn4 forms of the repressor. Here we show that, as cells transition from zinc sufficiency to deficiency, operons regulated by Zur are derepressed in three distinct waves. The first includes the alternative RpmEB(L31) and RpmGC(L33) ribosomal proteins, which mobilize zinc from the ribosome, whereas the second includes the ZnuACB uptake system and the YciC metallochaperone. Finally, as zinc levels decrease further, the Zur2:Zn3 form loses Zn(II) leading to derepression of RpsNB(S14) and FolE2, which allow continued ribosome assembly and folate synthesis, respectively. We infer that zinc mobilization from intracellular zinc stores takes priority over energy-dependent import, and our results link the biochemistry of zinc sensing by Zur to the molecular logic of the zinc deprivation response. A hallmark of life is the ability to adapt to changing environmental conditions, often by altering the expression of DNA through the action of transcription factors. In many systems, we understand in atomic detail how activators and repressors are converted from their inactive to their active forms, and genomics-level approaches, such as transcriptomics and chromatin immunoprecipitation (ChIP), provide an overview of the complete suite of genes (regulon) that is affected by individual transcription factors. A quantitative understanding of transcriptional control mechanisms also requires an appreciation for the ways in which transcription factors are regulated (often involving integration of multiple inputs), the combinatorial nature of their interactions at specific regulatory regions and the graded responses resulting from variations in operator occupancy, as a function of transcription factor activity. Here we focus on a relatively simple system, the regulation of the zinc deficiency response mediated by the Bacillus subtilis zinc uptake regulator (Zur) [1][2][3] . B. subtilis Zur is representative of a large class of bacterial metal-sensing transcription factors (metalloregulatory proteins) whose DNA-binding activity is regulated by the reversible binding of metal ions 4 . These proteins therefore function as 'one-component' regulators that serve to directly couple changes in concentration of a simple ligand (an inorganic ion) to DNA occupancy 5 . B. subtilis Zur, like other members of the Fur family of metalloregulatory proteins, is a dimer in solution and requires a structural Zn(II) ion for protein folding and dimerization 3 . When Zn(II) levels are sufficient, Zn(II) is additionally bound to a regulatory site within each monomer and the resulting, fully metallated protein (designated Zur 2 :Zn 4 ) is an active repressor. Zur is expressed at relatively uniform level across a range of growth conditions 6 , and its activity is regulated primarily, if not exclusively, by the reversible binding of Zn(II) ions to this regulatory binding site. As cells transition from Zn(II) sufficiency to deficiency, Zur transitions first to a partially metallated dimer (Zur 2 :Zn 3 ) and finally to the inactive resting form with Zn(II) bound only at the structural sites, Zur 2 :Zn 2 . The presence of the Zur 2 :Zn 3 intermediate results from the B20-fold negative cooperativity between the binding of the first and second Zn(II) ions to the regulatory sites in the Zur dimer 3 . Zur is known to regulate at least seven operons encoding proteins that facilitate adaptation to Zn(II) limitation 2,7-9 . As might be anticipated for a stress response activated when an essential metal becomes limiting for growth, a key part of this adaptive response is the expression of a high-affinity ABC transporter uptake system encoded by the znuACB operon, which likely functions together with a Zn(II) scavenging lipoprotein, ZinT [10][11][12] , and a highly abundant member of the COG0523 family of metallochaperones (YciC) conserved in Bacteria, Eukarya and Archaea 13 . Adaptation to Zn(II) deficiency also commonly involves the expression of alternative ribosomal protein paralogues that functionally replace ribosomal proteins that require Zn(II) for function 14,15 . In B. subtilis, two Zur-regulated ribosomal proteins are paralogues of L31 (designated L31 and encoded by rpmEB) and L33 (designated L33, encoded by rpmGC). Induction of L31 and L33* serves to displace the cognate Zn(II) containing proteins from the surface of the assembled ribosome thereby mobilizing Zn(II) into the cytoplasm for redistribution 16,17 . The expression of a third ribosomal protein paralogue (S14, encoded by rpsNB) allows for de novo ribosome synthesis, which might otherwise become limited by the cell's ability to synthesize the Zn-requiring S14 protein 7,18 . Another process that becomes limiting when cells are zinc deficient is folate biosynthesis due to the Zn dependence of the FolE1 GTP cyclohydrolase. Zur also regulates an alternate, Zn-independent FolE2 enzyme 19 . In general, the replacement of enzymes dependent on one metal ion by an alternative enzyme dependent on a different metal or a non-metal cofactor is a widely conserved mechanism to facilitate adaptation to metal-limiting growth conditions 20 . We previously speculated that the presence of negative cooperativity in Zn(II) binding might provide a mechanism for a graded response to changing Zn(II) levels 3 . Here we have tested this hypothesis and monitored the relative sensitivity of each of the Zur-regulated operons to zinc deprivation. Our results indicate that as cells transition from zinc sufficiency to zinc limitation operons regulated by Zur are derepressed in three distinct waves. First, induction of RpmEB(L31) and RpmGC(L33) mobilizes Zn(II) from the ribosome and ZinT primes the cell for Zn(II) import. Next, the high-affinity ZnuACB transporter for Zn(II) import is expressed together with the YciC metallochaperone. Finally, RpsNB(S14) and FolE2 are expressed. Studies with strains expressing a mutant Zur protein (C84S) that can form the partially active Zur 2 :Zn 3 form, but which is impaired in its ability to transition to the fully active Zur 2 :Zn 4 form, indicate that this partially active form is still sufficient to repress the rpsNB and folEB genes in vivo and still binds with high affinity to the corresponding regulatory regions in vitro. We conclude that the negative cooperativity in Zn(II) binding to Zur is responsible for the transition between the middle and late genes derepressed as part of the zinc limitation adaptive response. Results The Zur regulon is derepressed in a stepwise manner. We used an S1 nuclease protection assay to monitor the induction of six different Zur-regulated operons in cells exposed to the zinc chelator TPEN for various periods of time (Fig. 1a). The results reveal that Zur's target genes are induced in three waves which we assign as early (zinT and rpmEB), middle (yciC and znuA) and late genes (folEB and rpsNB) with respect to their time of induction after chelator addition. While these genes differ in their basal level of expression under zinc sufficient conditions (no added TPEN), they are all highly induced by TPEN but with different kinetics. On the basis of these findings, we extended this analysis to include an additional Zur-regulated gene, rpmGC, which is a pseudogene in B. subtilis 168 due to a frameshift mutation. Although rpmGC does not encode a functional protein in this genetic background, its expression is still regulated by Zur 7 . Since the rpmEB gene (encoding the L31* ribosomal protein) is induced early, we hypothesized that rpmGC (encoding L33) would also be induced early. Indeed, rpmGC is induced at a time intermediate between the two early genes and the middle genes noted above (Fig. 1a). These results are supported by studies in which cells were treated for a fixed time (5 min) with increasing amounts of chelator (Fig. 1b). Average values from three independent experiments are presented in Supplementary Figs 1 and 2 and representative full gel images are shown in Supplementary Figs 7 and 8, which revealed a similar stepwise derepression. Stepwise derepression correlates with operator occupancy. The simplest model to account for the stepwise induction of the Zur regulon is that binding of Zur to its various operator sites is differentially affected as a function of zinc depletion. To monitor the occupancy of Zur protein on its operator sites in vivo, we used ChIP. As predicted, the operator occupancy at the six tested sites declined in parallel with the observed increase of mRNA expression level (Fig. 2). Indeed, the same three sets of early, middle and late genes can be discerned. These results are supportive of a model in which mRNA levels of Zur-regulated operons are modulated primarily, if not exclusively, by the relative occupancy of Zur at its operator sites. Role of negative cooperativity of Zn(II) binding to Zur. Previously, we demonstrated that Zur forms a dimer and that efficient dimerization requires a structural Zn(II) site 3 . The Zur 2 :Zn 2 protein binds operator DNA with low affinity and mutations that prevent binding of Zn(II) to the regulatory site are non-functional in repression in vivo. The affinity of Zur for a cognate operator site (folEB) was shown to increase by 1,000-fold in the presence of Zn(II) sufficient to form the active Zur 2 :Zn 4 form of the repressor 3 . Biochemical measurements revealed that the binding of Zn(II) to the regulatory sites (one in each monomer) occurs with negative cooperativity with a B20-fold difference in the measured affinities. Further, we found that a C84S mutation to one of the Zn(II) ligands of the regulatory metal binding site did not significantly affect binding of the first Zn(II), but greatly impaired the second binding event (a decrease in affinity of B10 4 -fold). Since the C84S mutant protein was still able to repress a Zur-regulated target (yciC) in vivo, we concluded that this protein likely retained some repressor activity despite being impaired in the transition from the Zur 2 :Zn 3 to the Zur 2 :Zn 4 form of the repressor 3 . One possible mechanism to enable stepwise repression of Zur-regulated genes would be differential activity of the Zur 2 :Zn 3 and Zur 2 :Zn 4 forms of the repressor. To explore this possibility, we compared the ability of FLAG-tagged variants of the wild-type (WT) Zur and the C84S mutant to repress each of the six Zur-regulated operons using S1 nuclease protection ( Fig. 3; full gel images in Supplementary Fig. 9). As controls, we also included Zur mutants shown previously to be defective for binding the structural Zn(II) ion (C98S) or compromised for protein dimerization (H124A). To monitor the basal level of repressor activity in standard growth conditions, cells were grown to mid-logarithmic phase (OD 600 ¼ 0.5) in Luria-Bertani (LB) medium. Zinc limitation was imposed by the treatment with 2 mM EDTA for 1 h, and washed cells were then resuspended in LB medium amended with 25 mM Zn(II) for 1 h to restore repression (Fig. 3a). Unlike TPEN, EDTA does not enter cells and imposes zinc limitation by chelating extracellular zinc. Under these conditions, WT Zur was able to repress all six operons with the ambient zinc levels in our LB medium (B18 mM), consistent with the previous work 1,2,7 . All six tested operons were induced by EDTA, leading to mRNA accumulation, and this was reversed by resuspension in zinc-amended medium (Fig. 3b, lanes 1-3). In contrast, in cells expressing the C84S Zur mutant both the early and middle genes were partially derepressed in LB medium, whereas the late genes (folEB and rpsNB) were fully repressed. Again, all six operons were derepressed by EDTA and repression was complete in the zinc-amended medium (lanes 4-6). These results suggest that the C84S mutant, which is likely to be largely restricted to the Zur 2 :Zn 3 form in unamended medium, is still active in repression of the late genes, but not the early and middle genes. In contrast to the C84S mutant, a Zur protein defective for dimerization (H124A), is reduced in activity at all six genes (lanes 7-9), and a mutant defective for coordinating the structural zinc ion (C98S) is grossly defective in function (lanes 10-12), despite similar levels of expression of all Zur proteins under all tested conditions ( Supplementary Fig. 3e). Table 2). Enriched DNA values for each Zur target gene from three independent experiments were normalized to the input (1% of the amount used for each ChIP experiment), and are presented as the relative per cent (%) of binding compared with the nontreated sample (full Zur binding; 100%). NATURE COMMUNICATIONS \| DOI: 10.1038/ncomms12612 ARTICLE To confirm that the effects observed at the level of mRNA are reflective of Zur operator occupancy, we conducted ChIP experiments using the four FLAG-tagged Zur variants (Fig. 3c). The fractional enrichment of each operator site (relative to that observed for WT Zur in LB medium) is consistent with the observed mRNA levels: EDTA treatment leads to a substantial (but not full) decrease in operator occupancy (approximately twofold) for both WT and C84S, consistent with the observed derepression. Further, the difference in repression efficiency in LB medium noted for C84S Zur (Fig. 3b) was also reflected in the promoter occupancy levels (Fig. 3c, lane 4). As expected, the H124A Zur (dimerization defective) had decreased operator occupancy under most conditions (but could be rescued by excess zinc), and the C98S Zur (disrupted in the structural Zn-binding site essential for protein folding and dimerization) was essentially inactive in this operator binding in this assay. Stepwise Zn(II) binding and operator discrimination. The above results are consistent with the hypothesis that decreasing zinc availability leads to a decrease in fully metallated Zur (Zur 2 :Zn 4 ) and that this leads to a selective induction of early and middle genes, whereas the remaining Zur 2 :Zn 3 population may suffice for repression of late genes. To further explore the basis for this graded response, we sought to determine the biochemical affinity of Zur proteins for each operator. All four Zur proteins (WT, C84S, C98S and H124A) were purified after overproduction in Escherichia coli and biochemically characterized using assays to monitor purity, dimerization and Zn content as described previously for Zur and other Fur family proteins 3,21,22 ( Supplementary Fig. 4). Next, the DNA-binding affinity was monitored for each active protein (the C98S protein was inactive in binding DNA) at the six operator sites using an electrophoretic mobility shift assay (EMSA) optimized with WT Zur ( Supplementary Figs 5 and 6). The results, summarized in Table 1, indicate that WT Zur bound to all six operator sites with similar affinities (K d values of B12-25 nM) that differed by no more than approximately twofold. The H124A Zur protein, which was very poorly active as a repressor in vivo (Fig. 2), bound with much reduced affinity (K d values B100 nM). Of note, the C84S mutant protein was essentially unaffected in its ability to bind to the operators of the two late genes, folEB and rpsNB, but had reduced affinity for the other four operator sites. These findings support the inference that declining cellular zinc levels lead to an initial transition of Zur 2 :Zn 4 to Zur 2 :Zn 3 , and the residual activity of the Zur 2 :Zn 3 form can account for the comparatively slow induction of late genes. However, the operator affinities in this assay do not provide an explanation for the difference in sensitivity between early and middle genes. We note that early and middle genes clearly differ in their operator occupancy as a function of zinc depletion in vivo (Fig. 2), yet do not detect any difference in operator affinity in vitro that can account for these differences, at least under fully zinc sufficient conditions (Table 1). Stepwise repression is independent of local chromosomal context. Although our working model is that the temporal induction of genes in the Zur regulon is due exclusively to binding of Zur to the promoter proximal operator site(s), we cannot exclude the possibility that other factors might influence their regulation and, specifically, their sensitivity of zinc depletion. For example, it is increasingly appreciated that chromosome position can affect gene function in bacteria 23 , and regulators and nucleoidassociated proteins can function over extended distances to impact regulation. To determine whether the local sequence context of each operon is sufficient to determine its sensitivity to zinc-mediated repression, we generated transcriptional fusions of representative early, middle and late promoters to fluorescent protein reporters and integrated these at three distinct sites around the chromosome. In both strains tested, the reporter fusions retained their relative sensitivity to zinc depletion (Fig. 4). This observation indicates that chromosome context is not determinative, and further reveals that all information needed for the stepwise induction of the Zur regulon is contained within the B500 bp regions used for fusion construction. Discussion The observation that Zur-regulated genes are derepressed in a distinct temporal order as cells transition from sufficiency to deficiency provides new insights into the molecular logic of the zinc deprivation response (Fig. 5). Our results indicate that as intracellular zinc availability declines the first response is the mobilization of zinc from surface-associated, zinc-associated ribosomal proteins. Both L31 and L33 have been previously implicated as part of a widely distributed but underappreciated mechanism of zinc storage 7,14,16,17,24 . Just as cells may store iron in times of sufficiency within ferritin and ferritin-like proteins, many bacteria appear to use small zinc-binding proteins that associate with the surface of the ribosome as a mobilizable Zn(II) store 15 . These storage proteins are notable since they are small peptides (o50 amino acids in length) that coordinate Zn(II) through two Cys-X-X-Cys motifs. In many organisms, there are paralogues of nearly identical sequence that lack many or all of these cysteine ligands and these non-Zn(II) binding paralogues are nearly always regulated by Zur or analogous sensors of zinc deprivation 15 . Biochemical studies reveal that these paralogous ribosomal proteins (for example, L31) can displace the endogenous ribosomal protein from the surface of the ribosome to mobilize the associated Zn(II) 16 . Genetic and physiological studies support a model in which this mobilized Zn(II) is important for growth under conditions of zinc limitation 7 . It is presently unknown whether Zn(II) exchange from the displaced L31 is spontaneous or whether it is facilitated by specific low-molecular-weight ligands or protein chaperones or by proteolysis. Regardless of mechanism, due to the high abundance of ribosomes in bacterial cells (B2-6 Â 10 4 per cell depending on growth rate), it can be calculated that ribosomally associated Zn(II) represents a substantial fraction of total cell-associated Zn(II). The other protein induced as part of the early response to Zn(II) deprivation is ZinT. ZinT is a surface-associated lipoprotein hypothesized to function as an accessory factor for Zn(II) import [10][11][12] . It is somewhat surprising that ZinT would be induced before the ZnuACB uptake system with which it presumably functions. It may be that ZinT primes the cell for Zn(II) import by scavenging Zn(II), or ZinT may function with other transporters. As cellular zinc levels decline further, and Zn(II) mobilized from the ribosomal pool is consumed by continued protein synthesis, the middle genes are induced. These include the high-affinity Zn(II) ABC transporter, ZnuACB, and the putative metallochaperone YciC. The function of YciC is presently unclear, but this protein is representative of a highly conserved family of GTPases implicated as cofactors for metal insertion into target metalloproteins 13 . YciC may be involved in the intracellular trafficking of Zn(II), which likely becomes more critical when zinc is limiting for growth. One possible role of YciC is to prioritize Zn(II) utilization by making sure that Zn(II) is delivered to those enzymes essential for growth. An analogous problem arises with respect to iron deprivation. In this case, B. subtilis activates an iron-sparing response involving the small RNA FsrA and accessory proteins FbpABC 25 . This system translationally represses numerous abundant iron-containing enzymes and complexes including succinate dehydrogenase, glutamate synthase and lactate dehydrogenase with the presumed function of enabling the limiting amounts of iron to be available for incorporation to more essential enzymes. An analogous iron-sparing response is present in E. coli mediated by the RyhB sRNA 26,27 . There is, to date, no evidence for an analogous zinc-sparing response in B. subtilis, and we speculate that one role of YciC may be to help the cell prioritize the allocation of limited zinc. However, zinc-sparing responses are present in other organisms: one well-documented example is in the model eukaryote Saccharomyces cerevisiae where abundant Zn(II)-containing alcohol dehydrogenases are downregulated in response to zinc limitation 28 . The ability of the ZnuACB system to overcome Zn(II) limitation relies, of course, on the availability of extracellular zinc. As zinc levels fall further, the final two genes are induced, rpsNB and folEB. Repression of these genes is maintained at intermediate levels of intracellular Zn(II) by their unique ability to be efficiently repressed in vivo by the Zur 2 :Zn 3 form of Zur. The rpsNB gene encodes an S14 paralogue (S14). Unlike the L31 and L33 proteins, which are dispensable, S14 is essential and is required at an early stage of ribosome biogenesis. Thus, if cellular Zn(II) levels decline to levels that no longer support Zn(II) acquisition by S14 cells will be unable to assemble new ribosomes. The induction of rpsNB allows the synthesis of a replacement, Zn-independent S14 protein that enables new ribosome synthesis (a 'failsafe' pathway) 18 . Mutant strains lacking rpsNB can still grow under severe zinc limitation, but their inability to synthesize new ribosomes results in linear rather than exponential growth 7 , consistent with the prediction that each new daughter cell inherits only half the ribosome complement of its parent. Like rpsNB, folEB encodes a replacement function for a protein that fails when zinc levels are low. The FolE1 enzyme (product of folEA) encodes a Zn(II)-dependent GTP cyclohydrolase required for folate biosynthesis. The folEB product is a non-orthologous replacement that allows continued folate synthesis even under conditions of severe zinc depletion 19 . In summary, our results indicate the transition from zinc sufficiency to zinc deprivation derepresses (i) early proteins for the mobilization Zn(II) from the surface of the ribosome (L31* and L33) and to prime the cell for zinc uptake (ZinT), (ii) middle proteins for high-affinity zinc import (ZnuACB) and an accompanying metallochaperone (YciC), and (iii) late proteins to replace critical functions that fail as zinc levels decline further, RpsNB and FolE2. Numerous other sites in B. subtilis are thought to be associated with Zur in vivo, as judged by ChIP, but the significance of this extended regulon is not clear, and many sites did not appear to significantly affect gene expression 9 . A graded response to zinc deprivation is also likely to occur in other bacteria. Indeed, a graded response of Zur has been previously documented in Streptomyces coelicolor 29 , although the molecular basis for this response, and its physiological implications, were not clear. Similarly, the E. coli Zur regulon contains operator sites of widely varying affinities and this has been correlated with the overall magnitude of the derepression response but not yet with a defined temporal order of induction 30 . It is likely that graded responses are controlled by a combination of factors that ensure that genes are derepressed in an optimal order. Here we have documented the role of negative cooperativity in zinc binding to the two regulatory sites in each dimer in helping distinguish early and middle genes from late genes. However, our results have not yet provided a mechanism to explain the transition from early to middle genes, although this transition is also correlated with Zur operator occupancy (Fig. 2). In the case of some other Fur homologues (including B. subtilis Fur 31 , S. coelicolor Zur 29 and Magnetospirillum gryphiswaldense Fur 32 ) a second regulatory site within each monomer (site 3) may serve to fine-tune operator binding, with some sites requiring occupancy of this additional site and others not. An added level of complexity results from cooperativity in binding of multiple dimers to a single regulatory region. Although Fur proteins were originally proposed to bind to a 19 bp Fur box recognition sequence 33 , subsequent studies demonstrated that each a minimal Fur-binding site is a 7-1-7 inverted repeat and the classic 19 bp Fur box represents two overlapping repeats that can bind a dimer of dimers 34 . Other Fur family members likely share a similar architecture, at least at some operators 4 . Recent structural studies have highlighted the role of electrostatic interactions between the two dimers in cooperative binding by E. coli Zur 30 and M. gryphiswaldense Fur 32 . Thus, another possible mechanism to impart a preferred order for derepression is a variable number of dimers or a variable level of cooperativity between dimers at a particular operator region. Indeed, in the B. subtilis Fur regulon, some operons are regulated by a single 7-1-7 repeat and bind a single Fur dimer, whereas other operons, such as dhb, are tightly repressed by a dimer of dimers 34,35 . The idea of a graded response occurring as a function of increasing levels of a functional transcription factor is well precedented, but in only a minority of cases is the underlying logic apparent. One notable example is the regulation of gene expression in response to declining nutrient availability by the Spo0A transcription factor in B. subtilis. This is a complex system, involving multiple kinases, phosphotransfer proteins and phosphatases that integrates a variety of stress signals. As the level of the active Spo0ABP transcription factor increases, cells are proposed to sequentially activate a motility response (sliding), biofilm formation and ultimately they enter into sporulation 36 . In this and related systems, a remaining challenge is to understand the molecular basis of this sequential gene regulation that may involve different forms of the transcription factor and variable affinities of activated factor for its operator sites, combinatorial effects with other regulators and complex promoter architectures. Methods Bacterial strains and culture conditions. All B. subtilis strains used in this study were isogenic with common laboratory strains listed in Supplementary Table 1. B. subtilis CU1065 was grown on LB medium and modified glucose minimal medium (20 g l À 1 (NH 4 ) 2 SO 4 , 183 g l À 1 K 2 HPO 4 3H 2 O, 60 g l À 1 KH 2 PO 4 , 2 g l À 1 MgSO 4 *7H 2 O, 10 g l À 1 sodium citrate, 0.5% glucose, 0.5 mM CaCl 2 and 5 mM MnCl 2 ) was used for PY79 at 37°C. When appropriate, antibiotics were included at: 100 mg ml À 1 spectinomycin, 5 mg ml À 1 chloramphenicol, 10 mg ml À 1 kanamycin, 5 mg ml À 1 tetracyclin and 1 mg ml À 1 erythromycin plus 25 mg ml À 1 lincomycin for the selection of various B. subtilis strains. DNA was transformed into B. subtilis using a modified version of a previously published protocol 37 . E. coli DH5a was used for routine DNA cloning 38 . Unless indicated otherwise, liquid media were inoculated from an overnight pre-culture and incubated at 37°C with shaking at 200 r.p.m. Preparation of total RNA. Total RNA was isolated from B. subtilis strains that were cultured to mid-logarithmic phase (at an OD 600 of 0.4-0.5) in LB medium. For Zn(II)-depleted condition, various amounts of TPEN were treated for 5 or 40 min. Total RNA was extracted by the 'hot phenol method' as described 39 . The total amount of RNA and its quality were measured by absorbance spectroscopy and confirmed by resolving RNA samples on 1.3% formaldehyde agarose gels. S1 nuclease mapping analysis. Gene-specific DNA oligonucleotide probes for zur, zinT, znuA, yciC, folEB, rpsNB and rpmEB transcripts were used for PCR amplification using B. subtilis wild-type genomic DNA as template. The appropriate primer pairs are listed in Supplementary Table 2. An amount of 100 mg of total RNA was pelleted and lyophilized. Each specific DNA probe was radiolabelled with (g-32 P) ATP and T4 polynucleotide kinase, and 30,000-40,000 c.p.m. of labelled probe was used in each reaction. The total RNA pellet was carefully resuspended in 20 ml hybridization buffer (40 mM PIPES (pH 6.4), 400 mM NaCl, 1 mM EDTA, 80% (v/v) formamide). Individual samples were incubated at 95°C for 25 min and slow cooled to 42°C. Following incubation overnight, 300 ml of S1 nuclease mix containing 100 units of S1 nuclease in S1 nuclease buffer (280 mM NaCl, 30 mM NaOAc (pH 4.4), 4.5 mM ZnOAc) was added and incubated at 37°C for 45 min. The reaction was terminated by addition of 75 ml of S1 nuclease termination solution (2.5 M NH 4 OAc, 0.05 M EDTA). The DNA-RNA hybrid was precipitated by adding 400 ml of isopropanol and the pellet was washed with 70% (v/v) ethanol, vacuum dried and resuspended in 13 ml alkaline loading dye. The protected DNA fragments were then resolved by 6% (wt/vol) polyacrylamide gels containing 7 M urea. The dried gels were exposed to a phosphor imaging screen (Typhoon FLA 7000; GE) and bands were quantified using Multi Gauge V3.0 (Fuji). Quantitative in vivo crosslinking and immunoprecipitation. In vivo crosslinking of DNA to Zur proteins and subsequent immunoprecipitation with specific antibody to FLAG were carried out. Cells were grown in LB medium to mid-logarithmic phase (at an OD 600 of 0.5-0.6) and each 30 ml aliquots were spun down and the pellets were saved at À 80°C. For crosslinking, pellets were resuspended in buffer CA (10 mM Na 2 HPO 4 , 2 mM KH 2 PO 4 , 137 mM NaCl and 2.7 mM KCl (pH 7.4)) and 30 ml of 34% formaldehyde was added and the samples were gently rocked at room temperature for 10 min. To quench the crosslinking, 133 ml of 1 M glycine pH 7.5 (final molar concentration 0.133 M) was added, and the cells were shaken gently at 4°C for 30 min. Cells were collected by centrifugation, and the pellets were resuspended and washed twice with buffer CB (50 mM Tris-HCl (pH7.4), 150 mM NaCl and 1 mM EDTA). Washed cells were resuspended in 0.5 ml buffer CB and lysed by sonication. Total 400 ml of supernatant was collected after centrifugation and frozen at À 80°C after aliquot. A measure of 1 ml of the lysate was diluted with 9 ml of the dilution buffer and the mixture was saved at À 80°C, to serve as the input-control (1% of input DNA). For immunoprecipitation, a-FLAG M2 magnetic agarose beads (M8823-Sigma) were washed and resuspended in 400 ml buffer CB. A measure of 200 ml cell lysate (total 200 mg of proteins) was mixed and incubated at 4°C for overnight on rotation mixer. The bead slurry was recovered by using the magnetic stand for 1 min standing and washed twice with 400 ml buffer CB. The Zur-DNA complexes were eluted from the beads by addition of 0.1 M glycine-HCl (pH 3.0) and neutralized by buffer CC (500 mM Tris-HCl (pH 8.0) and 1.5 M NaCl). The input-control samples were also diluted with the elution buffer to the same final volume as the samples. Enriched target DNAs were purified using PCR Clean-Up Kits (Thermo Fisher Scientific) as per the manufacturer's instructions. Each 1 ml volume of eluted DNAs was quantified using the qPCR kit (Bio-Rad) followed by the manufacturer's instructions with primer pairs of zur, zinT, znuA, yciC, folEB, rpsNB, rpmEB and 16S rRNA in Supplementary Table 2. The data were normalized to values from 1% input. Resuspension experiments. BsZur variants with a C-terminal FLAG-tag were integrated at the amyE locus of CU1065 zur::tet (1). The strains were grown in LB medium to mid-logarithmic phase (at an OD 600 0.5) with appropriate antibiotics. Each 40 ml from cell culture was spoon down and saved before and after 2 mM EDTA treatment for 1 h for the nontreated control sample or Zn(II)-depleted cells. Remained cell culture was collected and the pellet was washed three times with equal volume of chelated LB medium. Resuspended cells with an equal volume of pre-warmed fresh LB medium containing 25 mM ZnSO 4 were further incubated for another 1 h before harvesting. Structure modelling. Modelled three-dimensional structure of BsuZur dimer was constructed after sequence alignment of the structure of the closely related Zur protein of S. coelicolor (PDB ID 3MWM) 29 Electrophoretic mobility shift assays. Each Zur target promoter DNA probes of B45 bp containing Zur-binding sites were isolated using crush and soaking method from the polyacrylamide gel 41 after annealing with each primer pairs in Supplementary Table 2. The purified DNAs were labelled at 5 0 -ends with (g-32 P) ATP using T4 polynucleotide kinase. Binding reactions were performed with B1 fmol of labelled DNA fragments and 0.075-616 nM of purified Zur proteins in 20 ml of the reaction buffer (20 mM Tris-HCl (pH 6.4), 50 mM KCl, 1 mM DTT, 0.1 mg of bovine serum albumin per ml, 5% glycerol and 0.1 mg of poly(dI-dC), with 0.1 mM ZnSO 4 ). Following incubation at room temperature for 20 min, the binding mixture was subjected to electrophoresis at 4°C on a 5% polyacrylamide gel in TA (pH 6.4) buffer. After electrophoresis, the dried gels were exposed and quantified by a phosphor image analyser (Typhoon FLA 7000). A band intensity of unbound DNA probes was measured against Zur concentration using Multi Gauge V3.0 software. Digitalized data were fit to binding curves through SigmaPlot 2001 program (SPSS Inc.). Apparent K d values, corresponding to the concentration of variables (Zur) at half-maximal upshift of DNA probes, were determined from at least three independent sets of experiments. Construction of fluorescence fusion strains. For construction of reporter fusions between the representative Zur target promoters corresponding to groups with differential Zn(II) sensitivity and the three different fluorescent protein encoding genes (ecfp, egfp and mCherry), each promoter region was amplified by PCR using DNA oligonucleotides shown in Supplementary Table 2. To integrate at different loci, the PCR products were digested with XmaI restriction enzyme and ligated into three different integration vectors digested by same restriction enzyme. Constructs were confirmed by Sanger sequencing and transformed into B. subtilis PY79. For selection of correctly integrated strains, appropriate antibiotics were used and candidates were verified using diagnostic genomic DNA PCR and Sanger sequencing of PCR products. Fluorescence spectroscopy. For measurement of fluorescence intensity from reporter fusion strains, fresh cultured cells were used to inoculate 5 ml of modified glucose minimal media. That was cultured overnight while shaking at 37°C. An aliquot was then diluted 1:100 in 100 ml glucose minimal media supplemented as necessary with antibiotics. These cells were incubated shaking at 37°C until reaching an OD 600 B0.3, whereupon they were pelleted after exposure of different concentration of TPEN from 10 to 100 mM, washed with two volume of 1 Â PBS buffer (pH 7.4) before fixation. And then, the cell pellet was soaking in the 4% paraformaldehyde buffer for 20 min at room temperature 42 . After that, fixed cells were washed twice and resuspended in 5 ml 1 Â PBS buffer (pH 7.4). OD 600 readings were taken of the resuspended cells, and each cell sample was diluted to an OD 600 of 0.2 with 1 Â PBS buffer (pH 7.4). Fluorescence intensity of each 1 ml sample was taken with a Perkin-Elmer LS55 luminescence spectrometer. Western blot analysis of BsZur-FLAG. Cells were prepared under the same condition, as described in Fig. 2 for Supplementary Fig. 3d or Fig. 3b for Supplementary Fig. 3e. The cell pellet was resuspended in 0.5 ml of PBS buffer and lysed by sonication. Cell debris were removed by centrifugation and the resulting supernatant was mixed with SDS-PAGE loading buffer, and boiled at 95°C for 10 min before being resolved by 13% SDS-PAGE. The proteins were transferred to a membrane at 60 mA for 40 min. The membrane was then blocked with blocking solution (Dry Milk dissolved in 20 mM Tris-HCl (pH 7.8), 150 mM NaCl and 0.1% Triton X-100) overnight. The membrane was incubated with a 1:1,000 dilution of polyclonal anti-FLAG Antibody (Sigma Chemical Co., SAB4301135)f or 1 h, washed with TBST (20 mM Tris, 150 mM NaCl and 0.1% Triton X-100), and incubated with 1:5,000 dilution of anti-rabbit IgG secondary antibody conjugated with alkaline phosphatase antibody (Santa Cruz Biotech, SC-2004) for 1 h. The membrane was then developed with 5 ml AP buffer (100 mM Tris-HCl (pH 9.5), 100 mM NaCl and 5 mM MgCl 2 ), 1:100 dilution of NBT, and BCIP. The Zur-FLAG protein has a molecular weight of B17 kDa, in agreement with the observed mobility. Data availability. The authors declare that the data supporting the findings of this study are available within the article and its Supplementary Information files, or from the corresponding author on request.	8,545.6	2016-08-26T00:00:00.000	[ "Biology" ]
Insights on entanglement entropy in 1 + 1 dimensional causal sets Entanglement entropy in causal sets offers a fundamentally covariant characterisation of quantum field degrees of freedom. A known result in this context is that the degrees of freedom consist of a number of contributions that have continuum-like analogues, in addition to a number of contributions that do not. The latter exhibit features below the discreteness scale and are excluded from the entanglement entropy using a ‘truncation scheme’. This truncation is necessary to recover the standard spatial area law of entanglement entropy. In this paper we build on previous work on the entanglement entropy of a massless scalar field on a causal set approximated by a 1 + 1D causal diamond in Minkowski spacetime. We present new insights into the truncated contributions, including evidence that they behave as fluctuations and encode features specific to a particular causal set sprinkling. We extend previous results in the massless theory to include Rényi entropies and include new results for the massive theory. We also discuss the implications of our work for the treatment of entanglement entropy in causal sets in more general settings. Introduction An important open question that quantum gravity aims to answer is: what are the microscopic degrees of freedom or states that describe black hole entropy?We know that the classical Bekenstein-Hawking black hole entropy is finite and proportional to the area of the event horizon [1][2][3].The proportionality constant is also known and is 1/4 when some fundamental constants are set to 1. Whatever the microscopic nature of this entropy is and however many states there are then, they too must be finite, and yield an entropy equal to 1/4 of the event horizon area in the same fundamental units.A similar statement applies to other horizons, such as cosmological ones, with analogous thermodynamic properties. Entanglement entropy of quantum fields is one promising candidate microscopic source of this entropy.In fact, while entanglement entropy has a wide array of applications today, it was originally conceived of to study this very question of the microscopic nature of black hole entropy [4].From its earliest days [4,5] it was noticed that entanglement entropy also generically scales as the area of the boundary of the entangling subregion, be this the event horizon in a black hole spacetime or more generally some surface dividing the entire quantum system into two complementary subsystems.The proportionality constant in these so-called spatial "area laws" is not fixed, and depends on the details of the UV physics.Another crucial observation made early on in these studies was that the entanglement entropy diverges in the absence of a UV cutoff.Thus a UV cutoff is necessary to obtain a finite entanglement entropy and for it to have a chance to be the correct microscopic picture of black hole entropy. If this UV cutoff is furnished by a theory of quantum gravity such as causal set theory, this would bring us closer to understanding the horizon degrees of freedom fundamentally.Moreover, if we wish to understand the 1/4 proportionality constant in addition to the area scaling, then our fundamental theory (as well as our entanglement entropy definition) must be covariant.The study of entanglement entropy in causal sets allows one to do exactly this.Causal set theory is equipped with a covariant discreteness scale which serves as the UV cutoff.Furthermore, a definition due to Sorkin [6] of entanglement entropy in terms of spacetime correlation functions, makes possible the use of this covariant cutoff in counting quantum field degrees of freedom.Hence it is of interest to study this subject further. The most well understood calculation of entanglement entropy in causal set theory is in the context of a Gaussian massless scalar field in a 1 + 1D causal set approximated by a causal diamond in Minkowski spacetime [7].The entanglement was considered between the field restriction to a smaller subdiamond and its complement in the larger one in which the global pure state was defined.Some challenges were met along the way, but with insights from analytic calculations in the continuum [8][9][10][11][12], these were overcome and the expected area laws of the entanglement entropy were obtained.The challenges stemmed from a number of unexpected, and dominating, contributions to the entanglement entropy. A key ingredient in the calculation, are the eigenfunctions and eigenvalues of the Pauli-Jordan function or spacetime commutator.Previous work in the 1 + 1D continuum diamond [8,10] has shown that the eigenfunctions are (approximately) linear combinations of plane waves and the eigenvalues are inversely proportional to the wavenumbers which have known values following a power law.In the causal set we also see discrete analogues of these plane wave eigenfunctions and their eigenvalues, all the way down to wavelengths near the discreteness scale.The contributions do not stop here, however.Instead, we also observe numerous eigenfunctions that do not look like plane waves and have eigenvalues that do not follow a power law.This latter set of contributions is poorly understood and it is an open question whether or not they have any physical significance.As first discovered in [7], this family of contributions must be excluded from the entanglement entropy calculations in order to recover the conventional area laws.The exclusion is done via a "truncation scheme", whereby these solutions are eliminated at two stages of the calculation.Without the truncations, a spacetime volume law is obtained. Attempts have been made to extend the results of [7] to more general settings such as higher dimensions, nonlocal field theories, disjoint regions, and spacetimes with curvature [13][14][15].These attempts have had various degrees of success.What these studies have shown is that a better understanding of the extra contributions in the causal set, either through closer investigations in the causal set or through further analytic studies in the continuum, is necessary to motivate unique generalisations of the truncation scheme. In this work, we return to the best understood case of the 1 + 1D causal diamond, with the intention to gain further insights from the causal set.We study more closely the eigenfunctions that do not have continuum-like counterparts and ask whether they are in some sense fluctuations below the discreteness scale.We find evidence in favor of this.Specifically, we consider averages of the eigenfunctions over many sprinklings.We find that the continuum-like contributions have persistent features while the others do not.The details of this are presented in Section 5. The plan of the remainder of the paper is as follows.We begin by reviewing the spacetime entanglement entropy formalism, as well as the expected scaling laws in 1 + 1D that we will compare to, in Section 2. In Section 3 we reproduce the results of the massless theory, and include new results on Rényi entropies.We then discuss the massive theory in Section 4. The extension to the massive theory is nontrivial, as we lack analytic results there.Upon close inspection of the Pauli-Jordan function eigenvalues and eigenfunctions in this case, we find that the knowledge from the massless theory is adequate to employ a meaningful truncation here.This is encouraging and has implications for more general extensions of these and similar results.These implications are discussed in Section 6. Entanglement Entropy Many different techniques exist for computing entanglement entropy.Below we review the spacetime formulation of entanglement entropy for a Gaussian scalar field theory, defined in [6], which we use throughout this paper.This formulation is special because it is in terms of spacetime correlation functions and hence allows for a covariant treatment of the degrees of freedom and UV cutoff. Spacetime Definition of Entropy We will consider a Gaussian scalar field theory.Therefore, everything, including the entanglement entropy, can be determined from the two-point correlation function or Wightman function.‡ The entropy associated to the scalar field in a spacetime region R (where R can be a full spacetime or causal set, or a subset of it) is ‡ Interestingly, it has been shown that the Wightman function is also enough to determine the entanglement entropy in non-Gaussian and interacting theories up to first order in perturbation theory [16]. where λ is a solution to under the condition ∆v = 0. W is the Wightman function W (x, x ) = 0\| φ(x)φ(x ) \|0 and i∆, which is equal to one half of the imaginary part of W , is the Pauli-Jordan function or spacetime commutator i∆(x, x ) = [φ(x), φ(x )], where x, x ∈ R.That the expression (1) agrees with the standard S(ρ) = −Tr ρ lnρ can be readily seen from its derivation via the replica trick, given in [16], which we review in Appendix A. Analogously, the Rényi entropy of order α is where each term in the sum (4) accounts for a pair of eigenvalues λ and (1 − λ). If the global state is taken to be the SJ state, W = W SJ , we have a pure state and the entropy vanishes.The SJ Wightman function § is defined to be the restriction of i∆ to its positive eigenspace, W SJ ≡ Pos(i∆).This results in all the eigenvalues in (1) being either λ = 0 or λ = 1 and the entropy vanishing, as mentioned.If, however, W and i∆ are restricted to subsets of the full degrees of freedom or subregions within R, we can obtain a non-zero entropy.In particular, this occurs if the subregion has a non-zero causal complement, and the entropy in this case can be regarded as the entanglement entropy between the subregion and its causal complement. As mentioned above, entanglement entropy diverges in the absence of a UV cutoff.Therefore to obtain a meaningful entanglement entropy, we need a UV regulator or fundamental cutoff.The kind of cutoff that can be applied is also dependent on the manner in which the entanglement entropy is calculated.For example, when the calculation is done on a spatial hypersurface (as it often is), common choices of cutoffs include a minimum spatial distance from the entangling boundary or a spatial lattice spacing.The formulation presented in this subsection allows for the implementation of a spacetime cutoff.W and i∆ are both spacetime functions, though infinite dimensional in the continuum.One possible way to implement a covariant spacetime cutoff, as was done in [11], is to expand W and i∆ in the eigenbasis of i∆ up to a minimum eigenvalue which can be related to a minimum wavelength or maximum wavenumber and defined as the cutoff.In the causal set, as we will see below, we are automatically equipped with the discreteness scale as our cutoff, and both W and i∆ are finite dimensional matrices.§ We will review the SJ state in greater detail in the next section. The closure of the image of the i∆, or its eigenfunctions with non-zero eigenvalues, span the full solution space of the Klein Gordon equation [17]. Scaling Laws in 1 + 1D We will study both a massless and massive scalar field theory in 1 + 1D.We will compare our results for the entanglement and Rényi entropies with known scaling laws that have been derived and numerically verified in various places in the literature.Below we summarise the relevant scalings for later reference.Most of these scaling laws have been studied in settings where the entanglement entropy is calculated on a spatial hypersurface.These results can be compared to those in our spacetime formulation, if the spacetime regions considered are domains of dependence (or causally convex subsets of the domains of dependence, containing the Cauchy surface) of the spatial regions.It is worth highlighting that the spacetime formulation reviewed in the previous subsection does not require for the spacetime or subregions to be domains of dependence of spatial Cauchy hypersurfaces.This merely facilitates things such as comparisons to conventional results. The entanglement entropy of a massless scalar field confined to a spatial interval of length ˜ within a larger spatial interval of length L >> ˜ , is [18] S ∼ 1 3 ln where a is the UV cutoff, and c 1 is a non-universal constant.This scaling is for the case where the subinterval ˜ has two boundaries across which the entanglement occurs and in the limit of small UV cutoff a. Similarly, the Rényi entanglement entropy of order α is [18] where c α are non-universal constants. The entanglement entropy of a massive scalar field with mass m, where a where c1 is once again a non-universal constant.Note that these logarithmic scalings are consistent with the general arguments that lead to area laws for the entanglement entropy [19].The scalings ( 5) and ( 6) have been obtained in the continuum using the spacetime formulation of the previous subsection in [11].The scaling (5) has been obtained in the causal set using the same formulation, in [7,12].Below we present results confirming all of these scalings in the causal set. Massless Scalar Field Theory A Gaussian scalar field theory on a causal set can be set up using the Sorkin-Johnston (SJ) prescription [8,17,20].This prescription uses the entire spacetime and yields a unique and covariant Wightman function, the SJ Wightman function W SJ .This Wightman function can then be used in (2) to obtain the entanglement entropy.The starting point of this prescription is the retarded Green function ¶, which for a scalar field with mass m in D-dimensional Minkowski spacetime satisfies where x = (t, x) and G (D) R (x, x ) = 0 unless x causally precedes x, denoted by x ≺ x [21,22].For a massless scalar field in 1 + 1D Minkowski spacetime, this Green function has the simple form where θ is the Heaviside step function and τ = (∆ x) 2 − (∆t) 2 is the proper time.In other words, this Green function is only non-zero if x is in the past lightcone of x and takes the constant value of 1/2 if this is the case.The causal matrix C can be used to define a discrete analogue of this Green function in the causal set.The causal set retarded Green function is where C is the causal matrix The causal set Green function (10) agrees identically with the continuum Green function (9) evaluated at the causal set elements.The spacetime field commutator, or Pauli-Jordan function, is then given by where G A , the advanced Green function, is simply the transpose of G R .i∆ is anti-symmetric and Hermitian and has real non-zero pairs of eigenvalues, ± λi and eigenvectors given by where λi > 0. i∆ can then be expanded in its non-zero eigenbasis as ¶ The SJ prescription defines a unique W SJ in globally hyperbolic spacetimes (or causal sets approximated by globally hyperbolic spacetimes), since only these spacetimes possess a unique retarded Green function. Restricting only to the positive eigenspace of i∆, the SJ Wightman function is Once we have W , we have all we need to compute the entropy using ( 1) and ( 2).We will first review the results of [7] and the truncation scheme, before presenting new results.Throughout this paper, we will work with a causal set approximated by a causal diamond in 1 + 1D Minkowski spacetime, and we will consider the entanglement entropy arising from the restriction of W SJ to a smaller subdiamond concentric to the larger one.These regions are shown in Figure 1, with the subdiamond in red.The properties of W SJ and the eigenspectrum of i∆ in these causal diamonds has been well studied [9][10][11].In the massless theory, working with a compact region such as a causal diamond cures the IR divergence otherwise present in 1 + 1D.In addition, the causal diamond is globally hyperbolic (hence there will be a unique G R ). Its eigenfunctions f in the continuum satisfy The full set of functions with non-zero eigenvalues satisfying ( 17) is known and form a two-set family of eigenfunctions, f k and g k .These are where K = {k ∈ R\|tan(kL) = 2kL and k = 0} [7,8].The eigenvalues for both sets are Similarly in the causal set, we see discrete analogues of these eigenfunctions and eigenvalues, up to wavenumbers comparable with the discreteness scale.Figure 2 shows a comparison of the causal set and continuum eigenvalues of i∆ for a 2000 element causal set.+ As evident in the figure, the larger eigenvalues in the causal set and continuum agree well with one another, but at some point, which corresponds to wavenumbers or wavelengths around the discreteness scale (horizontal dashed line in the figure), the causal set eigenvalues start to deviate from a powerlaw.The smaller causal set eigenvalues, appearing as the trend curves down, no longer have continuumlike counterparts.In Section 5 we will further discuss the nature of the causal set eigenfunctions corresponding to these small eigenvalues.2 .The horizontal dashed line corresponds to λmin , where the wavenumber in the eigenvalue (19) is comparable to the discreteness scale. + The eigenvalues in the continuum have dimensions of area while in the causal set they are dimensionless.For the comparison, we have rescaled the causal set eigenvalues using the appropriate density factor. It has been shown [7,12] that the inclusion of the contribution of these small eigenvalues and their eigenfunctions, which do not have continuum-like counterparts, generically leads to spacetime volume laws rather than the conventional spatial area laws for the entanglement entropy.Since a main motivation of the study of entanglement entropy is to understand black hole entropy, it is of great interest to understand if and how the conventional area laws can be recovered.It has been shown that the area law is recovered after the exclusion of these components from the calculations, using a truncation scheme, which we discuss next. A Covariant Truncation Scheme In a causal set * with density ρ = N/V , where N is the mean number of elements sprinkled into a spacetime volume V , the minimum distance that is meaningful is of the order of the discreteness scale, 1 √ ρ in 1 + 1D.All structures below this scale do not have any physical reality in the causal set.In the present context, knowing what we know about the spectrum of i∆ and its relation to a wavenumber, we can argue for a maximum meaningful wavenumber (or minimum wavelength) and therefore minimum meaningful eigenvalue in the causal set. If, as mentioned already, we recognise that the minimum possible wavelength on the causal set, Λ min , is given by the discreteness scale we find that the maximum wavenumber supported on the causal set is Substituting this into the expression for the eigenvalues (19), we find that the minimum magnitude of an eigenvalue that is meaningful in the causal set is where the superscript cont is a reminder that we are expressing the eigenvalues in their dimensionful form, as they appear in the continuum.The eigenvalues in the continuum have dimensions of length squared, while those in the causal set are dimensionless.The two are related to one another through multiplication by a factor of ρ.Therefore, in the causal set, we should aim to retain a minimum (in magnitude) eigenvalue The eigenvalues with magnitudes smaller than λ cs min correspond to features below the discreteness scale and should be excluded [7].This argument assumes that * For a review of causal set theory, see [23][24][25][26][27]. the eigenvalues beyond the bend in Figure 2 can be continued to be interpreted approximately as L/k, even though this is not guaranteed to be the case.In Section 5 we will provide further evidence that these contributions have features below the discreteness scale. Hence, when carrying out the entropy calculation using the formalism of Section 2, the spectrum of i∆ needs to be "truncated" such that the magnitude of its minimum eigenvalue is (23).This truncation is performed at two stages of the calculation.The first is when i∆ and W SJ are initially defined on the full causal set.The minimum eigenvalues that enter their definition must be consistent with (23).The second truncation is performed after the restriction of i∆ and W SJ to the subregion (the red subdiamond in Figure 1).The N in (23) corresponds to the number of elements in the region where the truncation is taking place.Hence N would be different in each of the two truncations: in the first truncation it would be the number of elements in the larger diamond N L and in the second truncation it would be the number of elements in the smaller subdiamond N .Due to the double application of the truncation, this scheme is sometimes referred to as a "double truncation".For more details on the double truncation, see [7,12]. As this truncation concerns the smallest scales in the theory, near the discreteness scale, the same truncation can be implemented in the massive scalar field theory provided that m 2 ρ.We will discuss this further in Section 4. It is an open question how to motivate a minimum eigenvalue of the spectrum of i∆ in general spacetimes, in the absence of analytic results telling us how to relate it to something like a wavenumber (such as in (19)).Our work below in the massive theory, where such analytic results are not available, gives some insight on potential generalisations. Entropy Results The causal set setup of our calculations is Figure 1.The UV cutoff a that we will study the entanglement entropy scalings with respect to is the discreteness scale a ≡ 1/ √ ρ.We will vary a by holding fixed the volumes of the causal diamonds ( /L = 1/2 and L = 1/2) and varying the number of elements sprinkled into them.We consider 130 sprinklings from 15000 to 40000 elements in increments of 1000 with 5 sprinklings per increment.The results for the entanglement entropy (1) and Rényi entropies (4) of order 2 to 5 are shown in Figures 3 and 4. Since the causal diamonds in Figure 1 are the domains of dependence of their spatial diameters, our results for the entanglement entropy and Rényi entropy can be compared to ( 5) and ( 6) respectively.In all our calculations we used (1) and ( 4) and applied the double truncation scheme described in the previous subsection. The results for the entanglement entropy versus /a are shown in Figure 3.In agreement with the results of [7] and (5), a logarithmic scaling with a coefficient consistent with the expected 1/3 is obtained.More precisely, the scaling coefficient is 0.332 ± 0.009 and the non-universal constant is 1.104 ± 0.037, where the uncertainties were calculated by taking the square root of the diagonal elements of the covariance matrix of the fit. In future work it would be interesting to study these non-universal constants in greater detail. Massive Scalar Field Theory The retarded Green function for a massive scalar field in 1+1D Minkowski spacetime is G (2) where J 0 is the zeroth order Bessel function of the first kind [28].Notice that there is a non-trivial dependence on the proper time, compared to the massless expression (9).The causal set retarded Green function for the massive theory is where m 2 ρ, and I is the identity matrix.The Pauli-Jordan function in the continuum theory is Due to presence of the Bessel function in (26), the integral eigenvalue problem (17) in a causal diamond is considerably more complicated in the massive theory.Instead of trying to solve for the eigenfunctions directly, we can extrapolate (18) to the massive case by introducing ω 2 = \|k\| 2 + m 2 .This analogy gives These eigenfunctions can then be empirically compared to those obtained directly from the causal set, by substituting in guess values of k, as shown in Figure 5 for a massive scalar field with m = 10 on a causal set sprinkling of 10000 elements with 2L = 1.The eigenfunction shown corresponds to the first antisymmetric eigenfunction and was thus compared to f k,m with a guess value of k = 7.However, in contrast to the massless case, it was found through numerical integration that the eigenvalue equation ( 17) is no longer fulfilled with the cos term included in the g family of eigenfunctions.One should also notice that ( 27) is not a solution of the massive Klein-Gordon equation due to the presence of the cos term.Hence, (27) should be modified to We also find that for "the right values of k", the left and right hand sides of the eigenvalue equation ( 17) are in almost complete agreement if we keep away from the left and right corners of the diamond, as shown in Figure 6.† † In this case, the eigenvalues, λk,m are † † This is also consistent with the result that in the infinite volume spacetime, the SJ state is the Minkowski vacuum and the positive (negative) eigenvalue SJ modes are linear combinations of positive (negative) frequency plane wave solutions to the Klein Gordon equation [29]. given by where the √ 2 factor can be understood simply by setting m = 0 in (28).While we were unable to find exact and closed form expressions for the eigendecomposition of i∆ in the massive theory, our approximate results are encouraging.The form of the eigenvalues λk,m , ( 29), tells us that as long as m 2 ρ, the massive theory's eigenvalues approach the massless theory's eigenvalues in the UV limit (as the eigenvalues become smaller).This is exactly the regime we need to understand in order to apply a meaningful truncation scheme for the entanglement entropy.Therefore, since the spectrum ( 29) approaches the massless one (19) in the UV, in the massive theory we can expect to retain the same minimum magnitude eigenvalue as in the massless theory.This is also illustrated in Figure 7, where the spectrum of i∆ for two massive theories (m = 5 and m = 10) are shown together with spectrum in the massless theory.The eigenvalues of i∆ are only significantly different for large λ and are essentially the same when λ ∼ λmin . For small values of k (or large λk,m ), the mass plays a significant role in determining the magnitude of λk,m .However, for larger values of k, unless the mass is itself of the same order as √ ρ (which would not be a well-defined theory in the causal set), Thus, as λmin corresponds to a UV limit, for m √ ρ, we have that where λmin,m corresponds to the minimum eigenvalue to retain in the causal set truncations in the massive case.This is what we use in our calculations below. Entropy Results The setup of our calculations in the massive theory is the same as in Section 3.2.As motivated in the previous subsection, we use the same minimum eigenvalue truncation scheme as in the massless theory.First, we hold the mass fixed (m = 5) and vary a. Once the truncations are applied, the scaling with respect to the UV cutoff a = 1/ √ ρ follows S = 0.33 log (1/ma) + 1.17, in agreement with the expected form (7).These results are shown in Figure 8 along with the best fit log scaling with a coefficient of 0.33 ± 0.01 and non-universal constant 1.17 ± 0.04. We also studied the entanglement entropy scaling with respect to the mass, in the range 5 ≤ m ≤ 15 in a diamond with side length 2L = 1 and ρ = 20000.The results are shown in Figure 9, along with the best fit log scaling S = 0.333 log (1/ma) + 1.993.The coefficient of the log fit, 0.333 ± 0.006, is in good agreement with the expected result (7).The non-universal constant is 1.993 ± 0.017. We have thus extended some of the main results of [7,12] to the massive scalar field theory.Knowledge of the form of the eigenvalues in the massive case was vital in obtaining a physical understanding of the spectrum of i∆, from which a meaningful truncation scheme was defined.As shown above, we obtained the expected (1 + 1D log) "area law" scalings for the massive scalar field theory.The universality of the spectrum of i∆ in the UV regime, where truncations need to be implemented, may hold true more generally.This could make extensions to more general theories and spacetimes possible as long as the UV limit of the massless theory in D-dimensional flat spacetime is well understood.We discuss this further in Section 6. Fluctuations So far we have focused on the eigenvalues that we wish to keep in the truncation (the left of the bend in the spectrum of Figures 2 and 7), and their corresponding eigenfunctions which are plane wave-like.In this Section we turn our attention to the discarded eigenvalues and eigenfunctions. From the spectra in Figures 2 and 7, it is already evident that the latter set has a qualitatively different behaviour.The spectrum no longer follows a power law in this regime.This is seen by the trend in the Figures 2 and 7 curving down sharper and sharper as we move to the right. It is also instructive to compare a typical eigenvector corresponding to a large (within the power law regime) and a small (past the power law regime) eigenvalue in the causal set.These are illustrated in Figure 10.At the top, in Figure 10a, is the real part of the eigenvector corresponding to the 10 th largest positive eigenvalue in a massless theory on a causal set with N = 1000 elements and a diamond with L = 1/2.The eigenvector values have been interpolated between the values they take on causal set elements.As expected from (18), the behaviour is consistent with a linear combination of plane waves, and the oscillations are smooth and above the discreteness scale.At the bottom, in Figure 10b, is the real part of the eigenvector associated with the 425 th largest positive eigenvalue in the same causal set.This eigenvector has a much more jagged behaviour, with features near the discreteness scale, and does not resemble a linear combination of plane waves with variations above the discretenesss scale.The comparison of the imaginary parts of the eigenvectors follows the same trend.While there is no objective or unique measure of smoothness or jaggedness, the qualitative contrast of smooth oscillations above the discreteness scale versus sharp variations over scales at or below the discreteness scale is a persistent feature of the two branches of eigenvectors in the causal set. The jagged and chaotic nature of the small eigenvalue components prompts the question of whether they can be viewed as some kind of fluctuation arising from a particular causal set sprinkling.We investigate this question as follows: We consider a fixed coarse grained (by 10%) causal set of N c elements sprinkled into a diamond with L = 1/2.We then consider an ensemble of 20 denser sprinklings with N elements (where N c /N = 9/10), including the original N c .Namely, the N c elements will be shared by all causal sets in the ensemble, while the remaining N − N c elements will be different.For each of these causal sets, we compute the eigenvectors and eigenvalues of i∆ for the massless scalar theory.We then use the following strategy to study whether an eigenvector is a fluctuation occurring in a particular causal set or else an eigenvector with physical reality in every causal set in the ensemble: First, for each sprinkling, we average the n = 10 largest (in absolute value) entries of each j th eigenvector.† Then, for each j, we take the average of this quantity over all the sprinklings in the ensemble. Second, we set the j th eigenvector of each sprinkling to the same phase.For example, this can be done by requiring the imaginary part to vanish at the same fixed element (which must be one of the elements in the coarser causal set) in all the sprinklings.Then, the average of each in-phase j th eigenvector on the N c elements is taken across all the sprinklings, producing an "averaged eigenvector".This averaging is possible because we have fixed the N c elements across all the sprinklings.Had we used entirely different sprinklings, there would not be a meaningful notion of taking the average at an element(s), because the different elements in the different sprinklings cannot not be identified with one another.Next, we repeat a similar procedure as in the first step: we average the absolute values of the n = 10 largest entries of the averaged eigenvectors.This boils down the properties of our averaged eigenvector to a single number, which we can compare to the analogous number we obtained in step one (where we didn't first average over N c ). Finally, we divide the result obtained in the second step by the result obtained in the first step.An example of the results of this process are shown in Figure 11.A ratio close to 1 indicates that the behaviour of a single eigenvector captures well other sprinklings' eigenvectors corresponding to the same eigenvalue.On the other hand, lower ratios show that the eigenvectors are more particular to a specific causal set, i.e. they fluctuate and cancellations occur, leading to an averaged vector with a smaller magnitude.The latter could be interpreted as a "fluctuation" arising in a particular sprinkling. As evident in Figure 11, the ratio first monotonically decreases before approaching a constant.This means that there is an eigenvalue beyond which the eigenvectors do not become more "jagged" or fluctuation-like, according to our prescription. Relation to Truncations The trend in Figure 11 shows a steady decrease, signalling an approach to more fluctuation-like behaviour, followed by a stagnation in the fluctuation-like regime.These two regimes of decrease and stagnation are approximately marked by lines drawn in the figure.An interesting question is, could the transition between these two regimes be related to the transitions in Figures 2 and 7 from the power law regime to the non-power law regime?Viz, could this transition be another signature of the point at which we must truncate?Let us investigate this. We consider several collections of sprinklings, each collection with a different density, and apply the procedure laid out in the previous subsection to each collection.We then approximate the transitions from the decreasing to the constant regimes ‡ and record the eigenvalue magnitude, λ transition , at which this occurs.Our results for λ transition are shown in Figure 12 versus the number of elements N in their respective sets.We considered sprinklings up to 2200 elements and the best fit power law for our results was λ transition = (0.13 ± 0.03)N (0.46±0.04) , noting that the errors given should be understood as lower bounds since there is an uncertainty going into the estimation of the two regimes of Figure 11. In order to reduce this uncertainty, the slope of the first regime was plotted against the number of elements, as shown in Figure 13.As can be seen, a relatively clear minimum is present which removes some of the error in the distinguishing of the linear and constant regimes.The transition point was then decided by taking the average of these minima. The results for λ transition are close to our usual truncation rule in (23).This gives evidence to the suggestion that there is a connection between the truncation and the ‡ The approximation is done via the intersection of a linear and constant fit, as shown in Figure 11.transition to fluctuation-like eigenvectors.We have also checked that our results persist if we consider slightly different N c and n values. The connection between the transition to fluctuations and the truncation point is practically useful, as it provides an independent truncation scheme (using the eigenvectors) which generalises to any causal set (i.e.any dimension and with or without curvature).This is especially significant in cases where we lack a physical understanding of the eigenvalues and eigenfunctions of i∆, for example in terms of wavenumbers and plane waves, to reveal their relation to the discreteness scale. Discussion and Conclusions In this work we extended previous work on the entanglement entropy of a massless scalar field theory on a 1+1D causal set to Rényi entanglement entropies and the massive scalar field theory.In all the cases that we studied, we obtained results consistent with the expected area law scalings (log scalings in 1 + 1D).We used a spacetime definition of entropy and the covariant causal set discreteness scale as our UV cutoff.We also arrived at two key insights that will facilitate future generalisations of these studies to higher dimensions and causal sets approximated by curved spacetimes. The first insight was in our study of the massive theory, where we observed that the spectrum of the Pauli-Jordan function in the massive theory (where m 2 ρ) approaches the spectrum in the massless theory as we go deeper in the UV, towards the discreteness scale.The significance of this is that our knowledge of truncations, which are necessary to obtain meaningful entanglement entropies, can be carried over from the well studied massless theory to the less understood massive one.These truncations need to be implemented near the discreteness scale, which is precisely when the massless and massive spectra resemble one another.A conjecture is that this universality is more general.Namely, that the spectra in a different shaped compact region, in the presence of mass, curvature, and/or interactions, all approach that of the massless theory in the causal diamond, provided that all these other scales (mass, curvature, interaction strength, and system size) are far from the discreteness scale.This would mean that a general understanding of the spectrum of the spacetime commutator or Pauli-Jordan function in the UV would arise from our understanding of it in causal sets approximated by flat spacetime in every dimension of interest. The second novel insight that we obtained was regarding the nature of the truncated eigenvectors as fluctuations.We considered a particular averaging of the eigenvectors in an ensemble of coarse grained causal sets, and found evidence that the truncated (or discarded) eigenvectors have few features that persist after the averaging, whereas the non-truncated eigenvectors have large-scale/smooth variations and persistent features after the averaging procedure.This indicates that these components behave as fluctuations particular to individual causal sets.The physical significance of this requires further thought.Could this be a consequence of treating the causal set as a fixed background?Which quantities have a physical reality in every causal set realisation, and which quantities should only be regarded as physical if they persist over an ensemble of sprinklings?A full answer to these questions will require further development of causal set theory as a theory of quantum gravity.In this particular context of entanglement entropy, since we would like to recover the conventional area laws and study their relation to black hole entropy, it is the case that we need to exclude these fluctuations from our calculations.We have shown in this work how this can more generally be done. Rewriting this in terms of µ we arrive at the same expression as in (A.4).Hence the entropy (1) agrees with the one in (A.4).Passing over to the full field theory, we sum over all eigenvalues of (i∆) −1 W to arrive at the entropy (1). Similarly, the formulas for Rényi entropies in terms of the eigenvalues of (i∆) −1 W can be shown to be where each term in the sum (A.9) accounts for a pair of eigenvalues λ and (1 − λ). Figure 1 : Figure 1: Causal set of a smaller causal diamond (with side length 2 ) nested within a larger causal diamond (with side length 2L) of 2000 elements generated via a Poisson sprinkling. 2 : Comparison (on a log-log scale) of the positive spectrum of i∆ in the continuum and the causal set for a causal set of 2000 elements and L = 1 Figure 3 : Figure 3: Massless scalar field entanglement entropy versus the half side length of the subdiamond in units of the discreteness length a. Sprinklings from 15000 to 40000 elements were considered and /L = 1/2.The data fits S = 0.332 log ( /a) + 1.104, shown in red. Figure 4 : Figure 4: Rényi entropies of order 2, 3, 4, and 5 for a massless scalar field on a causal set.Sprinklings from 15000 to 40000 elements were considered and /L = 1/2.The best fit curves are shown in red. Figure 5 : Figure 5: Left: a plot of the interpolation of the real part of an eigenfunction of i∆ for a massive scalar field with m = 10 on a causal set sprinkling of 10000 elements with 2L = 1, against u and v. Right: a plot of Re{f k,m } with a guess value of k = 7 against u and v. Figure 6 : Figure 6: Plot of Re{i∆(u − u , v − v )g k,m (u, v)} in blue and Re{ λk,m g k,m (u , v )} in red mesh against u and v for k = 3 and m = 10. 7 : Log-Log plot of each n th largest λ against n, for L = 1/2 and a sprinkling density of ρ = 7000 for m = 0 in blue, m = 5 in orange, and m = 10 in green.The red dashed horizontal line corresponds to λmin . (a) Real part of the eigenvector corresponding to the 10 th largest eigenvalue for a causal set of 1000 elements and L = 1/2.(b) Real part of the eigenvector corresponding to the 425 th largest eigenvalue for a causal set of 1000 elements and L = 1/2. Figure 10 : Figure 10: Comparison of the eigenvectors corresponding to a large and a small eigenvalue of i∆. Figure 11 : Figure 11: Ratio of averages for causal sets of N = 2000 elements, along with a constant and a linear fit.j corresponds to the number of an eigenvalue (when ordered from largest to smallest). Figure 12 : Figure12: λ transition at which there is a transition from the decreasing to constant regime in the eigenvector average ratios.Sprinklings up to 2200 elements were considered in a diamond with side length 2L = 1.The results fit λ transition = (0.13 ± 0.03)N (0.46±0.04) , represented in blue.For comparison, the √ N /(4π) truncation curve is also included, represented in orange. Figure 13 : Figure 13: Example plot of the slope of the linear fit representing the first regime as a function of the number of eigenvectors taken into account for N=1300.	9,767.8	2021-11-10T00:00:00.000	[ "Physics" ]
Maximum Power Point Tracking using Particle Swarm Optimization Algorithm for Hybrid Wind-Tidal Harvesting System on the South Coast of Java Received Oct 24, 2016 Revised Jan 23, 2017 Accepted Feb 7, 2017 This paper proposes a hybrid wind-tidal harvesting system (HWTHS). To extract maximum power from the wind and tidal, HWTHS implements particle swarm optimization (PSO) algorithm in maximum power point tracking (MPPT) method. The proposed HWTHS had been tested on the range of possible input appropriate to the characteristics of the southern coast of Java. The presented result shows that by using PSO-based MPPT algorithm, maximum power point can be achieved. Thus the efficiency of HWTHS is 92 %, 94 % in wind section and 91 % in tidal section. By using PSO-based MPPT, HWTHS can respond well to changes in wind and tidal speed, whether it's a change from low speed to a higher speed or change from high speed to lower speed wherein time to reach new steady state is ± 0.1 s. At varied wind and tidal speed, PSO algorithm can maintain Cp of the system in the range of 0.47 0.48 so that power can be extracted to the maximum. Keyword: INTRODUCTION Indonesia is an archipelago that 2/3 of its territory is ocean. It has the longest coastline in the world, about ± 80791.42 km, which is a potential area for development of wind and ocean power plants. Indonesia has average wind speed about ± 5 m/s. The wind speed of 4 m/s to 5 m/s is classified as low-sized with a potential capacity of 1-100 kW. Moreover, the extent of marine areas of Indonesia is also a potential source of ocean wave energy. Several marine areas in Indonesia have an average wave height between 0.5 to 3 m. To optimize the power generation systems, two or more types of energy sources can be combined. Each of the energy sources can overcome the weakness of the other. The intermittent nature of wind energy can be compensated by the predictable nature of ocean wave energy. However, the existence of maximum power point tracker in renewable energy power plants are still essential to ensure that the maximum power can be extracted [1], [2]. On [3], a step size which is used is a fixed value so that it will affect the speed of achieving convergent. Choosing the appropriate value of a step size is essential in designing MPPT. A small step size values will minimize the occurrence of oscillations but the system will take a long time to achieve convergent. Large step size values will shorten the time to achieve convergent but oscillation will occur around the optimum point so it will produce losses. Thus, the use of adaptive step-size MPPT algorithm is the right solution to optimize speed to achieve convergent and losses due to oscillations [4][5][6]. Research surrounding adaptive MPPT is getting a lot done, some algorithms which have been used are neuro-fuzzy, genetic algorithms, simulated annealing, and PSO. The use of adaptive MPPT algorithm shows a good performance to maximize the output power of renewable energy power plants. By using PSO algorithm, efficiency of wind energy conversion systems can be increased, and showed a good performance in responding to changes in wind speed, as presented in [7]. In this research, hybrid wind-tidal harvesting system (HWTHS) was composed of wind and tidal turbine system. To maximize the output power of HWTHS, PSO algorithm was used to control the duty cycle of the buck-boost converter. MPPT process was performed on each system, wind energy system (WES) and tidal energy system (TES), so that the maximum power of each system can certainly be extracted. Furthermore, total extracted power from the two systems were transferred to the load. In this study, HWTHS was tested in accordance with the prevailing winds and ocean waves on the south coast of Java island. HWTHS's MPPT HWTHS was composed of wind turbine, tidal turbine, rectifier, buck-boost converter, and load as shown in Figure 1. Wind and tidal turbine was used to convert wind and tidal energy into mechanical energy according to (1). The coefficient of performance of the turbine (Cp) represents the power extraction efficiency from turbines. In theory, the maximum value of Cp is 0.59, but in practice the Cp values only in the range of 0.4 -0.45 [8]. At each turbine there is a spesific operating point where the mechanical power can be extracted to the maximum, the point is commonly referred to optimum tip speed ratio (TSR). TSR (λ) is the ratio between the rotational speed to wind (or tidal) speed (4). In the state of wind and tidal vary over the time, the TSR should be maintained so the value will always be at the optimum point. Thus, the Cp of the system can be kept constant at the optimal point so that maximum electrical power can be extracted. where P m is mechanical power, ρ is air and seawater density, R is rotor diameter, V is wind or tidal speed (m/s), β is picth angle, and ω m is rotational speed. Figure 2 shows the power characteristics of wind and tidal turbine in which at any wind and tidal speed there is a certain rotational speed where the maximum power generated. The point becomes the target of MPPT so Cp can be kept constant at the optimal point and the generated power is always maximum. To be able to extract the maximum power, duty cycle of the buck-boost converter was equipped with MPPT algorithm. The output power of the buck-boost converter was the basis variable for evaluating the duty cycle. In this study, the PSO algorithm used 3 particles wherein the particles represent a duty cycle (d). While speed (Ф) represents step-size of duty cycle. MPPT process was carried out with reference to Figure 3. Where w is the momentum factor (w = 0.15), r 1 and r 2 are random values between 0 and 1, c 1 and c 2 are acceleration constants (c 1 = c 2 = 0.5 and 1.6). Wind and tidal turbine parameters used in this research can be seen in Table 1 along with parameters of PMSG. Figure 4 shows wind and sea waves characteristic which were the basis for testing the proposed system. The wind gradient in the region of Indonesia generally blows from the southeast -Southwestern with wind speeds ranging between 2.5 -10 m/s. With the highest wind speeds were in the Western Indian Ocean South Sumatra to East Java, Andaman Sea, South China Sea, Java Sea and the Eastern Pacific Ocean Philippines. Figure 4 (a) is a sample data of wind speed on the South Coast of Java taken from 2 -5 September 2016. The wind speed varied between 1.03 -5.66 m/s with an average wind speed of 3.81 m/s. At that time the dominant wind speed was 4-5 m/s. This was influenced by the character of the monsoon east where the wind blew from the continent of Australia to the Asian continent through the desert in the northern part of Australia and only through the narrow sea. So the wind was dry which resulting territory of Indonesia suffered drought and in general had a relatively stable wind speed. When the wind blows over the surface of the sea, some of its energy is transferred to the sea water through friction between the air molecules and the water molecules. Data showed that several marine areas in Indonesia had the potential waves with an average height of 0.5 -2 m, which is a potential source of energy to generate electricity. Sample data of ocean wave height on the South Coast of Java can be seen in Although the process of formation of waves influenced by the wind, but both have different characteristics. As can be seen on Figure 4(b), ocean wave is more predictable and stable, so the nature of ocean wave energy can be used to compensate the intermittent nature of wind energy. RESULTS AND ANALYSIS After the analysis of the design and MPPT mechanism for HWTHS, simulation using Simulink was conducted to verify the proposed method as shown in Figure 5. In this simulation, the parameters used in HWTHS refer to Table 1. The load used in this study is a resistive load with a value 10 Ω. Performance of PSO MPPT Algorithm for Each System Based on simulation results, the greater the wind speed, the output power of the WES became greater. This also applies to TES, the greater the speed of the tidal, the output power of TES also increased. PSO algorithm which was implemented on HWTHS could increase the output power of each system, both WES and TES. Thus, the efficiency of WES could be increased from 71% to 94% while TES's efficiency increased from 66% to 91%. In the TES, in addition to improve the efficiency of 24%, the use of the PSO algorithm also could maintain system efficiency at 91% where TES's efficiency that did not use MPPT varies between 49-86%. Performance of WES and TES at any wind speed and tidal speed can be seen in Figure 6. Performance of PSO MPPT Algorithm for HWTHS Based on [9], tidal speed is a function of the ocean wave height as stated in (5). Where U is tidal speed, m was average beach slope (m= 0.033), g is acceleration of gravity (g= 9.8 m/s 2 ), H is ocean wave height, and α was the wave breaker angle (α= 15˚). CONCLUSION The proposed HWTHS was composed of wind turbine, tidal turbine, rectifier, buck-boost converter, and load. MPPT process was performed on each system, WES and TES. The proposed system had been tested on the range of possible input appropriate to the characteristics of the southern coast of Java. The presented result shows that by using PSO-based MPPT algorithm, maximum power point can be achieved. Thus the efficiency of HWTHS is 92 %, 94 % in wind section and 91 % in tidal section. By using PSO-based MPPT, HWTHS can respond well to changes in wind and tidal speed, whether it's a change from low speed to a higher speed or change from high speed to lower speed wherein time to reach new steady state is ± 0.1 s. At varied wind and tidal speed, PSO algorithm can maintain Cp of the system in the range of 0.47 -0.48 so that power can be extracted to the maximum.	2,427.6	2017-04-01T00:00:00.000	[ "Computer Science" ]
Haemodynamic Effects and Complications of Unilateral Spinal versus Standard Spinal Anesthesia in Elderly with Low Ejection Fraction Undergone Lower-Limb Surgery Background: Cardiovascular system may be profoundly affected by spinal anaesthesia due to unavoidable sympathetic blockade which is more prominent in elderly.A restricted sympathetic block during spinal anesthesia may minimize hemodynamic changes. Objective: To assess whether a unilateral spinal anaesthesia using 0.5% hyperbaric bupivacaine will restrict the sympathetic block to avoid the undesired cardio vascular effects. Materials and method: In this prospective study 60 ASA Ill and IV patients aged between 60-90 years undergoing unilateral lower limb surgery were included. Patients were divided into two groups. In group-A, dural puncture was performed with the patient in the lateral decubitus position with 1.5 mL of hyperbaric bupivacaine. In group-B, it was performed with the patient in a seated position using 1.5 mL hyperbaric bupivacaine. Each patient was then placed in supine position. The speed of injection was 1 mL/30s. Patients were placed in the lateral position with operated side down and kept in this position for 10 minutes. Motor and sensory levels were assessed, and haemodynamic alterations were monitored just after block, 5, 10, 15 and 30 minutes of spinal anaesthesia. Results: The demographic data were found similar in both groups. The time to the onset of the sensory and motor block was significantly shorter in group-B. The duration of motor and sensory block was significantly shorter in group-A. Haemodynamically all the parameters revealed better out come in unilateral spinal anesthesia. The incidence of complications (nausea, headache, and hypotension) was also lower in group A. Conclusion: When unilateral spinal anesthesia was performed using a low-dose, low-volume and low-flow injection technique, it provides adequate sensory-motor block and helps to achieve stable hemodynamic parameters during surgery on a lower limb. Furthermore, this technique avoids unnecessary paralysis on the non-operated side. Correspondence: Dr. Md.Mushfiqur Rahman.e-mail<EMAIL_ADDRESS>see the cardiovascular effects of spinal blockade.Hypotension is the most frequent side effect of spinal anaesthesia, occurring in more than 30% of patients. 1rd et al. 2 reported a decrease in mean arterial blood pressure of 21.3% of the base line following spinal anaesthesia.He also reported that a level of spinal anaesthesia to T5 resulted in an increase in heart rate by 3.7%.The cardio accelerator fibres originate from T1-T4, so the level of spinal anaesthesia affecting these dermatomes may cause bradycardia.The unilateral spinal anaesthesia has been claimed by many as an alternative technique, to restrict the undesired sympathetic block. 3There are many benefits to this technique including fewer hemodynamic changes, less urinary retention, more satisfied patients, better motility during recovery and the restriction of selective nerve block to the relevant limb. This study was undertaken to evaluate whether unilateralspinal anesthesia can avoid the undesired cardiovascular effects and patient satisfactions accompanying the conventional spinal anaesthesia in elderly cardiac compromised patient. Materials and method This prospective study included sixty adult patients scheduled for unilateral lower limb surgery. The patients were divided in two randomized groups: A and B. In group-A, unilateral spinal anesthesia was used with 7.5 mg of 0.5% hyperbaric bupivacaine.In group-B, standard spinal anesthesia was used with 7.5 mg of 0.5% hyperbaric bupivacaine.Patient age ranged from 60 to 85 years.The patients were in ASA class III or IV.The duration of Nil per os (NPO) time and the sedation regimen were the same in both groups. Patients with any absolute contraindication for regional anesthesia were excluded from the study. Ethical approval for this study was provided by Hospital Ethics Committee, BIRDEM General Hospital, Shahbagh, Dhaka.Informed consent was obtained from each patient to ensure that he or she understood that the technique used for spinal anesthesia would be modified.An IV cannula was inserted, and then a mL/kg intravenous infusion of lactate Ringer's solution was administered over 20 min.All patients underwent standard monitoring, including electrocardiography, non-invasive blood pressure measurements and pulse oximetry. In group-A, spinal anesthesia was performed with the patient in the sitting position at the L3-L4 inter space using a 25-G Quincke spinal needle in sterile condition.Once intrathecal placement had been confirmed, 7.5 mg of hyperbaric bupivacaine 0.5% was injected.The patient was then placed in the supine position. In group-A, the patients were placed in the lateral decubitus position with the operating limb in the lower position.Similar to the technique used for group-A, the L3-L4 inter-vertebral space was detected, and then spinal anesthesia was performed with a 25-G Quincke spinal needle.After the confirmation of intrathecal needle placement, 7.5 mg of hyperbaric bupivacaine 0.5% was injected at a speed of 5mg every 30 s.The patient was kept in this position for 10 minutes.After confirming a unilateral sensory block up to T10 dermatome, surgery was allowed to proceed. In group-B, spinal anesthesia was performed with the patient in the sitting position at the L3-L4 inter space using a 25-G Quincke spinal needle in sterile condition.Once intrathecal placement had been confirmed, 7.5 mg of hyperbaric bupivacaine 0.5% was injected.The patient was then placed in the supine position.To reduce patient anxiety, 2 mg of midazolam was injected IV. Sensory block was assessed by using ice packs, bilaterally, at every 5 minutes, for 30 minutes after spinal anaesthesia.Motor block of lower limbs was also assessed bilaterally using Bromage scale at baseline and at every five minutes, for 30 minutes after spinal anesthesia. The Bromage scale was, 0=no paralysis; 1=inability to raise extended leg; 2=inability to raise flex knee; 3=inability to do dorsi flexion of foot but can wiggle toes and 4=inability to move at all.The patient was asked to raise the extended leg, flex the knee, flex the ankle and was rated from 0 -4.The haemodynamic variables such as blood pressure and heart rate were monitored before spinal anesthesia, just after spinal anaesthesia and then every 5 minutes interval for 30 minutes.If systolic blood pressure decrease by more than 30% of base line or systolic blood pressure less than 90 mmHg, ephedrine 5 mg in incremental doses with maximum of 50 mg was used.Bradycardia, was defined as heart rate less than 50 beats/minute and was treated with atropine 0.5-1 mg given intravenously. The duration of surgery and any complications regarding anaesthesia were also recorded.All patients were shifted to recovery room after surgery.Patients were discharged from recovery room when they were able to move both legs and their haemodynamic status was stable. The clinical data including the onset of sensory and motor block, hemodynamic changes, duration of sensory and motor block and the complications of spinal anesthesia were evaluated using SPSS version 19.6 for windows.Unpaired t-test, chi-square and Fisher's exact test were used where applicable.p value of <0.05 was considered as significant. Results Demographic data of patients are shown in Table I.No demographic variables differed significantly between Group-A and Group-B. Table I: Demography Intra-operative period pulse rate was slower at each time interval in Group-B in comparison with Group-A.In Group-A mean pulse rate ranged from 79±6.2 beats/min to 69±2.2 beats/min while it ranged between 76±6.5 (beats/min) to 68±7.4 (beats/min) in Group B. Significant decrease in group-B was observed at 15 and 30 minutes after block (Table II). Table II: Comparison of heart rate between groups Table III shows the status of systolic and diastolic blood pressure in two groups.The mean change of systolic blood pressure varies in Group-B from 142.36+12.89mmHg to 100.6±10.8mmHg and in Group-A from 140.83+12.25 mmHg to 120.40+8.82mmHg.The mean change of diastolic blood pressure varies in Group-B from 79.6±4.1 mmHg to 70.8±1.8mmHg and in Group-A 80.1±4.5 mmHg to 75.8±6.8mmHg.Both mean diastolic and systolic blood pressure significantly decreased in group-B in comparison to group-A at 5, 10, 15 and 30 minutes after block. Table III: Comparison of blood pressure between groups The change of mean arterial pressure varies in Group-B from Group-A.Significant changes were observed in three occasions (5 minutes, 10 minutes and 15 minutes after block) (Table V). Table V: Comparison of mean arterial pressure between groups Table VI shows the distribution of complications between the groups where no complication developed in Group-A. Table VI: Distribution of complications between groups Better outcome was observed regarding duration and quality of anaesthesia where Group-A showed better outcome (Table VII). Discussion The patient's position during and immediately after spinal anesthesia influences the spinal distribution of drugs.If an anesthetic drug solution is hyperbaric with respective the cerebrospinal fluid, it is possible to create a unilateral block.Moreover, the distance between the left and right nerve roots in the lumbar and thoracic region is about 10-15 cm, which makes it possible to achieve unilateral spinal anesthesia. 4Kusniemi et al. reported that hyperbaric bupivacaine is more effective in achieving unilateral spinal anesthesia than plain bupivacaine. 5However, determining the optimal time for lateral positioning is difficult when a high dose of hyperbaric bupivacaine (12-20 mg) is used.The anesthetic drug may migrate even when the patient is placed in the lateral position for 30-60 min.Conversely, if a low dose (5-8 mg) of anesthetic solution is used, putting the patient in the lateral position for 10-15 min may prevent migration of the anesthetic drug.In this study, we injected 1.5 ml (7.5 mg) of hyperbaric bupivacaine 0.5% to achieve unilateral spinal anesthesia.The patient was kept in the lateral position for 10 min, which led to unilateral spinal anesthesia in 94.45% of cases.In two cases, the anesthetic drug spread to the other side, resulting in bilateral spinal anesthesia.In a study performed by Esmaoglu, the patient was in the lateral position for 05 min.This approach yielded an 85.7% success rate.This discrepancy in terms of the success rate seems to be dependent on the duration of time spent in the lateral position. 6otably, none of the patients in the unilateral spinal anesthesia group experienced hypotension, but six patients in the bilateral group had hypotension and were treated by IV ephedrine.Chohan and Afshan administered unilateral spinal anesthesia prior to lower-limb surgery in elderly patients with ASA classificationof III or IV (average age, 60).The authors found no significant hemodynamic changes.They used hyperbaric bupivacaine 0.5% (1.1-1.2 mL). 7In our study, there was no bradycardia in the unilateral group, but in the bilateral group, 5 patients had bradycardia.On average, the time to the onset of anesthesia and immobility was faster in the bilateral as compared to the unilateral spinal anesthesia group.The sensory and motor block lasted for less time in the unilateral as compared to the bilateral group.Unilateral spinal anesthesia is therefore suitable for outpatient surgery.Valanne used 4 or 6 mg of bupivacaine to induce unilateral spinal anesthesia in 106 patients scheduled to undergo knee arthroscopy.While both doses were sufficient for sensory and motor block, 4 mg of bupivacaine achieves a more rapid regression of motor function. 8 Conclusion Unilateral spinal anesthesia with a low dose (7.5 mg) limited volume (1.5 mL) induces sufficient sensory and motor block with anappropriate level of analgesia for lowerlimb surgery.This technique is a good alternative of conventional spinal and achieves stable hemodynamics, particularly in elderlyASA classIII/IV and patients with low ejection function.	2,525.6	2017-02-04T00:00:00.000	[ "Medicine", "Biology" ]
Omi, a recessive mutation on chromosome 10, is a novel allele of Ostm1 Large-scale N-ethyl-N-nitrosourea (ENU) mutagenesis has provided many rodent models for human disease. Here we describe the initial characterization and mapping of a recessive mutation that leads to degeneration of the incisors, failure of molars to erupt, a grey coat colour, and mild osteopetrosis. We mapped the omi mutation to chromosome 10 between D10Mit214 and D10Mit194. The Ostm1 gene is a likely candidate gene in this region and the grey-lethal allele, Ostm1gl, and omi mutations fail to complement each other. We show that om/om mice have reduced levels of Ostm1 protein. To date we have not been able to identify the causative mutation. We propose that omi is a novel hypomorphic mutation affecting Ostm1 expression, potentially in a regulatory element. Introduction Large-scale N-ethyl-N-nitrosourea (ENU) mutagenesis has provided many rodent models for human disease, including several forms of deafness, Branchio-Oto-Renal syndrome, and CHARGE syndrome (Hrabe de Angelis et al. 2000;Smits et al. 2005;Bosman et al. 2005Bosman et al. , 2009Calvert et al. 2011;Hilton et al. 2011). Due to the nature of production of these mutant mice, it is likely that additional mutations are present in the mouse lines studied because mutations are generated at random in the genome and are usually noticed only when they lead to a detectable phenotype. Some mutations may be without any obvious consequence, but other mutations may have additional phenotypes of varying degree that may go undetected because they are not screened for or they may appear in later generations if they are recessive. It has been estimated that an average of around 150 mutations occur in each F1 offspring from an ENU-mutagenized male (Hrabe de Angelis et al. 2000). Many of these additional mutations will be lost by dilution in successive generations through crossing to the wild-type background (Hilton et al. 2011) or by selection against the mutant phenotype; however, some will remain. We report here a new recessive phenotype that was found in a line selected for a completely different phenotype originating from an ENU mutagenesis screen. The phenotype was initially detected due to high lethality just after weaning in some matings within this mouse line. Here we describe the initial characterisation and mapping of this mutation, that we named omi (om). Omi is a recessive mutation that leads to degeneration of the incisors, failure of molars to erupt, a grey coat colour, and mild osteopetrosis. We mapped the omi mutation to chromosome 10 between D10Mit214 and D10Mit194. The Ostm1 gene is a likely candidate gene in this region, and the grey-lethal allele, Ostm1 gl , and omi mutations fail to complement each other. We show that om/om mice have reduced levels of Ostm1 protein. To date we have not been able to identify the causative mutation. We propose that omi is a novel hypomorphic mutation affecting Ostm1 expression, potentially in a regulatory element. Mice Animal husbandry and experiments were carried out in accordance with UK Home Office regulations. The omi mutation was detected in a line of mice derived from a large-scale mutagenesis screen (Hrabe de Angelis et al. 2000). Male C3HeB/FeJ mice were injected with three doses of 80 mg/kg ENU at weekly intervals, allowed to recover, and mated with uninjected C3HeB/FeJ females. F1 offspring were screened for a variety of dominantly inherited defects, including deafness and vestibular (balance) defects. The recessive omi mutation was discovered in a mouse line initially isolated because of its dominantly inherited mild head-bobbing behaviour (ABE9, also known as Bob). We were unable to map the original feature of mild head bobbing to any chromosome and no obvious malformation of the inner ear could be detected (Bosman and Steel, unpublished results). Offspring from the original ENU-mutagenized male were outcrossed to wild-type C3HeB/FeJ mice (never exposed to ENU) at least five times, diluting out other mutations resulting from the ENU treatment, before the omi phenotype was discovered. The omi phenotype was not linked to the head bobbing, and mice described here showed no sign of any balance defect. Affected omi mice were provided with a Pico-Vac Ò soft dietary supplement (LabDiet) which enabled them to survive until adulthood. For all experiments control animals were fed normal diet pellets ad libitum. For all phenotypic analyses the mice were studied on their original C3HeB/ FeJ genetic background and unaffected wild-type or heterozygous littermates were used as controls. Grey-lethal mice were obtained from Prof. T. Jentsch and were genotyped as described previously (Chalhoub et al. 2003). Genetic mapping Om/om animals on a C3HeB/FeJ genetic background were outcrossed to C57BL/6J mice. ?/om offspring were backcrossed to affected (om/om) C3HeB/FeJ animals. Backcross mice were collected around weaning time. The mice were killed by cervical dislocation, the teeth were photographed, and tissues were taken for DNA purification. The DNA from 28 backcross offspring exhibiting tooth defects was used in a genome scan to link the tooth defect with a chromosome. A panel of 69 markers spanning the autosomes was used to detect polymorphisms between C3HeB/FeJ and C57BL/6J mice (Supplementary Table 1). For further fine mapping, another ten polymorphic markers were used (Supplementary Table 2). Single nucleotide polymorphisms (SNPs) between C57BL/6J and C3HeB/FeJ in the region of interest were identified using a SNP database (http://www.informatics. jax.org/javawi2/servlet/WIFetch?page=snpQF). Primers spanning the SNPs were designed using Primer3 (http://frodo. wi.mit.edu/) with standard settings (Supplementary Table 2 for primer sequence). PCR products were then sequenced with the forward and reverse primers to detect the SNP using standard techniques. Phenotypic analysis Adult mice were killed by cervical dislocation. For histology, organs of 4-week-old male and female mice (n = 10; five om/om and five ?/om) were collected and fixed in 10 % neutral buffered formalin. Heads were subsequently decalcified in 10 % EDTA (pH 8.0) for 7 days, followed by paraffin embedding. Embryos (E16.5, n = 13 from heterozygous 9 homozygous timed matings) were dissected in ice-cold PBS and decapitated, and the heads were fixed in 10 % neutral buffered formalin for 2 days and processed for paraffin embedding. P7 pups (n = 10 from heterozygous 9 homozygous matings) were decapitated and the heads were fixed in 10 % neutral buffered formalin and decalcified in 10 % EDTA (pH 8.0) for 7 days, followed by paraffin embedding. Serial sections at 7 lm were cut and stained with haematoxylin and eosin by standard procedures. To analyse the skeletons, mice (seven male ?/om, four male om/om, seven female ?/om, and four female om/om) were weighed and anaesthetised by intraperitoneal injection of either Avertin (1.25 %) or 100 mg/ kg ketamine/10 mg/kg xylazine (dosage of 0.1 ml/10 g mouse weight) at 6-8 and at 12 weeks of age. When ketamine/xylazine was used, the anaesthesia was reversed using a solution of 1 mg/kg Antisedan (0.1 ml/10 g mouse weight). Mice were scanned in a MX20 Specimen Radiography System (Faxitron X-Ray Corporation, Lincolnshire, IL, USA) in combination with Faxitron SR v1.2 software. At 12 weeks the density of the tissues was measured using a PIXImus Densitometer in combination with Lunar PIXImus 2 2.1 software. At 16 weeks of age, all nonfasted mice used for the radiographic skeletal analysis were subjected to a terminal retro-orbital bleed under anaesthesia induced by 100 mg/kg ketamine/10 mg/ kg xylazine (dosage of 0.1 ml/10 g mouse weight). A complete blood count and clinical chemistry panel were analysed (http://www.sanger.ac.uk/mouseportal/phenotyp ing/MAHN/plasma-chemistry/). Whole blood was analysed with an automatic haematology analyser which uses spectrophotometry and volume impedance principle to measure white and red blood cell counts, mean corpuscular volume, haemoglobin, erythrocyte indices (haematocrit, mean corpuscular haemoglobin, mean corpuscular haemoglobin concentration, red blood cell distribution width), platelet counts, and mean platelet volume (scilVet Animal Blood Counter, RAB 015 A Ind.E, 22.02.01). Plasma chemistry was analysed using an Olympus AU400 for the following parameters: sodium, potassium, chloride, glucose, triglycerides, cholesterol, high-density lipoprotein, low-density lipoprotein (LDL), NEFAC, glycerol, amylase, alanine aminotransferase, alkaline phosphatase, creatine kinase, aspartate aminotransferase, total bilirubin, total protein, albumin, creatinine, urea, calcium, magnesium, iron, phosphate, lactate dehydrogenase, and uric acid. Western blotting Adult mice (n = 2 per genotype) were killed by cervical dislocation and adult brains were frozen in liquid nitrogen and stored at -80°C until further use. Brain tissue (7 ml per 0.379 g tissue) was lysed by homogenization in an ice-cold Dounce homogenizer in ice-cold DOC buffer [1 % sodium deoxycholate in 50 mM Tris-HCl (pH 9.0)] containing Complete protease inhibitors (Roche Applied Science, Indianapolis, IN, USA) according to the manufacturer's instructions. The lysate was incubated in ice for 1 h, spun at 4,000 rpm for 1 h, and then cleared through a 5-lm filter. Protein concentration was determined and similar quantities were loaded on a 4-12 % acrylamide gel (Bio-Rad, Hercules, CA, USA) and blotted. Western blots were blocked for 4 h at 4°C in TBS ? 0.1 % Tween-20 (TBST) containing 5 % low-fat milk powder, rinsed in TBST, and then incubated in anti-OSTM1 (catalog No. HPA010851 from Atlas Antibodies, Stockholm, Sweden) diluted 1:2,000 in TBST ? 3 % low fat milk overnight at 4°C. Blots were rinsed and washed 3 9 10 min with TBST at room temperature, then incubated with anti-rabbit-HRP (Horse radish peroxidase) diluted 1:50,000 in TBST ? 3 % low fat milk overnight at 4°C. Blots were rinsed and washed 3 9 10 min and 2 9 30 min in TBST. An ECL Advance kit (Amersham GE Healthcare Life Sciences, Piscataway, NJ, USA) was used to detect HRP activity. Blots were then stripped, blocked, and incubated with anti-PSD95 (catalog No. MA1-046 from Affinity Biosciences, Burwood, VIC, Australia) as described above, except for the use of anti-mouse-HRP as a secondary antibody. Results Omi is a recessive mutation affecting teeth and coat colour The recessive omi founders were discovered in a mouse line initially discovered because of dominant mild headshaking behaviour (ABE9), although the balance defect was not linked to the tooth defect. In a subset of matings of healthy parents, very small animals were found around the time of weaning, some of which died or had to be culled for welfare reasons. Providing these small mice with a special diet of soft food improved their survival rate. Analysis of these affected mice and normal littermates showed that these small mice lacked most incisors (Fig. 1a, b). In addition, we found that affected mice had a more grey (or less yellow) coat colour than their normal littermates (Fig. 1c, d). We analysed the breeding records to establish inheritance. Of matings between healthy, unaffected individuals, we obtained either normal offspring only or a mixture of affected and normal offspring (Table 1). In the latter case, we found a ratio of affected:unaffected of approximately 1:3, indicating recessive inheritance. We crossed affected animals with normal animals from the colony and found that the offspring could be either all normal or a mixture of affected (41.4 %) and normal (58.6 %) ( Table 1). Next we crossed an affected male with an affected female and we found that all ten offspring were affected, supporting the recessive inheritance. When breeding these mice we noticed that some affected (om/om) mice appeared to have a shortened life span. In crosses where normal, unaffected males were used, these were in good health but culled due to a decrease in breeding performance around 7 months (n = 10; Table 2). However, seven om/om males were found dead and another 18 had to be culled because of ill health by 6 months (24 %; total n = 104). Similar lethality was found for om/ om females (16 %; 13 of 79 females). Necropsy on these mice did not identify any obvious malformations. However, some nonbreeding om/om males (n = 2) survived to at least 1 year of age. Teeth are formed in om/om mice but are abnormal To analyse the tooth defect and other skeletal problems over time, we took X rays at 6-8 and 12 weeks of age of ?/ om and om/om mice (Fig. 2). ?/om males and females had normal incisors and molars at 6-8 and 12 weeks (Fig. 2a, data not shown). However, the morphology of incisors and molars was very abnormal in om/om mice (Fig. 2b). The incisors appeared short and the molars were not regularly aligned as in control mice. In addition, it appeared that some molars from om/om mice were still covered by other tissue indicating that these do not erupt (Fig. 2b). Sometimes upper molar 3 was visible within the buccal cavity but appeared damaged. We sectioned heads of 3-week-old ?/om and om/om animals. In normal control ?/om mice, all teeth, including molars, were present and erupted (Fig. 3a, data not shown). In om/om mice, we found that incisors had erupted but appeared damaged (data not shown). We also identified molars, but these were still covered with bone and an epithelial cell layer (Fig. 3b). This confirmed that molars failed to erupt in om/om mice. To determine the onset of the incisor defect, we examined pups from ?/om 9 om/om matings at several time points. Serial sections of teeth at E16.5 (n = 13) showed no obvious abnormalities in any of the embryos (data not shown). We observed 26 offspring from om/om 9 ?/om matings from P14 until P30-32 and recorded the status of their incisors daily (Table 3). At the end of the time course, 13 mice were scored as affected (om/om) and 13 as normal (?/om) littermates. At P14, five mice, later identified as om/om, had apparently normal incisors. However, six om/ om pups had small or thinner teeth and two mice had missing teeth. At P21, 2 of 13 om/om mice were scored as having normal incisors, two mice had small or thin teeth, and the remaining nine all had missing teeth. At P30-32, this trend continued and 11 mice had some teeth missing and two had no teeth visible at all. This indicated that the incisors were formed but not maintained. om/om mice have an abnormal bone morphology and increased bone mass Analysis of the skeletal structure by X ray showed that om/ om mice had an abnormal skull shape as a result of malformed/eroded teeth. In the maxillary area the bone around the roots of the damaged upper incisors appeared rough. This sclerosis in the maxillary area appeared to get progressively worse, with a denser appearance at the 12-week time point (Fig. 2a-d, data not shown). The zygomatic process of the maxilla was thicker and continued to be Fig. 1 Coat colour and teeth of normal (?/? or ?/om) and omi (om/om) adult mice. a Normal mice have two upper and two lower incisors, b om/om mice lack most incisors. Any remaining incisors need clipping to prevent overgrowth, c A normal agouti coat colour in mice, d om/om mice have a less yellow coat Table 1 The omi mutation is a recessive mutation: numbers of normal and affected mice were analysed to determine mode of inheritance Cross predicted genotypes Unaffected 9 unaffected ?/om 9 ?/om Unaffected 9 unaffected ?/? 9 ?/? Affected 9 affected om/om 9 om/om thick halfway into the zygomatic bone (Fig. 2c, d). Proximal ribs 1-8 were thickened, especially on the dorsal parts (data not shown). Long bones had a layered appearance and there was an increased density of the deltoid tuberosity of the humeri (Fig. 3e, f) and of the diaphyses of the femurs in females (Fig. 2g, h). In addition, the deltoid process of the humerus had an abnormal morphology (Fig. 3e, f), and femurs were shorter but thicker and had increased density at the distal epiphysis. DEXA scan revealed that om/om mice had a significantly reduced lean and fat mass (Table 4). At 8 weeks, om/om females had a low bone mineral density (BMD), measured over the full body or the femur, when compared to ?/om females (Table 4). At 12 weeks, the BMD of om/om females was significantly higher compared to that of ?/om controls mice. BMD at the right knee of om/om females was higher at both time points compared to that of ?/om females (Table 4). This suggests progressive osteopetrosis in om/om females. The BMD of om/om males was decreased at 8 weeks. At 12 weeks, the BMD of om/om males was similar to that of ?/om males. This implies that the progressive osteopetrosis seen in young females is not present in the males at this age. Blood from ?/om control and om/om mice was analysed. Blood counts were normal (Supplementary Table 4). Although the level of total alkaline phosphatase (TAP) can be influenced by many factors, TAP is always raised in osteopetrosis patients and mouse models (Sanger et al. 1966;Carolino et al. 1998;Kornak et al. 2001). The higher alkaline phosphatase levels of om/om females may be consistent with the higher BMD seen at 12 weeks. Lower levels of triglycerides, LDL, and haemoglobin in homozygote mice are likely to be an indication of poor nutritional status in om/om mice but could be due to other underlying metabolic problems (Supplementary Table 4). We analysed haematoxylin and eosin-stained sagittal sections of the heads from 4-week-old ?/om and om/om mice for tooth and bone abnormalities. In control mice molars had erupted, but in om/om mice molars were still covered by tissue (Fig. 3a, b). The jaw bones of control mice consisted of spongy bone and compacted bone areas (Fig. 3c, e). In the compacted bone, no blood vessels were seen and the bone had a regular structure (Fig. 3c). In contrast, the areas normally Fig. 2 Analysis of tooth and bone abnormalities in normal (?/om) and om/om adult female mice. a, b Lateral view of the skull of 12-week-old normal (?/om) and omi (om/om) mice. In normal mice, the incisors (in) are clearly visible, and well-defined molars (mo) are visible in the oral cavity. In om/om mice, the incisors are reduced to stumps and misshapen molars appear covered by a tissue layer (arrow). In addition, fibrotic tissue is visible in the maxillary bone (star), c, d Dorsal view of the skull of 12-week-old normal (?/om) and omi (om/om) mice. Compared to normal mice, the zygomatic bone (zy) of omi mice is thickened. Fibrotic tissue is visible in the maxillary bone (star), e, f Lateral view of the front leg of 12-week-old normal (?/om) and omi (om/om) mice. The deltoid tuberosity (de) and deltoid process (arrow) are depicted. Note the difference in morphology of the deltoid process between normal and omi mice (arrow), g, h Dorsal view of the hind leg of 12-week-old normal (?/ om) and omi (om/om) mice. In omi mice femurs are shorter but thicker and have increased density at the diaphysis (di) and the distal epiphysis (arrow) formed by compact bone were invaded by blood vessels and had a very irregular structure in om/om mice (Fig. 3d). In addition, the spongy bone appeared disorganised (Fig. 3f). The omi mutation maps to chromosome 10 To map the omi mutation, we crossed om/om mice with C57BL/6J mice. All offspring (n = 43) were normal (?/ om) (Table 5). These outcrossed mice were then backcrossed to an affected om/om mouse. A total of 182 backcross mice were collected; 97 (53.3 %) had the tooth defect and a grey appearance (om/om) and 85 (46.7 %) were normal in appearance (?/om) ( Table 5). These crosses indicate an autosomal recessive inheritance. DNA from 28 backcross offspring that had the tooth defect and a grey appearance was used to identify chromosome/trait linkage. Analysis of 69 polymorphic markers (Supplementary Table 1) distributed throughout the autosomes indicated clear linkage of the omi phenotype to chromosome 10 (data not shown). The highest percentage (92 %) of homozygosity for the C3H-type polymorphism was found at marker D10Mit106. To narrow down the region, Fig. 3 Analysis of tooth and bone abnormalities by histology in 3-week-old normal (?/om) and om/om mice. a, b Sections through molars of normal (?/ om) and omi (om/om) mice. In normal mice molars (mo) have erupted and are located in the buccal cavity, whereas the molars of om/om mice remain in the jaw under several tissue layers, c, d Sections through incisors (in) of normal and om/ om mice. The maxillary bone of normal mice has a compacted and regular structure (compacted bone, cb) with no blood vessels. In om/om mice the bone is disorganized (arrowhead) and has many blood vessels penetrating (arrow). Scale bar 100 lm Missing teeth a 2 9 11 No teeth 0 0 2 All control mice had normal teeth at all time points. Incisors erupted in most om/om mice but degenerated after eruption a Either top or bottom had teeth missing. Of these, six animals had two lower incisors, four animals had one lower incisor, one animal had one top and one lower incisor, and none had only top incisors we used additional markers on chromosome 10 of the same 28 backcross animals (Supplementary Table 2; data not shown). This narrowed the interval to the region of chromosome 10 between D10Mit214 and D10Mit194, with the highest percentage of homozygosity at D10Mit3 (100 %). An additional set of 44 om/om backcross samples was used to confirm the data and to narrow the genomic interval. The haplotypes of the animals defining this critical region are shown in Fig. 3b. One animal was heterozygote for D10Mit3 but homozygote for D10Mit194, narrowing the region to this interval. This corresponds to a 17.6-Mb physical region from 28.8 to 46.5 Mb on chromosome 10. We then used four SNPs polymorphic between C57 and C3H that are located in the region between D10Mit3 and D10Mit194 (Fig. 4). The six animals that set the critical interval were used (Fig. 4). Two animals (S7 and S144) were heterozygote for C3H and C57 alleles at SNPs rs29330142, rs13480575, and rs13480578. All animals were homozygous for the C3H allele of SNP rs8244299 at 43.0 Mb. Therefore the omi mutation lies in the 12-Mb region between SNP rs13480578 and D10Mit194. Omi and grey-lethal are noncomplementing mutations The region on chromosome 10 between SNP rs13480578 and D10Mit194 contains approximately 100 genes. Several genes in the region could lead to a severe phenotype as seen in the om/om mice, but one gene was of particular interest. Ostm1 mutations have been identified in mouse and human as causing very severe osteopetrosis. Grey- Fig. 4 Genetic mapping of the omi mutation. A genome-wide association scan using Mit markers located the mutation on chromosome 10. Fine mapping using Mit markers and SNPs polymorphic between C3H and C57 mice showed that the mutation lies between rs13480578 and D10Mit194. S(number) sample name, black sample is heterozygous for a marker, white sample is homozygous for a marker, Mb position of marker on chromosome 10 in megabases lethal (gl) mice have a recessive null mutation in Ostm1 that leads to a grey coat colour and early lethality due to severe osteopetrosis (Chalhoub et al. 2003;Pangrazio et al. 2006). Due to the similarity in phenotype between the grey-lethal and omi phenotypes, we tested for noncomplementation. Offspring of crosses between om/om and ?/gl mice were collected. Of 28 pups, 10 had a grey appearance and lacked incisors at P21 and the remaining pups were all normal (Fig. 5a, b). These ten abnormal pups were confirmed to be heterozygous for the grey-lethal allele by genotyping (and thus ?/gl, ?/om). In contrast to gl/gl mice, that are small at P21 and lethal around 3-4 weeks of age, these compound heterozygous mice were viable on a Pico-Vac Ò soft diet for at least 12 weeks. om/om mice do not have a mutation in the coding sequence and 500-bp promoter region of Ostm1 but have reduced Ostm1 protein levels We sequenced the coding region of Ostm1 in om/om, ?/om, and control mice from wild-type C3HeB/FeJ colony (?/?) mice. We were unable to detect a mutation within the coding sequence or the 5 0 and 3 0 UTRs. Work by Meadows et al. (2007) has shown that Ostm1 expression is regulated by the MIcrophthalmia Transcription Factor (MITF). Analysis of the alignment of human and mouse areas upstream (500 bp) of the transcription start site identified one putative MITF binding site (M box) and three weaker binding sites (E boxes). In addition, Ets and Pu.1 binding sites were identified (Meadows et al. 2007). We sequenced this region for control and om/om mice and did not find any mutation in this promoter region. Next we analysed Ostm1 protein levels of the brains of ?/om, om/om, and gl/gl mice. The OSTM1 antibody recognized several bands ranging from 35 to 90 kDa, with the strongest band just above 64 kDa. This band was weaker in om/om mice and weakest in gl/gl mice compared to the loading control (Figs. 5,6). This suggests that om/om mice have less Ostm1 protein than normal mice but more than gl/gl mice. Discussion Here we described a novel allele (omi) of Ostm1 that results in tooth, bone, and coat colour abnormalities. All homozygote mutant (om/om) mice lose incisors in the first month of life and most molars fail to erupt. We have demonstrated that female om/om mice may suffer from progressive osteopetrosis at a young age. Male om/om mice have decreased BMD at 8 weeks of age, but bone mass increases significantly between 8 and 12 weeks of age, even though mice suffer from poor nutritional status (indicated by low triglycerides, LDL, reduced lean and fat mass). In addition, om/om mice have a shortened life span. It is possible that bone mass would increase further with age in om/om males and an osteopetrotic phenotype could underlie the shortened life span, but further detailed analysis of this complex bone phenotype is required. The omi mutation is likely to have been induced by ENU mutagenesis. A spontaneous mutation rate of 1.1 9 10 -8 per base per generation was estimated for the mouse (Drake et al. 1998), resulting in an average of 28.6 mutations per genome per generation. In contrast, the mutation rate induced by ENU can be approximately 10,000 times higher than the natural background mutation rate (Salinger and Justice 2008;http://cshprotocols.cshlp. org/content/2008/4/pdb.prot4985.full), for an average of 3.97 9 10 7 mutations in the first generation. After five generations of outcrossing with unexposed C3HeB/FeJ, one would expect 93.75 % of the mutations to have been diluted out, leaving around 2.48 9 10 5 ENU-induced Fig. 5 Genetic complementation. Homozygous omi (om/om) mice were crossed with heterozygous grey-lethal (?/gl) mice and offspring were analysed for gross morphological defects. a Three-week-old ?/ om; ?/? mice have normal teeth, b A 3-week-old littermate (?/ om; ?/gl) has missing incisors and coat colour abnormalities. Unlike gl/gl mice, these compound heterozygotes are viable until at least 12 weeks of age mutations. Therefore, it is likely that the mutation was ENU-induced, but we cannot exclude the possibility that it arose spontaneously. The phenotype of the om/om mice is similar but not identical to the previously reported grey-lethal mouse. Grey-lethal is a recessive mutation in Ostm1 but has a more severe phenotype than the om/om phenotype across all aspects of the phenotype. Incisors and molars do not erupt in grey-lethal mice. Some teeth do erupt in om/om mice but most are not maintained and are lost just after weaning. In vitro cultures of tooth germs showed that teeth of greylethal mice can develop normally, and the failure to erupt in vivo is due to an absence of bone remodelling (Ida-Yonemochi and Saku 2002). Grey-lethal mice are not viable and die around weaning time, even when the mice are on an appropriate soft diet, whereas omi mice are viable and fertile on a soft gel diet. The grey coat colour is another phenotype that is more severely affected in grey-lethal mice than in omi mice. The grey-lethal mutation has been bred onto various genetic backgrounds, including C3H and C57BL/6J. The phenotype of the grey-lethal mouse is less severe on a Mus spretus genetic background, indicating that modifier genes exist (Vacher and Bernard 1999). The omi mutation arose on an inbred C3HeB/FeJ genetic background and backcrossing to C57BL/6J did not alter the phenotype significantly. Grey-lethal was discovered by Grüneberg (1935) in a stock segregating for Tyr c-e . It arose as a spontaneous mutation in a mixed genetic background. Based on mapping data, Vacher and Bernard (1999) suggested that the mutation arose on a 129 or related genetic background. The grey-lethal mutation was identified as a genomic deletion of the 5 0 region of the gene. The deletion spans 7.5 kb and includes the promoter, the first exon, and part of the first intron. A 460-bp sequence corresponding to the 3 0 UTR of a LINE1 retrotransposon element was inserted at the deletion breakpoints (Chalhoub et al. 2003). RT-PCR showed that this Ostm1 mRNA is not expressed in gl/gl tissues, making it likely that this is a complete null mutation. One possibility is that the difference in the type of mutation could cause the difference in phenotype. Omi was identified in a mouse line that came from an ENU mutagenesis program. Therefore, it is likely to be a point mutation rather than a retroviral integration as is the greylethal mutation. Omi could be a hypomorphic allele rather than a complete null mutation, which would explain the difference in phenotype. Our data suggest that om/om mice have reduced levels of Ostm1 protein, confirming that it is not a null mutation. Our current hypothesis is that the omi mutation could lie in a regulatory element and lead to a reduction in Ostm1 expression. As enhancer elements have been identified at many kilobases up-or downstream from the transcription start site, further fine mapping of the omi mutation in combination with new sequencing technologies will be required to identify the causative mutation. Osteopetrosis in humans was described in the early 1900s by Albers-Schönberg, and by the 1940s it was clear that the severity of the phenotype varied greatly. In the most severe forms, deficient bone resorption leads to haematological failure, cranial nerve compression, short stature, and brittle bones. Various inheritance patterns have been described for the different types of osteopetrosis. Autosomal dominant osteopetrosis is variable but usually mild. Various forms of recessive osteopetrosis have been reported, some of which are associated with the most severe phenotypes (including those caused by mutations in the TCIRG1, CLC7, and OSTM1 genes), whereas mutations in other genes (CAII and PLEKHM1) give a milder osteopetrotic phenotype. Recently, long-term survival in infantile malignant autosomal recessive osteopetrosis secondary to homozygous p.Arg526Gln mutation in CLCN7 was described (Kantaputra et al. 2012). Interestingly, the phenotype of omi mice greatly overlaps the description of a case study of recessive mild osteopetrosis (Kahler et al. 1984). As in omi mice, patients presented with oligodontia, striation of the long bones, abnormalities of the zygomatic arch, sclerosis in the skull, and mild anaemia. To date, only four different human OSTM1 mutations have been Fig. 6 Ostm1 protein levels in ?/om, om/om, and gl/gl mice analysed by Western blotting. a Brain lysates of 3-week-old ?/om, om/om, and gl/gl mice were analysed by Western blotting for the Ostm1 protein. Blots were stripped and incubated with a PSD95 antibody to determine sample loading, b Genotypes of grey-lethal (gl/gl) samples were verified by PCR analysis identified, always leading to recessive malignant osteopetrosis. This does not exclude the possibility that other hypomorphic OSTM1 mutations could lead to a mild osteopetrotic phenotype in humans. The recent identification of a similar mutation in CLC7 (Kantaputra et al. 2012) confirms the idea that many more subtle mutations may be identified in the human population. Therapy for osteopetrosis is presently unsatisfactory and much work needs to done to unravel the gene defects and to identify new treatments to improve symptoms. Recent efforts to identify novel genes involved in bone homeostasis by using ENU mutagenesis screening in mice have identified one novel candidate for osteopenia (Barbaric et al. 2008) and one for osteopetrosis (Ochotny et al. 2011). Although omi mice have a mutation that affects a gene already known to be involved in bone homeostasis, it is likely to be hypomorphic making it a suitable model for validating novel therapeutic treatments.	7,448	2012-11-17T00:00:00.000	[ "Biology" ]
The Way of Pooling p-values Pooling p-values arises both in practical (in any science and engineering applications) and theoretical (statistical) issues. The p-value (sometimes p value) is a probability used as a statistical decision quantity: in practical applications, it is used to decide if an experimenter has to believe that his/her collected data confirm or disconfirm his/her hypothesis  about the “reality” of a phenomenon. It is a real number, determination of a Random Variable, uniformly distributed, related to the data provided by the measurement of a phenomenon. Almost all statistical software provides p-values when statistical hypotheses are considered, e.g. in Analysis of Variance and regression methods. Combining the p-values from various samples is crucial, because the number of degrees of freedom (df) of the samples we want to combine is influencing our decision: forgetting this can have dangerous consequences. One way of pooling p-values is provided by a formula of Fisher; unfortunately, this method does not consider the number of degrees of freedom. We will show other ways of doing that and we will prove that theory is more important than any formula which does not consider the phenomenon on which we have to decide: the distribution of the Random Variables is fundamental in order to pool data from various samples. Manager, professors and scholars should remember Deming’s profound knowledge and Juran’s ideas; profound knowledge means “understanding variation (type of variation)” in any process, production or managerial; not understanding variation causes cost of poor quality (more than 80% of sales value) and do not permits a real improvement. Introduction Let's consider a "continuous" probability distribution F(t)=P[Tt], T being a random variable. U=F -1 (T) is a random variable uniformly distributed: G(u)=P [Uu]. pvalue is the probability 1-G(u)=P [U>u], WHEN F(t) is related to a statistical test THEN the p-value is a result of a statistical test. The p-value (sometimes p value) [1,2] is a statistical decision quantity: it is used, in practice, to decide if we have to believe that our data confirm or disconfirm our hypothesis about the "reality" of a phenomenon. It is a real number, determination of a Random Variab le related to the data provided by the measurement of a phenomenon. In this section, we provide the general ideas about Hypothesis Testing, while in section 2 we describe the pvalues. We consider here only the parametric version of the method. We ask the reader to look at figure 1, which shows a typical situation that anybody can be confronted with. For example, let's assume that we want to decide about the following simple case : We want to test if people weigh more than 75 kg, assuming that we want to risk only =5% and assuming that the standard deviation of the weight is 8 kg. Connect this case to figure 1 which refers to any general setting related "Hypothesis Testing". There you see the (above mentioned) risk  (not its value 5%) related to the Hypothesis H0 (left unspecified in the case) and the hypothesis H1=(not shown the value) people weigh (in mean) more than 75 kg; regarding the Probability Model the information we have is limited to the standard deviation is 8 kg. Before arriving to the decision based on the data in our hand, we describe the general framework of figure 1. Any statistical decision is referred to a Probability Model assumed (either on technical or theoretical reasons) to rule the data we are going to collect. The Probability Model is generally specified by the Cumulative Distribution which depends on various "parameters". To explain figure 1, we consider only 1 parameter. Let  be the parameter we want to "test"; previous to any collection of data we should state two Hypotheses and a probability , named the "significance level": 1. The "Null Hypothesis", named H0, where we assume, before any collection of data, a value for the parameter ; we indicate it with the symbol 0; 0 is a number, while  is the symbol of the parameter: we write H0: [=0] 2. The "Alternative Hypothesis", named H1, where we assume, before any collection of data, another value for the parameter ; we indicate it with the symbol 1; 1 is a number different from 0, while  is the symbol of the parameter: we write H1: [=1] 3. The probability , the "significance level" that we assume, before any collection of data and analysis of the data, is the probability that we accept of being wrong IF we, after the collection and the analysis of the data, claim the Null Hypothesis H0: [=0] is Rejected, when actually (and nobody knows it!) the Null Hypothesis H0: [=0] should not be Rejected . From the points 1, 2, 3, using the Theory, we can, before any collection and analysis of the data, find two items:  A "formula", named Test Statistics s, that will provide us with a number, after the analysis of the data  And an interval of the real line (real numbers) C, named Critical Region (or Rejection Region) such that we Reject "the Null Hypothesis H0: Let's assume that we collect the data and analyse them, according to the Theory (probability and statistic), and compute the number s (computed from the Decision Function); IF s  C, then we, according to the Theory, should Reject the Null Hypothesis H0: [=0]; IF s  C, we Accept [but really we do not reject] the Null Hypothesis H0: [=0]. NOTICE that the Test Statistics s is the "determination (=estimate)" of the estimator S, which is a Random Variable! In figure 1 it is clearly shown that, in order to take a decision, we need the Probability Model (on the top of the figure); the model has to be suitable to the analysis of our data for the parameter we want to test! Moreover, above the yellow box, you see the Hypothesis H1 and not the risk ; this depends on the fact that in various cases  is not stated and one wants to compute the power of the test 1-. For example, in the above case, the parameter we want to test is the "mean weight" ; we have to assume a Probability Model depending from the parameter  (and other parameters). Deming and Juran [3,4,5] ("It is a hazard to copy". "It is necessary to understand the theory of what one wishes to do or to make". "Without theory, experience has no meaning". "A figure without a theory tells nothing". «The result is that hundreds of people are learning what is wrong. I make this statement on the basis of experience, seeing every day the devastating effects of incompetent teaching and faulty applications») and Galetto [ Notice that the sample size n can be computed when both risks  (related to H0) and  (related to H1) are stated. If  is not stated as in figure 1, the sample size is defined by the experimenter. From these last two points, according to the Theory, anybody can derive that the Probability Model for decision; since we assumed N(,  2 ) as Probability Model of the data, in figure 1 the Probability Model for decision is the Normal distribution with mean  and variance  2 /n, N(, 2 /n). p-values In section 1, we saw the general ideas of Testing of Hypothesis (see figure1). The decision there depends on the Test Statistic s and on the Critical Region C which is related to the risk  (significance level): Test Statistic s and Critical Region C were the two statistical decision quantities. The Test Statistic s is a real number, determination of a Random Variable S, depending on the Probability Model. Now we see a different statistical decision quantity: the p-value (sometimes p value) [1,2]. It is used to decide if we have to believe that our data confirm or disconfirm our hypothesis about the "reality" of a phenomenon. It is a real number, determination of a Random Variable related to the data provided by the measurement of a phenomenon. For the same purpose almost every statistical software provides the p-value (probability value). In our example, it is the probability that the Random Variable Mean (indicated generally with ̅ ) is greater than the empirical (observed) mean ̅ = 79.25, that is From the example we derive the definition of the p-value: it is the probability, under the Null Hypothesis H0 (opposed to the Alternative Hypothesis H1 about the distribution of the random variable), that the variate (RV) to be observed takes values equal to or more extreme than the value observed. The p-value can be viewed as an index of the "strength of evidence" against the Null Hypothesis H0; a small pvalue indicates an unlikely event and, hence, an unlikely hypothesis. Let f(x) be the pdf of the estimator T and t the determination of T; after the elaboration of the collected data, we can compute the integral which is a real number. The value of the integral  is the determination of the Random Variable , related to the  "parameter p-value". The author thinks the opposite. Let's see why. IF, with other 25 data, ̅ = 77.75 then ̅ ∈ C and, AGAIN in this new case, we reject H0 with 5% significance level, same as before. BUT, from (1), the new p-value (estimation of the same parameter p-value ) is 0.0428: the "strength of evidence" is less than before… What is the true p-value for H0=people weigh (in mean) 75 kg=0 based on both samples? We do not know. What is the true significance level? =0.05 for both decisions! Notice that the example used the normal distribution, and the decision function depended on that. For example, the same numerical procedure cannot be used (but Six Sigma Professionals do not know it) for the following case: Pooling p-values To explain the point, we consider the data of table 1, drawn form a reliability test on the same type of systems. They are times to failure (hours) of 5 items in sample 1 (sample size 13) and on 10 items in sample 2 (sample size 17). According to figure 1, we set H0=[0<100 h] where  is the MTTF (Mean Time To Failure of the items) and =0.05; the probability model depends on the exponential distribution; the alternative hypothesis is H1=[1>100 h]. We have to find the test statistic and the Critical Region. According [7][8][9][10][11][12][13][14][15][16], the test statistics is the "Total Time On Test" t1, the determination of the RV T1(t); "Total Time On Test" is the sum of all the data of the items on test. In this case, the Probability Model for decision f(x), in figure 1, is the Erlang distribution with mean 5, and the decision is Reject H0, because the Critical Region is C=>915.3 and t1=925. 3 is a Random Variable. We call Π the "Random Variable p-value". The p-value 0.047 is the estimate of the  "parameter pvalue". Consider secondly sample 2; there are 10 failures and 17 items on test. The p-value is 0.063=∫ f(x) ∞ 1521.5 dx. This confirms that we have to Accept H0, because 0.063>0. 05 We have two contradictory decisions with risk 5%, because different samples (as obvious) provide different data and any statistical analysis depends on the data. Notice that we used the same value =0.05 for both tests. Since we have two independent comparisons, we have to consider the Bonferroni Correction: in order to avoid a lot of spurious positives, the alfa value needs to be lowered to account for the number of comparisons being performed. In this case, we have two comparisons so that modified=0.025. With this modification, we have to Accept H0, because 0.047>0.025 and 0.063>0.025; with this ideas the "two decisions" are no longer contradictory. To decide which of the "two (contradictory) decisions" could be the right one we think to test a new hypothesis According [7][8][9][10][11][12][13][14][15][16] the test statistics is the "Ratio of the Total Time On Test t1/t2,", the determination of the RV T1(t)/T2(t); in this case the Probability Model for decision f(x), in figure 1 Hence we should Reject H0…. The two samples (combined) tell us that the "true mean" should be >100 h. Comparison of "Pooling p-values" and Theory In this section, we test a new "Null Hypothesis" H0=Fisher's method and Reliability Theory provide the same decisions about combining two samples as in § 2, based on exponential distributed data, with 5 failures out of 13 items and with 10 failures out of 17 items . The dfs are important both for exponential data and for normal data. We consider these two cases. Using Fisher's method, two small p-values P1 and P2 combine to form a smaller p-value. The yellow-green boundary defines the region where the "combined" p-value is below 0.05. For example, if both p-values are around 0.10, or if one is around 0.04 and one is around 0.25, the "combined" p-value is around 0.05. The comparison of Fisher's method and the Theory [7][8][9][10][11][12][13][14][15][16] is shown in figure 3. It is clear that Fisher's p-value is higher than the value computed from Theory, taking into account the dfs! That means wrong decisions depending on the data… We should reject the "Null Hypothesis" H0=Fisher's method and Reliability Theory provide the same decisions about combining two samples as in § 3, based on exponentially distributed data, with 5 failures out of 13 items and with 10 failures out of 17 items and we should reject the "Null Hypothesis" H0=Fisher's method and Reliability Theory provide the same decisions about combining two samples as in § 3, based on Normally distributed data, with 5 failures out of 13 items and with 10 failures out of 17 items. The same for others distributions of the data and different numbers of degrees of freedom. This is in line with W. E. Deming's "profound knowledge" [3,4] and Juran's ideas [5]. Conclusions We saw that Theory is more important than formulae which do not consider the phenomenon on which we have to decide (distribution of the data). The number of degrees of freedom of the samples we want to combine is influencing our decision: forgetting this can have dangerous consequences. Manager, professors and scholars should remember	3,319.4	2020-03-01T00:00:00.000	[ "Mathematics" ]
A novel deep proteomic approach in human skeletal muscle unveils distinct molecular signatures affected by aging and resistance training The skeletal muscle proteome alterations to aging and resistance training have been reported in prior studies. However, conventional proteomics in skeletal muscle typically yields wide protein abundance ranges that mask the detection of lowly expressed proteins. Thus, we adopted a novel deep proteomics approach whereby myofibril (MyoF) and non-MyoF fractions were separately subjected to protein corona nanoparticle complex formation prior to digestion and Liquid Chromatography Mass Spectrometry (LC-MS). Specifically, we investigated MyoF and non-MyoF proteomic profiles of the vastus lateralis muscle of younger (Y, 22±2 years old; n=5) and middle-aged participants (MA, 56±8 years old; n=6). Additionally, MA muscle was analyzed following eight weeks of resistance training (RT, 2d/week). Across all participants, the number of non-MyoF proteins detected averaged to be 5,645±266 (range: 4,888–5,987) and the number of MyoF proteins detected averaged to be 2,611±326 (range: 1,944–3,101). Differences in the non-MyoF (8.4%) and MyoF (2.5%) proteomes were evident between age cohorts, and most differentially expressed non-MyoF proteins (447/543) were more enriched in MA versus Y. Biological processes in the non-MyoF fraction were predicted to be operative in MA versus Y including increased cellular stress, mRNA splicing, translation elongation, and ubiquitin-mediated proteolysis. RT in MA participants only altered ~0.3% of MyoF and ~1.0% of non-MyoF proteomes. In summary, aging and RT predominantly affect non-contractile proteins in skeletal muscle. Additionally, marginal proteome adaptations with RT suggest more rigorous training may stimulate more robust effects or that RT, regardless of age, subtly alters basal state skeletal muscle protein abundances. INTRODUCTION Aging adversely affects skeletal muscle physiology as evidenced by a reduction in muscle stem (or satellite) cell content, a loss of myofibrillar protein, and a loss in motor units and myofibers [1][2][3][4].Muscle aging is also associated with impairments in mitochondrial function, dysfunctional redox balance, and altered proteostasis [5][6][7].The culmination of these events likely contributes to a loss in muscle mass, which according to a recent AGING review, is accelerated in all the body's musculature past the age of 50 years old [8]. Resistance training can reverse certain aspects of skeletal muscle aging.For instance, weeks to months of resistance training in older participants has been shown to increase tissue-level and myofiber hypertrophy and muscle satellite cell content [9].Resistance training also catalyzes skeletal muscle mitochondrial biogenesis and remodeling in older participants [10][11][12][13][14], and weeks to months of resistance training alters nuclear and mitochondrial DNA methylation patterns in older participants which may lead to "rejuvenating" effects on global mRNA expression patterns [11,15,16]. Proteomic investigations intend to examine the entire detectable protein expression signature of a given tissue under various experimental conditions.While other -omics-based approaches exist (i.e., genomics, epigenomics, transcriptomics, and metabolomics), it has been posited that proteomic signatures likely best translate to cellular and tissue phenotypes [17].Past proteomic investigations have provided tremendous insight as to how myofiber type, aging, and exercise training affect the skeletal muscle molecular milieu [1,14,[18][19][20][21][22].Notwithstanding, skeletal muscle-based proteomics poses technical challenges.For instance, skeletal muscle tissue processing with general lysis buffers results in the clearance of insoluble (e.g., contractile) proteins [23], and if standard bottom-up proteomics is employed thereafter, the relative abundances of these proteins will ultimately be inaccurate.Even if care is taken in isolating the poorly soluble contractile and soluble non-contractile protein fractions, another pitfall lies in lowly-abundant proteins being masked by highly abundant proteins in each fraction [1].Single fiber isolation techniques have increased the depth of proteins detected [21,24].However, certain disadvantages with this method exist including the burdensome process of tissue digestion and fiber dissection, the need for pooling myofibers to obtain adequate protein for proteomics, and the inability to detect proteins enriched in the extracellular matrix or stromal cells. A novel deep proteomics approach in human plasma was recently published whereby unique nano-bio interaction properties of multiple magnetic nanoparticles (NPs) was leveraged for automated protein separation (referred to as the Proteograph assay; Seer, Inc. Redwood, CA, USA) [25].Downstream digestion followed by liquid chromatography coupled to mass spectrometry (LC-MS) analyses enabled the identification of over 2,000 plasma proteins and this provided approximately a 10fold increase in depth compared to prior studies that utilized other strategies to deplete plasma of highly abundant proteins [26,27].However, this approach has not been performed in human skeletal muscle.Thus, we sought to leverage this technology, along with our prior method of muscle tissue fractionation [23], to examine the proteomic signatures of the myofibrillar (MyoF) and non-myofibrillar (non-MyoF) fractions from the vastus lateralis (VL) muscle of a subset of younger (Y, 22±2 years old, n=5, all of whom were females) and middleaged (MA, 56±8 years old, n=6, 3 men and 3 women) participants.We also sought to determine how eight weeks of unilateral knee extensor resistance training affected the MyoF and non-MyoF proteomic signatures in the MA cohort.Given some of our past work in this area [1], we hypothesized that more non-MyoF proteins would be altered by aging when comparing MA and Y participants.We also hypothesized that resistance training in MA participants would affect more non-MyoF versus MyoF proteins.However, we did not adopt an a priori hypothesis regarding which proteins or biological processes would be affected between comparisons given the novelty of interrogating skeletal muscle using the Proteograph assay. Characteristics of proteins identified in the MyoF and non-MyoF fractions A total of 6,445 non-MyoF proteins and 4,421 MyoF proteins were identified in at least one participant (Figure 1A).Across all participants, the number of non-MyoF proteins detected averaged to be 5,645 ± 266 (range: 4,888-5,987) and the number of MyoF proteins detected averaged to be 2,611 ± 326 (range: 1,944-3,101).Notably, there was a numerically lower number of MyoF proteins detected in MA (preintervention) versus Y participants that approached statistical significance (p=0.066), and a non-significant greater number of non-MyoF proteins detected in MA (pre-intervention) versus Y participants (p=0.112; Figure 1C). MYH isoform peptide identification information Myosin heavy chain isoforms have been intensely studied in human skeletal muscle for fiber typing purposes and prominent isoforms include the slowtwitch type I isoform (encoded by the MYH7 gene) as well as the fast-twitch IIA (encoded by the MYH2 gene) and IIX (encoded by the MYH1 gene) isoforms [28].However, other MYH isoforms were highly enriched in the MyoF fraction according to data presented in Figure 1C.Because of this, we opted to provide the peptide sequences used for detecting some of these isoforms in Table 1 below. Alternative proteins isoforms in the MyoF and non-MyoF fractions of MA and Y The enhanced depth of detection provided by proteomics revealed the presence of numerous isoforms in both protein fractions; specifically, there were 175 isoforms for 82 MyoF proteins and 375 isoforms for 173 non-MyoF proteins.The MyoF proteins with the most isoforms included titin (TTN), myosin-binding protein C (MYBPC), and MICOS complex subunit MIC60 (IMMT); each of these targets had four isoforms detected.Given the vast research interest in titin [29,30], associated isoform data are plotted in Figure 3A; notably no significant aging or training effects were noted (p>0.01 for all comparisons).The non-MyoF proteins with the most isoforms included Gelsolin (GSN, four isoforms), IMMT (4 isoforms), and Reticulon-4 (4 isoforms); again, no significant age effects were noted for these targets (data not plotted). We plotted significantly different alternative protein isoform abundances between Y and MA in both protein fractions given that bioinformatics on the non-MyoF fraction indicated "regulation of RNA splicing" (GO:0043484) was predicted to be upregulated in the older cohort (Figure 3B, 3C).MA (pre-intervention) Legend: these data contain the peptide sequences used for alignment to identify the several highly abundant myosin heavy chain isoforms in the MyoF fraction. versus Y comparisons indicated that only three alternative MyoF protein isoforms were different between age groups (all higher in Y, p<0.01).However, 14 alternative non-MyoF protein isoforms were significantly different between age groups (11 higher in MA, p<0.01), and while not depicted in Figure 3C, 32 additional alternative non-MyoF protein isoforms were numerically different between age groups (25 higher in MA, p<0.05).Further, when examining the abundances of small nuclear ribonucleoproteins belonging to spliceosome complexes in the non-MyoF fraction, three reached the p<0.01 significance threshold as being more enriched in MA (pre-intervention) versus Y participants (SNRP200, SNRPE, SNRPF), and several others were numerically greater in MA participants (SNRP40/70/A/A1/B/C/D2/D3, p<0.05; Figure 3D).Notably, training did not alter the expression of any SNRP in Figure 3D (p>0.05 for all), and most of these proteins were not detected in the MyoF fraction. MyoF and non-MyoF protein differences prior to and following resistance training in MA participants In MA, knee extensor resistance training significantly altered 13 MyoF proteins (11 upregulated and two downregulated, p<0.01; Figure 4A), and 64 non-MyoF proteins (56 upregulated and eight downregulated, p<0.01; Figure 4B, 4C).These alterations represented ~0.3% of the MyoF proteome and ~1.0% of the non-MyoF proteome.Bioinformatics within each fraction were attempted, albeit no pathways were predicted to be significantly affected. Proteolysis targets manually interrogated in both fractions Based on bioinformatics indicating that proteostasis was predicted to be altered with aging (Table 3), we manually interrogated proteolysis-related protein targets (i.e., calpain-1/2, and the summed spectra of 26S proteasome subunits).Figure 5 DISCUSSION Using a novel analytical approach, we examined the deep proteomic signatures of the MyoF and non-MyoF fractions in younger adults as well as middleaged participants before and after eight weeks of knee extensor resistance training.More non-MyoF proteins differed between age cohorts compared to MyoF proteins (8.4% versus 2.5% of the respective protein pools).More non-MyoF proteins (447/543) were also more highly abundant in MA versus Y and bioinformatics predicted that several biological processes were more operative in the older participants.A greater abundance in alternative variants, proteins associated with spliceosomes, and proteolysis-related proteins were also evident in the non-MyoF fraction of MA versus Y, and these observations corroborated certain bioinformatics findings.Although resistance training in MA non-significantly increased VL crosssectional area (+6.5%) and significantly increased knee extensor strength (+8.7%), training marginally affected the MyoF and non-MyoF proteomes and no biological processes were predicted to be affected in either fraction.These findings will be expanded upon in the paragraphs below. As stated above, several studies have performed proteomic analyses on skeletal muscle to compare molecular signatures that exist between younger and older adults or to examine how resistance training affects this aspect of the muscle-molecular milieu [1, 14,[18][19][20]22].The novelty of the current study was the proteomics approach utilized and the knowledge gained relative to these prior investigations. Fractionation of muscle into solubilized MyoF and non-MyoF homogenates enabled the detection of unique proteins in each fraction, which has only been attempted in one other study to our knowledge [1].In this prior study, we performed bottom-up LC-MS-based proteomics on each fraction from younger resistancetrained, younger untrained, and older untrained men (n=6 per group).We identified a total of 810 proteins in both fractions that were expressed in at least one participant.In the current study we detected a total of 10,866 proteins in both fractions.While most of identified proteins were present in both fractions, we were able to identify 2,217 unique non-MyoF proteins and 193 unique MyoF proteins.This robust increase in detection depth (~13.4-fold) is insightful for numerous reasons.First, it was revealed that metabolic enzymes constituted the top class of proteins in both fractions as well as proteins that overlapped in both fractions.Hence, although myosin heavy chain isoforms, troponin, titin, and actin were the most enriched in the MyoF fraction, these data counter the notion that the MyoF fraction contains mainly contractile proteins.Several nuclear proteins were also identified in the MyoF fraction (e.g., histones and other chromatinbinding proteins) indicating that our MyoF isolation method likely pellets nuclei.Finally, we were able to identify numerous proteins that are not commonly reported in previous skeletal muscle proteomic studies (see Table 1 and Figure 2 for example).To this end, several MYH isoforms beyond the three common 7/2/1 isoforms were highly enriched in the MyoF fraction, and this may be due to the persistence of non-conventional or developmental isoforms in certain Notes: Certain p-values not meeting the significance criteria for these data were presented due to the visual differences observed between cohorts.AGING regions of adult myofibers or in transitioning fibers as discussed by Schiaffino et al. [31].Both fractions contained most of the large (~40) and small (~30) ribosomal subunit proteins, mitochondrial oxidative phosphorylation enzymes, and mitochondrial ribosomal proteins (>70), all of which likely represents the presence of sarcoplasmic and intermyofibrillar mitochondria.Both fractions also contained numerous alternative isoforms for several proteins, various transcription factors (e.g., MEF2D, F-box proteins, SMAD1/2/3, NFAT isoforms, and several others), DNA and RNA polymerase subunits, various growth factors and their receptors (e.g., EGFR, VEGFA, FGF2/13, PDGFRa/b, TGFB1/2, and others), dozens of eukaryotic initiation/ elongation factors, nearly 50 solute carrier family member proteins (i.e., nutrient and metabolite transporters), cytoplasmic and mitochondrial aminoacyl tRNA ligases, and hundreds of signal transduction proteins of interest to skeletal muscle biologists (e.g., mTOR, RPTOR, p70s6k, AMPK subunits, cyclindependent kinases and inhibitors, and others).We believe this enhanced level of detection was due to muscle fractionation, and more importantly, the utilization of NPs prior to mass spectrometry.Indeed, this same contention has been posited by others using this technology to increase detection depth of circulating proteins in human plasma [25,32]. Notable MyoF and non-MyoF proteome signatures between age groups were also evident.For instance, aging seemingly affects the non-MyoF protein signature more so than the MyoF fraction.This finding agrees with our past proteomic study where we reported that 37 non-MyoF proteins (versus only 18 MyoF proteins) were differentially expressed between college-aged and older men (mean age 62 years old) [1].However, the increased detection depth in the current study indicated that 112 of the 4,421 identified MyoF proteins met the p<0.01 significance threshold between age cohorts.Moreover, all but one of these proteins (TMPO, Lamina-associated polypeptide 2 isoforms beta/gamma) were greater in Y versus MA indicating either a loss or decreased expression of ~2.5% proteins belonging to the MyoF fraction.Although no biological processes were predicted to be affected between age cohorts based on this list of 111 proteins, several of these targets were notable.For example, three heat shock proteins (HSPB1, HSP90AB1, HSPD1) were more lowly abundant in MA participants and this agrees in principle with past literature indicating the expression of heat shock proteins in skeletal muscle is dysregulated with aging [33,34].Several MyoF mitochondrial proteins were also lower in MA participants (DECR1, NDUFS8, ETFDH, GPD2, STOML2, ALDH2, COQ5, IARS2, ALDH1B1, MRPL32).This also agrees with past literature indicating either a decrease in mitochondrial volume density or decreased mitochondrial function with skeletal muscle aging [35], and more specifically agrees with a study by Callahan et al. [36] who reported that older participants presented reductions in the size of intermyofibrillar mitochondria. Strikingly, 543 proteins (~8.4%) of the non-MyoF proteome were different between MA and Y participants, and unlike the trends observed in the MyoF fraction, most of these proteins (447) were significantly enriched in the MA cohort.These figures agree with a proteomic investigation by Robinson et al. [14] who reported that more muscle proteins (220/347) were higher in untrained older versus younger individuals prior to a period of exercise training.These robust differences between age cohorts in the current study also revealed that several biological processes were predicted to be upregulated in MA participants.Some of these processes either contradict each other or agree with past literature reporting similar aging phenotypes. Regarding the former, while more proteins associated with translation elongation were more abundant in MA participants (which would potentially promote muscle anabolism), proteins associated with ubiquitin-mediated proteolysis were also more enriched.One interpretation of these data could be that aging increases skeletal muscle protein turnover.However, this is likely not the case given that a variety of studies ranging from human tracer studies to nematode models have indicated that protein turnover in response to feeding or in a basal state is impaired with aging [37,38].This aspect of our data also agrees with a report by Ubaida-Mohien et al. [22] who showed that 31% of proteins related to proteostasis were altered with age (24 underrepresented and 50 overrepresented, p<0.05) in healthy older versus younger adults.Hence, we posit that proteins associated with these processes may have been more abundant in MA participants in a compensatory attempt to counter age-related declines in muscle protein turnover.It is also interesting that proteasome subunits and calpains were more enriched in the MyoF fraction of Y versus MA, whereas these same proteins were more enriched in the non-MyoF MA versus Y.While speculative, an enrichment of proteolytic proteins in the MyoF fraction might play a role in functional proteostasis, while an enrichment of these proteins in the non-MyoF fraction might be indicative of a gradual dysregulation in proteostasis. Non-MyoF proteins associated with mRNA export and splicing were also elevated in MA participants, and this agrees with other reports.For instance, a recent review by Park et al. [39] cites a variety of cell culture evidence to suggest that the nuclear pore complex is disrupted with aging and that this leads to a dysregulation in mRNA export.A rodent study by Mobley et al. [40] AGING also suggests that mRNA levels linearly decrease in skeletal muscle with increasing age.Hence, again, a higher abundance in non-MyoF proteins associated with mRNA export could also be a compensatory response to offset these age-associated effects burdened by myonuclei.The greater abundance of proteins associated with mRNA splicing is striking and agrees in principle with a report by Rodriguez et al. [41] who showed that the skeletal muscle of aged mice possessed ~4 times more RNA splice variants than younger counterparts.Our data also agree with the abovementioned proteomics report by Ubaida-Mohien et al. [22] who showed that proteins related to alternative splicing were more abundant in healthy older (versus younger) adults.We also performed a follow-up analysis showing that several small nuclear ribonucleoproteins (SNRPs, or snRNPs) that make up spliceosomes and alternative non-MyoF protein isoforms (indicative of increased spliceosome activity) were elevated in MA versus Y participants.This is particularly insightful given that dysfunctional spliceosome activity and the aberrant RNA and protein expression of splice variants have been linked to age-associated maladies such as cellular senescence [42,43].Hence, these independent reports of ageassociated increases in splice RNA and protein variants, along with the current data suggesting that the relative abundances of spliceosomes are greater in older participants, warrant future research elucidating the causes and consequences of this phenomenon. A final noteworthy topic was the observation of marginal MyoF and non-MyoF alterations with eight weeks of knee extensor training in MA participants.Although this may have been due to the modest training regimen that only lasted eight weeks in duration, limited proteome plasticity with aging and/or the limited ability of resistance training to alter the muscle proteome could also be plausible explanations.Support for both phenomena come from Robinson et al. [14] who employed deep proteomics to report that ~200 muscle proteins were altered in older participants after 12 weeks of resistance training (p<0.05), and this was less than the ~300 proteins that were altered in the younger participants.Hence, an aging effect was noted.However, assuming the authors identified >3,000 muscle proteins, which was not reported to our knowledge, this represents less than 10% of the detectable proteome being altered with resistance training.Deane et al. [13] used a different proteomics approach to examine the non-MyoF proteome adaptations in older participants following 20 weeks of resistance training in younger and older adults.Although their depth of detection was limited to ~160 proteins, resistance training only increased five non-MyoF proteins in older participants when a p<0.05 significance threshold was employed (i.e., ~3% of the detectable proteome).Interestingly, this effect was not confined to older participants given that the younger participants in their study only presented an elevation in four non-MyoF proteins with training.Our laboratory also used proteomics to examine non-MyoF protein adaptations in college-aged men following ten weeks of resistance training [1].Only 13 proteins were shown to be altered with training (12 up, one down, p<0.05) and this represented ~3.4% of the detectable non-MyoF proteome.Hence, these two latter studies do not support the aging hypothesis and, instead, provide evidence of limited muscle proteome plasticity with resistance training.Despite marginal non-MyoF proteome alterations in MA participants with training, there were interesting targets that were altered.For instance, the UBR7 E3 ligase was upregulated, and recent evidence suggests that an E3 ligase in this same protein family (UBR5) is required for load-induced skeletal muscle hypertrophy [44].Additionally, the knockdown of another member of this family (UBR4) promotes hypertrophy in Drosophila and mice [45].SRC was upregulated and this non-receptor tyrosine kinase has been implicated in interacting with vitamin D to promote anabolic signaling in skeletal muscle [46].HDAC4 was upregulated and this adaptation could be operative in muscle-metabolic adaptations and ultrastructural remodeling to resistance training given that HDACs have been implicated in controlling the expression of various metabolic and contractile protein genes [47,48].Two protein phosphatases were also upregulated (PPP1R12A and PIP4P2), and both have been shown to be involved in aspects of insulin and growth factor signaling [49,50]. There are notable limitations to the current study.First, the MA and Y participant pools were small, and Y participants were all women.Furthermore, the lack of training data in Y participants to examine differential age-associated training effects is a limitation.We lacked remaining skeletal muscle to perform downstream analyses (e.g., examining RNA splice variants and/or proteasome activity assays) which may have provided additional insight.Moreover, having additional age groups (e.g., older participants) would have provided more depth to the current study.Despite these limitations, the primary objective of this publication was to feature our novel proteomic approach as we believe that this will add tremendous insight into the field of skeletal muscle biology.Likewise, the data from this study can be used to generate hypotheses for other age-related or resistance training proteomic or targeted protein approaches moving forward. In conclusion, we provide preliminary evidence to support that muscle aging predominantly affects the non-MyoF protein pool and that this is associated with AGING biological processes which may act to counteract dysfunctional cellular homeostasis.We also provide preliminary evidence of limited MyoF and non-MyoF proteome plasticity to shorter-term resistance training in middle-aged participants, and this agrees with prior proteomic investigations.Finally, and most importantly, we believe that the utilization of skeletal muscle tissue fractionation protocols and NP-based protein corona formation prior to downstream proteomics has the potential to add incredible insight in identifying novel protein targets affected by exercise training, aging, and various disease states. Ethical approval and study design Muscle specimens were obtained from two studies whereby approval was obtained from the Auburn University Institutional Review Board.The first protocol in untrained MA participants (approved protocol #21-461 MR 2110) involved investigating the effects of a dietary supplement (312 mg of combined Wasabia japonica extract, theacrine, and copper (I) niacin chelate) versus a placebo on potential blood marker responses over an eight-week period.A unilateral leg resistance training (two days/week) protocol was implemented to perform non-supplementation secondary analyses as presented herein.The six MA participants included in the current study were in the placebo group; thus, no confounding effects of dietary supplementation were expected.Additionally, both the pre-intervention and post-intervention biopsies were collected the same time of day following at least a four-hour fast, and the post-intervention muscle biopsy was collected 72 hours following the last training bout.Y participant muscle tissue was banked from a prior study examining how ten weeks of daily peanut protein supplementation affected resistance training outcomes in untrained individuals (approved protocol #19-249 MR 1907) [51].Notably, muscle tissue from these participants was collected in the basal state prior to the intervention.Hence, again, there were no potential confounding effects of supplementation.Study procedures for both projects were in accordance with the most recent revisions of the Declaration of Helsinki except for the MA study not being pre-registered as a clinical trial. Knee extensor resistance training The resistance training intervention consisted of supervised unilateral leg extensions (two days/week for eight weeks), and the intervention was preceded and followed by strength and VL muscle assessments (described in later paragraphs).All MA participants trained their right legs whereby each training session consisted of five sets of 12 repetitions.The beginning training load was established at ~40% of the participants' three-repetition maximum (3RM).After each set, participants verbally articulated their perceived repetitions in reserve (RIR) [52], and training load was adjusted accordingly.RIR values of 0-2 after a set resulted in no training load change in each session.RIR values of 3-5 for consecutive sets resulted in the training load being increased by 5-10%.For RIR values ≥6 after one set, the training load was increased by 10-20%.If the weight could not be performed with full range of motion, or the participant could not complete 12 repetitions for a given set, the training load was decreased accordingly. Strength testing The first and last workout of the eight-week training paradigm consisted of maximal leg extensorflexion torque assessments using isokinetic dynamometry (Biodex System 4; Biodex Medical Systems, Inc., Shirley, NY, USA) and 3RM leg extensor strength testing.Prior to dynamometer testing, the participant's lateral epicondyle was aligned with the axis of the dynamometer's lever arm, and the hip was positioned at 90°.The participant's shoulders, hips, and leg were strapped and secured for isolation during testing.Following three warm-up trials at a submaximal effort, participants completed five maximal voluntary isokinetic knee extension and flexion actions at 60 degrees/second.Participants were provided verbal encouragement during each contraction.The isokinetic contraction resulting in the greatest peak torque value was used for analyses.Approximately five minutes following isokinetic dynamometry testing, participants performed 3RM strength testing using a free-weight apparatus.Prior to testing, participants were given a warm-up load and instructed to complete 10 repetitions.After participants recorded their RIR for the warmup set, the weight was adjusted accordingly for another warm-up set of five repetitions.RIR was recorded again to determine the participants starting load for a 3RM attempt.The load was incrementally increased 5-10% per 3RM attempt until 3RM testing concluded, indicated by failure of full range of motion on any of the repetitions, or if RIR recorded was 0. Participants were allowed a full three minutes of recovery between attempts.The isokinetic dynamometry and 3RM testing described was similar for both the first and final workout. Urine specific gravity testing for hydration Participants performed a testing battery prior to the start of training (PRE) and 3-5 days following the last resistance training workout (POST).Participants arrived AGING for testing at a minimum of 4 hours fasted and well hydrated.Upon arrival participants submitted a urine sample (~5 mL) for urine specific gravity assessment (USG).Measurements were performed using a handheld refractometer (ATAGO; Bellevue, WA, USA).USG levels in all participants were ≤ 1.020, indicating sufficient hydration [53]. Body composition testing Body composition was assessed using multi-frequency bioelectrical impedance analysis (InBody 520, Biospace, Inc., Seoul, Korea).From the scan, body fat percentage was recorded.Previously determined test-retest reliability yielded an intraclass correlation coefficient (ICC3,1) of 0.99, standard error of the measurement (SEM) of 0.87%, and minimal difference (MD) of 1.71% for body fat percentage. Ultrasonography assessment for muscle morphology A detailed description of VL assessments using ultrasonography has been published previously by our laboratory [54,55].Briefly, real-time B-mode ultrasonography (NextGen LOGIQe R8, GE Healthcare; Chicago, IL, USA) using a multifrequency lineararray transducer (L4-12T, 4-12 MHz, GE Healthcare) was used to capture VL muscle cross-sectional area (mCSA).Prior to scans, the mid-thigh location was determined by measuring the total distance from the mid-inguinal crease in a straight line to the proximal patella, with the knee and hip flexed at 90°, a mark was made using a permanent marker at 50% of the total length.From that location, a permanent marker was used transversely to mark the mid-belly of the VL.This marking is where all pre-intervention ultrasound images were taken as well as the muscle biopsy (described below).All post-intervention images were taken at the pre-intervention biopsy scar to ensure location consistency between scans.During mCSA scans, a flexible, semirigid pad was placed around the thigh and secured with an adjustable strap to allow the probe to move in the transverse plane.Using the panoramic function of the device (LogicView, GE Healthcare), images were captured starting at the lateral aspect of the VL and moving medially until rectus femoris was visualized, crossing the marked location.All ultrasound settings were held constant across participants and laboratory visits (frequency: 10 MHz, gain: 50 dB, dynamic range: 75), and scan depth was noted and held constant across time points per participant.Images were downloaded and analyzed offline using ImageJ software (National Institutes of Health, Bethesda, MD, USA).All ultrasound images were captured and analyzed by the same investigators at each timepoint.Previously determined test-retest reliability on 10 participants measured twice within 24 hours (where BAR captured images and JSG analyzed images) yielded an intraclass correlation of 0.99 and standard error of measurement of 0.60 cm 2 . Collection of muscle tissue Muscle biopsies from all participants were obtained from the mid-belly of the right VL, and sampling time of day was standardized for MA participants at pre and post resistance training intervention.Lidocaine (1%, 1.0 mL) was injected subcutaneously above the skeletal muscle fascia at the previously marked location.After five minutes of allowing the anesthetic to take effect, a small pilot incision was made using a sterile Surgical Blade No. 11 (AD Surgical; Sunnyvale, CA, USA), and the 5-gauge biopsy needle was inserted into the pilot incision ~1 cm below the fascia.Approximately 30-50 mg of skeletal muscle was removed using a double chop method and applied suction.Following biopsies, tissue was rapidly teased of blood and connective tissue, placed in pre-labeled foils, flash frozen in liquid nitrogen, and subsequently stored at −80° C until processing described below. MyoF and non-MyoF protein fractionation The MyoF and non-MyoF protein fractions were isolated per methods published by our laboratory and others [23,56].On the day of homogenization, muscle tissue was powdered on a liquid nitrogencooled ceramic mortar and pestle.Approximately 30 mg of tissue was homogenized using tight-fitting pestles in 500 µL of 25 mM Tris, pH 7.2, 0.5% Triton X-100, with added protease inhibitors (Promega, cat# G6521; Madison, WI, USA).Samples were centrifuged at 1,500 g for 10 minutes at 4° C, supernatants (non-MyoF fraction) were transferred to new 1.7 mL tubes, and tubes were stored at -80ºC until shipment on dry ice to Seer, Inc. Remaining MyoF pellets were kept on ice and thoroughly aspirated with micro-pipet tips to remove residual supernatant.Thereafter, 300 μL of solubilization buffer was added which contained 20 mM Tris-HCl, pH 7.2, 100 mM KCl, 20% glycerol, 1 mM DTT, 50 mM spermidine with added protease inhibitors (Promega, cat# G6521).Samples were then homogenized using tight-fitting pestles and stored at -80ºC until shipment on dry ice to Seer, Inc. Proteograph assay Proteomics analysis was performed at Seer, Inc. (Redwood City, CA, USA).For each sample, 250 μL of received sample was subjected to the Seer Proteograph Assay protocol.After loading samples onto the SP100 Automation Instrument, protein corona formation and processing was initiated to generate desalted purified AGING peptides for protein identification using Reversed Phase (RP) LC-MS.To form the protein corona, Seer's proprietary NPs were mixed with the samples and incubated at 37° C for 1 hour.Unbound proteins were removed prior to downstream wash, reduction, alkylation, and protein digestion steps which were performed according to Seer's Proteograph Assay protocol [25]. LC-MS configuration Peptides obtained from each of the five NP mixtures were separately reconstituted according in a solution of 0.1% formic acid and 3% acetonitrile [57] spiked with 5 fmol μL PepCalMix from SCIEX (Framingham, MA, USA).Reconstitution volumes varied by NP types to allow for constant peptide quantity for MS injection between samples regardless of starting volume (240 ng: NP1, 400 ng: NP2, 360 ng: NP3, 120 ng: NP4, and 320 ng: NP5). 4 µL of each sample were analyzed with a Ultimate3000 RLSCnano LC system coupled with a Orbitrap Fusion Lumos mass spectrometer (Thermo Fisher Scientific; Waltham, MA, USA).Peptides were loaded on an Acclaim PepMap 100 C18 (0.3 mm ID × 5 mm) trap column and then separated on a 50 cm μPAC analytical column (PharmaFluidics, Zwijnaarde, Belgium) at a flow rate of 1 μL/min using a gradient of 5-25% solvent B (100% ACN) mixed into solvent A (100% water) over 26 minutes.The mass spectrometer was operated in Data Independent Acquisition (DIA) mode using 10 m/z isolation windows from 380-1200 m/z and 3-second cycle time.MS1 scans were acquired at 60k resolution and MS2 at 30k resolution. Data processing DIA LC-MS data were processed using Proteograph Analysis Suite (PAS) v2.1 (Seer, Inc) using the DIA-NN search engine (version 1.8.1) in library-free mode searching MS/MS spectra against an in silico predicted library based on Uniprot's Homo Sapiens reference database (UP000005640_9606, download December 9, 2022).Library-free search parameters included trypsin digestion allowing for one missed cleavage, N-terminal methionine excision, fixed modification of cysteine carbamidomethylation, peptide length of 7-30 amino acids, precursor range of 300-1800 m/z, and fragment ion range of 200-1800 m/z.Heuristic protein inference was enabled, MS1 and MS2 mass accuracy was set to 10 ppm.Precursor FDR was set to 0.01, and PG q-value was set to 0.01.Quantification was performed on summed abundances of all unique peptides considering only precursors passing the q-value cutoff.PAS summarizes all NP values for a single protein into a single quantitative value.Specifically, a single protein may have been measured up to five times, once for each nanoparticle.To derive the single measurement value, PAS uses a maximum representation approach, whereby the single quantification value for a particular peptide or protein group represents the quantitation value of the NP which most frequently has measured any given proteins across all samples. The relative abundances of protein targets were obtained by normalizing raw spectra values for each identified protein to total spectra within-subject.After normalization, undetected protein abundance values were set at zero.Protein values are presented as spectra-normalized values in all figures and results.MyoF and non-MyoF proteome data can be found in Supplementary Tables 1, 2. Statistics and bioinformatics Data processing and statistical analysis were performed A total of 4 , 228 proteins overlapped in both fractions yielding 2,217 unique non-MyoF proteins, 193 unique MyoF proteins, and 6,638 unique proteins identified.Using the PANTHER Classification System classifications, the top five protein classes of MyoF proteins, non-MyoF proteins, and proteins in both fractions are presented in Figure 1B.The top 15 enriched MyoF and non-MyoF proteins in MA (preintervention) and Y are presented in Figure 1D, 1E.None of the 15 MyoF or non-MyoF proteins met the p<0.01 significance criteria between age cohorts. Figure 3 . Figure 3. MyoF and non-MyoF alternative protein isoform differences detected with proteomics.Legend: Data presented for Y and MA (pre-and post-intervention) include the identified titin isoforms in the MyoF fraction (A), alternative MyoF protein isoforms affected by aging (B), and alternative non-MyoF protein isoforms affected by aging (C), and small nuclear ribonucleoproteins that make up spliceosomes between cohorts (D).Data are presented as mean ± standard deviations for individual protein spectra values (normalized to total run spectra values) and y-axes were scaled as log10 for improved visualization.Symbols: #, indicates lower in MA versus Y at one or both time points (p<0.01);, indicates greater in MA versus Y at one or both time points (p<0.01);Φ, indicates greater in MA versus Y at one or both time points for panel d only (p<0.05).Notes: (ND), indicates that the isoform number was not provided from the Uniprot's Homo Sapiens reference database (UP000005640_9606). Figure 4 . Figure 4. MyoF and non-MyoF proteins altered with resistance training in MA participants.Legend: Data presented for MA prior to and following eight weeks of knee extensor training include proteins in the MyoF fraction (11 up-regulated, 2 down-regulated; (A), the top 15 up-regulated proteins in the non-MyoF fraction (B), and all 8 down-regulated proteins in the non-MyoF fraction (C).Data are presented as mean ± standard deviations for individual protein spectra values (normalized to total run spectra values), and y-axes were scaled as log10 for improved visualization.Notes: No biological processes were predicted to be affected with training based on these alterations. Figure 5 . Figure 5. MyoF and non-MyoF proteasome and calpain proteins.Legend: Data presented for Y and MA prior to and following eight weeks of knee extensor training include proteasome subunits and calpains 1/2 in the MyoF fraction (A-C) and non-MyoF fraction (D-F).Data are presented as mean ± standard deviations for individual protein spectra values (normalized to total run spectra values).Symbols: #, indicates lower in MA versus Y at one or both time points (p<0.01);, indicates greater in MA versus Y at one or both time points (p<0.01).Notes: Certain p-values not meeting the significance criteria for these data were presented due to the visual differences observed between cohorts. Table 2 . Top 15 of 111 MyoF proteins greater in Y versus MA participants. Legend: Data are presented as mean ± standard deviations for individual protein spectra values (normalized to total run spectra values) in younger (Y) and pre-intervention middle-aged (MA) participants.Protein targets are sorted from most to least abundant in the Y cohort. Table 3 . Top 30 of 543 non-MyoF proteins different between Y versus MA participants. Legend: Data are presented as mean ± standard deviations for individual protein spectra values (normalized to total run spectra values) in younger (Y) and pre-intervention middle-aged (MA) participants.Protein targets more abundant in Y versus MA are sorted from most to least abundant in the Y cohort.Protein targets more abundant in MA versus Y are sorted from most to least abundant in the MA cohort. Table 4 . Biological processes affected based on non-MyoF proteins different between Y versus MA participants. Legend: Biological pathways presented to be up-and down-regulated between age groups based on differential non-MyoF protein expression differences between age cohorts (i.e., Table3data). ). AGING protein.In the non-MyoF fraction, the summed spectra of the 42 detected 26S proteasome subunits was not significantly different between Y versus MA participants at either time point.However, both calpains were higher in the non-MyoF fraction of MA at both time points versus Y participants (CAPN1 was significant in MA pre-intervention versus Y, p<0.01), and training did not significantly affect either protein. [58,59]icrosoft Excel for Microsoft 365 (Redmond, WA, USA) and GraphPad Prism version 9.2.0 (San Diego, CA, USA).Independent samples t-tests were used for Y versus MA (pre-intervention) to determine age effects, and dependent samples t-tests were used to determine training effects in MA.All data in tables and figures are presented as mean ± standard deviation (SD) values.Training phenotypes were considered significantly different at p<0.05, although approaching values (i.e., p<0.100) were discussed as "numerical" changes due to limited n-sizes.Conversely, significant aging and training effects for protein targets were established as p<0.01 for enhanced stringency given the high number of identified proteins, although again approaching values (i.e., p<0.05) were discussed in certain circumstances due to limited n-sizes.Bioinformatics was performed using PANTHER v17.0[58,59].First, protein lists from each fraction were characterized using the functional classification tool.Next, overrepresentation tests of PANTHER GO-Slim biological processes were performed between Y and MA participants and in MA participants from pre-to-post training.Parameters for statistical overrepresentation tests included the following: i) entered proteins had to meet the aforementioned p<0.01 significance threshold, ii) protein lists were entered separately based on being up-or downregulated to generate a list of biological processes that were predicted to be directionally affected, and iii) Fisher tests with Bonferroni adjusted p<0.05 values were used as significance thresholds.variousaspects of proteomics analysis.C. B. Mobley, T. N. Ziegenfuss, A. D. Fruge, and A. N. Kavazis were intellectually involved in various stages of the project.All co-authors critically contributed to the writing of this manuscript, and all co-authors approved the final submitted version.	9,128	2024-04-19T00:00:00.000	[ "Medicine", "Biology" ]
On Certain Sufficient Condition Involving Gaussian Hypergeometric Functions Recommended by Teodor Bulboac˘ a The authors define a new subclass of A of functions involving complex order in the open unit disk U. For this new class, we obtain certain inclusion properties involving the Gaussian hypergeometric functions. Introduction and Motivation Let A be the class of functions f normalized by a n z n , 1.1 which are analytic in the open unit disk As usual, we denote by S the subclass of A consisting of functions which are also univalent in U. A function f ∈ A is said to be starlike of order α in U 0 ≤ α < 1 , if and only if This function class is denoted by S * α .We also write S * 0 : S * , where S * denotes the class of functions f ∈ A that are starlike in U with respect to the origin. International Journal of Mathematics and Mathematical Sciences A function f ∈ A is said to be convex of order α in U 0 ≤ α < 1 if and only if The class of convex functions is denoted by the class K α .Further, K K 0 , the well-known standard class of convex functions.It is an established fact that f ∈ K α ⇐⇒ zf ∈ S * α . 1.5 A function f ∈ A is said to be in the class UCV of uniformly convex functions in U if f is a normalized convex function in U and has the property that, for every circular arc δ contained in the unit disk U, with center ζ also in U, the image curve f δ is a convex arc.The function class UCV was introduced by Goodman 1 . For functions f ∈ A given by 1.1 and g ∈ A given by g z z ∞ n 2 b n z n , we define the Hadamard product or Convolution of f and g by a n b n z n , z ∈ U. 1.6 Furthermore, we denote by k − UCV and k − ST two interesting subclasses of S consisting, respectively, of functions which are k-uniformly convex and k-starlike in U. Thus, we have 1.7 The class k − UCV was introduced by Kanas and Wiśniowska 2 , where its geometric definition and connections with the conic domains were considered.The class k − ST was investigated in 3 .In fact, it is related to the class k − UCV by means of the well-known Alexander equivalence between the usual classes of convex and starlike functions; see also the work of Kanas and Srivastava 4 for further developments involving each of the classes k − UCV and k − ST .In particular, when k 1, we obtain where UCV and SP are the familiar classes of uniformly convex functions and parabolic starlike functions in U, respectively see for details, 1, 5 .In fact, by making use of a certain fractional calculus operator, Srivastava and Mishra 6 presented a systematic and unified study of the classes UCV and SP . A function f ∈ A is said to be in the class P τ γ A, B ⊂ A if it satisfies the inequality 1.9 The class P τ 0 A, B was introduced earlier by Dixit and Pal 7 .Two of the many interesting subclasses of the class P τ γ A, B are worthy of mention here.First of all, by setting Secondly, if we put we obtain the class of functions f ∈ A satisfying the inequality which was studied by among others Padmanabhan 9 and Caplinger and Causey 10 .Finally, many of the authors have also studied the class P 1 γ A, B .For details of these works one can refer to the works of Ding Gong 11 , R. Singh and S. Singh 12 , Owa and Wu 13 , and also the references cited by them.Although, many mapping properties of the class P 1 γ A, B have been studied by these authors, they did not study any mapping properties involving the hypergeometric functions. The Gaussian hypergeometric function F a, b; c, z , z ∈ U is given by is the solution of the homogeneous hypergeometric differential equation and has rich applications in various fields such as conformal mappings, quasiconformal theory, and continued fractions. International Journal of Mathematics and Mathematical Sciences Here, a, b, c are complex numbers such that c / 0, −1, −2, −3, . .., a 0 1 for a / 0, and for each positive integer n, a n a a 1 a 2 • • • a n − 1 is the Pochhammer symbol.In the case of c −k, k 0, 1, 2, . .., F a, b; c; z is defined if a −j or b −j, where j ≤ k.In this situation, F a, b; c; z becomes a polynomial of degree j in z.In particular, the close-to-convexity in turn the univalency , convexity, starlikeness, for details on these technical terms we refer to 5 , and various other properties of these hypergeometric functions were examined based on the conditions on a, b, and c in 8 .For more interesting properties of hypergeometric functions, one can also refer to 20, 21 . Let f z and g z be analytic in U and g z univalent.Then we say that f z is subordinate to g z written as f z ≺ g z if f 0 g 0 and f U ⊂ g U .For f ∈ A, we recall that the operator I a,b,c f of Hohlov 22 which maps A into itself defined by where * denotes usual Hadamard product of power series.Therefore, for a function f defined by 1.1 , we have 1.17 Using the integral representation, we can write Indeed, I 1,1,2 f and I 1,2,3 f are known as Alexander and Libera operators, respectively. Let 0 ≤ k < ∞, and let f ∈ A be of the form 1.1 .If f ∈ k − UCV , then the following coefficient inequalities hold true cf. 2 : where which is the extremal function for the class P p k related to the class k − UCV by the range of the expression where P 1 P 1 k is given, as above, by 1.22 . Similarly, if f of the form 1.1 belong to the class k − ST , then cf. 3 where P 1 P 1 k is given, as above by 1.22 . Properties of P τ γ A, B Theorem 2.1.Let f ∈ S and be of the form The estimate is sharp. Proof.Since f ∈ P τ γ A, B , we have where w z is analytic in U and satisfies the condition w 0 0 and \|w z \| < 1 for z ∈ U. Hence, we have International Journal of Mathematics and Mathematical Sciences Using By equating the coefficients, we observe that the coefficient a n in the right-hand side depends only on a 2 , a 3 , . . ., a n−1 on the left-hand side of the above expression.This gives 2.7 By letting r → 1, we conclude that By making use of the fact that −1 ≤ B < 1, we get 2.11 The result is sharp for the function 2.13 The result is sharp for the function Proof.In view of 2.13 , which is clearly less than or equal to zero for all \|z\| r, 0 < r < 1. Letting r → 1, we get Thus, f ∈ P τ γ A, B . International Journal of Mathematics and Mathematical Sciences Results Involving Gaussian Hypergeometric Function Proof.zF a, b; c; z has the series representation given by In view of Theorem 2.2, it suffices to show that S a, b, c, γ : 3.4 From the fact that \| a n \| ≤ \|a\| n , we observe that c is real and positive, under the hypothesis 3.6 Using the fact that 3.8 From 1.14 , 3.9 By using the Gauss summation theorem Proof.Let f be of the form 1.1 belong to the class P τ γ A, B .By virtue of Theorem 2.2, it suffices to show that International Journal of Mathematics and Mathematical Sciences Taking into account inequality 2.1 and the relation which is bounded previously by A − B \|τ\|, in view of inequality 3.12 . Repeating the previous reasoning for b a, we can improve the assertion of Theorem 3.2 as follows. B , and if the inequality In the special case when b 1, Theorem 3.2 immediately yields the following new result. where k ∈ N. 3.17 Proof.Let f ∈ S. Applying the well-known estimate for the coefficients of the functions f ∈ S, due to de Branges 23 , we need to show that The left-hand side of 3.18 can be written as The second expression of 3.19 , by virtue of the triangle inequality for the pochhammer symbol Now, making use of the relation 3.7 , we get where we are writing n 2 2 n n − 1 5n 4. By repeating the use of 3.7 and the Gauss summation formula, we have International Journal of Mathematics and Mathematical Sciences As a next step, we consider the first expression of equation.By making use of the triangle inequality for the pochhammer symbol as stated in evaluating S 1 , we get 3.23 Now making use of relation 3.7 , we obtain 3.24 where we write n 2 2 n 1 2 2 n 1 1.By repeating the use of 3.7 and the Gauss summation formula, we have 3.25 The proof of Theorem 3.5 now follows by an application of the inequalities of the terms dealing with S 1 , S 2 and inequality 3.17 . Repeating the previous reasoning for b a, we can improve the assertion of Theorem 3.5 as follows. Theorem 3.6.Let a, b ∈ C \ {0}.Also, let c be a real number such that c > max{0, 2R a 3 }.If f ∈ S, and if the inequality Proof.By means of 1.17 and 2.13 , the following inequality must be satisfied: Results regarding F a, b; c; z when R c − a − b is positive, zero, or negative are abundant in the literature.In particular when R c − a − b > 0, the function is bounded.This and the zero balanced case R c − a − b 0 are discussed in detail by many authors see 14, 15 .The hypergeometric function F a, b; c; z has been studied extensively by various authors and it plays an important role in Geometric Function Theory.It is useful in unifying various functions by giving appropriate values to the parameters a, b, and c.We refer to 8, 16-19 and references therein for some important results. 1 . 19 When f z equals the convex function z/ 1 − z , then the operator I a,b,c f in this case becomes zF a, b; c; z .For a 1, b 1 δ, c 2 δ with R δ > −1 then the convolution operator I a,b,c f turns into Bernardi operator Theorem 3 . 4 . Let a ∈ C \ {0}.Also, let c be a real number such that c > \|a\| 1.If f ∈ P τ γ A, B , and if the inequality Theorem 3 . 5 . Let a, b ∈ C \ {0}.Also, let c be a real number such that c > \|a\| \|b\| 3.If f ∈ S, and if the inequality is satisfied, then I a,b,c f ∈ P τ γ A, B .	2,734.2	2009-12-16T00:00:00.000	[ "Mathematics" ]
Zinc finger protein 468 up-regulation of TFAM contributes to the malignant growth and cisplatin resistance of breast cancer cells Background Because of the progress on the diagnosis and treatment for patients with breast cancer (BC), the overall survival of the patients has been improved. However, a number of BC patients cannot benefit from the existing therapeutic strategies as the essential molecular events triggering the development of BC are not well understood. Previous studies have shown that abnormal expression of zinc finger proteins is involved in the development of various malignancies, whereas it remains largely unclear on their significance during the progression of BC. In this study, we aimed to explore the clinical relevance, cellular function and underlying mechanisms of zinc finger protein 468 (ZNF468) in BC. Methods The clinical relevance of ZNF468 and TFAM was analyzed based on TCGA database. Overexpression or knockdown of ZNF468 and TFAM were performed by transfecting the cells with overexpression plasmids and siRNAs, respectively. Overexpression and knockdown efficacy was checked by immunoblotting. CCK-8, colony formation, transwell and apoptosis experiments were conducted to check the cellular function of ZNF468 and TFAM. The content of mtDNA was measured by the indicated assay kit. The effects of cisplatin on BC cells were detected by CCK-8 and colony formation assays. The regulation of ZNF468 on TFAM was analyzed by RT-qPCR, immunoblotting, dual luciferase activity and ChIP-qPCR assays. Results ZNF468 was overexpressed in BC patients and inversely correlated with their prognosis. Based on overexpression and knockdown assays, we found that ectopic expression of ZNF468 was essential for the proliferation, growth and migration of BC cells. The expression of ZNF468 also negatively regulated the sensitivity of BC cells to the treatment of cisplatin. Mechanistically, ZNF468 potentiated the transcription activity of TFAM gene via direct binding on its promoter. Lastly, we demonstrated that ZNF468 up-regulation of TFAM was important for the growth, migration and cisplatin resistance in BC cells. Conclusion Our study indicates that ZNF468 promotes BC cell growth and migration via transcriptional activation of TFAM. ZNF468/TFAM axis can serve as the diagnostic and therapeutic target, as well as the predictor of cisplatin effectiveness in BC patients. Supplementary Information The online version contains supplementary material available at 10.1186/s13008-024-00113-1. Background Breast cancer (BC) is one of the most common malignant tumors threatening women's health in the world, with the incidence ranking first and the mortality ranking second in female malignant tumors [1].Based on the expression of the estrogen receptor (ER), progesterone receptor (PR), or epidermal growth factor 2 (HER2), BC patients are divided into three types, including ER+/HER2−, HER+ and ER−/PR−/HER2− type.ER+/HER2− type accounts for 70% of all the cases, while the remaining 15% cases are HER2+and ER−/PR−/HER2− (triple-negative breast cancer, TNBC), respectively [2].Recently, increasing evidences have identified the critical molecular events driving the development of BC.For example, poly (ADP-ribose) polymerase (PARP), epidermal growth factor receptor (EGFR) and vascular endothelial growth factor (VEGF) are important drivers during the progression of TNBC and BRCA1-or BRCA2-mutated BC [3].Benefiting from these studies, the inhibitors of PARP1 were under clinical evaluation for the treatment of BC patients [4][5][6].Despite these progresses, a large number of TNBC patients are left without effective medication.Thus, it is urgent for us to identify novel driver genes which can be applied as the diagnostic or therapeutic target of BC patients. The family of zinc finger proteins (ZNFs) comprises of thousands of proteins, exhibiting either transcription or post-transcription activity based on their interaction with DNA, RNA or proteins [7].Previous studies have shown that ZNFs play important roles during the progression of cancers.For example, ZNF306 is amplified and overexpressed in colorectal cancer (CRC) patients.Overexpression of ZNF306 drives the tumorigenesis and 5-fluorouracil resistance of CRC through transcriptional activation of integrin β4 and VEGF [8].Another study reported that ZNF306 functioned as an oncogene in multiple myeloma by modulating the expression of Cyclin D2 [9].Similar with ZNF306, ZNF468 protein has C2H2type zinc finger motif and is predicted to enable DNAbinding transcription factor activity by interacting with RNA polymerase II.However, the significance of ZNF468 in the development of cancers, including BC, remains to be understood. In this study, we explored the clinical significance, cellular function and molecular mechanisms of ZNF468 in breast cancer cell growth and cisplatin sensitivity based on TCGA database, overexpression and knockdown of ZNF468, as well as CCK-8, colony formation, Transwell, luciferase activity and ChIP-qPCR assays.We demonstrated that ZNF468 up-regulation of TFAM not only contributed to the growth and migration of breast cancer cells but also reduced the sensitivity of the cells to cisplatin treatment. ZNF468 is overexpressed in BC patients Firstly, we investigated the clinical relevance of ZNF468 in pan-cancer.Based on TCGA database, ZNF468 was overexpressed in different cancer types, including bladder cancer (BLCA), Cervical squamous cell carcinoma and endocervical adenocarcinoma (CESC), glioma (GBM), lung adenocarcinoma (LUAD), breast cancer (BC) and other cancer types (Fig. 1A, B).We focused on BC and further found that LumA and LumB patients had higher expression of ZNF468 than HER+ and TNBC patients (Additional file 1: Fig. S1A).In addition, BC patients with high expression of ZNF468 had significantly poorer survival than those with low expression of ZNF468 (Fig. 1C).High expression of ZNF468 was significantly correlated with the poor survival of LumA patients, but not other subtypes (Additional file 1: Fig. S1B-E).To verify the results in Chinese patients, we collected the normal and cancer samples from BC patients and subjected them to immunoblotting analysis of ZNF468.The results showed that ZNF468 was highly expressed in the cancer tissues than normal tissues (Fig. 1D).Lastly, the expression of ZNF468 was also increased in BC cancer cell lines as compared with breast immortalized cells MCF-10A (Fig. 1E).Collectively, ZNF468 likely functions as an oncogene in BC. ZNF468 overexpression is essential for BC cell growth and migration Next, we intended to determine the cellular function of ZNF468 based on overexpression and knockdown experiments.Because MDA-MB-231 cells had relatively higher expression of ZNF468 than MCF-7 cells, we performed knockdown assay in MDA-MB-231 cells and overexpression assay in MCF-7 cells.Western blot demonstrated that ZNF468 was successfully overexpressed in MCF-7 cells by transfecting with overexpression vectors of ZNF468 (Fig. 2A).CCK-8 and colony formation results showed that ectopic expression of ZNF468 promoted the proliferation and growth ability of MCF-7 cells (Fig. 2B, C).Furthermore, we transfected MDA-MB-231 cells with siRNAs to knock down ZNF468.Immunoblotting results confirmed the knockdown efficacy (Fig. 2D).As expected, ZNF468 downregulation suppressed the viability of MDA-MB-231 cells (Fig. 2E, F).The apoptosis of MDA-MB-231 cells was obviously induced by knockdown of ZNF468 (Fig. 2G).In addition, ZNF468 knockdown also inhibited the migration ability of MDA-MB-231 cells (Fig. 2H).These results suggest that ZNF468 acts as an oncogenic protein in BC by promoting the proliferation and migration capacity of BC cells. ZNF468 expression dictates the sensitivity of BC cells to cisplatin treatment Cisplatin is a chemotherapeutic drug applied for the treatment of patients with advanced HER+ and TNBC breast cancer in clinic [10,11].To investigate whether ZNF468 dysregulation modulates the effectiveness of cisplatin, the BC cells transfected with ZNF468 siRNAs or overexpression vectors were treated with different concentrations of cisplatin and subjected to CCK-8 analysis of cell viability.As shown in Fig. 3A, when cisplatin inhibits the viability of MCF-7 cells transfected with empty control vectors at a dosage dependent manner, it exhibited lower inhibitory effects on the cells transfected with ZNF468 overexpression vectors (Fig. 3A).By contrast, ZNF468 knockdown enhanced the cytotoxicity of cisplatin on MDA-MB-231 cells, especially at higher concentration (Fig. 3B).These results indicated that ZNF468 overexpression protected MCF-7 cells from the toxicity of cisplatin, while ZNF468 knockdown promoted the inhibitory effect of cisplatin to MDA-MB-231 cells.To confirm the results, BC cells treated with cisplatin were subjected to colony formation experiments.As 10 μM of cisplatin had nearly 50% and 30% inhibitory effect on the proliferation of Ctrl MCF-7 and siCtrl MDA-MB-231 cells for 48 h, we applied this dosage in colony formation assay.We found that cisplatin exhibited higher cytotoxicity on the colonies of MCF-7 cells transfected with empty control vectors as compared with those with ZNF468 overexpression (Fig. 3C, D).In addition, knockdown of ZNF468 significantly potentiated the inhibitory effect of cisplatin on the colony formation ability of MDA-MB-231 cells (Fig. 3E, F).These results indicate that ZNF468 expression confers the sensitivity of BC cells to cisplatin treatment. ZNF468 potentiates the transcription activity of TFAM gene Because ZNF468 is a transcription factor, the downstream effectors participating in its function on BC growth and cisplatin should be illustrated.We analyzed the positively correlated genes of ZNF468 in BC patients based on TCGA database and found that RBM12, ADNP, DDX46, TFAM and RBM13 were positively correlated with ZNF468 (Fig. 4A).Previous studies demonstrated that down-regulation of TFAM was associated with the enhanced sensitivity of BC cells to cisplatin treatment [12].Furthermore, we found that knockdown of TFAM not only suppressed the growth and proliferation of BC cells, but also promoted to toxicity of cisplatin in vitro and in vivo [13].Thus, we intended to explore whether ZNF468 supported BC growth and cisplatin resistance through regulating TFAM.According to TCGA data, TFAM was positively correlated with ZNF468 and it was highly expressed in BC patients (Fig. 4B, C).RT-qPCR and immunoblotting assays showed that ZNF468 overexpression promoted, while its down-regulation suppressed the expression of TFAM at both mRNA and protein levels (Fig. 4D, E).We also checked the mtDNA content and found that ZNF468 positively regulated the levels of mtDNA in BC cells (Fig. 4F).Luciferase reporter results found that ZNF468 positively regulated the transcription activity of TFAM gene (Fig. 4G).Importantly, ZNF468 directly interacted with the promoter sequence of TFAM gene (Fig. 4G).These results indicate that ZNF468 transcriptionally activates TFAM gene in BC cells. ZNF468 contributes to the growth, migration and cisplatin resistance of BC cells through up-regulation of TFAM Lastly, we performed experiments to investigate the involvement of TFAM in ZNF468 triggering BC growth and cisplatin treatment.TFAM was knocked down in ZNF468 overexpressing MCF-7 cells and was ectopically expressed in ZNF468 silencing MDA-MB-231 cells.The cells were treated with or without different concentrations of cisplatin and then subjected to CCK-8 detection of cell viability.We found that TFAM knockdown largely reversed the enhanced cell proliferation of MCF-7 cells by ZNF468 overexpression (Fig. 5A, B).Interestingly, the reduced sensitivity of MCF-7 cells to cisplatin caused by ZNF468 was restored by knockdown of TFAM (Fig. 5C).By contrast, TFAM overexpression had promoting function not only on the proliferation but also on cisplatin resistance in ZNF468 silenced MDA-MB-231 cells (Fig. 5D-F).We also assessed the significance of ZNF468/TFAM axis on BC cell migration and found that knockdown of TFAM significantly reduced the migration ability of MCF-7 cells which was potentiated by overexpression of ZNF468 (Fig. 5G, H).Opposite results were shown after overexpressing TFAM in ZNF468 silenced MDA-MB-231 cells (Fig. 5I, J).These results suggest that ZNF468 promotes BC cell growth, migration and cisplatin resistance through up-regulation of TFAM. Overall, ZNF468 promotes BC cell proliferation and cisplatin resistance through transcriptional activation of TFAM gene (Fig. 6). Discussion There are approximately 1600 transcription factors (TFs) expressed in eukaryotes, including Kruppel-like family, specificity protein (SP), zinc finger proteins (ZNFs) and etc. [14,15].Normal expression of TFs exhibits essential functions on cellular metabolism, proliferation, differentiation, survival and death [16].Dysregulation of TFs is involved in the development of cancers [17].To our knowledge, the significance of ZNFs in carcinogenesis is beginning understood.For example, ZNF704 acts as an oncogene in BC [18], chondrosarcoma [19], and uveal melanoma [20] by regulating circadian rhythm and AKT/ mTOR signaling.Overexpression of ZNF488 contributes to the malignant progression of nasopharyngeal carcinoma [21] and cervical cancer [22] through activating Wnt and MEK/ERK signaling pathway.In this study, we demonstrated that ZNF468 was overexpressed in BC tissues and cells.High expression of ZNF468 predicted the poor prognosis of BC patients.Up-regulation of ZNF468 contributed to the growth, migration and cisplatin resistance of BC cells. Mitochondrial transcription factor A (TFAM) is a nuclear-encoded transcription factor that plays an important role in mtDNA replication and damage repair [23].TFAM is closely associated with the tumor suppressor P53 signaling.Knockdown of TFAM promotes the binding between p53 and E3 ubiquitin protein ligase MDM2, thereby inhibiting p53 expression [24].TFAM is also involved in p53-mediated apoptosis via direct interacting with p53 [25].The function of TFAM in carcinogenesis is controversial in different cancer types.While down-regulation of TFAM suppressed the proliferation of HepG2 and U-2 OS cells [26], loss of TFAM facilitates the malignancy of head and neck cancer [27].Consistent with the promoting function of TFAM in HepG2 and U-2 OS cells, we previously found that TFAM served as an oncogenic protein in BC.However, the upstream regulator of TFAM in BC remains to be determined.In this study, we showed that ZNF468 positively regulated the transcription activity and the protein expression of TFAM gene in BC cells.Mechanistically, ZNF468 interacted with the promoter sequence of TFAM gene, causing the enhanced transcription activity.However, the precise binding area of TFAM promoter was not explored and should be identified in the follow up study.In BC patients, there was a positive correlation between ZNF468 and TFAM.The content of mtDNA, which is regulated by TFAM, was potentiated by ZNF468 overexpression and was suppressed by its knockdown.Importantly, we demonstrated that TFAM up-regulation was critical for the accelerated proliferation and migration ability of BC cells triggered by ZNF468.These results suggest that ZNF468 ectopic expression led to dysfunction of mitochondria and enhanced cell growth capacity via up-regulating TFAM. Dysregulation of TFAM not only contributes to the progression of malignant tumors, but also regulates the sensitivity of cancer cells to chemotherapeutic drugs, including cisplatin.Down-regulation of TFAM promoted the effectiveness of NSCLC cells to the treatment of cisplatin [28].In ER positive BC cells, cisplatin resistance was acquired by TFAM overexpression [13].However, another study showed that knockdown of TFAM attenuated cisplatin-induced cell death in ovarian cancer cells [29].The results imply that the significance of TFAM in cisplatin effectiveness maybe different in various cancer types.Although we have shown that ZNF468 up-regulation of TFAM is critical for BC cell growth, the involvement of TFAM in cisplatin resistance induced by ZNF468 overexpression needs further validated.As demonstrated by CCK8 and colony formation assays, TFAM knockdown by siRNAs in ZNF468 overexpressed MCF-7 cells significantly restored the sensitivity of cisplatin.Because silence of ZNF468 gene would not be last for 10 days by siRNAs transfection, our study suggested that transient downregulation of ZNF468 could obviously promoted the cytotoxicity of cisplatin, revealing the important function of ZNF468 in BC cell growth and chemotherapy.Nevertherless, stable cell lines with ZNF468 knockdown or overexpression should be constructed in the follow up study.By contrast, TFAM overexpression reduced the cisplatin cytotoxicity in ZNF468 silenced MDA-MB-231 cells.Thus, we confirm that ZNF468 confers cisplatin resistance via up-regulation of TFAM. Conclusion In conclusion, we provided the evidence that ZNF468 contributed to the proliferation and migration of BC cells.Overexpression of ZNF468 alleviated cisplatin effectiveness.Mechanistically, ZNF468 activates the expression of TFAM at transcriptional level.There was a positive correlation between ZNF468 and TFAM in BC patients.Importantly, ZNF468 exerted the tumor-promoting function via up-regulation of TFAM.Our study highlights the potential of ZNF468/TFAM axis as the diagnostic and therapeutic target for BC patients. The Cancer Genome Atlas (TCGA) analysis and BC specimens The clinical relevance of ZNF468 and TFAM was analyzed in BC based on the public database, including UAL-CAN, GEPIA and The Cancer Genome Atlas (TCGA).In brief, the expression of ZNF468 in pan-cancers was analyzed using TCGA data from the UALCAN database (https:// ualcan.path.uab.edu/ analy sis.html).The expression of ZNF468 and TFAM, their correlation, as well as their relationship with BC patients' overall survival, were analyzed from the GEPIA database (http:// gepia.cancer-pku.cn/), which included the data not only from TCGA.A total of 1085 tumor samples and 291 normal tissues were included.For overall survival analysis, the patients were divided into ZNF468 high (n = 212) and low (n = 213) expression group.The results presented in Additional file 1: Fig. S1 were analyzed by using the data from TCGA. We also collected a total of 12 cancer and normal samples from BC patients who received operation at Gui-Hang Guiyang Hospital.Each patient signed the informed consent before surgery.This study was approved by the Clinical Research Ethics Committee of GuiHang Guiyang Hospital.The tissues were subjected to immunoblotting analysis of ZNF468. Overexpression of ZNF468 or TFAM was conducted by transfecting the cells with 5 μg overexpression plasmids containing the coding sequence of indicated gene by using 2 μL VigoFect (T001, Vigorous Bio, Beijing, China).The cells transfected empty plasmids represented the Ctrl group.The cells were subjected for other experiments 48 h after transfection. Detection of mtDNA Total DNA was extracted from the cells with the QIAquick Nucleotide Removal Kit (28,304,QIAGEN). Copy number of mtDNA was measured by using the Human Mitochondrial DNA Copy Number Assay Kit (Detroit R&D, MCN1), according to the manufacturer's instructions and the protocols described previously [31]. Cell Counting Kit-8 (CCK-8) The effect of ZNF468 and TFAM on BC cell proliferation and cisplatin sensitivity was detected by CCK-8 (C0039, Beyotime Biotechnology).Briefly, the cells transfected with overexpression plasmids or siRNAs were seeded in 96-well plates, treated with or without different concentrations of cisplatin.One to three days later, 10 μL CCK-8 solution was incubated with the culture medium at 37 °C for 2 h.Cell viability was detected by measuring the OD450 value. Colony formation assay A total of 500 Ctrl and ZNF468 overexpressing MCF-7 cells, and 1000 siCtrl, siZNF468#1 and siZNF468#2 MDA-MB-231 cells were seeded into 6-well plates.The cells were treated with or without 10 μM cisplatin for 7-10 days.When cell colonies formed, the plates were mildly washed by PBS, fixed by methanol for half an hour, and stained by crystal violet.Lastly, the plates were washed by clean water and the colonies were photographed by a camera. Transwell analysis of cell migration A total of 60,000 MDA-MB-231 cells (siCtrl, siZNF468#1, siZNF468#2 and siZNF468#1+TFAM) and 120,000 MCF-7 cells (Ctrl, ZNF468 and ZNF468+siTFAM) in 200 μL FBS-free culture medium were seeded in the upper surface of the transwell chamber.The lower surface was covered onto a total of 500 μL complete medium in 24-well plates.One day later, migrated cells on the lower surface was subjected to methanol fixation and crystal violet staining.Cells attached on the upper surface were removed by the cotton ball.Migrated cells were photographed under the microscope. Dual luciferase reporter assay The coding sequence of ZNF468 was cloned into pCDNA3.1 plasmids (Addgene).The promoter sequence of TFAM gene was cloned into pGL3.basicplasmids (Addgene).MCF-7 cells were transfected with empty control and ZNF468 overexpressing plasmids.MDA-MB-231 cells were transfected with siRNAs against negative control and ZNF468.Meanwhile, the cells were transfected with pGL3.basicplasmids containing the promoter of TFAM gene and TK plasmids.Relative dual luciferase activity was determined by using the assay kit from Promega (E1960), according to the manufacturers' protocols. Statistical analysis The data were shown as mean ± standard error of mean (SEM) and statistical analysis was conducted on Grpah-Pad Prism software.Student's t test was used to analyze the difference between two groups.One-way ANOVA followed by followed by a Tukey's post hoc test was applied to compare the difference among multiple groups.P value less than 0.05 was considered as statistical significance. Fig. 1 Fig. 1 BC samples and cells have higher expression of ZNF468.A The transcript level of ZNF468 was analyzed in pan-cancer based on the database from TCGA.Most of the cancers showed higher expression ZNF468.n.s, no significance.p < 0.05.B Analysis of ZNF468 transcript was conducted in cancer (n = 1085) and normal tissues (n = 291) derived from BC patients.p < 0.05.C Overall survival was analyzed on BC patients who were divided into ZNF468 high expression (n = 212) and low expression (n = 213) group.p = 0.022.D Cancer (n = 12) and normal tissues (n = 12) of BC patients were collected for immunoblotting analysis of ZNF468.Quantification of ZNF468 protein abundance was adjusted to GAPDH.p < 0.01.E The expression of ZNF468 was determined by immunoblotting assay in breast immortalized cells MCF-10A and in BC cells MCF-7, MDA-MB-468 and MDA-MB-231.For immunoblotting assays, the blots were cut prior to hybridisation with antibodies Fig. 4 Fig. 5 Fig. 4 ZNF468 potentiates the expression of TFAM at transcriptional and protein level.A The positive genes correlated with ZNF468 in BC patients were analyzed based on TCGA database.B Spearman correlation between ZNF468 and TFAM was analyzed in BC patients from TCGA database.p < 0.001.C Analysis of TFAM transcript was performed in cancer (n = 1085) and normal tissues (n = 291) derived from BC patients.D, E The mRNA (D) and protein (E) level of TFAM was detected by RT-qPCR in Ctrl and ZNF468 overexpressing MCF-7 cells, and in siCtrl, siZNF468#1 and siZNF468#2 MDA-MB-231 cells.p < 0.01.F Relative mtDNA copy number was measured by RT-qPCR assay in Ctrl and ZNF468 overexpressing MCF-7 cells, and in siCtrl, siZNF468#1 and siZNF468#2 MDA-MB-231 cells.p < 0.05.G Dual luciferase activity of TFAM promoter was checked in Ctrl and ZNF468 overexpressing MCF-7 cells, and in siCtrl, siZNF468#1 and siZNF468#2 MDA-MB-231 cells.p < 0.01.H ChIP-qPCR assay was performed to determine the binding of ZNF468 on the promoter of TFAM gene in MCF-7 cells.p < 0.01.For immunoblotting assays, the blots were cut prior to hybridisation with antibodies Fig. 6 A Fig. 6 A schematic diagram of ZNF468/TFAM axis in promoting breast cancer cell proliferation, migration and cisplatin resistance	4,884.6	2024-03-01T00:00:00.000	[ "Medicine", "Biology" ]
Understanding the links among magnetic fields, filament, the bipolar bubble, and star formation in RCW57A using NIR polarimetry The influence of magnetic fields (B-fields) in the formation and evolution of bipolar bubbles, due to the expanding ionization fronts (I-fronts) driven by the Hii regions that are formed and embedded in filamentary molecular clouds, has not been well-studied yet. In addition to the anisotropic expansion of I-fronts into a filament, B-fields are expected to introduce an additional anisotropic pressure which might favor expansion and propagation of I-fronts to form a bipolar bubble. We present results based on near-infrared polarimetric observations towards the central $\sim$8'$\times$8' area of the star-forming region RCW57A which hosts an Hii region, a filament, and a bipolar bubble. Polarization measurements of 178 reddened background stars, out of the 919 detected sources in the JHKs-bands, reveal B-fields that thread perpendicular to the filament long axis. The B-fields exhibit an hour-glass morphology that closely follows the structure of the bipolar bubble. The mean B-field strength, estimated using the Chandrasekhar-Fermi method, is 91$\pm$8 {\mu}G. B-field pressure dominates over turbulent and thermal pressures. Thermal pressure might act in the same orientation as those of B-fields to accelerate the expansion of those I-fronts. The observed morphological correspondence among the B-fields, filament, and bipolar bubble demonstrate that the B-fields are important to the cloud contraction that formed the filament, gravitational collapse and star formation in it, and in feedback processes. The latter include the formation and evolution of mid-infrared bubbles by means of B-field supported propagation and expansion of I-fronts. These may shed light on preexisting conditions favoring the formation of the massive stellar cluster in RCW57A. Though bipolar bubbles are the natural outcome of anisotropic expansion of ionizing fronts from Hii regions hosting massive O/B-type star(s) located in filaments (Minier et al. 2013;Deharveng et al. 2015;Zhang et al. 2016), details involved in their formation and evolution processes remain poorly understood. According to 2D (Bodenheimer et al. 1979) and 3D (Fukuda & Hanawa 2000) hydrodynamic simulations, the evolution of an Hii region in a filamentary cloud induces supersonic and subsonic I-fronts along the minor and major axes of the filament, respectively. This results the distribution of low and high density ma-terial along the minor and major axes. Thus, the I-fronts experience more hindered flows along the major axis than along the minor axis. As a consequence of this anisotropic expansion, a bipolar bubble will be formed in a filament (Figure 1 of Deharveng et al. 2015). Though a few studies (e.g., Minier et al. 2013;Deharveng et al. 2015) were devoted to the study of bipolar bubbles, there exists no study focused on exploring the connection between the B-fields anchored in the filaments and their role in the formation and evolution of bipolar bubbles. Fukuda & Hanawa (2000) did include B-fields (parallel to the filament long axis) in their simulations, but they studied only the influence of B-fields on shape, separation, and formation epoch of the cores that formed along the filament. B-fields are believed to guide the contraction of cloud material to form filaments in molecular clouds, and thus play a crucial role in controlling cloud stability and collapse, fragmentation into cores, and internal star formation (Pereyra & Magalhães 2004;Alves et al. 2008;Li et al. 2013;Planck Collaboration et al. 2014;Franco & Alves 2015). A few studies (e.g., Pereyra & Magalhães 2007 towards the IRAS Vela Shell and Wisniewski et al. 2007 towards supershell NGC 2100 in the LMC) suggest the importance of B-fields in the dynamical evolution of the expanding shells driven by O/B-type stars. The interplay between the expansion of an ionized nebula and the influence of B-fields has been the subject of several studies (e.g., Pavel & Clemens 2012;Santos et al. 2012Santos et al. , 2014Planck Collaboration et al. 2015;Kwon et al. 2010Kwon et al. , 2011Kwon et al. , 2015. These studies hint that B-fields permeated through filamentary molecular clouds would also influence the expanding I-fronts or outflowing gas, from an Hii region created in the filament. RCW57A (also known as NGC 3576, G291. 27-0.70, or IRAS 11097-6102) is an Hii region associated with a filament and bipolar bubble and is located at a distance of 2.4 -2.8 kpc (Persi et al. 1994;de Pree et al. 1999). We adopt 2.4 kpc, which is within uncertainties of both the kinematic and spectroscopic determinations (see Persi et al. 1994). Figure 1 depicts the overall morphology of RCW57A. It contains optically bright nebulosity with several dark globules and luminous arcs (Persi et al. 1994). It is one of the massive star -Tricolor image of the RCW57A region made using WISE 4.6 µm (red), 2MASS K s -band (green), and DSS2 R-band (blue) images. Various contours, extracted from Purcell et al. (2009), are overplotted. Yellow contours enclosing white arrows represent the bipolar bubble-like structure observed in mid-infrared (MSX, Spitzer, and WISE) images. This feature appears as an extended widened loop in the R-band, as shown with an extended yellow contour in the Northern part. Thick cyan contour represents the 3.4 cm radio free-free emission from ionized gas, delimiting the extent of the Hii region. The white contours represent the 1.1 mm dust continuum emission (Hill et al. 2005), showing an elongated dusty filament from NE to SW which crosses the H ii region. The extent of the molecular cloud, from 13 CO(1 -0) line emission (Purcell et al. 2009), is depicted with magenta outline (the horizontal line at the bottom of the magenta outline is the limit of the mapped area). Infrared sources (IRS; Frogel & Persson 1974) and water/methanol maser sources (cf. Purcell et al. 2009) are shown with open black squares and green crosses, respectively. Possible directions of the outflowing gas (cf. de Pree et al. 1999) are shown with white arrows. Red square box denotes the area of 7. 7 × 7. 7 observed with SIRPOL. forming regions in the southern sky, hosting an Hii region (cyan contour) embedded in a filament (white contour) from which a widely-extended bipolar bubble (yellow contours) is emerging. A deeply embedded near-IR cluster, consisting of more than 130 young stellar objects (YSOs) is associated with this region (Persi et al. 1994). The observed ratios of the infrared fine-structure ionic lines (Ne ii, Ar iii, and S iv) (Lacy et al. 1982) indicate that at least eight O7.5V stars are necessary to account for the ionization of the region. However, even these stars may not be sufficient to account for the Lyα ionizing photons inferred from radio data (Figuerêdo et al. 2002;Barbosa et al. 2003;Townsley 2009). Based on the newly discovered cluster of stars, using X-ray data, Townsley et al. (2014) suggested that an additional cluster of OB stars might be deeply embedded that were not known before because of heavy obscuration. This cluster is located slightly NW of the center of the near-IR cluster. The 10 µm map (cf., Frogel & Persson 1974), reveals the presence of five infrared sources (IRS; black squares in Figure 1) near the center of the Hii region. These, together with water and methanol maser sources (green crosses in Figure 1) distributed along the filament, are indicative of active ongoing star formation in RCW57A (Frogel & Persson 1974;Caswell 2004;Purcell et al. 2009). Therefore, RCW 57A is an ideal target to investigate the morphological links among filaments, bipolar bubbles, and B-fields so as to understand the star formation history. Here, near-infrared (NIR) polarimetric observations towards the central region of RCW57A have been carried out to map the plane-of-sky B-field geometry in the star forming region. When unpolarized background starlight passes through an interstellar cloud, dust particles partially aligned by a B-field linearly polarize the light by a few percent. Virtually all possible mechanisms responsible for the dust grain alignment with respect to the B-fields (Davis & Greenstein 1951;Andersson et al. 2015;Lazarian et al. 2015) yield directions of the polarized light tracing the average B-field orientation projected on the plane-of-sky. The aim of this paper is to understand whether (a) the B-field structure in the molecular cloud guides feedback processes, such as expansion and propagation of outflowing gas and ionization fronts, which lead to the formation of bipolar bubbles, or (b) feedback processes regulate the resulting B-field structure. The B-field is active in the former scenario, while it is passive in the latter. The structure of this paper is as follows. Section 2 describes the observations and data reduction. Detailed analyses are presented in Section 3, the main purpose of which is to identify and remove the YSOs from the list of fore-ground and background stars by utilizing NIR and mid-infrared (MIR) colors, NIR polarization measurements, and polarization efficiencies. Results are presented in Section 4. This section presents estimates of the B-field strength, and detailed comparisons between B-field pressure to turbulent and thermal pressures. A possible evolutionary scenario, based on the morphological correspondences among the filament, bipolar bubble, and B-fields is discussed in Section 5. Furthermore, based on the scenario that best characterizes our observed B-fields in RCW57A, we discuss the possible preexisting conditions that might favored the formation of star cluster in RCW57A. Conclusions are summarized in Section 6. One set of observations consisted of 10 s exposures at four HWP position angles (0 • , 22.5 • , 45 • and 67.5 • ) at 10 dithered sky pointings. Such sets of 4×10 images were repeated towards the same sky coordinates to increase signal-to-noise. Sky frames were also obtained between target observations. The total integration time was 400 s per HWP angle. The average seeing was 1. 54 (J), 1. 44 (H) and 1. 31 (K s ). Master flats were created by utilizing the evening and twilight flat-field frames on the same night of the observations. We processed the data using the dedicated data reduction pipeline 'pyIRSF' 2 . This pipeline used the raw data from the target, sky, dark, and flat field frames as inputs. The data reduction tasks included dark subtraction, flat-field correction, median sky subtraction, frame registration, and averaging. The final products of this pipeline were the average combined four intensity images (I 0 , I 22.5 , I 45 , and I 67.5 ) corresponding to the four positions of the HWP. Aperture photometry of point sources We performed aperture photometry of pointlike sources on the intensity images for the HWP angles (I 0 , I 22.5 , I 45 , and I 67.5 ) of each of the JHK s -bands using iraf 3 and the idl Astronomy Library (Landsman 1993). Point sources with peak intensities above 5σ (where σ is the rms uncertainty of the particular image pixel values) of the local sky background were detected using DAOFIND. Aperture photometry was performed using the PHOT task of the DAOPHOT package in IRAF. An aperture radius was chosen to be nearly the FWHM of the point-like sources, i.e., 3.4, 3.2, and 2.9 pixels in the J-, H-, and K s bands, respectively. The sky annulus inner radius was set to be 10 pixels with a 5 pixel width. A source that matched in center-to-center positions to within 1 pixel radius (i.e., 2 pixel diameter equivalent to ∼ 1 ) in all four HWP images was judged to be same star. The Stokes I intensity of each point-like source was calculated using Aperture polarimetry Polarimetry of point-like sources (aperture polarimetry) was performed on the combined intensity images at each HWP angle. Extracted intensities were used to estimate the Stokes parameters of each star using The aperture and sky radii were the same as those used in the aperture photometry on total intensity (I) images. To obtain the Stokes parameters (Q and U ) in the equatorial coordinate system, an offset rotation of 105 • (Kandori et al. 2006;Kusune et al. 2015) was applied. We calculated the degree of polarization P and the polarization angle θ as follows: The errors in P and θ were also estimated by propagating the errors in Stokes parameters according to equations 4 and 5. Since the polarization degree, P , is a positive definite quantity, the derived P value tends to be overestimated, especially for low S/N cases. To correct for this bias, we calculated the de-biased P using P db = √ P 2 − δP 2 (Wardle & Kronberg 1974), where δP is the error in P . It should be noted here that, hereafter, we consider P db as P . The absolute accuracy of the offset polarization angle for SIRPOL was estimated to be less than 3 • (Kandori et al. 2006). The polarization efficiencies of the wire grid polarizer are 95.5%, 96.3%, and 98.5% at the J-, H-, and K s -bands, respectively, and the instrumental polarization is less than 0.3% over the field of view at each band (Kandori et al. 2006). Due to these high polarization efficiencies and low instrumental polarization, no additional corrections were made to the data. To obtain the astrometric plate solutions, we matched the pixel coordinates of the detected sources and the celestial coordinates of their counterparts in the 2MASS Catalog (Skrutskie et al. 2006), and applied ccmap, ccsetwcs and cctran tasks of the iraf imcoords package to the matched lists. The rms error in the returned coordinate system was ∼0.03 . However, since the 2MASS astrometry system is limited about to 0.2 , the stellar positional uncertainties should take that into account. A total of 919 stars had polarization detection in at least one band with P/σ P ≥ 1. Complete data sample The polarimetric data for the 919 stars were merged with the photometry of the 702 stars, and the resultant data are presented in Table 1. Star IDs are given in column 1. The photometric data of the stars, those do not have them from IRSF, are extracted from 2MASS (Skrutskie et al. 2006) and are denoted with an asterisk along with star IDs in column 1. The coordinates, aperture polarimetric and photometric results of each source are presented in columns 2 -12. Column 13 reports our classifications of the stars, as described in section 3 below. Though we have presented all the stellar data satisfying the P/σ P ≥ 1 criterion in Table 1 Black dots are the total sample in the entire region. The blue and red filled circles represent Class I and Class II sources, respectively in the RCW57A region. The black circles surrounding some of the points are three Class I and thirty two Class II sources having 2σ polarization measurements. The slanted dotted line denotes the boundary between Class I and Class II YSOs. The average error in the colors is shown. Identification of YSOs and foreground stars The B-field inside the molecular cloud can be probed by using polarimetry of background stars whose light becomes differentially extincted and linearly polarized by the dust in the cloud. Hence, a sample of background stars needs to be identified among all the stars with measured polarizations. For this purpose, we identify and exclude the following sources: YSOs (Class I, Class II, HAe/Be stars and NIR excess sources) based on MIR and NIR color-color diagrams, stars with Lband excess using the data from Maercker et al. (2006), stars projected against the nebulous region where their polarizations might be contami-nated by scattering (especially at the cluster center), the stars with non-interstellar polarization origin due to reflected light from their circumstellar disks/envelopes (whose properties are not discerned based on MIR and NIR colors), and the stars that have suffered depolarization effects due to multiple B-field components (associated with multiple dust layers) along their lines of sight. The latter two are identified using polarization efficiency diagram. Below, we present MIR and NIR two color diagrams (TCDs) and polarization efficiency (P versus H − K s ) diagram to identify and exclude the above mentioned stars with intrinsic polarization. Finally, the remaining sample, that are free from intrinsic polarization, is sub-categorized further into foreground (FG) and background (BG) stars, based on their NIR colors and polarization characteristics. YSOs identified by their mid-infrared colors Since the RCW57A region is still enshrouded in its natal molecular cloud, as evident from the 13 CO(1 -0) map (magenta contour in Figure 1 and red background in Figure 7), YSOs in the region can be deeply embedded. Therefore, the Spitzer MIR observations were used to probe deeper insight into the embedded YSOs. These occupy distinct regions in the Spitzer IRAC color plane, which makes MIR TCDs a useful tool for the classification of YSOs. Since 8.0 µm data are not available for the region, we used [[K s ] − [3.6]] 0 and [[3.6] − [4.5]] 0 (cf. Gutermuth et al. 2009) to identify deeply embedded YSOs, as shown in Figure 3. The zones of Class I and Class II YSOs, based on the color criteria of Gutermuth et al. (2009), are also depicted. Three sources were identified as Class I and thirty two sources as Class II and so were excluded from analyses. Identified Class I and Class II sources are mentioned in column 13 of Table 1. YSOs identified by their near-infrared colors NIR colors are also important tools to classify YSOs in star forming regions. Figure (Bessell & Brett 1988), respectively. The dotted blue line indicates the locus of unreddened CTTSs. The parallel blue dashed lines are the reddening vectors drawn from the tip (spectral type M4) of the giant branch (left reddening line), from the base (spectral type A0) of the MS branch (middle reddening line), and from the tip of the intrinsic CTTS line (right reddening line) with the extinction ratios A J /A V = 0.265, A H /A V = 0.155, and A K /A V = 0.090 (Cohen et al. 1981). The blue plus symbols on the reddening vectors show increments of A V by 5 mag each. The sources located in the "F" region could be either reddened field stars or Class II and Class III sources with small NIR excesses. The sources distributed in the "T" and "P" regions are considered to be mostly Classical T-Tauri stars (CTTSs; Meyer et al. 1997) or Class II sources with relatively large NIR excesses and likely Class I sources, respectively (for details see Pandey et al. 2008;Chauhan et al. 2011a,b). Typical errors in the plotted colors are comparable to the size of the symbols. (Maercker et al. 2006). Two of the stars found to be Class I also have L-band excess, and similarly 3 stars found to be Class II also have L-band excess. Star identified as having NIR excess, Lband excess, and nebulous background are noted in column 13 of Table 1. The remaining 342 stars that appear to be free from intrinsic polarization and distributed in the zone "F" in Figure 4 are (i) lightly reddened foreground field stars (FG), (ii) moderately reddened cluster members and highly reddened background stars (hereafter both cluster members and background stars referred to as BG), and (iii) weak line T-Tauri stars (WTTSs) or Class III sources. The locations of the WTTSs in the NIR TCD overlap with reddened background stars. Generally WTTSs exhibit almost negligible NIR-excess, as their disks will have been evaporated already or very optically thin (eg., Adams et al. 1987;Andrews & Williams 2005;Cieza et al. 2007). Therefore, it is reasonable to assume that these WTTSs are not intrinsically polarized by scattered light from the negligible amount of circumstellar material. In order to estimate the level of FG contamination, a control field (α = 11 h 18 m 12. s 22, δ = −61 • 59 38. 92 [J2000]) of angular size 7. 7×7. 7 located ∼60 from the RCW57A region was chosen. Comparing the distributions of [H − K s ] colors of the control field with colors in the RCW57A region shows that the field stars to have [H − K s ] < 0.15. Therefore, the 108 RCW57A stars with [H − K s ] < 0.15 and lying in the 'F' region of the NIR TCD are judged to be FG star candidates and remaining the 234 stars with [H − K s ] ≥ 0.15 and lying in the 'F' region are judged to be BG star candidates. Final confirmed FG and BG stars based on polarization efficiency diagrams Polarization efficiency is defined as the ratio of polarization degree to interstellar reddening, such as P (λ)/E(λ 1 − λ 2 ). Observed polarization efficiencies of dust grains are found to vary from one line of sight to the another. Reddening (E(B−V )) and polarization in V -band (P V ) are correlated and can be represented by the observational upper limit relation: (Serkowski et al. 1975). Polarization measurements that exceed this upper limit relation are generally attributed to the intrinsic polarization (non-magnetic polarization) caused by scattered light from dust grains in circumstellar disks or en- (Serkowski et al. 1975). We also seek to identify and exclude the stars that suffer depolarization due to their starlight's passing through multiple B-field components along the line of sight. For optically thin clouds, P (λ) is expected to be proportional to the column density if the field structure is uniform. In contrast, presence of multiple B-fields with different orientations along a line of sight will result depolarization (Martin 1974) and hence weakens the correlation between polarization and extinction (eg., Eswaraiah et al. 2011Eswaraiah et al. , 2012. Therefore, polarization efficiency diagrams (eg., Kwon et al. 2011;Sugitani et al. 2011;Hatano et al. 2013) will help identify stars with either intrinsic polarization or depolarization. Figure 5 show P versus [H−K s ] diagrams of FG and BG candidate stars (Section 3.2) with H-band data. These diagrams contain no known YSOs as we excluded them in the above Sections 3.1 and 3.2. The BG candidate stars show an increasing trend in polarization in accordance with an increasing [H − K s ] and well distributed below the upper limit interstellar polarization efficiency relation (gray line), Serkowski et al. 1975, also see Hatano et al. 2013 for detailed description on deriving polarization efficiency relations in NIR wavelengths.). To find the stars with non-magnetic polarizations (non interstellar origin of polarizations) and depolarization affect, first, we fit a straight line to polarization versus extinction for the BG stars with σ P (H) < 0.3% (encircled filled circles). Second, we drew dashed lines with slope of twice (upper dashed line) and half (lower dashed line) of the fitted slopes. Third, we identified the stars falling above and below these lines as probable BG stars having influenced by non-magnetic origin of polarization and depolarization affect, respectively. The remaining stars falling between the two dashed lines are ascertained as confirmed BG stars. Therefore, the final number of confirmed BG stars was found to be 178 in H-band. Additional 64 candidate BG stars falling above and below the dashed lines (plus symbols) are mentioned as probable BG stars with intrinsic polarization or depolarization in Table 1. FG candidate stars (open circles) exhibit two distributions and those with P (H) > 3% (plus symbols on open circles) may be affected by additional polarization caused by scattering or simply with larger uncertainties. We consider them as probable FG stars. Remaining 97 stars with P (H) ≤ 3% are considered as confirmed FG stars in H-band. These, confirmed and probable FG stars with P (H) > 3% are mentioned in Table 1. In the further analysis, we used only the confirmed 97 FG and 178 BG stars with Hband data. The fitted slope of P (H)/E(H − K s ) = 5.5±0.8% mag −1 (thick line) suggest that polarization efficiency of dust grains in the star forming cloud RCW57A are relatively higher than those of other Galactic line of sights (e.g., Hatano et al. 2013;Kusune et al. 2015). Separation of foreground and background magnetic fields FG stars towards any distant cluster are, generally, less polarized and extincted than BG stars (Eswaraiah et al. 2011(Eswaraiah et al. , 2012Pandey et al. 2013). This is because the dust layer(s) between the observer and the FG star contains fewer aligned dust grains, and because the degree of polarization and extinction are linearly correlated (Aannestad & Purcell 1973;Serkowski et al. 1975;Hatano et al. 2013). If the B-field orientation of the foreground dust layer is sufficiently different from that of the cloud, then the foreground stars would exhibit different polarizations (P and θ) from those of the background stars. Therefore, foreground and background stars can also be distinguished based on their polarization characteristics (P vs θ or Q versus U ), as well as with their NIR colors (Medhi et al. 2008(Medhi et al. , 2010Eswaraiah et al. 2011Eswaraiah et al. , 2012Medhi & Tamura 2013;Pandey et al. 2013;Santos et al. 2014). Figures 6(a)-(d) depict that the 97 confirmed FG and 178 confirmed BG stars are well separated in their polarization characteristics and [H − K s ] colors. Foreground stars have [H − K s ] < 0.15, P (H) < 3% with a peak around 1.0%, and a single Gaussian distribution of θ(H) (Gaussian mean of 65 • with a standard deviation of 14 • ). Background stars have [H − K s ] ≥ 0.15, P (H) between 0.5 -12% with a peak ∼ 2% and an extended tail towards greater P values and widely distributed θ(H) (range from 0 • to 180 • ) indicating the presence of complex B-field structure in RCW57A with a prominent peak at θ(H) ∼163 • . About 55% of stars are distributed between 110 -180 • , whereas the remaining 45% of stars are distributed between 0 -110 • . This suggests that the plane-of-sky B-field structure of RCW 57A is dominated by a component with θ(H) ∼163 • , which is nearly orthogonal to the foreground B-field component (65 • ) as well as being nearly orthogonal to the position angle of the major axis of the filament (∼60 • ). Figure 6(d) also reveals that the distributions of θ(H) of the confirmed FG and BG stars In order to account for the influence of FG polarization on the polarization angles of the confirmed BG stars, below we performed the FG subtraction. First, the polarization measurements of confirmed 97 FG stars were converted into Stokes parameters. Second, weighted mean FG Stokes parameters were computed and found to be Q F G = −0.332±0.008% and U F G = 0.845±0.007% (with standard deviation of 0.64% in Q F G and 0.70% in U F G ) and were subtracted vectorially from those of each confirmed BG star. Third, FGsubtracted Stokes parameters of the BG stars were then converted back to θ(H) FG−corrected . The Gaussian mean and standard deviation, for the distribution of offset polarization angles of the BG stars (θ(H) − θ(H) FG−corrected ), are found to be 1 • and 7 • , respectively. Hence, there was no obvious change in the FG corrected B-field morphology inferred from the BG stars. Therefore, we ignored the FG contribution to the BG polarization measurements. Figure 7 shows the vector map of the H-band polarization measurements of the 178 confirmed BG stars. The polarization vector map reveals a systematically ordered B-field that is configured into an hour-glass morphology which closely resembles the structure of the bipolar bubble. Magnetic field geometry The direction of expanding I-fronts or outflowing gas is revealed based on a large NS velocity gradient in a radio recombination line observations (de Pree et al. 1999). These high velocity outflows signatures are further traced in 13 CO(1 -0) position-velocity diagrams ( Figure 11; shown below). Shocked gas is visible at the SE and NW ends of the white arrows in Figure 7 (also see Figure 1), highlighting the outflow-cloud interaction. Townsley et al. (2011, see their Figures 3 and 4) have also showed that outflows from the deeply embedded protostars are responsible for the soft X-ray emission observed in the SE part of the cloud. These outflows have created the cavities depicted with enclosed yellow contours ( Figure 1). Interestingly, the directions of the expanding I-fronts and/or outflowing gas, as depicted with white arrows (Figure 7), follow the B-field orientations. B-field are compressed along the rim of BRC (located inside of white square box in Figure 7) and more details of which are addressed in Appendix B. Extracting essential parameters for estimating magnetic field strength To test whether B-fields play an active role in the formation of bipolar bubbles, it is essential to estimate the B-field strength by using the Chandrasekhar & Fermi (1953) relation (hereafter CF Method). For this purpose, below we determine the dispersion in θ(H) for the background stars, the gas velocity dispersion using CO data, and the dust and gas volume density using Herschel data. Dispersion in polarization angles: σ θ(H) The presence of multiple components of B-fields in RCW57A can be seen from the H-band polar-ization vector map overlaid on the 13 CO(1 -0) (Purcell et al. 2009) total integrated intensity image ( Figure 8) and the histogram of θ(H) (continuous histogram in Figure 6(d)). Each component follows its own spatial distribution and exhibits its own θ(H) distribution and dispersion. For this reason, we divided the observed area into five regions, named, A, B, C, D, and E, as shown in Figure 8 and Table 2. The spatial extent of each region is visually chosen in such a way that it should not include multiple components of θ(H). This criteria constrains the dispersion of θ(H) to be less than 25 • (Ostriker et al. 2001), so as to apply the CF method to estimate the B-field strength. Some portions are not included because they either have polarization data but not CO data (the area right to the C region), or have CO data but has multiple θ(H) distributions (area slightly above the E region), or contains few randomly oriented vectors (area distributed between D and E regions). We believe that the bias involved in the selection of five regions may not have significant impact on the final scientific results. A Gaussian function was fit to the θ(H) distribution for each region, as shown in Figure 9. The fitted means and dispersions in polarization angles, along with their errors, are listed in columns 7 and 8 of Table 2, respectively. Mean θ(H) (=µ θ(H) ) values of regions A, B, C, D, and E regions found to be 26 • , 164 • , 153 • , 132 • , and 74 • respectively. Dispersions in θ(H) (=σ θ(H) ) of five regions lie between 10 -16 • . Velocity dispersion: σ V LSR To derive the velocity dispersion in regions A, B, C, D and E, we have used 13 CO(1 -0) emission line mapping (Figure 8) performed using the Mopra telescope by Purcell et al. (2009). The observed 13 CO(1 -0) data has a velocity resolution of 0.4 km s −1 and a spatial resolution of 40 . CASA software was used to extract the mean velocity (V LSR ) versus brightness temperature (T ) spectrum of each region, which are plotted in Figure 10 using small filled circles. All the regions exhibit at least two velocity components, with clear asymmetric spectra towards higher velocities. These asymmetric, spatially extended, spectra correspond to outflowing gas from the embedded YSOs, expanding I-fronts, or both. Each of these spectra (T vs. V LSR ) were then fitted by Fig. 7.-H-band polarizations (red vectors) of the 178 confirmed BG stars overlaid on the tricolor image constructed from the 13 CO(1 -0) total integrated emission map (red) (Purcell et al. 2009), Spitzer IRAC Ch2+Ch3 combined image (green), and SIRPOL J + H + K s -band combined image (blue). Reference vector with P = 6% and θ = 90 • is plotted. Thick black contour represents SIMBA 1.2-mm dust emission (Hill et al. 2005) with a flux level of ∼3 Jy/beam. Thin black contours denote ATLASGAL 0.87-mm thermal dust emission ranging from 2 Jy/beam to 22 Jy/beam (beam size of 18. 2) with an interval of 1 Jy/beam. Cyan contour represent the extent of the Hii region, as traced in 3.4-cm radio free-free emission (de Pree et al. 1999). Yellow contours delineate the morphology of the bipolar bubble. Magenta contour depicts the extent of the molecular cloud, as traced by 13 CO(1 -0) emission using Mopra telescope (Purcell et al. 2009). White square box shows the location of a Bright Rimmed Cloud. White arrows represent possible orientations of expanding I-fronts or outflowing gas. Green plus marks and black squares distributed along the filament are the water/methanol masers and IRS sources, respectively. Blue diamonds correspond to the seven massive cores (namely, S1-M1, S1-M2, S3-M4, S3-M5, S3-C3, S4-M6 and S5-M6) identified by André et al. (2008). All the information on various contours, and the location of masers and IRS sources is extracted from Purcell et al. (2009), by permission. Yellow thick lines represent 'Cut 1' and 'Cut 2' along the ionization front or outflowing gas and cloud main axis, respectively, as described in the text and used as the basis for Figure 11. A reference length scale with 2 corresponding to 2.2 pc is shown as the horizontal white, labeled line. a multi-Gaussian function using the custom IDL routine 'gatorplot.pro' 4 . Details regarding this multi-Gaussian fitting are described in Appendix C. The resulting multi-Gaussian fitting parameters of each spectrum are presented in Table 3 and are plotted with black lines in Figure 10. The resulting combined spectrum of each region, the sum of all fitted multi-Gaussians (orange line), closely match the observed spectrum. The difference between observed and fitted spectra, the residuals, are shown with open squares and are closely distributed around zero T . Among the multiple velocity components of each region, we attribute the one related to the peak of the spectrum to the turbulent cloud component of that region (red Gaussian curve). These are used to estimate the B-field strength. The remaining Gaussian velocity components (black lines) are ascribed to the expanding 4 http://www.astro.ufl.edu/~warner/GatorPlot/ I-fronts and outflowing gas. These values are also given in column 9 of Table 2. The dashed line drawn at V LSR = −26.5 km s −1 (Figure 10) corresponds to the line center velocity over the areas covered by ABCD regions of RCW57A. This V LSR = −26.5 km s −1 component closely matches center of the red Gaussian components of all regions except for region E. However, based on the dense gas tracers (NH 3 , N 2 H + , and CS), Purcell et al. (2009) witnessed that the peak V LSR of the filament remain constant around −24 km s −1 (see their Figure 13), which is slightly different from the V LSR = −26.5 km s −1 . This is mainly because of the following reasons: (a) CO is optically thick and hence traces the less dense outer parts of the cloud, and (b) the ABCD regions, located slightly away from the cloud center, contain less dense gas that is influenced by the Hii region. Below, we try to resolve and quantify the contributions from turbulent cloud component, outflowing gas, and expanding I-front using positionvelocity (PV) diagrams. PV diagrams are useful diagnostics of gas kinematics in a star forming molecular clouds (e.g., Purcell et al. 2009;Zhang et al. 2016). Figures 11(a) and (b), show the P V diagrams corresponding to the two cuts shown and labeled as '1' and '2', respectively, in Figure 8. The cut '1' is along the orientation of expanding ionization front or outflowing gas (parallel to the minor axis of the filament) and the cut '2' is along the major axis of the filament. The velocity range over the whole region, based on the extent of the contours shown in Figures 11(a) and (b), spans −28.5 to −16.5 km s −1 a width ∆V 12 km s −1 . This implies that the entire region is dynamically active under the influence of Hii region. However, among all parts of the cloud that we observed, one of the most quiescent clumps S4 (André et al. 2008;Purcell et al. 2009) is situated to SE part of the filament. This clump is situated at ∼250 and at ∼ −26.5 km s −1 in Figure 11(b). Dust temperature T d (∼19 -24 K; André et al. 2008) and kinetic temperature T k (∼15 -25 K; Purcell et al. 2009) of the cores in S4, suggest that this clump is dynamically quiescent as being far from the Hii region. Gas velocity around this clump spans over ∼ −25 to ∼ −28 km s −1 (centered at ∼-26.5 km s −1 ) and a width of 2 -3 km s −1 (Figure 11(b)). Therefore, Table 3. According to the collect and collapse model of triggered star formation (see Torii et al. 2015, and references therein), the velocity separation due to expanding ionization front based on the P V diagrams to be ∼4 km s −1 . Therefore, Observed data points are shown with filled dots. The spectrum of each region was fit with multiple Gaussians shown with black lines. The combined spectrum is shown with a thick orange line. The mean velocity corresponding to the RCW57A region is drawn at −26.5 km s −1 using a dashed line. the outflowing gas may have remaining contribution of ∼5 -6 km s −1 in addition to the cloud turbulent component of ∼2 -3 km s −1 . Below we decompose the relative contribution of various velocity components in each region. In Figure 11(a), there exist two prominent redshifted, high-velocity components with peak intensities at −20 km s −1 and −23 km s −1 . These components are concentrated close to the cloud center. While the former component is attributed to the outflowing gas, the latter one is to the expanding I-front. The signatures of both the components prevail in A, B, C, and D regions. However, as shown in the bottom panel of Figure 10, the E region seems to be less influenced by the above two components, but has other two prominent components: (a) ∼ −25 km s −1 is related to the expanding I-front, and (b) ∼ −23 km s −1 is attributed to the turbulent cloud region. Moreover, spatially, the E region, in comparison to the others, is located far from the area influenced by the outflowing gas as depicted with white arrows (cf. Figure 7). This implies that the E region, containing a BRC, has a contribution from the expanding I-fronts alone, while the rest of the regions had the influences of both expanding I-fronts and outflowing gas. Therefore, the gas velocity dispersion due to cloud turbulent component (red lines in Figure 10) is well decomposed from other components using multi-Gaussian spectrum fitting. Number density: n(H 2 ) Herschel PACS (Poglitsch et al. 2010) 160 µm and SPIRE (Griffin et al. 2010) 250 µm, 350 µm, and 500 µm images of RCW57A were used to construct a dust column density map by using the modified black body function (details are given in Appendix D). The resolution of the final column density map is 36.4 . To obtain the number densities, we further used Plummer-like profile fitting on column density profiles to obtain the volume density map as described below. To construct the column density (N (H 2 )) profiles, first we manually identify the cloud spine using the pixels that have maximum column densities and distributed along the filament. Thus identified spine, depicted with green squares in Figure 12(c), divides the entire cloud region into two parts NW and SE. Second, for each spine element, we identify the N (H 2 ) profile elements distributed on a line that bisects the filament long axis and passes through that spine element. The dimensions of spine elements as well as N (H 2 ) profile elements are chosen as 36.4 × 36.4 . To achieve the Nyquist sampling, separation between adjacent elements was chosen as 18.2 . Third, we have constructed N (H 2 ) profiles as a function of projected distances from the filament spine elements towards both NW and SE regions, and are plotted using gray circles in Figures 12(a) and (b), respectively. Every point and its uncertainty, in this plots, corresponds to the mean and standard deviation of column density values in a particular N (H 2 ) profile element. The mean column density profiles, using distance wise averaged column densities of the N (H 2 ) profile elements distributed parallel to the spine, are shown with red filled circles. Similarly, the error bars are corresponding to the standard deviations in corresponding mean column densities of N (H 2 ) profile elements. Assuming that the filament follows an idealized cylindrical model, we have extracted the volume density (n(r)) profile correspond to each N (H 2 ) profile by using Plummer-like profile (Nutter et al. 2008;Arzoumanian et al. 2011;Juvela et al. 2012;Palmeirim et al. 2013) The factor, A p = π, is estimated using the equation A p = +∞ −∞ [(1 + u 2 ) p/2 cos i] −1 du (Arzoumanian et al. 2011) and by assuming that the filament is on the plane of the sky implying an inclination angle of i = 0, and p = 2. In equation 7, N (r) is the column density profile as a function of offset distance (r) from the cloud spine. The fitting was performed using IDL mpfit nonlinear least-squares fitting programme (Markwardt 2009) and extracted three parameters: (i) n c the central mean volume density of the spine point, (ii) R f lat the radius within which the N (H 2 ) profile is flat, and (iii) p the power-law index that determines the slope of the power law fall beyond R f lat . Plummer-like profile fit was performed only to the data spanning to ∼5 pc (7.3 arcmin) from the spine, beyond which background column density dominates (horizontal dashed lines in Figures 12(a) & (b)). The fitted values of n c , R f lat , and p are (7.6±1.3)×10 4 cm −3 , 0.24±0.11 pc and 2.1±0.1 for NW part ( Figure Fig Contours are plotted at CO intensity ranging from −3 K to 10 K with a step of 1 K. Blue dotted line connects the three brightest intensity points along the y-axis denoting a velocity gradient of 13 km s −1 pc −1 . (b) Same as (a) but for a cut along the cloud major axis (cut '2' with a position angle = 65 • , and central offset coordinates are α = 11 h 12 m 04 s , δ = −61 • 18 44 [J2000]). Contours are plotted at CO intensity ranging from −2 K to 26 K with a step of 2 K. Blue dotted line connects the brightest intensity points along the x-axis denoting a velocity gradient of 0.35 km s −1 pc −1 . Unit of the color scale in both panels is K. 12(a)), and (5.1±0.9)×10 4 cm −3 , 0.37±0.18 pc and 2.0±0.1 for SE part (Figure 12(b)). These fit results, as indicated in the plots, are similar. Volume density map, created using n(r) profiles of all the filament spine elements, is shown in Figure 12(c). Since we considered spine point on the filament as the zeroth element while performing Plummer-like profile fitting for both the NW and SE regions, each spine point will have two volume density values and are averaged in the final map. The minimum and maximum n(H 2 ) values on the filament spine lie between 2.48×10 3 cm −3 and 1.52×10 5 cm −3 . The mean n(H 2 ) values for pixels at the extreme-end of the NW and SE regions are 111±26 cm −3 and 228±74 cm −3 , respectively. This implies that the RCW57A is still embedded in a diffuse background cloud. We found that more than 95% of the polarization vectors are centered outside of the blue isodensity contour corresponds to volume density n(H 2 ) = 13,000 cm −3 . It implies that our NIR polarimetry fails to trace adequately the highly extincted parts as as shown in Figure 12(c). There-fore, we exclude the pixels with n ≥ 13,000 cm −3 while estimating the mean volume densities in A, B, C, D, and E regions. Means and Poisson errors (standard deviation divided by square root of the number of measurements) of the volume densities for the five regions (A, B, C, D, and E) are estimated and are listed in column 10 of Table 2. More details on producing volume density map from column density can also be found at Smith et al. (2014) and Hoq et al. (2017). It should be noted here that there exist 12 stars within the high column density regions (within the contour of volume density n(H 2 ) = 13,000 cm −3 or column density N (H 2 ) 1×10 23 cm −2 ). Only one star (star ID = 290; Table 1) to the cloud. Remaining 11 stars are consistent with either reddened cluster members embedded in the cloud or reddened field stars lying in the outskirts of the cloud but projected on the high density parts of the cloud region. Magnetic field strength using Chandrasekhar-Fermi method Using the dispersion in the polarization angles (σ θ(H) ; column 8 of Table 2), velocity dispersion values (σ V LSR ; column 9 of Table 2) and mean volume densities (n(H 2 ); column 10 of Table 2), for the five regions of RCW57A, we estimated the B-field strength using the Chandrasekhar & Fermi (1953) relation: The mass density ρ = n(H 2 ) m H µ H2 , where n(H 2 ) is the hydrogen volume density, m H is the mass of the hydrogen atom, and µ H2 ≈ 2.8 is the mean molecular weight per hydrogen molecule and includes the contribution from helium. The correction factor Q = 0.5 is included based on the studies using synthetic polarization maps generated from numerically simulated clouds (Ostriker et al. 2001;Heitsch et al. 2001) which suggest that for σ θ ≤ 25 • , B-field strength is uncertain by a factor of two. Uncertainties in B-field strength were estimated by propagating the errors in σ θ(H) , σ V LSR and n(H 2 ) values. The estimated B-field strengths and corresponding uncertainties are given in column 11 of Table 2. The uncertainties in B-fields, derived by propagating errors in n(H 2 ), σ V LSR , and σ θ H , are considerably large (see column 11 of Table 2). We also estimate the B-field strengths and corresponding uncertainties without considering errors in σ θ H values. The resultant B-field strengths, given in column 12 of Table 2, for A, B, C, D, and E regions are to be 123±21µG, 89±13µG, 63±15µG, 108±27µG, and 74±8µG, respectively. Mean B-fields strength over five regions is estimated to be 91±8µG. To understand the importance of B-fields with respect to turbulence, we estimated the magnetic pressure and turbulent pressure using the relations P B = B 2 /8π and P turb = ρσ turb 2 (where σ turb = σ V LSR is given in Table 2), respectively, and the results are given in Table 4. The mean P B /P turb is estimated to be 2.4±0.6. These estimated parameters suggest that the magnetic pressure is more than the turbulent pressure at least by a factor of ∼2 in all five regions, signifying the dominant role of B-fields over turbulence. For the five regions together, the mean magnetic and turbulent pressures are estimated to be (35±7)×10 −11 (dyn cm −2 ) and (15±3)×10 −11 (dyn cm −2 ), respectively. Danziger (1974) estimated the mean electron density, n e , as ∼20 cm −3 , and mean electron temperature, T e , as ∼10000 K for the RCW57A region. For the entire region the thermal pressure P th using the relation P th 2n e kT e (where k is the Boltzmann constant), is estimated to be 6×10 −11 dyn cm −2 , which is smaller than the mean magnetic pressure. Ratio of mean magnetic to thermal pressure, P B /P th , is found to be ∼6. A lower value of P th again suggests dominant magnetic pressure over thermal pressure. Therefore, in comparison to thermal and turbulent pressures, magnetic pressure is playing a crucial role in guiding the expanding ionization fronts or outflowing gas in RCW57A. The implications on the relative importance of B-field pressure in comparison to turbulent and thermal pressures on the formation and evolution of bipolar bubble is discussed in the Section 5.3. Anisotropic distribution of material in the cloud and the bipolar bubble Based on the CO and Spitzer images, number of stars with polarization detection, and the column density profiles we infer on the distribution of gas and dust in RCW57A. The quiescent molecular cloud gas traced by CO (red background in Figure 7) is absent at the foot points of, as well as along, the bipolar bubbles. Moreover, the spatial distributions of the CO gas and the bubbles (green background by Spitzer images) appear to be anti coincident with each other. This implies that cloud material has been either eroded or blown out by the expanding I-fronts, outflowing gas, or strong stellar winds from the embedded early type protostars. Therefore, CO and Spitzer maps suggest that bipolar bubble likely blown by the massive stars emanated stellar winds, outflowing gas, and expanding ionization fronts (see Townsley 2009). While the SE bubble appears to be closed towards the SE, the NW bubble is widely extended towards the NW (Figures 1 and 7) and exhibits large-scale loops, likely due to anisotropic distribution of material towards SE and NW regions. This is further corroborated by the following facts. Although the area together covered by regions A, B and C (located in NW) is similar to that of D (located in SE), the total number of background stars with polarization detection (51) in A, B, and C exceeds those in D (21) (see Table 2, and Figures 7 and 8). Furthermore, the mean N (H 2 ) profiles suggests that anisotropic distribution of gas and dust in RCW57A i.e., SE part (N (H 2 ) ∼1.3×10 22 cm −2 ) has relatively more background material than in the NW (N (H 2 ) ∼0.70×10 22 cm −2 ) as depicted with dashed lines in Figures 12(a) and (b). Similar asymmetric distribution of column density profiles was seen towards other region (e.g., Taurus B211/3 filament; Palmeirim et al. 2013). An X-ray study (Townsley et al. 2011, see their Figures 3 and 4) revealed the presence of OB stellar association (NGC 3576OB) towards NW of the main embedded star forming region RCW57A (or NGC 3576GHIIR). Furthermore, presence of hard X-ray emission and a pulsar (+PSRJ1112-6103 & PWN) in NGC 3576OB region may suggests a fast occurrence of supernovae event. This supernova together with OB stars of NGC 3576OB stellar association, most probably, evacuated the material in the NW part of the RCW57A. This could be the plausible reason behind the low density material observed in the NW region. The observed large scale loops in the NE, possibly, formed due to the newly formed material as a result of expanding Ifronts driven by the embedded massive protostars of RCW57A. A high velocity outflow signature discernible close to the eastward from the center, while the same feature is absent to the westward as shown in Figure 11(b). This could attribute to the confined ionized gas flow towards west from center possibly due to the presence of a dense clump (black thin contours in Figure 7). High velocity component is only seen towards the east which implies dearth of dense material there and hence ionized gas expanded with relatively more velocity. de Pree et al. (1999) and Purcell et al. (2009) have also shown that the Hii region expands more freely towards the east and is bounded to the west. Column density profiles: filament width and B-field support We have fitted a Gaussian to the mean column density profiles spread over ∼5 pc radii and resultant FWHM, corresponds to the filament characteristic width, is found to be 1.55±0.09 pc. However, a constant filament width ∼0.1 pc has been determined for the central dense sections of filaments of the nearby star forming regions in the Herschel Gould Belt survey by fitting a Gaussian to the inner sections (radii upto 0.3 -0.4 pc) of the filaments by (Arzoumanian et al. 2011). Therefore, we fit a Gaussian to the inner 1 pc width of the clubbed mean column density profiles of NW and SE regions and the resultant σ and FWHM are found to be 0.21±0.08 pc 0.50±0.18 pc, respectively. Within the error, the filament width (0.50 pc; green lines in Figures 12(a) and (b)) of RCW57A is higher than the constant filament width ∼0.1 pc (Arzoumanian et al. 2011). Filament widths larger than 0.1 pc (spanning over 0.1 pc -1 pc with the typical widths of ∼0.2 -0.3 pc) has also been witnessed in a number of studies (eg., Hennemann et al. 2012;Juvela et al. 2012). However, the filament width of RCW57A is not resolved because of the limited resolution (36. 4 corresponds to 0.42 pc as shown with cyan lines) of the column density map. The mean power-law index p = 2.02±0.05, derived from Plummer-like profile fitting for RCW57A, is consistent with other studies (Arzoumanian et al. 2011;Juvela et al. 2012;Palmeirim et al. 2013;Hoq et al. 2017). This similar value of index (p = 2) is expected if the filament is supported by B-fields (Hennebelle 2003;Tilley & Pudritz 2003) which is in accordance with our observational results supporting active role of Bfields in the formation and evolution of filament as described in Section 5.3 below. Evolutionary scenario of filament and bipolar bubble Based on the morphological correlations among (i) dense filamentary cloud structure seen in sub-mm (at 0.87-mm ATLASGAL and 1.1-mm SIMBA dust continuum emission maps), (ii) bipolar bubble seen in mid infrared images, and (iii) the B-field morphology of RCW57A trace by NIR polarimetry, we postulate a scenario as illustrated with four schemas in Figure 13. Each schema describes our observational evidence along with our prediction at a particular evolutionary state of RCW57A. Schemas A and B: As per the background star polarimetry, the filament (position angle ∼ 60 • ; Figure 7) is oriented perpendicular to the Bfields (the dominant component of θ(H) = ∼ 163 • ; Figure 6(d)). From this observational evidence, we speculate the formation history of filament and the role of B-fields in them. Initially a subcritical molecular cloud is supported by B-fields as shown in schema A, later compresses into a filamentary molecular cloud due to B-fields guided gravitational contraction as per schema B. Although the role of B-fields in the formation and evolution of filamentary molecular clouds is still a matter of debate, its morphology with respect to the geometry of filaments is well established. The geometry of B-fields in a molecular cloud is mainly governed by the relative dynamical importance of magnetic forces to gravity and turbulence. If the B-fields are dynamically unimportant compared to the turbulence, the random motions dominate the structural dynamics of the clouds, and the field lines would be dragged along the turbulent eddies (Ballesteros-Paredes et al. 1999). In that case, B-fields would exhibit chaotic or disturbed B-field structures. Based on the facts that the B-field structure is more regular with an hourglass morphology and B-field pressure dominates over turbulent pressure (see Section 4.4), we do not favor weak B-field scenario in RCW57A. If B-fields are dynamically important then the support to the molecular cloud against gravity Fig. 13.-Schematic diagram illustrates the possible connection between filament, bipolar bubble and B-fields. Initially, subcritical molecular cloud permeated by uniform B-fields (schema A) could have gravitationally compressed along the B-fields into a filament (schema B). Later, as mass accumulated at the center of the filament, gravitationally contracting cloud could have pulled the B-fields along the cloud material towards the central region. Thus, B-fields followed a clear hour-glass morphology when the collapsing cloud reached supercritical state (schema C). Dense and high pressurized Hii region formed due to the formation of massive stars at the cloud center (schema D). Further, propagation of ionization fronts and stellar winds is rather controlled by the an isotropic distribution of gas and dust in the filament. Presence of B-fields in an hour-glass morphology is further introduced an additional anistropic pressure in favor of expanding ionization fronts or outflowing gas so as to form the bipolar bubble (schema D). is rendered predominantly by the B-fields. In this case, B-fields would be aligned, preferentially, perpendicular to the major axis of the cloud (Mouschovias 1978). This is expected as the cloud tends to contract more in the direction parallel to the B-fields than in the direction perpendicular to the it. Recent polarization studies have also demonstrated that B-fields are well ordered near dense filaments and perpendicular to their long axis (Moneti et al. 1984;Pereyra & Magalhães 2004;Alves et al. 2008;Chapman et al. 2011;Sugitani et al. 2011;Palmeirim et al. 2013;Li et al. 2013;Franco & Alves 2015;Cox et al. 2016;Planck Collaboration et al. 2016). Numerical simulations have also shown that filamentary molecular clouds form by gravitational compression guided by B-fields (e.g., Nakamura & Li 2008;Li et al. 2013). Therefore, we believe that the fil-ament in RCW57A could have formed due to the gravitational compression of cloud along the Bfields. Schema C: Due to the accumulation of more mass at the center, the filament is achieved a supercritical state from its initial B-field supported subcritical state. B-fields are dragged along the gravity-driven material contraction towards the strong gravitational potential at the cloud center, and subsequently they configured into an hour-glass morphology (schema C). Similar hourglass morphology of B-fields were observed towards Orion molecular cloud 1 (Schleuning 1998;Vaillancourt et al. 2008;Ward-Thompson et al. 2017) and Serpens cloud core (Sugitani et al. 2010) signifying ongoing gravitational collapse. In RCW57A, B-fields in some portions exhibit clear evidence for hour-glass morphology. In the NE part of the filament (covered by two regions A and D; see Figure 8), B-fields are bent by ∼ 90 • as the mean θ(H) for region A is ∼26 • and for D is ∼132 • (see column 7 of Table 2). Therefore, B-fields are distorted in the NE part of the filament, signifying a collapsing filamentary molecular cloud. However, this similar distortion is not clearly apparent in the SW part of the filament, partly because of the light from the background stars is obscured by the dense cloud material. Therefore, the NIR polarimetric observations were unable to detect sufficient numbers of background stars in the SW part to delineate the detailed structure of the B-field there. Schema D: Gravitational instability caused the filament fragmentation into seven cores (blue diamonds in Figure 7) as evident from dust continuum emission maps (André et al. 2008) as well as dense gas tracers (Purcell et al. 2009). Star formation has began at the central region due to the onset of gravitational collapse during which B-fields were not strong enough to stop the cloud collapse. Ionization and shock fronts, driven by the Hii region, are propagated in to the ambient medium. Interaction between the Hii region and dense cloud material is evident from Figure 7 and as a consequence of this, several IRS sources are formed at the boundary of interaction signifying triggered star formation at the edges of Hii region of RCW 57A. Our observational results suggest that B-field pressure dominates over thermal pressure (see Sec-tion 4.4), implying active role of B-fields in governing the feedback processes. Since the B-fields observed to be responsible for the formation of filament in RCW57A (Schema B), it is expected that B-fields anchored through the filament may also have significant influence on the expanding I-fronts. In addition to the anisotropic pressure that the I-fronts experience while they expand anisotropically in the filament, they undergo an additional anisotropic pressure in the presence of strong B-fields (Bisnovatyi-Kogan & Silich 1995). As a result I-fronts experience accelerated flows along the B-fields (i.e., low pressure) and hindered flows (i.e., high pressure) in the perpendicular direction to the B-fields (Tomisaka 1992;Gaensler 1998;Pavel & Clemens 2012;van Marle et al. 2015). Therefore, I-fronts expand into greater extent along the hour-glass shaped B-fields and eventually form the bipolar bubble as shown in schema D of Figure 13. Similar applications for the influence of B-fields on I-fronts have found at other environments. Bubbles around young Hii regions and supernova remnants have been found to be elongated along the Galactic B-field orientation (Tomisaka 1992;Gaensler 1998;Pavel & Clemens 2012). Falceta-Gonçalves & Monteiro (2014) have studied that the strong B-fields can lead to form the bipolar planetary nebula. van Marle et al. (2015) have investigated the influence of B-fields on the expanding circumstellar bubble around a massive star using magneto-hydrodynamical simulations. They found that that the weak B-fields cause circumstellar bubbles become ovoid rather than spherical shape. On the other hand, strong B-fields lead to the formation of a tube-like bubble. Therefore, according to the proposed evolutionary scenario (Figure 13), initially B-fields in molecular clouds may be important in guiding the gravitational compression of cloud material to form a filament. In the later stage, hour-glass B-fields might have guided the expansion and propagation of I-fronts to form the bipolar bubble. B-fields favoring the formation of massive cluster at the center of the filament Based on the presence of 29 NIR excess sources (Persi et al. 1994) and finding of 51 stars earlier than A0 (Maercker et al. 2006), it has been found that RCW57A hosts a massive infrared cluster at the center (within the cyan contour encloses Hii region in Figure 7) of the filament. Furthermore, the Chandra X-ray survey by Townsley et al. (2014, see also Townsley et al. 2011 has provided a first detailed census about the members in the embedded massive young stellar cluster (MYSC; see their Figures 9c and 9d). Their study unraveled many highly-obscured but luminous X-ray sources (one source exhibits photon pile-up) that are likely the cluster's massive stars ionizing its Hii region. This highly-obscured X-ray cluster is situated at the SW edge of the infrared cluster. Formation history of this cluster is poorly understood. Below, we propose that the large scale B-fields traced by NIR polarimetry may also govern crucial role in the formation of this massive cluster. Initial turbulence in the molecular clouds may decay rapidly. To maintain turbulence in the cloud and hence to govern cloud stability against rapid gravitational collapse, supersonic turbulence should be replenished by means of prostellar outflows. In the model of outflow driven turbulence cluster formation (Li & Nakamura 2006;Nakamura & Li 2007, and references therein), Bfields are dynamically important in governing two processes. First, the outflow energy and momentum are allowed by the B-fields to escape from the central star forming region into the ambient cloud medium. Second, B-fields guide the gravitational infalling material (just outside of the outflow zone) to the central region. Therefore, the competition between protostellar outflow-driven turbulence and gravitational infall regulates the formation of stellar clusters in the presence of dynamically important B-fields. Observational signatures towards the Serpens cloud core (Sugitani et al. 2010) is in accordance with the scenario of B-field regulated cluster formation: (a) B-fields should align with the short axis of the filament, outflows and gravitational infall motions, and (b) outflow injected energy should be more than the dissipated turbulence energy in order to maintain the supersonic turbulence in the cloud. In the case of RCW57A, the presence of a deeply embedded cluster, and alignment of the I-fronts, outflowing gas, and bipolar bubbles with the NIR-traced Bfields, trace the pre-existing conditions in favor of cluster formation according to the protostellar outflow driven turbulence cluster formation in the presence of dynamically strong B-fields. Summary and Conclusions Though there exist few studies regarding the formation of bipolar bubbles, none of them has explored the importance of B-fields. Our aim was to understand the morphological correlations among the B-fields, filament, and bipolar bubbles, and their implications for the star formation history in RCW57A. We conducted NIR polarimetric observations in the JHK s -bands using SIRPOL to delineate the B-field structure in RCW57A. We employed various means (NIR and MIR color-color diagrams, polarization efficiency diagrams, and the polarization characteristics) to exclude YSOs with possible intrinsic polarization. Through our analyses, we separated 97 confirmed foreground stars from 178 confirmed background stars having H-band polarimetry. Below we summarize the results of our present work. • The foreground dust dominated by a single component B-field having a mean θ(H) ∼ 65 • , which is different from the position angle (101 • ) corresponding to the Galactic plane. • The polarization values of the background stars consistent with the dichroic origin of dust polarization and exhibit relatively higher polarization efficiencies, suggesting efficiently aligned dust grains in RCW57A. • The polarization angles of the reddened background stars reveal that the B-field in RCW57A is configured into an hour-glass morphology which follows closely the structure of the bipolar bubbles. The dominant component of the B-field (θ(H) ∼ 163 • ) is perpendicular to the filament major axis. The orientations of both outflowing gas from the embedded protostars and the expanding ionization fronts from the Hii region are aligned with the B-fields. • The B-field strength averaged over five regions across RCW57A based on the Chandrasekhar-Fermi method, is 91±8µG. The B-field pressure is estimated to be more than turbulent and thermal pressures. • Morphological correlations among B-fields, filament, and bipolar bubbles as well as the dominance of B-field pressure over turbulent and thermal pressures suggest a scenario in which B-fields not only play an important role in formation the filamentary molecular cloud but also in guiding the expansion and propagation of I-fronts and/or outflowing gas to form bipolar bubbles. In this picture, B-fields impart additional anisotropic pressure to expanding I-fronts, from Hii regions, in order to be expanded and propagated into greater extent. • Protostellar outflow driven turbulence and gravity in the presence of dynamically important B-fields might be responsible for the cluster formation. Therefore, our study, based on the link between B-fields, filament, and bipolar bubble, traces the preexisting conditions where B-fields might also be important in the formation of massive cluster in RCW57A. Acknowledgments We thank Prof D. P. Clemens for helpful discussions and providing constructive suggestions in the improvement of the draft. We thank the anonymous referee for his/her careful reading and insightful comments which have improved the contents of the paper. CE, SPL and JWW are thankful to the support from the Ministry of Science and Technology (MoST) globule prior to the formation of BRC would be ∼163 • (which corresponds to the dominant component of B-fields in the cloud before the onset of star formation, see Figure 6(d) and Section 4.1) and is shown with a broken cyan line in Figure 15. The direction of propagating ionizing radiation from the Hii region with respect to the BRC makes a position angle of 95 • as shown with a green arrow. This implies that initial B-fields around BRC (∼ 163 • ) are oriented nearly orthogonal to the direction (∼ 95 • ) of I-fronts. Thus, the estimated B-field strength (74±8 µG) and its initial orientation around the BRC are similar to those of R5 models of Mackey & Lim (2011a) involving perpendicularly oriented B-fields with moderate strength (50µG). We believe that this BRC might be formed due to a portion of ionizing radiation that is leaked through the low density parts of the molecular cloud, otherwise it would have evaporated if all the radiation from Hii region is impinged on BRC. Therefore, this BRC might experienced the influence of thermal pressure equivalent to that emanated from a single O-type star. Mackey & Lim (2011a) also used only a single O-type star as the ionizing source in their models. According Mackey & Lim (2013), at 400 kyr, the models with moderate B-fields (∼ 50 µG), vertical motions along the B-fields are initiated (see left panel of Figure 1 of Mackey & Lim 2013). Due to this, gas flows in response to photo-evaporation flows along the B-fields result in the formation of dense ridge filament elongated parallel to the B-fields. At 500 kyr a clear cometary globule is formed in which B-fields are aligned with their bright rim similar to our results. Thus, our results are in accordance with RMHD models of Mackey & Lim (2011a with moderate B-field strength (initially oriented perpendicular to the direction of the radiation propagation). Observational evidences similar to our present results includes compressed B-fields along but behind the bright rim SFO 74 (Kusune et al. 2015). C. Multi-Gaussian fitting on T vs. V LSR spectra Initial guess values were provided by visual inspection of observed spectra. The number of Gaussian components are determined based on the number of peaks appeared in each spectrum. Since the 'gatorplot.pro' does not provide the errors in the fitted multi-Gaussians, we further used an output of 'gatorplot.pro' as an input to 'mpfitpeak.pro' of Markwardt IDL suite of functions (Markwardt 2009) to ascertain the uncertainties to the fitted measurements. Instead of varying all three parameters of one Gaussian (peak:T peak , center: V LSR and width: σ V LSR ) we fixed two of them and varied one (similarly, in case of three Gaussians, out of 9 parameters three were varied and six were kept constant). Likewise we constrained all the multi-Gaussian components of each spectrum along with their fitted errors using reduced chi-square minimization method. Equal weights are given to all points of an observed spectrum while fitting was performed. D. Column density map using Herschel data Column density map has been constructed using the following relation (Kauffmann et al. 2008) where S beam ν is the flux per beam (mJy beam −1 ), N H2 is the column density and K ν (=0.1 ν(GHz) where A pixel (Sr) is the pixel area. All the images were smoothed to match the resolution of 36. 4 corresponds to the SPIRE 500µm image. Equation D1 is fitted to the pixel-wise flux values of four images using 'mpfit.pro of Markwardt IDL suite of functions (Markwardt 2009). Equal weight is given to four input fluxes while fitting is performed. (Skrutskie et al. 2006) The polarization values given in columns 4-6 are de-biased The uncertainties in the polarization angles given in columns 7-9 are not accounted for overall angular calibration uncertainty of 3 • , which is a systematic uncertainty which would affect all the measured polarization angles Column 13: 'Foreground' means confirmed foreground stars Column 13: 'Background' means confirmed background stars or cluster members Column 13: 'Foreground † ' means probable foreground stars with P (H) > 3% (see section 3.3) Column 13: 'Background † † ' means probable background stars or cluster member with excess polarization (see section 3.3) Column 13: 'Background ‡ ' means probable background star or cluster member with depolarization (see section 3.3) The entries those do not have 2MASS photometric uncertainties are left blank A portion of the table is given here, entire table will be available online Table 2: Central coordinates, widths, number of stars, mean (µ θ(H) ) and dispersion (σ θ(H) ) in polarization angles (using Gaussian fitting), velocity dispersion (σ V LSR ) using 13 CO(1 -0) data, volume density (n(H 2 )) using Herschel data and Plummer-like profile fitting on column density maps, and magnetic field strength estimated using Chandrasekhar-Fermi method for the regions A, B, C, D and E. Note: † Errors in B-fields strengths were estimated by propagating uncertainties in n(H2), σ θ(H) , and σ V LSR . ‡ Errors in B-fields strengths were estimated by propagating uncertainties only in n(H2) and σ V LSR . Table 3: Multiple Gaussian fitted components of 13 CO brightness temperature (T ) versus V LSR spectra of Table 4: Magnetic and turbulent pressures along with their uncertainties in the regions A, B, C, D and E. Region P B P turb P B /P turb 10 −11 (dyn cm −2 ) 10 −11 (dyn cm −2 ) (1) (2) (3) (4) A 60 ± 20 15 ± 5 4 ± 2 B 32 ± 9 20 ± 6 2 ± 1 C 16 ± 8 5 ± 2 3 ± 2 D 46 ± 23 22 ± 11 2 ± 1 E 22 ± 5 11 ± 3 2 ± 1 Note: In this table, B-field pressure values and their uncertainties are derived using B-field strength and corresponding uncertainties given in column 12 of Table 2.	16,935.8	2017-10-03T00:00:00.000	[ "Physics", "Environmental Science" ]
An Iterative Scheme of Arbitrary Odd Order and its Basins of Attraction for Nonlinear Systems In this paper, we propose a fifth-order scheme for solving systems of nonlinear equations. The convergence analysis of the proposed technique is discussed. The proposed method is generalized and extended to be of any odd order of the form 2n 1. The scheme is composed of three steps, of which the first two steps are based on the two-step Homeier’s method with cubic convergence, and the last is a Newton step with an appropriate approximation for the derivative. Every iteration of the presented method requires the evaluation of two functions, two Fréchet derivatives, and three matrix inversions. A comparison between the efficiency index and the computational efficiency index of the presented scheme with existing methods is performed. The basins of attraction of the proposed scheme illustrated and compared to other schemes of the same order. Different test problems including large systems of equations are considered to compare the performance of the proposed method according to other methods of the same order. As an application, we apply the new scheme to some real-life problems, including the mixed Hammerstein integral equation and Burgers’ equation. Comparisons and examples show that the presented method is efficient and comparable to the existing techniques of the same order. Introduction One popular research area in mathematics is to find the solution a ¼ ða 1 ; a 2 ; . . . ; a n Þ t of the system of nonlinear equation . . . ; f n x ð ÞÞ t , and X ¼ ðx 1 ; x 2 ; . . . x n Þ t 2 R n . This type of problems occurs in many applied sciences like engineering, physics, biology and chemistry. Many researchers developed iterative methods for solving this kind of systems using different techniques. The most popular iterative method for solving system of nonlinear equations is the well-known Newton's method which has second order of convergence [1]. To improve the order of convergence and increase the accuracy of the solution obtained, many researchers tried to improve Newton's method. Some authors used different forms and modifications based on Adomian decomposition technique for solving systems of nonlinear equations, see for instance [2][3][4][5][6]. Another way to improve some schemes for systems of nonlinear equation is by using homotopy analysis method and homotopy perturbation method, see for example [7,8]. Grau-Sánchez et al. [9] used the harmonic mean of the derivative to improve an iterative scheme for solving systems of nonlinear equations. By applying some quadrature formulas, some researchers implement their techniques to solve systems of nonlinear equations, for instance [10][11][12]. Also, some derivative-free schemes for systems of nonlinear equations were proposed, see for example [13][14][15] and the references therein. One of the well-known modifications of Newton method is Jarratt method of order four. Cordero et al. [16] extended Jarratt method to solve systems of nonlinear equations preserving the same order of convergence. Many variants of Jarratt type methods have been developed, see for example [17][18][19] and the references therein. Many other different orders of convergence schemes for nonlinear modules can be found in the literature, see for example [20,21] and the references therein. Some techniques to improve the order of convergence of the iterative schemes for systems of nonlinear equations have been proposed, for instance, see [22,23]. In general, obtaining a higher-order iterative method is not the only important thing; as the computational and the time cost are crucial issue also. So, establishing a high order iterative method based on low computational and time cost is very important. In this paper, we develop a new multi-step scheme of arbitrary odd order for nonlinear equations. The proposed method can be used in the multidimensional case preserving the same order. The convergence analysis of the new scheme is discussed. Several examples are given to show the efficiency of the generalized method and its comparison with other iterative schemes of the same order. To confirm the applicability of the new technique, we apply the new technique to some real-life problems. The Proposed Method In this section we will derive the proposed technique for nonlinear modules. We begin by writing the function f x ð Þ as: As we want f x ð Þ ¼ 0, and by using midpoint quadrature formula and writing the equation as an iterative scheme, one gets Now, to write the iterative scheme (2) in explicit form, is the Newton step. So, scheme (2) becomes: The iterative method (3) was proposed by Frontini and Sormani [10]. The multidimensional case of scheme (3) was discussed by Homeier [24] and can be written as: where F 0 ðX n Þ À1 and F 0 ðY n Þ À1 are the inverse of the first Fréchet derivative of F X n ð Þ and F Y n ð Þ respectively. Scheme (4) is of third-order of convergence, and requires at each iteration the evaluation of one function, two Fréchet derivatives and two matrix inversions. In order to increase the convergence order and the computational efficiency of scheme (4) Sharma et al. [20] proposed a new scheme of the fifth-order of convergence by adding one step to scheme (4): W n ¼ X n À F 0 ðY n Þ À1 FðX n Þ; X nþ1 ¼ W n À ð2F 0 ðY n Þ À1 À F 0 ðX n Þ À1 ÞFðW n Þ: Per iteration, scheme (5) requires the evaluations of two functions, two Fréchet derivatives and two matrix inversions. Now, to derive the new scheme for solving systems of nonlinear equations, we start by composing scheme (3) to additional Newton step, that is: Now, to reduce number of functional evaluations at each iteration, we will use divided difference approximation to write the derivative f 0 w n ð Þ using some already computed functions from the previous steps. To do that, one can write f 0 y n ð Þ ¼ f y n ð Þ À f x n ð Þ y n À x n ¼ f y n ; x n ½ %f 0 x n ð Þ; in the same manner, we have f 0 y n ð Þ ¼ f y n ð Þ À f w n ð Þ y n À w n ¼ f y n ; w n ½ %f 0 w n ð Þ; by adding (7) and (8), one easily can conclude that Now, if we substitute (9) in (6), then we will have a new scheme for solving nonlinear equations: To generalize scheme (10) to the multidimensional case to solve systems of nonlinear modules, the scheme becomes: W n ¼ X n À F 0 ðY n Þ À1 FðX n Þ; X nþ1 ¼ W n À ð2F 0 ðY n Þ À F 0 ðX n ÞÞ À1 FðW n Þ: Scheme (11) requires at each iteration the evaluation of two functions, two Fréchet derivatives and three matrix inversions. The proposed scheme is of fifth-order of convergence as we will see in the next section. Order of Convergence We will discuss in this section the order of convergence of the proposed schemes (11) and (12). Assume for the next theorems that C j ¼ 1 j! F 0 ðaÞ À1 F j ð Þ a ð Þ, j ! 2, and e n ¼ X n À a. Theorem 1 Let a be the solution of the system F X ð Þ ¼ 0 where F : D R n ! R n be a sufficiently differentiable function on a neighborhood D of a. Suppose that F 0 X ð Þ is continuous and nonsingular in a. If X 0 2 D is an initial approximation which is close enough to a, then the sequence X n f g n!0 obtained by scheme (11) converges to the root a, and the order of convergence equals 5, with asymptotic equation Proof. By using the Taylor expansion of F X n ð Þ we can write F X n ð Þ ¼ F 0 a ð Þ e n þ C 2 e 2 n þ C 3 e 3 n þ À C 3 e 3 n þ C 4 e 4 n þ . . .Þ. Now, we use the following Mathematica code to show the convergence order of scheme (11) for m ¼ 3: In [ À Á e 5 n . By this, we show that scheme (11) is at least of fifth-order of convergence. Now, we want to discuss the order of convergence of the generalized scheme given by (12). Theorem 2 Let a be the solution of the system F X ð Þ ¼ 0 where F : D R n ! R n be a sufficiently differentiable function on a neighborhood D of a. Suppose that F 0 X ð Þ is continuous and nonsingular in a. If X 0 2 D is an initial approximation which is close enough to a, then the sequence X n f g n!0 obtained by scheme (12) converges to the root a, and the order of convergence equals 2m À 1, for any integer m ! 3, with asymptotic equation of the form e kþ1 ¼ 1 Proof. We will use the mathematical induction to prove the convergence order of scheme (12). Firstly, we will prove that scheme (12) is convergent for m ¼ 3, and the convergence order satisfies 2 3 ð Þ À 1 ¼ 5. Note that for m ¼ 3, scheme (12) reduces to scheme (11) which we have been proved that it has the fifth-order of convergence in the previous theorem. Now, to complete the proof using the mathematical induction, suppose that scheme (12) is true and converges for all m r for some positive r > 3 and satisfy the given asymptotic equation. We need to show that the scheme converges for m ¼ r þ 1, and satisfy the given asymptotic equation. To do so, consider the following code of Mathematica: In[2]:= y ¼ e À Series 1 2 ðF 0 ½eÞ À1 F½e; fe; 0; 6g In [5]:= e nþ1 ¼ x½r À ð2F 0 ½y À F 0 ½eÞ −1 F½x½r //FullSimplify Hence, this shows that for m ¼ r þ 1, we have e kþ1 ¼ 1 Computational Efficiency In this section, we compare the efficiency index of our proposed method with other methods in the literature. Commonly in the literature, the efficiency index EI ¼ p 1 d is used, where p is the order of convergence of the iterative scheme, and d is the number of functions needed to be found per iteration in the iterative scheme. Another common index that can be used in the comparison between iterative scheme is the computational efficiency index CEI ¼ p To find the number of functions required to be found per iteration in an iterative scheme, the following rules applied: Any computation of F X ð Þ needs n evaluations of scalar functions. Any computation of the Jacobian F 0 X ð Þ needs n 2 evaluations of scalar functions. Also, the floating points for obtaining the LU factorization are 2 3 n 3 , and to solve the triangular system we need n 2 floating points operations. Finally, n 2 operations required to find a matrix-vector multiplication, and n 3 operations needed to find a matrixmatrix multiplication. We compare the efficiency index and the computational efficiency index for the proposed method (PM) (11) to the following iterative schemes: • The third-order Frontini-Sormani method (FS) [10] given by (4). • The fifth order scheme (CHMT) proposed by Cordero et al. [23], given by • The fifth order scheme (MMK) proposed by Waseem et al. [4], which is given by: • The fifth-order iterative scheme (SG) presented by Sharma et al. [20], which is defined by scheme (5). A comparison of the number of functional evaluations of the selected iterative schemes is illustrated in Tab. 1. Also, the computational efficiency indices of the selected schemes are compared (for n ¼ 2; 3; 4; 5; 10; 20; 50), see Fig. 1. Note that the proposed scheme does not attain the best efficiency in this comparison, especially for small n. We will see in the next two sections that this issue does not affect the scheme negatively when applied to some numerical tests. 2 illustrates the efficiency indices for the selected methods. Note that CHMT, SG and our proposed method have the same efficiency indices. However, this does not guarantee that they have the same behavior, accuracy and computational time cost. Basins of Attraction The concept of basins of attraction is a method to show how different starting points affect the behavior of the function. In this way, we can compare different root-finding schemes depending on the convergence area of the basins of attraction. In this sense, the iterative scheme is better if it has a larger area of convergence. Here, we mean by the area of convergence, the number of convergent points to a root a of f x ð Þ in a selected range. To check the stability and the area of convergence of our proposed method, we select the case m ¼ 3 of scheme (12). We denote the proposed method by PM 5 . For comparison, we compare PM 5 with the following schemes of the same order of convergence: The scheme CHMT given by Cordero et al. (13), the scheme SG proposed by Sharma and Gupta (5), and the scheme MMK presented by Waseem et al. (14). We choose three test examples to visualize the basins of attraction. All examples are polynomials with roots of multiplicity one. The test polynomials are Basins of attraction of PM 5 shows that the proposed method is comparable to other methods of the same order, with an area of convergence which is larger or the same as the areas of convergence of the other methods used in the comparison. Numerical Tests and Applications In this part, we consider some numerical problems to clarify the computational efficiency and convergence behavior of the proposed scheme. All calculations have been performed using 4000 significant digits on Mathematics 9. For comparisons, we find the number of iterations n needed to satisfy the stopping criterion k X n À X nÀ1 k þ k F X n ð Þ k< 10 À150 for each selected method. Also, we use the approximated computational order of convergence for each iterative scheme, which can be found by ACOC % ln k X nþ1 À x n ð Þk= k X n À X nÀ1 ð Þk ð Þ ln k X n À X nÀ1 ð Þk= k X nÀ1 À X nÀ2 ð Þk ð Þ : Finally, we compare for the selected schemes the distance between two consecutive iterations k X n À X nÀ1 k and the value of k F X n ð Þ k for n ¼ 1; 2; 3. To be consistent in the comparison, we compare the proposed scheme PM 5 defined by scheme (12) for m ¼ 3, to the original method which we derived from, that is, FS scheme given by (4). Also, we use the following fifth-order iterative schemes in the comparison: CHMT 5 method defined by scheme (13), MMK method defined by (14), and SG method defined by (5). To test the efficiency of the extension of our proposed scheme to higher orders schemes, we compare the proposed scheme PM 7 of seventh-order given by scheme (12) for m ¼ 4, to the extension of CHMT 5 to the seventh-order scheme CHMT 7 given by: Numerical Tests To be not selective in our examples, we choose most test problems from the same papers which contain the schemes used in the comparisons, see [4,20,23]. Also, we choose two distinct initial guesses for all problems to test the validity and the applicability of iterative schemes. We consider the following test problems and applications: Example 1 Consider the following system of two nonlinear equations: x þ 1 À e y ¼ 0; x þ cos y ð Þ À 2 ¼ 0; & with initial guesses X 0 ¼ f0; 0g t and X 0 ¼ f2; 2g t . The exact solution of this problem is a ¼ f1:3401918575555883401 . . ., 0:8502329164169513268 . . . g t . Example 3 Consider the following system Assume the following partitioning for the interval 0; 1 ½ : is the step size, m is the system size. Let y i ¼ y u i ð Þ for i ¼ 0; 1; 2; . . . ; m þ 1. We use the finite difference method to solve the problem, in which the second derivative y 00 will be replaced by the central difference y 00 % y iÀ1 À 2y i þ y iþ1 h 2 , i ¼ 1; 2; . . . ; m. By this, we obtain m Â m system given by: . . . ; m: We solve this system for m ¼ 10 by selecting X 0 ¼ À1; À1; . . . ; À1 f g t zfflfflfflfflfflfflfflfflfflfflfflfflfflffl ffl}\|fflfflfflfflfflfflfflfflfflfflfflfflfflffl ffl{ 10Àtimes , and X 0 ¼ Method X 0 n k X 1 À X 0 k k X 2 À X 1 k k X 3 À X 2 k k F X 1 ð Þ k k F X 2 ð Þ k k F X 3 ð Þ k ACOC Method X 0 n k X 1 À X 0 k k X 2 À X 1 k k X 3 À X 2 k k F X 1 ð Þ k k F X 2 ð Þ k k F X 3 ð Þ k ACOC Method X 0 n k X 1 À X 0 k k X 2 À X 1 k k X 3 À X 2 k k F X 1 ð Þ k k F X 2 ð Þ k k F X 3 ð Þ k ACOC Applications To check the applicability of the proposed scheme on real-life problems, we apply it on the mixed Hammerstein integral equation and Burgers' equation. Comparisons in Tab. 7 show that the proposed schemes have a good functioning and comparable to the other iterative methods of the same order. Problem 2 Consider the following Burgers' equation selected from [25]: We use discretization to solve this problem. Let h ¼ b À a N , and k ¼ T M be the spatial and temporal step sizes respectively, where N and M are numbers of subintervals in x and t directions respectively. Therefore, for this problem we select points x i ; t j À Á from a grid of domain 0; 1 . . . ; M and t j ¼ 0 þ jk, j ¼ 0; 1; . . . ; N . Let w ij be the approximate solution at x i ; t j À Á . By applying the central differences to u x and u xx , and the backward difference to u t , we get the following nonlinear system: If we let r ¼ Dk h 2 , and z i ¼ w ij , then we are trying to solve the following system of equations: For the unknowns z 1 ; z 2 ; . . . ; z M . The last term w i;jÀ1 is known from the previous time step. The first and the last equations in the system can be replaced by using the given boundary conditions. In our problem, the first and the last equations in the system are We find the approximate solution of the problem using the proposed method PM 5 at x ¼ 0:5 and t ¼ 0:2. To check the effect of the temporal step sizes on the solution, we select different values for k, which means a different number of steps to reach the wanted time. Consider for our problem that the diffusion coefficient D ¼ 0:05, a ¼ 5, and b ¼ 4. The exact solution of the given Burgers' equation is given by: The exact solution for this problem is u 0:5; 0:2 ð Þ¼0:2277071734 . . .. Based on spatial step size equals 1 10 , we choose X 0 ¼ 0:6; 0:6; Á Á Á ; 0:6 f g zfflfflfflfflfflfflfflfflfflfflfflfflfflffl}\|fflfflfflfflfflfflfflfflfflfflfflfflfflffl{ 10Àtimes . Tab. 8 illustrates the numerical results of this problem. The effect of the selected k is clear. The results become better whenever we have a smaller temporal step size. Based on that, we will compare our proposed schemes to the other schemes for h ¼ 0:1 and k ¼ 0:01. We compare the approximate solutions at x ¼ 0:5 and t ¼ 0:2 for n ¼ 3 and n ¼ 4, that is X 3 and X 4 . Also, we find the norms of the functions F X 3 ð Þ and F X 3 ð Þ. Finally, we find the norm of the difference between the two consecutive iterations X 4 À X 3 for each selected method. Comparisons results are shown in Tab. 9. It is clear that the proposed schemes perform in a good way, and in general, give results which are better than the other selected schemes. Conclusion In this study, we have proposed an iterative scheme for systems of nonlinear equations of fifth-order of convergence. We have improved the proposed scheme to a generalized scheme of arbitrary odd order. The proposed method is based on Frontini-Sormani iterative method and developed using additional step with the usage of first derivative approximation. The software Mathematica has been used to show the order of convergence of the proposed method. Different comparisons were used to compare our proposed scheme to the other schemes of the same order, including the efficiency index, computational efficiency index, basins of attractions and several numerical problems. Comparisons show that the efficiency index and the computational efficiency index need not be proper tools for the efficiency of the iterative scheme. As an application, we test the proposed method on the mixed Hammerstein integral equation and Burgers' equation. Comparisons show that the proposed scheme is of excellent performance and overall, it is comparable to the other iterative techniques used in the comparisons regarding the convergence speed, accuracy and the area of convergence in the basins of attraction. Funding Statement: We are grateful for the financial support from UKM's research Grant GUP-2019-033. Conflicts of Interest: The authors declare that they have no conflicts of interest to report regarding the present study.	5,290.6	2021-01-01T00:00:00.000	[ "Mathematics" ]
Possibilities of Integrated Fabrication of Insulation Systems in Electric Drives by Injection Molding of Thermosets Due to the increasing demand for electro mobility and specifically for electrified vehicles, the demand for electric drive technology is expanding significantly with changing requirements in terms of the process and the application. The electrical insulation system of the stator is an essential part of the fabrication process with a high impact on the application properties. Due to limitations—for example, in terms of suitable materials for the stator insulation—a new technology of integrated fabrication by injection molding of thermosets has been founded. In this study, two epoxy (EP) types with different fillers were investigated to prove their suitability in terms of the material properties in the fabrication process and the application. A general realization of the integrated fabrication of insulation systems in electrical engineering by injection molding was proved. Further, the differences regarding the suitability of the two materials are portrayed. It was demonstrated that mainly the filler material influences the fabrication process and the properties in the application, leading to differing suitability in terms of the EP 3162 EMG within the fabrication process and in terms of XW 6640-1 within the application properties of the thermal conductivity and the thermal linear expansion. It was further shown that the filler within the material system is required to increase the thermal conductivity needed for the application. The inclusion of the filler influences the reaction kinetics and the viscosity behavior. A fabrication of the material with fillers is however still possible. Introduction The increasing demand for electro mobility especially in terms of electrified vehicles and powertrains is expanding the development of electric drive technology significantly. Further, the requirements not only in terms of the product but also in terms of the fabrication process is changing. To reduce the manufacturing costs and achieve economic efficiency, a high degree of automation in the production with low rejection rates is required to face the increasing demand. Concerning the product, a high power density and optimum efficiency of the electric drive unit must be reached while realizing low weight and small installation space [1]. The stator insulation system determines the power of the motor with respect to the entire life cycle of the application due to thermal and electrical restrictions [2]. Therefore, the electrical insulation system reveals a crucial position within the development of electro mobility applications. The classic stator insulation system consists of two groups of insulation-the primary insulation covering the conductors and the secondary insulation in the slot and of the full stator by impregnation. The insulation of the conductors provides the basic electrical insulation to avoid electrical flashovers between wires, which would lead to a reduction in performance. Common insulation materials of the conductors are varnishes consisting of polyamide-imide (PAI) or foils of polyimide (PI). The foils are wrapped several times system with up to 85 mass% of fillers. These fillers allow a wide range of modifications of the material behavior and the properties achieved in the application. For example, the thermal conductivity of thermosets can be increased from 0.2 W m·K to over 2.5 W m·K [8]. The basic requirement for using thermoset injection molding for the insulation of stators is the complete impregnation of the electrical conductors by the thermoset. Impregnation is defined as the flow of a liquid through a porous medium [9]. The porous medium in the stator is covered by the windings and the slots between the conductors. With reference to [10], in the two-dimensional perspective the flow path of the liquid and thus, the impregnability is proportional to the viscosity of the liquid, the permeability of the porous medium, the impregnation time, and the processing pressure. These impact factors can be adaptable to the injection molding process with the exception of the viscosity of the liquid, since a thermoset molding compound is not a Newtonian fluid, as assumed by [10]. The complete impregnation of conductor alike structures with highly filled thermosets in the injection molding process was shown on carbon rovings by [11]. It was demonstrated that several process parameters directly influence the impregnation quality. An increasing tool temperature, for example, reduces the viscosity leading to an improvement in the impregnation and realizing complete insulation, analogous to [12]. The general use of the injection molding process to fabricate a stator has already been realized to some extent [12,13]. However, only small dimensions of the stator with short flow paths or stator segments have been implemented so far. The aim of this paper is to investigate the suitability of epoxy-based resins for the application of insulation of stators by injection molding. The main material requirements are summarized in Table 1. The target values of these parameters are based on the standard process conditions as well as the application requirements. For example, the low viscosity is a precondition for long flow paths as this allows not only full insulation of the wires due to a high fluidity, but further a longer time with that flow path in the presence of curing parameters. The requirement for the reaction kinetics is rather complex and is therefore here not further defined. The evaluation of the reaction kinetics is presented within Section 3. The high thermal conductivity is needed with respect to the application in stator systems, as temperature rise during the usage of the stator must be dissipated easily to reduce the thermal input in the system and increase the durability. The low thermal linear expansion is based on the stator assembly, which is based on different materials such as the copper in the wire or the metal sheets in the statorette. By reaching a low thermal linear expansion, higher eligibility between the different materials is given, which increases the lifetime again. The target values in terms of the average of the partial discharge and the partial discharge inception/extinction voltage go along with the application and indicate sufficient insulation of the wires. The paper investigated two commonly used epoxy resins and reveals their general material properties as well as the fabrication of stator segments with a long flow path, to evaluate their suitability in terms of the fabrication process and the application. Material The experiments were conducted using two commercial types of epoxy resin (EP)-the type EP 3162 EMG (Raschig GmbH, Ludwigshafen, Germany) and the type XW 6640-1 (Duresco GmbH, Witterswil, Swizerland), which differ in terms of the anorganic filler material, but are both based on bisphenol-A. Both materials are a premixed grey-black granulate with resin, hardener, catalyst, and some carbon black pigments. The exact composition of the mixture including the filler type and grade is a business secret of Raschig GmbH or Duresco GmbH and therefore confidential. Characterization of the pure filler systems is unfortunately not possible due to the business secret. Table 2 reveals the important properties of the used material based on our own measurements. Further characterization of the material is revealed in the datasheets of the materials [14,15]. To define the properties needed for fabrication of the materials, a wide characterization of the materials was conducted in this study. Figure 1 depicts the IR spectrum of the ATR method of the two types of epoxy resin with the main difference in the bands highlighted. The different bands are related to possible elements according to [16]. As it can be seen from the IR-spectrum there is a significant difference in the components and the proportion of resin and hardener in the two epoxy types with respect to the different bands. However, the IR spectrum is similar for the range of wavenumbers between 2.500 and 4.000 cm −1 . The difference in bands of the two epoxy types is also a clue with respect to the filler type chosen. Taking the NIST Standard reference database into account, the characteristic peak (position 3) of EP 3162 EMG suggests the use of boron nitride, whereas the characteristic peak (position 6) of XW 6640-1 depicts the use of aluminum oxide. Both filler types are chosen mainly with respect to their high thermal conductivity, which is one of the demands in terms of the utilization of resins within the integrated fabrication of insulation systems in electrical engineering. Further, the filler in EP 3162 EMG has a high electric insulation property, which serves another demand in terms of the application. While the filler itself could not be characterized due to the business secret, the material behavior was compared to a pure epoxy resin type EP 3681 E (Raschig GmbH, Ludwigshafen, Germany), which is based on the same resin, hardener, and catalyst without fillers. The density of this type is 1.225 g/cm 3 . Fabrication of the Test Specimens The test samples were produced pressure controlled by a Krauss Maffei KM 80-380 CX DUR/03 injection molding machine (KraussMaffei Group, Munich, Germany) with a screw diameter of 30 mm. Two types of test samples were produced; in terms of the main material characterization, plates with the dimension of 60 × 60 × 2 [mm 3 ] were fabricated in a dual cavity to prepare the test samples needed for the different characterizations. Further, test samples-so-called single slot samples-were produced to evaluate the suitability of the material with respect to the application. The assembly of the tool is shown in Figure 2, where one stack of metal sheets together with two wires are inserted into the tool before the injection molding process starts. The cavity is sealed on the opposite side of the gate using silicone pads to ensure accessible terminals. Both filler types are chosen mainly with respect to their high thermal conductivity, which is one of the demands in terms of the utilization of resins within the integrated fabrication of insulation systems in electrical engineering. Further, the filler in EP 3162 EMG has a high electric insulation property, which serves another demand in terms of the application. While the filler itself could not be characterized due to the business secret, the material behavior was compared to a pure epoxy resin type EP 3681 E (Raschig GmbH, Ludwigshafen, Germany), which is based on the same resin, hardener, and catalyst without fillers. The density of this type is 1.225 g/cm 3 . Fabrication of the Test Specimens The test samples were produced pressure controlled by a Krauss Maffei KM 80-380 CX DUR/03 injection molding machine (KraussMaffei Group, Munich, Germany) with a screw diameter of 30 mm. Two types of test samples were produced; in terms of the main material characterization, plates with the dimension of 60 × 60 × 2 [mm 3 ] were fabricated The processing parameters were set as shown in Table 3. Due to different test sample volumes, the processing parameters for the two types of samples differ slightly in terms of the mold temperature and the holding pressure. However, the main parameters were kept constant for both samples. 15 15 in a dual cavity to prepare the test samples needed for the different characterizations. Further, test samples-so-called single slot samples-were produced to evaluate the suitability of the material with respect to the application. The assembly of the tool is shown in Figure 2, where one stack of metal sheets together with two wires are inserted into the tool before the injection molding process starts. The cavity is sealed on the opposite side of the gate using silicone pads to ensure accessible terminals. The processing parameters were set as shown in Table 3. Due to different test sample volumes, the processing parameters for the two types of samples differ slightly in terms of the mold temperature and the holding pressure. However, the main parameters were kept constant for both samples. Characterization To evaluate the suitability of the material in terms of the demand of the application the material was characterized with respect to the injection molding process, the fabrication in the test sample, and the resulting properties in the application. For each characterization a material recommendation is given respectively. Characterization To evaluate the suitability of the material in terms of the demand of the application the material was characterized with respect to the injection molding process, the fabrication in the test sample, and the resulting properties in the application. For each characterization a material recommendation is given respectively. Specific Heat Capacity c According to ISO 11357-4 To evaluate the material impact, specifically due to the fillers, on the temperature control and the flow and curing process, the specific heat capacity c was determined at 25 • C using the C80 calorimeter (type: 3D-Calvet calorimeter; TA Instruments, New Castle, DE, USA). Although the specific heat capacity c is temperature dependent, the measured parameters are assumed to be valid in terms of evaluating the impact of the material on the curing. Thermal Conductivity a According to DIN EN 821 To analyze the thermal conductivity a in terms of the process as well as the application properties, plate-like samples of the dimension 12.7 × 12.7 [mm 2 ] were prepared from the middle of the test samples of the plate. The measurements were carried out for the temperature sets of 23, 80, 120, and 160 • C, to consider the temperature dependence within the analysis. Three samples per temperature set were examined using a Nanofash (type: LFA 447; Netzsch-Group, Selb, Germany). Thermal Linear Expansion ∆L According to ISO 7991 The thermal linear expansion ∆L was determined after the preparation of a sample with the dimension of 4 × 4 [mm 2 ] from the test sample of the plate, using a thermomechanical analyzer (type: TMA 450; TA instruments, New Castle, DE, USA). The analysis was set between −20 and 200 • C with a heating rate of 3 • C per minute. The thermal linear expansion ∆L has mainly an impact on the application properties. To investigate the temperature dependent reaction kinetics in terms of the process conditions of the material, a differential scanning calorimetry (DSC Q100; TA Instruments, New Castle, DE, USA) was used. Samples of about 5 mg were placed in DSC aluminum pans and heated with a constant rate of 10 • C per minute from 0 • C to 240 • C. The experiments were conducted in a nitrogen atmosphere with a flow rate of 50 mL per minute. To characterize the curing process, the specific enthalpy ∆H ges; 1 and the peak temperature T peak were determined. Further, the reaction turnover α was calculated considering Equation (1) where ∆H j is the specific enthalpy at the temperature T j and ∆H ges;1 is the total specific enthalpy in the first heating cycle [7]. 2.3.5. Determination of the Viscosity Using a Rotational Viscometer According to DIN EN 6043 To characterize the viscosity, as one of the most important processing conditions in terms of thermosets, a rotational viscometer (Discovery Hybrid Rheometer 2; TA Instruments, New Castle, DE, USA) was used with DIN EN 6043. The viscosity was determined with respect to increasing temperature (dynamic behavior) and to time dependence (isothermal). The assembly was based on two plates with a shearing load rotating with a constant frequency of 1 Hz. In the case of the dynamic measurements, the temperature range was set between 90 • C and 200 • C with a constant heat rate of 5 • C per minute. The minimum of the viscosity η min and the corresponding temperature Tη min were analyzed. The isothermal measurements started at a certain temperature, which was held constant, and the change of the viscosity depending on the time was determined. This isothermal plateau was set at a temperature of η min , which was 120 • C in the case of EP 3162 EMG and 110 • C for XW 6640-1, and was further increased in steps of 20 • C up to 180 • C ongoing from 120 • C. The route of the viscosity η relative to the time with respect to a constant temperature follows an s slope. The time t pot between the beginning of the calculation and the turning point-so-called pot life-was analyzed. Microscopy To analyze the position of the wire within the stack and the insulation of the wire due to the injection molding process, small strips of the single slot sample were removed using a water-cooled saw with minimal temperature input. The strips were taken from positions near and far away from the gate, to evaluate the change of the position of the wires and the polymer along the flow path. The strip samples were embedded in cold-curing epoxy resin (type: Epofix; Struers GmbH, Ottensoos, Germany) and polished. Afterwards, the samples were characterized by a stereo microscope (type: Axio Zoom.V16; Carl Zeiss AG, Oberkochen, Germany) to create an overview image and further characterized by a reflected light microscope (type: Axio Imager.M2; Carl Zeiss AG, Oberkochen, Germany) to create images in more detail. Average of Partial Discharge and Partial Discharge Inception/Extinction Voltage The evaluation of the partial discharge (PD) and the partial discharge inception (PDIV) as well as the extinction voltage (PDEV) were realized using a test equipment with two parallel electrodes, which is defined by IEC 60243-1, and a measuring system within a of the partial discharge level, the partial discharge inception, and the extinction voltage were defined using the testing profile as shown in Figure 3. The testing time for the partial discharge inception voltage (PDIV) was defined with 15 s, and for the partial discharge level (PD) for 30 s. Further, the voltage incline was 100 V per second. The experiments were held at a room temperature of 20 • C with a humidity of 50%. a Faraday cage according to IEC 60270 (type: Omicron MPD 600; Omicron elect GmbH, Klaus, Austria). The test samples were provided by the material supplier an the dimension of 150 × 150 × 4 [mm 2 ] in terms of EP 3162 EMG as well as 100 × 1 [mm 3 ] for XW 6640-1. As the thickness of the plates is the only dimension, which im the measurement, the different cross sections of the plates in terms of the two ma do not affect the testing. After clamping the test samples within the test setting, the of the partial discharge level, the partial discharge inception, and the extinction v were defined using the testing profile as shown in Figure 3. The testing time for the discharge inception voltage (PDIV) was defined with 15 s, and for the partial disc level (PD) for 30 s. Further, the voltage incline was 100 V per second. The experi were held at a room temperature of 20 °C with a humidity of 50%. Specific Heat Capacity c According to ISO 11357-4 The specific heat capacity c reaches 0.997 J g· • C in terms of the material EP 3162 EMG and 0.899 J g· • C in terms of XW 6640-1. For the pure epoxy resin c reaches 1.616 J g· • C , which shows that the fillers implemented in the resin reduce the specific heat capacity c. As c should reach a high value with respect to the application, as shown in Table 1, EP 3162 EMG reveals slightly higher suitability in the application, although the difference between the two materials is less. Thermal Conductivity a According to DIN EN 821 The thermal conductivity a, as shown in Figure 4, is reduced with increasing temperature T independent of the material. Further, the standard deviation is higher in terms of the temperature of 23 • C for both materials compared to the other temperature sets. The material XW 6640-1 reaches about 40% higher values in terms of the low temperature sets. The difference in the values between the two materials is reduced to 25% higher values for higher temperatures. Nevertheless, the material XW 6640-1 reaches higher values for a and reveals therefore higher suitability in terms of Table 1. It can be also seen that the fillers in both materials increase a compared to the pure epoxy resin, which reveals a significant lower value for a at a temperature of 23 • C. terms of the temperature of 23 °C for both materials compared to the other temperature sets. The material XW 6640-1 reaches about 40% higher values in terms of the low temperature sets. The difference in the values between the two materials is reduced to 25% higher values for higher temperatures. Nevertheless, the material XW 6640-1 reaches higher values for a and reveals therefore higher suitability in terms of Table 1. It can be also seen that the fillers in both materials increase a compared to the pure epoxy resin, which reveals a significant lower value for a at a temperature of 23 °C. Thermal Linear Expansion ΔL According to ISO 7991 The thermal linear expansion ΔL increases between −20 °C and 200 °C for both materials with different gradients. For low temperatures up to 60 °C, the linear route of ΔL reveals a higher gradient in terms of EP 3162 EMG, which starts at a lower value at −20 °C compared to XW 6640-1. Above 60 °C, the thermal linear expansion ΔL increases much faster in terms of EP 3162 EMG, which leads to inhomogeneous behavior over the whole temperature range. The route of the thermal linear expansion ΔL compared for both materials can be seen in Figure 5. With respect to the material requirements and Table 1, XW 6640-1 complies with the demand with a higher amount due to the lower ΔL and a more homogeneous route compared to EP 3162 EMG. Thermal Linear Expansion ∆L According to ISO 7991 The thermal linear expansion ∆L increases between −20 • C and 200 • C for both materials with different gradients. For low temperatures up to 60 • C, the linear route of ∆L reveals a higher gradient in terms of EP 3162 EMG, which starts at a lower value at −20 • C compared to XW 6640-1. Above 60 • C, the thermal linear expansion ∆L increases much faster in terms of EP 3162 EMG, which leads to inhomogeneous behavior over the whole temperature range. The route of the thermal linear expansion ∆L compared for both materials can be seen in Figure 5. With respect to the material requirements and Table 1, XW 6640-1 complies with the demand with a higher amount due to the lower ∆L and a more homogeneous route compared to EP 3162 EMG. In comparison to the thermoset material, the stack metal sheets reveal a thermal linear expansion ΔL of 28 µ at 150 °C with only a small range over the temperature range. Differential Scanning Calorimetry (DSC) According to ISO 11357 The route of the DSC measurements as well as the specific enthalpy ΔHges;1 and the peak temperature Tpeak (A) together with the reaction turnover α (B) are shown in Figure Figure 5. Thermal linear expansion in the temperature range of -20 up to 200 • C compared for two EP types. In comparison to the thermoset material, the stack metal sheets reveal a thermal linear expansion ∆L of 28 µm m at 150 • C with only a small range over the temperature range. Differential Scanning Calorimetry (DSC) According to ISO 11357 The route of the DSC measurements as well as the specific enthalpy ∆H ges;1 and the peak temperature T peak (A) together with the reaction turnover α (B) are shown in Figure 6 in comparison for the two different EP types of EP 3162 EMG and XW 6640-1 as well as the pure epoxy resin. The reaction kinetics of the two materials with fillers reveal different values in terms of the specific enthalpy ∆H ges;1 with only 50% needed in terms of XW 6640-1 compared to EP 3162 EMG, but similar behavior regarding the reaction turnover α and the peak temperature T peak . The reaction turnover α illustrates an s slope as expected. The level of the specific enthalpy ∆H ges;1 and the peak temperature T peak is mainly influenced by the material in terms of the different heat capacity c. With that, XW 6640-1 requires less applied heat in the curing process. As the difference between the two materials within the reaction kinetics is only present in terms of the level of ∆H g es;1 , a similar behavior during the fabrication regarding the process parameters is likely. However, EP 3162 EMG depicts a higher standard deviation, leading to more unstable process conditions. Compared to the pure epoxy resin, the specific enthalpy ∆H g es;1 is significantly reduced by the fillers as they do not participate within the hardening process. As the thermal conductivity a is increased by the filler systems, the reaction turnover α is shifted to lower temperatures in terms of the filler systems compared to the pure epoxy resin. With that, the presence of the fillers in the two material systems reduces the specific enthalpy ∆H ges;1 needed for the reaction and changes the temperature of the curing with respect to the reaction turnover α. Polymers 2022, 14, 5352 11 of 17 Figure 6. Route of DSC measurements with specific enthalpy ΔHges;1 and peak temperature Tpeak (A) as well as reaction turnover α (B) compared for two EP types and the pure epoxy resin. Determination of the Viscosity Using a Rotational Viscometer According to DIN EN 6043 The route of the complex viscosity as dynamic behavior is shown in Figure 7A for the two different EP types. EP 3162 EMG reveals about half a decade lower viscosity relative to XW 6640-1 with a slightly higher temperature of the minimum of viscosity. After reaching the minimum of the viscosity, EP 3162 EMG increases over a small range of temperature and gains much higher values compared to XW 6640-1. Compared to the pure resin the minimum of the viscosity is increased in terms of the filler systems by at least three decades. Figure 7B further shows the pot life tpot for both EP types, where XW 6640-1 reaches higher values compared to EP 6640-1. With respect to the pure epoxy resin the pot life tpot is only increased for low temperatures. With increased temperature the difference between the material systems with and without filler is reduced. Overall, this Figure 6. Route of DSC measurements with specific enthalpy ∆H ges;1 and peak temperature T peak (A) as well as reaction turnover α (B) compared for two EP types and the pure epoxy resin. Determination of the Viscosity Using a Rotational Viscometer According to DIN EN 6043 The route of the complex viscosity as dynamic behavior is shown in Figure 7A for the two different EP types. EP 3162 EMG reveals about half a decade lower viscosity relative to XW 6640-1 with a slightly higher temperature of the minimum of viscosity. After reaching the minimum of the viscosity, EP 3162 EMG increases over a small range of temperature and gains much higher values compared to XW 6640-1. Compared to the pure resin the minimum of the viscosity is increased in terms of the filler systems by at least three decades. Figure 7B further shows the pot life t pot for both EP types, where XW 6640-1 reaches higher values compared to EP 6640-1. With respect to the pure epoxy resin the pot life t pot is only increased for low temperatures. With increased temperature the difference between the material systems with and without filler is reduced. Overall, this results in a lower sensitivity of XW 6640-1 regarding the impact of time and temperature during the curing. As the difference in the values of t pot between the two materials is low, the significantly lower viscosity in terms of EP 3162 EMG is the crucial factor to evaluate the viscosity behavior of the two EP types. Therefore, EP 3162 EMG is defined as more suitable in terms of the flow conditions, as the low viscosity is more likely to realize long flow paths. The higher pot life t pot makes the material system more sensitive in terms of the impact of time and temperature. However, it can be used in terms of low fabrication times leading to cost reduction in the production process. The high increase of EP 3162 EMG after reaching the minimum of the viscosity refers to fast curing after reaching a certain temperature set-up. This can be used in terms of low fabrication times, however it has to be taken into account that this might also hinder the achievement of long flow paths. The low viscosity enhances the possibility of long flow paths, but fast curing reduces them. Microscopy The assembly of the single slot sample is shown in Figure 8 with the copper wire itself (A), the copper wire inserted into the stack metal sheets (B), which functions as the inserts in the injection molding process, and the sample after fabrication (C). The sample reveals full filling of the cavity and the possibility of sealing through silicon pads, as the material load on the copper wire stops at a defined position. Microscopy The assembly of the single slot sample is shown in Figure 8 with the copper wire itself (A), the copper wire inserted into the stack metal sheets (B), which functions as the inserts in the injection molding process, and the sample after fabrication (C). The sample reveals full filling of the cavity and the possibility of sealing through silicon pads, as the material load on the copper wire stops at a defined position. Microscopy The assembly of the single slot sample is shown in Figure 8 with the copper wire itself (A), the copper wire inserted into the stack metal sheets (B), which functions as the inserts in the injection molding process, and the sample after fabrication (C). The sample reveals full filling of the cavity and the possibility of sealing through silicon pads, as the material load on the copper wire stops at a defined position. Further, Figure 9 depicts the position of the conductor in the cavity near the gate (A) and far away from the gate at the end of the flow path (B). It can be seen that the position of the wires changes along the flow path leading to a replacement of the conductor on the opposite side of the gating system. While the straight wire terminals are clamped near position B, the curved ends near the gating (A) remain unfixed. This change in the position might occur due to the forces of the injection process. It can be assumed that the optimization of the process parameter can reduce this movement. At this moment, full insulation of the conductor is yet not given at the end of the flow path. Further, Figure 9 depicts the position of the conductor in the cavity near the gate (A) and far away from the gate at the end of the flow path (B). It can be seen that the position of the wires changes along the flow path leading to a replacement of the conductor on the opposite side of the gating system. While the straight wire terminals are clamped near position B, the curved ends near the gating (A) remain unfixed. This change in the position might occur due to the forces of the injection process. It can be assumed that the optimization of the process parameter can reduce this movement. At this moment, full insulation of the conductor is yet not given at the end of the flow path. The details-shown in Figure 10-reveal, that general insulation in between the two wires is possible. However, the amount of material in the gap is reduced along the flow path. With that, a general insulation of the conductor using the injection molding process is possible, but further improvement has to be made in terms of the length of the flow path to ensure proper insulation along the flow path length. So far, the two EP types did not reveal a difference in the realization of the insulation and the behavior along the flow path. The details-shown in Figure 10-reveal, that general insulation in between the two wires is possible. However, the amount of material in the gap is reduced along the flow path. With that, a general insulation of the conductor using the injection molding process is possible, but further improvement has to be made in terms of the length of the flow path to ensure proper insulation along the flow path length. So far, the two EP types did not reveal a difference in the realization of the insulation and the behavior along the flow path. Average Level of Partial Discharge and Partial Discharge Inception/Extinction Voltage The electric strength of thermosets is crucial to ensure the suitability of the electrical machines during the application. To evaluate the qualification of the two EP types, the average level of partial discharge (PD) (A), the partial discharge inception voltage (PDIV) (B), and the partial discharge extinction voltage (PDEV) (C) are shown in Figure 11. The partial discharges should reach a low value to prevent electrical breakdowns due to degradation and to ensure the durability of the insulating material over the lifetime. With that, EP 3162 EMG has a slightly lower level of PD. By defining a margin of safety according to IEC60034-18-41 high PDIV and PDEV values ensure a safe operation of the electric machines without crucial PDs. In accordance with the PD-level the EP 3162 EMG has a higher value of PDIV and PDEV compared to XW 6640-1. It can be assumed, that the difference in the partial discharge resistance goes along with the change of the components within the two EP types which can have a decisive influence in electrical treeing inside the thermoset. Those two EPs probably do not only differ in terms of the filler type but further in terms of the amount of filler. Average Level of Partial Discharge and Partial Discharge Inception/Extinction Voltage The electric strength of thermosets is crucial to ensure the suitability of the electrical machines during the application. To evaluate the qualification of the two EP types, the average level of partial discharge (PD) (A), the partial discharge inception voltage (PDIV) (B), and the partial discharge extinction voltage (PDEV) (C) are shown in Figure 11. The partial discharges should reach a low value to prevent electrical breakdowns due to degradation and to ensure the durability of the insulating material over the lifetime. With that, EP 3162 EMG has a slightly lower level of PD. By defining a margin of safety according to IEC60034-18-41 high PDIV and PDEV values ensure a safe operation of the electric machines without crucial PDs. In accordance with the PD-level the EP 3162 EMG has a higher value of PDIV and PDEV compared to XW 6640-1. It can be assumed, that the difference in the partial discharge resistance goes along with the change of the components within the two EP types which can have a decisive influence in electrical treeing inside the thermoset. Those two EPs probably do not only differ in terms of the filler type but further in terms of the amount of filler. Figure 11. Average of partial discharge (PD) (A) and the partial discharge inception (PDIV) (B) as well as extinction voltage (PDEV) (C) compared for two EP types. Conclusions Based on the investigations of this paper, the suitability of epoxy based resins for the application of insulation of stators by injection molding could be determined. The main material requirements according to Table 1 can be divided into impact factors of the fabrication process and the application itself. It was shown that a general realization of insulation of stators by injection molding of epoxy based resins is possible. However, the EP type and regarding that, mainly the filler material influences the fabrication process and the properties in the application. In terms of the investigation, EP 3162 EMG reveals higher suitability in terms of the fabrication process and-in more detail-in terms of the low viscosity and the possibility of long flow paths, so far. However, the impact of the fast-curing process has to be taken into account, which might hinder long flow paths relative to the chosen process parameters. Further, EP 3162 EMG reveals advantages in the application due to a higher partial discharge resistance. However, this EP type has a deficiency regarding the thermal conductivity and the thermal linear expansion. The high suitability of XW 6640-1 in terms of the thermal conductivity and the thermal linear expansion could further be used by adopting the process. Further investigations are to be held in terms of the process parameters and the local definition of the copper wires position in the slot. It is assumed, that a general improvement of the insulation of the wires increases the suitability of XW 6640-1 in terms of the application. Further changes in the Figure 11. Average of partial discharge (PD) (A) and the partial discharge inception (PDIV) (B) as well as extinction voltage (PDEV) (C) compared for two EP types. Conclusions Based on the investigations of this paper, the suitability of epoxy based resins for the application of insulation of stators by injection molding could be determined. The main material requirements according to Table 1 can be divided into impact factors of the fabrication process and the application itself. It was shown that a general realization of insulation of stators by injection molding of epoxy based resins is possible. However, the EP type and regarding that, mainly the filler material influences the fabrication process and the properties in the application. In terms of the investigation, EP 3162 EMG reveals higher suitability in terms of the fabrication process and-in more detail-in terms of the low viscosity and the possibility of long flow paths, so far. However, the impact of the fast-curing process has to be taken into account, which might hinder long flow paths relative to the chosen process parameters. Further, EP 3162 EMG reveals advantages in the application due to a higher partial discharge resistance. However, this EP type has a deficiency regarding the thermal conductivity and the thermal linear expansion. The high suitability of XW 6640-1 in terms of the thermal conductivity and the thermal linear expansion could further be used by adopting the process. Further investigations are to be held in terms of the process parameters and the local definition of the copper wires position in the slot. It is assumed, that a general improvement of the insulation of the wires increases the suitability of XW 6640-1 in terms of the application. Further changes in the properties can be realized by different filler systems, which can-for example-even increase the thermal conductivity or the heat capacity in terms of EP 3162 EMG. The comparison of the two EP types regarding the material properties in terms of the fabrication process and the application are shown in Table 4 and reveal an intermediate result. It was also shown that the inclusion of fillers within the epoxy resin is needed to improve the thermal conductivity a. This inclusion of fillers within the material systems has an impact on the reaction kinetics and the viscosity behavior as shown. Table 4. Evaluation of the suitability of two EP types in terms of the main material requirements regarding the fabrication process and the application for the usage of thermosets to impregnate stators by injection molding [x: higher suitability; -: lower suitability]. Material Property Qualification of the Material Further investigations plan to determine the influence of the process parameters mainly in terms of insulation of the conductor and the reduction of the change of the wire position along the flow path. Moreover, a method of defined positioning of the conductor in the slot during the fabrication process needs to be found, to use the advantages of XW 6640-1 in the application. In addition, the influence of the filler system on the process conditions and the application properties will be analyzed by integrating a defined amount of fillers in pure EP material systems.	9,277.8	2022-12-01T00:00:00.000	[ "Materials Science" ]
Deciphering the Bacterial Microbiome of Citrus Plants in Response to ‘Candidatus Liberibacter asiaticus’-Infection and Antibiotic Treatments The bacterial microbiomes of citrus plants were characterized in response to ‘Candidatus Liberibacter asiaticus’ (Las)-infection and treatments with ampicillin (Amp) and gentamicin (Gm) by Phylochip-based metagenomics. The results revealed that 7,407 of over 50,000 known Operational Taxonomic Units (OTUs) in 53 phyla were detected in citrus leaf midribs using the PhyloChip™ G3 array, of which five phyla were dominant, Proteobacteria (38.7%), Firmicutes (29.0%), Actinobacteria (16.1%), Bacteroidetes (6.2%) and Cyanobacteria (2.3%). The OTU62806, representing ‘Candidatus Liberibacter’, was present with a high titer in the plants graft-inoculated with Las-infected scions treated with Gm at 100 mg/L and in the water-treated control (CK1). However, the Las bacterium was not detected in the plants graft-inoculated with Las-infected scions treated with Amp at 1.0 g/L or in plants graft-inoculated with Las-free scions (CK2). The PhyloChip array demonstrated that more OTUs, at a higher abundance, were detected in the Gm-treated plants than in the other treatment and the controls. Pairwise comparisons indicated that 23 OTUs from the Achromobacter spp. and 12 OTUs from the Methylobacterium spp. were more abundant in CK2 and CK1, respectively. Ten abundant OTUs from the Stenotrophomonas spp. were detected only in the Amp-treatment. These results provide new insights into microbial communities that may be associated with the progression of citrus huanglongbing (HLB) and the potential effects of antibiotics on the disease and microbial ecology. Introduction Huanglongbing (HLB), the most devastating citrus disease worldwide, is vectored by phloem-feeding insects and caused by fastidious bacterial pathogens (Candidatus Liberibacter) [1,2,3]. The three species of the pathogen have been identified by their 16S rRNA sequences: Candidatus Liberibacter asiaticus (Las), the most prevalent species in Asia and America [4,5], Candidatus Liberibacter africanus (Laf) in Africa [6], and Candidatus Liberibacter americanus (Lam) in South America [7]. In the U.S., citrus HLB was first discovered in August of 2005 in South Florida. Currently, HLB exists in all 34 citrus-producing counties in Florida and has caused an estimated $3.63 billion in lost revenues and 6,611 lost jobs by reducing orange juice production [8]. To date, there is no effective strategy to control citrus HLB after it is established [5,9]. Soon after a bacterium was reported to be associated with HLB, antibiotics were first used to control the pathogen. Different types of antibiotics, such as tetracycline and penicillin, were injected into infected citrus trees to temporarily relieve HLB symptoms and decrease Las bacterial titers [10]. Injecting antibiotics was recommended as a part of the integrated management program in India [11]. In our previous studies, different kinds of antibiotics were tested for efficacy against the HLB bacterium while assessing their phytotoxicity to citrus. A combination of a Beta-lactam antibiotic, penicillin, and an aminoglycoside, streptomycin, has been shown to act synergistically against the bacterium and facilitate the aminoglycoside's uptake, which leads to bacterial cell death [12,13]. The microbial community structure in Las-infected field citrus plants treated with the above antibiotic combination has been analyzed. Our previous data detected 7,028 known Operational Taxonomic Units (OTUs) in citrus leaf midribs using the PhyloChip TM G3 array, of which Proteobacteria was constantly the dominant bacterial phylum, with the Alphaproteobacterial and the Gammaproteobacterial classes vying for prevalence. Bacterial cells in close proximity may be able to modify their microenvironment, making the composition of the microbial community an important factor in the ability of Las to cause HLB progression [14]. The microbial diversity associated with citrus HLB in planta has also been reported by other research groups [15,16,17]. Some plant growth-promoting bacteria, such as Bacillus and Burkholderia, were detected in non-infected leaf samples [15], while bacteria such as Methylobacterium and Sphingobacterium were present in root samples from HLB-affected trees [16]. A powerful oligonucleotide microarray of high-density 16S rRNA genes, the PhyloChip microarray, has been developed and effectively used to study bacterial diversity, especially from environmental samples [18]. In this article, we aim to decipher the bacterial microbiome in HLB-affected citrus versus non-infected citrus as well as in citrus plants treated with ampicillin and gentamicin using PhyloChip-based metagenomics. Plant materials and treatments HLB-affected budsticks were sampled from severely HLBaffected field rough lemons (Citrus limonum 'Lemon #76') at the USDA-ARS-USHRL farm in Fort Pierce, FL and tested positive for Las by real-time qPCR. They were soaked in the antibiotic treatments; ampicillin sodium at a concentration of 1.0 g/L (Amp, Sigma-Aldrich, St. Louis, MO) or gentamicin sulfate at a concentration of 100 mg/L (Gm, Sigma-Aldrich, St. Louis, MO) and water as the diseased control (CK 1 ), overnight in a fume hood under ventilation and lighting. Las-free budsticks, which tested negative by qPCR from healthy rough lemons, were also soaked in water as the healthy control (CK 2 ). The budsticks were grafted onto two-year-old healthy grapefruit (Citrus paradisi 'Duncan') rootstocks and covered using plastic tape for three weeks. Each experiment was replicated for three times with 45 scions. To improve scion growth, new flush from the rootstocks was removed after grafting and then allowed to grow. All experimental plants were grown in an insect-proof greenhouse. The first leaf samples from scions (rough lemon) and rootstocks (grapefruit) for DNA extraction were taken four months after inoculation, and second samplings were taken at six month after inoculation. The leaves were washed in tap water and then rinsed three times with sterile water. The midribs of the leaves were excised, frozen in liquid nitrogen, and stored at 280uC. The midribs of five leaves from each sample were pooled, and DNA was isolated for qPCR analysis as described previously [13,19]. The scion growth rate (%) was defined as the number of scions that had newly emerging leaves or flushes out of the total grafted scions. The scion infection rate (%) was defined as the number of Las infected scions with threshold cycle (Ct) values lower than 32.0 out of the total grafted scion number. The Las transmission rate (%) was defined as the number of the grafted plants' rootstocks that tested Las positive by qPCR with Ct values less than 32.0. Data were analyzed by a generalized linear mixed model using the SAS procedure GLIMMIX. Differences among treatment levels were determined with the LINES option of the LSMEANS statement. PCR amplification of 16S rRNA genes DNA for the PhyloChip TM G3 analysis, which was extracted from all scion samples of the same treatment, was pooled in equal amounts and quantified by the PicoGreenH method. PCR amplifications of 16S rRNA genes were conducted as described previously [14]. PhyloChip TM G3 analysis The PhyloChip TM G3 analysis was conducted by Second Genome Inc. (San Bruno, CA). The 16S rRNA amplicons and a mixture of amplicons at known concentrations (spike-mix) were combined, fragmented using DNAse1 (Invitrogen, Carlsbad, CA), and biotin-labeled using the recommended protocol for Affymetrix Prokaryotic Arrays. Labeled products were hybridized with three replicates overnight at 48uC at 60 rpm. The arrays were washed, stained, and scanned as described in Hazen et al. (2010) [18]. Data analyses Preliminary data analyses were performed by Second Genome (San Bruno, CA) as described in Hazen et al. (2010) [18]. Briefly, to calculate the summary intensity for each feature on each array, the central nine pixels of individual features were ranked by intensity and the 75 th percentile was used. Probe intensities were background-subtracted and scaled to the PhyloChip TM Control Mix. Array fluorescence intensity was collected as integer values ranging from 0 to 65,536 (2 16 ). Fluorescent intensities for sets of probes complementing an OTU were averaged after discarding the highest and lowest, and the means were log2 transformed. Thus, they are decimal numbers ranging from 0 to 16. For compatibility with some statistical operations, the HybScore was multiplied by 1,000 and then rounded to the nearest integer allowing a range of 0 to 16,000. Thus, if an OTU's HybScore changes by 1,000, it indicates a doubling in the fluorescence intensity. An OTU is defined by a group of highly similar 16S rRNA gene sequences. In most OTUs, the intra-OTU similarity is .99%. The data was reduced to only the bacterial OTUs meeting criteria for confirmed presence as described in Hazen et al. (2010) [18]. After the OTUs were identified for inclusion in the analysis, the values used for each OTU-sample intersection were populated in two distinct ways: i) Abundance metrics were used directly (AT); ii) Binary metrics were created where 1's represent presence and 0's indicate absence (BT). The HybScore was averaged from all present OTUs in a taxonomic family, such as Alcaligenaceae. The families containing more than 1% of the OTUs present were used for pairwise comparisons and construction of circular trees. The ratios were calculated as follows: Ratio~( HSi{HSj ) 1000 , where HS represents the average HybScore of OTUs in each family, i represents one treatment and j represents another treatment. The five comparative trees, CK 1 versus CK 2 , Amp versus CK 1 , Amp versus CK 2 , Gm versus CK 1 and Gm versus CK 2 , were constructed by the NJ method in BioNJ [20,21], and these were used as the initial trees for the maximum likelihood method in PhyML [22,23]. The resulting phylogenetic trees were uploaded to the iTOL website (http://itol.embl.de/) and reconstructed as circular trees [24,25]. The number of OTUs and their family ratios are presented in the circular trees. Bacterial microbiome in response to ampicillin and gentamicin Of more than 50,000 bacterial OTUs in the PhyloChip TM G3 array, 7,407 were detected in midribs from the tested citrus plants, of which 585 OTUs (7.90%) were shared by all samples. A total of 6,356 OTUs (85.8%) found in the Gm-treated samples were significantly more than the number of OTUs found in the ampicillin treatment (Amp, 1,795 OTUs, 24.2%), the disease control (CK 1 , 2,099 OTUs, 28.3%) and the healthy control (CK 2 , 1,306 OTUs, 17.6%). After subtracting the OTUs also shared in the controls (CK 1 and CK 2 ), 589 OTUs (32.8%) and 4,472 OTUs (70.4%) were detected in the Amp-and Gm-treatments, respectively ( Fig. 1A and Table S1). In total 53 phyla were detected, of which five phyla were comprised of more than 150 OTUs (more than 2% of the total detected OTUs), Proteobacteria (38.7%), Firmicutes (29.0%), Actinobacteria (16.1%), Bacteroidetes (6.2%) and Cyanobacteria (2.3%). The relative proportions of the above five dominant phyla differed between the antibiotic treatments. Higher percentages of Proteobacteria, Firmicutes and Bacteroidetes were detected in the Gm-treated samples than in the Amp-treated samples. However, higher percentages of Actinobacteria and Cyanobacteria occurred not in the Gm-treated samples but in the Amp-treated samples ( Fig. 1B and Table S1). Among the proteobacterial OTUs, the greatest numbers of unique OTUs were affiliated with Betaproteobacteria (17.0%), followed by Gammaproteobacteria (14.2%), Alphaproteobacteria (5.5%), Deltaproteobacteria (1.7%) and Epsilonproteobacteria (0.4%). Within the orders of Alphaproteobacteria, Rhizobials, to which the Las bacterium belongs, had the largest proportion, accounting for 2.1%, of the total detected OTUs, and this was due to their especially large percentages in the Gm-treatment and CK 1 . Within the Betaproteobacteria and Gammaproteobacteria, the families Comamonadaceae and Pseudomonadaceae, respectively, showed the highest OTU numbers and proportions in the Gmtreatment. The OTU62086, representing 'Ca. Liberibacter', was detected only in the inoculated plants from the Gm-treatment and the disease control (CK 1 ), which showed typical HLB symptoms (Fig. 2). Antibiotic efficacy against Las bacterium and phytotoxicity to citrus Amp and Gm were tested for their efficacy against the Las bacterium and evaluated for their phytotoxicity to citrus using scion growth rates. The Las-infected scions treated with Amp had .70% new growth as measured by emerging leaves or new flushes. However, only 47.5% and 50% of the scions had new growth when treated with Gm (Table 1) and water (disease control CK 1 ), respectively. Variance analysis showed that there were significant effects of the antibiotic treatments (Pr = 0.000) on HLB bacterial titers, scion infection rates, and Las bacterial transmission rates in the fixed model. All graft-inoculated plants, whose HLBaffected scions were treated with Amp or whose scions were Lasfree (CK 2 ), tested negative for the Las bacterium via qPCR (Ct<40.0), which indicates an estimated bacterial titer of ,100 cells/g of plant tissue. No scions were infected and no Las bacteria were transmitted into the rootstocks, indicating that Amp successfully eliminated Las from the HLB-affected scions. The inoculated plants from the scions treated with Amp displayed normal growth, green leaves, and no HLB-like symptoms, similar to the plants grafted with Las-free scions (CK 2 ) (Fig. 2). However, plants (scions and rootstocks) graft-inoculated using HLB-affected scions treated with Gm and water (CK 1 ) had higher Las scion infection rates, transmission rates, and bacterial titers (approximate 1.4610 6 cells/g of plant tissue) ( Table 1). The results indicate that Gm applied alone was not effective in eliminating the Las bacterium, and the plants showed typical HLB symptoms, such as yellow shoots and vein corking on leaves, in the rootstock (Fig. 2). The HybScore of OTU62086 in the Gm-treated samples indicated a fluorescent intensity greater than twice that measured in the Amp-treated samples and the healthy control (CK 2 ) ( Table 1). Specific OTUs associated with the diseased status and the antibiotic treatments In a pairwise comparison of the disease (CK 1 ) and healthy control (CK 2 ), only 500 OTUs (total OTUs number in Fig. 3A-B 1 and 3A-C 1 or Fig. 3B-B 2 and 3B-C 2 ) in 114 families were present in the Las-free CK 2 but absent in the Las-infected CK 1 , including 23 OTUs from Alcaligenaceae (only present in Fig. 3A-C 1 or Fig. 3B-C 2 ). However, 1,283 OTUs (total OTUs number in Fig. 3A-E 1 and 3A-F 1 or Fig. 3B-E 2 and 3B-F 2 ) in 155 families were present in CK 1 but absent in CK 2 , including 190 OTUs from Comamonadaceae, 128 from Staphylococcaceae and 120 from Flavobacteriaceae (Fig. 3). When the abundance or hybridization scores (HybScores) of the detected OTUs was taken into account, we found that the relative abundance (ratio) of several bacterial OTUs is a more important indicator of disease status than the exclusive presence of specific bacterial OTUs. Circular trees comparing CK 2 and CK 1 showed that 18 families had more than 1% of the 500 OTUs detected in Table 1. Ca. L. asiaticus (Las) and its transmission in grapefruit graft-inoculated with Las-infected lemon scions treated with ampicillin at 1.0 g/L (Amp), gentamicin at 100 mg/L (Gm), or water (disease control, CK 1 ) as well as grafted with the Las-free lemon scions (healthy control, CK 2 ). (Figure S1-B). However, only Alcaligenaceae, especially Achromobacterxy losoxidans, was more abundant in CK 2 than in CK 1 . In CK 1 , 14 families had more than 1% of the total 1,283 OTUs detected and the families of Methylobacteriaceae and Propionibacteriaceae were more abundant (Fig. S1-A). Plants graft-inoculated with HLB-affected scions treated with Amp appear Las-free and healthy. Our analyses in Fig. 3A showed that 18 out of 107 families contained over 1% of the 1,049 OTUs (total OTUs number in Fig. 3A-A 1 and 3A-F 1 ) detected in the Amp-treatment when compared to the healthy control (CK 2 ), while 20 out of 132 families had over 1% of the 702 OTUs number in Fig. 3A-A 1 and 3A-B 1 detected when compared to the disease control (CK 1 ). From the circular tree analysis, three families, Xanthomonadaceae, Propionibacteriaceae and Cyanpbacteriaceae, were abundant in the Amp-treatment (Fig. S3-E), while only Alcaligenaceae was abundant in CK 2 (Fig. S3-F). However, two families, Methylobacteriaceae and Staphylococcaceae, had HybScores with doubled fluorescence intensities in CK 1 (Fig. S2-D), while only Xanthomonadaceae was abundant in the Amp-treatment ( Fig. S2-C). Antibiotics Plants graft-inoculated with HLB-affected scions treated with Gm appeared diseased and contained a higher titer of the Las bacterium. Our analyses in Fig. 3B showed 18 out of 218 families had over 1% of the 4,679 OTUs (total OTUs number in Fig. 3B-A 2 and 3B-B 2 ) in the Gm-treatment when compared to CK 1 , while 15 out of 248 families had over 1% of the 5,422 OTUs (total OTUs number in Fig. 3B-A 2 and 3B-F 2 ) detected when compared to CK 2 . When compared to the disease control, all of the families, except Lachnospiraceae and Ruminococcaceae, were abundant in the Gm-treatment (Fig. S4-G), while only Methylobacteriaceae was more abundant in CK 1 (Fig. S4-H). All families except Ruminococcaceae were abundant in the Gmtreatment (Fig. S5-I), while only Alcaligenaceae was more abundant at ratio of 4.0 in CK 2 (Fig. S5-J). Discussion A microbial community analysis provides an accelerated approach to understanding the structure and function of bacterial communities, and it may result inthe isolation and identification of novel bacteria [26]. This study provides a comprehensive survey of the richness and composition of microbial communities in the leaf midribs of HLB-affected citrus and healthy citrus plants as well as HLB-affected scions treated with antibiotics in greenhouse. In a microbial community, more than 99% of the microorganisms have not been cultured [27], including the viable but nonculturable [28,29] and the dormant [30]. The updated PhyloChip TM G3 effectively studies bacterial diversity and composition, and it is an improvement over the last version in a number of ways. These include an increase from 500,000 to 1.1 million probes, the inclusion of strain-specific probe sets, the ability to detect over 50,000 OTUs instead of ,9,000 OTUs [18], and the utilization of over 320,000 sequences in the reference database, which is over 10 times greater than that for the PhyloChip TM G2. Many endophytic bacteria have been isolated from citrus [14,15,31]. Compared to the 15 phyla detected in citrus leaves in a previous report using PhyloChip TM G2 [15], we have detected 53 phyla in the HLB-affected citrus in greenhouse using the updated PhyloChip TM G3. A total of 7,407 bacterial OTUs were detected from the bacterial microbiome, of which 585 OTUs were present in all samples. A total of 6,356 OTUs were detected in the Gm-treatment, which was significantly higher than those detected in the Amp-treatment (Amp, 1,795 OTUs), the disease control (CK 1 , 2,099 OTUs), and the healthy control (CK 2 , 1,316 OTUs). In our previous report, 7,028 known OTUs were detected in citrus leaf midribs of the HLB-affected citrus treated by antibiotic combinations (PS and KO) in the field using the same PhyloChip TM G3 array. These OTUs were from 58 phyla, of which five contained 100 or more OTUs, Proteobacteria (44.1%), Firmicutes (23.5%), Actinobacteria (12.4%), Bacteroidetes (6.6%) and Cyanobacteria (3.2%). In the antibiotic treated samples, the number of OTUs decreased to a total of 5,599 [14]. A metagenomic analysis of citrus phloem alone showed that only the Las bacterium was associated with HLB [17]. Thus, these other families are most likely present in tissues other than the phloem and may relate to secondary proliferation in declining leaves rather than relating to initial HLB development. Comparative analysis of microbial community provides an approach to understanding community structure and function. Some microorganisms isolated from plant tissues exhibit potential as biocontrol agents against phytopathogens [32], promote plant growth, and hasten plant development [33]. However, there are no reports of synergistic interactions between endophytic microorganisms and phytopathogens that result in a plant disease. In a pairwise comparison (Fig. 3, Fig. S1-B, Fig. S3-F, Fig. S5-J and Table S1), 23 OTUs from the family Alcaligenaceae were abundant only in the healthy control (CK 2 ), including most OTUs of Achromobacter xylosoxidans (14378, 14510, 14570, 14691,14717, 14737, 14789, 15105, 15502, 15845 and 15854) and other Achromobacter spp. A. xylosoxidans has been reported to inhibit the growth of plant pathogens by the production of chitinase, or other inhibitory substance [32,34], or through iron competition [35]. In a previous report, an increased abundance of Alcaligenaceae was reported in the asymptomatic samples when compared to the symptomatic samples of Las-infected citrus [15]. Due to the limited amount of soluble iron in the rhizosphere, microbes and plants scavenge for iron using highly sophisticated iron binding and uptake mechanisms [36]. The acquisition of iron is recognized as one of the key steps in the survival of any pathogen in its host [37]. Our results indicated Las-infected plants were deficient in zinc, iron, nitrogen, and phosphorus, and they produced more potassium and boron than uninfected plants (unpublished data). These findings may warrant further investigation on whether A. xylosoxidans plays a significant role in suppressing HLB disease symptoms. The results (Fig. 3, Fig. S1-A, Fig. S2-D, Fig. S4-H and Table S1) presented here also indicated that 12 OTUs (59185, 59212, 59404, 59410, 59417, 59549, 59601, 59718, 59757, 59917, 59976 and 60144) from the genus Methylobacterium in the family of Methylobacteriaceae were more abundant in the disease control (CK 1 ) when compared to the other treatments. Methylobacterium was also detected in the root samples from HLB-affected citrus plants [16]. The genus Methylobacterium resided in the xylem vessels of citrus plants, and abundant Methylobacterium spp. in citrus triggers CVC disease by a synergistic interaction with X. fastidiosa [31]. Figure 3. Distribution of the bacterial OTUs in response to antibiotic treatments. In the Venn diagram, the numbers in parentheses represent the number of bacterial OTUs that occurred in each antibiotic treatment [ampicillin (Amp) and gentamicin (Gm)], disease controls (CK 1 ), healthy control (CK 2 ) and their intersections. Pie charts A to G correspond to the appropriately labeled Venn diagram areas (A 1 to G 1 for the Amp treatment and A 2 to G 2 for the Gm treatment) and show families that contained over 1% of the total OTUs in each area. In pie charts A to G, the names of the families are followed by their frequencies as a percentage (%). doi:10.1371/journal.pone.0076331.g003 Therefore, the abundance of the endophytic Methylobacterium may be associated with HLB progression. The antibacterial activity of an antibiotic is influenced by the state of bacterial responsiveness, the physicochemical environment at the infection site, and the degree of cooperation with the host defenses [38,39]. The results presented here indicate that Amp but not Gm was effective in eliminating the Las bacterium. The grafted lemon scions had much more severe HLB symptoms with higher Las titers following Gm treatment (Table 1 and Fig. 2). Understanding the structure and species composition of bacterial communities is necessary for evaluating the influence of the applied antibiotics. From a pharmacodynamic point, the intracellular concentration of the antibiotic is critical for Gm and the time of exposure is important for Amp. It is reported that Gm kills bacterial cells by inhibiting 30S ribosomal protein synthesis and disrupting lipopolysaccharides in the outer membrane [39]. Amp belongs to the penicillin group of betalactam antibiotics and acts as a competitive inhibitor of the transpeptidase to prevent bacterial cell wall synthesis in binary fission, which ultimately leads to bacterial cell lysis [40]. Ten abundant OTUs (16112, 16171, 16258, 16452, 16529, 16992, 17063, 17215, 17247, and 17254) from Stenotrophomonas spp. in the family of Xanthomonadaceae were detected only in the Amp-treatment, but not in the controls (Fig. 3A, Fig. S2-C, Fig. S3-D and Table S1). Xanthomonadaceae is a wide-spread family of bacteria belonging to the gamma subdivision of the Gram-negative proteobacteria, which includes two plant-pathogenic genera, Xanthomonas and Xylella, and the related genus Stenotrophomonas. Stenotrophomonas was abundant only in the Amp-treatment, and Xylella was not detected in any sample. It has been reported that Stenotrophomonas spp. produce antifungal antibiotics and have growth promoting activities on plants [41]. It is intriguing that the number and abundance of OTUs in the Gm-treatment were much more than those in the Amp-treatment and the controls. Over 85% of the total detected OTUs were found in the Gm-treatment. All the families with over 1% of the total OTUs in the Gm-treatment were abundant except Lachnospiraceae and Ruminococcaceae (Fig. 3B, Fig. S4-G and Fig. S5-I). However, Gm-treatment had lower percentage of Actinobacteria and Cyanobacteria. Cyanobacteria was reported to produce antimicrobial compounds against several Gram-positive bacteria, such as Bacillus subtilis, Bacillus pumulis, Enterococcus faecalis, Staphylococcus aureus and Staphylococcus epidermidis, and Gramnegative bacteria, such as Escherichia coli, Pseudomonas aeruginosa and Klebsiella pneumoniae [42]. We hypothesize that Gm-treatment might break the existing balance of bacterial communities in the citrus, and result in more OTUs detectible by killing or suppressing some critical OTUs in the balance. Further verification of this hypothesis is necessary to address this finding. Although an antibiotic treatment may be effective to a given number of bacterial diseases, it is critical to measure its ecological effects in addition to the effects on pathogens. In this research, we revealed the bacterial communities of citrus, with and without HLB infection, along with different antibiotic treatments, which has provided new insights into HLB progression, and the bases for the development of more effective and eco-friendly HLB control strategy. Availability of supporting data The data sets supporting the results of this article are available in the Geo repository, GSE46728 http://www.ncbi.nlm.nih.gov/ geo/query/acc.cgi?acc = GSE46728. Figure S1 Comparative trees of CK 1 versus CK 2 . Phylogenetic trees of families with over 1% of the total detected Operational Taxonomic Units (OTUs) from the bacterial community of leaf midribs of scions from grapefruit graftinoculated with HLB-affected lemon scions (disease control, CK 1 ) and with Las-free scions as the healthy controls (CK 2 ). The half-circle A) OTUs present in CK 1 and absent in CK 2 ; B) OTUs present in CK 2 and absent in CK 1 . (DOCX) Figure S2 Comparative trees of Amp versus CK 1 . Phylogenetic trees of families with over 1% of the total detected Operational Taxonomic Units (OTUs) from the bacterial community of leaf midribs from grapefruit graft-inoculated with HLB-affected lemon scions treated with ampicillin (Amp) and water (disease control, CK 1 ). The half-circle C) OTUs present in Amp and absent in CK 1 ; D) OTUs present in CK 1 and absent in Amp. (DOCX) Figure S3 Comparative trees of Amp versus CK 2 . Phylogenetic trees of families with over 1% of the total detected Operational Taxonomic Units (OTUs) from the bacterial community of leaf midribs from grapefruit graft-inoculated with HLB-affected lemon scions treated with ampicillin (Amp) and with Las-free scions were the healthy controls (CK 2 ). The half-circle E) OTUs present in CK 2 and absent in mp; F) OTUs present in Amp and absent in CK 2 . (DOCX) Figure S4 Comparative trees of Gm versus CK 1 . Phylogenetic trees of families with over 1% of the total detected Operational Taxonomic Units (OTUs) from the bacterial community of leaf midribs from grapefruit graft-inoculated with HLB-affected lemon scions treated with gentamicin (Gm) and water (disease control, CK 1 ). The half-circle G) OTUs present in Gm and absent in CK 1 ; H) OTUs present in CK 1 and absent in Gm. (DOCX) Figure S5 Comparative trees of Gm versus CK 2 . Phylogenetic trees of families with over 1% of the total detected Operational Taxonomic Units (OTUs) from the bacterial community of leaf midribs from grapefruit graft-inoculated with HLB-affected lemon scions treated with gentamicin (Gm) and with Las-free scions were the healthy controls (CK 2 ). The half-circle I) OTUs present in CK 2 and absent in Gm; J) OTUs present in Gm and absent in CK 2 .	6,361.6	2013-11-08T00:00:00.000	[ "Biology", "Environmental Science" ]
Cavity-QED-based scheme for verification of the photon commutation relation We propose an experimental scheme to prove the photon commutation relation, [a, a†]=1. The scheme exploits interaction between a single-mode cavity field and three two-level atoms, two of which are prepared in an entangled state. The success of the scheme is subject to observation of the atoms in predetermined states after the interaction. We show, in particular, that a reasonably high success probability can be obtained with our scheme by preparing the cavity field in a superposition of two Fock states and choosing the interaction times appropriately. Introduction Recent progress in quantum state engineering has made the experimental realization of singlephoton creation operation a † and annihilation operation a possible [1]. The field states generated by such operations are usually termed photon-added and photon-subtracted states, respectively. Photon-added coherent and thermal states were first considered by Agarwal and Tara [2,3], who showed that these states exhibit non-classical properties that are absent in the original coherent and thermal states. The photon creation operation a † on coherent and thermal states was performed experimentally by Zavatta et al [4]- [6]. The same group achieved sequences of photon creation and annihilation operations, namely photon-annihilation-thencreation operation a † a and photon-creation-then-annihilation operation aa † [7]. They showed, in particular, that operations a † a and aa † applied to the same field state produce states exhibiting different photon statistics, proving experimentally the non-commutativity of operators a and a † . This, however, does not prove the exact commutation relation [a, a † ] = 1. The exact commutation relation can only be proven if one constructs a scheme that performs a coherent superposition of the two operations a † a and aa † . One such scheme was recently proposed [8] and realized experimentally [9], in which a coherent superposition of the two operations is achieved in a single-photon interferometer setup employing beam splitters and a parametric amplifier. The commutation relation [a, a † ] = 1 is the most fundamental relation that defines quantum optics, and it is thus of fundamental importance to prove it experimentally. In this paper, we propose a cavity-quantum electrodynamics (QED)-based scheme to prove the commutation relation [a, a † ] = 1. A theoretical analysis of cavity-QED-based schemes that perform individual photon creation and annihilation operations and their combinations, including the commutation operation, has been given recently [10]. A cavity-QED-based scheme to realize the photon-annihilation-then-creation operation a † a and the creation-thenannihilation operation aa † has been proposed [11,12], exploiting interaction between two atoms and a cavity field. The scheme we propose here exploits interaction between three atoms, two of which are entangled initially, and a cavity field. A coherent superposition of the two operations, a † a and aa † , results from the initial entanglement of the two atoms and post-selection of the states of the atoms that erases information on whether photon subtraction occurs before or after photon addition. Like the scheme proposed by Kim et al [8], our scheme is probabilistic as it relies on post-selection of the states of the atoms after their interaction with the cavity field. We pay particular attention to the issue of the success probability of the scheme. In section 2, we briefly review the fundamentals of the interaction between a two-level atom and a monochromatic radiation field. The scheme we propose is described in section 3. It is shown that the final state of the cavity field after its interaction with the three atoms is approximately equal to the state that results when the commutation operation aa † − a † a is applied to the initial cavity state, if the interaction times are chosen sufficiently short. By showing that the final state of the cavity field is the same as the initial cavity state, the commutation relation can thus be proven. The success probability of the scheme, however, is generally very small. In section 4, we show that the success probability of the scheme can significantly be enhanced if consideration is restricted to the case when the cavity field is prepared initially in a superposition of two Fock states. In section 5, we describe the experimental feasibility of the proposed scheme. Finally, section 6 presents a summary. Atom-field interaction We briefly review the atom-cavity field interaction and introduce our notation. The time evolution of the atom-field system is determined by In (1), g is the atom-field coupling constant assumed to be real, σ − and σ + are atomic ladder operators defined as σ − = \|g e\|, with \|g and \|e denoting the lower and upper states of the atom, and the photon annihilation and creation operators a and a † are given by Note that we leave f n of (3) unspecified, because our purpose is to propose an experimental scheme that proves the commutation relation [a, a † ] = 1. If the initial state of the atom-cavity system is given by \|g n c n \|n , corresponding to the situation where initially the atom is prepared in \|g and the cavity field is in an arbitrary pure state n c n \|n , (1) yields If the initial state is given by \|e n c n \|n , then (1) yields In (4) and (5), C g (t), S g (t), C e (t) and S e (t) are field operators defined as S e (t) = a † sin(gt Equations (4) and (5) can, respectively, be expressed, with the help of (3), as and The scheme The scheme we propose to prove the commutation relation consists of three two-level atoms and a single-mode cavity. The three atoms enter the cavity and interact with the cavity field one after another. We label them atoms 1, 2 and 3 in the order they enter the cavity. Atoms 1 and 3 are prepared initially in a singlet Bell state 1 √ 2 (\|e 1 \|g 3 − \|g 1 \|e 3 ), while atom 2 is prepared in its upper state \|e 2 . The cavity field, assumed to be resonant with the atomic transition, can be in any arbitrary pure state, \|ψ c (t = 0) = n c n \|n . The experiment proceeds as atoms 1, 2 and 3, one after another, enter the cavity and interact with the cavity field for times t 1 , t 2 and t 3 , respectively. Measurement is made on the three atoms after the interactions are over, and the experiment is accepted only if the atoms are measured to be in the state \|e 1 \|g 2 \|e 3 . Straightforward algebra yields that the final state of the cavity field, contingent upon postselection of the three atoms in \|e 1 \|g 2 \|e 3 , is given by where N {} signifies that the unnormalized state inside the curly bracket should be normalized. The physical interpretation of (9) is straightforward. The first term inside the curly bracket on the right-hand side results from a series of interactions with atom 1 prepared in \|e 1 remaining in \|e 1 [C e (t 1 )], atom 2 making a transition \|e 2 → \|g 2 [S e (t 2 )], and atom 3 prepared in \|g 3 making a transition \|g 3 → \|e 3 [S g (t 3 )], whereas the second term results from a series of interactions with atom 1 prepared in \|g 1 making a transition \|g 1 → \|e 1 [S g (t 1 )], atom 2 making a transition \|e 2 → \|g 2 [S e (t 2 )], and atom 3 prepared in \|e 3 remaining in \|e 3 [C e (t 3 )]. The − sign between the two terms arises from the phase difference of π in the initial singlet Bell state of atoms 1 and 3. Substituting (6a)-(6d) into (9) and using (3), we can express (9) as We desire this state to be equal to the state \|ψ desired = N {(aa † − a † a) n c n \|n }. The commutation relation can then be proven by showing experimentally that this state \|ψ c (t 1 , t 2 , t 3 ) coincides with the initial cavity state n c n \|n . It can be easily seen that, if one takes short interaction times such that wheren denotes the average number of photons in the initial cavity field, (10) becomes and our mission is accomplished. (Strictly speaking, the limit of short interaction times represented by (11) should be given in terms of n max , the maximum number of photons in the initial radiation field for which \|c n \| is non-negligible, instead of in terms ofn. We assume, however, that the radiation field has a smooth, continuous distribution of photon numbers, in which case the condition \| f n max \|gt 1 implies \| fn\|gt 1.) We can understand this result by noting that, in the limit of short interaction times, we have C g (t) 1, C e (t) 1, S g (t) (gt)a and S e (t) (gt)a † , and (9) becomes (12). Note that, due to the normalization in (12), the commutator [a, a † ] = aa † − a † a is realized by our scheme only up to a constant. By showing that the cavity field state after the interaction with the three atoms, with the interaction times t 1 , t 2 and t 3 chosen to be sufficiently short to satisfy (11), is identical to the initial cavity state, we can thus prove [a, a † ] = K , where K is an arbitrary constant. To complete the proof, one needs to show that the constant K is indeed one. One way to prove K = 1 is to measure the success probability, i.e. the probability of observing the atoms in the desired atomic state \|e 1 \|g 2 \|e 3 after the interaction. The success probability of our scheme can be easily calculated to be given by In the limit (11) of short interaction times, we obtain Noting that \| f n+1 \| 2 = n\|aa † \|n and \| f n \| 2 = n\|a † a\|n , (14) reduces to Measurement of the success probability thus determines the value of the constant K . As described above, in order to prove the commutation relation with our scheme, one needs to go to the limit (11) of short interaction times. Since the success probability P ∝ (gt) 2 (gt 2 ) 2 , our scheme suffers from a very low success probability. The problem of low success probability also exists in the optical scheme proposed by Kim et al [8]. This problem of low success probability may not be a serious threat to the optical scheme, because the traveling optical field can be provided at a high pulse rate. For the cavity-QED-based scheme, however, atoms are provided at a relatively low rate, and therefore it is desirable to find a way of enhancing the success probability. In the next section, we demonstrate that if the cavity field is prepared initially in a superposition of a small number (two) of Fock states, the success probability can be significantly enhanced. The case when the cavity field is prepared in a superposition of two Fock states In this section, we consider the simple case in which the cavity field is initially prepared in a superposition of two Fock states, α\|N 1 + β\|N 2 . Substituting c n = αδ n N 1 + βδ n N 2 (δ denotes the Kronecker delta function) into (10), we obtain In order for our scheme to realize the commutator [a, a † ], this state needs to be the same as the The success probability in this case is given, according to (13), by If (17) is satisfied, the success probability reduces to The problem reduces to finding the interaction times t 1 , t 2 and t 3 that satisfy (17). One obvious choice is the limit (11) of short interaction times. With such short interaction times, one sees that (17) is satisfied with Q (gt)(gt 2 ) and (19) , which is consistent with (14). As pointed out earlier, 7 however, this success probability is very low. It should be noted from (19) that P is of the order of Q 2 . It can be seen from (17) that if one can find non-small values of t 1 , t 2 and/or t 3 that satisfy (17), then Q and consequently P may not necessarily be small. In order to pursue this possibility, we consider the simplest case of N 1 = 0 and N 2 = 1. In this case, condition (17) becomes Let us take the limit (11) of short interaction times for t 1 and t 3 (\| fn\|gt 1 = \| fn\|gt 3 ≡ \| fn\|gt 1) but leave t 2 arbitrary. Equation (20) then becomes Although f 1 and f 2 are left unspecified, it can be seen that it is not too difficult to find nonsmall values of gt 2 that satisfy (21). Q is then of the order of (gt) and P is of the order of (gt) 2 . Although this success probability is still small, it represents a significant enhancement over P ∝ (gt) 2 (gt 2 ) 2 , the success probability as given by (15) In summary, our scheme to prove the commutation relation [a, a † ] = K with the success probability significantly higher than the scheme of section 3 goes as follows. As in the scheme of section 3, we prepare atoms 1 and 3 in the entangled state 1 √ 2 (\|e 1 \|g 3 − \|g 1 \|e 3 ) and atom 2 in \|e 2 . The cavity, however, is prepared in a superposition of two Fock states, for example, in a superposition of the vacuum and the one-photon state. We then let the three atoms interact with the cavity field one after another for interaction times t 1 , t 2 and t 3 , respectively. We post-select the case when the atoms after their interaction with the cavity field are observed to be in the state \|e 1 \|g 2 \|e 3 . We choose the interaction times in such a way that (17) is satisfied. For the case when the cavity is prepared initially in a superposition of the vacuum and the one-photon state, we can choose gt 2 = 3.7613, while gt 1 = gt 3 ≡ gt is chosen sufficiently small to satisfy (11). The success probability of the scheme will then be of the order of (gt) 2 . Or we can choose gt 1 = gt 2 = gt 3 = 2.70 if an error of ∼ 15% can be tolerated. The success probability of the scheme will then be ∼ 0.014. By showing that the final cavity state after interaction with the atoms is identical to the initial cavity state, we prove experimentally that [a, a † ] = K . To prove K = 1, we can exploit interaction between a single atom and a cavity field. We prepare the atom in \|e and the cavity field in vacuum, \|ψ(t = 0) = \|e \|0 . We let the atom 8 interact with the cavity field for time t and post-select the case when the atom is observed to remain in \|e after the interaction. The probability of finding the atom in \|e is given by Noting that \| f 1 \| 2 − \| f 0 \| 2 = 0\|aa † − a † a\|0 = K and that a\|0 = 0 and therefore f 0 = 0, equation (22) can be written as One can therefore determine the value of K by monitoring the periodic variation of the success probability as the interaction time is varied and determining the period of the variation. Experimental feasibility In this section, we discuss the experimental feasibility of our scheme by giving realistic considerations to some experimental issues. In order to prove the commutation relation with the proposed scheme, it is necessary to verify that the final cavity state produced by the scheme is identical to the initial cavity state. Since the scheme operates in the strong coupling regime, direct measurement of the field state inside the cavity is not allowed. One thus needs to rely on probe atoms to obtain information on the field state. One such method that allows direct measurement of the Wigner distribution function of the cavity field is proposed [13] and demonstrated [14] experimentally for the vacuum and the one-photon state. The theory given in the previous sections assumes a lossless cavity and neglects atomic decay. In this regard, we note that our scheme needs to operate in the strong coupling regime in which the atom-field coupling constant (or the vacuum Rabi frequency) g is large compared with the cavity decay rate κ and the atomic decay rate γ . For the scheme described in section 3, the interaction time t needs to be chosen sufficiently small that gt 1. Since in the strong coupling regime g is large compared with κ and γ , the inequalities κt 1 and γ t 1 are well satisfied. For the scheme of section 4, the parameter gt may be of order unity, gt 1, but neglecting the cavity decay and atomic decay can still be justified. Let us take as an example the rubidium atom in a high-Q microwave cavity operating on hyperfine levels. We have g 10-100 kHz, whereas κ 0.01-1 kHz (corresponding to Q 3 × 10 10 -3 × 10 8 ) and γ 0.1-10 kHz [15]- [17], and thus the conditions κt 1 and γ t 1 can easily be satisfied. If we consider cesium atoms operating in the optical regime, we have, for example, g 34 MHz, κ 4.1 MHz and γ 2.5 MHz [18]. Although both cavity decay and atomic decay are more important in the optical regime, it is still possible to operate under the condition where they play relatively insignificant roles. Perhaps the most demanding part of the scheme is the initial preparation of the atomic states and the cavity field states. Let us first consider the initial atomic states. Two atoms are required to be prepared in a maximally entangled singlet Bell state. We recall that it is exactly the entangled atoms in a singlet Bell state that were generated in the experiment in which generation of entanglement in massive particles was first demonstrated [19]. Subsequently, various elaborate cavity-QED methods of generating entangled atoms have been proposed and shown to be experimentally feasible [20]- [25]. Recently, entangled atoms or ions have been generated to teleport unknown atomic states [26,27]. One can therefore say that it is within the reach of current experimental technology to prepare atoms in the entangled state required by our scheme. There, however, remains a problem because, due to experimental errors, the 9 entangled state actually prepared may not exactly coincide with the desired entangled state. One possible source of error is the uncertainty in the interaction time arising from the finite resolution of the atomic velocity. Let us consider, for example, the scheme of [19] to prepare two atoms in the singlet Bell state. Two atoms, atoms A and B, prepared in the excited state and in the ground state, respectively, interact one after the other with a cavity prepared initially in vacuum. The interaction times, t A and t B , for atoms A and B, respectively, are chosen such that gt A = π 4 and gt B = π 2 . Due to experimental error, however, one may actually have gt A = π 4 (1 + A ) and gt B = π 2 (1 + B ), where A 1 and B 1. The state of the system, atom A + atom B + cavity field, after the interactions then becomes 1 Post-selection of the cavity field in vacuum results in the two atoms in the entangled state 1 Atoms A and B now play the role of atoms 1 and 3 in our proposed scheme. The initial state of the three atoms for our scheme is then given by 1 where ≡ A represents the error arising from a finite velocity resolution of the π 4 atomic pulse. Through straightforward algebra, we obtain that the final state of the cavity field produced by our scheme of section 3 is given, under the limit (11) of short interaction times, by The fidelity of this state with respect to the desired state \|ψ desired = (aa † − a † a) n c n \|n = n c n \|n , defined as F = \| ψ desired \|ψ c (t 1 , t 2 , t 3 ) \| 2 , is then given by F 1 − π 2 8 2 (2n + 1) 2 . This indicates that if our scheme is to work accurately against the error , the inequality 2 (2n + 1) 2 1, i.e.n 1/2 , must be satisfied. Taking 0.005 corresponding to a typical velocity resolution of 0.5% [28], we must haven 100. The experimental error in the initial preparation of atomic entanglement puts restrictions on the average photon number of the initial field state. We can also carry out a similar analysis for the scheme of section 4. Taking the initial atomic state to be given by (24) and taking the initial cavity field state as α\|0 + β\|1 , we obtain for the final cavity state, under the condition of short interaction times t 1 and t 3 , The fidelity of this state with respect to the desired state \|ψ desired = α\|0 + β\|1 is given by The fidelity remains very close to 1 for 0.005. We can conclude that our scheme is not very sensitive to the error in the initial preparation of atomic entanglement, as long as the average photon number of the initial cavity state is not too large. In this respect, our scheme favors smalln. As described in section 4, if our scheme is to perform with a reasonably high success probability, the cavity field needs to be prepared in a superposition of two Fock states. Various schemes to generate an arbitrary finite superposition of Fock states have been proposed [29]- [36] and demonstrated experimentally [37]- [39] in the past. In particular, an arbitrary superposition of the vacuum and the one-photon state can be generated using the optical quantum scissors device or other methods, as already demonstrated experimentally [37]- [39]. It appears therefore that the requirement to prepare the cavity field in a superposition of two Fock states is also within the reach of current experimental technology, although it is certainly not an easy requirement to be fulfilled. Finally, we consider the effect due to the error in atomic state detection. Atomic detectors have miscounts of 2-5% [16,40]. This is potentially a serious problem for our scheme, because the miscount probability may be significantly higher than the success probability of the scheme and, as a consequence, it may be difficult to discern the success of the scheme from misfiring of the detector. We note, however, that the miscount rate can be made arbitrarily low at the expense of the detector efficiency and higher miscount probabilities of complementary detectors [40]. In order to quantitatively estimate the effect of the error in atomic state detection, we consider the case in which atoms 1 and 2 are post-selected correctly but atom 3 is post-selected incorrectly, i.e. the case when atom 3 in \|g 3 is incorrectly identified as in \|e 3 . The incorrect post-selection of atom 3 yields, for the final state of the cavity field, as opposed to (9), which results when the correct post-selection is made. Under the limit (11) of short interaction times, (27) reduces to \|ψ c (t 1 , t 2 , t 3 ) N {a † n c n \|n }. The fidelity of this state with respect to state (aa † − a † a) n c n \|n = n c n \|n is given by which reduces, for the initial coherent state, to F =n/(n + 1). One thus sees that ifn is large, the final cavity state produced by our scheme is insensitive to the error in atomic state detection. This insensitivity is rather undesirable, because erroneous data resulting from the detection error need to be discarded. The insensitivity makes it difficult to judge whether the coincidence of the final state produced by the scheme with the desired state actually proves the commutation relation or results simply from the insensitivity. One therefore sees that, in this respect, too, our scheme favors smalln. Ifn is large, it is difficult to detect the detection error simply by looking at photon statistics. One may have to rely on other more sensitive means, such as looking for the negativity of the Wigner distribution function. Another problem in atomic state detection in addition to misreading of atomic states is the possibility of atoms not being detected. In order to avoid errors arising from having more than one atom in a detection window, one needs to have atoms well separated in space from one another. When atoms are not detected, the experiment can be aborted. The detection efficiency of atomic detectors typically ranges from 50 to 90% [28], [40]- [44]. Taking, for example, the detection efficiency to be 0.8, the success probability of the scheme will be smaller by a factor of (0.8) 3 0.5 (the detection efficiency of one detector should be cubed because our scheme requires detection of three atoms) than the success probability with perfect detectors. The finite detection efficiency only lowers the success probability and it does not cause any error, as long as there is one or less than one atom in each detection window. Considering, however, that the proposed scheme generally has a low success probability, further lowering of the success probability is undesirable. It appears therefore that, for successful execution of the proposed scheme, it is important to minimize the error arising from misreading of atomic states and at the same time to maximize the detection efficiency of the atomic detectors. Summary We have proposed a cavity-QED-based scheme to prove the photon commutation relation [a, a † ] = 1. The scheme exploits interaction between a cavity field and three atoms, two of which are prepared in an entangled state. The proof is carried out in two steps. Firstly, the relation [a, a † ] = K , where K is an arbitrary constant, is proven by showing that the field state after the interaction, contingent upon post-selection of atoms in the desired state, is identical to the initial cavity state. Secondly, the constant K is proven to be one by measuring the success probability, i.e. the probability to find atoms in the desired state. If the scheme is to work for any arbitrary initial state of the cavity field, one needs to choose sufficiently short interaction times t 1 , t 2 and t 3 , as described in section 3. The choice of the short interaction times, however, leads inevitably to the problem of low success probability of the scheme. If we restrict our consideration to the case when the initial cavity state is prepared in a superposition of only two Fock states, it is then generally possible to choose non-small values of the interaction times that yield the final cavity state identical to the state that results when the commutation operation is applied to the initial cavity state, as described in section 4. The corresponding success probability is significantly higher than that of section 3. This higher success probability, however, comes at the expense of having to prepare the cavity field in a superposition of two Fock states.	6,540.8	2010-03-01T00:00:00.000	[ "Physics" ]
The longest path in the Price model The Price model, the directed version of the Barabási–Albert model, produces a growing directed acyclic graph. We look at variants of the model in which directed edges are added to the new vertex in one of two ways: using cumulative advantage (preferential attachment) choosing vertices in proportion to their degree, or with random attachment in which vertices are chosen uniformly at random. In such networks, the longest path is well defined and in some cases is known to be a better approximation to geodesics than the shortest path. We define a reverse greedy path and show both analytically and numerically that this scales with the logarithm of the size of the network with a coefficient given by the number of edges added using random attachment. This is a lower bound on the length of the longest path to any given vertex and we show numerically that the longest path also scales with the logarithm of the size of the network but with a larger coefficient that has some weak dependence on the parameters of the model. the longest path in the price model tim S. evans * , Lucille calmon & Vaiva Vasiliauskaite the price model, the directed version of the Barabási-Albert model, produces a growing directed acyclic graph. We look at variants of the model in which directed edges are added to the new vertex in one of two ways: using cumulative advantage (preferential attachment) choosing vertices in proportion to their degree, or with random attachment in which vertices are chosen uniformly at random. In such networks, the longest path is well defined and in some cases is known to be a better approximation to geodesics than the shortest path. We define a reverse greedy path and show both analytically and numerically that this scales with the logarithm of the size of the network with a coefficient given by the number of edges added using random attachment. This is a lower bound on the length of the longest path to any given vertex and we show numerically that the longest path also scales with the logarithm of the size of the network but with a larger coefficient that has some weak dependence on the parameters of the model. The Price model 1,2 is one of the oldest network models and it was motivated by the pattern of citations in academic papers. In a citation network, each node represents a document while every entry in the bibliography of a document t is represented by a directed edge from an older document, node s, to node t. One of the key features of a citation network, one inherent in the Price model, is that there is a fundamental arrow of time in the network; bibliographies can only refer to older documents. This means that there are no cycles in the network, you can never find a path from a node that returns to that node. Thus a citation network is an example of a Directed Acyclic Graph (DAG). Mathematically, DAGs have some distinctive properties and one of them is that for any pair of connected nodes there is a well defined and meaningful longest path length, for example see Fig. 1. Contrast this with, for example, undirected networks, where you can often find many paths between two given vertices that visit most of the nodes in a component so longest paths are often as long as the component is big, if all nodes in the path must be distinct, and infinite, if multiple visits to the same node were allowed. In directed graphs with cycles, the longest path is infinite, if multiple visits to a node are allowed. Both of these definitions of the longest path coincide if the network is acyclic, as the absence of cycles ensures that in any path, a node can only occur once. In a citation network, it is not clear how useful the shortest path is. For instance, in writing this paper, the oldest citation we have is to a paper by Price 1 . The shortest path to Price's paper from this work has length one. On the other hand, most of the knowledge of that work contained in this paper did not come directly from that paper. We only reread Price's paper to check one detail while working on this project. So the length of the shortest path to that paper seems largely irrelevant. Rather, the information in this early bibliometrics paper by Price has reached us through a sequence of other work, much of it not explicitly referenced in our paper. We drew on much more recent documents such as the reference book by Newman 4 which in turn cites papers which developed various aspects of the Price model. Indeed there is much evidence 3,5,6 that typically 70% or so of a bibliography may not have been used directly when producing the work in an academic paper. So our thesis is that for DAGs the longest path plays a much more important role than the shortest path. In simple models the longest path has been shown to be the best approximation to the geodesic for models of DAGs embedded in Minkowski space 7 where there is a single time direction. This has been exploited in real data sets where dimension and curvature of a DAG can be measured 8,9 enabling us to embed DAGs such as citation networks in Minkowski space 10 . A similar rigorous link for undirected networks has only been made for the shortest path in networks embedded in Euclidean space where there is no arrow of time 11,12 ]. The properties of the longest path have been investigated in the context of simple models known as Cube Spaces 13 which include those built from Poisson Point Processes in Minkowski spaces where all causally connected points are connected to form a network. However these are examples of transitively complete DAGs, that is if there is a path between two points then there is always an edge connecting those two points directly. However, that is not true for a citation network where the limited size of a bibliography means no document ever cites every older paper to which it has some connection. What we seek to do in this paper is to look at the Analytic results In the Price model 2 (for instance see Sect. 14.1 of Newman 4 ) we start from a network G(t) defined at an integer 'time' t. We create a new graph G(t + 1) by first adding one new vertex, which we label with the time (t + 1) . This new node, (t + 1) , is connected to m existing vertices s in the graph G(t). These m existing vertices {s} are each chosen with probability �(t, s) . We will use a convention that these edges point from older to newer vertices, from s to (t + 1) . Once these edges have been added we have our new graph G(t + 1) . The process is then repeated. For an example of how a network grows according to the Price model, see Fig. 2. The mathematical and numerical simplicity of this model comes from the simple definition of �(t, s) . To define the probability �(t, s) we first define N(t) = N 0 + t be the number of nodes in the graph G(t) for Fig. 1. An illustration of a Price-model style DAG where the longest, shortest and reverse greedy paths from last point to the first are distinct. The longest path from the source node to the sink node is highlighted in blue dot-dash line; the reverse greedy path is the dotted green path. Note the first edge is the same for both-the green-blue edge. As illustrated here, the Price model produces DAGs which are neither transitively complete nor transitively reduced. In a transitively complete DAG, all nodes which are connected by a path are connected by a direct edge. Likewise, except for the case of one-incoming edge per node, the model is not transitively reduced 3 , that is some edges could be removed without removing a path between any pair of nodes. www.nature.com/scientificreports/ some constant N 0 . The number of edges in the graph G(t), after all m edges have been added to node t, is E(t) = E 0 + mt where E 0 is some constant. Finally in the graph G(t) let the node created at time s have out-degree k (out) (t, s) , the number of edges leaving s and connecting it to later nodes. In this model, the connection of edges to new node (t + 1) is made in one of two ways. With probability p node (t + 1) is connected to an existing vertex s chosen in proportion to the number of edges leaving s for later nodes in G(t), k (out) (t, s) . Price called this cumulative advantage and, after normalisation, we have that the probability of choosing s is k (out) (t, s)/E(t) . The second process happens with probability p = (1 − p) and in this case we choose the source vertex s uniformly at random from the set of vertices in G(t), i.e. with probability 1/N(t). If we start the process at time equal to 1, then the probability of connecting the vertex (t + 1) to existing vertex s is �(t, s) where unless s = t = 1 when �(t, s) = 1 , otherwise �(t, s) = 0 . Note that in his original paper, Price considered p = m/(m + 1) where � ∝ k (out) + 1 . This more general form for the attachment probability �(s, t) in Eq. (1) has been used in many related contexts since Price, see Newman 4 for a review. There is an issue about the starting point for this process. The usual form for , the t ≥ s > 1 form in Eq. (1), leaves us with a problem for �(t = 1, s = 1) when looking at the attachment to the second vertex, t = 2 . Our solution is to demand that �(t = 1, s = 1) = 1 . This fixes the cumulative probability ≤ to have a consistent value which is in fact all we need for this calculation. However, we will also assume that the number of nodes and number of edges are given by N(t) = t and E(t) = mt respectively. This is only needed for t ≥ 2 so in principle we must allow multiple edges between nodes starting with m edges added between node 2 and node 1. Again this cannot be true for at least the first node at t = 1. We also note that our analytic calculations allow our networks to contain multiple edges (node pairs linked by more than one edge). Of course, a real citation network and many numerical calculations of this model (though not our numerical calculations) do not have multiple edges. However, in the long time limit the effect of such edges becomes negligible as they form a small fraction of the edge population, a fraction that dies off as a power law in time 4 . Now we would like to define the longest path algebraically. Unfortunately, finding the longest path requires global knowledge of all the paths. This is extremely hard to do algebraically (though is surprisingly straightforward numerically). So the first stage of our calculation is to decide to calculate a path defined with local knowledge only. That is we define what is called a reverse greedy path using an iterative process where at each stage we only need to know about the properties of the next vertex in the path. We will denote the length of the reverse greedy path from the source vertex s = 1 to a target vertex t as ℓ(t) . The length of the longest path from the source vertex to a target vertex t will be denoted as L(t). The reverse greedy path to a node t is a path running from the source node at the initial time t = 1 to node t. This always exists and it is unique. To define it suppose that we have found the reverse greedy path to all earlier nodes. The last step on the reverse greedy path to node t is made along the edge arriving at t from its most recent predecessor node, say s. The idea is that the most recent predecessor of node t, furthest from the source node in terms of the time, is also the most likely to be the predecessor node furthest from the sink node in terms of network path lengths. There is no guarantee that our reverse greedy path is identical to the longest path, so the reverse greedy path length is a lower bound on the longest path length. A more formal definition is given in Appendix A.1 in the Supplementary Information. Of course in any one instance of the Price model, this reverse greedy path length will fluctuate if we look at nodes of similar ages, not least because ℓ is integer valued. We will use a mean field approach so our ℓ(t) is www.nature.com/scientificreports/ an average over many realisations of the model though for simplicity we will not include the expectation value notation . . . . For that reason our ℓ(t) will be a real valued monotonically increasing function of time t. We can find the long-time behaviour using the following simple argument. On average, there are pm edges added with cumulative advantage at each time step. Suppose we are adding a new node at time (t + 1) and we are looking for source nodes s for the m edges we are going to add. The source nodes chosen with cumulative advantage are those with largest degree and those are the oldest nodes created at small values of time s (since k (out) (t, s) ∝ (t − s) p , for instance see Newman 4 ). So the youngest source node chosen, nodes created at the largest value of time s, is likely to be one of the m rnd = mp nodes chosen uniformly at random. The probability that all these m rnd randomly chosen source nodes are chosen between time 1 and time ŝ inclusive is (ŝ/t) m rnd . Suppose we consider the time ŝ 1/2 where with probability one half the time coordinate of the largest randomly chosen source node is ŝ 1/2 or less, then this sets the scale for the birth date of the youngest source node connected to node (t = 1) , namely that ŝ 1/2 = µt where µ = 2 −m rnd . This is the previous node on the reverse greedy path from the initial node to node t. We can then estimate the numbers of steps it takes to get back to the source node at t = 1 as µ ℓ t ≈ 1 which leads to The simplicity of the attachment probability in the Price model means we can also produce more detailed derivation using a mean-field approach. Let the probability that the length of the reverse greedy path, ℓ , from new node (t + 1) to the initial node at t = 1 , be P(ℓ, t) . Then the master equation is of the form Here � max (t, s) is the probability that of the m predecessor nodes connected to a new node at (t + 1) , the oldest of them is s. In terms of the generating function G(z, t) = ∞ ℓ=0 z ℓ P(ℓ, t) we find that the exact solution in the Price model is (see Appendix A.3 in the Supplementary Information for details) Exact forms for the expected reverse greedy path length can be found from this expression, especially for specific small values of m. However, the leading order contribution for large times is always of the form where it is implicit that there is no contribution from the term with the sum for the case of m = 1 . Here ψ(z) is the digamma function and ζ(z) is the Riemann zeta-function. The details of the calculation are given in Appendix A.3 of the Supplementary Information. Finally, the scaling properties of the longest path in the Price model suggests that the properties of height antichains are also very simple. The height of a node in a DAG is the length of the longest path to a node from a source node, any node with zero in-degree. Thus in the Price model, the height of a node is simply the length of the longest path length from the initial node to the given node, our L. Nodes connected by a path cannot be of the same height. Thus the subset of all nodes at the same height form an antichain, a set of nodes in which no two are connected by a path 14 . The scaling properties of these height antichains are simple to estimate if we conjecture that the average longest path L of a node t, its average height, scales as ln(t) . This suggests that if the median index of a node in an antichain of integer valued height h is t mid = (µ) h then the mean index of nodes in the antichain will scale as cosh( √ µ)(µ) h , the variance in the index of nodes in the antichain will be roughly (1/ √ 3) sinh( √ µ)(µ) h , and the number of nodes in the antichain will vary as in the 2 sinh( √ µ)(µ) h . numerical methods and results In the master Eq. (3), multiedges (attaching two edges from the new vertex to the the same vertex) are not excluded. In our numerical implementation code we also allowed multiedges to be created. However the probability of attaching one edge from new vertex (t + 1) to any existing vertex s is decreasing as (s/t) p (for instance see p. 489-90 Newman 4 ). So the creation of a multiedge becomes negligible at large times hence our networks are essentially the same as implementations of the Price model in which multiedges are excluded. The first few steps of the numerical implementation of the Price model have some subtleties which are worth mentioning. The problems noted analytically with the initial node at t = 1 , which is the only node with zeroin-degree, exemplify the issue. The earliest nodes are those with the shortest values of our ℓ and L path lengths to the first node. Since the path lengths of the first few nodes will be added to any other path routed via one of these early nodes, we expect the initial graph to give a constant contribution to the path lengths we measure but not to alter the growth in length scales over long-times. We chose to start our simulations from a complete graph of (2m + 1) nodes, labelled t = 1 to t = (2m + 1) . All pairs of nodes are connected in this initial graph, with the edge direction from earlier to later node. This initial graph ensures that E(t) = mt for all graphs generated numerically, G(t) for t ≥ (2m + 1) . The out-degree www.nature.com/scientificreports/ distribution is fairly even but the in-degree is not fixed to be m for nodes in this initial graph. Further notes on the effect of the initial graph are given in Appendix B.4 in the Supplementary Information. In order to simplify and accelerate the numerical analysis, for each new node (t + 1) added we drew nodes uniformly at random from an "attachment list" which we maintain. After we have chosen the source edges for the m edges attached to the new vertex, we update our attachment list by adding each source node once for every edge, and we add (p/pm) references to the new node. This means we restrict our results to cases where (p/pm) is an integer. Drawing nodes uniformly from our attachment list means that we are choosing vertices according the the probability Eq. (1). For the special case where there was no cumulative advantage, p = 0 , the attachment list was simply a list of the existing vertices where each is referenced once. A more detailed explanation is given in Appendix B.1 of the Supplementary Information. For each node s, we also record values for the lengths of the reverse greedy path ℓ(s) and the longest path L(s) . When adding a new node t, it is simple to look at the values of the lengths of these paths to the m nodes attached to the new node. From that information, it is simple to record the lengths of these paths to the new vertex, ℓ(t) and L(t) . Storing and manipulating these results proved to be more of a limiting factor than the speed to produce them. We produced results for networks of up to 10 8 nodes. The results for these path lengths are quite noisy for any one node as shown by an exemplary run in Fig. 3. Despite the relatively large fluctuations in results for any one node, there is a clear trend in the nodes created at later times. The fluctuations of the path length scaling are greatly reduced when averaged over multiple networks as shown in Fig. 3. So we use 100 runs for each set of parameter values in our work. In order to compare our numerical data with the analytical results we fitted the path lengths found to the function f(t) where The fit was made by using a non-linear fitting routine based on the optimisation of the chi-squared measure of goodness of fit (for instance see 15 ) as described in more detail in Appendix B.3 of the Supplementary Information. Errors on parameters were estimated from the covariance matrix produced by such a method. Given that our analytical work only studies the long-time limit and that the early times in the numerical simulation do not satisfy all the conditions of the analytical work, it is not surprising that in Fig. 4 we still see significant difference between the numerical results and analytical predictions for the length of paths from t = 1 to those nodes created at early times. So when fitting to our numerical data we only use data for nodes created from time t 0 = 1, 000 up to the last node at t = 10 8 . The effect of this cutoff is discussed in Appendix B.3 but we found varying this lower cutoff had little effect on our results since we had so many data points from the region where the asymptotic growth dominates. The dependence of the coefficient of the ln(t) term found from the fit, a, on the model parameters is shown in Figs. 5 and 6. The next-to-leading order coefficient, b of Eq. (6), showed no clear trends. We also considered a non-linear fit with a term of c/t added to the expression in the Eq. (6). We found that in practice, this term had little influence on the remaining parameters of interest, namely, a and b. Furthermore, the errors in c were found to be relatively large in comparison to the errors of the parameters a. Finally, it is clear that the longest path length is scaling as ln(t) to a good approximation. As noted above this then implies that the properties of height antichains in the Price model should follow a regular pattern which depends on the height of nodes in each antichain. Numerical confirmation of these patterns are given in Appendix B.6 of the Supplementary Information. Discussion The numerical results for the leading behaviour of the path length scales are striking. Within the margin of numerical error, our results in Fig. 5 show that the length of the reverse greedy path scales asymptotically as mp ln(t) for a wide range of parameter values. This is consistent with both the simple argument and the detailed analytical calculation presented in the Analytic Results (see also Appendix A of the Supplementary Information). The analytical approach also shows that for long times, the distribution of lengths of reverse greedy paths in the Price model is Poisson distributed with mean equal to mp ln(t) (see Appendix A). The reverse greedy path length is a lower bound on the length of the longest path so it is no surprise that the longest path length also scales as ln(t) with a coefficient, a max , which is larger than the corresponding scaling factor for the reverse greedy path length, a gr . Interestingly this coefficient of the ln(t) term, a max , for the longest path shows some additional weak dependence on the parameters beyond the mp found for the reverse greedy path, as both Figs. 5 and 6 clearly show. The Price model is not in itself a very realistic model for any particular context. For instance, a true citation network often shows many other features such as a preference to cite recent papers, for example see 6,[16][17][18][19][20][21] . The choice of a simple linear form for the attachment probability Eq. (1) appears to be part of this simplification, a form linear in degree motivated by the need for mathematical simplicity. At first sight, this form seems unrealistic No author can know exactly how many citations a paper has let alone the total number of citations in the network. However, this form emerges naturally in many situations as the result of doing local searches on the network, see 6,[21][22][23][24][25][26][27] and references therein. In more realistic models, the cumulative advantage, the p term in emerges from doing a local search back through the current citation network, while the random attachment, the p term in , represents a simple model of other possible processes. So like all good models, the emphasis in the Price model on the linear form for in terms of degree does capture an important and realistic feature of many real situations. This linear form of the attachment probability is also the critical feature in the analysis of undirected versions of this model, such as the Barabási-Albert model 28 where the cumulative advantage aspect is known as preferential attachment and the original example worked with p = 1/2 in our notation. However, the Price model, simple as it is, also emphasises another critical aspect of a citation network, and that is the inherent arrow of time in this context. Citations (almost) always point backwards in time. Typical data sets 3,8 suggest that less than 1% of citations are to documents which are labelled as being published later than the citing document. The networks created in the Price model are realistic in this way, they always produce directed acyclic graphs. This acyclic property is lost when the edge direction is ignored, as in the Barabási-Albert implementations of this model. Since many analyses work in the undirected version, they have missed this key feature of the Price model and of real-world citation networks. For instance, the length of the shortest path between two nodes is a natural measure for undirected networks since in some circumstances it can be related to the geodesic of networks embedded in Euclidean space, for example see 11,12 . For an undirected version of the Price model, the LCD model of Bollobás and Riordan 29,30 (a more precisely defined version of the Barabási-Albert model 28 ), it is known that the diameter, the largest length of any shortest path between two nodes, scales as ln(t)/(ln ln(t)) if m > 1 while the diameter scales as ln(t) for the special case of m = 1 29 (see also theorem 18 of 30 ). For the case m > 1 , Bollobás points out that while in any random graph we expect to see the small-world effect 31 and a ln(N) scaling of lengths (for N nodes in the network), for the undirected version of this model "one might expect the diameter to be even smaller: the unbalanced degree distribution pushes up the number of small paths, and thus, perhaps, pushes the diameter down" (see Bollobás 30 section 13, p 25). That is, the unusual slow scaling of the shortest path distance scale in this undirected version of this model is due to the effect of the very high degree nodes created because of the cumulative advantage (preferential attachment) process. However, the situation is completely different when we take account of the direction of edges in this model. First, the link between shortest path lengths and geodesics in Euclidean space used in Ref. 11,12 is lost. The natural order of nodes in a DAG, the arrow of time, means we should compare the path lengths network against geodesic lengths for network models embedded in Minkowski space, and indeed there is a proven relationship between these two 7,10,32 . Following on from this, when using the longest path in the directed form of these models, our analysis has shown that the longest path is likely to be created by edges created from random attachment not those formed using the cumulative advantage mechanism, the opposite of what is suggested for the shortest path in the undirected form of these models. Thus the fat-tailed nature of the degree distribution in the Price model (or directed versions of the Barabási-Albert/LCD models) is not a factor for the longest path and so, using Bollobás' insight 30 , we should expect the longest path to scale simply as ln(t) , and not something slower than that. That is, indeed what we have shown in our work here. Looking more widely, we note that Bollobás 30 (p 10) suggested that "For these models the orientation is not very interesting". Our conclusion is the opposite. Namely that for any directed network in which vertices are Fig. 6. The ratio of a max /a gr where a is the coefficient of ln(t) in the best fit of the numerical path length data to Eq. (6), a max for the longest path data and a gr for the reverse greedy path data. These values were obtained by fitting the form to nodes created between t = 1, 000 and t = 10 8 from 100 realisations. As a result the errors on the fitted values of a, as estimated from the covariance matrix of the linear fitting algorithm, were smaller than the marker size and so these are not shown. www.nature.com/scientificreports/ added sequentially, the arrow-of-time inherent in these growing network models is both physically relevant and this vertex order produces new and distinctive features. Our analytical and numerical analysis of the longest path length is just one illustration of what is possible.	7,186.2	2019-03-08T00:00:00.000	[ "Mathematics" ]
Hydrocortisone Mitigates Alzheimer’s-Related Cognitive Decline through Modulating Oxidative Stress and Neuroinflammation Alzheimer’s disease (AD), an age-related degenerative disorder, is characterized by β-amyloid deposition, abnormal phosphorylation of tau proteins, synaptic dysfunction, neuroinflammation, and oxidative stress. Despite extensive research, there are no medications or therapeutic interventions to completely treat and reverse AD. Herein, we explore the potential of hydrocortisone (HC), a natural and endogenous glucocorticoid known to have potent anti-inflammatory properties, in an Aβ1–42-induced AD mouse model. Our investigation highlights the beneficial effects of HC administration on cognitive impairment, synaptic function enhancement, and neuronal protection in Aβ1–42-induced AD mice. Notably, HC treatment effectively suppresses the hyperactivation of microglia and astrocytes, leading to a reduction in proinflammatory factors and alleviation of neuroinflammation. Furthermore, HC intervention demonstrates the capacity to mitigate the generation of ROS and oxidative stress. These compelling findings underscore the potential therapeutic application of HC in AD and present promising opportunities for its utilization in AD prevention and treatment. The implications drawn from our findings indicate that hydrocortisone holds promise as a viable candidate for adjunctive use with other anti-AD drugs for the clinical management of patients presenting with moderate to severe AD. Introduction Alzheimer's disease (AD) is a progressive age-related neurodegenerative disease characterized by memory impairment, behavioral dysfunction, and emotional irritability [1]. Pathologically, AD is defined by β-amyloid deposition, abnormal phosphorylation of tau proteins, and the formation of neuronal fibrillary tangles (NFTs), ultimately resulting in synaptic dysfunction and neuronal death [2,3]. With an estimated global prevalence of 50 million patients, AD poses a significant threat to public health, overwhelmingly impacting affected patients, their families, and society at large [4][5][6]. Despite increasingly extensive research, the current treatment landscape remains inadequate, leaving a hugely unfulfilled need as no medications or interventions offer complete remission or reversal of AD. Consequently, the development of effective therapeutic strategies is of utmost urgency in addressing this formidable health challenge. AD is a complex neurodegenerative condition with an uncertain pathogenesis. Numerous mechanistic hypotheses, including the amyloid hypothesis, tau hypothesis, inflammation hypothesis, and cholinergic hypothesis, have been proposed to explain the disease's underlying mechanisms [7][8][9][10][11][12]. As research has progressed, neuroinflammation The SH-SY5Y cells were seeded into 96-well plates and incubated overnight until they reached about 70% confluence. The cells in the control group were cultured in DMEM for 36 h, and the model group were cultured with DMEM for 12 h and treated with 10 µM Aβ 1-42 for 24 h. In the HC treatment group, cells were pretreated with 0.5, 1, and 2 µM HC for 12 h, and then treated with 10 µM Aβ 1-42 for 24 h. Finally, the cell viabilities were detected using the MTT method and the absorbance was measured with a microplate reader (Molecular Devices, SpectraMax Mini, San Jose, CA, USA). Animals and Drug Administration All male C57/BL6 mice (6 weeks old, weight 18-22 g) were purchased from SpePharm Biotechnology Company (Beijing, China). The animal culture and procedures were approved by the Pharmaceutical Laboratory Animal Center of China Pharmaceutical University. Recombinant human Aβ 1-42 peptide (Beyotime Biotechnology, Shanghai, China) was dissolved in sterile PBS at a concentration of 2 mg/mL and incubated for 24 h at 37 • C to obtain aggregated Aβ . All the mice were randomly divided into three groups (n = 8 per group): the control group, the Aβ 1-42 (intracerebroventricular injection, i.c.v.) group, and the Aβ 1-42 (intracerebroventricular injection, i.c.v.) + HC (25 mg/kg, intragastric administration, i.g.) group. The control group was injected with 5 uL saline and the other mice were injected with 5 uL aggregated Aβ 1-42 into the lateral ventricle through the brain stereo-positioning instrument (Harvard). HC (dissolved in saline) was intragastrically administrated after surgery for 14 days. The mice in the model and control group were gavaged with only normal saline. Morris Water Maze (MWM) During days 9-14, the Morris water maze test was performed for cognitive function measurement [38]. The MWM system mainly consisted of a black circular pool, a circular hidden platform which was submerged 1 cm below the water surface, and a video analysis system. The experimental maze was divided into four quadrants and the target platform was placed in the middle of the third quadrant. The experiment lasted 6 days and was conducted in two stages, including the hidden platform training for 5 consecutive days and the probe trial on day 6. In the hidden platform training, the mice were allowed to swim for 90 s to find the target platform which was 1 cm above the water. Each mouse entered the pool from the same position which was opposite the third quadrant. The video analysis system automatically recorded the time when each mouse climbed onto the platform. If the mouse failed to reach the platform within 90 s, it was manually guided to the target and left to rest on the platform for 10 s to remember the position. The probe trial was carried out on the 6th day, and the latency, path length, swimming velocity, target quadrant residence time, and traveled trajectory were recorded and analyzed through the video tracking system. RNA Extraction and Real-Time PCR The mice under deep anesthesia were perfused with frozen PBS (PH = 7.4) transcardially and the brains were collected and stored at −20 • C. The total RNA was purified from the cerebral cortex with Trizol Reagent (Vazyme, Nanjing, China) and reversed-transcribed to cDNA with HiScript III RT SuperMix (Vazyme, Nanjing, China) according to the manufacturer's instructions. The cDNA samples were amplified using the CFX Opus Real-Time PCR System (Bio-Rad, v2.2) using Taq Pro Universal SYBR qPCR Master Mix (Vazyme, Nanjing, China). Relative expression changes were analyzed using the 2 −∆∆Ct method and the target gene expression levels were normalized to GAPDH. The specific primer sequences for RT-PCR are listed in Supplementary Table S1. Determination of ROS Levels SH-SY5Y cells were seeded into 24-well plates and incubated overnight. After modeling and hydrocortisone intervention, the cells were incubated with a 10 µM dihydroethidium (DHE) (Beyotime, Shanghai, China) solution for 30 min at 37 • C, followed by rinsing with PBS 3 times. The brain sections were subjected to a series of prepared protocols. The frozen brain sections were also incubated with 10 µM DHE for 30 min at 37 • C. The images of cells and brains were captured using a microscope (BioTek, Hong Kong, China, Cytation5) and analyzed with Image J software (v1.54f). Western Blot Analyses Western blot analyses was performed as described previously [31]. Cerebral cortex tissues were lysed with RIPA buffer (Beyotime, Shanghai, China) supplemented with phenylmethanesulfonyl fluoride (PMSF). After centrifugation at 12,000 rpm/min at 4 • C for 30 min, the supernatant was collected and a BCA kit (Beyotime, Hong Kong, China) was used for protein quantification. The protein was isolated by sodium dodecyl sulfatepolyacrylamide gel electrophoresis (SDS-PAGE) and transferred onto polyvinylidene difluoride fluoride (PVDF) membranes (Bio-Rad, Hercules, CA, USA). After blocking with 5% skim milk for 1 h at room temperature, the membranes were incubated with the primary antibody (Phospho-NF-kB p65, AffinitY, AF2006, 1:1000) overnight at 4 • C. After washing with TBST 5 times, the membranes were then incubated with the secondary antibodies (HRP-conjugated anti-rabbit IgG) for 1 h. Protein expression was detected with an enhanced chemiluminescence (ECL) method with a gel imaging system (Bio-Rad, ChemiDoc MP, USA). Densities were normalized to GAPDH intensity levels and ImageJ software (v1.54f) was used for quantification. Statistical Analyses All the histograms and line charts were made with GraphPad Prism 8.0. The results are expressed as means ± SEM. At least three biological replicates were performed for all experiments to ensure consistency. Significant differences between two different groups were obtained using an unpaired Student's t-test. A one-way ANOVA and the original FDR method of Benjamini and Hochberg were used to compare multiple independent groups. Statistical significant differences are indicated as * p-value < 0.05. Meta-Analysis in AD Patients Inflammation has been identified as a critical factor in the pathogenesis of Alzheimer's disease (AD), and previous studies have demonstrated that anti-inflammatory treatments targeting neuroinflammation can ameliorate AD symptoms in mice [31]. To delve deeper into the impact of anti-inflammatory treatment strategies on AD symptoms in human patients, a meta-analysis was conducted to assess the potential benefits of anti-inflammatory interventions. Initially, a total of 780 articles were identified, of which 374 duplicates were removed. Following title and abstract screening, 394 articles were excluded, leaving 12 for full-text screening. Ultimately, four articles were included for the meta-analysis ( Figure 1A,B) [37,[40][41][42]. The results of the meta-analysis revealed that anti-inflammatory strategies may offer a certain degree of relief for AD patients (RR; 95% confidence interval [CI], 0.98 to 1.13; p = 0.04). Considering the prominent anti-inflammatory effects of hydrocortisone (HC), this prompted further investigation into its potential as a drug candidate for AD intervention and treatment. These findings underscore the importance of anti-inflammatory approaches in AD management and provide a basis for exploring HC as a potential therapeutic agent for AD in future research. Behavioral and Cognitive Function In Vivo To gain insights into the potential of HC as a potent anti-inflammatory drug in AD treatment, we initiated investigations of its effects in SH-SY5Y cells. The MTT results demonstrated that HC showed no significant cytotoxicity at concentrations of 500, 1000, or 2000 nmol/L, as cell viability remained unchanged within this range ( Figure 2A). When SH-SY5Y cells were exposed to Aβ 1-42 , the cell viability significantly decreased compared to the control group. However, pretreatment with HC at various concentrations for 12 h increased cell viability, indicating a dose-dependent protective effect of HC against Aβ 1-42 -induced damage ( Figure 2B). Furthermore, to confirm the protective effect of HC in AD, we evaluated its impact on cognitive function in Aβ 1-42 -induced mice using the Morris water maze (MWM) behavioral test ( Figure 2C). Both control and Aβ 1-42 -induced mice were orally administered either a vehicle solution or HC for two weeks, followed by behavioral testing for one week. To ensure that the mice did not exhibit motor impairments during the test, we measured average swimming speed and distance traveled on the first day ( Figure 2D,E), which confirmed the absence of movement disorders in all mice. Notably, the Aβ 1-42 -induced mice displayed longer times to locate the hidden platform compared to the control mice, indicative of lower memory levels. In contrast, the HC-treated Aβ 1-42 -induced mice exhibited improved performance, finding the hidden platform faster than the Aβ 1-42induced mice ( Figure 2F). Behavioral and Cognitive Function In Vivo To gain insights into the potential of HC as a potent anti-inflammatory drug in AD treatment, we initiated investigations of its effects in SH-SY5Y cells. The MTT results demonstrated that HC showed no significant cytotoxicity at concentrations of 500, 1000, or 2000 nmol/L, as cell viability remained unchanged within this range (Figure 2A). When SH-SY5Y cells were exposed to Aβ1-42, the cell viability significantly decreased compared to the control group. However, pretreatment with HC at various concentrations for 12 h increased cell viability, indicating a dose-dependent protective effect of HC against Aβ1-42-induced damage ( Figure 2B). Furthermore, to confirm the protective effect of HC in AD, we evaluated its impact on cognitive function in Aβ1-42-induced mice using the Morris water maze (MWM) behavioral test ( Figure 2C). Both control and Aβ1-42-induced mice were orally administered either a vehicle solution or HC for two weeks, followed by behavioral testing for one week. To ensure that the mice did not exhibit motor impairments during the test, we measured average swimming speed and distance traveled on the first day ( Figure 2D,E), which confirmed the absence of movement disorders in all mice. Notably, the Aβ1-42-induced mice Moreover, the HC-treated Aβ 1-42 -induced mice spent more time in the hidden platform quadrant in the probe trial than the Aβ 1-42 -induced mice ( Figure 2G). The HC-treated group also demonstrated a more linear and targeted search strategy compared to the Aβ 1-42 -induced mice ( Figure 2H). Overall, these results indicate that HC ameliorated Aβ 1-42 -induced cell damage and enhanced the behavioral and cognitive functions of the Aβ 1-42 -induced mice, suggesting the potential beneficial role of HC in AD. Synaptic Dysfunction in Aβ 1-42 -Induced Mice Synaptic damage represents a prominent pathological feature of AD. The accumulation of Aβ and the subsequent increase in free radicals contribute to synaptic loss and damage, ultimately leading to neuronal impairment and cognitive deficits [43]. In this context, enhancing synaptic density and improving synaptic function have emerged as potential strategies to prevent and ameliorate cognitive impairment in AD [44]. To investigate the potential benefits of HC in alleviating synaptic damage, we examined the mRNA expression levels of synaptic-function-related factors in the cortexes of mice brains. Notably, the mRNA expression of GluA1 and GluA2, two subunits of the α-amino-3-hydroxy-5methyl-4-isoxazolepropionic acid (AMPA) receptor involved in neural signaling; CaMdependent protein kinase IIα (CamKIIα); CaM-dependent protein kinase IIβ (CamKIIβ), critical for synaptic plasticity; and synaptophysin (SYP), a synaptic vesicle protein associated with synaptic remodeling, were significantly down-regulated in the Aβ 1-42 -induced mice. However, HC treatment significantly increased the mRNA expression of GluA1, GluA2, CaMKIIα, CamKIIβ, and SYP in vivo [45][46][47]. Moreover, postsynaptic density protein-95 (PSD95), encoded by Dlg4, indicates synaptic function [48]. The HC treatment substantially increased the mRNA expression level of Dlg4, and the immunofluorescence assay indicated that the expression levels of PSD95 were also significantly increased in the cerebral cortexes and hippocampi of mice following the HC treatment ( Figure 3F,G). These observations and data underscore the critical role of HC in repairing synaptic dysfunction and enhancing synaptic function, pointing towards its potential therapeutic significance in AD treatment. The beneficial effects of HC on synaptic integrity may hold promise for promoting cognitive improvements and ultimately contributing to the management of AD pathology. Cells 2023, 12, x FOR PEER REVIEW 7 displayed longer times to locate the hidden platform compared to the control mice, ative of lower memory levels. In contrast, the HC-treated Aβ1-42-induced mice exh improved performance, finding the hidden platform faster than the Aβ1-42-induced ( Figure 2F). Moreover, the HC-treated Aβ1-42-induced mice spent more time in the hidden form quadrant in the probe trial than the Aβ1-42-induced mice ( Figure 2G). The HC-tr group also demonstrated a more linear and targeted search strategy compared to th 42-induced mice ( Figure 2H). Overall, these results indicate that HC ameliorated A induced cell damage and enhanced the behavioral and cognitive functions of the A induced mice, suggesting the potential beneficial role of HC in AD. The Inflammatory Responses in Aβ1-42-Induced Mice Neuroinflammation mediated by microglia and reactive astrocytes plays a significant role in promoting the pathological process of AD, leading to synaptic dysfunction and cognitive impairment [49]. In order to evaluate the anti-inflammatory effect of HC in Aβ1- 42-induced mice, we examined the mRNA levels of various proinflammatory factors. Re- The Inflammatory Responses in Aβ 1-42 -Induced Mice Neuroinflammation mediated by microglia and reactive astrocytes plays a significant role in promoting the pathological process of AD, leading to synaptic dysfunction and cognitive impairment [49]. In order to evaluate the anti-inflammatory effect of HC in Aβ 1-42 -induced mice, we examined the mRNA levels of various proinflammatory factors. Results from the RT-PCR assay demonstrated that the HC treatment significantly decreased mRNA levels of pro-inflammatory factors, including TNF-α, MCP-1, IL-1β, and IL-6 when compared to the Aβ 1-42 -induced mice ( Figure 4A-D). The immunofluorescence analysis of IBA1 and GFAP, markers for microglia and reactive astrocytes, respectively, revealed that Aβ 1-42 induced excessive activation of microglia and astrocytes in different hippocampal regions such as DG, CA1, CA2, and CA3. However, the HC intervention effectively reduced the number of activated microglia and astrocytes ( Figure 4E-H). Additionally, NF-κB activation, an inflammatory transcription factor, plays a crucial role in neuroinflammation and the accumulation of Aβ plaques, contributing to AD pathogenesis [50]. Our results indicated that the HC treatment effectively blocked the Aβ 1-42 induction's function in increasing protein levels of phosphorylated NF-κB ( Figure S1). Collectively, these data strongly suggest that HC may prevent the activation of microglia and astrocytes, exerting anti-inflammatory effects, and thereby safeguarding synaptic function and ameliorating cognitive impairment. These findings highlight the potential of HC as a therapeutic agent to target neuroinflammation in AD and underscore its significance in modulating the intricate pathogenesis of the disease. Oxidative Stress and Neuroprotection Similar to neuroinflammation, multi-factor-induced oxidative stress is also considered a core pathogenesis of AD [26]. Chronic oxidative stress can lead to impaired nerve cell function and an enhanced inflammatory response [51]. To investigate the potential role of HC in ameliorating oxidative stress, we examined the expression of ROS after HC treatment both in vitro and in vivo. Our findings demonstrated that the ROS levels in SH-SY5Y cells significantly increased after Aβ 1-42 induction, and the HC treatment effectively reduced the ROS levels in a dose-dependent manner ( Figure 5A,B). Similarly, the ROS fluorescent signal was also diminished in the hippocampi and cortexes of the HC-treated Aβ 1-42 -induced mice, indicating less oxidative stress injury ( Figure 5C-E). Furthermore, to examine the neuroprotective effect of HC, we assessed the mRNA levels of the neurotrophic factors BDNF and GDNF, which promote neuronal survival and repair neuronal damage. Following the HC treatment, the mRNA levels of BDNF and GDNF were significantly increased ( Figure 5F,G), further supporting the benefits of HC in inhibiting oxidative stress and protecting neurons. Our results demonstrate that HC treatment can ameliorate synaptic dysfunction and protect neurons by inhibiting neuroinflammation and oxidative stress, ultimately leading to improved learning and cognitive function in Aβ 1-42 -induced mice ( Figure 6). These findings highlight the potential therapeutic value of HC in mitigating the multifaceted pathogenesis of AD and offer promising avenues for future research and clinical intervention. in increasing protein levels of phosphorylated NF-κB ( Figure S1). Collectively, these data strongly suggest that HC may prevent the activation of microglia and astrocytes, exerting anti-inflammatory effects, and thereby safeguarding synaptic function and ameliorating cognitive impairment. These findings highlight the potential of HC as a therapeutic agent to target neuroinflammation in AD and underscore its significance in modulating the intricate pathogenesis of the disease. icantly increased ( Figure 5F,G), further supporting the benefits of HC in inhibiting ox tive stress and protecting neurons. Our results demonstrate that HC treatment can ameliorate synaptic dysfunction protect neurons by inhibiting neuroinflammation and oxidative stress, ultimately lead to improved learning and cognitive function in Aβ1-42-induced mice ( Figure 6). These fi ings highlight the potential therapeutic value of HC in mitigating the multifaceted pa genesis of AD and offer promising avenues for future research and clinical interventi Discussion Alzheimer's disease (AD) remains a significant global public health challenge, with Discussion Alzheimer's disease (AD) remains a significant global public health challenge, with its pathogenesis attributed to various factors, including Aβ senile plaque deposition, hyperphosphorylation of tau proteins, and neuroinflammation [52]. However, the existing medications for AD merely offer temporary symptomatic relief without achieving a complete reversal of the underlying pathological processes. Notably, the failure of singletargeted anti-AD drugs has led to a shift towards exploring multi-targeted drugs or drug combinations as crucial alternatives for AD treatment. Particularly, the combination of hormonal anti-inflammatory drugs with other therapeutic agents shows promising potential as a treatment option for future clinical AD patients. Several therapeutic approaches have been developed, focusing on the inhibition of activated microglia and astrocytes, which significantly contribute to the exacerbation of neuroinflammation [53]. Our meta-analysis reinforces the significance of anti-inflammatory drugs as a promising avenue for AD treatment. In conclusion, the pursuit of multi-targeted therapeutic strategies and combinations may offer new hope in effectively addressing the complexities of AD pathogenesis and improving patient outcomes in the future. HC, an essential natural glucocorticoid and potent anti-inflammatory agent, holds considerable significance in various conditions, including rheumatoid arthritis, allergic diseases, severe infections, and shock. Moreover, HC is a suitable candidate for inhibiting inflammatory responses within the brain. Studies have demonstrated the beneficial impact of HC on learning and memory function and its potential application in treating central nervous system injury by targeting microglia and astrocytes has been explored [54]. Despite these promising findings, the specific therapeutic effect of HC on AD remains uncertain and necessitates further investigation. As a potential avenue for AD treatment, understanding HC's role in disease pathology and its interactions with neuroinflammatory processes may pave the way for novel therapeutic strategies in managing AD and improving cognition. Our results provide support for the neuroprotective effect of HC in vitro, as demonstrated by its non-cytotoxic nature over a wide concentration range and its ability to alleviate Aβ 1-42 -induced injury in SH-SY5Y cells. Our investigation further revealed that HC exerted a considerable inhibitory effect on the pro-inflammatory cytokines induced by Aβ 1-42 in mice. The immunofluorescence assay confirmed HC's ability to impede the activation of microglia and astrocytes, which play critical roles in neuroinflammation. Our findings revealed HC's ability to attenuate phosphorylated NF-κB protein levels ( Figure S1). The NF-κB signaling pathway has been recognized as a compelling therapeutic target for AD, given its critical role in promoting microglial and astrocytic activation and consequent neuroinflammation [39,55,56]. The accumulation of Aβ triggers the activation of NF-κB, instigating an inflammatory response, while pro-inflammatory factors further increase NF-κB levels, creating a detrimental cycle. In light of these observations, our study provides valuable insights into HC's potential role in AD pathogenesis by inhibiting microglia and astrocyte activation via the NF-κB signaling pathway, ultimately modulating neuroinflammation. These findings hold promise for advancing our understanding of HC's therapeutic potential in mitigating AD-associated neuroinflammatory processes and offer potential avenues for future therapeutic interventions. Oxidative stress is a pivotal factor in the pathogenesis of Alzheimer's disease (AD), as it plays a critical role in bridging multiple pathways [26]. Reactive oxygen species (ROS) serve as a prominent characteristic of oxidative stress, exerting detrimental effects on synaptic function and neurotransmission, thereby leading to neurodegeneration and cognitive impairment [57]. Additionally, ROS further triggers microglia and astrocytes to release pro-inflammatory factors, promoting neuroinflammation, which reciprocally amplifies ROS levels [58]. The interplay between ROS and neuroinflammation results in neuronal dysfunction and cell death. Consequently, reducing ROS levels represents a critical avenue for intervening in AD and has garnered extensive investigation. In our study, ROS levels were assessed in Aβ 1-42 -induced SH-SY5Y cells in vitro and in an acute AD mouse model, revealing a decrease in ROS levels following the HC treatment. Furthermore, the HC treatment was observed to significantly increase the levels of neurotrophic factors and synapse function-related proteins. These findings suggest that HC may confer neuroprotection by reducing ROS levels, thereby mitigating oxidative stress-induced injury and presenting a potential therapeutic strategy for AD treatment. The modulation of ROS levels and its interaction with neuroinflammatory processes through HC administration hold significant implications for advancing AD research and fostering the development of novel therapeutic interventions. The complexity and diversity of the pathological mechanisms involved in Alzheimer's disease (AD) have highlighted the limitations of single-target medications in meeting therapeutic needs. Consequently, there has been an increasing emphasis on multi-target and combinatorial medications as promising approaches. Notably, the combination of memantine with cholinesterase inhibitors like donepezil and galantamine has gained widespread recognition for its substantial cognitive benefits [59,60]. Additionally, the combination of dexamethasone with acyclovir, an anti-herpes virus drug, has demonstrated efficacy in improving cognitive impairment induced by Aβ 1-42 in mice [30]. Given these promising findings, the prospect of combining lower hydrocortisone (HC) doses with anti-AD drugs, such as donepezil, emerges as a viable approach. This combination reduces the required dose of donepezil and acts as a synergistic agent, potentially mitigating adverse effects and reducing overall toxicity. While our study has yielded promising results regarding the neuroprotective effects of hydrocortisone (HC) in inhibiting neuroinflammation and oxidative stress, important limitations warrant further investigation. Notably, several studies have indicated that endogenous HC levels are elevated in the brains of AD patients, and excessive HC doses have been linked to hippocampal toxicity [61]. Considering the complex effects of HC, our current study has solely examined the high-dose effects of HC on cells, necessitating additional in vivo data to substantiate these findings. Our research has been limited to cell and animal models, warranting further exploration through clinical studies to establish robust references for future applications. Conclusions Our study underscores the potential of HC in protecting synapses and neurons through its capacity to inhibit neuroinflammation and oxidative stress, leading to improvements in cognitive and learning deficits in Aβ 1-42 -induced mice. These findings suggest that HC may hold promise in AD applications and provide encouraging prospects for future preventive and therapeutic approaches in AD management. By addressing the limitations and exploring the translational potential of HC in clinical settings, we can advance our understanding of its therapeutic efficacy and optimize its usage for the benefit of AD patients. The combined efforts of further research and clinical evaluation will pave the way for potential AD prevention and treatment strategy advancements.	5,811.4	2023-09-25T00:00:00.000	[ "Medicine", "Environmental Science", "Biology" ]
Robust Cell Size Checkpoint from Spatiotemporal Positive Feedback Loop in Fission Yeast Cells must maintain appropriate cell size during proliferation. Size control may be regulated by a size checkpoint that couples cell size to cell division. Biological experimental data suggests that the cell size is coupled to the cell cycle in two ways: the rates of protein synthesis and the cell polarity protein kinase Pom1 provide spatial information that is used to regulate mitosis inhibitor Wee1. Here a mathematical model involving these spatiotemporal regulations was developed and used to explore the mechanisms underlying the size checkpoint in fission yeast. Bifurcation analysis shows that when the spatiotemporal regulation is coupled to the positive feedback loops (active Cdc2 promotes its activator, Cdc25, and suppress its inhibitor, Wee1), the mitosis-promoting factor (MPF) exhibits a bistable steady-state relationship with the cell size. The switch-like response from the positive feedback loops naturally generates the cell size checkpoint. Further analysis indicated that the spatial regulation provided by Pom1 enhances the robustness of the size checkpoint in fission yeast. This was consistent with experimental data. Introduction In order to maintain proper size, dividing cells need to time mitosis carefully. Previous analyses performed in fission yeast suggested that there is a homeostatic mechanism that can maintain the appropriate cell size [1][2][3]. The cell is allowed to enter mitosis only after it reaches a critical size (size checkpoint). Experimental data also showed that cells smaller than critical size had to grow until they reached the threshold value. This period is called the size-dependent phase, or sizer. Then, after a fixed period, called as timer, the cells completed mitosis. Daughter cells that are larger than critical size when produced can undergo mitosis without going through the sizer phase [3]. Some higher eukaryotes such as Xenopus laevis [4,5], Drosophila [6], animal cells [7], and HeLa cells [8] also have similar methods of size control. Biological experimental data indicate that the rate of cyclin protein synthesis may increase as the cell grows [9]. This may be one mechanism underlying size control. Previous mathematical models have explored the nonlinear dynamic properties of the temporal regulation of cell cycle events [10][11][12]. The cyclin protein synthesis rate is assumed to increase as the cell grows, and it exhibits a bistable relationship with MPF. This bistability, which is generated by the positive feedback loops in the cell cycle, is responsible for the mitosis initialization [9,13]. In this way, cell size is linked to entry into mitosis. Recent evidence has shown that the cell polarity protein kinase Pom1 forms a polar gradient from the ends of the cylindrical cell to its center [14,15]. In this way, it can provide spatial information that can be used to regulate the mitosis inhibitor Wee1. This spatial regulation links cell size directly to mitosis, and it may play a critical role in size control. In summary, cell size is coupled to the progression of the cell cycle through the rates of synthesis of cyclin proteins and the direct spatial information provided by Pom1. The results of the present study show that when spatial regulation and the rate of synthesis rate are both coupled to temporal positive feedback loops, a bistable response generates the cell size checkpoint. Bifurcation analysis shows that the concentration of MPF can exhibit a bistable steady-state relationship with the rate of synthesis of cyclin proteins or the concentration of Cdr (downstream of Pom1) alone. The size checkpoint is naturally built into the system in the form of dual regulations of the rate of synthesis and the Pom1 gradient. Stochastic analysis then showed that the direct spatial regulation can allow temporal positive feedback to enhance the robustness of the cell size checkpoint in fission yeast, which is consistent with the experimental data. The rate of synthesis of Cdc13 increases as the cell grows ( Figure 1(b)). Besides, Cdr proteins also couple cell growth to cell division through a size sensing mechanism involving Pom1. Several previous works have identified the function of the Pom1 pathway [14][15][16]. The cell polarity protein kinase Pom1 is a cell polarity protein kinase, which can form a spatial gradient that is greatest at the ends of the cylindrical cell and least in the middle of the cell. Cdr which locates near the center of the cell can suppress the activity of Wee1 and so promote mitosis. Pom1 phosphorylates Cdr to inhibit its activity. The size-dependent relief of this inhibition can repress Wee1 to promote the initialization of mitosis. Results and Discussion The network was then transferred into a set of ordinary differential equations using the principles of biochemical kinetics. The initial size of a WT daughter cell was normalized to 1. The model was adapted from the models constructed by James Ferrell's group [17] and Novak-Tyson's group [11]. However, different from their models, we also took the spatial information provided by Pom1 into consideration. A detailed mathematical model is presented in Section 4. Bifurcation Analysis. Experimental observations have provided some evidences of size checkpoint [2]. If the initial size of a cdc2-33 fission yeast cell was smaller than 12 m, a marked negative relationship was observed between the extension length and the initial size. However, the extension length was not found to be significantly related to cell size at initial sizes larger than 12 m. This critical size that determined whether the cell could begin mitosis was the size checkpoint. Besides, Rupeš and colleagues also showed that cells smaller than critical size had to grow until they reached the threshold value. If the birth size of the fission yeast is larger than the critical size, the cell can undergo mitosis without additional time delay [3]. This critical size also indicates the existence of the size checkpoint. Earlier experimental studies have revealed that the steady state of MPF shows a hysteretic steady-state response relationship with the concentration of cyclin B [13]. Mathematical models have established that the concentration of MPF has a bistable relationship with the rate of synthesis of the cyclin proteins. This bistability is attributable to the positive feedback loops (active Cdc2 promotes its activator Cdc25 and suppresses its inhibitor Wee1) [18,19]. In our model, the positive feedback loops rely both on the rate of synthesis and on the spatial regulation involving Wee1. In this way, the coupling between cell size and cell division is more realistic in this model. To demonstrate how the rate of synthesis of Cdc13 and the concentration of Cdr both affect the activation of MPF, we first calculated the steady state of MPF for a given rate of Cdc13 synthesis and a given concentration of Cdr, when Cdc13 was made nondegradable (during the period prior to mitosis, the concentration of APC remains at a low and constant level) [20]. The results are presented in three-dimensional space (the green surfaces in Figures 2(a), 2(c), and 2(e)). If the regulation related to the rate of synthesis is solely considered, then the vertical plane Cdr = 0.2 (which represents a fixed concentration of Cdr, which occurs when Pom1 spatial regulation is blocked) intersects with the surface at an S-shaped curve ( Figure 2(a)). The bifurcation analysis shows that the steady state of MPF has a bistable relationship with the rate of synthesis (Figure 2(b)). As the cell grows, the rate of synthesis of Cdc13 increases and the concentration of MPF accumulates in turn. When the rate of synthesis passes point K2 in Figure 2(b) as the cell grows, the low stable branch disappears and the MPF has to jump to the upper stable branch (arrow (1)). And the mitosis begins. In this way, the rate of synthesis of Cdc13 contributes to the function of the size checkpoint. Similarly, if the regulation related to the Pom1 pathway is solely considered, the vertical plane synthesis rate = 0.009 (which represents a fixed synthesis rate of Cdc13) intersects with the bent surface along an S-shaped line ( Figure 2(c)). The bifurcation analysis shows that the steady state of MPF also has a bistable relationship with Cdr (the downstream of Pom1) when the direct spatial regulation is linked to the positive feedback loops (Figure 2(d)). As the cell grows, the concentration of Cdr increases (due to a reduction in regulation provided by Pom1). Then the concentration of MPF also accumulates along the lower branch. When the concentration of Cdr passes point C2 as the cell grows ( Figure 2(d)), the low stable branch disappears and MPF has to jump to the upper stable branch (arrow (2)). Then mitosis begins. In this way, the direct spatial regulation provided by the Pom1 pathway also contributes to the function of the size checkpoint. In the real-world cell cycle, these two previous regulations both contribute to the coupling of the cell size and cell division. The steady state of MPF in real-world systems was assessed as follows. First, the relationship between the rate of synthesis of Cdc13 and the concentration of Cdr as the cell grows was calculated. Then the vertical surface, which represents the variation in the rate of synthesis rate and the concentration of Cdr as the cell grows, was intersected with the steady state surface (Figure 2(e)). As in the sole regulation scenarios, the line of intersection is S-shaped. This means that the steady state of MPF continues to exhibit bistability with cell size when the spatial regulation and the rate of synthesis are both involved in the positive feedback loops. After that we directly linked the steady state of MPF to the cell size through bifurcation analysis (Figure 2(f)). Figure 2(f) shows that the concentration of MPF increases as the cell grows. When the cell size reaches the threshold, about 1.5 (size checkpoint, S2 in Figure 2(f)), MPF switches to the upper branch (arrow (3)). Then the cell undergoes mitosis. Then we further summarized the relationship between the bifurcation analysis and the size checkpoint. The bifurcation analysis shows that MPF exhibits the bistability with the cell size. There is a critical cell size S2 (corresponding to saddle node point SN2): if the cell size is smaller than S2, MPF stays in low level; if the cell size passes S2 point, the low stable branch disappears and MPF has to jump to the upper stable branch. As we mentioned previously, experimental studies [1,2,21,22] have shown that a cell will not begin mitosis until it grows to a critical size. Therefore, this saddle node point SN2 naturally performs the role of a check point: before the size reaches S2, cell remains in G2 state (low MPF); once the cell passes S2 point, MPF can jump to the upper branch to trigger mitosis. After that, a numerical simulation was used to check the size checkpoint ( Figure 3). The initial size of the model varied from 0.25 to 4. The result shows that when the initial cell size is smaller than 1.5, the cycle time shows a significantly negative relationship with the initial size. However, if the initial size exceeded 1.5, then the cycle time was mostly independent of the initial cell size. This result accords with the previous experimental data in yeast (the inserted figure in Figure 3 [2]). In this way, 1.5 is established as the size checkpoint. The result of the simulation is also consistent with the bifurcation analysis shown in Figure 2(f), where the critical size for the mitosis initialization is about 1.5. In summary, the concentration of MPF exhibits a bistable steady-state relationship with cell size, which depends on the spatiotemporal positive feedbacks. This bistability naturally produces the size checkpoint. Stochastic Analysis. Experimental evidence showed that some intrinsic stochastic noise (such as random cell production and collisions between molecules) and extrinsic stochastic noise (such as variations in the environment) will result in fluctuations in gene expression [23]. In this way, processes related to the cell cycle may vary from cell to cell within a population, over time, and even within a single cell. The present study not only coupled cell size to the rate of synthesis of Cdc13 but also to the direct spatial regulation provided by the Pom1 pathway. This direct spatial regulation may help the size checkpoint resist interference from different sources and keep cell size coupled to cell division. To evaluate the impact of random fluctuation on the cell cycle, some stochastic noise was introduced to the present model: (1) each parameter in the deterministic model was multiplied by a stochastic factor, which was randomly chosen from the normal distribution with = 1 and = 0.016 ( represents the mean value and represents the variance of the distribution). In this way, the cell cycle can fluctuate near the deterministic value. into two identical daughter cells. Asymmetrical division was characterized by a normal distribution with = 0.5 and = 0.016. Then the model was used to determine if the spatial regulation can help the size checkpoint resist the fluctuations of the system. When the stochastic factor was disturbed, the size check point was calculated 100 times with and without the Pom1 spatial regulation. Results are shown in Figure 4. the presence of stochastic noise. However, if spatial regulation is taking place in the system, the size checkpoint changes from 1.48 to 1.56, which is much narrower than the one shown in Figure 4(a). In this way, even with the interference produced by stochastic noise, the cell must still exceed a strict size checkpoint. The comparison indicates that the direct spatial regulation provided by the Pom1 pathway can ensure tight coupling between cell size and the cell division. Because direct spatial regulation works through the mitosis inhibitor Wee1, wee1Δ may show a weak ability to resist interference. This mathematical model was used to assess the size checkpoint in wee1Δ ( 5 = 0.15, decreases from 2 in WT, the stochastic noise presents as mentioned previously). Bifurcation analysis shows the size at which cells undergo mitosis in wee1Δ to be about half of that in WT (Figure 5(a)). This result accords with the previous experimental data that the wee1Δ cell divided at a half size of WT [2]. However, in the presence of random disturbances, the width of the size checkpoint in wee1Δ was found to be twice of that of WT. These theoretical results are consistent with observations made in earlier experiments [2]. Experimental observation showed that wee1Δ fission yeast exhibited larger variance in the duration of the cell cycle for any given initial size (Figure 1(b) of a previous study) [2]. Because the duration of the cell cycle includes the time required for the cell to reach the size checkpoint (sizer phase) and the fixed time, which is independent of other factors (timer phase) [24]. This indicates that the variation of the size checkpoint is larger in wee1Δ than in WT. Furthermore, Table 1 in [2] summarized that the variation in the division length was about twice as large in wee1Δ as in WT, and the variation in cycle time was increased in the similar way. Therefore, it means that the size control in wee1Δ is not as strict as that in WT. In this way, the size checkpoint in wee1Δ is less robust than in WT. Then the numerical simulation results of wee1Δ and WT fission yeast were compared to those produced in earlier experiments. During the simulation, stochastic factors continued to act on the cell. And the initial concentrations of proteins and initial size were given in the deterministic model. After every division, the system listed the initial cell size and initial concentrations of relevant proteins for the next cell cycle. The results of the simulation are shown in Figure 5(b): the overall range of the duration of the cell cycle was similar in wee1Δ and WT. However, for a given initial size, the range of the duration of the cell cycle was always larger in wee1Δ than in WT. The large variation in the length of the cycle time can be attributed to weakness in the size checkpoint control. These results are consistent with those of a previous experiment published by Novak and Tyson [2]. In conclusion, the direct spatial regulation provided by Pom1 can enhance the robustness of the size checkpoint and couple cell size to cell division. Discussion Although a group of models have investigated the temporal regulation of cell cycle [10,[25][26][27], most of them did not consider the direct spatial regulation provided by Pom1. In our model, we take this regulation into account. Vilela and colleagues built a mathematical model incorporating the Pom1 pathway [16]. However, they paid more attention on the formation of the Pom1 gradient and overpassed the link between bistability and size checkpoint. In our model, we specified that the critical size S2 (corresponding to saddle node of the lower branch SN2) is the cell size checkpoint and focused on the robustness of the size checkpoint. Since the function of the Pom1 pathway has not been understood until 2009, our previous work related the size checkpoint to the cytoplasmic-to-nuclear size ratio. In the present work, we revealed that the underlying mechanism of size checkpoint is the saddle node bifurcation. Bifurcation analysis of wee1Δ ( Figure 5(a)) showed that although the range of the cell size checkpoint is larger than that of WT, it is still narrower than that of systems without spatial regulation (Figure 4(a)). This is because the spatial regulation is still assumed to work in wee1Δ ( 5 remains 7.5% of WT, not 0 in wee1Δ). Even this weak spatial regulation can enhance the robustness of size checkpoint significantly. Therefore, the direct spatial regulation provided by Pom1 is thought to play a more important role in coupling cell growth to cell division. Cell size checkpoints are present in many kinds of cells. A robust cell size checkpoint is required for the maintenance of appropriate cell size during proliferation. Although only spatial regulation was reflected in the present model of fission yeast, other cells, such as frog eggs, may also have similar ways of transferring spatial information directly, but this has not been experimentally established. Unlike that of fission yeast, the spatial regulation of oocytes takes place in a spherical space [12]. In the present study, a mathematical model was used to investigate the manner in which cell size can be coupled to the cell division in fission yeast. As the cells grow, the rate of synthesis of Cdc13 increases. However, the relief gradient offered by Pom1 can reduce the concentration of Cdr, which reduces the ability of Cdr to inhibit Wee1. The novel dynamics shown in the present model can be used to evaluate the direct spatial regulation provided by Pom1 and to examine its impact on cell checkpoints. The positive feedback loops were found to depend on spatial regulation and to generate a switch-like MPF response, which naturally produces the endogenetic size checkpoint. This direct spatial relation was found to protect the size checkpoint from fluctuations in gene expression. Methods and Materials The mathematical models of the cell cycle have been extensively studied. We adapted the parameters from the classic models of Ferrell's and Tyson's group. Besides, we also added the effect of the Pom1 pathway on Wee1 regulation. In other words, we introduced the spatial regulation to the system. The ordinary differential equations for this mathematical model are as follows: Cdc25 ( ) = 13 * MPF( ) 14 rel = * 7, 1 = exp (− * (rel /2)) exp ( * (rel /2)) + exp (− * (rel /2)) , 2 = exp ( * (rel /2)) exp ( * (rel /2)) + exp (− * (rel /2)) , Cdr = cdr on cdr on + cdr off * pom1 / (pom1 + ) In our model, the change rate of Pom1 at location is given by a kinetic equation, where the first term is the rate of diffusion and the second term is its rate of degradation. Here represents the diffusion coefficient of Pom1 in fission yeast. And represents the degradation coefficient of Pom1. Here rel represents the amplified size of the fission yeast. These parameters are estimated from the model of Vilela and colleagues [16]. It is notable that the birth size of the fission yeast is normalized to 1 in our model. However, the birth size of fission yeast is 7 m in Vilela's model. Therefore, we amplified the cell size to 7-fold when applying the parameters of Vilela's model: 8 BioMed Research International The activity of Cdr is repressed by the Pom1 at the central zone of the fission yeast. Thus we only need to consider the concentration of Pom1 at the center of the fission yeast: pom1 = 1 * exp ( * rel 2 ) + 2 * exp (− * rel 2 ) . Similarly, to simplify the model, we assumed that the concentration of Cdr is set in steady state. Thus the ordinary equation can be transformed into an algebraic equation: Cdr = cdr on cdr on + cdr off * pom1 / (pom1 + ) . (20) The parameters are as follows:	4,891.6	2013-07-11T00:00:00.000	[ "Biology", "Mathematics" ]
Transitional Channel Flow: A Minimal Stochastic Model In line with Pomeau’s conjecture about the relevance of directed percolation (DP) to turbulence onset/decay in wall-bounded flows, we propose a minimal stochastic model dedicated to the interpretation of the spatially intermittent regimes observed in channel flow before its return to laminar flow. Numerical simulations show that a regime with bands obliquely drifting in two stream-wise symmetrical directions bifurcates into an asymmetrical regime, before ultimately decaying to laminar flow. The model is expressed in terms of a probabilistic cellular automaton of evolving von Neumann neighborhoods with probabilities educed from a close examination of simulation results. It implements band propagation and the two main local processes: longitudinal splitting involving bands with the same orientation, and transversal splitting giving birth to a daughter band with an orientation opposite to that of its mother. The ultimate decay stage observed to display one-dimensional DP properties in a two-dimensional geometry is interpreted as resulting from the irrelevance of lateral spreading in the single-orientation regime. The model also reproduces the bifurcation restoring the symmetry upon variation of the probability attached to transversal splitting, which opens the way to a study of the critical properties of that bifurcation, in analogy with thermodynamic phase transitions. Context How laminar flow becomes turbulent, or the reverse, when the shearing rate changes, is a problem of great conceptual interest and practical importance. This special issue is focused on the case when the transition is characterized by the fluctuating coexistence of domains either laminar or turbulent in physical space at a given Reynolds number Re (control parameter), a regime called spatiotemporal intermittency, relevant to wall-bounded flows in particular. Several years ago, Y. Pomeau [1] placed that problem in the realm of statistical physics by proposing its approach in terms of a non-equilibrium phase transition called directed percolation (DP). This process displays specific statistical properties defining a universality class liable to characterize systems with two competing local states, one active, the other absorbing, with remarkably simple dynamical rules: any active site may contaminate a neighbor and/or decay into the absorbing state, and an absorbing state cannot give rise to any activity [2]. The coexistence is regulated by the contamination probability, and a critical point can be defined above which the mixture of active and absorbing states is sustained and below which the active state recedes, leaving room for a globally absorbing state. The fraction of active sites is a measure of the global status of the system. The subcritical context typical of wall-bounded flows, initially pointed out by Pomeau, seems an interesting testbed for universality [3,4]. Here, turbulence plays the role of the active state and laminar flow, being linearly stable, represents the absorbing state. DP has indeed been shown relevant to simple shear between parallel plates (Couette flow) [5] and its stress-free version (Waleffe flow) [6]. The most recent contributions to the field can be found in [7]. In this paper we will be interested in plane channel flow (also called plane Poiseuille flow), the flow driven by a pressure gradient between two parallel plane plates, which is not fully understood despite recent advances. In this context, universal properties are notably difficult to extract from experiments, since they relate to the thermodynamic limits of asymptotically large systems in the long time limit, whereas what plays the role of microscopic scales involves already macroscopic agents, e.g., roll structures in convection or turbulent streaks in open flows, and the turnover time associated with such structures. However, universality focuses on quantitative aspects of systems sharing the same qualitative characteristics, in particular symmetries and the effective space dimension D in which these systems evolve. Delicate questions can thus be attacked by modeling attempts that implement these traits appropriately. This approach involves simplifications from the primitive equations governing the problem, here the Navier-Stokes equations, to low-order differential models implementing the building blocks of the dynamics [8], to coupled map lattices (CML) in which the evolution is rendered by maps and space is discretized [9,10], to cellular automata for which local state variables are also discretized, and ultimately to probabilistic cellular automata (PCA), where the evolution rule itself becomes stochastic [11]. The absence of a rigorous theoretical method supporting the passage from one modeling level to the next, such as multi-scale expansions or Galerkin approximations, makes the simplification rely on careful empirical observations of the case under study, which somehow comes and limits the breadth of the conclusions drawn. Physical Context: Plane Channel Flow Of interest here, the transitional range of plane channel flow displays a remarkable series of steps at decreasing Re from large values where a regime of featureless turbulence prevails. It has been the subject of numerous studies and references to them can be found in the article by Kashyap, Duguet, and Dauchot in this special issue [12]; see also [13]. Our own observations based on numerical simulations are described in [14,15] and summarized in Figure 1. [14]. Re g ≈ 700. Transversal splitting sets in at Re ∼ 800 (event A). The extrapolated 2D-DP threshold is Re DP 984. The "one-sided → two-sided" transition takes place at Re 2 1011. localized turbulent bands (LTBs) exist up to Re ≈ 1200 (event B), beyond which a continuous laminar-turbulent oblique pattern prevails up to the threshold for featureless turbulence Re t ≈ 3900. The Reynolds number used to characterize the flow regime is defined as Re = U c h/ν, where 2h is the gap between the plates, U c is the mid-gap stream-wise speed of a supposedly laminar flow under the considered pressure gradient, and ν the kinematic viscosity. This definition using U c is appropriate for our numerical simulations under constant pressure-gradient driving. Other definitions involve the friction velocity U τ , or the stream-wise speed averaged over the gap U b . They are related either empirically, vis., U b vs. U c , or theoretically, vis., Re τ = √ 2Re to be used in particular for connecting to the work presented in [12], and some other articles. See [14] for details. Below a first threshold Re t , featureless turbulence leaves room for a laminar-turbulent, oblique, patterned regime (upper transitional range) that next turns into a sparse arrangement of localized turbulent bands (LTBs) propagating obliquely along two directions symmetrical with respect to the general stream-wise flow direction, experiencing collisions and splittings ("two-sided" lower transitional regime). Event B in Figure 1 corresponds to the opening of laminar gaps along the intertwined band arrangement observed in the tight laminar-turbulent network regime, and the simultaneous prevalence of downstream active heads (DAHs) driving the LTBs. Upon decreasing Re further, a symmetry-breaking bifurcation takes place at a second threshold Re 2 , below which a single LTB orientation prevails. Figure 2 displays snapshots of the flow illustrating these last two stages. Figure 1 of [14]. The domain size is 250 × 500 (span-wise × stream-wise). The flow is from left to right. Left: Strongly intermittent loose continuous LTB network at Re = 1200 (∼event B). Centre: Two-sided regime at Re = 1050 (Re Re 2 ). Right: One-sided regime at Re = 850. Downstream active heads (DAHs) are easily identified in the two right-most panels; a single one is visible in the upper left corner of the left image, marking the transition between sustained regular patterns and loose intermittent ones. Images here and in Figures 2 and 3 are adapted from snapshots taken out of the supplementary material of reference [14]. A significant result in [14] was that the decrease of turbulence intensity with Re below event B followed expectations for directed percolation in two dimensions but that, controlled by the decreasing probability of transversal splitting, the bifurcation at Re 2 prevented the flow to reach the corresponding threshold. The latter could nevertheless be extrapolated to a value Re DP < Re 2 . The ultimate decay stage takes place at Reynolds numbers below the point whereat transversal splitting ceases to operate. Figure 3 illustrates an extremely rare occurrence of transversal splitting at a Reynolds number roughly corresponding to event A in Figure 1. At lower Re, deprived of the possibility to nucleate daughters' LTBs of opposite propagation orientation, LTBs are forcibly maintained in the "one-sided" regime that eventually decays below a third threshold Re g , marking the global stability of the laminar flow. Corresponding flow patterns are illustrated in Figure 4, the right panel of which displays the surprising result that the turbulent fraction decreases as a power law with an exponent β of the order of that for directed percolation in one dimension, despite the fact that the flow develops in two dimensions [16]. The objective of the present work is the design of a minimal PCA model for these two last stages that is applicable to flow states for Re below event B, incorporates the anisotropy features visible in Figures 2-4, and accounts for the specific role transversal splitting above event A, in view of providing clues to their statistical properties in relation to dimensionality and universality issues. Figure 2. The domain size is now 500 × 1000, the stream-wise direction is vertical, and the flow upwards. (Right): Used as a proxy for the turbulent fraction, E y = V −1 u 2 y dV is displayed as a function of 1/Re; inset: same data raised at power 1/β with β = 0.28 suggesting decay according to the DP scenario in 1D, adapted from [16]. Modeling Context: Directed Percolation, Probabilistic Cellular Automata, and Criticality Issues Various modeling approaches to transitional wall-bounded flows have received considerable attention recently, from low-order Galerkin expansions of the primitive equations [17,18], to phenomenological theories based on a deep physical analysis of the processes involved in a reaction-diffusion context [19], to analogical systems expressed in terms of deterministic coupled map lattices [6,10], and to more conceptual models implementing the dynamics of cellular automata with probabilistic evolution rules (PCA) [20][21][22]. The model developed below belongs to this last category, implementing rules that focus on the main qualitative features seen in experiments. Such models are based on the conventional modeling of DP [2] which is most appropriate to account for the absorbing versus active character of local states. Let us briefly recall the PCA/DP framework. In the most general case, the activity at site j at time t + 1, call it S j ∈ {0, 1}, depends on the activity at sites in a full D-dimensional neighbor V j of that site at time t and the status of the links, permitting or not the transfer of activity within the neighborhood. For convenience a (D + 1)-dimensional lattice is defined with one-way (directed) bonds in the direction corresponding to time so that D-dimensional directed percolation is often presented as a special (D + 1)-dimensional percolation problem. In the simplest case of one space dimension (D = 1), the neighborhood of a lattice site at j is the set of sites with j ∈ [j − r 1 , j + r 2 ], comprising r 2 + r 1 + 1 sites, and it is supposed that contamination of the state at j at time t + 1 depends on the status of full configuration, the sites' activity, and the bonds' transfer properties ("bond-site" percolation [23]). In some systems, the propagation rule is totalistic in the sense that the output only depends on the number of active sites in the neighborhood and not on their positions, i.e., ς j = ∑ j ∈V j S j ; an interesting example is given in [24]. The question is whether, depending on the value of p, once initiated, activity keeps continuing in the thermodynamic limit of infinite times in an infinitely wide system. An answer is readily obtained in the mean-field approximation where actual local states are replaced by their mean value, neglecting the effect of spatial correlations and stochastic fluctuations (we follow the presentation of [20]). The spatially-discrete Boolean variables S j are, therefore, replaced by their spatial averages S = S j (t) and this mean value is just the probability that any given site is active. It is then argued that the probability to get a future absorbing state, 1 − S , is given by activity not being transmitted (1 − pS) 2 , which yields the mean-field equation: Equilibrium states correspond to the fixed points of (1): S = S = S * , which gives a nontrivial activity level S * = (2p − 1)/p 2 when p ≥ p c = 1/2. Close to threshold, defining ε = (p − p c )/p c = 2p − 1 one gets S * ≈ 4ε. In the mean-field (MF) approximation S * is the order parameter of the transition supposed to vary as ε β , which defines the critical exponent β, here β MF = 1. Directed percolation is the prototype of non-equilibrium phase transitions and, as such, is associated with a set of critical exponents (see [2]). Both the critical probability p c and the mean activity S * are affected by the effects of fluctuations, with p c ≈ 0.6445 > 1/2 expressing that a probability larger than the mean-field estimate is necessary to preserve activity, and β DP ≈ 0.276 when D = 1. The simple mean-field argument is not sensitive to the value of D in contrast with reality: β DP ≈ 0.584 when D = 2, ≈ 0.81 when D = 3, and trends upwards to 1 reached at D = 4 = D c = 4, called the upper critical dimension (see [2] for a review). Quite generally, mean-field arguments are valid for D > D c . We are interested in another critical exponent, α. When starting from a fully active system exactly poised at p c , the turbulent fraction is observed to decrease with time (the number of iteration steps) as S ∝ t −α with α ≈ 0.159 when D = 1 and 0.451 when D = 2, whereas the mean-field prediction, easily derived from (1), is α MF = 1. Scaling theory shows that α = β/ν , where ν is the exponent accounting for the decay of time correlations while ν ⊥ describes the decay of space correlations [2]. Universality is a key concept in the field of critical phenomena characterizing continuous phase transitions. It leads to the definition of universality classes expressing the insensitivity of critical properties to specific characteristics of the systems and retaining only properties linked to the symmetries of the order parameter and the dimension of space. For directed percolation, universality is conjectured to be ruled by a few conditions put forward by Grassberger and Janssen: that the transition is continuous into a unique absorbing state and characterized by a positive one-component order parameter, and that the processes involved are short-range and without weird properties such as quenched randomness; see [2]. Universality issues are discussed at length elsewhere in this special issue, in particular by Takeda et al. [25]. In this first approach, we shall examine how universality expectations hold for the ultimate decay stage of transitional channel flow at Re g , as described in Section 1.1, and limit the discussion to the consideration of exponents β and α. This will be done in Section 3, the next section being devoted to the derivation of the model and its mean-field study. Section 4 focuses on its ability to account for the symmetry-breaking bifurcation at Re 2 , and our conclusions are presented in Section 5. Context The approach to be developed is not new in the field of transitional flows. For example, studying plane channel flow, Sano and Tamai [21] introduced a plain 2D-DP model dedicated to support their experimental results, with a simple spatial shift implementing advection and a uniformly turbulent state upstream corresponding to their setup. Earlier, a similarly conceptual model was examined by Allhoff and Eckhardt [20], who introduced a PCA with two parameters accounting for persistence and lateral spreading appropriate for the symmetries of plane Couette flow, developed its mean-field treatment, and performed simulations to illustrate the spreading of spots and decay of turbulence in agreement with expectations. In a similar spirit but introducing more physical input, Kreilos et al. [22] analyzed the development of turbulent spots in boundary layers as a function of the residual turbulence level upstreams, separating a deterministic transport step from a stochastic growth/decay step with probabilities extracted from a numerical experiment, gaining insight into the statistics of boundary layer receptivity. Following the lines of research suggested by those works, we developed a 2D model designed to interpret the decay of channel in the LTB regimes from two-sided to one-sided at decreasing Re, just qualitatively proposing a plausible variation of probabilities introduced as functions of Re. In our approach, the elementary agents are the LTBs themselves either propagating to the left or to the right of the stream-wise direction. To them we attach variables analogous to spins in magnetic phase transitions problems. Even if in computations, numerical values S = ±1 will be used, for descriptive and graphical convenience we shall associate them with colors-specifically: blue (B) and red (R) for right-and left-propagating LTBs, respectively. Laminar sites will be denoted using the empty-set symbol ∅, will have value 0, and will be graphically left blank. These agents will be seated at the nodes of a square lattice with coordinates (i, j), i.e., S (i,j) with S → {R, B, ∅} at the given site. As seen in Figure 5a, we place the stream-wise direction along the first diagonal of the lattice so that the LTBs will move along the horizontal and vertical axes; see Figure 5b. A strong assumption is that an LTB as a whole corresponds to a single active state, while the discretization of space coordinates (i, j) ∈ Z 2 , and time t ∈ N tacitly refers to an appropriate rescaling of time and space. Furthermore, interactions are taken as local, with configurations limited to nearest neighbors in each space direction. Accordingly, the dynamics at a site (i, j) only depend on the configuration of its von Neumann neighborhood V (i,j) := {(i, j), (i ± 1, j), (i, j ± 1)}, Figure 5c, while evolution is driven by a random process. We now turn to the definition of rules that mimic the actual continuous space-time, subcritical and chaotic, Navier-Stokes dynamics governing the LTBs' propagation, decay, splitting, and collisions, via educated guesses from the scrutiny of simulation results, in particular those in the supplementary material attached to [14]. Design of the Model Let us first give a brief description of the processes to be accounted for. Below Re ≈ 800 (event A) only decay and longitudinal splittings are possible. Not visible in the snapshots of Figure 4 (left) but observable in the movies is the fact that a daughter LTB resulting from longitudinal splitting runs behind its mother along a track that may be slightly shifted upstream. This shift is negligible when Re is small (in-line longitudinal splitting) but as Re increases it becomes more and more visible while the general propagation direction is unchanged (off-aligned longitudinal splitting). On the other hand, Figure 4 clearly illustrates the fact that, upon transversal splitting, the new-born LTB systematically develops on the downstream side of its parent. Importantly, the propagation of LTBs is a dynamical feature different from advection treated as a deterministic step in [22]. Accordingly, it will be understood as a statistical propensity to move in a given direction resulting from an imbalance of stochastic "forward" and "backward" processes along their directions of motion. Other complex processes also seen in the simulations, such as fluctuating propagation with acceleration, slowing down, or lateral wandering, will be included only in so far as they can be decomposed into such more elementary events. All the events to be included in the model can be translated into the language of reaction-diffusion processes, persistence or death, offspring production, and coalescence, common in the field of DP theory [2]. On general grounds the governing equation reads: where C(i, j) is the neighborhood configuration of site (i, j) at time t, C one of the possible configurations, and R C a stochastic variable taking value 1 with probability p C corresponding to configuration C and value 0 with probability 1 − p C . The Kronecker symbol δ C C is here to select the configuration C that matches C. Depending on C and C , the output S (i, j) can be B or R. Figure 6 illustrates the set of possible single-colored neighborhoods, either B (upper line) or R (lower line). Following the indexation in Figure 5c, the order of the columns is based on the physical condition and respects the upstream/downstream distinction illustrated in Figure 5a, making configurations with the same index physically equivalent. ? ? ? ? Figure 6. Single-color configurations: from the overall geometry depicted in Figure 5a, the downstream side of a state is to the top for B states and to the right for R states. Each colored square indicates the active state in the configuration at time t of site (i, j) at the center. The question mark features the probabilistic outcome (time t + 1). These single-color elementary configurations will be denoted as C i with i ∈ [1:5]. They will be described as Later, more complicated configurations will not be given a name but just a description following the same rule, e.g., [∅ BBR ∅]. Importantly, we make the assumption that the future state at a given node, the question marks in Figure 6, is the result of the probabilistic combination of the independent contributions of elementary configurations involving a single active state in its neighborhood. First of all, the void configuration C 0 ≡ [∅ ∅ ∅ ∅ ∅] obviously generates an empty site with probability 1, hence an occupied site with probability p C 0 = 0, in order to preserve the absorbing character of the dynamics. All the other configurations evolves according to probabilities that are free parameters just constrained by empirical observations. Let us now interpret probabilities associated with the five situations depicted in Figure 6, focusing on the case of B states: corresponds to the natural propagation of the active state along its own motion direction. Accordingly, the active site B at (i − 1, j) is expected to be found at (i, j) and time t + 1 with a high probability, p C 5 = p 5 1, which corresponds to the near-deterministic propagation of an active state as observed for Re ≥ Re g . With probability 1 − p 5 1, site (i, j) will not turn active, which means that the LTB has decayed or experienced a speed fluctuation that delayed its propagation. The corresponding R configuration is corresponds to an active site B at (i, j) that is not supposed to stay in place but move to (i + 1, j) with probability p 5 and leave site (i, j) empty at time t + 1. The probability p 1 that (i, j) is still active at time t + 1, therefore, generally corresponds to the creation of a novel active state by in-line longitudinal splitting at the rear of the active state that has effectively moved. Persisting activity at (i, j) and time t + 1 can also be the result of state at (i, j) and time t experiencing a speed fluctuation leaving it stuck at the same place with probability 1 − p 5 as argued above for configuration C 5 . The presence of parameter p 1 undoubtedly makes the dynamics richer. The corresponding R-configuration is corresponds to an active state B at site (i, j + 1) that contaminates backwards and laterally upstream the site at (i, j) in addition to its likely propagation to (i + 1, j + 1) with probability p 5 . This is precisely what is sometimes observed for longitudinal splitting, where the daughter follows a track parallel to that of the mother but slightly shifted upstream, i.e., off-aligned. Configurations C 1 and C 2 both account for longitudinal splitting but the latter hence introduces some lateral diffusion. Along this line of thought, numerical simulation results in [14], illustrated in Figure 4, suggest that probability p 2 is tiny close to Re g but increases with Re. The corresponding R-configuration is is supposed to advance further at (i + 2, j) with probability p 5 . Persisting activity at (i, j), therefore, means longitudinal splitting ahead but now with the opening of a wide laminar gap between the offspring left behind at (i, j) and the parent that has advanced, with probability p 5 , at (i + 2, j). Else, activity at (i, j) and t + 1 could result from activity at (i + 1, j) and time t propagating backwards to (i, j) at time t + 1. These circumstances have not been observed and appears unlikely or impossible, which suggests to take 5. In configuration C 4 ≡ [∅ ∅ ∅ B ∅], the state B at (i, j − 1) and time t is expected to be at (i + 1, j − 1) at time t + 1. State at (i, j) being active at t + 1 means contamination backwards and laterally downstream, which is never observed in the simulations; hence, Still about configuration C 4 , the situations described in the previous items all imply single-colored evolution, which is guaranteed below the onset of transversal splitting, i.e., R 800. When Re 800, as illustrated in Figure 3, this splitting produces an R offspring at (i, j) out of a B parent at (i, j − 1) or B offspring from an R parent at (i − 1, j), as sketched in Figure 7 (left). A probability p 4 = 0 will be associated with it, where the prime is meant to recall that it involves states of different colors. To summarize, as it stands the model involves four parameters: p 1 mainly governs longitudinal splitting and p 2 additional lateral diffusion, p 5 is for propagation, and p 4 for transversal splitting. The propagation of active states along their own direction involves probabilities associated with elementary configurations C 1 and C 5 while the overwhelming contribution of p 5 favors one direction. Configuration C 3 that could have contributed to the balance is empirically found negligible, saving one parameter as indicated above. Neighborhoods with more than one active site are treated by assuming that the future state S of the central node (i, j) is the combined output of its elementary ingredients, each contribution being considered as independent of the others, i.e., without memory of the anterior evolution, of which the considered configuration is the outcome. The computation of the probability attached to the output of a given single-colored neighborhood is then straightforward. The argument follows the lines given for directed percolation, bearing on the probability that the state at the node will be absorbing (empty) and leading to Equation (1) in the mean-field approximation [20,24]. Things are a little more complicated when the neighborhood is two-colored since in all mixed-colored cases some configurations correspond to collisions and others allow for the nucleation of a differently colored offspring when p 4 = 0. For an elementary configuration, non-contamination of site (i, j) from an active neighboring state in position k ∈ [1:5] takes place with probability (1 − p k ) and of course with probability 1 if the corresponding site is empty. This gives the general formula (1 − p k S k ), where S k = 1, when the site is active, either B or R, and S k = 0 when it is absorbing (∅). For a configuration C x = [S 1 , S 2 , S 3 , S 4 , S 5 ], where S = B, R, or ∅, the probability to get an absorbing state is ( To deal with two-colored neighborhoods properly, we must be a little more specific and write the probability of the state S of a given color S as where it is understood that if S = B, thenS = R or the reverse, and S j = 0 for j = 1, 2, 5, orS 4 = 0 if the corresponding states are ∅. Figure 7 (right) illustrates the most interesting two-state configurations with different colors corresponding to collisions (C1) and offspring generation (C2). Such a situation is dealt with by adding a supplementary rule: 7. When the general expression (3) gives non-zero probabilities to S andS the resulting superposition of states is not allowed and a choice has to be made. It might seem natural to keep the state with the maximum probability but, depending on circumstances hard to decipher, collisions sometimes appear to cause the decay of both protagonists or else reinforce the dominance of one color in a given region of space. A similar bias can affect transversal splitting. These peculiarities are not taken into account here: for simplicity, in all conflicting cases, we make the assumption that the result is non-empty and random with probability 1/2. The model is now complete with parameters clearly related to empirical observations, plausible relative orders of magnitude and sense of variation: Probability p 5 is the main ingredient for the built-in propagation of the two families of LTBs (active states). In turn p 1 is obviously related to the behavior of the system close to decay at and slightly above Re g . The value given to probability p 2 will appear crucial to the 1D reduction of DP in a 2D medium as observed experimentally (Figure 4, right). Finally, we can anticipate that probability p 4 will control the one-sided/two-sided symmetry-restoring bifurcation, as it continuously grows from 0 beyond Event A at R ≈ 800. Mean-Field Approach The explanatory potential of the model is first examined by means of a mean-field approximation which mainly relies on the replacement of fluctuating quantities by space-averaged values and the neglect of correlations. The observables involved in the mean-field expressions are the ensemble averages of the microscopic states S(i, j) . Their values at t + 1 are obtained by taking averages of the governing Equation (2) using the expression of the configurational probabilities given in (3). By assumption/definition S is the mean outcome of p C x averaged over all the possible configurations, where space dependence (i, j) is temporarily kept: S (i, j) = p [S 1 ,S 2 ,S 4 ,S 5 ] . This gives a set of two equations: The approximation now enters the evaluation of the products on the right hand side of the equation. Each variable is replaced by its average and the spatial dependence is dropped: B(i, j) → B and R(i, j) → R . Further, correlations are neglected so that the average of a product is just the product of averages. The expansions of (4) and (5) in powers of B and R are then readily obtained. Forgetting for a moment the intricacy linked to transversal splitting/collisions, the general expression for the dummy variables S and S reads: with p k ∈ {p 1 , p 2 , p 4 , p 5 } and where h.o.t. stands for the higher order terms, formally cubic, quartic, etc. The first sum in (6) corresponds to the contribution of the elementary configurations introduced in Figure 6, and the second sum to binary configurations, in particular the nontrivial ones corresponding to transversal splittings and collisions examined in Figure 8 (right). Orders of magnitude among the p k , further support neglecting the contribution of configurations populated with three or more active sites, involving products of three or more probabilities p k , and among contributions of a given degree, those not containing p 5 when compared to those that do, recalling the assumption p 5 1 and {p 1 , p 2 } 1 implied by the nearly deterministic propagation of states in position 5 of Figure 6. A number of terms can, therefore, be neglected in the expanded forms of (4) and (5), which after simplification read: This system presents itself as the discrete time counterpart of the differential system introduced in [14] to interpret the symmetry-breaking bifurcation observed at decreasing Re in the simulations. As a matter of fact, subtracting B and R on both sides of (7) and (8) respectively, one gets: to be compared with system (1,2) in [14], reproduced here for convenience: where X ± represents what are now the densities B and R . The coefficients in (11) and (12) are then related to the probabilities introduced in the model as a ∝ p 1 + p 2 + p 5 − 1, b ∝ p 5 (p 1 + p 2 ), c ∝ p 4 , and d ∝ p 2 5 . By omitting the common proportionality constant that accounts for the time-stepping inherent in the discrete time reduction (featured by the denominator of left-hand sides in (9) and (10) as "(dt ≡ 1))," constants a, b, c, and d will serve as short-hand notation for the corresponding full expressions in terms of the probabilities p k . Since fixed points given by the condition S = S is strictly equivalent to dX ± /dt = 0, we can next take advantage of the analysis performed in [14] and predict a supercritical symmetry-breaking bifurcation for an order parameter \| B − R \| (denoted "A" in [14]) at a threshold given by c cr = a(d − b)/(d + 3b). This symmetry-breaking bifurcation takes place for p 4 = c > 0, but the model can deal with the regime below event A at Re ≈ 800 for which p 4 ≡ 0. In that case the bifurcation corresponding to global decay at Re g takes the form of two coupled equations generalizing (1) for DP. Using the abridged notation, these equations read: In addition to the trivial solution R 0 = B 0 = 0 corresponding to laminar flow, we have two kinds of non-trivial solutions, either single-sided ( * ) with R = 0 and B = 0 or B = 0 and R = 0, the non-vanishing solution being S * = a/b, with S = R or B, or double-sided ( * * ) with B * * = R * * = a/(b + d). A straightforward stability analysis of the fixed points of iterations (13) shows that the one-sided solution is stable when b < d and unstable otherwise whereas the reversed situation holds for the two-sided solution. Returning to probabilities, the global stability threshold is thus given for a = 0; hence, (p 1 + p 2 + p 5 ) cr = 1 and the one-sided solution is expected when b < d; i.e., p 1 + p 2 < p 5 . Results of the mean-field approach adapted from [14] to the present formulation will be illustrated in Figure 14 below. Numerical Simulations While serving as a guide to the exploration of a vast range of parameters, the simplified mean-field theory developed above is not expected to give realistic results relative to the critical properties expected near the transition point, whether decay at Re g or symmetry restoration above Re 2 . For example, observations suggest that LTB propagation is a dominant feature; hence, p 5 1 and {p 1 , p 2 } is small, leading us to expect stable one-sided solutions systematically. This conclusion, however, strongly relies on neglecting all terms beyond second degree in (4) and (5) in the evaluation of the contribution of densely populated configurations, leading to (7) and (8). This is legitimate only when S n S 2 , i.e., S 1, that is, close to decay in the case of a continuous (second-order) transition but not necessarily elsewhere in the parameter space, in particular at the one-sided/two-sided bifurcation where both R and B are of the same order of magnitude but may be large. Even when keeping the assumption of independence of contributions to the future state at a given lattice node, this problem is not easily addressed and, at any rate, has to be properly accounted for in the presence of stochastic fluctuations, which will be done numerically. The translation of the probabilistic rules introduced in Section 2.2 using Matlab R is straightforward once the "B/R/∅" convention is appropriately translated into "+1/−1/0". No assumption is made other than the independence of the contributions of the different configurations to the outcome at a given lattice node, by strict application of the rules expressed through (2) A preliminary study of the model in a small domain has shown that the different transitional regimes and the symmetry-breaking bifurcation were indeed present as expected from the simplified mean-field approach. (We remind that the model contains nothing appropriate for organized laminar-turbulent regimes for Re > 1200 and is relevant only for the strongly intermittent sparse LTB networks pictured in Figures 2 and 3). In [14], we argued that the onset of transversal splitting was the source of genuinely 2D behavior. Accordingly we shall consider the stochastic model in two steps, below and above the onset of transversal splitting, here associated with p 4 ≡ 0 and p 4 > 0 respectively. Furthermore, in the simulations the LTBs were seen to propagate obliquely with respect to the background downstream current. This propagation is nearly all contained in the probability attached to configuration C 5 (p 5 for propagation and 1 − p 5 for decay or slowing-down), and to a lesser extent influenced by the contribution of configuration C 1 , mostly associated with in-line longitudinal splitting. We shall account for the limited sensitivity of the propagation speed to the value of Re to fix p 5 constant and close to 1, more specifically p 5 = 0.9, and let other parameters vary. The role of p 2 and p 4 , both related to 2D features, will be studied separately in the two next sections. Coarsening from Two-Sided Initial Conditions In the absence of transversal splitting, changes in the population of each state only comes from transversal collisions. As documented in [14], when starting from an initial condition with two similarly represented orientations, collisions lead to the formation of domains uniformly populated by one of each species, following from a majority rule, with interactions limited to the domain boundaries. A coarsening takes place with one species progressively disappearing to the benefit of the other, leaving a single-sided state at large times. The process is illustrated here using simulations of the model with p 5 = 0.9, p 1 = 0.1, p 2 = 0.07, values known from the preliminary study to produce a sustained nontrivial final state. The decay from a fully active state populated with a random distribution of B and R states in equal proportions is scrutinized in a 256 × 256 domain with periodic boundary conditions. Figure 8 illustrates a particularly long transient displaying the different stages observed during a typical experiment. The upper panel displays the time series of the turbulent fractions for each species, B and R, for a two-sided high-density initial condition, F t (B) + F t (R) = 1, F t (B) F t (R) 0.5. Contrasting with the monotonic variation observed when starting from one-sided initial conditions, either increasing from a low density of active states (F t = 0.05) or decreasing from a fully active configuration (F t = 1), the turbulent fractions change in a more complicated way that is easily understood when looking at the bottom line of snapshots. The total turbulent fraction first decreases due to the dominant effect of collisions. These collisions tend to favor a spatial modulation of the activity amplifying inhomogeneities in the initial conditions. This distribution results from the majority effect expressing the local stability of one-sided states predicted by the mean-field analysis. A periodic pattern already appears at t = 100, with bands oriented parallel to the second diagonal of the square domain. B states move right along the horizontal axis, and R states up along the vertical axis, at the same average speed so that the pattern drifts along the first diagonal of the domain. Regions where B or R dominate are locally stable against destructive collisions and activity is limited to B/R interfaces. After a while, splittings begin to counteract collisions and an overall activity recovers, here for t ≈ 250. The local density of B and R states increases inside bands that become better defined, reaching a sustained regime with two R-B alternations, wide and narrow, at t 1500. This configuration is nearly stable and slowly evolves only due to the erosion of narrowest bands at the R/B interfaces. At t ≈ 5500 these bands disappear by merging, leaving two bands, B wide and R narrow. The same slow erosion process leads to the final homogeneous B regime by decay of the R band at t ∼96,000. The two successive band decays take place at roughly constant total turbulent fraction with fast adjustment at the band decay, up to the final single-sided turbulent fraction. The asymptotic state is independent of the way it has been obtained, from one-sided or two-sided initial conditions. Time series of the turbulent fractions for a simulation from a fully active initial configuration with B and R states in equal proportions-blue and red in graphs, respectively; the dotted black trace is for the total turbulent fraction. Two simulations starting from low (F t = 0.05, cyan) and high (F t = 1, magenta) density one-sided states are displayed for comparison. (Bottom): Snapshot of state during the simulations from the two-sided initial condition, at t = 100 during initial decay, at t = 5500 with two pairs of active bands of each color, at t = 13,000 when the narrowest bands merge and disappear, at t = 96,000 when the R active band disappears, leaving a uniform B state. The long duration of the transient taken as an example is due to the near stability of the rather regular pattern building up after the initial fast decay. This property is in fact the result of a geometrical peculiarity of the square domain: B and R states travel statistically at the same speed through the domain, horizontally and vertically, respectively, so that the band integrity is maintained despite propagation and the evolution controlled by collisions at the B-R and R-B interfaces only. The observed slow erosion process only results from large deviations among collisions. In rectangular domains, the propagation times become different and the symmetry of the two interfaces is lost. A bias results, which induces a systematic erosion of bands and a shorter transient duration. Whatever the aspect ratio, one of the states is always ultimately eliminated and the last stage of the transient corresponds to a trend toward a statistically uniform saturated one-sided regime with a turbulent fraction strictly independent of the shape. Accordingly, to save the time corresponding to the transient, in the next section we will study the decay of the one-sided regime by starting from random one-sided initial conditions. All these features nicely fit the empirical observations discussed at length in [14] where similar transients were obtained below the onset of transversal splitting-in much smaller effective domains and with far fewer interacting LTBs, however ( Figure 2, right panel, and Figure 4, left panels). Decay: 1D vs. 2D The model is designed to exemplify a decay according to the DP scenario in a two-dimensional setting, with specificities linked to the anisotropic propagation properties of the LTBs in transitional channel flow, and, in particular, propose an interpretation for the observation of 1D-DP exponents in the absence of transversal splitting (p 4 = 0). Accordingly, we examine the role of transverse diffusion (parameter p 2 ) modeling the slight upstream shift that may affect LTBs at longitudinal splitting. We focus on a set of experiments with p 5 = 0.9, p 2 fixed, and control parameter p 1 . When p 2 cancels exactly, it is easily understood that transversal expansion is forbidden: An active B state at (i, j + 1) or R state at (i + 1, j) at time t cannot give birth to an active state of the same kind at (i, j) at t + 1. The evolution stems from processes associated with configuration C 5 with probability p 5 or C 1 with probability p 1 . These processes change occupancy only along direction i for B states, and j for R states, precisely in the direction corresponding to the single-sided regime considered (after termination of the transient). The dynamics are, therefore, strictly one-dimensional and decay is expected to follow the 1D-DP scenario. In contrast, introducing some transverse diffusion (p 2 = 0) immediately gives some 2D character to the dynamics. This is illustrated in Figures 9-12. We consider first p 2 non-zero and relatively large p 2 = 0.1. Figure 9 displays the behavior of the turbulent fraction as a function of p 1 . Having a good estimate of the threshold one can next consider the decay of the turbulent fraction, which is supposed to decrease as a power law at criticality, p 1 = p c 1 : F t ∼ t −α DP with α DP = β DP /ν DP where ν DP ≈ 1.295; hence, α DP ≈ 0.451 [2]. Figure 10 (left) shows that this is indeed the case for the compensated turbulent fraction F t × t α DP , up to the moment when fluctuations become too important due to size effects and lack of statistics. When p 1 is different from p c 1 but stays sufficiently close to it, the variation of the turbulent fraction keeps trace of the critical situation, except that the number of steps needs to be rescaled by the distance to threshold due to critical slowing down: the time scale τ diverging as (p 1 − p c 1 ) −ν , number of steps is rescaled upon multiplying it by (p 1 − p c 1 ) ν . Figure 10 (right) indeed shows a good collapse of the compensated curves as a function of the rescaled number of steps when using the exponents corresponding to 2D-DP, α DP ≈ 0.451 and ν DP ≈ 1.295 [2]. We now consider p 2 = 0 which, as argued earlier, should fit the critical behavior of directed percolation when D = 1. In that case, when using square or nearly-square rectangular domains, size effects turn out to be particularly embarrassing as will be illustrated quantitatively soon. However, we can take advantage of the fact that, assuming propagation in the one-sided regime, e.g., along the direction for B active states, N B being the corresponding number of sites involved, the computed turbulent fraction is, in fact, the average of the activity over N R independent lines in the complementary direction, while still being sensitive to size effects controlled by N B . Accordingly, at given computational load (proportional to N B × N R ), one can freely increase the size artificially in considering a strongly elongated domain D = [(N B × k) × (N R /k)], with k sufficiently large that the average over N R /k independent lines still make sense from a statistical point of view, while postponing size effects. With reference to a (192 × 192) domain, we have obtained good results with k = 16, i.e., 3072 × 12 up to k = 64, i.e., 12288 × 3. Though this choice is a bit extreme, we present here results about 1D-DP criticality with the 12288 × 3 domain in Figure 11. The left panel displays the variation of the mean turbulent fraction with p 1 , which has been fitted against the expected power law, F t = a(p 1 − p c 1 ) β . One gets a = 1.473 (1.446, 1.5), p c 1 = 0.2682 (0.2682, 0.2683), β = 0.2701 (0.2664, 0.2738). This value of β is quite compatible with the value β DP ≈ 0.276 when D = 1 [2]. Furthermore, accepting this value, a linear fit of F t 1/β with p 1 then provides an extrapolated threshold p c 1 = 0.26817. As seen in the right panel of Figure 11, in the neighborhood of p c 1 a good collapse is obtained for the compensated turbulent fraction as a function of the rescaled number of steps when using the exponents α = 0.159 and ν = 1.734 corresponding to 1D-DP [2]. Size effects already alluded to above are illustrated in Figure 12. Displaying the turbulent fraction as a function of the number of steps for linear size N B from small systems to relatively large ones (N B = 64 up to 768) in lin-log scale, the left panel illustrates the late stage of decay right at criticality as obtained from the previous study summarized in Figure 11. It is seen that, in the time-window considered (0, 10 5 ) the exponential dependence observed at small sizes is progressively replaced by the power-law behavior expected at criticality at infinite size. Size effects are also ruled by scaling theory; see, e.g., [2] for DP. They relate to correlations in physical space that are associated with exponent ν ⊥ . The ratio z = ν /ν ⊥ is called the dynamical exponent and theory predicts that, for finite size systems, scaling functions depend on time with the number of sites as t D/z /N where N is the total number of sites. In the (quasi-)one-dimensional regime we are interested in, D = 1, N is just N B and z = 1.58 [2]. The right panel of Figure 12 indeed shows extremely good collapse of the traces corresponding to those in the left panel, once the number of steps is rescaled as t/N 1.58 B and the turbulent fraction is compensated for decay as F t (t) × t 0.159 , both exponents taking on the 1D-DP values already mentioned. Of interest in the context of channel flow decay, the crossover from 2D behavior for p 2 sizable (e.g., p 2 = 0.1, Figures 9 and 10) to 1D behavior for p 2 = 0 is of interest since p 2 is associated with the progressive importance of off-aligned longitudinal splitting as Re increases. A series of values of p 2 , decreasing to zero roughly exponentially, has been considered and the corresponding DP threshold has been determined as given in Table 1 and shown in Figure 13 (left). Table 1. Values of p 1 at criticality at given p 2 (p 5 = 0.9 and p 4 = 0). The results for p 2 = 0, evolving as t α 2D −α 1D are marked with ( * ) and ( * * ) are obtained in the 192 × 192 domain and in the 3072 × 12 quasi-1D domain, respectively. In the time span considered here, the latter is free from finite-size effects which is not the case of the former with the corresponding compensated data decaying exponentially at the largest times. It is easily seen that, except for p 2 = 0, the compensated time series display a wide plateau indicating that 2D behavior holds for a certain amount of time. Whereas traces for p 2 = 0.1 and p 2 = 0.05 cannot be distinguished, for smaller values of p 2 the plateau regime starts at larger and larger times and develops after having followed the 1D trace for longer and longer durations, clearly indicating the influence of the anisotropy controlling the effective dimensional reduction. A similar consequence of the crossover affects the decrease of the mean turbulent fraction with the distance to threshold but, apart from this qualitative observation, no reliable information can be obtained on exponent β owing to the difficulty to reach the relevant critical regime. We shall not document the case when p 1 = 0 and p 2 varies. This situation is not observed in the simulations since off-aligned longitudinal splitting is conspicuous only sufficiently above Re g , in the vicinity of which decay is fully accounted for by in-line longitudinal splitting modeled by a variable p 1 = 0, but the possibility remains, at least conceptually. The decay when p 1 = 0 happens to follow the same 1D-DP scenario though the argument is slightly less immediate. It relies on the observation that no growth is possible in the propagation direction of a given LTB species, whereas off-aligned longitudinal splitting (p 2 = 0) permits growth and diffusion in the transverse direction. Under the combined effects of transversal diffusion (p 2 small) and propagation (p 5 large), near-threshold, the sustained turbulent regime is made of quasi-1D clusters that are aligned with and drift along the diagonal of the lattice, i.e., the stream-wise direction, and get thinner and thinner when decaying, supporting the reduction to a "D = 1" scenario. Here, the trick used for p 2 = 0 does not work, and simulations in square domains are necessary with no escape for size effects which hinders the observation of the critical regime. Nevertheless, p c 2 when p 1 = 0 seems close to p c 1 when p 2 = 0, suggesting some symmetry between p 1 and p 2 . The relevance of the results with p 4 ≡ 0 to transitional channel flow will be discussed in the concluding section. We now turn to the general two-sided case with transversal collisions and splittings. Beyond Onset of Transversal Splitting, P 4 > 0 In statistical thermodynamics systems, critical properties at a second order phase transition leads to define a full set of exponents governing the variation of macroscopic observables close to criticality [26]. The concept of universality was introduced to support the observation that these systems can be classified according to the value of their exponents depending on a few qualitative characteristics, the most prominent ones being the symmetries of the order parameter and the dimension of physical space. This viewpoint can be extended to far-from-equilibrium systems such as coupled map lattices (CMLs) displaying nontrivial collective behavior. The associated ordering properties present many characteristics of thermodynamical critical phenomena at equilibrium. Universality classes beyond those known from equilibrium thermodynamics have been shown to exist with different sets of exponents. An additional criterion, the synchronous or asynchronous nature of the dynamics, has been found relevant to distinguish among them [27]. In the context of the present model, as soon as probability p 4 grows from zero, fully one-sided configurations previously reached after the termination of a possibly long transient are now unstable against the presence of states with the complementary color. The stationary regime that develops in the long term can be, either ordered, i.e., one-sided with one dominant active state (B or R), or disordered, i.e., two-sided with statistically equal fractions of each active state (B and R). Furthermore, a transition at some critical value p 4 c is expected to take place on general grounds. This gives us the motivation to study the response of the model to the variation of p 4 as a critical phenomenon beyond the mean-field expectations of Section 2.3. The results of the mean-field approach, system (11) and (12), rephrased from [14], are depicted in Figure 14 (left). Upon variation of parameter c representing p 4 up to an unknown rescaling factor, all along the one-sided regime (c < c cr ), the total turbulent fraction is seen to decrease while the order parameter measuring the lack of symmetry similarly decreases to zero according to the usual Landau square-root law. Obviously symmetrical, the two-sided regime (c > c cr ) is then characterized by a total turbulent fraction that regularly grows due to the contribution of splitting, whatever the type of active state. From now on, we shall simply refer to the turbulent fractions and other statistical quantities as their time average over a sufficiently long duration, up to 2 × 10 6 simulation steps, after elimination of an appropriate transient, up to 10 5 steps, the largest values being necessary close to the transition point owing to the well-known critical slowing down. On the one hand, the total turbulent fraction is obviously defined as F t (B) + F t (R), where the over-bar denotes the time averaging operation. (Later on, we shall omit this over-bar when no ambiguity arises between the instantaneous value of a quantity and its time average, especially for the axis labelling in figures.) On the other hand, the lack of symmetry can be measured by the signed difference averaged over time F t (B) − F t (R), able to distinguish global B orientation from its R counterpart, or rather its absolute value F t (B) − F t (R) since we are only interested in the amplitude of the asymmetry (called 'A' in [14]) and not in which orientation is dominant, the two being equivalent a priori for symmetry reasons. However, due to the finite size of the system, in the symmetry-broken regime close to threshold, orientation reversals can be observed as illustrated later (Figure 15), so that blind statistics in the very long durations are no longer representative of the actual ordering. Like in thermal systems [28] or their non-equilibrium counterparts [27], it is thus preferable to define the order parameter through the mean of the unsigned difference: \|F t (B) − F t (R)\| . Corresponding simulation results are displayed in Figure 14 (right) for a system of size (256 × 256). The general agreement between the two diagrams is remarkable, up to an unknown multiplicative factor translating c into p 4 , as discussed earlier. One can notice that the order parameter is minimal but not zero in the two-sided regime, which is due to fluctuations and the fact that the two operations of averaging over time and taking the absolute value do not commute. Finite-size effects are also apparent as a rounding of the graph at the location of the would-be critical point in the thermodynamic limit. (11) and (12) after [14]. The total turbulent fraction is F t (B) + F t (R) and the order parameter characterizing the transition is abs(F t (B) − F t (R)). The current justification for taking the absolute value is that the time between orientation reversals diverges with the system size and the phase transition only takes place once we have taken the thermodynamic limit of infinitely large systems studied over asymptotically long durations [28]. Accordingly, very long well-oriented intermissions can be considered as representative of the symmetry-broken regime. The problem is illustrated in Figure 15 displaying the time series of F t (B) − F t (R) and histograms of \|F t (B) − F t (R)\| for p 4 = 0.0121, still in the one-sided but already alternating regime, next for p 4 = 0.0125 and 0.0126, where one can notice a change in the shape of the histogram, and finally for p 4 = 0.0140, sufficiently deep inside the two-sided regime where the histogram displays a sharp maximum at the origin. On this basis one could use the histograms of the "order parameter" and determine the threshold from the position of its most probable value, whether non-zero in the symmetry-broken state or at the origin when symmetry is restored. This procedure would give p 4 c ≈ 0.01255. The symmetry-breaking bifurcation can now be studied beyond the mean-field description as other collective phenomena studied in equilibrium and far-from-equilibrium statistical physics: In addition to the order parameter, the variation of which leads to the definition exponent β in the ordered regime, another observable of interest is the susceptibility measuring the response to an applied field conjugate to the order parameter, vis. M = χH with the magnetization M coupled to magnetic field H in the case of magnets. The susceptibility diverges near the critical point, with leads to the definition of two exponents γ and γ in the disordered and ordered regime, respectively. Universality implies γ = γ , as can already easily be derived in the mean-field framework. When a conjugate field cannot be defined, one uses the property that fluctuations take the instantaneous value of the order parameter away from its average value, which can be understood as resulting from the response to a conjugate field. This helps one to relate the susceptibility to the variance of fluctuations of the order parameter. The identification is up to a multiplication by the "volume" of the system that has to be introduced in order to compare the results from systems with different sizes. This is what will be done here; hence, χ = N B N R × var (\|F t (B) − F t (R)\|). As shown in Figure 16 (top), this quantity displays a sharp maximum, indicative of the singularity expected at the thermodynamic limit. In a finite-size but large system, the critical point is then estimated from the position of the maximum of the susceptibility. Here, this gives p 4 c ≈ 0.0123 slightly smaller but compatible with the value obtained above from the examination of the histograms. Unfortunately, this discrepancy due to size-effects forbids us to determine exponents β and γ with some confidence. The histograms were all built using 75 bins and contain the same number of points for 10 5 < t < 2 × 10 6 , but the vertical scales are not identical. Having in mind results of the mean-field approach, namely, β = 1/2 and γ = 1, we can, however, estimate the range where stochastic fluctuations have nontrivial effects. The bottom-left panel of Figure 16 displays the variation of the order parameter already shown in Figure 14 (right), but now squared in order to show that, far from the critical point, the system fulfils the mean-field square-root prediction to an excellent approximation, with an extrapolated threshold p 4 MF 0.0133, shifted upwards with respect to the estimates obtained from the simulations p 4 c. 0.0123-0.0125. In the same way, the divergence of the susceptibility with exponents γ = γ = 1 expected from the mean-field argument shows up upon retreating the data already given in Figure 16 (top) and plotting 1/χ as a function of p 4 . This is done in Figure 16 (bottom-right) showing the same linear variation of 1/χ below and above the transition point, in agreement with the theory. The extrapolation of the linear fits on both sides of the transition yield p 4 c ≈ 0.0127 in reasonable agreement with the value obtained from the order parameter variation in the same conditions and definitely larger than the empirical values. Clearly, deviations seen in the boxed parts of these two figures warrant further scrutiny, motivating our current approach via finite-size scaling theory [28] in search for universality. On going work attempts at a full characterization of the critical regime through exponents determination. Though local agents do not behave as Ising spins, symmetries are basically identical, so that the equilibrium 2D Ising universality class or its non-equilibrium extension [27] might be relevant. We shall discuss this further below. Discussion and Concluding Remarks Coming long after a conjecture by Pomeau [1], empirical evidence is growing that the ultimate stage of decay of wall-bounded turbulent flows towards the laminar regime follows a directed-percolation scenario. The evidence comes from laboratory experiments and direct numerical simulation of the Navier-Stokes equations but this support is still far from a theoretical justification. The recognition of the globally subcritical character of nontrivial states away from laminar flow and the elucidation of the structure of coherent structures involved in these nontrivial states [29] were first steps in this direction. The next ones would be the elucidation of special phase space trajectories from sustained localized turbulence accounting for the decay to laminar regime, on one side, and to proliferation via splitting, on the other side, using specific algorithms for the detection of rare events and the determination of transition rates that can be attached to them (see [30] for an illustrative example and references). These are heavy, and possibly not much rewarding, tasks but it would be nice to be able to attach numbers to specific events such as the splittings illustrated in Figure 3 or Figure 4. We have chosen to short-circuit such studies through analogical modeling, by which seemed more appropriate to make further progress regarding the thermodynamic limit and associated universality issues. One should though consider this practice as providing hints and not a demonstration that the results will apply to the case under study. In the present paper, the problem has been considered from this last viewpoint, assuming that the ultimate decay stages were amenable to the most abstract level of implementation in terms of probabilistic cellular automata [2], following [20,22]. We focussed on the specific case of channel flow that offers a particularly rich transitional range. Its upper part displays regular non-intermittent laminar-turbulent patterns that can better be described using the tools of pattern-forming theory [15,25,31]. The lower transitional range is characterized by their spatiotemporally intermittent disaggregation, to which the considered type of modeling is particularly relevant. The analogy alluded to above has, however, been severely constrained to fit the empirical observations. The main assumptions were the introduction of two types of active agents attached to each kind of localized turbulent bands propagating in one of the two possible orientations with respect to the stream-wise direction. Interactions were assumed local so that the probabilistic cellular automata evolved simple nearest neighbors von Neumann neighborhoods (Figures 5-7). Scrutiny of simulation results lead to the introduction of a certain number of probabilities governing the fate of single-occupancy neighborhoods. Multiple-occupancy was treated as a combination of single-occupancy configurations supposedly independent, reducing the number of parameters to be introduced and drastically simplifying the interactions (at any rate impractical to estimate in detail). A clear-cut physical interpretation was, however, given to each parameter in the set reduced to four, accounting for every possible stochastic event affecting the agents, namely, propagation, decay, and splitting, either longitudinal or transversal. A mean-field study of the model, neglecting the nontrivial effects of stochastic fluctuations, reproduced the empirical bifurcation diagram of channel flow at a qualitative level ( Figure 14). Transitions have been studied quantitatively by numerical simulation of the stochastic model considering variations of these parameters as putative functions of the Reynolds number Re, highlighting three situations: In the two first cases, the parameter p 4 associated with transversal splitting, i.e., the nucleation of a daughter with orientation opposite of its mother, was switched off, as inferred from observations for Re 800, where the single-sided regime is well established. The coarsening observed when starting from two-sided initial conditions was faithfully reproduced (Figure 8) and decay seen to follow the directed-percolation expectations. The specific conclusion was that, when parameter p 2 is no-zero, with p 2 attached to longitudinal but upstream-shifted splitting, the scenario is typical of a 2D system with a high level of confidence, whereas when it is strictly zero, i.e., the daughter strictly aligned with the mother, the decay is 1D. A cross-over is observed when p 2 is reduced, that manifests itself as a transient reminiscent of 1D behavior, the longest the closest p 2 is to zero. Simulations of channel flow have shown that exponent β controlling the ultimate decay of the turbulent fraction was that of 1D directed percolation [16]. Since parameter p 2 is attached to the slight upstream trajectory shift experienced by a daughter upon splitting from its mother, this observation strongly suggests that the trajectory shift is mostly irrelevant and that localized turbulent bands propagate along independent tracks so that the end result is just a mean over the direction complementary to their propagation direction. The last situation we have considered corresponds to p 4 = 0, with transversal splitting on. This parameter measures the frequency of transversal splitting and is expected to increase with Re. Accordingly, the system can change from one-sided when p 4 is zero or small, to two-sided when it is large. The transition has indeed been observed and mean-field predictions were well observed far from the transition point. Unfortunately, while the effect of fluctuations close to that point was obvious, strong size effects have forbidden us to approach it and evaluate critical corrections. This is the subject of on-going work within the framework of finite-size scaling theory [2,27,28]. This follow-up should allow us to establish the universality class to which this transition belongs. Here, the left-right symmetry of localized turbulent bands with respect to the stream-wise direction is reminiscent of the up-down symmetry of magnetic systems at thermodynamic equilibrium, which may lead to conjecture the relevance of the 2D Ising class [26]. This class appears also applicable to coupled map lattices with the same up-down symmetry when updated asynchronously, one site after the other, close to randomization by thermal fluctuations. In contrast, another universality class is obtained with synchronous update [27]. Here, the situation is unclear: on the one hand, configurations are treated as a whole in a simulation step, which tips the scales in favor of a synchronous update model (in line with what is expected for a problem primitively formulated in terms of partial differential equations); on the other hand, spatial correlations generated by the deterministic dynamics governing the coupled map lattices are weakened by the independence of random drawings at the local scale, which can be viewed as a source of asynchrony in the probabilistic cellular automata. In its application to the symmetry-breaking bifurcation in channel flow, this uncertainty is, however, only of conceptual importance in view of size effects: owing to the large and unknown time-scale rescaling that allowed us to pass from flow structures to local agents in the model and to the narrowness of the region where critical corrections are expected, the mean-field interpretation developed in [14] appears amply sufficient. In the three cases that were considered in detail (specific cuts in the parameter space), the transitions remained continuous. However, this may not always be the case since there are known example of similar systems displaying transitions akin to first-order ones [24]. Even while keeping the same general frame, a plethora of circumstances of physical interest can be mimicked: propagation can be made more stochastic by decreasing p 5 , splitting rules not observed in channel flow can be considered, e.g., with p 3 or p 4 different from zero, etc., though it seems hard to anticipate situations where the universal features pointed out here would not hold. In contrast, when dealing with highly populated configurations, even in the simple nearest-neighbor von Neumann setting, rules can be made more complicated by introducing the neighborhood's degree of occupation. This introduction might help us to account also for the upper part of the transitional range of wall-bounded flows characterized by the emergence of regular patterns in the same stochastic framework [11]. The construction of the present model is, of course, fully adapted to the study of universality in the framework of the theory of critical phenomena in statistical physics, especially directed percolation. Still, we are confident that the kind of approach illustrated here brings a valuable contribution to the understanding of the transition to turbulence, by rationalizing its key ingredients in an easily accessible way. Acknowledgments: P.M. wants to thank H. Chaté (CEA-Saclay, Gif-sur-Yvette, France) for pointing out a model that could help one uncover the universal contents of the symmetry-breaking bifurcation on the same footing as the decay near Re g . Conflicts of Interest: The authors declare no conflict of interest. Abbreviations 1D/2D/3D One/two-dimensional (depending on 1/2 space coordinates) DP Directed percolation (stochastic competition between decay and contamination) LTB Localized turbulent bands DAH downstream active head (part of an LTB controlling its propagation) PCA Probabilistic cellular automata CML Coupled map lattice (spatially-discrete iterative model)	16,815.4	2020-11-29T00:00:00.000	[ "Physics", "Engineering" ]
Optimization technique combined with deep learning method for teeth recognition in dental panoramic radiographs Computer-assisted analysis of dental radiograph in dentistry is getting increasing attention from the researchers in recent years. This is mainly because it can successfully reduce human-made error due to stress, fatigue or lack of experience. Furthermore, it reduces diagnosis time and thus, improves overall efficiency and accuracy of dental care system. An automatic teeth recognition model is proposed here using residual network-based faster R-CNN technique. The detection result obtained from faster R-CNN is further refined by using a candidate optimization technique that evaluates both positional relationship and confidence score of the candidates. It achieves 0.974 and 0.981 mAPs for ResNet-50 and ResNet-101, respectively with faster R-CNN technique. The optimization technique further improves the results i.e. F1 score improves from 0.978 to 0.982 for ResNet-101. These results verify the proposed method’s ability to recognize teeth with high degree of accuracy. To test the feasibility and robustness of the model, a tenfold cross validation (CV) is presented in this paper. The result of tenfold CV effectively verifies the robustness of the model as the average F1 score obtained is more than 0.970. Thus, the proposed model can be used as a useful and reliable tool to assist dental care professionals in dentistry. Visual examination by dental care experts during dental treatment alone cannot provide sufficient information to diagnose a number of dental anomalies. It is because of their location in the mineralized tissues (bone and teeth). Thus, it is indispensable to use digital radiographs during dental treatment. Immediate availability of digital images, limited radiation dose and the possibility of applying image processing techniques (such as image enhancement and image registration) are some of the advantages of dental radiographs. However, the possibility of computer aided analysis of digital radiographs is one of the most important aspects of using dental radiographs. Dental panoramic radiograph or simply pantomograph is a type of dental radiographs that provides the dentist an unobstructed view of the whole dentition (both upper and lower jaws). It contains detail information of the dentomaxillofacial anatomy 1 . The usage of dental radiographs is growing day by day and therefore, it is highly desirable to assist dentist with computer-aided analysis. Automatic recognition of teeth from dental radiographs can be a great way to aid dentists in dental treatment. It will not only reduce workload from dental professionals but also reduce interpretation error and diagnosis time, and eventually will increase the efficiency of dental treatment. Usage of machine learning (ML) and computer vision techniques is not new in dental radiographs. Nomir et al. 2 proposed an automatic system that can identify people from dental radiographs. The proposed system can segment the radiographs into individual teeth automatically, represent and match teeth contours and finally provides matching scores between antemortem (AM) and postmortem (PM) teeth. Nassar et al. 3 created a prototype architecture of automated dental identification system (ADIS) to address the problem of postmortem identification through matching image feature. This matching problem was tackled by high level feature extraction in the primary step to expedite retrieval of potential matches followed by image comparison using inherent features of dental images. To detect areas of lesions in dental radiographs, a semiautomatic framework is proposed by Li et al. 4 by using level set method. The framework, at first, segments the radiograph into three meaningful regions. It was done by using two coupled level set functions. Then, an analysis scheme influenced by a color emphasis scheme prioritizes radiolucent areas automatically. After that the scheme employed average intensity profile based method to isolate and locate lesions in teeth. The framework improved the interpretation in a clinical settings and enables dentist Scientific Reports \| (2020) 10:19261 \| https://doi.org/10.1038/s41598-020-75887-9 www.nature.com/scientificreports/ to focus their attention on critical areas. Local singularity analysis based teeth segmentation was proposed by Lin et al. 5 in periapical radiographs in order to detect periapical lesion or periodontitis. This method works on four different stages that include adaptive power law transformation (as image enhancement technique), Hölder exponent (for local singularity analysis), Otsu's thresholding and connected component analysis (as tooth recognition) and finally, snake boundary and morphological operations (for tooth delineation). The overall accuracy (considering true positive) was found to be near 90 percent. Kavitha et al. 6 employed a new support vector machine (SVM) method to diagnose osteoporosis (a disease that increases the risk of fractures in bone) at early stage to reduce the risk of fractures. They utilized dental panoramic radiographs to measure the thin inferior cortices of mandible which is very useful to identify osteoporosis in women. The limitation of the conventional ML techniques in processing raw natural data requires careful engineering to construct feature extractor in order to transform the raw data into a suitable representation for detecting or classifying input patterns. This limitation was overcome effectively by the introduction of deep learning techniques. Deep learning techniques are representation learning based techniques that allow a machine to be fed with raw data and then process the data in different layers to automatically discover necessary representations to detect or classify input data. The main advantage of deep learning is that these layers of features are learned directly from the raw data by using a general purpose learning procedure instead of design constructed by the human engineers 7 . It has thus caused remarkable improvements in artificial intelligence. Deep learning techniques beat records in image 8 and speech recognitions 9 , supersedes other machine learning techniques in analyzing particle accelerator data 10 , predicting activity in potential drug molecules 11 , reconstructing brain circuits 12 , and produced promising results in natural language understanding 13 . Transfer learning based convolutional neural network (CNN) was utilized by Prajapati et al. 14 to classify three kinds of dental diseases from dental radiographs. They have utilized a pretrained VGG16 15 as feature detector. Lin et al. 16 proposed an algorithm based on CNN to automatically detect teeth and classify their conditions in panoramic dental radiographs. In order to increase the amount of data, different data augmentation techniques such as flipping and random cropping are used. They claimed to achieve accuracy around 90% using different image enhancement techniques along with CNN. Chen et al. 17 proposed faster R-CNN technique that included three post processing steps on dental periapical films. The post processing steps included a filtering system, a neural network model and a rule-base module to refine and supplement faster R-CNN. Although, the detection rate was exceptionally well, there classification result was only very close to the level of a junior scientist even after applying three post-processing steps. Tuzoff et al. 18 proposed a model that used faster R-CNN for teeth detection, VGG-16 based convolutional network for classification and heuristic-based algorithm for result refinement. Although, their heuristic algorithm heavily depended on the confidence scores produced by the convolution network, adequate performance analysis of the convolutional network for teeth classification was not present. Muramatsu et al. 19 proposed a fully convolutional network (FCN) based on GoogleNet to detect teeth. A ResNet-50 based pretrained network was then used to classify tooth by its type i.e. incisors, canines, premolars, and molars as well as three different tooth conditions. They tried to improve the classification result by introducing double input layers with multisized image data. However, their final classification result i.e. 93.2% for teeth classification was fallen short to the required accuracy needed for clinical implementation. This paper proposes a residual network based faster R-CNN algorithm in panoramic radiographs for automatic teeth recognition. Faster R-CNN object detector is the modified and upgraded version of R-CNN 20 and fast R-CNN 21 . The main advantage of faster R-CNN is that it does not need a separate algorithm for region proposals; rather the same convolution network is used for region proposal generation and object detection and hence much faster than its predecessors are. Two variants of trained residual network i.e. ResNet-50 and ResNet-101 are utilized in this paper to increase the effectiveness of the proposed system. Residual network is widely known for mitigating infamous vanishing gradients problem in deep network 22 . This paper proposes a candidate optimization algorithm based on prior knowledge of the dataset to further refine the detection results obtained by residual network based faster R-CNN. The proposed candidate optimization method considers both the position patterns of detected boxes as well as the confidence scores of the candidates given by the faster R-CNN algorithm to refine the detected boxes. The proposed method, thus, combines an optimization algorithm with deep learning technique for teeth recognition in dental panoramic radiographs. The rest of the paper is structured as follows; "Materials" section describes about the data i.e. dental radiographs used in this research. It also describes the tooth numbering systems used for testing the performance of the proposed method. "Proposed method" section presents the method and the architecture of the proposed model as well as explanation of training and test datasets. It also includes description of the candidate optimization method proposed in this paper. "Results and discussions" section comprises of results and discussions include simulation results. Finally, the paper is concluded with the "conclusion" section, providing the gist of the paper and possible future work. Materials A total of 1000 panoramic radiographs were collected for this research. The dimension of the images was around (1400-3100) × (800-1536) pixels and stored as a jpeg format. The images were collected by Narcohm Co. Ltd. from multiple dental clinics under the approval of each clinic. The authors obtained the images from Narcohm Co. Ltd. with permission. Also, the images were anonymously collected so that no additional information, like age, gender or height was revealed. Figure 1 shows an example of a collected dental panoramic radiograph. For the sake of training and validation, all the images were labeled by putting a rectangular bounding box around each tooth with proper roots and shape. The panoramic radiographs consisted of normal teeth, missing teeth, residual roots and dental implants. This paper followed universal tooth numbering (UTN) system. In universal Scientific Reports \| (2020) 10:19261 \| https://doi.org/10.1038/s41598-020-75887-9 www.nature.com/scientificreports/ tooth numbering system, teeth count starts from upper right part to upper left part as 1 to 16 and then lower left to lower right as 17 to 32. Fédération Dentaire Internationale (FDI) 23 and Palmer notation (PN) 24 are other two notable tooth numbering systems. Figure 2 illustrates the UTN and FDI system simultaneously. Annotated 1000 radiographs were split into total tenfolds, each containing 100 radiographs; (1) training dataset consisted of total ninefolds and (2) test dataset consisted of onefold. Therefore, training and test datasets were consisted of 900 radiographs and 100 radiographs, respectively. Training dataset was used to train faster R-CNN, whereas test dataset was used to analyze and validate the performance of the proposed method. Proposed method The proposed method consists of two steps; (i) candidate detection and (ii) optimization. The first step detects candidates from the panoramic radiographs using faster R-CNN and in the second step, the detected candidates are refined using an optimization method. Image-based CNN detectors are used in this research and therefore, a brief overview of these detectors are given at first. Next, deep learning-based faster R-CNN technique, its architecture and criteria of introducing transfer learning techniques are presented and discussed. Finally, the proposed optimization method based on prior knowledge is explained along with brief descriptions of performance evaluation metrics. www.nature.com/scientificreports/ Image-based CNN detectors and candidate detection using faster R-CNN. Mainly two types of image-based CNN detectors have been developed, they are (i) single-stage methods and (ii) two-stage methods. Both types of method utilize multiple feature maps of different resolutions for object detection. These feature maps are generated by a backbone network e.g. AlexNet 8 or ResNet 22 . The single-stage name came from the fact that these kind of methods performed directly into these multi-scale feature maps for object detection. In contrast, two-stage methods at first work on the feature maps to generate region proposals from the anchor boxes. Anchor boxes are a set of predefined bounding boxes with different aspect ratio. The network that generates region proposals is known as region proposal network (RPN) in faster R-CNN. RPN produces region proposals by predicting whether the anchor boxes contain an object or not (without classifying which object). Region proposals with best confidence scores are then processed into the second stage for further classification and regression. Thus, the region proposals are classified and regressed twice and that is why usually these kinds of methods achieve higher accuracy. However, the second stage computation adds an extra computational burden and the system, thus, tends to be less efficient and slow 25 . In this research, faster R-CNN 26 technique is used for automatic teeth recognition. Faster R-CNN is modified and updated version of fast R-CNN 21 . It utilizes two different modules; one is deep convolutional network also known as region proposal network (RPN) for region proposals and second is fast R-CNN object detector that utilizes proposed regions. Two modules, however, work as single unified network for object detection. The main advantage of faster R-CNN from its predecessors R-CNN 20 and fast R-CNN is that it successfully alleviate the problem of needing a separate algorithm for region proposals and thus enabling a cost effective region proposals. A single unified network is used for both region proposals and object detection. The technique won 1st place in several tracks e.g. ImageNet detection, ImageNet localization, COCO detection and COCO segmentation of ILSVRC and COCO competitions 26 . Transfer learning. Transfer learning is a technique used in both machine learning and deep learning problems to improve the learning performance of a particular task through transferring the knowledge gained from a different task that has already been learned. Usually it works best when the tasks are quite similar. However, it has been found that this technique works well even though the tasks are completely different. There are many pre-trained architectures that are trained on huge datasets such as AlexNet 8 , VGG-16 15 , VGG-19 15 , Inception-V3 27 , ResNet-50 22 and ResNet-101 22 . Probably the most popular such dataset is ImageNet. It contains millions of data sample to classify 1000 different categories. Transfer learning technique enables the researchers to train models with minimal training data by fetching architecture and weights from some popular pre-trained model. Furthermore, it drastically reduces the computational cost and therefore the training time. ResNet-50 & ResNet-101. In order to construct the proposed model, the ResNet-50 and the ResNet-101 architectures are adopted by the faster R-CNN framework separately, i.e. the proposed model utilizes both the architectures separately for teeth recognition task. He et al. 22 presented a framework based on residual learning to overcome the difficulty of training deeper neural network. The degradation problem was addressed by introducing a deep residual learning framework. It showed that optimizing residual mapping is easier than to optimize the original and therefore, gained accuracy easily from network with greater depth. Residual network won the 1st place in classification task of ILSVRC 2015 competition 22 . ResNet-50 consists of four stages with total 50 layers and hence its name. ResNet-101 is the deeper version of ResNet-50, consisting of additional 17 blocks (3-layer block) in the third stage that made it total 101 layers. The architecture of ResNet-50 and ResNet-101 is summarized in Table 1. Activation layer selection. In order to use a pre-trained model for a completely different task, few pre-processing steps should be considered. The steps include the removal of the original classifier, add a new classifier according to the task and fine tune the model 28 . There are three strategies to fine tune the model. www.nature.com/scientificreports/ (i) The first strategy is to train the entire model, i.e. use only the architecture of the pre-trained model and train the model according to the available dataset. In short, training from the scratch. In order to achieve sufficient accuracy, large dataset is required for this strategy. It also involves huge computational cost. (ii) The second option is to train some layers of the model, while leaving other frozen. In general, lower layers keep information about general features, whereas the higher layers keep information about specific features. As general features are problem independent, lower layers can be left frozen in case of small dataset. The training then only be done in the higher layers (problem dependent). However, when large dataset is available, overfitting does not become an issue and lower layers can also be train with the higher layers. (iii) Third option is to freeze all the convolutional layers and thus, use only the classifier. This option can should only be considered where dataset is small and sufficient computational power is unavailable. Based on the above strategies, activation layer 40 (activation_40_relu) is used as a feature extraction layer for ResNet-50 and activation layer res4b22 (res4b22_relu) for ResNet-101. Candidate optimization. The residual network based faster R-CNN together with careful selection of parameters can provide very good recognition performance. However, there may still be a good number of false positives including double detections for a single tooth. In order to cope up with this problem, a candidate optimization algorithm based on prior knowledge is proposed in this research. This model selects the best combination of candidates in order to filter out the false positives and thus improving overall efficiency of the model. Assume that tooth x ( 1 ≤ x ≤ 32 ) has N(x) candidates detected by faster R-CNN. And, in some case, all candidates are false positives. Therefore, the selection is to find the best combination of candidates in 32 x=1 (N(x) + 1) combinations. The selection is done by optimizing Eq. (1). In this equation, the first term evaluates the confidence score, and the second term evaluates the relational position from other teeth. where, P = p 1 , p 2 , . . . , p 32 is the combination pattern (T1 to T32), ω c and ω p are weights of confidence score and coordinate score, respectively. µ c p x is confidence score of candidate p x obtained from faster R-CNN in the range of [0-1], whereas µ p p x is the positional relationship score of candidate p x calculated by using the following equation: where, p x is the tooth candidate under consideration and δ p x , p y is a customized function created using the prior knowledge of the dataset. The function evaluates the horizontal distance between the tooth p x with its neighboring teeth p y [where, p y = p x−2 , p x−1 , p x+1 , p x+2 ]. It then calculates and assigns score for each tooth defined as coordinate score following the equation below: Figure 3 shows the mechanism of calculating the coordinate score. The different colors refer to the different teeth number. The weights of confidence score and coordinate score are two of the parameters of this algorithm and they should be selected carefully. There may have multiple candidates for a single tooth that the optimization algorithm should fix. In that case, the candidates are numbered in accordance with its confidence value, i.e. candidate tooth with higher confidence value will be numbered first. For example, tooth T1 has two candidates with confidence value 0.95 and 0.7, they will be denoted as C 1,1 and C 1,2 , respectively. The first subscript refers to the tooth number. And, C x,0 represents the missing tooth candidate. The candidate optimization processes in three steps. Step 1: Initialize pattern P max = C 1,1 , C 2,1 , . . . , C 32,1 by choosing the candidate with the highest confidence value for each tooth. When there are no candidate at tooth x, C x,0 is used as the candidate. The score of pattern P max is calculated using Eq. (8), and let the calculated score be S max . (1) www.nature.com/scientificreports/ Step 2: For every tooth x, try all candidates of C x , where C x = (C x,0 , C x,1 , C x, 2 …) and calculate the score using Eq. (8). Update S max , if new best combination is found, and set the candidate to P tmp. Step 3: If there are update for S max , P max is replaced by P tmp and return to Step 2. Else, the algorithm finds the best combination of all. 29 is calculated for each category of tooth to evaluate the candidate detection performance of the proposed method. At first, the detected boxes are compared with the ground truth boxes by calculating the intersection-over-union (IOU) as shown in Fig. 4 and defined as below The IOU threshold value is set as 0.5 i.e. if the IOU value is greater or equal to 0.5 then the detected box is considered as true positive, otherwise is considered as false positive. To calculate the evaluation index, i.e. AP, precision and recall are calculated using the equations as follows. where, TP is defined as the number of ground truth boxes that overlap with the detected boxes with IOU ≥ 0.5; FP is defined as the number of detected boxes that overlap with the ground truth boxes with IOU < 0.5, and FN is defined as the number of teeth that are not detected or detected with IOU < 0.5. Finally, the model is tested with a test dataset of 100 images. The above mentioned metrics are used to evaluate the detected boxes. www.nature.com/scientificreports/ The overall proposed model with candidate optimization algorithm is depicted in Fig. 5. To evaluate the performance of candidate optimization algorithm, F 1 score is calculated. F 1 score is the harmonic mean of precision and recall and is defined by the following equation: Results and discussion This section presents the overall simulation results using residual network based faster R-CNN for teeth recognition. The proposed method was implemented in MATLAB 2019a software and executed with Ryzen 7 2700 Eight-Core Processors (16 CPUs) with clock speed ~ 3.2 GHz. The training and testing were done with TITAN RTX 24 GB display memory (VRAM). Table 2 refers to the parameter settings of faster R-CNN for teeth recognition task. Total number of epoch was set as 10. As each training image corresponds to each iteration, thus total number of iteration was 9000. Number of regions to sample from each training image was set as 256, whereas number of strongest regions to be used for generating training samples was set as 2000. Negative overlap range and positive overlap range were set as [0-0.3] and [0.6-1], respectively. To better explain the results of the proposed model, total three cases were considered. Test case 1: Total 900 panoramic radiographs were used to train the network, while 100 images were used for testing. Both, ResNet-50 and ResNet-101 networks were implemented separately as the base networks of faster R-CNN for the evaluation purpose. Figure 6 is presenting the comparison of candidate detection results between ResNet-50 and ResNet-101 for each tooth category. It can be seen from the Fig. 7 that only T1 achieves less than 0.900 AP while using ResNet-50. Other than that the AP of other teeth is above 0.900 and the mAP is 0.974. On the other hand, results obtained using ResNet-101 shows extremely good detection performance as total seven teeth categories achieve maximum average precision and mAP is 0.981, which is better than the results obtained by ResNet-50. Table 3 presents precision and recall value for each tooth category. Some of the teeth categories achieved perfect recall values for ResNet-101 based faster R-CNN i.e. there were no false negative for those teeth categories. Figure 7 shows recall-precision curves for different teeth categories using ResNet-101. For better visualization, the curves are shown in four different figures, and each figure shows recall-precision curves for eight www.nature.com/scientificreports/ categories of teeth. Almost all of the curves show ideal behavior and visibly it can be seen that the convergence performance is close to that of an ideal one. The recognition performance obtained by the proposed model is close to the level of an expert dentist. Test case 2: Second test case examines the compatibility and the feasibility of using candidate optimization algorithm along with faster R-CNN technique to improve overall recognition result. The candidate optimization algorithm based on prior knowledge was implemented along with faster R-CNN technique to refine the obtained detected boxes by faster R-CNN. In this paper, two sets of weights were considered. In set A, weight of confidence score (ω c ) was set as 0.8, whereas the weight of positional relationship score (ω p ) was set as 0.2. In set B, both weights were selected as 0.5. Table 4 shows the recognition results after applying the candidate optimization algorithm. From the table, it is clear that for both the networks, the overall F 1 scores improves. For ResNet-50 and ResNet-101, F 1 score improves from 0.965 to maximum 0.976 and 0.978 to 0.983, respectively. The optimization technique also effectively balances the difference between precision and recall, which indicates that the algorithm is fully compatible with the model and successfully improves its robustness. Figure 8 visualizes the result given in Table 3. In terms of F 1 score, set A performed better than set B for ResNet-50, whereas it remained in balance for ResNet-101. Test case 3: In order to check the robustness of the proposed method, K-fold cross validation (CV) technique was utilized in this experiment. In this paper, K is equal to 10 i.e. the whole dataset is divided into 10 different folds. Total 10 runs were required to perform tenfold CV. Both, ResNet-50 and ResNet-101 were used with faster R-CNN, separately to perform the recognition task. The overall results consisting of AP of all teeth categories in all 10 test datasets for ResNet-50 and ResNet-101 are summarized in Fig. 9. The obtained results of 10-CV presented in Fig. 9 shows that the residual network based faster R-CNN performed quite strongly and consistently with the average of mAP is 0.958 for ResNet-50 and 0.960 for ResNet-101. Furthermore, the robustness www.nature.com/scientificreports/ of the method in different test data shows that it is clinically applicable. The comparison of different residual networks presented in Fig. 9 shows ResNet-101 performs better for 7 folds, whereas ResNet-50 performs better in 3 other folds. The lowest AP achieved by ResNet-50 is 0.800 for T27 in K7 fold and 0.840 for T1 in K7 fold by ResNet-101. However, most of the lower detections came from the eighth number fold (i.e. K8) for both of the residual networks. To assess the feasibility and robustness of the proposed model after applying candidate optimization algorithm, tenfold CV technique was performed and the results are presented in Table 5. After applying candidate optimization, the average F 1 score improves from 0.962 to 0.971 for ResNet-50 and 0.975 to 0.976 for ResNet-101, respectively. Furthermore, for all cases, the candidate optimization algorithm refined the detected boxes successfully and thus, improved the overall recognition performance. The stand-alone residual network based faster R-CNN performed exceedingly well in recognizing tooth by its number. Two kinds of residual networks i.e. ResNet-50 and ResNet-101 were used as base networks of faster R-CNN. ResNet-101 is deeper network than ResNet-50 and it performs marginally better than its shallower counterpart. Although, ResNet-101 performs better in terms of recognition, it is computationally costlier than ResNet-50. As their recognition performance is not much different, authors recommend using ResNet-50 as a base network when computational cost is a concern. The inclusion of candidate optimization algorithm further improves the recognition performance of the proposed model. However, the optimization parameters should be chosen carefully based on the dataset in order to have a good impact on the overall recognition results. Figure 10 www.nature.com/scientificreports/ Table 5. Results (in F 1 score) of teeth recognition after applying candidate optimization algorithm for tenfold cross validation. Bold values indicate the best results in that particular row (particular section). Conclusion This research proposes a method for automatic teeth recognition in dental panoramic radiographs. The method is based on candidate detection with residual network based faster R-CNN and candidate optimization using a prior knowledge. Two versions of residual network i.e. ResNet-50 and ResNet-101 are used as base networks for faster R-CNN separately. The combination of residual network with faster R-CNN method successfully performs the recognition task with a high degree of accuracy. It achieves maximum 0.980 mAP. A prior knowledge based candidate optimization technique is also incorporated to improve the overall recognition performance. The introduction of the optimization method improves the F 1 score from 0.965 to maximum 0.976 and 0.978 to 0.983 for ResNet-50 and ResNet-101, respectively. The K-fold cross validation technique is also implemented with and without candidate optimization technique that effectively verifies the feasibility and the robustness of the proposed method. The level of performance achieved by the proposed model is close to an expert dentist and thus, clinically implementable. Finally, it can be said that the proposed model can be used as a reliable and useful tool to assist dental care professionals in dentistry. In future, we plan to extend the current model to include automatic dental condition evaluation and prosthetic detection features. Received: 1 June 2020; Accepted: 19 October 2020 Figure 10. Successful teeth detection in noisy panoramic radiograph. The detected boxes were generated by MATLAB 2019a software.	6,830.6	2020-11-06T00:00:00.000	[ "Medicine", "Computer Science" ]
Kazoo training for L2 pronunciation practice and reduced foreign accentedness? ABSTRACT This paper presents a controlled pretest–training–posttest study on using a small membraphone instrument called kazoo for L2 pronunciation practice and reducing foreign accentedness. Learners were recorded for free speech before and after a six-week pronunciation and oral skills course including L2 prosody training with personal kazoos. A control group did not receive any kazoo training but did similar tasks in regular speech. Samples of all learners’ free speech before and after the course were rated for foreign accentedness by L1 speakers. In addition, learners taking part in the kazoo group were interviewed in focus groups to ascertain their views on kazoo training. The results imply that the kazoo has potential as a pronunciation teaching and learning tool. Even if a group-specific analysis did not yield clear results, developing on an individual level was more common in the kazoo group. Moreover, many of the learners found positive aspects about kazooing. However, problems occurred as well; these are also reported on in this paper to hopefully help others who wish to conduct further teaching experiments and develop specialised tasks and materials for kazoo pronunciation practice. Introduction Kazoo (see Picture 1) is a musical instrument, to which one hums in order to make a buzzing sound. Mere blowing into the kazoo does not create a sound. Using kazoos for second/foreign language pronunciation teaching and learning is probably not very common or even familiar to many language teachers. However, kazooing has been recommended for teaching speech prosody already in the 1970s (Gilbert 1978). The recommendation is based on the kazoo's buzzing sound blurring the individual speech sounds, hence drawing attention to what can be heard clearly: prosody. In addition, kazoos are cheap to buy and add to the variety of pronunciation teaching techniques. Even though pronunciation has gradually gained increased attention within the field of second language research, as stated by e.g. Derwing (2010) and Pennington (2015), practical studies on teaching techniques are still scarce and hence extremely welcome (Derwing and Munro 2015, 25). Thus, we decided to take up a classroom study to investigate the potential of kazoo training for reducing foreign accentedness and as a pronunciation teaching and learning tool in general. This paper presents findings from a teaching experiment, in which L2 prosody was taught with the kazoo in a controlled pretest-training-posttest setting. To study the effects of kazoo training, participants' productions of free speech before and after the teaching experiment were rated for foreign accentedness by native speakers. Also, the participants were interviewed to ascertain their experiences in training with the kazoo. Kazoo's buzzing sound is produced with a vibrating membrane in the instrument. Hence, the kazoo is one type of a membraphone. The membraphone technique was also used in the eunuch flute (or onion flute, or mirliton) in the sixteenth and seventeenth century. However, membraphones are known already from Ancient Greece and traditional African societies, where they were used for impersonating voices of the dead in ceremonies. (Montagu 2007, 6.) In its current form, kazoo is known as an American invention from the 1850s, although documented proof of the origin of the instrument is not available. Kazoos may be mainly used as children's toys, but they had a significant role in blues and jazz in the twentieth century. The kazoo is worth considering a serious instrument, as there are thousands of kazoo bands in the United States alone, and well-established composers and artists such as Leonard Bernstein and Ringo Starr have used kazoos in their music. (New York Times, February 6, 1993.) Most research on kazooing has been done in the fields of music and music therapy. Anyone who can hum can play the kazoo without prior musical training or experience, and because of this democratic nature of the kazoo, they are commonly used in music therapy (see e.g. Wetherick 2014). Research suggests that as kazoos mask the voice, many inhibitions, such as the natural embarrassment of exposing one's voice out loud, can be removed (Bodner and Gilboa 2006). In another context, studying music and music theory has been found to benefit from kazooing. For example, young music theory students were assumed as being too intimidated for singing out loud, so kazoos were introduced in a teaching experiment by Snodgrass (2007). The students were motivated to practise on the kazoos, and their singing was strong. They developed in singing melodies by sight (i.e. prima vista, straight from the notes without hearing or practising the melody), and the teacher was able to correct those who were slightly off-pitch. Moreover, the student feedback included a comment on hiding behind the kazoo making a student feel more at ease with their own voice. Benefits of the kazoo discovered in the fields of music therapy and music teaching could potentially help second and foreign language learners as well. Especially removing or decreasing the embarrassment of speaking out loud would be welcome as for learners with low willingness to communicate (WTC, McCroskey andBaer 1985, for L2: MacIntyre andCharos 1996;MacIntyre et al. 1998) and/or high language anxiety, in particular. As scholars have suggested, willingness to communicate should be considered an important learning goal (MacIntyre et al. 1998;Dörnyei 2003). Many learners do not feel at ease speaking a second/foreign language in front of a class (e.g. Kjellin 2002, 139-140). Therefore, methods enabling learners to practise their pronunciation without anxiety or embarrassment are needed, and kazoo training may be one option. Gilbert (1978) bases her recommendation for using kazoos in language learning and teaching on that most productive teaching should appeal to both sides of the brain; whereas the left hemisphere is crucial for language, the right hemisphere is important for melody. Current knowledge on speech lateralisation suggests that the issue is more complex (see e.g. Watkins and Devlin 2008; Peelle 2012). Nevertheless, Gilbert (1978) suggests that as pronunciation processes are spatial and melodic tasks, the verbal approach may be appealing to the wrong side of the brain. Thus, she feels musical aids should not be merely nice supplements but rather a fundamental teaching approach. Gilbert recommends kazoos for intonation and stress, but also sees the potential of using kazoos to demonstrate emotional patterns. She points out that learners may not hear intonation from normal speech very well, because they are struggling with strange sounds, word meaning, and difficult grammar all at once. The kazoo offers a speech model that is stripped of everything but intonation and stress. Gilbert (1978) names the kazoo "probably the best tool a pronunciation teacher could have" and is of the opinion that the kazoo gets faster results than e.g. visual cues (cf. Gilbert 1980 below). She also states that students having personal kazoos may increase the effects. Gilbert (1980) has studied the effect of using personal kazoos on the perception of prosody. In a series of pilot studies, she tested if it was possible to identify a language based on prosody alone and how perceptive and production training affect the identification. A buzz-like resynthesis of free speech in Japanese, Cantonese and English was played to speakers of those languages. The listeners were to identify in a pre-test and a post-test which language was spoken. The listeners were divided into a passive group, active group, and a control group. Between pre-test and post-test, the passive group received perception training in the buzz-like resynthesis of the three languages, whereas the active group also imitated the stimuli using the kazoo. The control group received no training. After several adjustments to the research design concerning the time span of the pre-test-trainingpost-test setting, listeners' L1, the amount of training, and the length of training sessions, Gilbert was able to find significantly more improvement in the identification by those who had trained with the kazoo. This result was achieved by using L1 Japanese listeners who received four 15minute training sessions within one week, which was the time between pre-test and post-test. Gilbert concludes that people can tell the difference between the given languages based on prosody alone, and that trainingespecially active mimicry with the kazoocan increase this ability. As Gilbert (1978) only studied the effects of kazoo training on receptive skills, we decided to investigate kazoo's potential for L2 speech production. In the present study, L2 Swedish was taught to L1 Finnish-speaking learners. Half of the participants aimed for the Sweden-Swedish variety, whereas the other half aimed for Finland-Swedish, which is a distinct variety spoken in Finland. The learners studied on a course where they could themselves choose their target variety. As kazoo training was used for the learning of prosody, we will now briefly present the most important prosodic challenges and learning goals for Finnish-speaking learners of Swedish. These are the realisation of stress, rhythmic pattern and tonal word accents (in Sweden-Swedish). The challenges are based on the differences between Finnish and Swedish prosody. In Sweden-Swedish, duration is the main correlate of stress: stressed syllables are long, while unstressed syllables are more or less short, or at least shorter than stressed syllables (Bruce 2010, 85;Kuronen and Tergujeff 2018). In Finnish, the main correlates of stress are fundamental frequency and amplitude, while duration has a minor role as a stress correlate compared to Sweden-Swedish, partly because length is used extensively for phonological contrasts (see e.g. Ullakonoja et al. 2014). Thus, there is a greater prosodicrhythmic or durational, to be exactdifference between the stressed and the unstressed part(s) of speech in Sweden-Swedish than in Finnish. The acquisition of this rhythmic feature in Sweden-Swedish seems to be of great importance for perceived foreign accentedness (Kuronen and Tergujeff 2017). In addition, Sweden-Swedish has two tonal word accents (see e.g. Riad 2014, 181-191), while Finnish has no lexical tones. As tonal word accents occur only in stressed syllables, this feature also adds to the prosodic difference between stressed and unstressed syllables in Sweden-Swedish. Previous research has discovered that aspects of Sweden-Swedish prosody are difficult but possible to learn for L1 speakers of Finnish (Kuronen 2015;Kuronen, Ullakonoja, and Kautonen 2016;Kuronen and Tergujeff 2018). Finland-Swedish differs in some prosodic respects from Sweden-Swedish (see e.g. Helgason, Ringen, and Suomi 2013). Finland-Swedish resembles Finnish regarding stress correlates: fundamental frequency and amplitude are the main correlates of stress, and duration plays a minor role (Vihanta, Leinonen, and Pitkänen 1990). Still, the difference between stressed and unstressed syllables is greater in Finland-Swedish than in Finnish. Thus, even though learning Finland-Swedish can be considered less demanding for L1 Finnish-speaking learners (see e.g. Kautonen 2017), both varieties include prosodic learning goals, making them suitable target varieties for the present study. Aim and research questions The aim of this study is to investigate learners' experiences and the effect of kazoo training on reducing foreign accentedness. Thus, a teaching experiment was arranged with a pretest-trainingposttest setting, including an experimental group and a control group. The following research questions were set for the study: (1) How do learners experience the kazoo as a learning tool? (2) Can kazoo training help reduce the rate of foreign accent in L2? Due to lack of previous research, it is difficult to predict the results. Gilbert's (1978Gilbert's ( , 1980 work is encouraging, and previous studies (e.g. Kuronen and Tergujeff 2018) on L1 Finnish-speaking learners' development in Swedish prosody gives reason to believe that reducing foreign accentedness in these learners is possible. However, prosody is not the only aspect of pronunciation that influences the perception of foreign accent. Several studies (e.g. Boyd, Abelin, and Dorriots 1999;Boyd and Bredänge 2013;Kuronen and Zetterholm 2017) point to individual sounds playing a role in foreign accentedness of Swedish, along with prosodic features. As training with the kazoo will obviously be of no help in developing the pronunciation of individual sounds, the possible effects of kazoo training will have to do with prosody alone. Participants The participants of the present study were advanced L1 Finnish-speaking adult learners of Swedish (n = 20) and L1 Swedish-speaking listeners (n = 77). The learners had studied Swedish for at least six years before entering their current educational context, in which they pursue a university degree in Swedish language and/or second language teaching. Half of the learners (n = 10) aimed for the Sweden-Swedish variety, whereas the other half (n = 10) targeted Finland-Swedish. Similarly, the native-speaker listeners represented the Sweden-Swedish (n = 37) and Finland-Swedish (n = 40) varieties. In addition, native speakers of Sweden-Swedish (n = 2) and Finland-Swedish (n = 2) were used as control speakers. All native speakerslisteners and control speakerswere recruited in educational contexts in Sweden and in Finland, and they were either experts or students in a language or related field (e.g. Swedish as a second language, language education, phonetics). Teaching experiment The learners took part in an intensive oral skills course in Swedish, which combined explicit teaching of Swedish phonology and phonetics (including segmental and prosodic features of speech), conscious pronunciation training, and more general oral skills practice. The approach was interactional, including discussion on different aspects of oral language skills in Swedish. The major differences between the Sweden-Swedish and Finland-Swedish varieties as for pronunciation were covered during the course, and the learners got individualised feedback on their pronunciation according to their choice of target variety. The course took six weeks altogether, including two 90-minute sessions per week. Teaching took place in a language laboratory, which enabled recording learners' speech. Listening to one's own pronunciation was used as one of the teaching methods. The participants (n = 20) were volunteers from three teaching groups of the course, following the same course plan as for the contents and the overall working methods. The sole difference was that the participants (n = 10) in Group 1 trained with personal kazoos. The control group (n = 10) was formed of learners from two other groups (Group 2 and Group 3) to reach a balance between participants targeting Sweden-Swedish and Finland-Swedish. Groups 1 and 2 were taught by the same instructor, but Group 3 was taught by another instructor due to staff arrangements that were beyond our control. However, this slight difference is not considered crucial as for the learning results, because the contents and the overall teaching methods were identical in all groups. As no explicit kazoo-playing materials or teaching tips are generally available, we decided to take on the recommendation to use kazoos, and applied them to the kind of materials and tasks we would have used in teaching also without the kazoo. Simply put, the kazoo training consisted of working on the same materials and tasks that the controls practised on without the kazoos. For example, stress patterns in short dialogues were discussed, marked to the text and read aloud. Kazoo training was restricted to sessions that dealt with prosody (mainly rhythm) and the kazoos were used for approximately 10 min per session. Altogether, the experimental group trained with the kazoos for approximately 60 min, which is equivalent to the amount of training the participants received in Gilbert (1980). We wish to emphasise that the instructor had no prior experience in actually using kazoos in teaching; however, he had familiarised himself with the recommendation in the literature and how to play the kazoo. In addition, ideas on how to use the kazoo in the teaching experiment were discussed among the authors of this paper. The idea of the experiment was to simulate a situation in which a teacher decides to try out a new method based on a recommendation but without special training or ready-made tasks or materials designed for the method. Data and procedure The learners were recorded for free speech and read-aloud tasks (not dealt with in this study; see Kuronen andTergujeff 2017, 2018) before and after the course. The free speech task consisted of picture story-telling (ca. 3 min), samples of which were used as material in a listening test conducted for the present study. The learners were asked to tell freely about two images, which both pictured a yard with children engaged in different activities, such as reading a book, playing ball, and riding a bicycle. The two images featured different children and different activities. The same images were used for both pretest and posttest, and identical materials were collected from four native speakers of Swedish to be used as control items in the listening test. Speech samples from all participants were chosen so that they represented the best 20 s of the speaker's free speech task before and after the course. This was done because many of the participants struggled with producing grammatically accurate free speech, and we did not wish ungrammatical structures to interfere with listeners' judgments. This may have been the case if the speech samples had featured the participants' first 20 s of the speaking task, which is a methodological choice commonly made in previous research (e.g. Munro and Derwing 1999;Saito, Trofimovich, and Isaacs 2016). In determining the best 20 s, criteria included grammatical accuracy and overall fluency. Speech samples were admitted to a listening test with 77 native speakers of Swedish. Learners who aspired for Sweden-Swedish pronunciation were rated by speakers of Central-Swedish from the Stockholm area, whereas those who pursued Finland-Swedish were rated by speakers of that variety. The listeners gave scalar evaluations of the strength of the speakers' foreign accent (accentedness ratings) and of their perception of how easy or difficult the speaker is to understand (comprehensibility ratings; for terminology, see e.g. Derwing and Munro 2015, 3-5). However, the comprehensibility ratings yielded such good scores already in the pretest that the learners could simply not make much progress in this respect. Hence, the comprehensibility ratings were disregarded in this study. The ratings were done on a 7-point scale (1 = no accent, 7 = extremely strong accent) with an online survey tool. All participants agreed with transmitting their 20-second sound samples over the Internet. In addition to the ratings, the experimental-group learners were interviewed in focus groups of 3-4 learners to explore their views on using the kazoo as a teaching/learning tool. The interviews were semi-structured and thematic, and they were conducted at the end of the teaching experiment. The interviews focused on pronunciation learning during the course and the teaching methods used. Specific questions about the kazoo were included, e.g. what the learners thought of its effects on learning, whether they practised on it at home, and whether they would use it themselves as teachers. Analyses Data obtained from the focus group interviews were robustly transcribed, and a qualitative thematic analysis was conducted on them in a researcher triangulation procedure. Researcher triangulation is a means of strengthening reliability of qualitative research. The idea is not to rely on a single researcher's interpretation but to co-construct the analysis together in an expert group (Denzin 1978, 303). The authors plus another expert specialising in questions of L2 Swedish pronunciation were involved in the triangulation. Each of the four experts first coded the transcripts for the following themes rising from the data: benefits, challenges, using kazoos in teaching, and the fun factor. Next, the themes were discussed by the experts, who then co-constructed the analysis. Quantitative data obtained from the accentedness ratings were admitted to statistical analyses to investigate the possible effect of training. Statistical significance of the difference between learners' pre-and posttest speech sample ratings were tested by using a paired-samples t-test. Effect sizes were calculated using Cohen's d, which indicates the difference between means in terms of standard deviation units (Cohen 1988). Effect sizes were interpreted following the general guidelines: small effect >0.2, medium effect 0.5-0.8, large effect 0.8-1.3, very large effect >1.3. Learner interviews Participants were divided on the usefulness of kazoo training: some found it truly helpful (Example 1), whereas some thought it did not benefit their pronunciation learning. However, many of the participants that were sceptical about the benefits of the kazoo for their own pronunciation learning recognised its potential for various aspects. They emphasised that they understood the idea and stated that it could have worked for someone else. Novelty of the method was seen as an asset (Example 2), and kazoos were welcomed to add variety to L2 teaching methods and learning activities. Intonation (Example 3), rhythm and stress were mentioned as factors that may benefit from kazoo trainingin both perception and production. Thus, kazoos were seen to promote focus on prosody, even though it was mentioned that simply humming without the kazoo might have had the same effect. Example 1. "Kyllä se niinku oikeesti on hyödyllinen, et sitä ois voinu melkein ehkä sit vähän enemmänki -" "It was like truly useful, and we could have perhaps actually used it even more -" Example 2. "Ihan hauska idea ja tavallaan kyl siinä vähän ehkä just ehkä se sen niinkun menetelmän uuden-tavallaan niinku uutuus toi siihen semmosta lisää, et sen tavallaan huomas että hei tääki on tapa huomata-niinkun oppia tota painoa tossa lauseessa. Mut sit mä en osaa sit sanoo sitte kuinka kovaa se tehos." "Pretty nice idea, and in a way, the method being new, the novelty added to it, and you noticed that hey, this is one way to notice like stress in the sentence. But I don't know how well it worked." Example 3. "mä huomasin sen intonaation paljon paremmin [kazoolla] ku pelkästä puheesta." "I noticed the intonation a lot better [with the kazoo] compared to regular speech." Many learners stated in the interviews that practising with the kazoo was fun. It sounded funny in a positive sense but also invoked some negative fooling around in class (Example 4). In connection with mere humming having the same effect as kazooing, it was stated that practising with the kazoo is more fun (Example 5) and is more concrete. Learners also joked about using their personal kazoos for May Day Carnival celebrations, which is a clear indication of kazoos appealing to their sense of humour. Example 4. "Se meni vähän itellä sellaseksi hassutteluksi että tuli puhalleltua sellasia kaikenlaisia muita ääniä ku niitä varsinaisia repliikkejä mitä piti puhallella." "For me it slipped into fooling around a little bit. I whistled all kinds of sounds other than the lines we were supposed to." Example 5. "-Mä vaan sitten hymisin [ilman kazoota] sen saman homman. Se niinku mun mielestä ajoi saman asian mutta ei se nyt. Just onhan se sillä kazoolla on se nyt vähän hauskempaa kun se kuulostaa hassulta." "Then I just hummed the same thing [without the kazoo]. I thought it had the same effect but no, I mean it's more fun with the kazoo because it sounds funny." As the learners taking part in the kazoo experiment were mainly future language teachers, they were asked whether they considered the method something they would use in their own teaching. Learners were generally open to the idea of using kazoos, but they had trouble thinking of suitable learner groups for kazooing. Some considered kazoos suited for motivating young learners (Example 6), but also for adults that were ready to hurl themselves into playful activities (Example 7). Some suspected lower and upper secondary school pupils reluctant to take part in kazooing, and pupils fooling around too much with the kazoos was a common concern. Further, large groups and groups with restless pupils were mentioned as contexts in which kazoos might be better avoided. One interviewee pointed out that the method is worth trying, if even a couple of learners find it beneficial. Kazoos were recommended for demonstrating stress; some found the method more suitable for introducing topics rather than for actual pronunciation practice. Example 6. "Voi se ehkä motivoida jotain muksuja jotka ei muuten ehkä oo niin motivoituneita ruotsin opiskeluun jossain yläasteella tai jossain, tai sit se voi myös mennä aivan läskiks." "It might motivate some kids who are otherwise not so motivated to study Swedish, like in lower secondary school or so. Or it can turn into sheer fooling around." Example 7. "Ja sit täällä yliopistossa ehkä ainut, silleen alkaa olemaan sen verran avoin että okei no helvetti jos tää nyt pitäs toimia ni yritetää ees." "And here in the university, it might be the only place. One starts to be open, like okay, what the hell, if this should work, let's at least give it a try." When reflecting on the possible use of kazoo in their own teaching, the participants emphasised the importance of extensive instruction. This was probably due to themselves having severe challenges with getting started on the kazoo. Many of the participants indicated in the interviews that they found it difficult at first to produce the buzzing sound. It was mentioned, for example, that producing the sound required speaking somewhat louder than they were used to speaking, which was straining and distracted their focus (Example 8). Other challenges had to do with the amount of practice: kazoos were used too little in class (Example 9), and some participants were poorly motivated to kazoo in their free time. Participants also seemed to be afraid of or suffering from negative attitudes against the kazoo in their surroundings: one was worried about what their neighbours would think if they heard the kazoo through the walls, and another stated that their girlfriend did not like them kazooing at home. In addition, some participants found it challenging to take kazooing seriously. Example 8. "Se keskittyminen meni mulla ehkä siihen että mä saan sieltä pillistä jonkun äänen eikä siihen että miten mä teen ne äänenpainot. Mä ymmärrän sen että siitä olis siihen hyötyä ja sit ku mä vähän opin käyttään sitä ni se meni paremmin." "My concentration was on getting a sound out of it and not on how to produce the stress. I understand that it would be beneficial for stress, and when I learnt to use it a little, it worked better." Example 9. "siis periaatteessa oikeesti kyllä ymmärsin sen idean mikä siinä ois et se varmasti niinku auttanu, et loppupeleissä me ei kyl hirveesti taidettu käyttää sitä." "In principle, I did understand the idea of it and I'm sure it would have helped, but in the end, I don't think we used it that much." Accentedness ratings Overall, accentedness ratings did not reveal much difference between pretest and posttest speech samples. There was only a small but statistically significant difference between pretest and posttest mean scores for both the kazoo group and the control group. On average, the kazoo group was rated slightly less accented in posttest than in pretest, whereas the controls were rated slightly more accented (see Table 1). The difference between pretest and posttest was statistically significant for both groups, but effect sizes proved minimal. This means that the differences between pretest and posttest cannot be explained by chance, but the actual effect of the differences is only minute. Hence, we cannot draw conclusions based on the group comparisons despite the opposite directions of development, but will next zoom in to explore the ratings on the level of individual learners. Learner-specific accentedness ratings demonstrated that not everyone presented a change in their accentedness as a result of the six-week intensive course, which is clearly reflected in the group-specific results presented above. In fact, only 11 out of 20 learners reached a statistically significant difference between pretest and posttest scores. Detailed results describing the development of these 11 learners are presented in Table 2. The rest of the learners cannot be taken into account, as there was (practically) no difference between pretest and posttest mean scores, and the minor differences may be explained by chance; however, see Appendix for all results. As for the learners who went over a statistically significant development, only 5 learners were able to reduce their rate of foreign accent. The 6 remaining learners developed in the opposite direction. Effect sizes varied from small to large; in most cases, the effect size was medium or large. Reduced rate of foreign accentedness was found mainly among kazooers: 4 out of 5 learners reducing their accentedness were kazooers, meaning that only one control made this development. Out of the 6 learners that made a change in the opposite direction, 2 were kazooers and 4 were controls. Thus, it appears that kazooers succeeded better in reducing their foreign accent. In addition, reducing foreign accent seems to have worked slightly better with learners who targeted Finland-Swedish: 3 out of 4 kazooers targeting Finland-Swedish received lower mean scores for accentedness in the Table 1. Group-specific pre-and posttest mean scores for foreign accentedness as judged by native-speaker listeners on a 7-point scale (1 = no accent, 7 = extremely strong accent). posttest, whereas all 3 controls targeting Finland-Swedish were judged to have a higher rate of accentedness. This was confirmed by a paired-samples t-test: those aiming for Sweden-Swedish did not present a statistically significant difference between their pretest and posttest mean scores, whereas those aiming for Finland-Swedish did (see Table 3). Moreover, learners of Finland-Swedish made the change for the positive. The effect size was small. These results were obtained using all data; not only learners who made a statistically significant change from pretest to posttest. Discussion Group-specific analyses, i.e. kazooers vs. controls, did not indicate that kazoo-playing would clearly lower the rate of foreign accentedness. Overall, many of the learners did not demonstrate a statistically significant difference between their pretest and posttest scores, but their foreign accentedness was judged more or less the same both before and after the intervention. These results highlight the difference between our experiment and Gilbert's (1980) study: Gilbert's aim to reach receptive effects with 60 min of kazoo training was successful, whereas our investigation on reducing foreign accentedness in free speech with the same amount of kazooing did not yield such clear results. To apply the learnt skills into free speech would possibly have needed more practice; learners themselves stated in the interview that the kazoo was used only little (Examples 1 and 9). In a previous study (Kuronen and Tergujeff 2018), the same learners were able to make substantial prosodic developments during this teaching experiment, when assessed from read-aloud speech. This clearly indicates the difficulty of applying skills to free speech. Also, not all development may always appear directly after an intervention. On the contrary, learners may be less confident (Lintunen and Mäkilähde 2015) and feel puzzled by all new information and find it hard to concentrate on several things simultaneously, which was even mentioned in the interviews. Applying learnt features into free speech may take time. Hence, the lack of delayed posttest is an obvious limitation of the present study. Target variety specific analyses revealed that learners of Sweden-Swedish hardly demonstrated any change in their rate of foreign accentedness, whereas learners of Finland-Swedish did: kazoo group targeting Finland-Swedish reduced their foreign accentedness, but the control group developed for the negative. Finland-Swedish was assumed an easier target variety to begin with, because it has less prosodic learning goals compared to Sweden-Swedish. Learners could hence focus on rhythm and disregard tonal aspects, which means they have been under lighter cognitive strain during the learning process (cf. Ullakonoja 2011). The results imply that kazoo training works better if the prosodic learning goals are limited to rhythmor perhaps any one aspectwhereas focusing on two aspects (in this case: rhythm and tonal features) simultaneously may be too challenging. When analysed on the level of individual learners, reducing foreign accentedness was clearly more common among learners in the kazoo group. Thus, we see potential in the kazoo as a pronunciation teaching and learning toolespecially if used extensively and with tasks and materials explicitly designed for the kazoo. We wish to reiterate that our teaching experiment was conducted without much previous experience in practising on the kazoo, and that kazooing was applied to normal pronunciation tasks that otherwise were done in regular speech. Still, almost all learners who reduced their foreign accentedness were in the kazoo group. It is also noteworthy that segments as well play a role in perceived accentedness (e.g. Boyd, Abelin, and Dorriots 1999; Boyd and Bredänge Table 3. Target variety specific pre-and posttest mean scores for foreign accentedness as judged by native-speaker listeners on a 7-point scale (1 = no accent, 7 = extremely strong accent). Group Pretest 2013; Kuronen and Zetterholm 2017), and as many learners practising on the kazoo succeeded in reducing their accentedness, kazooing cannot be seen to negatively affect the development of segments. The possibility of the kazoo hindering or disturbing the development of segments occurred to us in planning the experiment but fortunately proved an unnecessary concern. However, if kazoo training is practised extensively, teachers should make sure that sufficient attention is paid to individual sounds as well. Interviews with the learners also support kazoo's potential as a pronunciation teaching and learning tool. Even though some of the learners did not feel they personally benefitted from kazooing, they were positive about the idea; some found it more suitable to other learner groups, and some found it valuable in terms of its novelty effect and in adding to the range of classroom activities. In our view, new ideas and techniques are welcome to language instruction, as varied activities are crucial for motivation and maintaining interest. In addition, the interviews clearly demonstrated that training on the kazoo was viewed as fun, which also increases motivation. The fact that kazoos mask the user's voice has been found to remove inhibitions such as embarrassment of speaking or singing in front of others (Bodner and Gilboa 2006;Snodgrass 2007). However, this issue was not raised by any of the interviewed learners. Hence, this study does not offer support to kazoos affecting language anxiety or willingness to communicate (WTC). Still, we want to point out that this aspect was not explicitly addressed by the interviewer, and we do not know whether any of the learners suffered from language anxiety or low WTC. Thus, the possibilities of kazoo training contributing to this matter calls for further investigation. Overall, more classroom research is needed on pronunciation teaching techniques, and we hope that our study will inspire teachers and researchers to conduct teaching experiments. As for future research regarding the kazoo, we would like to see kazoo teaching experiments conducted on children, who might form a better target group for kazooing despite the interviewees' concern for children losing focus and fooling around; children's natural enthusiasm for playful activities might deepen their engagement, and their lower level of target language proficiency might provide more opportunities for development. For example, participants with lower proficiency could be tested for comprehensibility, which is more important for communication in comparison to accentedness, and which is heavily influenced by prosody. In the present study, the participants were too advanced for this research purpose; hence, it was quite impossible for them to develop their comprehensibility, as they were very easy to understand already before the intervention. Future research should also focus on other languages, as this study only investigated the learning of Swedish by L1 Finnish speakers. The benefits of kazooing might come forth clearer in research designs with two languages diverging more radically from each other in regards to rhythm, e.g. L1 Spanish speakers learning Swedish or e.g. English. Conclusion The aim of this study was to shed light on using the kazoo as a learning and teaching tool for second language pronunciation and its possibilities for reducing foreign accentedness in second language (free) speech. The research entailed a practical teaching experiment with a controlled pretest-training-posttest setting, followed by listener ratings for foreign accentedness and learner interviews in focus groups. Participants were divided in their views on the usefulness of kazooing as for their own pronunciation learning, but they saw potential in it and could use it themselves in teaching with a suitable learner group. Kazooing was seen as a fun activity, adding variety to teaching techniques. Nevertheless, no indication of change as for willingness to communicate or embarrassment to speak out loud came up in the interviews. Accentedness ratings suggest that kazoo training has the potential of reducing foreign accentedness, especially in target languages with a limited amount of prosodic learning goals. At least in this study, kazooing mainly helped the learners of Finland-Swedish, whose prosodic learning goals were restricted to aspects of rhythm, whereas reducing accentedness was rare among learners of Sweden-Swedish, trying to tackle learning goals both rhythmic and tonal. In addition, this study gives no reason to suspect that small-scale kazooing would hinder the development of segmental features of pronunciation. Based on the results of this study, we recommend that second and foreign language teachers familiarise themselves with kazooing for prosodic training and try it out with suitable learner groups. Kazoos are cheap to buy and add to the variety of classroom activities. Moreover, their novelty effect and fun factor may increase motivation among learners. What our experiment taught us is not to overestimate the ease of playing the kazoo. Even adults can have difficulties in learning how to make a sound. Hence, learners would benefit from careful instruction on how to get started on the kazoo, and tasks designed for how to control the sound. In addition, teachers should test different kazoo types and choose the kind for their learners that plays with relatively weak volume and air stream. Based on our experience, the cheap plastic kazoos are the easiest to play. Disclosure statement No potential conflict of interest was reported by the authors. Funding This study is part of research project Fokus på uttalsinlärningen med svenska som mål-och källspråk (FOKUS, see www. jyu.fi/fokus), which is funded by the Society of Swedish Literature in Finland (SLS) 2015-2019. The foundation also supported Author A's work with an additional grant. Notes on contributors PhD Elina Tergujeff is a Postdoctoral fellow in the Department of Language and Communication Studies, University of Jyväskylä. Her research interests include second language pronunciation teaching and learning, comprehensibility, intelligibility and accentedness. As for this study, Tergujeff's research activities were funded by the Society of Swedish Literature in Finland (SLS). PhD, Senior lecturer Mikko Kuronen does research and teaches in the Department of Language and Communication Studies, University of Jyväskylä. His research interests include Swedish and Finnish phonetics and second language pronunciation learning. Kuronen is the principal investigator and head of a project concentrating on pronunciation teaching and learning with Swedish as L1 and L2 (www.jyu.fi/fokus). The project is funded by the Society of Swedish Literature in Finland (SLS). MA Maria Kautonen is a doctoral student in Swedish language in the Department of Language and Communication Studies, University of Jyväskylä. In her PhD study, she focuses on L2 pronunciation learning and assessment with Finland-Swedish as L2. Her research has been funded by the Society of Swedish Literature in Finland (SLS) 2015-2018.	8,972.6	2020-05-26T00:00:00.000	[ "Linguistics" ]
Covariance Matrix Reconstruction via a Subspace Method and Spatial Spectral Estimation for Robust Adaptive Beamforming Several robust adaptive beamforming (RAB) algorithms based on interference plus noise covariance matrix (INCM) reconstruction have been recently proposed, which can enhance the robustness of beamforming algorithms when certain mismatches occur in the model. However, some approaches ignore the resolution of the Capon spectral estimator (CSE), leading to reconstruction errors. This paper proposes a novel RAB algorithm formulated using the subspace projection method and spatial spectral estimation (SSE), which is named INCM-SSE. First, without using the CSE, the subspace projection matrix (SPM) is obtained through the integral of the angular sector where the signal of interest (SOI) is located. Subsequently, after estimating the direction of arrival (DOA) of incident signals using the multiple signal classi ﬁ cation (MUSIC) algorithm, we project the sample covariance matrix (SCM) onto the SPM to eliminate the SOI in ﬂ uence. Then, the estimation method of interference powers is derived. Moreover, the INCM is reconstructed based on the estimated powers and steering vector (SV) of interferences. The SV of the SOI is optimized by solving a quadratic convex optimization problem. The INCM-SSE algorithm not only employs SSE to improve the angular resolution but also reduces the in ﬂ uence of the SOI component by using SPM. Simulation results indicate that the proposed method is robust against various types of mismatches, thus achieving superior overall performance. Introduction Steering array antennas forming beam patterns according to certain optimal criteria are called smart antennas; they are also regarded as adaptive array antennas [1][2][3]. This description means that a computer can control the performance of the antennas, greatly improving the performance of the array system [4]. Adaptive array processing technology can adjust the weighting vector of array antennas in real time according to the signal environment. It employs adaptive beamforming algorithms to achieve a certain gain in the signal of interest (SOI) direction and suppress interferences in their directions [5][6][7]. The technology has been widely applied to the fields of radar, sonar, microphone array speech processing, medical imaging, internet of things, wireless communication, astronomy, and seismology [8][9][10][11]. However, adaptive beamforming algorithms are extremely sensitive to signal source mismatches and array geometry errors (e.g., signal direction error, sensor position perturba-tion, amplitude, and phase error). In particular, when components of the SOI exist in the snapshots, the beamformer performance is severely degraded [12]. Therefore, improving the robustness of beamforming is necessary for adaptive array processing. Over the decades, numerous robust adaptive beamforming (RAB) algorithms (e.g., diagonal loading method, eigen-subspace approach, uncertainty set constraint technology, and covariance matrix taper algorithm [13][14][15]) have been proposed. The foregoing methods are effective for resolving signal direction error and sensor position perturbation. Moreover, they are capable of improving the signal-tointerference-plus-noise ratio (SINR) under certain mismatch conditions. However, the SOI consistently exists in the sample covariance matrix (SCM) of these algorithms, limiting their performance at a high signal-to-noise ratio (SNR). Over the past few years, a number of RAB algorithms based on interference plus noise covariance matrix (INCM) reconstruction (in which the SOI component is effectively removed) have been proposed. The INCM reconstruction, first reported in [16], was achieved by combining the Capon spectral estimator (CSE) and integral of the region separated from the angular sector of the SOI. Subsequently, the steering vector (SV) of the SOI is modified by solving a quadratically constrained quadratic programming (QCQP) problem. The work in [17] describes an efficient RAB method based on the spatial power spectrum sampling (SPSS) approach. The computational complexity of the INCM-SPSS method is lower than that reported in [16]. However, this technique requires numerous sensors to achieve a similar performance as that in [16]. Reference [18] proposes a subspace-based RAB algorithm via residual noise elimination and interference power estimation. Although its performance is superior to other algorithms, the proposed algorithm has numerous integrals, matrix multiplications, and matrix inversion operations that increase the running time. The INCM reconstruction achieved through the interference SV lying at the intersection of two subspaces is presented in [19]. However, the foregoing requires multiple eigen-decomposition operations, increasing the complexity of the algorithm. Lowcomplexity spatial sampling processing and virtually received SV-based algorithm are proposed in [20]. The foregoing is realized by a projection matrix orthogonal to the signal subspace and retains interference plus noise in a higher dimension. This method can greatly improve the efficiency of the INCM reconstruction and beamforming robustness. The work in [21] describes a RAB method based on the reduction of interference matrix and the SV of SOI optimization; however, its performance degrades as the number of interferences increases. Nonetheless, some of the approaches mentioned here (such as those reported in [18,19,22]) ignore the resolution of the CSE. When the direction of arrival (DOA) of the interference is close, the CSE is unable to effectively identify it, leading to reconstruction errors. Although adjacent interferences can be solved by null broadening technology [23], this algorithm type may lead to an increase in the sidelobe level or broadening of the mainlobe of the beam pattern. Hence, its robustness (particularly when various signal model mismatches occur) remains to be verified. In this paper, a novel RAB method via the subspace projection method and spatial spectral estimation (SSE) is proposed. Different from previous RAB methods, the proposed algorithm obtains the subspace projection matrix (SPM) through the integral of the angular sector where the SOI is located and employs a high-resolution SSE algorithm to estimate the DOA of signals. In addition, the SV of the SOI is estimated by solving a quadratic convex optimization problem. The remaining portions of this paper are organized as follows. In Section 2, the signal model and problem background are briefly described. Section 3 introduces the details of the proposed method. The simulation and analysis results are presented in Section 4. Finally, the conclusions of the study are summarized in the last section. Signal Model and Problem Background Consider a uniform linear array (ULA) composed of M omnidirectional sensors, as shown in Figure 1. Suppose that L + 1 narrowband uncorrelated sources from a far field impinge on the ULA with directions θ 0 , θ 1 , ⋯, θ L . The M × 1 array observation vector at instant k has the following general form: where x s ðkÞ = a 0 ðθ 0 Þs 0 ðkÞ, x i ðkÞ = ∑ L l=1 a l ðθ l Þs l ðkÞ; nðkÞ denote the components of the SOI, interferences, and noise, respectively; s l ðkÞ, l = 0, 1, ⋯, L, denotes the signal waveform; nðkÞ represents the additive white Gaussian noise with zero mean and variance σ 2 n ; and a 0 ðθ 0 Þ and a l ðθ l Þ are the corresponding SVs of the SOI and interferences, respectively. These can be modeled as a l θ l ð Þ = 1, e j2πd sin θ l /λ ,⋯,e j2π M−1 where d = λ/2 represents the interelement spacing, λ denotes the wavelength of incident signals, and ð⋅Þ T is the transposed operator. The output of the beamformer at time k can be written as where w is the excitation weight vector of M sensors and ð⋅Þ H denotes the Hermitian transpose. The complex vector, w, can be solved by maximizing the output SINR of the ULA: where σ 2 0 = Efs 0 ðkÞs 0 * ðkÞg denotes the power of the SOI, ð⋅Þ * is the conjugate operator, and Eð⋅Þ represents the expectation operator. The M × M theoretical INCM is R i+n , which can be expressed as ... Wireless Communications and Mobile Computing where σ 2 l and I denote the power of the lth interference and M × M identity matrix, respectively. The maximization problem of (4) can be formulated as which is known as the Capon beamformer. The solution to (6) is given by The optimal SINR can be obtained by substituting (7) into (4): The beam pattern of the Capon beamformer is given as follows: where θ represents the angular sector, aðθÞ is the SV of θ, and lg ð⋅Þ denotes the logarithmic operator at base 10. However, a 0 ðθ 0 Þ may be inaccurate, and the exact covariance matrix, R i+n , is unavailable in practical applications. The two are usually replaced by the nominal SV (â 0 ðθ 0 Þ) and SCM, given byR where K is the number of snapshots received. Correspondingly, the complex weight vector can be obtained by the sample matrix inversion (SMI) beamformer expressed as The beam pattern and output power of the SMI beamformer are given by respectively. Proposed Algorithm 3.1. SPM Construction. References [18,24] present the estimate of residual noise power using the integral of the region where noise alone is present. This noise power can be removed from the Capon spectrum to derive the true powers of sources. The accurate covariance matrix of the SOI is reconstructed by integrating the region where the SOI is located: where b σ 2 n and Θ s are the estimated noise power and SOI region, respectively;â s ðθ s Þ represents the SV of θ s ; and S is the corresponding number of sample points. However, the above approach severely complicates the computations of the algorithm. By simply employing (15),R s can be obtained: Subsequently, the SPM is constructed through the eigendecomposition ofR s :R where ε m arranged in descending order denote the eigenvalues ofR s , v m represent the eigenvectors corresponding to ε m , U is the SPM, and V is composed of eigenvectors corresponding to B largest eigenvalues ofR s . Evidently,â s ðθ s Þ is orthogonal to the subspace of U, i.e., Z norm = kUâ s ðθ s Þk 2 2 = 0 , where k⋅k 2 represents the Euclidean norm. The comparability of the SPM obtained by the integral using and without using the CSE is shown in the following simulation such that a series of complex operations caused by the CSE can be avoided. As shown in Figures 2 and 3, the values of Z norm obtained by both methods are similar, and when B = 3, the results of Θ s are virtually zero. Therefore, matrix U obtained through (15) can also be adopted as the SPM to eliminate the SOI in the SCM. SSE. Most of the previous algorithms employ the CSE to estimate the power or DOA of sources for reconstructing the INCM [25]. However, the angular resolution of the CSE is unsatisfactory; consequently, it cannot distinguish the DOA when the incident angles of two signals are adjacent. Therefore, different from previous approaches, a high-resolution SSE algorithm is used here. In recent years, numerous SSE algorithms have been proposed, including the multiple signal classification (MUSIC) algorithm, estimation of signal parameters via rotational invariant technique (ESPRIT) algorithm and their variants, principle of maximum entropy power spectrum, sparse Bayesian learning, support vector machine, and discrete Fourier transform-based methods [26][27][28][29][30]. This subsection presents the MUSIC algorithm, which is classical and facile to implement. It is utilized to estimate the spatial spectrum, which can be formulated as follows: where G N represents the noise subspace composed of eigenvectors corresponding to several smaller eigenvalues ofR x . To demonstrate that the resolution of the MUSIC algorithm is higher than that of the CSE, a simulation is presented in this section. Assume that a desired 25 dB signal arrives at the ULA from −5°; three interference signals arrive at −30°, 24°, and 28°with 25 dB, K = 50, and the number of sources is supposed to be known. The spatial spectra estimated by the MUSIC algorithm and CSE are shown in Figure 4; the curves are the average of 100 Monte Carlo [31] experiments of the two algorithms. Figure 4 indicates that when the incident angles of two signals are close, the CSE cannot distinguish them. This means that the DOA estimated by the CSE inevitably leads to a mismatch in the INCM reconstruction and a reduction in the output SINR. However, the resolution of the MUSIC algorithm is higher than that of the CSE; hence, the algorithm can identify two adjacent signals to reconstruct the INCM more accurately. INCM Reconstruction and SV Estimation . MatrixR x can also be expressed aŝ By combining the foregoing with kU H a 0 ðθ 0 Þk 2 2 = 0, the SOI component can be eliminated through the SPM, U. Then, the following formula is derived: whereR i represents the matrix that preserves interferences and b σ 2 n is equal to the minimum eigenvalue ofR x , which can be obtained through the MUSIC algorithm without introducing further computational complexity. Functions ðU H Þ −1 and U −1 can reduce the influence of SPM on the interferences after the SOI is eliminated. By ignoring the influence of SPM on (20),R i can be approximately expressed in the following form: where A is the M × L matrix composed of a l ðθ l Þ, l = 1, 2, ⋯ , L; each diagonal element in Λ denotes the interference power. Accordingly, Equation (22) Wireless Communications and Mobile Computing where b Λ is a diagonal matrix that represents the estimated powers andÂ denotes the matrix composed of the estimated SV of interferences:Â In (23),â l represents the lth interference SV; it can be optimized by where δ is the bound value; e θ l , l = 1, 2, ⋯, L, represents the multiple spectral peaks obtained by searching in space using (18); andã l ð e θ l Þ denotes the SV corresponding to e θ l . Equation (24) can be solved using the Lagrange multiplier method:â where μ satisfies the following constraint: Different from the technique presented in [18], we used (22) to calculate the interference powers. The use of this equation avoids the matrix inversion operation to maintain low computational complexity. In the case of adjacent interferences, the estimated power is observed to be greater than that of two incident angles far apart, as demonstrated by the simulations reported in [22]. For RAB algorithms, the overestimated power facilitates the formation of a deeper null, which can better suppress the interference and improve the output SINR of the array. In this case, the overestimated power does not require correction. Accordingly, the INCM can be reconstructed based on (22) and (23): Next, the SV of the SOI is optimized. The convergence of nominal SVâ 0 ðθ 0 Þ to the SV of interferences can be prohibited byâ H 0 ðθ 0 ÞUU Hâ 0 ðθ 0 Þ = 0. However, the above equality constraint is nonconvex; thus, the following constraint is set: whereâ 0 ðθ 0 Þ + e ⊥ denotes the actual SV and e ⊥ (orthogonal toâ 0 ðθ 0 Þ) represents the error between the actual SV and a 0 ðθ 0 Þ. The right-hand side of (28) is a value tending to zero, ensuring that the termâ 0 ðθ 0 Þ + e ⊥ is restricted to the SOI region. The solution of e ⊥ can be derived as follows: The foregoing is a QCQP problem and can be solved by the CVX toolbox [32]. Then, the SV of the SOI and weight vector can be formulated as respectively. 3.4. Steps of Proposed Algorithm. In our approach, the computational complexity of the SPM construction is Oðmax Simulation Results To evaluate the capability of the proposed algorithm, a ULA with M = 10 is considered. The SOI is presumed to impinge on the ULA from −5°, the angular sector of the SOI is set to Θ s = ½−9°,−1°, and three interferences arrive at the array from −30°, 24°, and 28°with 25 dB. When comparing the performance of robust beamformers in terms of the output SINR versus input SNR, the number of snapshots is set as K = 50. In comparing the SINR and snapshots, the SNR is fixed to 10 dB. All angular sectors are sampled with a fixed [14], INCM-LINE method [16], INCM-SPSS approach [17], INCM-RNE beamformer [18], INCM-IMR beamformer [21], and INCM-ISV method [22]. For the EIG algorithm, the number of sources is assumed to be known. To achieve superior performance, accurate estimates of SV and INCM are necessary. In this regard, the proposed method applies a high-resolution SSE algorithm to estimate the DOA of interferences, and the INCM is reconstructed through the theoretical form. Consequently, its null depths in the interference directions are deeper than those of the other methods, and its performance is similar to that of the optimal beamformer. Although nulls can be formed by the INCM-IMR beamformer in all interference directions, their depths are shallow. Moreover, because the INCM-LINE beamformer integrates the region where the interferences are located, the reconstructed INCM contains all the information of the three interferences, ensuring that nulls are formed in all the interference directions. However, their null depths are shallower than that created by the proposed algorithm. Example 2 (Mismatch due to Look Direction Error). In the second example, the impact of random look direction mismatch on the performance of beamformers is considered. For each simulation run, the DOA mismatches of incident signals are uniformly distributed in ½−4°, 4°. Figure 7 illustrates the output SINR versus input SNR of all beamfor-mers. This figure shows that the proposed algorithm and INCM-IMR achieve the highest output SINR at a low SNR. Further, the performance of the EIG and INCM-IMR methods is severely degraded at a high SNR. The output SINR of the proposed method and INCM-LINE are higher than those of the other algorithms because the reconstructed Step 1. Calculate the SCM applying (10), and estimate the spatial spectrum by (18). Step 5. Calculate the complex vector of the proposed algorithm through (31). Algorithm 1: Steps of the proposed algorithm. Example 3 (Mismatch due to Array Geometry Error). In the third example, a scenario with a mismatch due to array geometry error is examined. Generally, the array geometry error is modeled as the sensor position perturbation, which is assumed to be uniformly distributed in ½−0:05λ, 0:05λ (i.e., the sensor position, d, used in each run is changed in interval ½d − 0:05λ, d + 0:05λ). Figure 9 depicts the output SINR versus input SNR of all beamformers. The figure shows that the proposed method performs best among the tested algorithms; its performance is similar to that of the INCM-IMR method at a low SNR. Moreover, the foregoing mismatch type has different extents of impact on the other algorithms (e.g., it severely degrades the performance of INCM-RNE). Figure 10 displays the curves of output SINR versus the number of snapshots. It also indicates that the proposed method has stable performance provided that the number of snapshots exceeds 30. However, the output SINR of INCM-LINE considerably fluctuates with the change in the number of snapshots. Example 4 (Mismatch due to Gain and Phase Error). In this example, the influence of arbitrary gain and phase perturbation on the RAB algorithms is considered. The m term in (2) can be modeled as where κ m denotes the gain error derived from the Gaussian distribution Nð1, 0:05 2 Þ and β m is the phase perturbation distributed in Nð0, ð5°Þ 2 Þ. Figures 11 and 12 4.5. Example 5 (Mismatch due to Incoherent Local Scattering). In the last example, the incoherent local scattering of the SOI is considered. Assume that the SOI has a time-varying signature. Accordingly, the model can be expressed as where θ η represents the scattering angle drawn from the Gaussian generator Nðθ 0 , 4°Þ, a η ðθ η Þ is the SV corresponding to θ η , and s η ðkÞ, η = 0, 1, 2, 3, 4, is independently drawn from Nð0, 1Þ. BecauseR s is no longer a rank-one matrix, the SINR in this scenario can be written as follows: Wireless Communications and Mobile Computing which can be maximized by where Φf⋅g represents the eigenvector of the matrix corresponding to the largest eigenvalue. Figures 13 and 14 show the output SINR versus input SNR and output SINR versus the number of snapshots, respectively. The two figures indicate that the proposed approach outperforms the other algorithms at a high SNR and considerably approximates the optimal SINR. A slight perturbation is observed in the proposed algorithm when the number of snapshots is less than 30. However, as the number of snapshots increases, the output SINR quickly exceeds that of the INCM-LINE beamformer. Conclusions This paper proposes a novel RAB algorithm formulated using the subspace projection method and SSE. Without using the CSE, the proposed method first obtains the SPM through the integral of the region where the SOI is located. Based on this, the SCM is projected onto the SPM to remove the SOI component. Then, the DOA of the sources is estimated by the MUSIC algorithm, and the estimation method of interference powers is derived. Subsequently, the INCM is reconstructed by the estimated powers, and the SVs of the interferences and SOI are optimized by solving a QCQP problem. Finally, the weight vector of the proposed algorithm is obtained. Simulation results indicate that the proposed RAB algorithm is superior to the other algorithms in terms of the overall performance in the following cases: comparison of beam patterns, look direction mismatch, sensor position perturbation, gain and phase error, and incoherent local scattering. In future work, the INCM-SSE algorithm can be applied to the RAB problem for other geometries of antenna arrays such as L-shape, planar, and circular array. Data Availability The data used to support the findings of this study are included within the article. Conflicts of Interest The authors declare that there is no conflict of interests regarding the publication of this paper.	5,069.4	2022-05-14T00:00:00.000	[ "Engineering", "Computer Science" ]
5G Ultra-Reliable Low-Latency Communication Implementation Challenges and Operational Issues with IoT Devices : To meet the diverse industrial and market demands, the International Telecommunication Union (ITU) has classiﬁed the ﬁfth-generation (5G) into ultra-reliable low latency communications (URLLC), enhanced mobile broadband (eMBB), and massive machine-type communications (mMTC). Researchers conducted studies to achieve the implementation of the mentioned distributions e ﬃ ciently, within the available spectrum. This paper aims to highlight the importance of URLLC in accordance with the approaching era of technology and industry requirements. While highlighting a few implementation issues of URLLC, concerns for the Internet of things (IoT) devices that depend on the low latency and reliable communications of URLLC are also addressed. In this paper, the recent progress of 3rd Generation Partnership Project (3GPP) standardization and the implementation of URLLC are included. Finally, the research areas that are open for further investigation in URLLC implementation are highlighted, and e ﬃ cient implementation of URLLC is discussed. Introduction With the everyday increase in data traffic requirements ranging from mission-critical to massive machine connectivity, the anticipation of the fifth-generation (5G) is growing at an exponential rate. Although such requirements open new doors to exciting features and business models, the provision of these requirements for such intense traffic and diverse services ( Figure 1) remains a challenge for the telecom industry. Introduction With the everyday increase in data traffic requirements ranging from mission-critical to massive machine connectivity, the anticipation of the fifth-generation (5G) is growing at an exponential rate. Although such requirements open new doors to exciting features and business models, the provision of these requirements for such intense traffic and diverse services ( Figure 1) remains a challenge for the telecom industry. The new NR facilitates communications between the base station and the user/mobile device. With such rapid changes in technology and user demands, NR is designed not only to facilitate the three defined categories of ITU, but also to provide space for future growth in technology, as shown in Figure 3. The three basic categories formulated by ITU for 5G [3] are as follows: 1 enhanced mobile broadband (eMBB), 2 ultra-reliable low-latency communications (URLLC), and 3 massive machine-type communications (mMTC). Importance of URLLC Although all three categories defined by ITU have their significance as shown in Table 1. URLLC has recently been very intriguing for researchers worldwide. A field experiment carried out by the NTT DOCOMO Inc. and Huawei on URLLC showed quite positive results [5]. The flexible frame structure of 5G new radio (NR) [2]. URLLC, ultra-reliable low latency communications; eMBB, enhanced mobile broadband; mMTC, massive machine-type communications; IoT, Internet of Things; NB, Narrowband. The new NR facilitates communications between the base station and the user/mobile device. With such rapid changes in technology and user demands, NR is designed not only to facilitate the three defined categories of ITU, but also to provide space for future growth in technology, as shown in Figure 3. The three basic categories formulated by ITU for 5G [3] are as follows: 1. massive machine-type communications (mMTC). These challenges are further enhanced by elements such as expectations of node-to-node or machine-to-machine communications requirements, the uncertainty of topology, diverse application requirements, backward compatibility, user equipment resource limitations, and the rapidly increasing number of devices. These elements exacerbate the technical complications of the implementation of future 5G networks. One of the crucial supporting technologies for the implementation of 5G is new radio (NR) [1], a new radio access technology. NR is the new radio interface and access method that is developed to facilitate the growing requirements of 5G in the coming future [2]. As shown in Figure 2, NR provides a flexible frame structure to support all 5G service requirements defined by the International Telecommunication Union (ITU). The flexible frame structure of 5G new radio (NR) [2]. URLLC, ultra-reliable low latency communications; eMBB, enhanced mobile broadband; mMTC, massive machine-type communications; IoT, Internet of Things; NB, Narrowband. The new NR facilitates communications between the base station and the user/mobile device. With such rapid changes in technology and user demands, NR is designed not only to facilitate the three defined categories of ITU, but also to provide space for future growth in technology, as shown in Figure 3. The three basic categories formulated by ITU for 5G [3] are as follows: 1 enhanced mobile broadband (eMBB), 2 ultra-reliable low-latency communications (URLLC), and 3 massive machine-type communications (mMTC). Importance of URLLC Although all three categories defined by ITU have their significance as shown in Table 1. URLLC has recently been very intriguing for researchers worldwide. A field experiment carried out by the NTT DOCOMO Inc. and Huawei on URLLC showed quite positive results [5]. Importance of URLLC Although all three categories defined by ITU have their significance as shown in Table 1. URLLC has recently been very intriguing for researchers worldwide. A field experiment carried out by the NTT DOCOMO Inc. and Huawei on URLLC showed quite positive results [5]. [3,4]. URLLC, ultra-reliable low latency communications; eMBB, enhanced mobile broadband; mMTC, massive machine-type communications. Category Basic Features eMBB eMBB focuses on a higher data rate, with a large payload and prolonged internet connectivity based applications. Potential applications could include cloud office/gaming, virtual/ augmented reality (VR/AR) and three-dimension/ultra-high-definition (3D/UHD) video. URLLC URLLC focuses on an ultra-responsive connection with ultra-low latency. The data rate is not expected to be very high in URLLC, but offers high mobility. Potential applications of URLLC include industrial automation, autonomous driving, mission-critical applications, and remote medical assistance. mMTC mMTC focus on providing connectivity to a large number of devices (IoTs), but with low reliability. It can provide long-range communication with energy efficiency and asynchronous access. Such features are very suitable for low power devices in a massive quantity. Achieving URLLC requirements is quite a challenge for 5G networks and will require massive modifications to the system design of the current telecom infrastructure. Owing to the encouraging results achieved with URLLC, it can play an integral role in the 5G era. Although current user requirements are initially based on high bandwidth, latency and reliability are also expected to play a vital role in real-time applications and mission-critical networks. Table 2 clearly highlights the importance of reliability and latency in future applications. Table 2. Importance of reliability and low-latency as per industrial/user requirements [6,7]. Industry Application Importance of Reliability and Low Latency Medical and Health Care Remote surgery/patient diagnosis. Remote surgery or remote patient's diagnosis might be carried out with the help of a robot. In such cases, the reliability of data transmitted as instruction for robot needs to be ultra-reliable because even a slight latency or delay could be very harmful to the patient. Media/ Entertainment/ Business Live reporting of an event, live sports events, online gaming, cloud-based entertainment (VR/AR). With the help of technology, the entire world is shrinking in terms of communications. Users desire to be up to date on world events and entertainment in real-time. Even in terms of business, the delay could make a huge impact on trades carried out in the world. In online gaming, the lag could be very frustrating for gamers. Transport Drone-based delivery, remote driving, self-driven cars, traffic management, sub-station management (system synchronization, traffic management) Through new features and attractions for users such as Amazon Prime Air [8] to deliver orders, it is very important for drones to respond in real-time. Similar to Amazon Prime Air, Google's self-driven car (WAYMO) [9] is quite important for the future automobile industry. The importance of reliability and latency is self-explanatory in such projects. Industrial Automation Control systems, automated assembly lines with robots, machine status reports, process surveillance, power grid management. In order to maximize productivity, industries have moved toward automation. Higher reliability and productivity can be obtained by replacing humans with robots in the manufacturing process. Apart from the manufacturing industry, the agriculture, journalism, and education sectors have also moved towards automation [10]. In the mentioned industrial areas, reliability will be a key factor. Such as that the automated car assembly line must have minimum latency to keep up with the moving tray and high reliability to avoid any damage to the car parts during assembly. Quality of Service (QoS) for URLLC URLLC-focused applications require an end-to-end (E2E) delivery of data with reliability, security, and minimum latency. Such requirements have driven the 3rd Generation Partnership Project (3GPP) to set desired quality of service (QoS) requirements such as an air interface latency of 1 ms and 99.999% system reliability for URLLC [11]. These QoS requirements for URLLC, depending on its various applications, are shown in Table 3. Table 3. Expected quality of service (QoS) requirements for URLLC [7,12]. Industry Error Rate/Reliability Latency (ms) Augmented/Virtual Reality 10 −3 -10 −5 5-10 Autonomies/guided vehicle The channel quality and lack of dedicated bandwidth can be an obstacle to meeting the desired latency requirement for URLLC [13]. To achieve the desired reliability in URLLC is also a challenge. As several mobile applications rely on different methods of retransmission, the retransmission of data in URLLC can degrade latency [11], unless the retransmission methods are designed as per URLLC requirements. The current 4G long-term evolution (LTE) and NR hybrid automatic repeat request (HARQ) are not quite appropriate to handle URLLC requirements. These methods depend on the complete or partial retransmission (RTX) of the error packet with additional time for HARQ processing, which is not suitable for time-critical applications [14]. An alternative method to achieve a low block error rate (BLER) and avoid RTX is to allocate high resources to the system; however, this might result in poor system capacity and low spectral efficiency. Therefore, there exists a trade-off between reliability and latency, which can be based on the application requirements. The physical layer plays a major role to achieve such a low latency and reliability; however, three major concerns exist [15]. First, system overhead in term of channel access, user schedule, and allocation of resources should be minimized. Second, the packet error probability should be minimized to achieve lower latency because the retransmission of packets can affect the latency, as mentioned earlier. Third, the transmission of URLLC packets should be prioritized, and they should be transmitted as soon as they are generated. Although the requirements mentioned by 3GPP and ITU for URLLC are based on one-way communication, latency should be defined on the basis of E2E communication [16]. Coexistence with eMBB The emerging 5G network must provide services to diversified applications with different requirements. Applications relying on URLLC require low latency with high reliability, whereas eMBB requires high data rates. For the existence of URLLC and eMBB in the same physical resource, as shown in Figure 4, an efficient coexistence method is needed to maintain the required QoS. Such coexistence on the same radio spectrum will open doors to new concerns in the scheduling optimization [17]. The proposed agile 5G frame structure [18] shows promising results for URLLC latency requirements by utilizing different transmission time intervals (TTIs) for URLLC and eMMB to meet their desired spectral efficiencies (SE). For example, URLLC traffic can be scheduled on a smaller TTI duration to achieve its low latency goal, and eMBB traffic can be scheduled with a long TTI duration The proposed agile 5G frame structure [18] shows promising results for URLLC latency requirements by utilizing different transmission time intervals (TTIs) for URLLC and eMMB to meet their desired spectral efficiencies (SE). For example, URLLC traffic can be scheduled on a smaller TTI duration to achieve its low latency goal, and eMBB traffic can be scheduled with a long TTI duration to maintain its extreme SE requirements. However, such a case will bring an additional overhead to the control signaling, which can result in the degradation of the control channel (CCH) capacity [17]. URLLC Packet Design Packet design is one of the key issues in URLLC. With an effective packet structure, the latency can be minimized in terms of packet processing time and packet transmission time [15]. Packet processing involves the time of acquiring a packet, accessing channel information, extracting scheduling (control) information, decoding the packet, and checking errors. As per URLLC requirements, the 5G NR system employs a non-square-shaped packet in the frequency domain with polar code for the control channel and low-density parity-check (LDPC) for the data channel to minimize the transmit latency. However, in LTE, a square-shaped packet is generally utilized for effective spectrum utilization [19]. URLLC Scheduling The scheduling of an unexpected packet generation by URLLC is one of the most significant issues. When user data arrives, it is stored in a user-specific transmission buffer, as illustrated in Figure 5. The transmission of each packet takes no less than one TTI. However, radio channel conditions, payload size, and availability of resources may force scheduling to increase the TTI of a packet [20]. The proposed agile 5G frame structure [18] shows promising results for URLLC latency requirements by utilizing different transmission time intervals (TTIs) for URLLC and eMMB to meet their desired spectral efficiencies (SE). For example, URLLC traffic can be scheduled on a smaller TTI duration to achieve its low latency goal, and eMBB traffic can be scheduled with a long TTI duration to maintain its extreme SE requirements. However, such a case will bring an additional overhead to the control signaling, which can result in the degradation of the control channel (CCH) capacity [17]. URLLC Packet Design Packet design is one of the key issues in URLLC. With an effective packet structure, the latency can be minimized in terms of packet processing time and packet transmission time [15]. Packet processing involves the time of acquiring a packet, accessing channel information, extracting scheduling (control) information, decoding the packet, and checking errors. As per URLLC requirements, the 5G NR system employs a non-square-shaped packet in the frequency domain with polar code for the control channel and low-density parity-check (LDPC) for the data channel to minimize the transmit latency. However, in LTE, a square-shaped packet is generally utilized for effective spectrum utilization [19]. URLLC Scheduling The scheduling of an unexpected packet generation by URLLC is one of the most significant issues. When user data arrives, it is stored in a user-specific transmission buffer, as illustrated in Figure 5. The transmission of each packet takes no less than one TTI. However, radio channel conditions, payload size, and availability of resources may force scheduling to increase the TTI of a packet [20]. The NR defined by 3GPP for 5G has proposed two scheduling schemes, instant scheduling and reservation-based scheduling, to handle URLLC packets [20]. The instant scheduling approach proposes to facilitate URLLC packets whenever they are generated. Therefore, this scheduling can interrupt the ongoing data transmission. Consequently, this approach can result in a drastic degradation of other services. The reservation-based scheduling is further divided into two types, semi-static and dynamic reservation, for effective handling of packets. Both approaches use a reservation-based frame for URLLC, which results in overheads in the control signaling. In the case of no URLLC data, the reserved slot may be wasted. Figure 5. User-specific transmission buffer [20]. UE, user equipment. BS The NR defined by 3GPP for 5G has proposed two scheduling schemes, instant scheduling and reservation-based scheduling, to handle URLLC packets [20]. The instant scheduling approach proposes to facilitate URLLC packets whenever they are generated. Therefore, this scheduling can interrupt the ongoing data transmission. Consequently, this approach can result in a drastic degradation of other services. The reservation-based scheduling is further divided into two types, semi-static and dynamic reservation, for effective handling of packets. Both approaches use a reservation-based frame for URLLC, which results in overheads in the control signaling. In the case of no URLLC data, the reserved slot may be wasted. Energy Efficiency Concern for End-User Device Most of the wireless devices employ a sleep mode operation to save energy. The devices need to act immediately upon receiving a packet from a network to avoid any delay. Similarly, the devices periodically check awaiting packets on the network to avoid latency [21]. The current energy-saving states defined for user equipment (UE) are not suitable for URLLC-based service. However, the UE can lose battery drastically as a result of the high frequency of data checks over the network. Handover Issues for URLLC Handover (handoff) is one of the most integral parts of any telecom infrastructure. NR for 5G must be able to support the mobility requirements illustrated in Table 4. In 5G NR, the basic handover process is quite similar to that of the LTE handover [23]. NR supports handover at two different levels to manage seamless handover. Cell level mobility is managed using a radio resource control (RRC) layer in the same way as in the LTE handover. In addition, the beam level mobility is handled using physical and medium access control (MAC) layers without involving RRC for low latency [24]. As NR adopts the same handover signaling procedures as the LTE [25], it inherits two unresolved issues of mobility robustness and mobility interruption time (MIT). Because of the mentioned concerns, enhancements in NR are considered and proposed to achieve zero handover interruption time (HIT) and handover failure (HOF) for URLLC [24]. Further studies are still needed to satisfy the requirements of URLLC QoS. Error Handling Owing to the faulty nature of the data channel [26], the handling of packet errors is another issue to meet along with the defined latency requirements for URLLC; for example, the 1 ms latency deadline [17]. The current LTE provides a very low error rate at the cost of higher latency, which is not suitable for URLLC [27]. As shown in Figure 6, when data arrives at the base station (BS) buffer, a request for resource grant (RG) is transmitted to the target UE. As data is received, the UE decodes the data and responds with either a positive or negative acknowledgment (ACK/NACK) based on the success of the data decoding. In a case where the UE fails to respond within the allocated time, the BS retransmits the data. Compared with the LTE, URLLC operates in a shorter TTI and requires a faster response from the UE to avoid retransmission. Wireless channel impairments can be another concern. If the BS does not receive an ACK/NACK within the mentioned deadline as a result of channel fading, the BS retransmits the data. This can increase latency, and hence the wastage of resources [27]. To overcome this issue, the stronger channel coding and multiple antenna technologies can be considered. However, the stronger channel coding may require longer decoding latency, for example, more iterations in an iterative channel decoding scheme. As a result of high power consumption and space limitation, multiple antennas cannot be equipped on the Internet of things (IoT) devices. Beamforming and mmWave Frequency Communications The next-generation mobile networks will operate with mmWave frequencies in order to increase bandwidth. In mmWave communications, beamforming between the UE and the BS becomes an important aspect, as shown in Figure 7. Beamforming and mmWave Frequency Communications The next-generation mobile networks will operate with mmWave frequencies in order to increase bandwidth. In mmWave communications, beamforming between the UE and the BS becomes an important aspect, as shown in Figure 7. 222 GHz link using LTE [32]. In 3GPP, the deployment of NR networks can be in standalone (SA) and devices over a wireless connection [33]. A massive multiple-input multiple-output (MIMO) 233 technology has recently become quite applicable to manage a massive number of devices. However, 234 the struggle to meet latency and reliability requirements remains problematic. The process of beam selection can affect the E2E performance and the QoS [28]. In order to counter the key issues of achieving precise beamforming for performance enhancement [29], the 3GPP NR standard included new MAC and physical layer (PHY) features. The new MAC and PHY features support directional communications [30], inter-network, and multi-network mechanism for LTE [31]. Despite the additional standards, some of the issues related with the directional communications and multi-connectivity are still unsolved. In the directional link, the requirement for precise beamforming can affect E2E performance. On the other hand, the provided solution [31] for multi-connectivity improves the mmWave network's E2E performance by merging a reliable sub-6 GHz link using LTE [32]. In 3GPP, the deployment of NR networks can be in standalone (SA) and non-standalone (NSA) modes. In the SA mode, the NR core and radio access network (RAN) are included, while LTE evolved packet core (EPC) and the LTE RAN are used in the NSA mode. However, the practical implementation of such systems to support beam management with SA and NSA is still an open issue [30]. Role of URLLC in Operating IoT Although mMTC is specifically categorized and designed to meet the IoT requirements, URLLC holds the key ingredients for effective IoT operations. When multiple operators control time-critical devices remotely, as shown in Figure 8, the latency and reliability play a vital role in the smooth operation of the IoT devices. It is quite challenging to operate mission-critical and real-time IoT devices over a wireless connection [33]. A massive multiple-input multiple-output (MIMO) technology has recently become quite applicable to manage a massive number of devices. However, the struggle to meet latency and reliability requirements remains problematic. Many potential real-time IoT operational issues can be overcome with the integration of the tactile Internet [34], URLLC, and MIMO radio access technologies. In the following subsections, some of the basic URLLC disputes are explained when operating IoT devices. URLLC and Massive Device Connectivity Present mobile services and specifications are not completely equipped to deliver URLLC costeffectively at scale [35]. Furthermore, they lack the capacity to deliver a reliable low latency communication to multiple users at the same time. It is particularly difficult to ensure link-level reliability and latency over a wide area and in a remote scenario, as shown in Figure 4. As wide-area cases involve many elements such as transitional nodes, backhaul, core/cloud, and fronthaul, they can play a vital role in degrading latency. However, the resources such as energy and computing power of IoT devices also play a vital role when operating over URLLC. To meet the latency requirements for URLLC, the IoT devices are forced to utilize excessive power and processing ability that is not appropriate for the life span of IoT Many potential real-time IoT operational issues can be overcome with the integration of the tactile Internet [34], URLLC, and MIMO radio access technologies. In the following subsections, some of the basic URLLC disputes are explained when operating IoT devices. URLLC and Massive Device Connectivity Present mobile services and specifications are not completely equipped to deliver URLLC cost-effectively at scale [35]. Furthermore, they lack the capacity to deliver a reliable low latency communication to multiple users at the same time. It is particularly difficult to ensure link-level reliability and latency over a wide area and in a remote scenario, as shown in Figure 4. As wide-area cases involve many elements such as transitional nodes, backhaul, core/cloud, and fronthaul, they can play a vital role in degrading latency. However, the resources such as energy and computing power of IoT devices also play a vital role when operating over URLLC. To meet the latency requirements for URLLC, the IoT devices are forced to utilize excessive power and processing ability that is not appropriate for the life span of IoT devices. However, most of the IoT devices have limited resources [36]. On-Device Artificial Intelligence and URLLC Traditionally, communication networks are designed with the concept of achieving high data rates with centralized management of resources. To accomplish the upcoming extreme latency and reliability requirements, the communication network architecture is now being pushed to be more non-centric and proactive. Most of the IoT devices are designed to be remotely controlled or to operate in a limited non-complex environment. However, some of the machines/applications require machine learning (ML) or artificial intelligence (AI) in order to be more effective and efficient to achieve the goals of the applications. Clearly, the customary machine learning approach based on the centralized architecture, as shown in Figure 9, is not very suitable for delicate latency applications [21]. However, most of the IoT devices have limited resources [36], and such devices may not be able to carry out ML or AI-based algorithms effectively while meeting the latency requirements. Consequently, researchers are investigating decentralized approaches such as distributed ML or AI on edge that involve collective problem solving [37]. Even with on-device machine learning, devices require a significant amount of storage and computational ability, which most of the IoT devices lack. URLLC and Massive Device Connectivity Present mobile services and specifications are not completely equipped to deliver URLLC costeffectively at scale [35]. Furthermore, they lack the capacity to deliver a reliable low latency communication to multiple users at the same time. It is particularly difficult to ensure link-level reliability and latency over a wide area and in a remote scenario, as shown in Figure 4. As wide-area cases involve many elements such as transitional nodes, backhaul, core/cloud, and fronthaul, they can play a vital role in degrading latency. However, the resources such as energy and computing power of IoT devices also play a vital role when operating over URLLC. To meet the latency requirements for URLLC, the IoT devices are forced to utilize excessive power and processing ability that is not appropriate for the life span of IoT devices. However, most of the IoT devices have limited resources [36]. On-Device Artificial Intelligence and URLLC Traditionally, communication networks are designed with the concept of achieving high data rates with centralized management of resources. To accomplish the upcoming extreme latency and reliability requirements, the communication network architecture is now being pushed to be more non-centric and proactive. Most of the IoT devices are designed to be remotely controlled or to operate in a limited non-complex environment. However, some of the machines/applications require machine learning (ML) or artificial intelligence (AI) in order to be more effective and efficient to achieve the goals of the applications. Clearly, the customary machine learning approach based on the centralized architecture, as shown in Figure 9, is not very suitable for delicate latency applications [21]. However, most of the IoT devices have limited resources [36], and such devices may not be able to carry out ML or AIbased algorithms effectively while meeting the latency requirements. Consequently, researchers are investigating decentralized approaches such as distributed ML or AI on edge that involve collective problem solving [37]. Even with on-device machine learning, devices require a significant amount of storage and computational ability, which most of the IoT devices lack. Most of the AI algorithms usually require a large data set to provide effective results. In URLLC, however, it is a challenge to provide such a big data set for the mission-critical IoT devices with reliability and low latency [38]. URLLC and Vehicle-to-Vehicle (V2V) One of the most promising and important applications of the future 5G network is V2V communication. V2V communication is one of the technologies that can lead to an intelligent transport system [39]. Naturally, for V2V, road safety (distance awareness to avoid any collision, speed limits, location-based traveling, environment information, road condition) plays a vital role and is extremely time-critical, as shown in Figure 10. Because of the safety concern, European Telecommunications Standard Institute (ETSI) has standardized safety protocols based on two awareness-based messages: decentralized environmental notification message (DENM) and cooperative awareness message (CAM) [40]. To reflect vehicles based on the mentioned safety standards, V2V communication should have the low latency characteristic of URLLC. As discussed in Section 3.6, handover is still an issue in URLLC implementation owing to the mobility of vehicles (as shown in Table 4) delivering safety standard messages. One of the most promising and important applications of the future 5G network is V2V communication. V2V communication is one of the technologies that can lead to an intelligent transport system [39]. Naturally, for V2V, road safety (distance awareness to avoid any collision, speed limits, location-based traveling, environment information, road condition) plays a vital role and is extremely time-critical, as shown in Figure 10. Because of the safety concern, European Telecommunications Standard Institute (ETSI) has standardized safety protocols based on two awareness-based messages: decentralized environmental notification message (DENM) and cooperative awareness message (CAM) [40]. To reflect vehicles based on the mentioned safety standards, V2V communication should have the low latency characteristic of URLLC. As discussed in Section 3.6, handover is still an issue in URLLC implementation owing to the mobility of vehicles (as shown in Table 4) delivering safety standard messages. IoT Energy Efficiency (EE) In IoT and machine-to-machine (M2M) communications, EE will play an important role, especially with sensor-type equipments with limited resources, for example, limited battery and computing power [41]. A number of URLLC applications require a lot of computation, which is not handled by some IoT devices [42]. From a PHY perspective, it is a challenge for URLLC to achieve low latency and high reliability in mission-critical IoT devices. The use of short packet in order to achieve low latency can degrade channel-coding gain, and it causes reliability issues in wireless channels. To mitigate reliability issue, re-transmission is required, but it involves additional resources and increases latency [43]. Base Station Densification and Device-to-Device (D2d) Communications In typical automated industry, clusters of sensors and actuators are working in a fixed area. One of the crucial use cases for the 5G URLLC is to support the wireless industrial automation (e.g., Industry 4.0 [44]). With the emerging industrial automation, M2M and D2D communications require URLLC features to deliver short messages from a controller to a cluster of sensors or machines. A reasonable amount of traffic is expected to be handled by WiFi and small-cell-technology based on mmWave frequencies, as shown in Figure 11. The METIS project estimated that dense metropolitan areas might have up to 200 devices per , with an expected data volume generated by each device could be 500 Gbyte/month [45]. Such an immense number of devices could force a drastic change in network infrastructure to avoid congestion and availability of service. With the limited frequency bands, improvement of the spectral efficiency (SE) could be an answer to support massive data. The massive MIMO network topology can theatrically support high-density traffic [46]. The theoretical performance and limitations of massive MIMO communications are extensively studied by a number of researchers [46][47][48]. IoT Energy Efficiency (EE) In IoT and machine-to-machine (M2M) communications, EE will play an important role, especially with sensor-type equipments with limited resources, for example, limited battery and computing power [41]. A number of URLLC applications require a lot of computation, which is not handled by some IoT devices [42]. From a PHY perspective, it is a challenge for URLLC to achieve low latency and high reliability in mission-critical IoT devices. The use of short packet in order to achieve low latency can degrade channel-coding gain, and it causes reliability issues in wireless channels. To mitigate reliability issue, re-transmission is required, but it involves additional resources and increases latency [43]. Base Station Densification and Device-to-Device (D2d) Communications In typical automated industry, clusters of sensors and actuators are working in a fixed area. One of the crucial use cases for the 5G URLLC is to support the wireless industrial automation (e.g., Industry 4.0 [44]). With the emerging industrial automation, M2M and D2D communications require URLLC features to deliver short messages from a controller to a cluster of sensors or machines. A reasonable amount of traffic is expected to be handled by WiFi and small-cell-technology based on mmWave frequencies, as shown in Figure 11. The METIS project estimated that dense metropolitan areas might have up to 200 devices per km 2 , with an expected data volume generated by each device could be 500 Gbyte/month [45]. Such an immense number of devices could force a drastic change in network infrastructure to avoid congestion and availability of service. With the limited frequency bands, improvement of the spectral efficiency (SE) could be an answer to support massive data. One of the main issues with massive MIMO is to manage data generated with unpredictable behavior [46]. It is essential to highlight that the building block of next-generation networks will be data packets, which show an unpredictable data generation behavior in non-streaming applications (such as social network applications and web browser). 3GPP Standardization for URLLC To guarantee the desired reliability, 3GPP and its allies are still working and planning for improvements at multiple aspects of 5G architecture. Some of the recent critical points highlighted by 3GPP Release 16 [49] include the following: Handover Handover, which is one of the most crucial prominent issues in supporting URLLC requirements, was also a part of the discussion in Release 16 of 3GPP. A way to support handover Figure 11. Illustration of a single user data packet and multiple user data packets relayed by a server/cluster head. The massive MIMO network topology can theatrically support high-density traffic [46]. The theoretical performance and limitations of massive MIMO communications are extensively studied by a number of researchers [46][47][48]. One of the main issues with massive MIMO is to manage data generated with unpredictable behavior [46]. It is essential to highlight that the building block of next-generation networks will be data packets, which show an unpredictable data generation behavior in non-streaming applications (such as social network applications and web browser). 3GPP Standardization for URLLC To guarantee the desired reliability, 3GPP and its allies are still working and planning for improvements at multiple aspects of 5G architecture. Some of the recent critical points highlighted by 3GPP Release 16 [49] include the following: Handover Handover, which is one of the most crucial prominent issues in supporting URLLC requirements, was also a part of the discussion in Release 16 of 3GPP. A way to support handover while keeping low latency and jitter remains a significant concern. In 3GPP TS 23.502 [50], the handover process requires a lossless handover. The source RAN node forwards data directly or indirectly to the target RAN node. The use of tunnel is the current approach, but it introduces additional jitter and latency. With this current approach, the issue remains open, and further studies were suggested by 3GPP on the matter. User Mobility The second issue on improving session stability while keeping UE application uninterrupted. The main focus of the discussion was to enhance runtime synchronization between UE and 5G systems to support application transfer without breaking application sessions and service continuity. In Release 15 of 3GPP, the mobility-related issue was partially solved by introducing a "notification" mechanism, but the purposed solution was not suitable for non-human devices. The proposed solution requires the UE to alter its behavior (e.g., reduction of speed in case of a vehicle) so that the RAN can restore the required QoS level with that cell. The purposed solution also involves UE to exchange numerous signal messages regardless of the potential link quality or congestion-related issues with the RAN to achieve the desired guaranteed bit rate (GBR). Then, it clearly degrades the E2E performance. QoS Monitoring to Support URLLC As mentioned in Table 2, URLLC services require high reliability and very low latency. Such requirements pose quite a challenge for 5G systems, because such requirements could affect E2E QoS performance. In the current 5G system, QoS notification control (NC) is maintained by a 5G access network (AN) to monitor GBR. Though the 5G-AN mechanism supports the guaranteed flow bit rate (GFBR) [49], which might not be sufficient for URLLC E2E services, features such as packet loss, jitter, and packet latency will also play a major part for URLLC services. To counter such QoS relates issues, 3GPP has branched out further aspects of studies, which include specification of the UE requirements for URLLC and improvement of QoS monitoring ability of the defined mechanisms. In Release 16 [49], a number of solutions to handle the mentioned issues were purposed and evaluated. The idea that server supporting user equipment should be kept close to the user (i.e., topologically, geographically) was proposed. With this idea, the transmission latency between the server and the base station could be minimized. In order to achieve high consistency, backhaul reliability was suggested for further improvement. Possible 5G Integration Plan by 3GPP Previous generations of cellular networks required access and core networks, which belong to the same generation, to be installed. It means that 4G systems were composed of LTE and evolved packet core (EPC). The deployment of a 5G system comes with the flexibility of integrating elements of previous cellular generations in different configurations [51]: 1. SA using only one radio access technology 2. N-SA is combining multiple radio access technologies. Standalone (SA) In SA, the evolved LTE radio or the 5G NR cells and the core network are operated alone, so that the NR or evolved LTE radio cells are used for both user and control planes. The SA provides a simple solution for operators to provide services to both 4G and 5G customers using normal inter-generation handover as shown in Table 5. The three variations of SA, as defined by 3GPP, are as follows: 1. 5GC and LTE ng-eNB access In NSA, the LTE radio cells and NR radio cells are combined using dual-connectivity to provide radio access. The core network could be either 5GC or EPC based on the operator's choice [52]. On the basis of the operators, they can provide 5GC for 5G customers or facilitate the existing 4G deployment combining NR radio and LTE resources with current EPC. NSA needs tight integration with the LTE RAN. Three variations of NSA, as defined by 3GPP, are as follows: 1. LTE eNB and EPC as master and NR en-gNB as secondary. 2. NR gNB and 5GC as master and LTE ng-eNB acting as secondary. 3. LTE ng-eNB and 5GC as master and NR gNB as secondary. Options defined by 3GPP for 5G deployment use either existing EPC [53] or the 5GC [54]. Both architectures follow different design principles. EPC can be considered as an evolution of earlier generation packet-based core networks. 5GC is designed based on "cloud-native" approach, with virtualization and cloud computing as its core. 5GC provides improved QoS features and superior network slicing. Future Research Areas In this paper, many key issues related to the URLLC implementation are discussed. Once the standards for 5G NR are matured, the researchers can work towards improving the core network, backhaul, and transport delays [39]. Issues related to UE energy efficiency (discussed in Sections 3.5 and 4.1) can be further discussed in the light of the new NR defined energy-efficient state called INACTIVE [55]. This state is defined as a state existing between the CONNECTED and IDLE states of a UE. The performance evaluation of the INACTIVE state remains open for further examination. One of the concerns related to the coexistence of URLLC with eMBB (discussed in Section 3.2) is not only limited to the existence of the mentioned categories of 5G. With the backward compatibility of 5G NR, the coexistence of URLLC with distributed system architecture and diverse application requirements will need further study to avoid latency concerns [55]. Likewise, handover related issues (discussed in Sections 3.6 and 4.3) can be further discussed in terms of the reestablishment of connection in the case of the radio link failure, keeping in view that LTE and NR traffic will coexist in the same carrier. Many areas need further investigation to operate IoT devices over URLLC. Although URLLC can provide low latency with high reliability, the characteristics of massive connectivity can degrade such requirements, as discussed in Section 4.1. As discussed in Section 4.2, on-device AI and ML will play a major role in an upcoming era, and hence researchers can further study the designing of AI/ML algorithms that do not require high resources and are better suited for devices with limited resources. A comprehensive summary of the issues discussed in the paper is presented in Table 6. Table 6. Summary of issues discussed in the paper. AI, artificial intelligence; ML, machine learning. Reference Section Summary QoS [11,13,15,16] In this Section 3.1, QoS requirements of URLLC (latency and reliability) and factors, which are a hindrance in achieving the desired QoS for URLLC, are discussed. Coexistence with eMBB [17][18][19] In the 5G networks, many different applications with diverse requirements will exist in the same physical medium. Such a coexistence of services will raise many challenges for telecom companies. In Section 3.2, the problems with the coexistence of eMBB and URLLC with different service requirements are discussed. URLLC Packet Design [15,20] Packet design plays a vital role in achieving low latency. Minimizing the packet processing time will be a key factor in enabling low latency for URLLC. Packet structure proposed by LTE and NR to achieve low latency is discussed in Section 3.3. URLLC Scheduling [20] Because of the unpredictable packet generation of URLLC, scheduling is a challenging task. In Section 3.4, some of the proposed scheduling schemes for URLLC and issues with those schemes are discussed. Energy issues for UE [21,56] To keep up with the latency requirement of URLLC, UEs are forced to perform extra tasks, which can result in low battery life for the UEs. Such power consumption related issues are discussed in Section 3.5. Handover issues for URLLC [22][23][24][25] Providing uninterrupted services to a mobile user is the most significant facility of any telecom infrastructure. Providing such an uninterrupted service to a user using URLLC based services is quite difficult. Issues related to handover when it comes to strict latency are discussed in Section 3.6. Error Handling [17,26,27] Wireless services are prone to many challenges, and providing highly reliable service in wireless communication is quite a tough task. The issues related to the handling of error packets and retransmission are covered in Section 3.7. Role of URLLC in operating IoT [33,34] IoT will play a major role in the coming era of technology. URLLC will play a vital role in supporting IoT services. In Section 4, the importance of URLLC to operate IoT is discussed. URLLC and Massive device connectivity [35,36] Although URLLC fulfills the basic requirement of reliability and latency for mission-critical IoT, it is a challenge for URLLC to provide simultaneous services to a vast number of devices. Section 4.1 covers the issues that URLLC brings in operating massive IoT devices. On-device AI and URLLC [21,[36][37][38] In earlier sections importance of URLLC for time-critical applications is highlighted. However, the provision of low latency service to massive devices is also a challenge, as cited in Section 4.1. It is provoking researchers to seek new solutions to achieve low latency with high reliability. Among such solutions developing intelligent machines is quite prominent. In Section 4.2, issues related to AI/ML-based machines and relying on URLLC services for such machines are discussed. URLLC and V2V [39,40] An automated vehicle is one of the most anticipated services of the upcoming era. However, providing highly reliable and time-critical connectivity is still a challenge for URLLC. V2V connectivity opens a whole new level of disputes. Among them, some issues are discussed in Section 4.3. Possible Solutions for Reliability and Latency Requirements Edge communications will play an essential role in future networks; researchers have provided a number of solutions to overcome the strict QoS requirements in URLLC. Current cellular networks follow a centralized approach, while edge communications bring resources close to the UE. Despite some issues, the edge communications based solutions, that is, mobile-edge computing (MEC) [57,58], local area communication [59,60], and wide-area large-scale communications [61,62], are very promising, as shown in Table 7. Reference Proposed Solutions Using MEC to Support URLLC [64] Minimizing E2E communication delay [65] Highlighting the MEC role to support URLLC in mission-critical applications with further optimization parameters for significant use cases Minimizing E2E communication delay [42] Proposing an algorithm for energy efficiency (EE) in mobile devices by optimizing queue complexity of the communication process [67] Reducing computation and latency for IoT devices using MEC In the papers explained in Table 8 and numerous other papers on edge computing, it is highlighted that the next generation networks should have a cloud-native approach in order to achieve high reliability and low latency. Although some challenges still exist in implementing the edge-computing systems (ECS) based approach for URLLC, ECS can further be improved and can provide better QoS for 5G URLLC networks. AI and 5G Networks Traffic Management AI can be integrated with 5G networks to improve the efficiency of resource and network management. The network architecture and user requirements for 5G networks, the traffic management will be a challenge [68]. AI including ML and deep learning can assist 5G networks in predicting and managing the unpredictable network traffic. AI can analyze and cope with the unpredictable requirements of 5G networks traffic [69]. Additionally, ML can play an integral part to support the MEC architecture [70]. However, the current AI-based research has some limitations [71]. In the papers explained in Table 8 and numerous other papers on edge computing, it is highlighted that the next generation networks should have a cloud-native approach in order to achieve high reliability and low latency. Although some challenges still exist in implementing the edge-computing systems (ECS) based approach for URLLC, ECS can further be improved and can provide better QoS for 5G URLLC networks. Minimizing E2E communication delay [42] Proposing an algorithm for energy efficiency (EE) in mobile devices by optimizing queue complexity of the communication process [67] Reducing computation and latency for IoT devices using MEC AI and 5G Networks Traffic Management AI can be integrated with 5G networks to improve the efficiency of resource and network management. The network architecture and user requirements for 5G networks, the traffic management will be a challenge [68]. AI including ML and deep learning can assist 5G networks in predicting and managing the unpredictable network traffic. AI can analyze and cope with the unpredictable requirements of 5G networks traffic [69]. Additionally, ML can play an integral part to support the MEC architecture [70]. However, the current AI-based research has some limitations [71]. Most of the existing studies focuses on utilizing AI/ML to solve core network and routing related issues. Only a few researchers have worked on applying AI to the semantic and application layer to propose traffic management solutions. 5G and Beyond With the expected deployment of global 5G networks in the 2020s, it is time to raise the imperative question on the future of mobile network, that is, beyond 5G or 6G. One of the prominent areas of beyond 5G will be software-defined network (SDN) and network function virtualization (NFV) [71]. The SDN and NFV will play a vital role in enabling management and control systems for E2E structures. Software-based transformation (i.e., softwarization) [72] cannot be implemented within the defined 5G implementation time frame. Other areas of beyond 5G include security [73,74], spectral and energy efficiency, resiliency (i.e., tolerance to interference and maintaining QoS), and MEC [58]. MEC and softwarization are among the promising candidates to provide a cost-effective, secure, manageable, and flexible architecture for 5G and beyond. Conclusions The main contribution of the paper is to provide researchers a fast and brief reference to some of the core issues in the implementation of URLLC. Keeping in view the importance of IoT in the coming era, this paper also covers a few most critical aspects of IoT and V2V communication over URLLC. On the basis of issues being covered in this paper, some of the areas that are still open for further investigation in URLLC improvements are also provided to readers. At the end of the article, a possible solution using edge computing is proposed for URLLC implementation. This paper can provide a comprehensive platform for researchers who are looking to study URLLC and its issues with diverse services and applications.	11,113.2	2019-09-02T00:00:00.000	[ "Engineering", "Computer Science", "Environmental Science" ]
Match_Motif: A rapid computational tool to assist in protein–protein interaction design Abstract In order to generate protein assemblies with a desired function, the rational design of protein–protein binding interfaces is of significant interest. Approaches based on random mutagenesis or directed evolution may involve complex experimental selection procedures. Also, molecular modeling approaches to design entirely new proteins and interactions with partner molecules can involve large computational efforts and screening steps. In order to simplify at least the initial effort for designing a putative binding interface between two proteins the Match_Motif approach has been developed. It employs the large collection of known protein–protein complex structures to suggest interface modifications that may lead to improved binding for a desired input interaction geometry. The approach extracts interaction motifs based on the backbone structure of short (four residues) segments and the relative arrangement with respect to short segments on the partner protein. The interaction geometry is used to search through a database of such motifs in known stable bound complexes. All matches are rapidly identified (within a few seconds) and collected and can be used to guide changes in the interface that may lead to improved binding. In the output, an alternative interface structure is also proposed based on the frequency of occurrence of side chains at a given interface position in all matches and based on sterical considerations. Applications of the procedure to known complex structures and alternative arrangements are presented and discussed. The program, data files, and example applications can be downloaded from https://www.groups.ph.tum.de/t38/downloads/. Match_Motif approach has been developed. It employs the large collection of known protein-protein complex structures to suggest interface modifications that may lead to improved binding for a desired input interaction geometry. The approach extracts interaction motifs based on the backbone structure of short (four residues) segments and the relative arrangement with respect to short segments on the partner protein. The interaction geometry is used to search through a database of such motifs in known stable bound complexes. All matches are rapidly identified (within a few seconds) and collected and can be used to guide changes in the interface that may lead to improved binding. In the output, an alternative interface structure is also proposed based on the frequency of occurrence of side chains at a given interface position in all matches and based on sterical considerations. Applications of the procedure to known complex structures and alternative arrangements are presented and discussed. The program, data files, and example applications can be downloaded from https:// www.groups.ph.tum.de/t38/downloads/. K E Y W O R D S binding interaction motifs, interaction design, protein design, protein-protein binding, protein-protein interface design \| INTRODUCTION Over the years, the database of protein structures has grown from a few atomic resolution structures in the 1980s to hundreds of thousands of entries. [1][2][3] It also includes the structures of many protein-protein complexes and multicomponent assemblies. 2 In recent years, the design of new protein structures and new protein-protein interactions to generate new proteins or complexes with a desired function has become a new research focus in protein science. [4][5][6][7] One powerful experimental tool for the stepwise adaptation of a protein sequence toward a desired function is the directed evolution method. It is based on random mutagenesis of a protein sequence or segment of a protein and subsequent selection according to a desired target function. 8,9 Typical applications are enzymes with known structure and function that are evolved toward new specificity and increased thermostability. 8,10 Application to design new pairwise or oligomeric interactions are also possible. 11 However, the process requires usually repeated mutation and selection cycles and depending on the required changes the process can be time consuming and requires significant experimental efforts. Enzyme and protein design can be supported by computational (rational) design methods. If the structure of the protein or protein-protein complex is known the effect of introducing a mutation in a protein can be studied using a variety of computational approaches. [12][13][14][15] Available methods range from molecular dynamics-based free energy simulation 16 to very simple computational scoring methods of replacing a residue based on just one structure or an ensemble of structures representing the wild type or mutated system. 14,17-20 A few approaches are available that allow for a complete redesign of a protein based on a backbone scaffold. 4 The most frequently used approach is based on the Rosetta program suite 21,22 and allows one to completely redesign the residues of a given backbone structure to achieve a stable folded protein using a Monte Carlo approach in combination with an appropriate scoring of the designed structures. 21,23 This method has been used in several applications to successfully redesign proteins or to create completely new stable folded proteins. 4,5,7,23 Combinations of rational design and directed evolution have also been used successfully to generate new or altered proteins. 9,24 In addition to the design of single proteins rational approaches have also been used to design stable protein-protein interactions. 6,7,25,26 A common method implemented in the Rosetta approach is based on identification of a putative interacting region, for example, a small putative interaction motif (Motifgraft), and the subsequent computational de novo design of a new protein partner around the motif and possibly additional interactions to the partner protein. 25,27 Such approach is, however, computational quite demanding and requires the computational evaluation of many in silico residue substitutions and further energy minimization and molecular modeling steps. The computational evaluation may also be of limited accuracy because of possible force field artifacts and the neglect of accounting for solvent effects in most modeling steps. In order to simplify the initial search for a putative desired protein-protein interface the computational Match_Motif tool has been developed. The approach is entirely based on interfaces of known protein-protein complexes. The user provides a desired initial geometric arrangement of two proteins that should include a putative contact surface. It may originate from a manual placement or can be obtained as putative complex by some protein-protein docking software. The tool then searches through a database of structural interaction motifs (based on the arrangement of short backbone segments in known complexes). All motifs that structurally match to a corresponding motif at the desired interface are collected. The desired contact surface can then be remodeled based on the side chains of the matching motifs. The procedure takes a few seconds on a standard workstation computer. Since the final contact surface is based on motifs extracted from known favorable proteinprotein interactions it likely also provides stable binding of the desired interface or at least provides a basis for further adjustment or refinement using more demanding methods. In a first part, the workflow of the method will be presented followed by several test applications to recover known highly stable interfaces or to design putative new arrangements. Additional possible applications of the tool and directions of further improvement are also discussed. \| RESULTS For the design of assemblies of proteins with a desired function often a specific binding geometry of interacting protein partners is required. Hence, for a given pair of stable proteins that act as building blocks a stable binding in a preselected arrangement is desired. In the Match_Motif approach the user needs to provide such desired geometry as an input. Based on the backbone geometry of both partners at the interface, the algorithm searches for structural motifs of the target arrangement and searches for similar motifs in a database extracted from known stable complexes. A structural motif is here defined as four consecutive interface residues in one and another four consecutive residues in the other partner within a threshold distance of 10 Å. The motif is characterized by the distances between backbone atoms within each partner (within each of the four residues) and between the two short stretches of residues ( Figure 1). The database stores $250,000 such structural interaction motifs extracted from 1,686 structures of protein-protein complexes extracted from the PDBbind. 28 In the following, a reduced set with $57,000 structural interaction motifs was employed that allows one to search through all motifs within <5 s for all tested cases on a standard workstation PC (see Section 4 for details). A match between an interaction motif in the target interface and a motif in the database is characterized by a small root mean square deviation of the backbone atom distances (dRMSD) and root-mean-square deviation (RMSD) of the superposition of target motif and database motif (it is typically set to <1 Å, but can be set as input parameter by the user). Once a match has been found it is stored and for constructing a new interface structure the interface segment is replaced by the database motif (only side chains are replaced that do not lead to sterical overlap with other residues at the interface). Upon completion of the algorithm, typically several side chains at the target interface are replaced by the side chains found in the matching motifs from the database. Since the side chains in the database motifs are all extracted from known stable protein-protein complexes, it is likely that the redesigned interface also forms a stable complex or at least is a good starting point for further adjustments or refinements. The process steps and workflow of the Match_Motif approach are outlined in Figure 1. \| Application to known proteinprotein complex structures As a first application, the method was tested on a diverse set of 10 known protein-protein complexes with different interfaces ( Table 1). The complexes were not included in the design of the database of interaction motifs. The partner proteins were placed in the native interaction geometry and for the matching procedure only target motifs with a center-of-mass distance between the four consecutive residues on each partner <10 Å were included. The RMSD threshold for a match was set to F I G U R E 1 Illustration of the Match_Motif approach. (a) Structural motifs of four consecutive amino acids in each partner with a center-of-mass distance <10 Å are extracted from known protein-protein complexes. The interaction motifs are stored in a database and are characterized by a set of backbone (Cα-Cα) distances. (b) The user provides a desired protein-protein interaction geometry that serves to extract putative interaction motifs (characterized by sets of distances between four residues on each partner). (c) A matching interface motif is selected. (d) The corresponding motif at the interface is replaced by the matching motif from the database providing side chains that fit sterically and allow favorable interaction in the desired geometry Matching motifs were identified in a collection of 57,000 motifs (not including the 10 test complexes, first 10 rows) with a backbone RMSD threshold of 1 Å. The last two rows indicate application to two complexes that were included in the motif database generation. For these cases numbers in parenthesis indicate the accepted motifs using a backbone RMSD threshold of 0.25 Å (all accepted motifs reproduce the original interface sequence). 1 Å. With these criteria for defining a structural match, the Match_Motif approach identified in each case several matches ( Table 1). The number of matches can vary depending on the size and type of protein-protein interface (for the considered cases between $20 and 160, Table 1). The matches may originate from homologous complexes but can also originate from complex structures that are not overall similar to the target complex structure. Many local recurrent structural interaction motifs occur in unrelated complex structures (e.g., segments of interacting α-helices, loop-loop, loophelix interactions, etc.). With the help of a molecule viewer such as Pymol, it is possible to easily visually inspect the structural interface matches (illustrated in Figure 2a,b for the PDB 3Q87 example case). In addition to the superimposed matching motifs, the Match_Motif approach also provides a new alternative complex structure that can include several substitutions of residues at the interface (indicates as accepted changes in Table 1). Note that the substitutions are entirely based on the collection of structural interface motifs in known complex structures. Only changes are accepted that do not result in significant sterical overlap with other residues of the partners and are selected based on the frequency of occurrence of a side chain in all matches for a given position at the interface (see also Section 4). Since the algorithm searches for interaction motifs from known complexes (that are usually well packed at the interface) and replaces a similar (backbone) matching motif at the desired interface the resulting alternative interface is typically also well packed. For example, in case of the PDB 2O25 complex with a relatively small interface, three substitutions in the buried interface region (accessible surface area, SASA, of side chains <10 Å 2 ) included a Glu-> Leu, an Arg-> Ala, and a Thr-> Ile substitution. In case of PDB 3Q87, with a larger interface, six substitutions at the buried interface were accepted (Leu-> Ser, Leu-> Met, Val-> Leu, Thr-> Ile, Val-> Ala, and Ile-> Tyr). Hence, both substitutions toward larger and sometimes smaller residues are observed such that the average size of interface residues does not change significantly. The approach in its present form does not include further (possibly costly) energetic evaluation of the substitutions. However, for the user it is possible to not only evaluate the suggested alternative interface but to inspect independently all suggested matching motifs (e.g., using a molecule viewer) and further evaluate the suggested changes using other computational tools or direct experimental testing. As a further test, the approach was also applied to two complexes that were also used to build the database of interface motifs (last two rows in Table 1). With the standard settings (with 1 Å backbone RMSD threshold for a match), several alternative matching motifs were found in these cases resulting also in a few substitutions at the interface. This is expected since for these examples, the database can contain matching motifs from other complexes. However, if the RMSD threshold for a match is sufficiently lowered (to 0.25 Å) exactly the residues found in the interface of the known complex are recovered (the match from the database corresponds to a motif from the same complex structure or a very close homolog!). \| Effect of small changes in the partner arrangement As a next step, the sensitivity of the approach to small changes of the placement of the partners relative to the native binding geometry was tested. For three test cases (Table 2), one partner protein was moved along a line connecting the center-of-mass of the interface atoms of each partner (atoms within 10 Å of the other partner) in steps of 1 Å. All residues of the protein partners were replaced by Ala residues and matching interface motifs were collected with the Match_Motif approach (Table 1). Due to the stepwise separation of the partners, a larger cutoff for atoms belonging to the interface (12 Å) in addition to the default of 10 Å was also used. This leads to overall more matches but as expected the number of matches decreases rapidly with increasing distance between protein partners (Table 1 and illustrated for the PDB 3P57 case in Figure 3). Already at displacements along the separation coordinate by 3 Å very few or no matches are found. The result is not unexpected since the distance between protein partners is critical for obtaining a sterical possible but also well-packed interface. If the distance becomes too large, there are simply no interaction motifs available that still include side chains in close contact with each other. It also indicates that the user needs to place the two protein partners fairly precisely (to within 1-2 Å) in order to obtain a reasonable number of interface structural motifs. However, since the application of the Match_Motif approach requires only a few seconds it is in principle possible to rapidly test slightly different interface geometries and compare the packing and stereochemistry of the obtained interface matches (see below). \| Design of putative protein-protein interfaces for example cases One possibility to generate alternative protein-protein interaction geometries at a reasonable distance of the partner proteins to yield a sufficient number of matching motifs is to apply a protein-protein docking approach. Such docking approach typically results in near-native solutions but also in a large number of alternative binding geometries that may not score as well as the near native solutions. In the present case, the ATTRACT docking program 29-31 was used to obtain alternative binding geometries for the PDB 1HE1 complex (using rigid partner proteins). In a systematic docking run the surface of the protein partners is systematically scanned for potential interaction geometries usually resulting in thousands of local energy minimum solutions. For the PDB 1HE1 case, the solution ranked at position 100 (top100 solution) significantly differs from the native geometry ( Figure 4a interface in the top100 complex relative to the same atoms in the native complex. The LRMSD measures the RMSD of one partner protein relative to the native placement after best superposition of the second protein (receptor) onto the receptor in the native complex. The ATTRACT program calculates for each docked complex a knowledge-based score in RT units (R: gas constant, T: temperature, RT unit indicates the mean energy per degree of freedom at a temperature T). For the native complex a score of À20.1 RT units (top1 solution) was obtained and À12.2 RT units for the top100 solution. The top100 solution does not score well because of several unfavorable polar-nonpolar contacts at the alternative interface and nonoptimal packing (Figure 4a, second row). Nevertheless, if we consider this geometry as a desired geometry one can use the Match_Motif approach to identify putative matching interface motifs for the top100 geometry. Indeed, the alternative complex structure obtained as output of the program contains several residue substitutions. The structure was energy minimized to remove any residual sterical overlap (see Section 4) and the redesigned partners were used as input for another systematic ATTRACT docking run. In this case, docking solutions very close to the new (top100-based) geometry were obtained that scored with an ATTRACT score of À18.8 RT units as "new" top1 solution. It also stayed close to the (old) top100 solution with a deviation from the (old) top100 geometry of IRMSD = 1.4 Å and LRMSD = 2.4 Å (Figure 4a, lower panels). Hence, at least the in silico analysis of the redesigned interface predicts a significant improvement by the Match_Motif approach. The same procedure was also tested on a docking solution for the complex PDB 1J2J (taking a solution that ranked as top94 solution in a systematic docking run with an ATTRACT score of À9.1 RT, compared to À15.1 RT for the native solution). Here, the interface redesign resulted in an alternative complex structure that upon systematic redocking gave a score of À13.4 RT units and scored at position top9 ( Figure 4b) again a significant predicted improvement of the new interface. F I G U R E 3 Search for interface motifs upon distance change between partner proteins. (a) Protein-protein complex PDB 3P57 with the interface region indicated by a black rectangle. (b) Native interface structure in stick representation. (c) Native interface with all side chains removed as start for the Match_Motif application. Rebuilt interface structure after application of Match_Motif. Note that the interface side chain structure differs at several positions from the native interface structure (shown in b) because Match_Motif identifies alternative putative interface residues occurring more frequently in the matching motifs. (e, f) Generated interface structure for different distances of the partner proteins relative to the native placement. The distance of the blue from the green partner was increased in steps of 1 Å (e), 2 Å (f), 3 Å (g) to up to 4 Å (h). For the case (e), an interface structure similar to the generation for the native placement (in d) was obtained but no matching motif was returned for the displacement by 4 Å (h) Finally, a manual placement of the partners at another putative interaction geometry (illustrated in Figure 4c) was also tested. The manual docking placement was achieved using the Pymol software 32 and checking of the protein-protein distance in the edited placement. The desired interface contains several charged and polar residues and not necessarily favorable contacts to nonpolar residues (Figure 4c) resulting in an ATTRACT score of À5.6 RT units. The Match_Motif approach generated an alternative complex interface structure that upon systematic docking scored as top32 solution with an ATTRACT score = À13.1 RT units. As indicated already in the previous paragraph, a manual placement of protein partners may not yield an optimal number of matching interface motifs (e.g., because in case of a too large or too small distance between partners no motifs that yield reasonable side chain contacts and/or no residue overlap are found in the motif database). Such case is illustrated in Figure 5a for an α-helix (blue in Figure 5a) that is desired to bind at the space between two helices of the second partner (green in Figure 5a). At the starting placement, no substitutions of interface residues are suggested and the interface is not well packed (Figure 5b). A systematic shifting of the single helix relative to the partner protein in all spatial directions yields one placement that moves the ligand protein closer to the partner protein (by $1 Å) resulting in several matches that produce an alternative interface structure with several hydrophobic residues better packed than the initial geometry (Figure 5c,d). \| DISCUSSION The large collection of proteins and protein-protein complexes with known structure forms a rich basis of F I G U R E 4 Examples of redesigned interfaces. (a) The top panel shows the native protein-protein complex PDB 1J2J (blue and green cartoon) and an alternative arrangement (magenta) obtained from a docking run using the ATTRACT program ranked as top100 solution (native complex ranked as top1). The interface region is indicated as black rectangle. In the second row, the enlarged interface at the alternative docking solution (top100, green and blue partner structures), and in the third row the redesigned interface using the Match_Motif approach is shown (here, the second partner is shown in magenta). (b) Same as (a) but for the complex PDB 1HE1 (an alternative docking solution scored as top94 was used). (c) Same as (b) but using a manually placed alternative arrangement for the second partner (magenta) extracting recurrent motifs at protein-protein interfaces. Such local motifs of contacting consecutive residues on both partner proteins occur not only in homologous complexes but can also form locally as part of interfaces in unrelated complexes. The Match_Motif approach rapidly compares all backbone interaction motifs at a desired interface with the known available motifs with sterically well-fitting intermolecular contacts. For a target interface of sufficient size that places both proteins at a reasonable distance to form a stable complex typically in the order of $100 structural interface motifs can be found. The process requires only a few seconds but could be further accelerated by a smarter algorithm to search through available motifs. In the future, this could be achieved for example by first searching through a structural classification of motifs that contains only a few representatives followed by an extended search through only the members of a few matching classes. It should be emphasized that the present approach can be used to redesign a known interface by suggesting appropriate substitutions. However, in such a case, it may also be useful to directly search for evolutionary conserved related complexes 33 in order to identify suitable residue substitutions. Even in such case, the Match_Motif approach can suggest appropriate substitutions if a structure of such related complex is available and used to provide motifs stored in the database. However, a desired focus of the Match_Motif approach is to offer the possibility to find appropriate matches even for a new interface with no evolutionary conserved homologs. Besides of a collection of matching motifs, the approach also provides a newly generated interface structure with new interface side chains extracted from the matching motifs. The selection of the side chains is solely based on frequency of occurrence in the matching motifs and sterical considerations. For the example cases, the algorithm resulted in changes that considerably improved the scoring of the redesigned interface based on a knowledgebased scoring function. This is a good hint but no guarantee that the exchanges at the interface indeed improved the binding affinity. In this regard, it is important to emphasize that the purpose of the Match_Motif program is not to automatically provide a new stable interface for every desired binding arrangement of two proteins. It can, however, be useful to provide a starting point for further refinement and to exclude certain arrangements that do not give any match to a structural motif found in know stable protein-protein complexes. Further refine-F I G U R E 5 Optimization of partner protein placement. (a) An initial manual and nonoptimal placement of an α-helical partner (blue) relative to a putative partner (green) gives only few interface matches. (b) The nonoptimal packing at the interface is illustrated by a van der Waals representation of the interface. (c) After optimization of the placement in (only) translational steps of 1 Å a more hydrophobic and sterically better packed interface is obtained by the Match_Motif approach (better van der Walls packing illustrated in (d)) ment indicates here that not all residue exchanges at a desired interface need to be tested but only a few positions of an interface that contains already residues extracted from close motifs found in other stable proteinprotein complexes. The number of structural interface motifs could also be significantly extended by considering sequence homologs of known protein-protein complexes that likely form the same protein-protein complex structure as the known complex. In addition, the motif database could be further improved by distinguishing between interaction motifs that correspond to binding hotspots in known complexes and may contribute also high affinity upon copying such motifs to matching segments in desired interfaces. \| Extracting protein-protein interaction motifs In the present study, an interaction motif is considered as four consecutive residues in one protein partner and four consecutive residues below a distance threshold in the second partner. The distance threshold needs to be large enough to include all interaction motifs such that residues in one four-residue segment are in contact with residues in the four-residue segment in the partner protein. A test with different threshold distances indicated that for distances larger than 11 Å, an increasing number of motifs are found that have no or only little contact between residues of the partner proteins. On the other side for threshold distance smaller than 11 Å, some motifs with large interacting side chains are excluded from the database. Hence, with a 11 Å cutoff structural motifs were extracted from a set of 1,686 protein-protein complexes extracted from the PDBbind database. 28 Structures with an interface of <1,000 Å 2 or one partner containing <40 residues were not included (the list of PDB-IDs is included in the download package, see below), it resulted in 255,000 interface motifs. The coordinates of each motif and the distance between all pairs of Cα backbone atoms are stored in the database. A second database with a slightly reduced set of structural motifs was generated by introducing a cutoff for the mean of Cα-Cα distance deviations (dRMSD) between motifs. A motif was only retained if it deviates from all other motifs by a sum of Cα-Cα distances larger than the cutoff of 1 Å. Second, the set was further reduced by including only interface motifs with at least three nonpolar or aromatic residues. The reduced set contains $57,000 motifs and allows rapid selection and was used for all applications in the present study. \| Matching interaction motifs to interface segments For a given protein-protein complex, the program Match_Motif identifies the interface regions of the protein partners. The interface region typically includes all residues within a distance threshold for interatom pairs between the protein partners (default is 10 Å). Similar to the collection of interface motifs extracted from known protein-protein complexes in a next step, all segments of four consecutive residues in one partner and four residue segments in the second partner within a centerof-mass distance below a threshold (typically 10 Å) are collected. The collected motifs are compared with all the motifs in the interaction motif database with respect to the backbone dRMSD. All structural motifs from the database below a dRMSD threshold are considered as possible matches and can replace the corresponding interface segment (with a different set of side chains but only small difference in the protein backbone). As a final check the RMSD of the target motif (backbone) relative to the database motif is calculated and used for selection (with a default threshold of 1 Å). With this threshold for a protein-protein interface (of a known complex) with typical size between 20 and 200 matches can be collected. The matching segments are stored in a separate file and can be inspected visually. However, the program also outputs a PDB-file with an alternative interface based on the matching motifs. An interface residue is replaced by the most frequently found residues for a given position found in all matching motifs if it does not overlap with other residues in the partner proteins. To remove residual overlap and optimize the alternative interface arrangement an energy minimization employing a molecular mechanics program is performed. For the present cases, it was performed using first the sander program of the Amber18 package 34 employing an implicit solvent model with a distance dependent dielectric constant (4r) and 400 steps of energy minimization followed by energy minimization using a Generalized Born implicit solvent model 35 (igb = 5 option in Amber, 2,000 steps) and the pmemd. cuda module of Amber18. The FF14SB force field 36 was used during all minimization steps. The Match_Motif program, a manual, and the data file of structural interface motifs as well as example applications can be downloaded from https://www.groups.ph. tum.de/t38/downloads/. ACKNOWLEDGMENTS The author thanks Dr J. Coles for helpful discussions. This work was supported by a grant of the Deutsche Forschungsgemeinschaft (DFG) SFB863/A10-111166240.	6,997.2	2021-10-14T00:00:00.000	[ "Computer Science", "Biology" ]
Evaluation of Candidates Performance in Core Subjects and Financial Accounting in the May/June 2014 Wassce in Egor and Oredo Local Government Areas, Edo State, Nigeria This study is designed to carry out an Evacuation of Candidates Performance in Core Subjects and Financial Accounting in the May/June 2014 WASSCE in Egor and Oredo Local Government Areas, Edo State, Nigeria. To be able to evaluate performance, the researcher raised five research questions, four of which were hypothesized. The statement of problem included the perceived attitude of trying to shift blames for failures in school examination to teachers. A comprehensive and empirical review of both related literatures and studies on West Africa Senior School Certificate Examination results were also done to enable the researcher paint a picture of what has been obtainable prior before now on performances of students especially in Financial Accounting. A Proforma titled “Core Subjects and Financial Accounting in the May/June 2014 WASSCE in Egor and Oredo Local Government Areas Edo State, Nigeria”,was used to collect data from the sampled secondary schools the public senior secondary schools in Egor and Oredo Local Government Areas. Data analysis was done by an expert in measurement and evaluation using descriptive statistics, ANOVA, t-test of independent sample etc.The findings of the study revealed that candidates performed relatively well in both the Core Subjects (including Financial Accounting) in the May/June 2014 WASSCE. The findings also revealed that the teachers were effective and efficient in carrying out their duties. Based on the findings, recommendations were made for improvement in future. Introduction Over the years, the performances of candidates in the West African Senior Secondary Certificate Examination (WASSCE), seems to have been a source of worry to both parents and teachers.Various attempts have been made to identify the likely causes of such poor performances.In Nigeria, there tended to have been evidences of poor performance in (WASSCE) as reported by various researchers as well as in the media. The need for education, cannot be over stressed, hence all countries, continue in the gradual quantitative and qualitative expansion of educational opportunities to enhance national development.In Nigeria, highly skilled professionals seem to be wanted in the education sector.Moreover, majority of persons might have expected tremendous improvement in schools in our country. It is known that education is regarded as major instrument for socio-economic development.In our country, many administrations have attempted to save the educational industry from collapse to no avail.Every government appears to express worries about failing educational standard because of inadequate educational facilities, poor funding, limited access, low capacity of staffing as well as poor policy implementation.Nigerian government appears to place a higher premium on education.This is why the issue of student's poor performance in West African Senior Secondary Certificate Examination (WASSCE) seems to attract public aspersion. It appears that curriculum implementation in secondary schools needs more attention.Examination seems to be one method of evaluation and assessing instructional and educational performance, according to Fafunula (1974).The performance of students in WASSCE tended to be very low.Perhaps, the studying habits of students as well as their time management call for serious concern.Most of these lapses in examination success might be attributed to insufficient attention towards studies by students. The very concept of academic failure varies in its definition.Castellanos (1986) considers academic failure as the situation in which the subject does not attain the expected achievement according to his or her abilities, resulting in an altered personality which affects all other aspects of life.Similarly, Tapia (2002) notes that, while the current Educational System perceives that the student fails if he or she does not pass, more appropriate for determining academic failure is whether the student performs below his or her potential. In general, the various studies which attempt to explain academic failure do so beginning with the three elements that intervene in education: parents (family causal factors), teachers (academic causal factors), and students (personal causal factors).Among personal variables most studied are motivation and self-concept.Motivation is considered to be the element that initiates the subject's own involvement in learning: when a student is strongly motivated, all his effort and personality are directed toward the achievement of a specific goal, thus bringing to bear all his or her resources.According to González (1997), a consensus exists among the diverse motivational theories and approaches inasmuch as they conceptualize motivation in terms of conscious beliefs and values.In the area of motivation there seems to exist all kinds of opinions and results.Over the years it appears that teachers were not adequately motivated to offer education service delivery which could enhance optimal performance of candidates in the external examinations in the country. Any organization, no matter how small, whether it is a trading business, a charitable organization or even a home deals with money, Financial Accounting is a Technical as well as a Vocational Subject.The keeping of Accounting Books is essentially an art and the aims of those who study the subject are essentially utilitarian.This curriculum, which is the original work of the Comparative Education Study and Adaptation Centre (CESAC), is designed to meet these aims. The chief aim of the Curriculum is to build on the work that has been done in Book-keeping over time and also provide the teacher with the framework within which the teaching is to take place. The aims of Financial Accounting in general are to provide: (a.) Specialized instruction to prepare students for career in book keeping and accounting field; (b.) Fundamental instruction to help students assume their economic roles as consumers, workers and citizens; (c.) Background instruction to assist students in preparing for other professional careers requiring advanced studies in book keeping and accounting; and (d.) Book keeping and accounting skills for personal use in future. These objectives are geared towards making the students to appreciate the rules and functions of accounting, lay a sound foundation for further study of accounting at higher level and also assess candidates' knowledge of basic accounting principles and their application to modern business activities. An examination of the School Certificate Syllabus in Financial Accounting exposes that a student needs time to study and become acquainted with the contents before he could succeed.The Senior Secondary School Curriculum is based on the National Policy on Education, with the quality of instruction oriented towards inculcating certain values.The Federal Republic of Nigeria ( 2004) stated the broad goal of the secondary school education is to prepare individuals for useful living within the society and higher education.To achieve this objective, secondary school education in Nigeria has six years duration given in two stages three years of junior secondary school followed by three years of senior secondary school, the curriculum designed for senior secondary school is comprehensive and broad based, aimed at broadening students' knowledge and outlook.Subjects offered in senior school are in three group's core subjects, vocational and nonvocational subjects.One of the vocational subjects is Financial Accounting.Financial Accounting is one of the vocational electives that students are expected to offer from the list of elective subjects in group B to give a maximum eight or nine subject.Apart from the core subjects, the students are expected to offer a minimum of one and maximum of three from the list of elective subjects in group B and C to give a maximum of eight or nine subjects. According to WAEC (2004), Financial Accounting is the most popular subject that students offer among the vocational elective subjects.Other subjects in this category include Agriculture, Commerce, Computer Education, Clothing and Textiles, Food and Nutrition, Home Management, Technical Drawing, Fine Art, Music, Marketing and Fishery.According to Asaolu (2002), Financial Accounting is the process of recording, classifying, selecting, measuring, interpreting, summarizing and reporting financial data of an organization to the users for objective assessment and decision making."Accounting data are processed into accounting information through the use of accounting principles and conventions.The accounting principles are the basic fundamentals which guide accountants in recording, appreciating and assessing accounting information as well as the preparation and interpretation of financial statements.The accounting information system is proven, time honoured, and its format is universally understood.Books of accounts prepared by accountants in one part of the world are easily understood by their counterparts in other parts of the world because the information system is based on principles that are widely accepted and globally used.There are also general accepted Financial Standards. According to the National Examination Council (NECO 2004), the general objectives of studying Financial Accounting at senior secondary school are as follows: 1.To enable senior secondary school students appreciate the basic rules, functions and principles of accounting 2. To lay proper foundation for further study of accountancy and allied courses at higher level and 3. To enable the students understand basic accounting principles, practice and their applications to modern business activities. It appears that these objectives were not achieved due to poor performance of candidates in external examinations as observed. Statement of the Problem The performance of students in Senior Secondary in Nigeria has remained an issue of concern to all stakeholders (Ajagun, 2000).The report by Ojerinde, (1998) on the survey of the performance of candidates in secondary schools in Nigeria over the years revealed a discernible decline.This perennial decline has remained a source of concern to educators, Accountants and the nation at large.The Senior School Certificate Accounting Examinations conducted by the West African Senior Secondary Certificate Examination (WASSCE) tended to have revealed a decline in the performance of candidates. Many approaches aimed at improving underachieving schools have served to further disadvantage them, largely by failing to take adequate account of their context and by locating the blame for failure squarely within the school.But reasons for school failure are rarely one -dimensional or singular.Consequently, the polarized debate of "school's fault" versus "society's fault" for failure is unlikely to take us very far.Attributing blame detracts from solving the problem.This problem seems to be deepened in Oredo Local Government of Edo State. As always said, the truth lies somewhere in-between.In some cases, schools are at the heart of the problem.Poor teaching is condoned, weak leadership is tolerated, and the dominant view is, "What can you do with these kids?" driven by low expectations.Could it be that the appalling performance is attributed to motivational issues as a whole?Are there Hypotheses To guide this study, the following null hypotheses were generated and tested at 0.05 Alpha Level.H01.There is no significant difference in the performance of candidates between the Core Subjects (English and Mathematics) and Vocational Subject (Financial Accounting) in the May/June 2014 WASCE.H02 There is no significant difference in the performance of candidates in Boys Schools, Girls Schools and Mixed Schools in Financial Accounting in the May/June 2014 WASSCE. Literature Review A core course of study can be an academic course that receives high school graduation credit in the following.One or a combination of these areas English, mathematics, natural/physical science, chemistry, biology, physics.A core course will be listed in program requirement and is a mandatory course required in one's degree.Unless you have received credit for prior learning specifically for a core course, you must complete all core course listed in your program this is said to be different from elective courses, elective course are chosen by a student from a number of optional subject or course as opposed to the required or compulsory course which a student must take. Core course are deemed special for an academic degree, they are course that are prized high by the society and which the society requires each of its citizens to take for the individual development and also the development of the society.In Nigeria for example the core course are English languages, mathematic, economics, physics, chemistry, biology, literature in English, information and communication technology, general metal work, general wood work, building engineering drawing, basic electricity, commerce. Core course of study refer to series of course that the student is required to complete before they can move to the next level in their education to earn a diploma.In secondary school care course include English languages, arts, Maths, science and social studies.In some school the core course may also entail additional credits requirement in specified subject areas such as languages, computer science but not all school may define their core course this way. The general educational purpose of a core course of study is to ensure that all students take and complete course that are considered to be academically and culturally essential i.e.The course that teach students the foundational knowledge they will need in careers and adult life.Yet depending on the structure of the academic program in a particular school, the core course of study may be different from some students.For example, some school offer distinct academic programs in parallel with their regular academics programs-such as international baccalaureate or theme based academic, among many other possible option and student enrolled I this program will likely have to satisfy different requirement to complete their program to earn a degree. Credit are awarded when student complete a course with a passing grade.Therefore increasing subject area credit requirement effectively increase course requirement.This is why state may attempt to influence the quality or effectiveness of academic programs by modifying state mandated credit requirement: school may offer a wide variety or math course and academic tracks, but they all offer course in the subject of math.Still, there is nuanced distinction between core academic course and credit requirement; some history course, for example, maybe elective in a school while other are considered part of the core course of study.To complete the corecourse of study and satisfy a school's graduation requirement then, student will need to pass the required history course, not just earn a specified number of history course. Pedagogically and philosophically, the core course of study, as a reform strategy is related to concept such as access, equity, high expectation and rigor.The basic rationale is that increasing requirement in the 'core' subject will not only improve student learning and skill acquisition but it will give graduate more educational and career option because they will graduate better educated and prepared.The core course of study as reform strategy, is related to learning standards (i.e the general educational intent is similar), but course requirement are distinct from standards; a core of study establishes minimum course requirement, why standard establish minimum learning requirement.Many learning standard may be addressed or taught in a course, but standards are not specific to a certain course (although they are typically organized by subject area and grade level) learning standards describe knowledge and skill expectation, but those standards can be met either without or outside of a course, less commonly, core course of study, learning standards and other attempt to standardized what gets taught in school may be perceived by some parents or public figures as a form of 'forced curriculum' i.e., an attempt to control what get taught of students.In most cases, such criticism mirrors larger political debaters and ideological fault lines in United States, such as whether and how schools should teach the science of evolution (a highly politicized topic).While core course of study and learning standards are, in fact, overt attempts to standardize education and ensure that students learn certain foundational knowledge and skill, the majority of educators do not see ominous or ideological intent behind these strategies. Method The research design adopted for the study was exploratory research design-an initial research design which analyses the data and explores the possibility of obtaining as many relationships as possible between different variables without knowing their ends applications.This was to establish as many specific relationships as possible between the variables of the study (Pannerselvem, 2004).The population of the study covered the entire public secondary schools in Egor and Oredo Local Government Areas of Edo State.This comprised of all the students' that wrote WAEC in Financial Accounting in the May/June 2014 WASSCE.There are 12 and 13 public secondary schools in Egor and Oredo Local Government Areas respectively, making the population of the study a total of 25 public senior secondary schools, (apx 5&6).Ten per cent of the population of public senior secondary schools from each of the two local government areas was sampled for the study.The researcher adopted both stratified and multistage random sampling techniques in the collection of research data.Four schools were used, two each from the two local government areas.The multi stage sampling technique was adopted thus: Stage 1: Stratified random sampling technique was employed in order to group the schools under study into public senior secondary schools.Stage 2 Purposive sampling technique was employed using the following criteria: (i) The schools must be offering Financial Accounting as a subject in S.S.S.3 and (ii) Any of the selected school must have a Financial Accounting teacher in their school.Stage 3: Simple random sampling technique was finally employed in order to randomly select two schools each from the public senior secondary schools in Egor and Oredo Local Government Areas of Edo State Nigeria.It was presumed that the sample taken represented the whole population of the public senior secondary schools' results in Financial Accounting. A pro forma titled "Evaluation of Candidates Performance in Financial Accounting" was developed and used for this study to collect and analyze the WAEC results of schools from the sampled secondary schools.The instrument consisted of subject, total number of candidates' registered, total number of results released, the WAEC grading and percentages.The result that was used was the May/June 2014 WASSCE. The instrument was thoroughly scrutinized by the research Supervisor; two other experts in Vocational and Technical Education at University of Benin, Benin City.The Pro forma "Evaluation of Candidates Performance in Financial Accounting" was taken on its face value having been validated by the research Supervisor and two other experts mentioned above. The researcher collected a letter of introduction from the Head, Department of Vocational and Technical Education, University of Benin, Benin City along with the already prepared pro forma for the purpose of identification which enabled the researcher have better approach with the school principals who provided the data; WAEC result of the school for May/June 2014 WASSCE .The researcher was interested in Financial Accounting and the Core Subjects which he used to evaluate performance.The results were collected within a reasonable time.The data (results) collected from the schools were analyzed using descriptive statistics such as mean (x), standard deviation, percentages, analysis of variance (ANOVA) and T-test of Independent samples.The mean and standard deviation were used to analyze the data for the research questions while ANOVA and t-test was used to test the hypotheses.All hypotheses were tested at 0.05 level of significance.The decision rule was based on the probability value (p).If p-value rule was less than or equal to 0.05, the null hypotheses was rejected, but if p-value was greater than 0.05, the null hypotheses was accepted.Table 2 shows the success rates of the schools sampled for the study in Financial Accounting for the May/June 2014 West African Senior School Certificate Examination.From the statistics, a total of 72 students took the exam from the four sampled schools out of which 0 percent of the candidates failed the subject and 100 percent passed.A further break down of the statistics reflected in the table shows that in total, the percentage of students who were able to make an 'A' across the four schools was 1.61 percent.This shows that in as much as all students across the four schools passed the examination, the percentage of those students who were able to make 'A' were very low.A satisfying percentage of 45.16 percent were able to fall within 'B' across the four schools, and 48.38 percent made 'C'.Pass (P) which includes a combination of D and E had a percentage of 3.22 percent.This reveals that in as much as there were very low percentage of candidates who had excellent in the exam (A), low percentage of students were also closer to failure 3.22 percent (D and E).A total of 0 percent failed the subject across the schools.In this respect the data reveals that the performance of candidates in this examination was average. Results and Discussion In Edo College Benin City, the results reveal a 100 percent pass rate while 0 percent of the candidates failed the examination.28 students registered and took the examination in the school, 0 percent had A, 67.85 percent had B and 32.14 percent had C.There were no D and E (passes).This shows that a larger population of the students who took the examination in this school can be categorized in the "very good" bracket (B).Comparing this with Federal Girls College Benin City, the results reveal that 70.37 percent of the candidates who took the examination in this school can be categorized as "Average" (C) students, while 7.40 percent had normal passes and 22.22 percent had B. This reveals that the ownership of the school whether State or Federal School did not affect the performance of the students in the examination. In Ihogbe College Benin City, 100 percent of the candidates who took the examination in this school passed.No failure was recorded in the subject in this school.In University of Benin Demonstration Secondary School Benin City, a percentage of 16.66 percent had A in the subject compared to 0 percent in other schools.50 percent had B and 33.33 percent had C.There were no normal passes or failure recorded in the subject in this school.The grade in this school represents a more spread students' ability and intelligence. Research Question One What is the level of performance of candidate in the May/June WASSCE 2014 in Financial Accounting in Egor and Oredo Local Government Areas, Edo state, Nigeria? To answer this research question, the standard format for grading that is used by the West African Examination Council was used as was already shown in Table 1.The results obtained from the data presented were based on the responses of 4 secondary schools to items in the Pro forma titled "Evaluation of Candidates' Performance in Financial Accounting" in the May/June 2014 WASSCE.The secondary schools that participated in the sampled were Edo College, Federal Government Girls College, Ihogbe Secondary School and University of Benin Demonstration Secondary School. Table 3 above revealed that the mean performance of candidates in Financial Accounting was 62.11 with a standard deviation of 7.40.Since the mean score falls within the Credit level of the grading system shown in Table 1, it can therefore be concluded that students' performance in the May/June 2014 WASSCE in Financial Accounting was Fair. Hypothesis one There is no significant difference in the performance of candidates between the Core Subjects (English & Mathematics) and Vocational Subject (Financial Accounting) in the May/June 2014 WASSCE.To test this hypothesis, the One-Way ANOVA statistics was used.Table 5 shows the ANOVA statistics for the mean performance of candidates in English, Mathematics and Financial Accounting. 5 above that there was a significant difference in the various subjects (English, Mathematics & Financial Accounting) across the various schools sampled for the study as P<0.05 in all the subjects. To tell were the difference lies, a Post Hoc Analysis using the Scheffe was subject carried out.Table 6 shows the Post Hoc Analysis for the various subjects across the sampled schools. (a).Levels of performance of candidates in the May/June 2004 WASSCE in Financial Accounting in Egor and Oredo Local Government Areas, Edo state. mass failures in May/June WASSCE over the years in Egor and Oredo Local Government Areas in Edo State?Research Questions To gain insight into this research, the following research questions are raised.1.What is the level of performance of candidates in Financial Accounting in the May/June 2014 WASSCE in Egor and Oredo Local Government Areas? 2. Are there differences in the level of performance of candidates between the Core Subjects (English and Mathematics) and Vocational Subject (Financial Accounting) in the May/June 2014 WASSCE? 3. Are there differences in the performance of candidates between Boys Schools, Girls Schools and Mixed Schools in Financial Accounting in the May/June 2014 WASSCE? Table 3 : Table 3 below shows the mean performance of candidates in Financial Accounting in the May/June 2014 WASSCE.Standard Mean Scores for Candidates in the various Subjects in School in the May/June 2014 WASSCE Table 4 : Table 4 below shows the mean scores for candidates in the various subjects in school.Mean Scores for Candidate in the various Subjects in school in the May/June 2014 WASSCE. Table 5 : One way ANOVA for the Mean Performance of Candidates in English, Financial Accounting and Mathematics Across the four School in the May/ June 2014 WASSCE. Table 6 Post Hoc Analysis for English, Financial Accounting and Mathematics across the various Schools in the May/June 2014 WASSCE.Looking atTable 6 above, in terms of English Language there was a significant difference between Ihogbe College and Edo College, Ihogbe College and Federal Girls College and Ihogbe College and University of Benin Demonstration Secondary School.The mean table in Table 4 revealed that the other three schools performed better than Ihogbe College in English Language.Also there was a significant difference between Edo College and Federal Girls College, Edo College and University of Benin Demonstration Secondary School and there was no significant difference between Federal Girls College and University of Benin Demonstration Secondary School.But looking at Table 4, University of Benin Demonstration Secondary School and Federal Girls College performed better than Edo College in English Language.	5,854.8	2024-08-20T00:00:00.000	[ "Education", "Economics" ]
The role of enabling actors in ecosystem service accounting When accounting for ecosystem services, it is important to distinguish between the flow of services and the flow of benefits (which can be part of economic accounts or not) generated by those services. To disentangle services and benefits, particular attention has to be paid in allocating each category of flows in the use table to those institutional sectors that generate the need for the services and have the power to modify them the enabling actors and to final beneficiaries. The general concept of use, without specifying whether services or benefits are referred to, could in fact lead to misinterpretations. This paper discusses the issue of the allocation of ecosystem services and the role of enabling actors through a practical example of water purification accounts in the Netherlands. In particular, the role of the agricultural sector as an enabling actor is disentangled from the cleaned water as benefit and from water supply companies as beneficiaries. The proper allocation of the flow of the service can in fact facilitate the establishment of a causal relationship between the actions of economic actors and ecological consequences and vice versa. ‡ § © La Notte A, Marques A. This is an open access article distributed under the terms of the Creative Commons Attribution License (CC BY 4.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source Introduction The importance of ecosystems, their services and sustainable use for long-term human well-being is being increasingly recognised (MA 2005, TEEB 2010).Consequently, a proper quantification of the contribution and reliance of human activities on ecosystem services is being pursued.Several international initiatives are contributing to this effort, for example: the work led by the United Nations Statistical Division which coordinates the System of integrated Environmental and Economic Accounts-Experimental Ecosystem Accounts (SEEA-EEA) (United Nations et al. 2014a), the work developed by World Bank with the Wealth Accounting and the Valuation of Ecosystem Services (WAVES) initiative (WAVES 2014) or the Integrated system for Natural Capital Accounting (INCA) project supported by the European Commission (European Commission 2016).After the adoption of the SEEA -Central Framework (SEEA-CF) as the first international standard for environmental-economic accounting (United Nations et al. 2014b), the SEEA-EEA approach has been endorsed as a reference structure for ecosystem services accounting, with calls for its testing, application and further development at the global scale. The SEEA-EEA defines ecosystem services as the contribution of ecosystems to benefits used in economic or other human activity and records them in Supply and Use tables (SUTs).The SUTs are a powerful tool in environmental accounting because they enable the establishment of linkages and dependencies between environmental and economic flows.In ecosystem services accounting, the Supply table records the amount of ecosystem services supplied by the different ecosystem types1 and the Use table records how much the different economic sectors and households use ecosystem services; this is commonly referred to as ecosystem services demand (see Wolff et al. 2015 for a discussion on the different concepts of ecosystem services demand). Testing the SEEA-EEA is fundamental to reveal its limitations and challenges and to foster its improvement towards its adoption as an official statistical standard.One of the areas of high priority for testing and further development is how to account for degradation in ecosystem accounting, even more so since degradation is left out from the SEEA-CF (Bartelmus 2013).The SEEA-EEA Technical Recommendations (United Nations Environmental Program et al. 2017) identify two approaches to assess degradation: (i) in physical terms through changes in ecosystem condition2; and (ii) in monetary terms through changes in the net present value of the actual use of ecosystems.Several challenges for quantification still remain; for example how to determine the appropriate reference condition to compare changes in ecosystem condition and how to determine future flows of ecosystem services (for the computation of the Net Present Value, NPV).The allocation of the degradation to the proper economic unit has also been identified as a challenging aspect for several reasons including distance and time (the impacts might occur in a different place from the one where the economic unit is located and in a different accounting period) and, as degradation will likely affect the supply of different ecosystem services, the attribution of overall impacts is very complex (United Nations Environmental Program et al. 2017, Hein et al. 2016).The notion of ecosystem capacity has been discussed in several studies as an essential metric to monitor the sustainable use of ecosystems and their services (for example Bagstad et al. 2014, Schröter et al. 2014, Villamagna et al. 2013). Hein et al. 2016 discuss the need to include the concepts of ecosystem capacity, capability and potential supply in the SEEA-EEA framework to define ecosystem assets and highlight their importance for monitoring ecosystem degradation beyond changes in NPV.Recently, a test case for implementation of the SEEA-EEA at the European scale also highlighted some challenges associated with the issues of sustainable use and degradation in ecosystem accounting (La Notte et al. 2017a).First, the study discusses that considering only the actual flow3 of ecosystem services is unlikely to be sufficient to analyse the sustainable use of ecosystem services; if the actual flow is higher than the natural regeneration or absorption rates, this will lead to over-exploitation and potentially degradation of ecosystems and their capacity to provide services.As a solution to tackle this issue, the study suggests to add information on the sustainable or potential flow of ecosystem services in the Supply table, keeping the actual flow recorded in the Use table.This would ensure that the official Supply = Use identity is preserved.Second, it emphasises the need for complimentarity and consistency between the SEEA-EEA, the SEEA-CF and the System of National Accounts, so that no room is left for ambiguities when accounting for ecosystem services.Finally, the authors highlight the fact that the separation between the benefits received from ecosystems (e.g.clean water, timber) and the actual ecosystem service (water purification, biomass growth) creates the need to differentiate between those benefiting from the outcome of the service (beneficiaries) and those whose activities create the need for the service and have the power to modify the service flow (enabling actors4). Sometimes beneficiaries and enabling actors overlap, but in other cases, there are clear differences that have to be considered in accounting terms.For example, when a factory releases a pollutant into a stream, it is enabling the generation of the water purification service, but the factory does not directly benefit from the cleaned water eventually generated; instead water companies and households will be the direct beneficiaries of cleaned water (La Notte et al. 2017a).The separation between enabling actors and beneficiaries does indeed generate an important modification in the Use Table because it determines where to allocate the service flow: this is going to be a remarkable change compared to the current frame.In order to contribute to this new stream of applied research, in this paper two accounting frameworks are compared: the one currently proposed by SEEA-EEA and an enlarged SEEA-EEA version that addresses the use of complementary information concerning the sustainable flow of ecosystem services and the allocation to enabling actors.The role of enabling actors is explored by using an ecosystem service accounting application already available for all European countries (La Notte et al. 2017b).After initially presenting the accounting tables with and without the allocation to enabling actors, the methods by which different accounting frameworks lead to different analyses and thus conclusions will be analysed. Water Purification Accounts The application reported here concerns a regulating ecosystem service characterised by being a sink-related service5: water purification.In-stream nitrogen (N) retention is used as proxy for the actual flow of water purification.Excessive N loading is a leading cause of water pollution which makes N a useful indicator for water quality (Rockström et al. 2009).N retention is defined as the process of temporary or permanent removal of nitrogen taking place in the river.This includes the processes of denitrification, burial in sediments, immobilisation and transformation or simply transport.To calculate the amount of N, the GREEN model (Geospatial Regression Equation for European Nutrient losses) was used (Bouraoui and Grizzetti 2014).GREEN contains a spatial description of N sources and physical characteristics influencing N retention.N sources are classified as diffuse sources and point sources.Diffuse sources include mainly mineral fertilisers, manure applications and crop fixation, while point sources consist of industrial and wastewater treatment discharges.For this reason, diffuse source is considered as representing the 'agriculture' sector and point sources representing 'other industries and households'.The amount of N that is retained and removed by rivers and lakes is then converted into a Constructed Wetland Equivalent Area (CWEA) expressed in hectares.A replacement cost approach is used to estimate the monetary value of the physical units produced by the CWEA.The full description of the biophysical assessment and monetary valuation of this ecosystem service is described in detail in previous publications (La Notte et al. 2017b, La Notte et al. 2012). Here the focus is on the accounting tables and their interpretation.Specifically, it is important to show how Supply and Use tables would look when applying the accounts as currently reported in SEEA-EEA (United Nations et al. 2014a) and when applying additional complementary accounts as proposed in La Notte et al. (2017a) with a specific emphasis on enabling actors.In La Notte et al. (2017b), when considering the difference between sustainable and actual flow, one of the countries with the most critical situation was the Netherlands and therefore this country is used as the case study throughout the paper. Water purification accounts according to SEEA-EEA -allocation to final beneficiaries According to the SEEA-EEA, SUTs record actual flows.In the Use table, actual flow is attributed to water supply companies that are the beneficiaries of the clean freshwater generated through water purification (Fig. 1).As in the SEEA-EEA Technical Recommendations (United Nations Environmental Program et al. 2017), the grey cells are marked wherever no data is going to be entered: ecosystem types provide services but they neither provide [section D] nor use [section H] products; economic units cannot provide ecosystem services [section A] but they can use them [section E].On the other hand, inter-ecosystem flows of services might occur [section F].The accounting of interecosystem flows strongly depends on the methodology applied to assess the service flow.In the specific case of water purification, the biophysical model used does not allow the reporting of any contribution to other ecosystem types.Sections C and D represent the System of National Accounts (SNA) Supply and Use tables that are completed according to the standard conventions (European Commission et al. 2009).No data were filled in this case since no standard SNA products are of relevance for water purification.The accounts for water purification, in fact, should not be confused with the accounts for water as reported in the SEEA-CF.The tables reported in Fig. 1 show that, over 20 years, the monetary value of the water purification service decreased.This negative trend indicates that, at the end of the analysed period, the service had less value than at the beginning of the period analysed, suggesting that according to standard economic theory, the water purification service had become less scarce.The main driver of the change in the biophysical model are N emissions; the outcome of the model is then translated into monetary terms by using a replacement cost technique.As N emissions decreased from 1985 to 2005, the monetary value of the water purification actual flow also decreased.The residual section was also included, this being part of the SEEA-CF [Section E and Section N].The reader should bear in mind that N emissions reported here are the same input variables used in the biophysical model adopted to assess the water purification service: there is an accounting linkage that inherently connects the N emission account to the water purification account.It is indeed through this linkage that the causal relationship between N emissions and the value of water purification is established: the more N emitted, the more N removal which is assessed by the biophysical model and thus valued as water purification (and vice versa).The environmental asset more closely related to water purification is inland water bodies.Data related to actual flow of water [section D] and gross total abstraction [section I] are withdrawn from Eurostat datasets.The reason to look at water resources is justified by the non-SNA benefit generated by water purification, i.e. clean water [section K].It should in fact be calculated as the fraction of N cleaned freshwater abstracted by water supply companies.At the moment, this indicator is not available: what matters in the frame is to show where to allocate the number in terms of "what" (i.e. the indicator of clean water related to the outcome of the biophysical model [N removed] and the water abstracted by water supply companies) and to "whom" (the beneficiary: water supply companies).To have residuals in the frame greatly helps in seeing the linkage with the service flow.The N input reported in the tables is the same input variable that runs into the model and should be consistent with Eurostat datasets.The frame is fully consistent.First, it is possible to link higher/lower N input [section E] to a changing level, as measured in the monetary value of water purification flow [section B], to detect which part of N is retained in soil (that is not part of the water purification service) and which part flows into the inland waters [section N].Second, by reporting the sustainable flow [section B] and actual flow [section G], it is possble to assess whether the current management is sustainable [section H, which is obtained by the difference between section B and section G].A negative sign shows that degradation is occurring and also its order of magnitude.The case of the Netherlands is very critical when checking the difference between the very low sustainable flow and the excessive actual flow.In Europe, the only countries where the sustainable flow is higher than the actual flow are Sweden, Norway and Finland (La Notte et al. 2017b).Trends show whether there are improvements over time: less N input occurs in the Netherlands, but the decrease is not sufficient to reach sustainability: the disparity between the two flows [section B versus section G] is too severe and the capacity of already degraded waterbodies to remove N is lower (which is consistent with the spatial model structure). The meaning of the negative numbers in [Section H] can be seen as depreciation of natural capital, i.e. the consumption of fixed capital in accounting terms.In other words, the overuse of water purification (negative sign) lowers the capacity of inland water ecosystems to provide the same amount of the water purification service in the coming years.The lost capacity can be tracked on the biophysical dataset when the N outflow for each subcatchment becomes gradually equal to the N inflow: this implies that N cannot be removed and thus that inland waters are being degraded (Bouraoui and Grizzetti 2014).Fig. 2 shows a hybrid account frame, where services [sections B, G and H] are expressed in monetary terms and benefits [sections I and K] and residuals [sections E and N] are expressed in physical terms.Ecosystem services can be reported in both physical and monetary terms.The reason to report here in monetary terms is justified by the need to link these numbers with economic accounts.The non-produced assets [section D] and SNA benefits [section I] can be reported both in physical and monetary terms, while non-SNA benefits and removals are likely to be reported in physical terms.The reason to report here water abstraction in physical terms is to show what part of this information (million m of water) will be linked to the non-SNA indicator related to clean water (less N tonnes/m of abstracted freshwater) [section K]. Outcome analysis In this section, the interpretation of the information reported on the Supply and Use Tables presented in the case study is provided.Results are analysed by addressing one specific issue: the causal relationship between the action of economic actors and ecological degradation.For the Netherlands, the assessment and valuation of the actual flow and the trend for the selected years suggest that when N emissions [section G and section N] are high, water purification's actual value [section G] is high, when N emissions are low, water purification's actual value is low (Fig. 2). A different trend and very different monetary values, emerge when assessing and valuing the sustainable flow.The more N emissions decrease [section G and section N], the more water purification's sustainable values increase [section B and section H].Since degradation has been an ongoing process for a long time, the value of sustainable flows is very low compared to the value of actual flow.In terms of sustainability assessment, considering only the actual flow would then provide misleading information: i.e.N emissions are good and enrich waterbodies.On the other hand, considering the sustainable flow provides the view that the sustainability path has only just begun and that the major player in this path is the reduction of N emissions.This clearly shows up in Fig. 2 by considering the residual accounts [section E and section N] and the difference between sustainable and actual flows [section H]. However, the most needed relationship to provide information on the sustainable use of ecosystem services relates to economic accounts (or in other words, information from the SNA).In order to understand how economic production affects the ability of inland waterbodies to provide water purification, the accounting tables filled in the previous section with information from the SNA are compared.In Fig. 3,the Value Added (VA) of the water supply related sectors is compared with the actual flow of water purification (i.e. the SEEA-EEA where the actual flow is attributed to water supply companies, as reported in Fig. 1).The source of all economic data is Eurostat National Accounts by 21 branchesaggregates at current prices (NACE rev2).By looking at these data alone, it is hard to find an answer to the relevance that water purification might have on economic production or vice versa. In Fig. 4, the total VA of all NACE activities is compared with the total water purification sustainable flow, in line with what was reported in Fig. 2, i.e. the SEEA-EEA with complementary information where the sustainable flow in the supply table is considered and where, as the user of the service, those who activate and modify the flow of the service are also considered.Once again, by looking at these data, it is difficult to find an answer to any relevance water purification might have on economic production or vice versa. In determining which economic data to relate with relevant ecosystem services, major drivers need to be considered.As can be seen in Fig. 2 [section E], the major source of N is agriculture.In a biophysical assessment, this source of pollution is referred to as a diffuse source, ad hoc measured in physical terms and then valued in monetary terms.When a question such as how economic production affects the ability of inland waterbodies to provide water purification needs to be addressed, in this case, the sector that is responsible for 90% of N emissions (i.e.agriculture) should be investigated and the ecosystem services withdrawn by it (i.e.diffuse source) should be considered.The ability of inland waterbodies to provide water purification is reported in the Supply table as sustainable flow [section B].Fig. 5 shows this trend and it can be seen that a causal relationship indeed exists.The attempt to establish a link between sustainable flow and water-related companies (Fig. 6) is not helpful, since the real relationship will, in this case, be the link between the non-SNA benefit clean freshwater (i.e. the percentage of N removal per m of water) and the water-related companies that are indeed the beneficiaries.In fact, the only common trend that can be tracked is that an increase in the sustainable flow of water purification moves in the same direction as that of water companies' value added.Their processing and provision of water for multiple uses is indeed already part of the SNA.It is important to consider that Value Added measurements also include compensation for employees, taxes on production and imports, subsidies and gross operating surplus.It might be interesting to look at measurements which specifically consider the value of output as the quantity of output multiplied by the price.Specifically, for water purification, it would matter to look at crop output and animal output (Fig. 7).In the Netherlands, crop production does not record a decrease in production, while the real decrease concerns animal production, even if not dramatically.The remarkable decrease in N emissions (check [Section E] as supply and [Section N] as use in Fig. 2) has been appropriately managed by the agricultural sector by not decreasing the crop production and by slightly decreasing animal production.If enabling actors were not disentagled from final beneficiaries, this kind of analysis could not take place. Conclusions An accounting system, as support for policy making, should provide information on relationships associated with homogeneous groups of actors in order to evaluate, analyse and forecast economic phenomena (European Commission 2014). Already the SEEA-CF acknowledges the presence of a melding of many disciplines, such as economics, statistics, energy, hydrology, forestry, fisheries and environmental science (ref. 2.3 United Nations et al. 2014b), each with its own concepts and structures.This multidisciplinary aspect becomes even more relevant for the SEEA-EEA.While the underlying structure remains the same as that used in the national accounts, the SEEA-EEA should integrate perspectives from ecology and natural science disciplines to properly measure and report about ecosystems and ecosystem services and thus provide an improved body of information for environmental-economic analysis.The role of enabling actors in ecosystem service accounting The ecological perspective in terms of 'users' of ecosystem services may require an additional effort to separate services and benefits.For some ecosystem services (especially the sink-related services), final beneficiaries might play no role in affecting the amount and increasing/decreasing trend of ecosystem services flow.When no causal relationship is established, it is not possible to plan and implement policies to address environmental issues.The example of water purification provided demonstrates that the agricultural sector has a common trend with water purification; implementing sustainable practices in agriculture can reduce degradation in water bodies and this can be measured through water purification.Specifically, reducing N outputs from agriculture would improve the capacity of inland waters to provide N reduction services and thereby reduce costs of water purification.It is a powerful means to provide evidence for a number of regulations and policy actions; in Europe for example, the Nitrate Directive and the agri-environment payments of the Rural Development Programmes. When no causal relationship is established, then it is not possible to develop strategies to reduce ecosystem services degradation.To use residual accounts already in the SEEA-CF is not enough.In the case of water purification, it is possible to read, through the accounting tables, that most of the N emissions are captured by soil and only part of them flow into water bodies.To only consider clean water as benefits is not correct because, in some cases, the beneficiaries of water do not need clean water (e.g.hydroelectric sector).The separation between services and benefits allows in turn the separation of enabling actors from final beneficiaries and thus disentangling and developing that logical step which, in the current SEEA-EEA frame, remains embedded. Testing of the SEEA-EEA by different specialists in different fields is essential for its improvement and to ensure that ecosystem services accounts hold the necessary data to convey the information on the interdependencies between economies and ecosystems. Endnotes In the SEEA-EEA, ecosystem types refer mostly to different land cover/ecosystem functional units. 1 According to SEEA-EEA, ecosystem condition reflects the overall quality of an ecosystem in terms of its characteristics (for example, biodiversity, vegetation etc.). 2 By actual flow, the authors mean the use of an ecosystem service in a given year. 3 The definition of "enabling actors" is inspired by the SEEA-EEA (ref. Annex 3 in United Nations et al. 2014b) where the authors introduce for regulating services the concept of "enabling factor".The evolution of this concept leads to the identification of the subjects responsible for the existence/occurrence of the factor itself and thus to the source of relevant causality.Data from the Eurostat website has been extracted from the "Renewable freshwater resources" [env_wat_res] Figure 1 . Figure 1.Supply and Use Tables for water purification in the Netherlands according to the SEEA-EEA. Figure 2 . Figure 2. Supply and Use Tables for water purification in the Netherlands according to the experimental proposal. enabling actors in ecosystem service accounting Figure 3 . Figure 3. Water related sectors Value Added and water purification actual flow in the Netherlands 1995-2005. Figure 4 . Figure 4. Total Value Added and water purification sustainable flow in the Netherlands 1995-2005. 3 Figure 5 . Figure 5.Primary sector Value Added and water purification sustainable flow in the Netherlands 1995-2005. Figure 6 . Figure 6.Water related sectors Value Added and water purification sustainable flow in the Netherlands 1995-2005.	5,782	2017-12-18T00:00:00.000	[ "Economics" ]
Primal Decomposition-Based Method for Weighted Sum-Rate Maximization in Downlink OFDMA Systems . We consider the weighted sum-rate maximization problem in downlink Orthogonal Frequency Division Multiple Access (OFDMA) systems. Motivated by the increasing popularity of OFDMA in future wireless technologies, a low complexity suboptimal resource allocation algorithm is obtained for joint optimization of multiuser subcarrier assignment and power allocation. The algorithm is based on an approximated primal decomposition-based method, which is inspired from exact primal decomposition techniques. The original nonconvex optimization problem is divided into two subproblems which can be solved independently. Numerical results are provided to compare the performance of the proposed algorithm to Lagrange relaxation based suboptimal methods as well as to optimal exhaustive search-based method. Despite its reduced computational complexity, the proposed algorithm provides close-to-optimal performance. Two main radio resource allocation (RRA) problems have been addressed in the literature. The first ones, consist of maximizing an increasing function of the user rates [5][6][7][8][9][10][11][12][13] subject to different power constraints, whilst the second ones consist of minimizing the transmit power subject to constraints on the minimum user rates [13][14][15][16][17]. A suboptimal greedy method for maximizing the smallest rate among the users has been proposed in [5]. A branch-andbound based algorithm for sum-rate maximization has been proposed in [6]. However, in practice the branch-and-bound method is still too computationally heavy for finding the global solution [23]. Computationally efficient algorithms for maximizing the sum-rate have been developed in [7,8]. A suboptimal method for characterizing the achievable rate region of the two-users frequency division multiple access (FDMA) channel have been presented in [10]. The general weighted sum-rate maximization problem has been used in [9] to characterize FDMA capacity region for a broadcast channel. Due to the nontractability of the original problem, a modified convex problem formulation, FDMA-time division multiple access (TDMA) was proposed (i.e., time sharing among users). Authors also considered algorithms to obtain optimal and suboptimal solutions to a particular variation of the original problem where the total power is evenly divided among the used set of subcarriers. Lagrangian relaxationbased approaches to obtain suboptimal algorithms for the weighted sum-power minimization problem has been introduced in [13,14]. A greedy algorithm is proposed in [15] to obtain and approximate solutions for the same problem. Recently, a Lagrangian relaxation-based method has been proposed in [13] for the weighted sum-rate maximization problem. A bisection search method was used to update the dual variable until the algorithm converges. Due to the nonconvexity of the optimization problem the optimality of the algorithm is not guaranteed. In this paper, we propose an alternative method based on the primal decomposition technique [24,25]. Using numerical simulations, its performance is compared to Lagrangian relaxation based algorithm [13] as well as to the optimal exhaustive search algorithm. Numerical results show that the proposed algorithm converges very fast. Although, the optimality of the final value cannot be guaranteed due to the nonconvexity of the problem, the simulations show that rate-region achieved by the proposed algorithm exactly matches with the one obtained using optimal exhaustive search algorithm. The rest of the paper is organized as follows. In Section 2 we present the system model and problem formulation. The proposed algorithm is presented in Section 3 and convergence properties are discussed in Section 4. Section 5 compares the complexity of the proposed algorithm to Lagrangian relaxation-based algorithm in [13] as well as to the optimal exhaustive search based method. The numerical simulation results are presented in Sections 6, and 7 concludes our paper. Notations. \|x\| denotes the absolute value of complex number x. x T represents the transpose of vector x and e i denotes the ith standard unit vector. CN (c, σ 2 ) stands for circularly symmetric complex Gaussian distribution with mean c, variance σ 2 /2 per dimension. For any real number r, [r] + denotes max{0, r}. System Model and Problem Formulation Consider a single antenna (Note that, under OFDMA assumption, extension to the multiple antennas case is straightforward [26].) OFDMA downlink transmission with K users and M subcarriers as shown in Figure 1. The signal received by user k in subcarrier m can be expressed as where k is the user index, m is the subcarrier index, S k denotes the set of subcarriers allocated to user k, x km is the transmitted signal, p km is the power allocated, h km is channel frequency response, and w km is the received noise. We assume that h km is time-invariant and its value is available at the base station. The noise samples are assumed to be independent and identically distributed as w km ∼ CN (0, σ 2 km ). We denote by c km = \|h km \| 2 /σ 2 km the channel signal-to-noise ratio (SNR) of kth user in subcarrier m and by β k the weight associated with the rate of user k. The weighted sum-rate maximization problem subject to a sum-power constraint P T can be formulated as [13] maximize K k=1 m∈Sk β k log 2 1 + p km c km subject to K k=1 m∈Sk where variables are p km and S k . It is also useful to introduce a virtual system where each subcarrier can be used by all users in the same time. This results in a general OFDMA downlink channel where the signal received by user k in subcarrier m is given by and the second term in the right-hand side represents the interference from other users. Assuming independent channel coding across users at the transmitter and independent decoding at receivers, the weighted sum-rate maximization problem for the virtual system can be formulated as where the optimization variables are p km . The constraints associated with orthogonal subcarrier allocations in problem (2) have been dropped out and the interference among users allocated to the same subcarrier is reflected in the objective. Here we can make several observations. First, any solution of problem (4) is such that the second constraint in problem (2) is automatically satisfied, for reasons that will be explained in the beginning of Section 3.2. In other words, any solution of problem (4) is feasible for problem (2). Moreover, at any of these solutions the objective function of problem (4) will be exactly the same as the objective function of problem (2). Based on these observations it can be concluded that any solution of the auxiliary problem (4) is a solution for the original problem (2) as well. The original problem (2) is combinatorial and it requires exponential complexity to find a global optimum. Although problem (4) is still nonconvex it is noncombinatorial. Thus, in the following, we focus on solving problem (4) instead of solving the original problem (2). A similar approach has been used in [7] to solve the (nonweighted) sum-rate maximization problem, that is, for the particular case β k = 1, k = 1, . . . , K. However, the methods proposed there do not apply to the general case of arbitrary weights, for reasons that will become clear in Section 3.3. Due to the nonconvexity of problem (4) finding the global optimum is intractable. Thus, an approximative method inspired from the primal decomposition technique is presented in Section 3. Primal Decomposition. To reveal the complicating constraints [24], we introduce M new variables p m = K k=1 p km , m = 1, . . . , M, and reformulate the problem (4) as follows: where the optimization variables are p km and p m . Note that p m represents the total power on subcarrier m. Treating p m as complicating variables, problem (5) can be decomposed [24,25] into a master problem and M subproblems, one subproblem for each subcarrier m = 1, . . . , M. For a given subcarrier m, the subproblem is given by where variables are p km , k = 1, . . . , K. The master problem can be expressed as where variables are p m and f m (p m ) represents the optimal value of subproblem (6) for fixed p m . Algorithm Derivation. Let us denote by P the feasible set of problem (6). Note that subproblem (6) is not a convex optimization problem (Since we maximize a convex function.). However, its objective function is convex with respect to (w.r.t.) optimization variables p 1m , . . . , p Km , its feasible set is a nonempty convex polyhedral set (i.e., a simplex [27]) and its objective is bounded above on P . Thus, by following the approach of [7, Section III], from [28, Corollary 32.3.4] (If a convex function f is bounded above on a convex set X ⊆ dom f , then the maximum of f relative to X is attained at one of the finitely many extreme points of X.) it follows that the solutions of subproblems (6) must be achieved at one of the vertices of the polyhedral set P . Consequently, the solutions of the M subproblems can be expressed as where j m represents the index of the user allocated to mth subcarrier, that is, Solution (8) confirms that, even though in subproblems (6) all users are allowed to use all subcarriers, the optimal power allocation consists of allocating only one user per subcarrier. This guarantees that solution (8) is feasible for the original problem (2). By substituting (8) and (9) in the objective of (6), f m (p m ) can be expressed as EURASIP Journal on Wireless Communications and Networking We note that the index j m depends on p m according to (9) and the function f m (p m ) is the pointwise maximum of a set of concave functions. Therefore f m (p m ) is not a concave function w.r.t. p m [27] in general. Thus, standard convex optimization tools (e.g., subgradient-based methods) cannot be directly applied to solve master problem (7). We propose an iterative method, where at each iteration i we first solve the M subproblems (6) to obtain an user-tocarrier allocation j (i) m for a given subcarrier power allocation Then, the objective of the master problem is approximated by the following lower bound where P denotes the feasible set of the master problem (7). The lower bound is concave w.r.t. p 1 , . . . , p M and the solution of the approximated master problem can be found by multilevel waterfilling algorithm [9]. The resulting solution is used as subcarrier power allocation for the next iteration. The proposed algorithm can be summarized as follows. (3) solve the following approximation of master problem and return the solution p m ; let p m (i+1) = p m . (4) check a stopping criteria; if it is satisfied EXIT, otherwise let i = i + 1 and go to step (2). The solution of problem (13) solved at step (3) is given by the following multilevel waterfilling expression [9]: where L is chosen such that the power constraint is satisfied with equality, that is, Particularization to the Sum-Rate Maximization. The problem of the sum-rate maximization (i.e., β k = 1 for all k = 1, . . . , K) in downlink OFDMA systems is solved in [7, Section III]. The solution method is exactly equivalent to only one iteration of the APD algorithm. Unlike the general weighted sum-rate maximization, in which user weights β k 's are different, in the sum-rate maximization (i.e., β k = 1 for all k = 1, . . . , K) the index j m will not depend on p m according to (9). Thus, by using (10) and (11) the function f m (p m ) can be found as f m (p m ) = log 2 (1 + p m c jmm ) = log 2 (1 + p m · max k c km ) which is concave w.r.t. p m (recall that the function f m (p m ) is not concave w.r.t. p m when the user weights β k 's are different). As a result, the inequality given in (12) holds with equality and solving problem (7) gives the optimal subcarrier power allocation [7, Section III]. Convergence Behavior and Exit Criterion In this section, we start by investigating the monotonicity of the proposed algorithm. Then we provide a specific exit criterion which certifies that algorithm converged to a fixed power and subcarrier allocation followed by a simple graphical illustration. Monotonic Behavior. The following theorem states the monotonic behavior of the proposed algorithm. Theorem 1. For any iteration that is, the proposed APD method is an ascent algorithm. Proof. From (10) and (11), it follows that the solution of (6) in iteration i is given by, Now we can write the following chain of relations, where the first inequality follows from (13), the second one follows trivially from the maximization over the users, and the last equality follows from (11) and (10), respectively. Exit Criterion. The exit criterion for such ascent algorithm is typically chosen heuristically, for example, the increasing in the objective between two successive iterations is below a certain predefined threshold. However, for the proposed algorithm we are able to find an exit criterion which certifies that algorithm converged to a fixed power and subcarrier allocation and further improvement is not possible. This is described by the following theorem. Theorem 2. If at iteration n +1, (n ≥ 1) we have j (n) m = j (n+1) m , m = 1, . . . , M then the following holds: That is, the algorithm converges to a fixed power and subcarrier allocation. Proof. Since c km 's are continuous random variables, the probability to have multiple solutions for (9) is zero. Thus, in the following we assume that j m given by (9) is unique (Equation (9) has multiple solutions if and only if c km = c lm for some l / = k. When p m = 0 we assign j m any arbitrarily user index.). Note that the objective function of (13) is strictly concave. Thus it has a unique solution [27]. Therefore, for all m, j (n) . Since (9) has a unique solution as well, p (n+1) , item (1) follows directly by induction. Furthermore, item (2) follows from item (1) by the uniqueness of the solution of problem (13). Finally, item (3) follows trivially from item (2). Thus the exit criterion checks if the subcarrier allocation between two successive iterations remains unchanged. Such point is a local optimum (possible global) in the sense that the objective cannot be increased by changing the power allocation or subcarrier allocation only. As a specific example, consider the simple OFDMA system with two subcarriers (i.e., m = 1, 2). By performing the variable transformations p 1 = (1 − t)P T and p 2 = tP T , t ∈ [0, 1], we can express the variation of 2 m=1 f m (p m ) on P as, which is plotted in Figure 2. According to Figure 2(a) global optimal is achieved at the iteration (i + 3). Achieving global optimality is not always possible because, quasiconcavity [27] of h(t) cannot be guaranteed with random channel SNR, c km . Consequently the APD algorithm can converge to a local optimal solution as shown in Figure 2(b). Complexity Analysis In this section, we analyze and compare the computational complexity of the proposed APD algorithm to Lagrangian relaxation-based algorithm [13] as well as to the optimal exhaustive search algorithm. With K users and M subcarriers, altogether we have K M user-subcarrier combinations. Therefore finding optimal subcarrier and power assignment requires K M searches. Combined with multilevel waterfilling at each instance of subcarrier assignment, O(MK M ) operations are required to find the solution. (a) Convergence to the global optimal solution The algorithm proposed in [13] for the weighted sumrate maximization problem requires O(MK) operations to obtain a suboptimal solution. The proposed APD algorithm described in Section 3 requires O(MK) operations in step (2) and O(Mlog 2 M) operations (This is the number of operations required in ordering.) in step (3). In practice, it is reasonable to assume that K log 2 M (The assumption is reasonable since the number of users simultaneously serviced by the system can be very large. For example, in a Wi-Max system M can be up to 2048 [29]. However, the value of log 2 M will not become very large (in a WiMax system log 2 M = 11 at most).). Therefore the complexity of the APD algorithm can be approximated by O(MK). Numerical Results The performance of the proposed APD algorithm is compared to the dual decomposition-based algorithm proposed in [13], denoted as WSRmax, as well as to the optimal algorithm based on exhaustive search. The WSRMax algorithm uses a bisection search method to update the dual variable λ [13, Section IV]. For initializing the bisection search interval, 6 EURASIP Journal on Wireless Communications and Networking [λ min , λ max ], we exploit the fact that the subgradient of the dual function can be analytically computed. Since the dual function is convex, the sign of its subgradients changes as we pass through the minimum point of the dual function [13, equation (11)]. Therefore, we use a grid search (with step size 1) to identify the interval in which the subgradient of dual function changes its sign, and it is used as initial bisection search interval. Thus, the interval [λ min , λ max ] is guaranteed to contain the optimal value of the the dual function and the width of the initial interval is one, that is, (λ max − λ min ) = 1. The proposed APD algorithm is initialized by allocating equal power to all subcarriers. In what follows, we compare the convergence behavior of the APD and the WSRMax algorithms. For a fair comparison, we define the following metric: the average normalized weighted sum-rate deviation, where C opt is the optimal weighted sum-rate value obtained using optimal exhaustive search, C subopt is the estimated objective value from either the APD algorithm or the WSRMax algorithm, and expectation E{·} is taken w.r.t. channel realization. An OFDMA system with M = 8 subcarriers and a uniform power delay profile with 4 channel taps is considered. We assume σ 2 km = σ 2 , k = 1, . . . , K, m = 1, . . . , M and define SNR per subcarrier as P T /(M · σ 2 ). Figure 3 shows the convergence behavior of the considered algorithms with SNR= 10 dB for K = 2 and K = 4 users. The weights of the users are [1,2] for K = 2 and [1, 2, 1, 2] for K = 4. The floor of the curves is due to the suboptimality of the algorithms. The results show that the APD algorithm converges faster than the WSRMax algorithm and provides smaller average normalized weighted sum-rate deviations. Specifically, for both cases, K = 2 and K = 4, the APD algorithm requires only 3 iterations on average to achieve an average normalized weighted sum-rate deviation of 10 −4 whilst the WSRMax algorithm requires around 15 iterations to reach the same accuracy level. It is intuitively obvious that the number of iterations required by the APD algorithm is sensitive to the nature of the surface of the objective function M m=1 f m (p m ) of problem (7), for example, see Figure 2. In general, it is a hard to quantify the number of iterations before convergence (or any bounds on the number of iterations) due to the nonconvexity of problem (5). However, the numerical results suggest that the APD algorithm often converges very fast in practice. It should be emphasized that the number of iterations required in the initialization of the WSRmax algorithm (i.e., the number of iterations required to find the initial bisection search interval) is not considered when drawing the curves. In particular, for the initialization process, the WSRMax algorithm requires a several number of steps (each step has complexity of O(MK)) and the proposed APD algorithm requires none. Moreover, it is hard to find good initialization methods for the WSRMax algorithm (i.e., initialization for bisection search method) compromising between the number of steps required in the initialization and the width of the initial searching interval (λ max − λ min ). Consequently, additional precautions are required and therefore, in practical implementations the APD algorithm is more favorable compared to the WSRMax algorithm. In the sequel, we compare the behavior of the APD and the WSRMax algorithms using the following metric: the probability of missing the global optimal, where ε is a small number which quantifies the maximum admissible deviation between C opt and C subopt . It is considered that the global optimum is missed if C subopt is more than ε away from C opt . Figure 4 uses the same simulation setup as that in Figure 3 and depicts the variation of probability of missing the global optimal, P ε with the number of iterations. The floor of probability P ε is again due to the suboptimality of both algorithms. The influence of ε on P ε is totally indistinguishable in case of the APD algorithm. This behavior shows that the proposed algorithm APD can arrive very close-tooptimal solutions within a very small number of iterations and then it remains there. The results further show that the P ε evaluated using the WSRMax algorithm is highly dependent on ε. That is, the smaller the deviations in the C subopt from the optimal C opt , the larger the number of iterations required by the WSRMax algorithm to reach the expected target value P ε . Therefore, independent from the ε, the APD algorithm allows to find a suboptimal solution within a small number of iterations at the expense of a slight increase in P ε . These observations are very useful in practice since they carry significant information in the system design point of view. For example, consider a design requirement P 10 more the design requirement as P 10 −6 ≤ 0.3, then the number of iterations required by the WSRmax increases to 24. In contrast, the APD algorithm always requires just one iteration. Figure 5 shows the rate region (The standard way to characterize the boundary points in the 2-user rate region is by solving problem (4) for β 1 = α and β 2 = 1 − α, where α ∈ [0, 1] [30].) computed by using all considered algorithms. The same simulation setup as in [13] T , respectively. Although the computational complexity of the proposed algorithm is much smaller compared to that of optimal exhaustive search-based method, Figure 5 indicates that the rate region obtained by the APD algorithm almost coincides with the optimal rate region. This behavior is expected since the average normalized weighted sum-rate deviation, (20) is in the order of 10 −4 as shown in Figure 3. In the following we compare the behavior of the APD and the WSRmax algorithm for large number of subcarriers and users. Since, for large number of users and subcarriers the complexity of evaluating C opt is prohibitively high, the metrics defined in (20) and (21) are not used. The behavior of the APD algorithm is compared with that of the WSRmax algorithm. In Figure 6, the evolution of the expected weighted sum-rate provided by the APD algorithm is compared to the resulting expected weighted sum-rate from the WSRmax algorithm, where the expectation is taken w.r.t. channel realization. An OFDMA system with M = 256 subcarriers, a uniform power delay profile with 128 channel taps, and K = 8, 16, 32, 64 users is considered. The weights of the users are taken from the sequence {1, 2, 1, 2, . . . , 1, 2}, (e.g., when K = 8, weights are [1, 2, 1, 2, 1, 2, 1, 2]). The SNR is assumed to be 10 dB. The results show that even for a large number of car- riers, the APD algorithm converges very fast as compared to the WSRMax algorithm independent of the number of users. Conclusions A joint subcarrier and power allocation algorithm which is inspired from primal decomposition techniques has been proposed for maximizing the weighted sum-rate in multiuser OFDMA downlink systems. Although the original problem is nonconvex, the proposed APD algorithm finds fast a 8 EURASIP Journal on Wireless Communications and Networking suboptimal, but still very close-to-optimal solution with very high probability (i.e., more than 90% of the time). Unlike the recently proposed WSRMax algorithm [13], the APD algorithm requires no additional precautions in the initialization, and convergence to a suboptimal solution is possible within a very small number of iterations. Although the proposed primal decomposition-based solution method does not rely on zero duality gap for proving the optimality in the case of large number of subcarriers, our computational experience with larger number of subcarriers suggests that the proposed APD algorithm is capable of finding the same solution as the WSRmax algorithm (which is asymptotically optimal when the number of carriers grows to ∞) even with very few iterations.	5,878.6	2010-04-01T00:00:00.000	[ "Computer Science", "Engineering" ]