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 |
|---|---|---|---|
Induction motor condition monitoring using infrared thermography imaging and ensemble learning techniques
In this paper, a novel noncontact and nonintrusive framework experimental method is used for the monitoring and the diagnosis of a three phase’s induction motor faults based on an infrared thermography technique (IRT). The basic structure of this work begins with this applying IRT to obtain a thermograph of the considered machine. Then, bag-of-visual-word (BoVW) is used to extract the fault features with Speeded-Up Robust Features (SURF) detector and descriptor from the IRT images. Finally, various faults patterns in the induction motor are automatically identified using an ensemble learning called Extremely Randomized Tree (ERT). The proposed method effectiveness is evaluated based on the experimental IRT images, and the diagnosis results show its capacity and that it can be considered as a powerful diagnostic tool with a high classification accuracy and stability compared to other previously used methods.
Introduction
Electrical systems condition monitoring plays a vital role in maintenance costs minimizing as well as reliability increasing. Recently, vibration signal analysis has been the most widely used method to monitor rotating machines and diagnose their faults. Several signal processing techniques have been developed such as those published by Zair et al., 1 Bettahar et al., 2 Nayana et al., 3 Glowacz et al., 4 Ikhlef et al. 5 Since vibration is considered a non-avoidable phenomenon in dynamic systems, the isolation and the diagnosis of coupled faults is generally difficult and not easy to establish due to the complexity of the structure and the machinery multiple components interactions. For this purpose, accelerometers are usually needed. However, they require being in contact with the object that needs to be monitored. Multiple challenges are encountered and must be considered before the installation of accelerometers, namely the high operating temperatures and greasy surfaces.
Infrared thermal imaging is a noncontact and nonintrusive measurement technique that can detect all the monitored system component temperature variation. This technique has been largely used in nondestructive examination, 6 medical science, 7 defense, 8 and automotive. 9 In recent years, it has been known that rich information is contained in IRT images and can be used as diagnosis data for several electrical machines such as bearing diagnosis of rotating machinery, 10 Grinder 11 diagnosis, and fault diagnosis of electric impact drills. 12 Electrical motor faults detection is performed based on IRT images as demonstrated by Glowacz and Glowacz 13 who developed a technique called Method of Area Selection of Image Differences (MoASoID) which is mainly based on analyzing infrared thermal images for three-phase induction motor different faults identification. Li et al. 14 adopted a technique to diagnose faults in rotating machinery using Conventional Neural Network (CNN) for fault features extraction from the captured infrared thermal images. After that, fault pattern is identified by feeding the obtained features into the Softmax Regression (SR) classifier. Devarajan et al. 15 propose a fault diagnosis method for induction machines, at first, by using the temperature pixels indicator of the thermal image, they addressed three types of faults that are known to provoke an increase in the stator temperature such as shaft misalignment, air gap eccentricity, and cooling system failure, for which different degree of temperature variation are directly related to pixels values of the IRT image. Then, the extracted features were classified using ANFIS structure model. The collected IRT images are analyzed in order to verify the accuracy of the proposed method.
Most of the methods presented above lack precision and stability of their system, this leads us to look for an efficient and more stable method for the monitoring and the diagnosis of a three phase's induction motor using an infrared thermography technique. The current proposed diagnosis method is a combination of fault extraction technology with a new machine learning method called ensemble learning for faults pattern classification.
In this work, a sophisticated approach of IRT images features extraction and indexing using SURF and BoVW is adopted. Speeded-Up Robust Features 16 (SURF) is a robust technique for image features extraction used to detect the interest points in IRT image and produce their descriptors. In addition to the distinctive and the resistance to noise and detection errors, the points of interest are also insensitive to geometric and photometric changes. They are key points with welldefined locations in the image's scale space and a rough representation of the of the image object.
Bag of words (BoW) had shown a notable efficiency in text retrieval, which extends it to image processing under the name of Bag of Visual Word 17 (BoVW). Like BoW, BoVW transforms an image in the form of histogram that can be defined as visual features frequencies of occurrences in the treated image. It is considered as a set of discrete words known for their unordered pattern and non-distinctiveness, this can be seen as an invariance to the spatial location of the objects in the image. Furthermore, the histogram of the visual words is considered as a bank of features to be used in image classification.
The next and the most important step after image features extraction is features classification. In order to detect and identify different fault in a rotating machine a robust and reliable classifier is needed. Classification is one of the top research subjects and issues in the machine learning discipline. In the area of fault diagnosis, several machine learning methods have been introduced, such as decision tree (DT), support vector machine (SVM), extreme learning machine (ELM), k-nearest neighbor (KNN), ., etc.
However, machine learning present some limits. For example, it requires lengthy offline/batch training and doesn't not learn incrementally or interactively in realtime. Furthermore, it has a poor transfer learning ability, reusability of modules, and integration. The opacity of the systems makes them very hard to debug. In order to overcome these problems, researchers are oriented toward a new machine learning techniques called ensemble learning. Ensemble methods help machine learning results improvement by combining multiple models. Using ensemble methods allows to produce better classification compared to a single model method.
There are many ensemble learning techniques that has been recently developed and embedded in the field of classification such as Random Forest 23 (RF) and Extremely Randomized Tree 18 (ERT). This paper's purpose is to propose a new intelligent method to diagnose faults in electromechanical systems based on IRT, image feature extraction using BoVW and SURF methods, and Extremely Randomized Tree (ERT) classifier. The proposed method had demonstrated its effectiveness in three phase's induction machine faults diagnosis.
In addition to conventional techniques (KNN, SVM, DT, LSSVM, RF), our method has been also compared with some recently developed AI techniques, namely Self-Organising Fuzzy logic classifier 19 (SOF) and Semi-Supervised Deep Rule-Based 20 (SSDRB) approach for image classification which are two classification methods recently developed and published in 2018.
Experiments indicate that, based on Extremely Randomized Tree (ERT) ensemble learning classifier, the discussed method achieves high accuracy and best stability in induction motor diagnosis and proves its superiority over the traditional methods and standard deep learning methods.
Image features extraction using Bag of Visual Words (BoVW)
The great success shown by ''Bag-of-Words'' (BoW) in text retrieval opened new horizons to its utility in other domains as a reliable features extraction method. Its extension in image processing is called ''Bag-of-Visual-Words'' (BoVW). Just like BoW, BoVW transforms an image in the form of histogram that can be defined as visual features in the treated image. This histogram is considered as the effective features bank that will be used later in image classification. 10 In this paper, the applied image processing technique is BoVW. It allows the extraction of features from the infrared thermographs images for three phases induction motor faults diagnosis. The BoVW method 14 is obtained by the two steps that are presented below: Step 1: Visual words extraction method There are many local features detectors and descriptors algorithms that have been used to procure the visual words from the interests regions in the infrared thermography images.
A pixel is basically described in an image by local feature descriptors via its local content. These descriptors must show a high robustness against localization errors and deformations, they have to ensure the identification of the corresponding pixel locations in images which capture the same kind of quantitative information about the spatial intensity patterns in various states modes.
In the next paragraph we succinctly explain the Scale Invariant Feature Transform (SIFT) and Speeded Up Robust feature (SURF) extraction methods of local features and comparison between them.
Scale invariant feature transform (SIFT)
This method had been firstly introduced by Lowe. 21 A 128-dimensional vector called SIFT descriptor which saves in an histogram of eight main orientations the gradients of 4 _ 4 locations around a pixel. A rotation invariant descriptor is given by the alignment of the gradients to the main direction. This descriptor becomes scale invariant through vector's computation in different Gaussian scale spaces. The Invariant rotation descriptors can lead to false matches in some implementations such as face recognition. If invariance with respect to rotation is not necessary, the descriptor gradients alignment can be oriented toward a fixed direction.
It is very important to detect and identify the stable locations of the interest points in a scale space. This can be done using the difference of Gaussian function scalespace extreme. 17 The scale space L (r, s) is known as the convolution of a variable scale Gaussian function G (r, s) with the original image I (r), it can be written as follows: Where * is the convolution operator and r = (x, y) is a point in the IRT image.
The Gaussian function G (r, s) is defined as: The scale-space extreme convolution using D (r, s), allows us to separate the difference of two scales using a multiplicative index k, which is given by the following expression: While à represents the convolution product. The local extreme of the function D can be detected by accurately localizing the interest points with respect to the proposed method known as Taylor expansion of the scale-space function D (r, s) that is shown in equation (4).
Speeded up robust feature (SURF) The SURF algorithm which was firstly presented by Bay et al. 16 is a novel detector and descriptor of scale and rotation invariant interest points. It generates a group of interest points for each image with a set of 128 dimensional descriptors for each point.
In addition to its conceptual similarity to the SIFT, the SURF descriptor likewise focusses on the gradient information distributed space within the interest point neighboring, where the localization and the description of the interest point itself can be done by its detection approaches or in a regular grid. SURF is characterized by its invariance to rotation, scale, brightness and, after reduction to unit length, contrast. That's why SURF computation is fast and it can increase distinctively, without losing its robustness to rotation of about 6 15°, which is typical for most face recognition tasks. The SURF descriptors are known for their higher robustness compared to the locally operating SIFT descriptors when it comes to dealing with different kinds of image perturbations.
SURF detector interest point localization is based on the Hessian matrix. If a point r = (x, y) in an IRT image I, is considered then, the Hessian matrix H = (r, s) at r and scale s is written as follows: While L xx (r, s) is the second order derivative of the Gaussian convolution ∂ 2 ∂ x 2 G(r, s) with the image I, and likewise for L xy (r, s) and L yy (r, s).
The SURF is close to second order Gaussian derivatives with box filters named average filter, by contrast the SIFT that is closer to Gaussian Laplacian (LoG) using the Gaussian difference (DoG), this can be rapidly calculated through integral images like shown in Figure 1. The selection of interest point's location and scale is realized by computing the Hessian matrix determinant. The application of non-maximum suppression in a 3 3 3 3 3 neighborhood allows us to localize the interest points in scale and image space.
By using the SURF descriptor, a circular region is constructed around the detected interest point and thus a unique orientation is assigned. Haar wavelet response in both x and y directions is used to compute the orientation, that helps to gain invariance to image rotations. Haar wavelets can be easily calculated by displaying integral images.
The SURF descriptors are created by extracting the square areas around the points of interest when the dominant orientation is evaluated and included in the information on the points of interest. Each windows underlying intensity pattern (first derivatives) is represented by a vector V and each sub-regions are split up in 4 * 4. Quick distinctive descriptors computation is considered one of the major SURF descriptor advantages. Furthermore, its invariance to common image transformations such as image rotation, scale and illumination changes, and small change in viewpoint makes it more reliable in this task.
Based on what has preceded, it is fairly justified to choose the SURF method as an adopted feature extractor in our experimental work.
Step 2: Histogram of SURF features The obtained features are encoded as a histogram which represents the occurrence frequency of these visual words. 17 Before encoding features, k-means clustering 16,22 is used to generate several interest points called vocabulary.
A specified vocabulary size can be obtained by clustering each faulty state extracted features and a set of k; clusters is learned. After that, the centers of the learned clusters are defined as the vocabulary. In this paper, a set of n dimensional vectors of SUFT features (x1, x2, ., xn) is done. Via k-means clustering, the n SUFT vectors are divided into k different groups such that S = {S1, S2, ., Sk} with minimized intracluster squared summed error (SSE), which is done as: Where mi are the mean vector in Si.
Extremely randomized tree (ERT) classifier
In both Random Forest 23 (RF) and Extremely Randomized Tree 18 (ERT) based on a great number of decision trees (DT) are used as proper classifiers in order to attain the ultimate classification. These methods consist of four major stages: identification of the input, selection of the optimal number of trees, the analysis of the votes, and the final decision.
There are several differences between Random Forests (RF) and Extremely Random Trees (ERT), of which one can quote. 24 a. Unlike the RF, the ERT uses all the training samples to build each decision tree. b. To bifurcate the decision tree, the ERT is completely random, while the RF uses a random subset.
Random forest
RF method was originally developed by Breiman. 23 It is considered as one of the most successful ensemble learning method. In order to perform a good classification and regression, a great number (hundreds or thousands) of independent decision trees are utilized. Since it is an ensemble methodology, RF employs a number of decision trees as weak classifiers or regressors, combines the concept of bagging, 24 and random feature selection. 25 The RF is a representative ensemble classifier built by a multitude of decision trees. Thanks to its excellent classification accuracy and high efficiency, RF classifier has been attracting increasing attention and wide using domains.
Extremely randomized trees
ERT method 18 is a group of randomized trees. The ERT maximizes RF algorithm randomization. Just like RF, ERT shows a computational effectiveness and a training capability even when fed with high dimensional input vectors. However, ERT surpasses the RF technique when the training time is taken into account. This rapidity is thanks to a simpler manner to choose the thresholds in ERT. Besides, unlike RF, ERT has a lowers variance that is ensured by its increased randomization. Furthermore, trees in ERT are autonomously trained using the total training data points. RF and ERT have different core methodologies even if they are both considered as ensemble learning models based on the decision tree algorithm. 18,26 Random forest subsamples the data with replacement. This subsampling enriches the data, thereby helping to form the model with highly distinctive learning data.
Oppositely, ERT uses the whole data and hence reduces bias which has a positive impact on the model's performance. Additionally, both methods have different ways in splitting nodes, RF splits nodes by finding the optimum split while ERT does it randomly. Data variance is lowered in extremely randomized trees by suing this random split. Thereby, both ERT and RF first goal is to ensure an optimal balance of bias and variance. RF and ERT are selected at first place for the reason that their algorithms are based on an ensemble of decision trees. This property gives them a higher performance than the traditional decision tree algorithm.
An ERT classifier is similar to RF but differs in how the randomness is introduced during the training. To train an ERT classifier multiple trees are trained, each tree is trained on all training data. Like the random decision forest, the optimal division at the node level is obtained by analyzing a subset of all available features. Instead of searching for the best threshold for each feature a single threshold for each feature is selected at random. Based on these random divisions, the division that leads to the highest increment in the scale score employed is then selected. The greater degree of randomness through training results in more independent trees and thus reduces the variance further. Due to that extremely randomized trees tend to give better results than random forests.
Experimental work
Experiments were conducted on a three phases 1.11 kW, 5 A, 220/230 V. In order to test and train the model, a healthy induction motor and eight different short circuit faults in the stator windings thermal images are captured and considered. All artifact generated defects in this dataset are internal faults, with no relation with external pieces or initial setup electrical components failure. 27 Thermal images acquisition was performed on an Electrical Machines Laboratory workbench by a Dali-tech T4/T8 infrared thermal image camera at an environment temperature of 23°. Thermal camera properties and induction motor specifications are representing in Table 1.
Nine sets of Images represent a health state and eight different faulty states of the Induction Machine. The %-stator stands for each phase short-circuit rate, while a-phase is the number of phases that contain shortcircuit defects as representing in Table 2.
The thermal images acquisition under healthy and defect conditions are shown in Figure 2. We can clearly notice that the detection and identification of the defect are almost impossible counting only on thermographs direct observation because healthy and faulty conditions thermal images are not distinctive and their In this context, we propose in this paper a new induction motor condition monitoring method based on extremely randomized trees (ERT) classifier combined with SURF-BoVW features extraction. Figure 3 shows the proposed method flowchart. After the collection of infrared image thermography (IRT), in the first step, features set (visual words) are extracted using the SURF algorithm. Secondly, the histogram is generated after the encoding of the extracted features using the k-mean clustering. Then, the obtained features set are randomly divided into training and testing samples.
Finally, ERT classifier is used for classification. After the training phase is finished and the parameters of the ERT classifier are fixed, the testing samples are classified based on the adopted parameters.
Obtained results and discussion
In the aim to prove the robustness of the adopted method, it is compared with other classification techniques that include KNN, DT, SVM, least squares support vector machine (LSSVM), self-organizing fuzzy logic classifier (SOF), Random Forest (RF), and Deep Rule Based (DRB) respectively.
Moreover, this method's classification stability had been analyzed based on the standard deviation of 10 experiments. Additionally, the average, maximum, and minimum values are taken to reduce the impact of the contingency.
All the above mentioned classification methods consider the features set extracted by BOVW and SURF as an input. In standard SVM, the penalty factor equals to 100, and the kernel function is 0.01. DTs minimum number of father nodes is 5. K = 5 is taken as the nearest neighbor number of KNN. The Gaussian kernel function of the LSSVM is 0.5, and its regularization parameter is 10,000. The obtained results from each classification method are represented in Table 3. Table 3 illustrates the low classification accuracy of KNN compared to the other methods. In addition to that, the standard deviation revealed the unsteadiness classification effect of this method on different testing samples, and the instability of its classification algorithm.
KNN is lower than SVM both in the average classification accuracy and its standard deviation by 3.7700% and 0.0015 successively. The SVM standard deviation classification accuracy is 0.0308. However its algorithm is unstable, with a 10.2 % lower minimum classification accuracy than that of ERT.
The maximum classification accuracy of SOF, DT, SVM, LSSVM, RF, and DRB is 100%, but the standard deviation of classification accuracy are higher than that of ERT classifier. The classification accuracy is greatly influenced by the input of different samples in SOF, DT, LSSVM, RF, and DRB.
Compared to ERT, the average classification accuracy of RF is considerably significant. Even so, if we focus on the standard deviation of classification accuracy, the RF classification stability is not as good as the proposed method.
Globally, the classification accuracy of our method is the highest of all the tested methods. Consequently, its classification result is the best and the most reliable.
In order to offer an intuitive illustration of the classification effects resulting from various methods, Figures 4 and 5 respectively shows the confusion matrix and classification results of the previously discussed methods in the eighth experiment.
From the figure of the KNN confusion matrix, we can see that the KNN has the lowest classification accuracy which equals 93.2099%, and we can also clearly see that the KNN has a misclassification in categories 2nd, 3rd, 7th, and 8th. So that around 8.7% of the test samples are misclassified for the second category, 15.79% and 23.53% of the test samples for the third and seventh category respectively, and 11.76% of the samples are misclassified for the category 8.
In the classification results of KNN, 6.7901% of testing samples are misclassified, of which 8.7% of In the classification results of proposed method (ERT), there are no misclassified samples, and the classification accuracy is 100%.
Furthermore, the evaluation of this experiment's results from different perspectives is performed using polygon Area Metric (PAM).
Classification evaluation using polygon area metric
In order to assess a single scale classifier's performance, Polygon Area Metric 28 is a novel method that is used thanks to its stability and its profound measure.
Six existing metrics including CA, SE, SP, AUC, JI, and FM are used to create a polygon, then the corresponding area for PAM is calculated. The theoretical formulas are mentioned bellow: Jaccard Index = JI = TN TN + FP ð11Þ Area Under Curve = AUC = Where TP, TN, FP, and FN are respectively known as the correctly predicted positive and negative samples numbers, and the incorrectly predicted positive and negative samples ones. The true-positive rate (SE) plot in function of the false-positive rate (1-SP) for different cut-off points R represents the receiver operating characteristic curve f(x). It should be known that SE and SP respectively refer to the ratios of correctly classified class 1 and class 2 samples total population.
As illustrated in Figure 6, the PAM calculation is done using the polygon's area created by CA, SE, SP, AUC, JI, and FM points in a regular hexagon. It worth noting that the regular hexagon is made up of 6 equilateral triangles and each side length is equal to 1. Therefore, it is fair to say that |OP 1 | = |OP 2 | = |OP 3 | = |OP 4 | = |OP 5 | = |OP 6 | = 1, with an area of 2.59807. |OP 1 |, |OP 2 |, |OP 3 |, |OP 4 |, |OP 5 |, and |OP 6 | lengths respectively are the values of CA, SE, SP, AUC, JI, and FM. The calculation of the PAM is done based on the following formula: PA is the polygon's area.
It should be mentioned that the normalization of the PAM into the [0, 1] interval is ensured by dividing the PA value by 2.59807. Figure 7 shows the visual results of Polygon area metric of each classification method in the eighth experiment. From these visual graphs, it can be observed among the PAM covered areas of all the tested classification methods, ERT has largest area, followed by that of the RF classifier.
Conclusion
A new induction motor fault diagnosis method based on SURF-BoVW and ERT classifier is proposed in this paper to diagnose various faults of induction motors. The IRT-based method is proven to be effective noncontact, nonintrusive, high sensitive, and stable to various faults.
The efficiency of the proposed method is validated by identifying eight sorts of induction motor stator faults (the rate of short-circuit in each phase and the number of phases included faulty phases in the induction motor).
Under the premise of the same input, the ERT classifier is always higher than that of RF, and the classification effect is better and stable. By comparing it with SVM, DT, KNN, LSSVM, and DRB, the ERT classifier had demonstrated the highest classification accuracy, stability, and the best fitted to be used in the diagnosis of induction motor faults, since its classification precision had reached 100%.
In order to evaluate the proposed method, polygon Area Metric (PAM) is used. The obtained results clearly show that the ERT Polygon Area Metric followed by the rest of the classifiers.
Compared to other existing classification methods, the obtained experimental results using ERT classifier indicate that the proposed method can be considered as a reliable alternative to monitor the state of an induction motor.
Declaration of conflicting interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The author(s) received no financial support for the research, authorship, and/or publication of this article. | 6,105.4 | 2021-11-01T00:00:00.000 | [
"Engineering"
] |
Cosmological Constraints on Unstable Particles: Numerical Bounds and Analytic Approximations
Many extensions of the Standard Model predict large numbers of additional unstable particles whose decays in the early universe are tightly constrained by observational data. For example, the decays of such particles can alter the ratios of light-element abundances, give rise to distortions in the cosmic microwave background, alter the ionization history of the universe, and contribute to the diffuse photon flux. Constraints on new physics from such considerations are typically derived for a single unstable particle species with a single well-defined mass and characteristic lifetime. In this paper, by contrast, we investigate the cosmological constraints on theories involving entire ensembles of decaying particles --- ensembles which span potentially broad ranges of masses and lifetimes. In addition to providing a detailed numerical analysis of these constraints, we also formulate a set of simple analytic approximations for these constraints which may be applied to generic ensembles of unstable particles which decay into electromagnetically-interacting final states. We then illustrate how these analytic approximations can be used to constrain a variety of toy scenarios for physics beyond the Standard Model. For ease of reference, we also compile our results in the form of a table which can be consulted independently of the rest of the paper. It is thus our hope that this work might serve as a useful reference for future model-builders concerned with cosmological constraints on decaying particles, regardless of the particular model under study.
Oct 2018 of observable consequences -especially if the final states into which these particles decay involve visible-sector particles. Indeed, electromagnetic or hadronic showers precipitated by unstable-particle decays within the recent cosmological past can alter the primordial abundances of light nuclei both during and after Big-Bang nucleosynthesis (BBN) [1][2][3][4], give rise to spectral distortions in the cosmic microwave background (CMB) [5,6], alter the ionization history of the universe [7][8][9][10], and give rise to characteristic features in the diffuse photon background. These considerations therefore place stringent constraints on models for new physics involving unstable particles.
Much previous work has focused on examining the cosmological consequences of a single particle species decaying in isolation, and the corresponding limits on the properties of such a particle species are now well established. Indeed, simple analytic approximations can be derived which accurately model the effects that the decays of such a particle can have on many of the relevant observables [2,5]. However, many theories for new physics involve not merely one or a few unstable particles, but rather a large -and potentially vast -number of such particles with a broad spectrum of masses, lifetimes, and cosmological abundances. For example, theories with additional spacetime dimensions give rise to infinite towers of Kaluza-Klein (KK) excitations for any field which propagates in the higher-dimensional bulk. Likewise, string theories generally predict large numbers of light moduli [11][12][13] or axion-like particles [14][15][16]. Collections of similar light fields also arise in supergravity [17]. There even exist approaches to the dark-matter problem such as the Dynamical Dark Matter (DDM) framework [18,19] which posit the existence of potentially vast ensembles of unstable dark-sector particles. It is therefore crucial to understand the cosmological consequences of entire ensembles of decaying particles in the early universe and, if possible, to formulate a corresponding set of analytic approximations which model the effects of these decays.
For a variety of reasons, assessing the effects of an entire ensemble of decaying particles with a broad range of masses, lifetimes, and cosmological abundances is not merely a matter of trivially generalizing the results obtained in the single-particle case. The decay of a given unstable particle amounts to an injection of additional electromagnetic radiation and/or other energetic particles into the evolution of the universe, and injection at different characteristic timescales during this evolution can have markedly different effects on the same observable. Moreover, since many of these observables evolve in time according to a complicated system of coupled equations, the effects of injection at any particular time t inj depend in a non-trivial way on the entire injection history prior to t inj through feedback effects.
In principle, the cosmological constraints on ensembles of decaying particles in the early universe can be evaluated numerically. Indeed, there are several publicly available codes [10,20,21] which can readily be modified in order to assess the effects of an arbitrary additional injection history on the relevant observables. Computational methods can certainly yield useful results in any particular individual case. However, another complementary approach which can provide additional physical insight into the underlying dynamics involves the formulation of approximate analytic expressions for the relevant observables -expressions analogous to those which already exist for a single particle species decaying in isolation. Our aim in this paper is to derive such a set of analytic expressions -expressions which are applicable to generic theories involving large numbers of unstable particles, but which nevertheless provide accurate approximations for the relevant cosmological observables. Thus, our results can serve as a useful reference for future model-builders concerned with cosmological constraints on decaying particles, regardless of the particular model under study.
In this paper, we shall focus primarily on the case in which electromagnetic injection dominates -i.e., the case in which the energy liberated by the decays of these particles is released primarily in the form of photons, electrons, and positrons rather than hadrons. We also emphasize that the approximations we shall derive in this paper are not ad-hoc in nature; in particular, they are not the results of empirical fits. Rather, as we shall see, they emerge organically from the underlying physics and thus carry direct information about the underlying processes involved.
This paper is organized as follows. In Sect. II, we begin by establishing the notation and conventions that we shall use throughout this paper. We also review the various scattering processes through which energetic photons injected by particle decay interact with other particles present in the radiation bath. In subsequent sections, we then turn our attention to entire ensembles of unstable particles, focusing on the electromagnetic injections arising from decays occurring after the BBN epoch. Each section is devoted to a different cosmological consideration arising from such injection, and in each case we ultimately obtain a simple, analytic approximation for the corresponding constraint. For example, in Sect. III we consider the constraints associated with the modification of the abundances of light nuclei after BBN, and in Sect. IV we consider limits on distortions of the CMB-photon spectrum. Likewise, in Sect. V we consider the constraints associated with the ionization history of universe and its impact on the CMB, and in Sect. VI we consider the constraints associated with additional contributions from unstable-particle decays to the diffuse photon background. Ultimately, the results from these sections furnish us with the tools needed to constrain decaying ensembles of various types. This is then illustrated in Sect. VII, where we consider how our results may be applied to two classes of ensembles whose constituents exhibit different representative mass spectra. Finally, in Sect. VIII, we conclude with a discussion of our main results and avenues for future work. For future reference, we also provide (in Table IV) a summary/compilation of our main results.
II. ELECTROMAGNETIC INJECTION: OVERVIEW AND CLASSIFICATION OF RELEVANT PROCESSES
Our aim in this paper is to assess the cosmological constraints on an ensemble consisting of a potentially large number N of unstable particle species χ i with masses m i and decay widths Γ i (or, equivalently, lifetimes τ i ≡ Γ −1 i ), where the index i = 0, 1, . . . , N − 1 labels these particle species in order of increasing mass m i . We shall characterize the cosmological abundance of each of the χ i in terms of a quantity Ω i which we call the "extrapolated abundance." This quantity represents the abundance that the species χ i would have had at present time, had it been absolutely stable. We shall assume that the total abundance Ω tot ≡ i Ω i of the ensemble is sufficiently small that the universe remains radiation-dominated until the time of matter-radiation equality t MRE ≈ 10 12 s. Moreover, we shall focus on the regime in which m i 1 GeV for all χ i and all of the ensemble constituents are non-relativistic by end of the BBN epoch. Within this regime, as we shall discuss in further detail below, the spectrum of energetic photons produced by electromagnetic injection takes a characteristic form which to a very good approximation depends only on the overall energy density injected [1]. By contrast, for much lighter decaying particles, the form of the resulting photon spectrum can differ from this characteristic form as a result of the immediate decay products lacking sufficient energy to induce the production of e + e − pairs by scattering off background photons [22,23]. The cosmological constraints on a single electromagnetically-decaying particle species with a mass below 1 GeV have recently been investigated in Ref. [24].
The considerations which place the most stringent constraints on the ensemble depend on the values of Ω i and τ i for the individual ensemble constituents. For ensembles of particles with lifetimes in the range 1 s τ i t now , where t now denotes the present age of the universe, the dominant constraints are those related to the abundances of light elements, to spectral distortions of the CMB, and to the ionization history of the universe. The effect of electromagnetic injection on the corresponding observables is sensitive to the overall energy density injected and to the timescales over which that energy is injected, but not to the details of the decay kinematics or the particular channels through which the χ i decay. Thus, in order to retain as much generality as possible in our analysis, we shall focus on ensembles for which all constituents with non-negligible Ω i have lifetimes within this range; moreover, we shall refrain from specifying any particular decay channel for the χ i when assessing the bounds on these ensembles due to these considerations. By contrast, the constraints on decaying ensembles that follow from limits on features in the diffuse photon background do depend on the particulars of the decay kinematics. Thus, when analyzing these constraints in Sect. VI, we not only present a general expression for the relevant observable -namely the contribution to the differential photon flux from the decaying ensemblebut also apply this result to a concrete example involving a particular decay topology.
Many of the constraints on electromagnetic injection are insensitive to the details of the decay kinematics because the injection of photons and other electromagnetically-interacting particles prior to CMB decoupling sets into motion a complicated chain of interactions which serve to redistribute the energies of these particles. In particular, the effects of electromagnetic injection on cosmological observables ultimately depend on the interplay between three broad classes of processes through which these photons interact with other particles in the background plasma. These are: • Class I: Cascade and cooling processes which rapidly redistribute the energy of the injected photons. Processes in this class include γ + γ BG → e + + e − and γ + γ BG → γ + γ, where γ BG denotes a background photon, as well as inverse-Compton scattering and e + e − pair production off nuclei. These processes occur on timescales far shorter than the timescales associated with other relevant processes, and thus may be considered to be effectively instantaneous. As we shall see, these processes serve to establish a non-thermal population of photons with a characteristic spectrum.
• Class II: Processes through which the non-thermal population of photons established by Class-I processes can have a direct effect on cosmological observables. These include the photoproduction and photodisintegration of light elements during or after BBN, as well as the photoionization of neutral hydrogen and helium after recombination.
• Class III: Processes which serve to bring the non-thermal population of photons established by Class-I processes into kinetic and/or thermal equilibrium with the radiation bath. Processes in this class include Compton scattering, bremsstrahlung, and e + e − pair production off nuclei.
We emphasize that these classes are not necessarily mutually exclusive, and that certain processes play different roles during different cosmological epochs. Any energy injected in the form of photons prior to last scattering is rapidly redistributed to lower energies due to the Class-I processes discussed above. The result is a non-thermal contribution to the photon spectrum at high energies with a normalization that depends on the total injected power and a generic shape which is essentially independent of the shape of the initial injection spectrum directly produced by χ i decays. This "reprocessed" photon spectrum serves as a source of for Class-II processes -processes which include, for example, reactions that alter the abundances of light nuclei and scattering processes which contribute to the ionization of neutral hydrogen and helium after recombination. Since all information about the detailed shape of the initial injection spectrum from decays is effectively washed out by Class-I processes in establishing this reprocessed photon spectrum, the results of our analysis are largely independent of the kinematics of χ i decay. This is ultimately why many of our results -including those pertaining to the alteration of light abundances after BBN, distortions in the CMB, and the ionization history of the universe -are likewise largely insensitive to the decay kinematics of the χ i .
The timescale over which injected photons can cause these alterations is controlled by the Class-III processes. Prior to CMB decoupling, these processes serve to "degrade" the reprocessed photon spectrum established by Class-I processes by bringing this non-thermal population of photons into kinetic or thermal equilibrium with the photons in the radiation bath. As this occurs, these Class-III interactions reduce the energies of the photons below the threshold for Class-II processes while also potentially altering the the shape of the CMB-photon spectrum. These Class-III processes eventually freeze out as well, after which point any photons injected by particle decays simply contribute to the diffuse extragalactic photon background.
III. IMPACT ON LIGHT-ELEMENT ABUNDANCES
We begin our analysis of the cosmological constraints on ensembles of unstable particles by considering the effect that the decays of these particles have on the abundances of light nuclei generated during BBN. We shall assume that these decays occur after BBN has concluded, i.e., after initial abundances for these nuclei are already established. We begin by reviewing the properties of the non-thermal photon spectrum which is established by the rapid reprocessing of injected photons from these decays by Class-I processes. We then review the corresponding constraints on a single unstable particle species [2]constraints derived from a numerical analysis of the coupled system of Boltzmann equations which govern the evolution of these abundances. We then set the stage for our eventual analysis by deriving a set of analytic approximations for the above constraints and demonstrating that the results obtained from these approximations are in excellent agreement with the results of a full numerical computation within our regime of interest. Finally, we apply our analytic approximations in order to constrain scenarios involving an entire ensemble of multiple decaying particles exhibiting a range of masses and lifetimes.
A. Reprocessed Injection Spectrum
As discussed in Sect. II, the initial spectrum of photons injected at time t inj is redistributed effectively instantaneously by Class-I processes. A detailed treatment of the Boltzmann equations governing these processes in a radiation-dominated epoch can be found, e.g., in Ref. [1]. For injection at times t inj 10 12 s, the resulting reprocessed photon spectrum turns out to take a characteristic form which we may parametrize as follows: The quantity dρ(t inj )/dt inj appearing in this expression, which specifies the overall normalization of the contribution to the reprocessed photon spectrum, represents the energy density injected by particle decays during the infinitesimal time interval from t inj to t inj + dt inj . The function K(E, t inj ), on the other hand, specifies the shape of the spectrum as a function of the photon energy E. This function is normalized such that It can be shown that for any process that injects energy primarily through electromagnetic (rather than hadronic) channels, the function K(E, t inj ) takes the universal form [1,25,26] K(E, t inj ) = 3) where K 0 is an overall normalization constant and where E C and E X are energy scales associated with specific Class-I processes whose interplay determines the shape of the reprocessed photon spectrum. The normalization convention in Eq. (3.2) implies that K 0 is given by Physically, the energy scales E C and E X appearing in Eq. (3.3) can be understood as follows. The scale E C represents the energy above which the photon spectrum is effectively extinguished by pair-production process γ + γ BG → e + +e − , in conjunction with interactions between the resulting electron and positron and other particles in the thermal bath. The energy scale E X represents the threshold above which γ + γ BG → γ + γ is the dominant process through which photons lose energy. By contrast, below this energy threshold, the dominant processes are Compton scattering and e + e − pair-production off nuclei. Note that while the normalization of the reprocessed photon spectrum is set by dρ(t inj )/dt inj , the shape of this spectrum is entirely controlled by the temperature T inj at injection. This temperature behaves like T inj ∝ t −1/2 inj in a radiation-dominated epoch. This implies that as t inj increases, the value of E C also increases. This reflects the fact that the thermal bath is colder at later injection times, and thus an injected photon must be more energetic in order for the pair-production process γ + γ BG → e + + e − to be effective. Numerically, the values of E C and E X at a given injection time t inj are estimated to be [1,2] where m e is the electron mass and T inj is the temperature of the thermal bath at t inj . The reprocessed photon spectrum in Eq. (3.1) is the spectrum which effectively contributes to the photoproduction and/or photodisintegration of light elements after the BBN epoch. In order to illustrate how the shape of this spectrum depends on the injection time t inj , we plot in Fig. 1 the function K(E, t inj ) which determines the shape of this spectrum as a function of E for several different values of the injection time t inj . Since the ultraviolet cutoff E C in the photon spectrum increases with t inj , injection at later times can initiate photoproduction and photodisintegration reactions with higher energy thresholds. The dashed vertical line, which we include for reference, represents the lowest threshold energy associated with any such reaction which can have a significant impact on the primordial abundance of any light nucleus which is tightly constrained by observation. As discussed in Sect. III B, this reaction turns out to be the deuterium-photodisintegration reaction D + γ → n + p, which has a threshold energy of roughly 2.2 MeV. Thus, the portion of the photon spectrum which lies within the gray shaded region in Fig. 1 has no effect on the abundance of any relevant nucleus. Since E C lies below this threshold for t inj ≈ 10 4 s, electromagnetic injection between the end of BBN and this timescale has essentially no effect on the abundances of light nuclei.
One additional complication that we must take into account in assessing the effect of injection on the abundances of light nuclei is that the reprocessed photon spectrum established by Class-I processes immediately after injection at t inj is subsequently degraded by Class-III processes, which slowly act to bring this reprocessed spectrum into thermal equilibrium with the background photons in the radiation bath. The timescale δt th (E, t inj ) on which these processes act on a photon of energy E is roughly As discussed in the text, this function determines the shape of the reprocessed photon spectrum for an instantaneous injection of photons of energy E at time tinj. The dashed vertical line at E ≈ 2.2 MeV indicates the lowest threshold energy associated with any photoproduction or photodisintegration reaction which can significantly alter the abundance of any light nucleus whose primordial abundance is tightly constrained by observation.
cross-section for the relevant scattering processes, which include Compton scattering and e + e − pair production off nuclei. The cross-sections for all relevant individual contributing processes can be found in Ref. [1]. Note that while at low energies σ th (E) is well approximated by the Thomson cross-section, this approximation breaks down at higher energies as other processes become relevant.
Given these observations, the spectrum of the resulting non-thermal population of photons at time t not only represents the sum of all contributions from injection at all times t inj < t but also reflects the subsequent degradation of these contributions by the Class-III processes which serve to thermalize this population of photons with the radiation bath. This overall non-thermal photon spectrum takes the form where is the Green's function which solves the differential equation The population of non-thermal photons described by Eq. (3.7) serves as the source for the initial photoproduction and photodisintegration reactions ultimately responsible for the modification of light-element abundances after BBN. We shall therefore henceforth refer to photons in this population as "primary" photons.
In what follows, we will find it useful to employ what we shall call the "uniform-decay approximation." Specifically, we shall approximate the full exponential decay of each dark-sector species χ i as if the entire population of such particles throughout the universe were to decay precisely at the same time τ i . As we shall see, this will prove critical in allowing us to formulate our ultimate analytic approximations. We shall nevertheless find that the results of our approximations are generally in excellent agreement with the results of a full numerical analysis.
Within the uniform-decay approximation, the contribution to dρ/dt(t inj ) from the decay of a single unstable particle species χ takes the form of a Dirac δ-function: where ρ χ (t inj ) is the energy density of χ at time t inj and where χ is the fraction of the energy density released by χ decays which is transferred to photons. It therefore follows that in this approximation, the primary-photon spectrum in Eq. (3.7) reduces to
B. Light-Element Production/Destruction
Generally speaking, the overall rate of change of the number density n a of a nuclear species N a due to the injection of electromagnetic energy at late times is governed by a Boltzmann equation of the form where H is the Hubble parameter and where C (p) a and C (s) a are the collision terms associated with two different classes of scattering processes which contribute to this overall rate of change. We shall describe these individual collision terms in detail below. Since n a is affected by Hubble expansion, it is more convenient to work with the corresponding comoving number density Y a ≡ n a /n B , where n B denotes the total number density of baryons. The Boltzmann equation for Y a then takes the form (3.13) The collision term C (p) a represents the collective contribution from Class-II processes directly involving the population of primary photons described by Eq. (3.11). The principal processes which contribute to C (p) a are photoproduction processes of the form N b +γ → N a +N c and photodisintegration processes of the form N a + γ → N b + N c , where N b and N c are other nuclei in the thermal plasma. The collision term associated with these processes takes the form where the indices b and c run over the nuclei present in the plasma, where σ a (E) respectively denote the cross-sections for the corresponding photoproduction and photodisintegration processes discussed above, and where E (ac) b and E (bc) a are the respective energy thresholds for these processes. Expressions for these cross-sections and values for the corresponding energy thresholds can be found, e.g., in Ref. [2].
By contrast, C (s) a represents the collective contribution from additional, secondary processes which involve not the primary photons themselves but rather a non-thermal population of energetic nuclei produced by interactions involving those primary photons. In principle, these secondary processes include both reactions that produce nuclei of species N a and reactions which destroy them. In practice, however, because the non-thermal population of any given species N b generated by processes involving primary photons is comparatively small, the effect of secondary processes on the populations of most nuclear species is likewise small. As we shall discuss in more detail in Sect. III C, the only exception is 6 Li, which is not produced in any significant amount during the BBN epoch but which can potentially be produced by secondary processes initiated by photon injection at subsequent times. Since these processes involve the production rather than the destruction of 6 Li, we focus on the effect of secondary processes on nuclei which appear in the final state rather than the initial state in what follows.
The energetic nuclei which participate in secondary processes are the products of the same kinds of reactions which lead to the collision term in Eq. (3.14). Thus, the kinetic-energy spectrum d n b (E b , t)/dE b of the nonthermal population of a nuclear species N b produced in this manner is in large part determined by the energy spectrum of the primary photons. In calculating this spectrum, one must in principle account for the fact that a photon of energy E γ can give rise to a range of possible E b values due to the range of possible scattering angles between the three-momentum vectors of the incoming photon and the excited nucleus in the center-of-mass frame. However, it can be shown [27] that a reasonable approximation for the collision term C (s) a for 6 Li is nevertheless obtained by taking E b to be a one-to-one function of E γ of the form [28] where E (bd) c is the energy threshold for the primary process N c + γ → N b + N d . In this approximation, is the energy-loss rate for N b due to Coulomb scattering with particles in the thermal background plasma. The exponential factor β b (E γ , t) accounts for the collective effect of additional processes which act to reduce the number of nuclei of species N b . The lower limit of integration in Eq. (3.16) is given by is the inverse of the function defined in Eq. (3.15). In other words, E −1 (E b ) is the photon energy which corresponds to a kinetic energy E b for the excited nucleus.
In principle, the processes which contribute to β a (E γ , t) include both decay processes (in the case in which N b is unstable) and photodisintegration processes of the form N b +γ → N c +N d involving a primary photon. In practice, the photodisintegration rate due to these processes is much slower that the energy-loss rate due to Coulomb scattering for any species of interest. Moreover, as we shall see in Sect. III C, the only nuclear species whose non-thermal population has a significant effect on the 6 Li abundance are tritium (T) and the helium isotope 3 He. Because these two species are mirror nuclei, the secondary processes in which they participate affect the 6 Li abundance in the same way and have almost identical cross-sections and energy thresholds. Thus, in terms of their effect on the production of 6 Li, the populations of T and 3 He may effectively be treated together as if they were the population of a single nuclear species. Although tritium is unstable and decays via beta decay to 3 He with a lifetime of τ T ≈ 5.6 × 10 8 s, these decays have no impact on the combined population of T and 3 He. We may therefore safely approximate β b (E γ , t) ≈ 0 for this combined population of excited nuclei in what follows.
The most relevant processes through which an energetic nucleus N b in this non-thermal population can alter the abundance of another nuclear species N b are scattering processes of the form N b + N f → N a + X, in which an energetic nucleus N b from the non-thermal population generated by primary processes scatters with a background nucleus N f , resulting in the production of a nucleus of species N a and some other particle X (which could be either an additional nucleus or a photon). In principle, processes of the form N b + N a → N f + X can also act to reduce the abundance of N a . However, as discussed above, this reduction has a negligible impact on Y a for any nuclear species N a which already has a sizable comoving number density at the end of BBN. Thus, we focus here on production rather than destruction when assessing the impact of secondary processes on the primordial abundances of light nuclei.
With this simplification, the collision term associated with secondary production processes takes the form where v(E b ) is the (non-relativistic) relative velocity of nuclei N b and N f in the background frame, where is the crosssection for the scattering process N b +N f → N a +X with corresponding threshold energy E (aX) bf , and where E(E C ) is the cutoff in d n b (E b , t)/dE b produced by primary processes.
C. Constraints on Primordial Light-Element Abundances
The nuclear species whose primordial abundances are the most tightly constrained by observation -and which are therefore relevant for constraining the late decays of unstable particles -are D, 4 He, 6 Li, and 7 Li. The abundance of 3 He during the present cosmological epoch has also been constrained by observation [29,30]. However, uncertainties in the contribution to this abundance from stellar sources make it difficult to translate the results of these measurements into bounds on the primordial 3 He abundance [31,32]. The effect of these uncertainties can be mitigated in part if we consider the ratio (D + 3 He)/H rather than 3 He/H, as the former is expected to be largely unaffected by stellar processing [33,34]. In this paper, we focus our attention on D, 4 He, 6 Li, and 7 Li, as the relationship between the measured abundances of these nuclei and their corresponding primordial abundances is more transparent.
The observational constraints on the primordial abundances of these nuclei can be summarized as follows. Bounds on the primordial 4 He abundance are typically phrased in terms of the primordial helium mass fraction Y p ≡ (ρ4 He /ρ B ) p , where ρ4 He is the mass density of 4 He, where ρ B is the total mass density of baryonic matter, and where the subscript p signifies that it is only the primordial contribution to ρ4 He which is used in calculating Y p , with subsequent modifications to this quantity due to stellar synthesis, etc., ignored. The 2σ limits on Y p are [35] 0.2369 < Y p < 0.2529 . (3.18) The observational 2σ limits on the 7 Li abundance are [36] 1.0 × 10 −10 < 7 Li H p < 2.2 × 10 −10 , (3.19) where the symbols 7 Li and H denote the primordial number densities of the corresponding nuclear species.
Constraining the primordial abundance of D is complicated by a mild tension which currently exists between the observational results for D/H derived from measurements of the line spectra of low-metallicity gas clouds [37] and the results obtained from numerical analysis of the Boltzmann equations for BBN [38] with input from Planck data [39], which predict a slightly lower value for this ratio. We account for these tensions by choosing our central value and lower limit on D/H in accord with the central value and 2σ lower limit from numerical calculations, while at the same time adopting the 2σ observational upper limit as our own upper limit on this ratio. Thus, we take our bounds on the D abundance to be While this upper bound is identical to the corresponding constraint quoted in Ref. [2], this is a numerical accident resulting from a higher estimate of the 6 Li/ 7 Li ratio (due to the recent detection of additional 6 Li in low-metallicity stars) and a reduction in the upper bound on the 7 Li/H ratio.
Having assessed the observational constraints on D, 4 He, 6 Li, and 7 Li, we now turn to consider the effect that the late-time injection of electromagnetic radiation has on the abundance of each of these nuclear species relative to its initial abundance at the conclusion of the BBN epoch. Of all these species, 4 He is by far the most abundant. For this reason, reactions involving 4 He nuclei in the initial state play an outsize role in the production of other nuclear species. Moreover, since the abundances of all other such species in the thermal bath are far smaller than that of 4 He, reactions involving these other nuclei in the initial state have a negligible impact on the 4 He abundance. Photodisintegration processes initiated directly by primary photons are therefore the only processes which have an appreciable effect on the primordial abundance of 4 He. A number of individual such processes contribute to the overall photodisintegration rate of 4 He, all of which have threshold energies E thresh 20 MeV.
The primordial abundance of 7 Li, like that of 4 He, evolves in response to photon injection primarily as a result of photodisintegration processes initiated directly by primary photons. At early times, when the energy ceiling E C in Eq. (3.5) for the spectrum of these photons is relatively low, the process 7 Li + γ → 4 He + T, which has a threshold energy of only ∼ 2.5 MeV, dominates the photodisintegration rate. By contrast, at later times, additional processes with higher threshold energies, such as 7 Li + γ → 6 Li + n and 7 Li + γ → 4 He + 2n + p, become relevant.
While the reactions which have a significant impact on the 4 He and 7 Li abundances all serve to reduce these abundances, the reactions which have an impact on the D abundance include both processes which create deuterium nuclei and processes which destroy them. At early times, photodisintegration processes initiated by primary photons -and in particular the process D + γ → n + p, which has a threshold energy of only E thresh ≈ 2.2 MeV -dominate and serve to deplete the initial D abundance. At later times, however, additional processes with higher energy thresholds turn on and serve to counteract this initial depletion. The dominant such process is the photoproduction process 4 He + γ → D + n + p, which has a threshold energy of E thresh ≈ 25 MeV. Unlike 4 He, 7 Li, and D, the nucleus 6 Li is not generated to any significant degree by BBN. However, a population of 6 Li nuclei can be generated after BBN as a result of photon injection at subsequent times. The most relevant processes are 7 Li + γ → 6 Li + n and the secondary production processes 4 He + 3 He → 6 Li + p and 4 He + T → 6 Li + n, where T denotes a tritium nucleus. All of these processes have energy thresholds E thresh ≈ 7 MeV. The abundances of 3 He and T, which serve as reactants in these secondary processes, are smaller at the end of BBN than the abundance of 4 He by factors of O(10 4 ) and O(10 6 ), respectively (for reviews, see, e.g., Ref. [41]). At the same time, the non-thermal populations of 3 He and T generated via the photodisintegration of 4 He are much larger than the nonthermal population of 4 He, which is generated via the photodisintegration of other, far less abundant nuclei. Thus, to a very good approximation, the reactions which contribute to the secondary production of 6 Li involve an excited 3 He or T nucleus and a "background" 4 He nucleus in thermal equilibrium with the radiation bath.
In Table I, we provide a list of the relevant reactions which can serve to alter the abundances of light nuclei as a consequence of photon injection at late times, along with their corresponding energy thresholds. Expressions for the cross-sections for these processes are given in Ref. [2]. While there exist additional nuclear processes beyond those listed in Table I 6 Li, and 7 Li in scenarios involving photon injection at late times, along with the corresponding energy threshold E thresh for each process. We note that the excited T and 3 He nuclei which participate in the secondary production of 6 Li are generated primarily by the processes listed above which contribute to the destruction of 4 He.
to the collision terms in Eq. (3.13), these processes do not have a significant impact on the Y a of any relevant nucleus when the injected energy density is small and can therefore be neglected. In should be noted that the population of excited T and 3 He nuclei which participate in the secondary production of 6 Li are generated primarily by the same processes which contribute to the destruction of 4 He. We note that we have not included processes which contribute to the destruction of 6 Li. The reason is that the collision terms for these processes are proportional to Y6 Li itself and thus only become important in the regime in which the rate of electromagnetic injection from unstable-particle decays is large. By contrast, for reasons that shall be discussed in greater detail below, we focus in what follows primarily on the regime in which injection is small and 6 Li-destruction processes are unimportant. However, we note that these processes can have an important effect on Y6 Li in the opposite regime, rendering the bound in Eq. (3.21) essentially unconstraining for sufficiently large injection rates [2].
Finally, we note that the rates and energy thresholds for 4 He + 3 He → 6 Li + p and 4 He + T → 6 Li + n are very similar, as are the rates and energy thresholds for the 4 He-destruction processes which produce the nonthermal populations of 3 He and T [2]. In what follows, we shall make the simplifying approximation that 3 He and T are "interchangeable" in the sense that we treat these rates -and hence also the non-thermal spectra d n a (E, t)/dE of 3 He and T -as identical. Thus, although T decays via beta decay to 3 He on a timescale τ T ≈ 10 8 s, we neglect the effect of the decay kinematics on the resulting non-thermal 3 He spectrum. As we shall see, these simplifying approximations do not significantly impact our results.
D. Towards an Analytic Approximation: Linearization and Decoupling
In order to assess whether a particular injection history is consistent with the constraints discussed in the previous section, we must evaluate the overall change δY a (t) ≡ Y a (t)−Y init a in the comoving number density Y a of a given nucleus at time t = t now , where Y init a denotes the initial value of Y a at the end of BBN. In principle, this involves solving a system of coupled differential equations, one for each nuclear species present in the thermal bath, each of the form given in Eq. (3.13).
In practice, however, we can obtain reasonably reliable estimates for the δY a without having to resort to a full numerical analysis. This is possible ultimately because observational constraints require |δY a | to be quite small for all relevant nuclei, as we saw in Sect. III C. The equations governing the evolution of the Y a are coupled due to feedback effects in which a change in the comoving number density of one nuclear species N a alters the reaction rates associated with the production of other nuclear species. However, if the change in Y a is sufficiently small for all relevant species, these feedback effects can be neglected and the evolution equations effectively decouple.
In order for the evolution equations for a particular nuclear species N a to decouple, the linearity criterion |δY b | Y b must be satisfied for any other nuclear species N b = N a which serves as a source for reactions that significantly affect the abundance of N a at all times after the conclusion of the BBN epoch. In principle, there are two ways in which this criterion could be enforced by the observational constraints and consistency conditions discussed in Sect. III C. The first is simply that the applicable bound on each Y b which serves as a source for N a is sufficiently stringent that this bound is always violated before the linearity criterion |δY b | Y b fails. The second possibility is that while the direct bound on Y b may not in and of itself require that |δY b | be small, the comoving number densities Y a and Y b are nevertheless directly related in such a way that the applicable bound on Y a is always violated before the linearity criterion |δY b | Y b fails. If one of these two conditions is satisfied for every species N b which serves as a source for N a , we may treat the evolution equation for N a as effectively decoupled from the equations which govern the evolution of all other nuclear species. We emphasize that N a itself need not satisfy the linearity criterion in order for its evolution equation to decouple in this way.
We now turn to examine whether and under what circumstances our criterion for the decoupling of the evolution equations is satisfied in practice for all relevant nuclear species. In Fig. 2 we illustrate the network of reactions which can have a significant effect on the values of Y a for these species. The nuclei which appear in the initial state of one of the primary production processes The nodes in the network represent different nuclear species: nodes represented by solid circles correspond to nuclei whose primordial comoving number densities Ya are reliably constrained by data, while nodes represented by dashed circles correspond to other nuclear species involved in these reactions. An arrow pointing from one node to another indicates that the nucleus associated with the node from which the arrow originates serves as a source for the nucleus associated with the node to which the arrow points. A solid arrow indicates that the corresponding reaction can potentially have a non-negligible impact on the abundance of the product nucleus, while a dashed arrow indicates that the effect of the corresponding reaction is always negligible. Note that 1 H nuclei -i.e., protons -are generated as a byproduct of many of the other reactions shown. However, the impact of these contributions to the 1 H abundance is negligible and the corresponding arrows have been omitted for clarity. A checked box superimposed on the arrows emerging from a node indicates that observational constraints enforce the linearity criterion |δYa| Ya for the corresponding nucleus. By contrast, an open box indicates that this criterion is not satisfied. The Boltzmann equation for a given nucleus Na effectively decouples, in the sense that feedback effects can be neglected in calculating Ya, whenever the linearity criterion is satisfied for all N b = Na which serve as a source for Na (though not necessarily for Na itself). Since no other nucleus serves as a source for 4 He or 7 Li, the Boltzmann equations for both of these species trivially decouple. The Boltzmann equation for D also decouples because 4 He, the one nucleus which serves as a significant source for D, is required to satisfy the linearity criterion. However, the Boltzmann equation for 6 Li does not decouple, as one of its source nuclei -in particular, 7 Li -does not satisfy the linearity criterion. Further details are described in the text. Table I are 4 He, which serves as a source for D and 6 Li, and 7 Li, which serves as a source for 6 Li. We note that while 3 He and T each appear in the initial state of one of the secondary production processes for 6 Li, it is the non-thermal population of each nucleus which plays a significant role in these reactions. Since the nonthermal populations of both 3 He and T are generated primarily as a byproduct of 4 He destruction, requiring that the linearity criterion be satisfied for 4 He and 7 Li is sufficient to ensure that the evolution equations for all relevant nuclear species decouple.
listed in
We begin by assessing whether the direct constraints on Y4 He and Y7 Li themselves are sufficient to enforce the linearity criterion. We take the initial values of these comoving number densities at the end of the BBN epoch to be those which correspond to the central observational values for Y p and 7 Li/H quoted in Ref. [42], namely Y p = 0.2449 and 7 Li/H = 1.6 × 10 −10 . Since neither 4 He nor 7 Li is produced at a significant rate by interactions involving other nuclear species, the evolution equation in Eq. (3.13) for each of these nuclei takes the form represents the rate at which Y b is depleted as a result of photodisintegration processes. This depletion rate varies in time, but depends neither on Y b nor on the comoving number density of any other nucleus. Since Y b decreases monotonically in this case, it follows that if this comoving quantity lies within the observationally-allowed range today, it must also lie within this range at all times since the end of the BBN epoch.
The bound on Y4 He which follows from Eq. (3.18) is sufficiently stringent that |δY4 He | Y4 He is indeed required at all times since the end of BBN for consistency with observation. Thus, our linearity criterion is always satisfied for 4 He. By contrast, the bound on Y7 Li in Eq. (3.19) is far weaker in the sense that |δY7 Li | need not necessarily be small in relation to Y7 Li itself. Moreover, while the contribution to δY6 Li from primary production is indeed directly related to δY7 Li , we find that the observational bound on Y6 Li is not always violated before the linearity criterion |δY7 Li | Y7 Li fails -even if we assume Y6 Li ≈ 0 at the end of BBN. The reason is that not every 7 Li nucleus destroyed by primary photodisintegration processes produces a 6 Li nucleus. Indeed, 7 Li + γ → 4 He + T and 7 Li + γ → 4 He + 2n + p contribute to the depletion of Y7 Li as well.
Since the energy threshold for 7 Li + γ → 4 He + T is lower than the threshold for 7 Li + γ → 6 Li + n, there will be a range of t inj within which injection contributes to the destruction of 7 Li without producing 6 Li at all. Furthermore, even at later injection times t inj 4.9 × 10 5 s, when the primary-photon spectrum from injection includes photons with energies above the threshold for 7 Li + γ → 6 Li + n, the 7 Li-photodisintegration rates associated with this process and the rates associated with 7 Li + γ → 4 He + T and 7 Li + γ → 4 He + 2n + p are comparable. Consequently, only around 30% of 7 Li nuclei destroyed by primary photodisintegration for t inj 4.9 × 10 5 s produce a 6 Li nucleus in the process. Thus, a large |δY7 Li | invariably results in a much smaller contribution to δY6 Li . Thus, 7 Li does not satisfy the linearity criterion when serving as a source for 6 Li.
That said, while our linearity criterion is not truly satisfied for 7 Li, we can nevertheless derive meaningful constraints on decaying particle ensembles from observational bounds on 6 Li by neglecting feedback effects on Y7 Li in calculating Y6 Li . Since the Boltzmann equation for Y7 Li takes the form given in Eq. (3.22), Y7 Li is always less than or equal to its initial value Y init 7 Li at the end of BBN. This in turn implies that the collision term C (p) 6 Li in the Boltzmann equation for 6 Li is always less than or equal to the value that it would have had if the linearity criterion for 7 Li had been satisfied. It therefore follows that the contribution to δY6 Li from primary production which we would obtain if we were to approximate Y7 Li by Y init 7 Li at all times subsequent to the end of BBN is always an overestimate. In this sense, then, the bound on electromagnetic injection which we would obtain by invoking this linear approximation for Y7 Li represents a conservative bound. Moreover, it turns out that because of the relationship between Y7 Li and Y6 Li , the bound on decaying ensembles from the destruction of 7 Li is always more stringent than the bound from the primary production of 6 Li. Thus, adopting the linear approximation for 7 Li in calculating C (p) 6 Li does not artificially exclude any region of parameter space for such ensembles once the combined constraints from all relevant nuclear species are taken into account.
Motivated by these considerations, in what follows we shall therefore adopt the linear approximation in which in calculating the collision terms C (p) a and C (s) a for any nuclear species N a = N b for which N b serves as a source. As we have seen, this approximation is valid for all species except for 7 Li, which serves as a source for primary 6 Li production. Moreover, adopting this approximation for 7 Li in calculating C (p) 6 Li yields a conservative bound on electromagnetic injection from decaying particle ensembles.
As discussed above, the advantage of working within the linear approximation is that the Boltzmann equations for all relevant N a effectively decouple and may be solved individually in order to yield analytic approximations for δY a . In the simplest case, in which the collision terms C where t 0 represents the time at the conclusion of the BBN epoch beyond which the initial abundance Y a (t 0 ) = Y init a generated by standard primordial nucleosynthesis remains essentially fixed in the absence of any subsequent injection. Moreover, even in cases in which C (p) a and C (s) a include both source and sink terms, we may still evaluate δY a in this way, provided that observational constraints restrict δY a to the region |δY a | Y a and therefore allow us to ignore feedback effects and approximate Y a ≈ Y init a as a constant on the right side of Eq. (3.13).
Since the Boltzmann equation for 6 Li contains no nonnegligible sink terms, and since observational constraints require that |δY a | Y a for both 4 He and D, it follows that δY a is well approximated by Eq. (3.23) for these species. Indeed, the only relevant nucleus which does not satisfy these criteria for direct integration is 7 Li. Nevertheless, since 7 Li is destroyed by a number of primary photodisintegration processes but not produced in any significant amount, the Boltzmann equation for this nucleus takes the particularly simple form specified in Eq. (3.22). This first-order differential equation can easily be solved for Y a , yielding an expression for the comoving number density at any time t ≥ t 0 : When this relation is expressed in terms of δY a rather than Y a , we find that it may be recast in the more revealing form In situations in which the linearity criterion |δY a (t)| Y init a is satisfied for the nucleus N a itself at all times t ≥ t 0 , Taylor expansion of the left side of this equation yields which is also the result obtained by direct integration of the Boltzmann equation for N a in the approximation that Y a ≈ Y init a . Comparing Eqs. (3.25) and (3.26), we see that if we were to neglect feedback and take Y a ≈ Y init a when evaluating δY a for a species for which this approximation is not particularly good, the naïve result that we would obtain for δY a would in fact correspond to the value of the quantity Y init Thus, given that a dictionary exists between the value of δY a obtained from Eq. (3.25) and the value obtained from Eq. (3.26), for simplicity in what follows we shall derive our analytic approximation for δY a using Eq. (3.26) and simply note that the appropriate substitution should be made for the case of 7 Li. That said, we also find that the constraint on Ω χ that we would derive from Eq. (3.26) in single-particle injection scenarios from the observational bound on Y7 Li differs from the constraint that we would derive from the more accurate approximation in Eq. (3.25) by only O(10%). Thus, results obtained by approximating δY7 Li by the expression in Eq. (3.26) are nevertheless fairly reliable in such scenarios -and indeed can be expected to be reasonably reliable in scenarios involving decaying ensembles as well.
E. Analytic Approximation: Contribution from Primary Processes
Having discussed how the Boltzmann equations for the relevant N a effectively decouple in the linear regime, we now proceed to derive a set of approximate analytic expressions for δY a from these decoupled equations. We begin by considering the contribution to δY a that arises from primary photoproduction or photodisintegration processes. The contribution from secondary processes, which is relevant only for 6 Li, will be discussed in Sect. III F.
Our ultimate goal is to derive an approximate analytic expression for the total contribution to δY a due the injection of photons from an entire ensemble of decaying states. However, our first step in this derivation shall be to consider the simpler case in which the injection is due to the decay of a single unstable particle species χ with a lifetime τ χ . We shall work within the uniformdecay approximation, in which the non-thermal photon spectrum takes the particularly simple form in Eq. (3.11). In this approximation, the lower limit of integration in Eq. (3.23) may be replaced by τ χ , while the upper limit can be taken to be any time well after photons at energies above the thresholds E (ac) b and E (bc) a for all relevant photoproduction and photodisintegration processes have thermalized. Thus, we may approximate the change in the comoving number density of each relevant nuclear species as In evaluating each δY a , we may also take advantage of the fact that the rates for the relevant reactions discussed in Sect. III C turn out to be such that the first (source) and second (sink) terms in Eq. (3.14) are never simultaneously large for any relevant nuclear species. Indeed, the closest thing to an exception occurs during a very small time interval within which the source and sink terms for D are both non-negligible. Thus, depending on the value of τ χ and its relationship to the timescales associated with these reactions, we may to a very good approximation treat the effect of injection from a single decaying particle as either producing or destroying N a .
With these approximations, the integral over t in Eq. (3.27) may be evaluated in closed form. In particular, when the source term in Eq. (3.14) dominates, we find that δY a is given by where we have defined .
By contrast, when the sink term dominates, we find that δY a is given by While the expressions for δY a in Eqs. (3.30) and (3.28) pertain to the case of a single unstable particle within the uniform-decay approximation, it is straightforward to generalize these results to more complicated scenarios. Indeed, within the linear approximation, the total change δY a in Y a which results from multiple instantaneous injections over an extended time interval is well approximated by the sum of the individual contributions from these injections. In the limit in which this set of discrete injections becomes a continuous spectrum, this sum becomes an integral over the injection time t inj . Thus, in the continuum limit, δY a is well approximated by where dY a /dt inj is the differential change in Y a due to an infinitesimal injection of energy in the form of photons at time t inj . The approximation in Eq. (3.31) allows us to account for the full exponential time-dependence of the electromagnetic injection due to particle decay in calculating δY a for any given nucleus. By extension, Eq. (3.31) gives us the ability to compare the results for δY a obtained both with and without invoking the uniform-decay approximation, thereby providing us insight into how reliably δY a can be computed with this approximation.
As an example, let us consider the effect of a single decaying particle χ with lifetime τ χ on the comoving number density of 4 He. Within the uniform-decay approximation, δY4 He is given by Eq. (3.30) because the sink term in Eq. (3.14) dominates. By contrast, when the full exponential nature of χ decay is taken into account, the corresponding result is where dρ χ (t inj )/dt inj denotes the rate of change in the energy density of χ per unit time t inj . Prior to t MRE , the energy density of an unstable particle with an extrapolated abundance Ω χ may be written is the critical density of the universe at present time and where t MRE is the time of matterradiation equality. The corresponding rate of change in the energy density, properly evaluated in the comoving frame and then transformed to the physical frame, is The corresponding expressions for continuum injection in cases in which the source term in Eq. (3.14) dominates are completely analogous and can be derived in a straightforward way.
In Fig. 3, we compare the results obtained for δY4 He within the uniform-decay approximation to the results obtained with the full exponential time-dependence of χ decay taken into account. In particular, within the (Ω χ , τ χ ) plane, we display contours of the corresponding 4 He mass fraction Y p obtained within the uniformdecay approximation (upper panel) and through the full exponential calculation (lower panel). For concreteness, in calculating these contours we have assumed an initial value Y p = 0.2449 for the 4 He mass fraction at the end of BBN, following Ref. [42].
Comparing the two panels of Fig. 3, we see that the results for Y p obtained within the uniform-decay approximation are indeed very similar throughout most of the parameter space shown to the results obtained through the full calculation. Nevertheless, we observe that discrepancies do arise. For example, we note that the constraints obtained within the uniform-decay approximation are slightly stronger for particles with lifetimes within the range 10 7 s τ χ 10 8 s and slightly weaker for τ χ above this range than the constraints obtained through the full calculation. These discrepancies are ultimately due to the fact that the reprocessed photon spectrum in Eq. (3.1) depends on the temperature of the radiation bath. Thus, there is a timescale t σ,max for which the reaction rate for a particular photoproduction or photodisintegration process is maximized for fixed dρ(t inj )/dt inj . Since all of the energy density ρ χ initially associated with the decaying particle is injected precisely at τ χ within the uniform-decay approximation, this approximation yields slightly more stringent constraints than those obtained through the full calculation when τ χ ∼ t σ,max . By the same token, the constraints obtained within the uniform-decay approximation are slightly less stringent than those obtained through the full calculation when τ χ differs significantly from t σ,max .
We also observe that for lifetimes τ χ 10 7 s, the uniform-decay approximation likewise yields constraints that are weaker than those obtained through the full exponential calculation. The reason for this is that the upper energy cutoff E C in the reprocessed photon spectrum is proportional to t 1/2 inj . For sufficiently early injection times, E C lies below the threshold energy E (bc) a for the photodisintegration reactions which contribute to δY4 He . Injection at such early times therefore has essentially no impact on Y p . This implies that within the uniform-decay approximation, a particle with a lifetime in this regime likewise has no effect on Y p . By contrast, within the full exponential calculation, injection occurs at a significant rate well after τ χ , leading to a non-negligible change in Y p even when the lifetime of the decaying particle is short.
In summary, the results shown in Fig. 3 indicate that other than in the regime where τ χ is sufficiently short that E C lies below the threshold energy for 4 He destruction, the result for δY4 He obtained within the uniform-decay approximation is very similar to the result obtained through the full exponential calculation for the same τ χ and Ω χ . We find this to be the case for the other relevant nuclei as well. Thus, having shown that the results for δY a obtained within the uniform-decay approximation accord well with those obtained through the full exponential calculation, at least for sufficiently long τ χ , we shall adopt this approximation in deriving our constraints on ensembles of electromagnetically-decaying particles. As we shall see, the advantage of working within the uniform-decay approximation is that within this approximation it is possible to write down a simple analytic expression for δY a . However, as we shall discuss in more detail below, we shall adopt an alternative strategy for approximating δY a within the regime in which τ χ is short and the results obtained within the uniform-decay approximation do not agree with those obtained through the full exponential calculation.
In order to write down our analytic expressions for δY a , we shall make one additional approximation: we shall treat the ratio of cross-sections S (bd) c (E) as not varying significantly as a function of energy between the threshold energy E (bc) a and E C . When this approximation holds, we may treat this ratio as a constant and pull it outside the integral over photon energies. In order to justify this approximation, we begin by noting that the cross-sections σ (bc) a (E) for primary processes typically peak at a value slightly above E (bc) a but fall precipitously with E beyond that point (see, e.g., Ref. [2]). In the vicinity of the peak, however, the variation of σ a (E). Moreover, because K(E, τ χ ) falls rapidly with E, the dominant contribution to the energy integral in Eq. (3.30) arises from photons just above threshold even within the approximation that S (bc) a (E) is constant. Thus, to a good approximation, we may replace S (bd) c (E) by a constant on the order of its peak value and take this quantity outside the energy integral.
Within this approximation, it is now possible to analytically evaluate the integral in Eq. (3.28). The form of the result depends on the relationship between the threshold energy E (bc) a for the scattering process and the energy scales E C (t inj ) and E X (t inj ) which determine the shape of the photon spectrum at time t inj = τ χ . There are three cases of interest: the case in which E a . Moreover, the relations in Eq. (3.5) imply that each of these cases corresponds to a specific range of τ χ . In particular, the respective lifetime regimes are τ χ < t Ca , t (bc) (3.35) Within the τ χ < t (bc) Ca regime, none of the photons produced by χ decay exceed the threshold for the photodisintegration process. The contribution to δY a within the uniform-decay approximation is therefore formally zero. As discussed above, this is the regime in which the uniform-decay approximation fails to reproduce the results obtained through the full exponential calculation. Thus, in order to derive a meaningful bound on decaying particles with lifetimes in this regime, we instead model the injection of photons using the full continuum expression in Eq. (3.32) with dρ(t inj )/dt inj given by Eq. (3.34). However, in order to arrive at a simple analytic expression for δY a , we include only the contribution from injection times in the range t (bc) Ca beyond leading order in the resulting expression. With these approximations, in this regime we find where the proportionality constant A (bc) a for each contributing reaction is independent of the properties of the decaying particle. This treatment ensures that we obtain a more reliable estimate for the contribution to δY a from particles with τ χ in this regime.
Within the remaining two lifetime regimes, the contribution to δY a within the uniform-decay approximation is non-vanishing. Thus, within these regimes, we obtain our approximation for δY a by integrating Eq. (3.28), as discussed above. For t (bc) is a proportionality constant and where β ≡ E X /E C ≈ 0.27 is the τ χ -independent ratio of the energy scales in Eq. (3.5). Likewise, for t (bc) We emphasize that the proportionality constant B Strictly speaking, Eq. (3.38) does not hold for arbitrarily large τ χ , since photons produced by extremely late decays are not efficiently reprocessed by Class-I processes into the spectrum in Eq. (3.1). In order to account for this in what follows, we shall consider each term in the sum in Eq. (3.38) to be valid only for injection times t inj < t f a is a characteristic cutoff timescale associated with the reaction. Photons injected after this cutoff timescale are assumed to have no effect on Y a . For most reactions, it is appropriate to take t (bc) f a ≈ 10 12 s, as this is the timescale beyond which certain crucial Class-I processes effectively begin to shut off and the reprocessed photon spectrum is no longer reliably described by Eq. (3.1).
F. Analytic Approximation: Contribution from Secondary Processes
The approximate analytic expressions for δY a which we have derived in Sect. III E are applicable to all of the primary photoproduction or photodisintegration processes relevant for constraining electromagnetic injection from unstable-particle decays after BBN. However, since secondary production can contribute non-negligibly to the production of 6 Li, we must derive analogous expressions for δY a in the case of secondary production as well. Moreover, since secondary production is fundamentally different from primary production in terms of particle kinematics, there is no reason to expect that these expressions should have the same functional dependence on τ χ as that exhibited by the expressions in Eqs. (3.36)- (3.38). Indeed, as we shall see, they do not.
We begin this undertaking by observing that within the uniform-decay approximation, the contribution to δY a from secondary production is given by .
(3.39)
We may simplify this expression by noting that the timedependence of the energy-loss rate b(E b , t) of excited nuclei due to Coulomb scattering is primarily due to the dilution of the number density of electrons n e (t).
Since n e (t) scales with t in the same manner as n c (t), the quantity is, to a good approximation, a comoving quantity, and hence independent of t. Thus, Υ bcf (E) may be pulled outside the time integral in Eq. (3.39). Within this approximation, the contribution to δY a reduces to In order to proceed further, we must first assess the dependence of the quantities S bcf (E b ) on the respective energy scales E γ and E b . However, we find that Υ (aX) bcf (E b ) varies reasonably slowly over the relevant range of E b [2,28]. Thus, to a good approximation, this quantity may also be pulled outside the integral over E b .
By contrast, we find that S (ab) c (E γ ) cannot reliably be approximated as a constant over the range of E γ relevant for secondary production. This deserves further comment -especially because we have approximated S (ab) c (E γ ) as a constant in deriving the expressions in Eqs. (3.36)-(3.38) for primary production. As we shall now make clear, there are important differences between the kinematics of primary and secondary production which enable us to approximate S (ab) c (E γ ) as independent of E γ in the former case but not in the latter.
For primary production, as discussed in Sect. III E, the rapid decrease of K(E γ , τ χ ) with E γ suppresses the contribution to δY a from photons with E γ E (ac) b . The dominant contribution to δY a therefore comes from a narrow region of the spectrum just above threshold within which S have little collective impact on δY a . Thus, in approximating the overall contribution to δY a from primary production, it is reasonable to treat S (ac) b (E γ ) as a constant. By contrast, for secondary production -or at least for the secondary production of 6 Li, the one nuclear species in our analysis for which secondary production can have a significant impact on δY a -photons with E γ well above the threshold energy E (bd) c for any relevant primary process play a more important role. One reason for this is that the energy threshold E (aX) bc for each secondary processes which contributes meaningfully to 6 Li production corresponds to a primary-photon energy ] well above the associated primary-process threshold E (bd) c . Indeed, the kinetic-energy thresholds for 4 He + 3 He → 6 Li + p and 4 He + T → 6 Li + n given in Table I (E γ ). Thus, it is really the energy threshold for the secondary process which sets the minimum value of E γ relevant for the secondary production of 6 Li. Since S (bd) c (E γ ) varies more rapidly with E γ at these energies than it does around its peak value, it follows that this variation cannot be neglected in determining the overall dependence of δY a on τ χ in this case.
There is, however, another reason why the variation of S (bd) c (E γ ) with E γ cannot be neglected in the case of secondary production -a deeper reason which is rooted more in fundamental differences between primary and secondary production than in the values of the particular energy thresholds associated with processes pertaining to 6 Li. The overall contribution to δY a from primary production in Eq. (3.28) involves a single integral over E γ . Thus, for primary production, the fall-off in K(E γ , τ χ ) itself with E γ is sufficient to suppress the partial contribution to δY a from photons with E γ E (bd) c . By contrast, the overall contribution to δY a from secondary production involves integration not only over E γ but also over E b . In this case, the fall-off in K(E γ , τ χ ) with E γ is not sufficient to suppress the partial contribution to δY a from photons with energies well above threshold. Thus, for any secondary production process, an accurate estimate for δY a can only be obtained when the variation of S (bd) c (E γ ) with E γ -a variation which can be quite significant at such energies -is taken into account.
We must therefore explicitly incorporate the functional dependence of S (bd) c (E γ ) on E γ into our calculation of δY a . We recall that S (bd) c (E γ ), as we have defined it in Eq. (3.29), represents the ratio of the cross-section σ (bd) c (E γ ) for the primary process N c + γ → N b + N d which produces the population of excited nuclei to the cross-section σ th (E γ ) for the Class-III processes which serve to thermalize the primary-photon spectrum. The cross-sections for the two primary processes 4 He + γ → 3 He + n and 4 He + γ → T + p relevant for secondary 6 Li production, expressed as a functions of the photon energy E γ , both take the form [2] σ (bd) c,0 ≈ 20.6 mb and σ (bd) c,0 ≈ 19.8 mb for these two processes, respectively. By contrast, σ th (E γ ) includes two individual contributions. The first contribution is due to e + e − pair-production off nuclei via Bethe-Heitler processes of the form γ + N a → X + e + + e − , where N a denotes a background nucleus and X denotes some hadronic final state. The cross-section for this process is where α is the fine-structure constant, where m e is the electron mass, and where σ T ≈ 661 mb is the Thomson cross-section. The second contribution to σ th (E γ ) is due to Compton scattering, for which the cross-section is Given the dependence of these cross-sections on E γ , we find that over the photon-energy range E (bd) c < E γ 1 GeV, the ratio S (bd) c (E γ ) for each relevant process is well approximated by a simple function of the form c,0 is a constant. In Fig. 4 we show a comparison between the exact value of S (bd) c (E γ ) for the individual process 4 He + γ → T + p and our approximation in Eq. (3.45) over this same range of E γ . We see from this figure that our approximation indeed provides a good fit to S (bd) c (E γ ) over most of this range. Moreover, while the discrepancy between these two functions becomes more pronounced as E γ ∼ O(1 GeV), we emphasize that S ] for the production of an excited T nucleus above the kineticenergy threshold for the secondary process 4 He+T → 6 Li+n. Thus, photons in the gray shaded region play no role in the secondary production of 6 Li. We note that the corresponding S (bd) c (Eγ) curves and thresholds for 4 He + γ → 3 He + n (the other primary process relevant for secondary 6 Li production) are nearly identical to those shown here. δY a , any discrepancy between Eq. (3.45) and the exact expression for S (bd) c at such energies may safely be ignored.
Armed with our approximation for S (bd) c (E γ ) in Eq. (3.45), it is now straightforward to evaluate the integral in Eq. (3.41) and obtain an approximate analytic expression for δY a . Just as it does for primary photoproduction and photodisintegration, the functional dependence of δY a on τ χ for secondary production depends on the relationship between τ χ and a pair of characteristic timescales determined by the energy thresholds for the relevant processes. As we have discussed above, it is ] rather than E (bd) c which sets the minimum E γ required of a photon in order for it to contribute to the secondary production of 6 Li. It therefore follows that the characteristic timescales for the secondary production of this nucleus are determined by . Thus, for secondary production, we define within the uniform-decay approximation, δY a is formally zero, as it is in the corresponding lifetime regime for primary production. Thus, in order to derive a meaningful constraint on unstable particles with lifetimes within this regime, we follow the same procedure as we employed in order to calculate the contribution to δY a from primary production for a particle with τ χ < t (bc) Ca . We model the injection of photons using the expression for secondary production appropriate for continuum injection, with dρ(t inj )/dt inj given by Eq. (3.34). In analogy to our treatment of the primary process in this regime, we include only the contribution from injection times in the range t Dropping terms beyond leading order in the dimensionless variable where the proportionality constant A (bcf ) a for each combination of primary and secondary processes is independent of τ χ . We note that the dependence of δY a on τ χ in this expression is exactly the same as in Eq. (3.36). By contrast, for t represents the ratio of the photon-energy threshold for primary production to the minimum photon energy needed to produce an excited nucleus of species N b with a kinetic energy above the energy threshold for the secondary process. Numerically, we find θ ≈ 0.46 for 4 He + γ → T + p followed by 4 He + T → 6 Li + n, while θ ≈ 0.52 for 4 He + γ → 3 He + n followed by 4 He + 3 He → 6 Li + p. Finally, for t (bcf ) Xa < τ χ , we find Thus, to summarize, we have derived a set of simple, analytic approximations for the change δY a in the abundance of a given nuclear species N a due to the late injection of photons by a decaying particle χ. For primary photoproduction or photodisintegration processes, we find that δY a is given by Eq. (3.36), Eq. (3.37), or Eq. (3.38), depending on the lifetime τ χ of the particle. Likewise, for secondary production, we find that δY a is given by Eq.
G. The Fruits of Linearization: Light-Element Constraints on Ensembles of Unstable Particles
We now turn to the task of extending these results to the case of an ensemble of decaying particles χ i with lifetimes τ i and extrapolated abundances Ω i . Indeed, we have seen that if the linearity criterion is satisfied both for N a itself and for all of its source nuclei N b = N a , all feedback effects on Y a can be neglected. Thus, in this regime, the overall change δY a in the abundance of a light nucleus is well approximated by the sum of the individual contributions associated with the individual χ i from each pertinent process. Indeed, it is only because we have entered a linear regime that such a direct sum is now appropriate. These, then, are the fruits of linearization.
While certain nuclei in our analysis -namely 6 Li and 7 Li -do not have the property that the linearity criterion is always satisfied for both the nucleus itself and its source nuclei, we emphasize that we are nevertheless able to derive meaningful bounds on the comoving number densities of these nuclei. As noted in Sect. III D, artificially adopting the linearity criterion for Y7 Li in the Boltzmann equation for 6 Li always yields a conservative bound on δY6 Li , regardless of the injection history. For 7 Li, the issue is that feedback effects on the photodisintegration rate due to changes in the comoving number density of 7 Li itself are not necessarily small. Thus, strictly speaking, δY7 Li is not well approximated by a direct sum of the contributions from the individual χ i . However, since this direct sum is simply an approximation of the integral in Eq. (3.25), it is equivalent to the quantity Y init 7 Li ln[1+δY7 Li (t)/Y init 7 Li ]. Thus, for all relevant N a , we find that an ensemble of decaying particles makes an overall contribution to δY a -or, in the case of 7 Li, to the quantity Y init a ln[1 + δY a (t)/Y init a ] -which can be approximated as a direct sum of the individual contributions from the individual ensemble constituents. Indeed, this yields either a reliable estimate for the true value of δY a or a reliably conservative bound on δY a .
In principle, each of these individual contributions to a given δY a can involve a large number of reactions with different energy thresholds and scattering kinematics, each with its own distinct fit parameters A (bc) Ca , etc. In practice, however, the number of reactions which contribute significantly to δY a for any one of the four relevant nuclear species is quite small, as can be seen from Table I. Moreover, it is often the case that many if not all of the reactions which have a nonnegligible impact on a given δY a have very similar energy thresholds and scattering kinematics. When this is the case, the contribution to δY a from these processes can, to a very good approximation, collectively be modeled using a single set of fit parameters. For example, the only reactions which have a significant impact on δY4 He are the primary photodisintegration processes 4 He + γ → T + p and 4 He + γ → 3 He + n, which have very similar energy thresholds and scattering kinematics. Thus, a fit involving a single set of parameters yields an accurate approximation for δY4 He . The reactions which contribute to δY7 Li differ more significantly in terms of their energy thresholds and scattering kinematics. Nevertheless, as we shall see, the collective effect of these processes is also well modeled by a single set of fit parameters. By contrast, for D there are two processes with qualitatively different energy thresholds and scattering kinematics which must be modeled using separate sets of fit parameters: primary photodisintegration via D + γ → n + p and primary photoproduction via 4 He + γ → D + n + p. Likewise, for 6 Li, two sets of fit parameters are required: one for primary photoproduction via 7 Li + γ → 6 Li + n and one for secondary production via both 4 He + 3 He → 6 Li + p and 4 He + T → 6 Li + n.
In general, then, we require at most two distinct sets of parameters in order to model the overall contribution to δY a for each relevant nucleus due to electromagnetic injection from an ensemble of decaying particles. Thus, in general, we may express this overall contribution as the sum of two terms (3.51) For each relevant nuclear species, one of these terms is associated with a primary process: primary photoproduction in the case of 6 Li and primary photodisintegration in the case of 4 He, 7 Li, and D. Thus, δY where A a , B a , t Ca , t Xa , etc., are model parameters whose assignments we shall discuss below. By contrast, the form of the second term in Eq. (3.51) differs depending on the nucleus in question. For 4 He and 7 Li, we simply have δY (2) a = 0. For D, this term is associated with primary production and thus takes exactly the same functional form as δY (1) a . In other words, we have where A * a , B * a , t * Ca , t * Xa , etc., represent an additional set of model parameters distinct from the parameters A a , B a , t Ca , t Xa , etc., in Eq. (3.52). For 6 Li, the term δY (2) a is associated with secondary production and therefore takes the form We note that for simplicity and compactness of notation, we have implicitly taken i = 1 for each χ i in formulating the expressions appearing in Eqs. (3.52)-(3.54). However, it is straightforward to generalize these results to the case in which i differs from unity for one or more of the χ i . In particular, the corresponding expressions for δY (1) a and δY (2) a in this case may be obtained by replacing Ω i for each species with the product Ω i i . Moreover, we remind the reader that for the case of 7 Li, a more accurate estimate of δY a can be obtained by replacing δY . We now determine the values of the parameters in Eq. (3.52). We begin by noting that the parameter t Aa for each relevant nucleus N a may be assigned essentially any value between the end of the BBN epoch and the timescale at which the first reaction which contributes to δY a becomes efficient. Thus, for simplicity, we choose a universal value t Aa = 10 4 s for all N a , which corresponds to the timescale at which the first reaction in Table I namely the photodisintegration process D + γ → n + p -effectively turns on. For t f a , we likewise choose a universal value t f a = 10 12 s for all N a . As discussed in Sect. VI, the reason for this choice is that many of the Class-I processes which serve to establish the reprocessed photon spectrum in Eq. (3.1) start to become inefficient around this timescale. As a result, the photon spectrum that results from electromagnetic injection at times t inj 10 12 s differs from that in Eq. (3.1), and the analytic approximation for δY Given these results, we now determine the values of the remaining parameters in Eqs. (3.52)-(3.54) by demanding that the constraint contour we obtain from each of these equations be consistent with the contour obtained from a more complete numerical analysis of the Boltzmann system in the limiting case in which our decaying "ensemble" comprises only a single particle species χ. In this single-particle case, our analytic expression for δY a becomes a function of only two variables: the abundance Ω χ and lifetime τ χ of χ. We determine the constraint contours corresponding to the observational limits quoted in Sect. III C by surveying (Ω χ ,τ χ ) space and numerically solving the full, coupled system of Boltzmann equations for δY a at each point. The undetermined parameters in Eqs. (3.52)-(3.54) are then chosen such that our analytic expression provides a good fit to the corresponding constraint contour for each nucleus. We list the values for all parameters obtained in this manner in Table II. In deriving these constraint contours, it is necessary to translate the observational constraints in Eqs. on the corresponding subsequent change δY a in each Y a . In so doing, we must specify a set of initial values of Y a at the end of BBN. For Y4 He and Y7 Li , we take Y p = 0.2449 and 7 Li/H = 1.6 × 10 −10 , as discussed in Sect. III D. For Y D , we take the value which corresponds to the central theoretical prediction D/H = 2.413 × 10 −5 for the primordial D/H ratio derived in Ref. [37]. Finally, since 6 Li is not produced in any significant amount during BBN, we take Y6 Li = 0. The values for the observational upper and lower limits δY min a and δY max a on δY a which correspond to these choices for the initial Y a are listed in Table II alongside the values for our fit parameters.
In Fig. 5, we display contours showing the results of our numerical calculation of δY a in (Ω χ , τ χ ) space for each relevant nucleus. More specifically, we display contours of Y p for 4 He, whereas we display contours of log 10 Y a for D, 7 Li, and 6 Li. Moreover, for purposes of illustration, we display separate contours for D production and destruction -i.e., contours evaluated considering either production or destruction processes in C (p) D only, with the cross-sections for the opposite set of processes artificially set to zero. Likewise, we also display separate contours for primary and secondary 6 Li production. We emphasize, however, that in assessing our overall constraints on decaying ensembles, we consider these processes together, allowing for the possibility of a cancellation between In all panels except that pertaining to 4 He destruction, we plot contours of log 10 Ya, while for 4 He destruction we plot contours of Yp. Note that for later reference, we have plotted separate contours for D production and destruction; likewise, we show separate contours for primary and secondary 6 Li production. Regions of the (τχ, Ωχ) plane which are found by numerical analysis to be excluded are those above and/or to the right of the gray dashed line in each panel, while the solid colored curve in each panel demarcates the analogous region found to be excluded by fitting the free parameters in our analytic approximations for δYa in Eqs. (3.52)-(3.54) to our numerical results. We see that the full numerical results and our analytic approximations agree very well in all cases except that of primary 6 Li production, where our analytic approximation diverges from the numerical result at large Ωχ and thus provides an even more conservative bound on the allowed parameter space.
the two individual contributions to δY a (in the case of D) or a reinforcement (in the case of 6 Li). The gray dashed line in each panel of Fig. 5 indicates the upper bound on Ω χ obtained through a numerical calculation of δY a which follows from the observational limits in Eqs. (3.18)-(3.21). By contrast, the solid curve in each panel represents the corresponding constraint contour obtained through our analytic approximation for δY a in Eq. (3.51). We see from Fig. 5 that in all cases except for that of primary 6 Li production, the constraint contours we obtain using the linear and uniform-decay approximations are quite similar to those obtained in the more realistic case in which we account for the fact that χ has a constant decay rate and in which we account for feedback effects in the collision terms C for each relevant nucleus. Moreover, as discussed in Sect. III D, we see that for primary 6 Li production the constraint contour obtained using these approximations indeed represents a conservative bound on Ω χ . We also note that with this sole exception, the results of our analytic approximations are in good agreement with the corresponding results in Ref. [2] within the regime in which Ω χ is small and the linear approximation holds. Once again, we note that the effects of additional processes which we do not incorporate into this analysis can have important effects on the Y a in the regime in which Ω χ -and therefore the overall magnitude of electromagnetic injection -is large. These include, for example, photodisintegration processes which contribute to the depletion of a previously generated population of 6 Li as well as additional processes which contribute to the production or destruction of D [2]. However, since the regions of (Ω χ , τ χ ) space in which these effects on D and 6 Li are relevant are excluded by constraints on the primordial abundances of other nuclei, our neglecting these processes will have essentially no impact on the overall constraints we derive on ensembles of decaying particles.
In Fig. 6 we plot our analytic approximations to the constraint contours obtained for each relevant nucleus together in (Ω χ , τ χ ) space. Note that the 6 Li contour includes both primary and secondary contributions to δY6 Li . In determining the D contour, we have included contributions from both photoproduction and photodisintegration processes, the interference between which is responsible for the "funnel" region at τ χ ≈ 10 6.5 s within which the bound on Ω χ is considerably weakened. However, since the upper and lower bounds on δY D derive from different considerations, different colors are used to distinguish the part of the contour which corresponds to the upper bound (dark blue) from the part which corresponds to the lower bound (light blue). We note that in all cases, these contours are in good agreement with the corresponding results in Ref. [2] within the 7 Li destruction (green), 4 He destruction (magenta), and 6 Li production (red), the latter including both primary and secondary contributions. In addition, we also plot constraints stemming from limits on potential distortions of the CMB-photon spectrum, both µ-type (orange) and ytype (black), as will be discussed in Sect. IV. In each case, the regions of parameter space above each contour are excluded.
regime in which Ω χ is small and the linear approximation holds.
IV. SPECTRAL DISTORTIONS IN THE CMB
In addition to altering the abundances of light nuclei, electromagnetic injection from the decays of unstable particles in the early universe can also have observable effects on the spectrum of CMB photons, distorting that spectrum away from a pure blackbody profile. Observational limits on these distortions therefore also constrain such decays. Once again, while the constraints on a single decaying particle species are well known, the corresponding constraints on an ensemble of decaying particles are less well established. Our aim in this section is to derive approximate analytic expressions for the CMB distortions which can arise due to entire ensembles of decaying particles -expressions analogous to those we derived in Sect. III for the change δY a in the comoving number density in each relevant nucleus after BBN.
One of the subtleties which arises in constraining distortions in the CMB-photon spectrum from ensembles of decaying particles with a broad range of lifetimes is that both the manner in which that spectrum is distorted and the magnitude of that distortion depend on the timescale over which the injection takes place. At early times, energetic photons produced by these decays are rapidly brought into both kinetic and thermal equilibrium by the Class-III processes discussed in Sect. II. As a result, the spectrum of photons retains a blackbody shape. However, at later times, photon-numberchanging processes such as double-Compton scattering and bremsstrahlung "freeze out," in the sense that the rates for these processes drop below the expansion rate H of the universe. Once this occurs, injected photons are no longer able to attain thermal equilibrium with photons in the radiation bath. Nevertheless, they are able to attain kinetic equilibrium with these photons through elastic Compton scattering. As a result, the photon spectrum no longer retains its original blackbody shape; rather, it is distorted into a Bose-Einstein distribution characterized by a pseudo-degeneracy parameter µ.
Eventually, at an even later time t EC ≈ 9×10 9 s, elastic Compton scattering also freezes out. Thus, photons injected at times t t EC attain neither kinetic nor thermal equilibrium with photons in the radiation bath. Nevertheless, the injected photons continue to interact with electrons in the radiation bath at an appreciable rate until the time of last scattering t LS ≈ 1.19 × 10 13 s. During this window, photons in the radiation bath are up-scattered by electrons which acquire significant energy from these interactions. The ultimate result is a photon spectrum which is suppressed at low frequencies and enhanced at high frequencies. This spectral distortion is characterized by a non-zero Compton y parameter y C . Finally, after last scattering, any additional photons injected from particle decay simply contribute to the diffuse photon background.
In order to derive analytic expressions for the constraints on the parameters µ and y C in the presence of a decaying ensemble, we shall once again make use of the uniform-decay approximation as well as a linear approximation similar to the approximation we invoked in calculating the δY a in Sect. III. The validity of such an approximation follows from the fact that both µ and y C are tightly constrained by observation. The most stringent limits on these parameters, which are currently those from COBE/FIRAS data [43], are |µ| < 9.0 × 10 −5 y C < 1.2 × 10 −5 . (4.1) Moreover, proposed broad-spectrum experiments, such as PIXIE [44], could potentially increase the sensitivity to which these distortions can be probed by more than an order of magnitude. These stringent constraints on the shape of the CMB-photon spectrum imply that the contribution δρ γ to the photon energy density from injection during the relevant epoch must be quite small, in the sense that δρ γ ρ γ . Thus, to a very good approximation, we may ignore feedback effects on the overall photon number density n γ and energy density ρ γ when analyzing distortions in that spectrum. This allows us to replace these quantities by their equilibrium values.
This linear approximation significantly simplifies the analysis of the resulting CMB spectrum [5,6,45,46]. For example, in this approximation we may study the evolution of the CMB-photon spectrum using the method of Green's functions [20,47]. In this approach, we may express a generic spectral distortion which constitutes a shift ∆I ν (t now ) in the present-day intensity of photons with frequency ν relative to the expected intensity in the standard cosmology in the form where Q is the energy density injected at time t and where G(ν, t, t ) is a Green's function which relates the intensity of photons with frequency ν injected at time t to the photon intensity at that same frequency at a later time t > t. In general, G(ν, t, t now ) is well approximated by a sum of terms representing an overall temperature shift, a µ-type distortion, and y-type distortion [20]. This Green's-function formalism provides a tool for numerically estimating the overall distortion to the CMB-photon spectrum from any arbitrary injection history, provided that the overall injected energy is small. We shall draw on this formalism extensively in deriving our analytic approximations for δµ and δy C .
A. The µ Parameter
The time-evolution of the pseudo-degeneracy parameter µ is governed by an equation of the form [5,6] 3) The first term in Eq. (4.3) is a source term which arises due to the injection of energetic photons. Within the linear approximation, this source term takes the generic form where n γ and ρ γ respectively denote the overall number density and energy density of photons. We emphasize that this expression is completely general and applies even in absence of any source of photon injection, in which case the two terms in Eq. (4.4) cancel. The second term in Eq. (4.3) is a sink term associated with processes which serve to bring the population of photons in the radiation bath into thermal equilibrium and thereby erase µ. As discussed above, the dominant such processes are double-Compton scattering and bremsstrahlung, the respective thermalization rates for which are well approximated by power-law expressions of the form [5] where t MRE once again denotes the time of matterradiation equality. The coefficients in this expression are given by where h is the Hubble constant, Ω B is the present-day abundance of baryons, and T now ≈ 2.7 K is the presentday temperature of the CMB. Up to this point, our expressions governing the evolution of µ are completely general and may be applied to any injection history, provided that the total energy density injected is sufficiently small that the linear approximation is valid. In order to derive an approximate analytic expression for the overall change δµ in µ due to an ensemble of unstable particles, we proceed in essentially the same manner as we did in deriving expressions for the changes δY a in the comoving number densities of light nuclei in Sect. III. We begin by deriving an expression for δµ in the presence of a single decaying particle species χ with mass m χ and lifetime τ χ . When we take into account the full exponential nature of the decay, we find where N (γ) χ is the average number of photons produced per decay and where χ is the fraction of the energy released by χ decays which is transferred to photons. By contrast, within the uniform-decay approximation in which dρ χ /dt is given by Eq. (3.10), we have In this approximation, a simple, analytic expression for δµ may be obtained by directly integrating Eq. (4.3) from the time t e ≈ 1.69 × 10 3 s at which electron-positron annihilation freezes out to the time t EC at which elastic Compton scattering freezes out. In particular, we find that a particle species with a lifetime τ χ anywhere within this range gives rise to a µ-type distortion of size where m p is the proton mass, where Ω B is the total present-day abundance of baryons, where η ≡ n B /n γ ≈ 6.2 × 10 −10 is the baryon-to-photon ratio of the universe, and where Ω χ once again refers to the extrapolated abundance of χ. For τ χ outside this range, δµ ≈ 0. Since the bremsstrahlung term in the exponential in Eq. (4.9) is generically negligible in comparison with the double-Compton-scattering term, we may safely neglect it in what follows. Moreover, for m χ and τ χ within our regimes of interest, the second term in the prefactor in Eq. (4.9) is subleading and can likewise be neglected. Thus, we find that δµ is well approximated by This latter quantity represents the timescale at which the damping due to double-Compton scattering becomes inefficient and contributions to δµ become effectively unsuppressed.
The dependence of δµ on τ χ has a straightforward physical interpretation. The exponential suppression factor reflects the washout of µ by double-Compton scattering at early times τ χ t µ . The additional polynomial factor reflects the fact that the equilibrium photon density grows more rapidly with time than does the matter density at earlier times. This implies that particles decaying at an earlier time yield smaller perturbations to the photon spectrum.
B. The Compton y Parameter
As discussed above, elastic Compton scattering serves to maintain kinetic equilibrium among photons in the radiation bath after double-Compton scattering and bremsstrahlung freeze out. However, elastic Compton scattering eventually freezes out as well, at time t ∼ t EC . An injection of photons which occurs after t EC but before the time t LS of last scattering contributes to the development of a non-vanishing Compton y parameter y C . Within the linear approximation, a reliable estimate for the rate of change of y C may be obtained from the relation [48] where dρ(t inj )/dt inj denotes the rate at which energy density is injected into the photon bath. We note that in contrast with the corresponding evolution equation for µ in Eq. (4.3), this equation contains no damping term. In deriving our analytic approximation for the change δy C from an ensemble of unstable particle species, we once again begin by considering the case of a single such species χ. When we take into account the full exponential nature of the decay, we obtain The corresponding result within the uniform-decay approximation is In integrating this latter expression over t in order to obtain our approximate expression for δy C , we must account for the fact that the epoch during which injection contributes to y C straddles the time t MRE of matter-radiation equality. Using the appropriate timetemperature relations 4.15) before and after the transition to matter-domination, where T MRE is the temperature at matter-radiation equality, we find (4.16) where η once again denotes the baryon-to-photon ratio of the universe.
C. The Fruits of Linearization: CMB-Distortion Constraints on Ensembles of Unstable Particles
Having derived analytic approximations for the spectral distortions δµ and δy C due to a single decaying particle species within the uniform-decay approximation, we now proceed to generalize these results to the case of an arbitrary injection history.
Within the linear approximation, the overall contribution to δµ from a decaying ensemble is simply the sum of the individual contributions from the constituent particles χ i with lifetimes in the range t e τ i t EC . Likewise, the overall contribution to y C is a sum of individual contributions from χ i with t EC τ i t LS . However, injection from constituents with lifetimes τ i ∼ t EC gives rise to intermediate-type distortions which are distinct from the purely µ-type or purely y-type distortions discussed above. While these intermediatetype distortions are not merely a superposition of µ-type and y-type distortions [46], isolating and constraining such intermediate-type distortions would require a more precise measurement of the CMB-photon spectrum than COBE/FIRAS data currently provide. We therefore find that -at least for the time being -the net effect of injection at times t ∼ t EC may indeed be modeled through such a superposition, with δµ and δy C given by modified versions of Eqs. (4.10) and (4.16). The appropriate modifications of these expressions can be inferred from the forms of the Green's functions in Ref. [20]. In particular, we model the overall contributions to δµ from the decaying ensemble by an expression of the form 1 − e −(t2µ/τi) 4αµ/3 .
(4.17)
Distortion t 0a (s) t Ba (s) t 1a (s) αa Aa Ba µ-type 3.3 × 10 6 2.4 × 10 9 1.2 × 10 10 1.0 × 10 0 5.4 × 10 −3 3.1 × 10 −1 yc-type 3.3 × 10 6 6.8 × 10 7 5.4 × 10 9 1.2 × 10 0 3.7 × 10 −5 2.7 × 10 −1 Likewise, for δy C we have In these equations we have introduced two sets of parameters A µ , B µ , α µ , t 0µ , etc., and A y , B y , α y , t 1y , etc., in order to characterize δµ and δy C , respectively. We note, however, that the timescales t 2µ ≡ t 1y are not independent model parameters but merely shorthand notation for quantities which are completely determined by t 1µ and t 1y , respectively. We also note that this parametrization accounts not only for µ-type distortions at times t t EC and y-type distortions at times t t EC , but also for subdominant contributions to δy C prior to t EC and to δµ after t EC . Finally, we note that in formulating Eqs. (4.17) and (4.18) we have once again implicitly taken i = 1 for all χ i , just as we did in Eqs. (3.52)-(3.54). The corresponding expressions for δµ and δy C may nevertheless be obtained for the case in which i differs from unity for one or more of the χ i by once again replacing Ω i for each species with the product Ω i i .
We now determine the values of the parameters in Eqs. (4.17) and (4.18) in a manner similar to that in which we determined the values of the parameters in Eqs. (3.52)-(3.54). Specifically, for each nucleus we demand that our analytic approximations yield constraint contours consistent with those obtained from numerical analysis in the case in which the ensemble comprises a single decaying particle with lifetime τ χ and abundance Ω χ . In particular, we obtain arrays of δµ and δy C values by surveying (Ω χ , τ χ ) space and computing these distortions numerically at each point using the Green'sfunction formalism of Refs. [20,47]. The undetermined parameters in Eqs. (4.17) and (4.18) are then chosen such that our analytic expressions for δµ and δy C provide good fits to the corresponding constraint contours. Our best-fit values for all of these parameters are quoted in Table III. In Fig. 7, we display contours showing the results of our numerical calculations for δµ (left panel) and δy C (right panel) within the (Ω χ , τ χ ) plane. The gray dashed line in each panel demarcates the upper boundary of the region within the (Ω χ , τ χ ) plane which is consistent with the limits on these distortions quoted in Eq. (4.1). By contrast, the solid curve in each panel represents the corresponding constraint contour obtained through our analytic approximation in Eqs. (4.17) and (4.18).
As we see, the results obtained through our analytic approximation represent a good fit to our numerical results for both δµ and δy C . The constraint contours obtained from our analytic approximations for both µ-and y-type distortions are also included in Fig. 6 in order to facilitate comparison with the constraint contours associated with the primordial abundances of light nuclei. We note that the constraint on δµ is more stringent than the constraint on y C for τ χ t EC , as expected, while the opposite is true for τ χ t EC . We also see that both of these constraints are subdominant in comparison with the constraints on the primordial abundances of light nuclei, except at late times τ χ 10 10 s.
V. MODIFICATIONS TO THE IONIZATION HISTORY OF THE UNIVERSE
As we have seen in Sect. IV, electromagnetic injection from late-decaying particles can give rise to certain distortions in the CMB-photon spectrum. However, such injections can affect the CMB in other ways as well. In particular, when such injection occurs during recombination, the resulting photons and electrons contribute to the heating and ionization of neutral hydrogen and helium as they cool, thereby expanding the surface of last scattering. This in turn leads to alterations in the pattern of CMB anisotropies, including a damping of correlations between temperature fluctuations as well as an enhancement of correlations between polarization fluctuations at low multipole moments [7]. These considerations therefore place additional constraints on decaying particle ensembles.
In a nutshell, these effects arise because the rates for many of the Class-I processes discussed in Sect. II are still sizable at times t ∼ t LS . These processes therefore continue to rapidly redistribute the energies of photons injected during the recombination epoch to lower energies, thereby allowing for efficient photoionization of neutral hydrogen and helium at a rate that exceeds the rate of cosmic expansion. However, there is only a somewhat narrow time window around t LS during which an injection of photons can have a significant impact on the surface of last scattering. At times t t LS , any additional ionization of particles in the plasma is effectively washed out. By contrast, at times t t LS , the relevant Class-I processes effectively shut off and photoionization is suppressed.
To a good approximation, then, we may regard any modification of the ionization history of the universe as being due to injection at t ∼ t LS . Thus, the crucial quantity which is constrained is the overall injection rate of energy density in the form of photons or other electromagnetically-interacting particles at t LS . Since the universe is already matter-dominated by the recombination epoch, the injection rate associated with a single decaying particle species χ with lifetime τ χ and extrapolated abundance Ω χ at t LS is given by 1) where χ once again denotes the fraction of the energy density released by χ decays which is transferred to photons. Thus, we model the constraint on such a decaying particle from its effects on the ionization history of the universe by an inequality of the form where t IH and Γ IH are taken to be free parameters. We next determine the values of Γ IH and t IH by fitting these parameters to the constraint contours obtained in Refs. [10,49] from a numerical analysis of the injection and deposition dynamics. The constraint contours obtained from subsequent numerical studies by other authors [50] are quite similar. In particular, we find Note that recent results from the final Planck data release [51] will affect the numerical values for Γ IH and t IH slightly but not dramatically. In Fig. 8, we compare the results of our analytic approximation with the numerical results of Refs. [10,49]. We observe that our approximation for the constraint contour is in good agreement with these numerical results for τ χ 10 15 s and deviates significantly from this contour only when τ χ ∼ t LS . However, as discussed in Ref. [10], the results to which we perform our fit yield an artificially conservative bound on Ω χ for lifetimes τ χ ∼ t LS , as the production of additional Lyman-α photons due to excitations from the ionizing particles was intentionally neglected. Other analyses of the ionization history [52] which include this effect yield constraint contours which more closely resemble that obtained from our analytic approximation.
The constraint in Eq. (5.2) may be generalized to an ensemble of decaying particles in a straightforward manner. Indeed, since this constraint is ultimately a bound on the overall injection rate at t ∼ t LS , the corresponding constraint on an ensemble of decaying particles χ i takes the form FIG. 8: Constraints on a decaying particle with lifetime τχ and extrapolated abundance Ωχ from modifications of the ionization history of the universe. These constraints correspond to the case in which χ = 1 and the entirety of the energy density released by χ decays is transferred to photons. The gray dashed contour represents the bound obtained in Refs. [10,49] from numerical analysis. The brown shaded region above the contour is excluded. The solid curve represents the constraint contour obtained by fitting the corresponding analytic approximation in Eq. (5.2) to these results. Dashed vertical lines indicating the time tLS of last scattering and the present age of the universe tnow have also been included for reference. We see that our approximation for the constraint contour is in good agreement with these numerical results for τχ 10 15 s and deviates significantly from this contour only when τχ ∼ tLS.
with values of Γ IH and t IH once again given in Eq. (5.3).
We emphasize that the constraint in Eq. (5.4) stems solely from the effect of electromagnetic injection from the decays of the ensemble constituents on the surface of last scattering. Since this effect arises primarily from injection at t ∼ t LS , the constraints are most relevant for ensembles in which the constituents have lifetimes roughly around this timescale. However, electromagnetic injection can affect the ionization history of the universe in other ways as well. For example, energy deposited in the intergalactic medium as a result of particle decays at timescales well after t LS serves both to elevate the gas temperature and to increase the ionization fraction of hydrogen. These effects can shift the onset of reionization to earlier times [49,53] and impact the 21-cm absorption/emission signal arising from hyperfine transitions in neutral hydrogen [54][55][56]. These considerations are relevant primarily for particles with lifetimes τ i t now -particles which would in principle contribute to the present-day dark-matter abundance. Since our primary focus in this paper is on ensembles whose constituents have lifetimes τ i t now , we do not investigate these additional considerations here. However, we note that they can play an important role in constraining ensembles of particles with longer lifetimes, such as those which arise within the context of the DDM framework.
VI. CONTRIBUTIONS TO THE DIFFUSE PHOTON BACKGROUND
At times t t LS , the universe becomes transparent to photons with a broad range of energies O(keV) E γ O(TeV) [7]. Photons within this energy range which are injected on timescales t t LS are not reprocessed by Class-I processes and do not interact with CMB photons at an appreciable rate. Thus, as discussed in Sect. IV, such photons therefore simply free-stream until the present epoch and contribute to the diffuse photon background. The total diffuse background of photons observed from the location of Earth includes both a galactic contribution from processes occurring within the Milky-Way halo (at redshifts z ≈ 0) and an extra-galactic contribution from processes which occurred within the more distant past. The decays of unstable particles with lifetimes τ i t now contribute essentially exclusively to the extra-galactic component of this total background. By contrast, particles with lifetimes τ i t nowparticles which would contribute non-negligibly to the present-day abundance of dark matter -in principle contribute to both components. Since our primary focus in this paper is on particles with lifetimes in the regime τ i t now , we focus on the extra-galactic component of the diffuse photon background. However, we also note that since the local dark-matter density within the Milky-Way halo is several orders of magnitude larger than the corresponding cosmological density, the galactic component of the diffuse photon background is typically the more important of the two for constraining particles with lifetimes τ i t now .
The overall diffuse extra-galactic contribution dΦ/dE γ to the differential flux of photons per unit energy observed by a detector on Earth from the decays of an ensemble of unstable constituent particles χ i is simply the direct sum of the individual contributions dΦ i /dE γ from these constituents. In order to evaluate these individual contributions, we begin by noting that the decays of a given species χ i , averaged over all possible decay channels, produce a characteristic injection spectrum of photons dN i /dE γ . The spectrum of photons arriving at present time at a detector on Earth is the integrated total contribution from decays occurring at different redshifts. A photon injected at redshift z with energy E γ is redshifted to energy E γ = (1+z) −1 E γ when it arrives at the location of Earth. Thus, we find that the observed flux of photons observed at a detector on Earth from the decay of an individual decaying species χ i may be written in the form where dΩ is the differential solid angle of the detector, where ρ i (z) is the energy density of χ i at redshift z, where κ(z) is the optical depth of the universe to photons emitted at redshift z, and where (z) is the line-of-sight distance away from Earth, expressed as a function of z.
We focus here primarily on photons with energies in the range O(keV) E γ O(TeV), for which the universe is effectively transparent at all times t t LS and opaque at times t t LS . Moreover, for simplicity, we shall approximate the optical depth for such photons with a function of the form where z LS ≡ z(t LS ) ≈ 1100 is the redshift at the time of last scattering. In other words, we shall assume that the universe is completely transparent to all photons within this energy range after last scattering and completely opaque beforehand. Moreover, since t LS t MRE , we approximate the line-of-sight distance (z) in a flat universe dominated by matter and dark energy as where c is the speed of light, where H now is the presentday value of the Hubble parameter, and where we have defined where Ω m and Ω Λ respectively denote the present-day abundances of massive matter and dark energy, the former including contributions from both dark and baryonic matter. The energy density ρ i (z) of a decaying particle can be expressed in terms of its extrapolated abundance Ω i as where the time t(z) which corresponds to redshift z is .
With these substitutions, Eq. (6.1) becomes Finally, in assessing our prediction for dΦ/dE γ for a decaying particle ensemble, we must account for the fact that at a realistic detector, the photon energies recorded by the instrument differ from the actual energies of the incoming photons due to the non-zero energy resolution of the detector. We account for this effect by convolving the spectrum of incoming photons with a smearing function R (E γ − E γ ), which represents the probability for the detector to register a photon energy E γ , given an actual incoming photon energy E γ . For concreteness, we consider a Gaussian smearing function of the form where is a dimensionless parameter which sets the overall, energy-dependent standard deviation σ(E γ ) = E γ of the distribution. Thus, we have our final result for the extra-galactic photon flux: with the individual fluxes dΦ i /dE γ given by Eq. (6.7) and with R (E γ − E γ ) given by Eq. (6.8).
Limits on additional contributions to the extra-galactic photon flux at energies E γ 800 keV can be derived from measurements of the total gamma-ray flux by instruments such as COMPTEL, EGRET, and Fermi-LAT. For photon energies in the range 800 keV E γ 30 MeV, COMPTEL data on the diffuse extra-galactic background spectrum [57] are well modeled by a powerlaw expression of the form MeV −1 cm −1 s −1 sr −1 . (6.10) Likewise, for energies in the range 30 MeV E γ 1.41 GeV, EGRET data [58] are well modeled by the power-law expression At even higher energies, the spectrum of the extragalactic diffuse photon background can be inferred from Fermi-LAT data. Within the energy range 100 MeV E γ 820 GeV, these data are well described by a powerlaw expression with an exponential suppression at high energies [59]: (6.12) Finally, we note that while we focus here primarily on photons with energies above 800 keV, estimates of the extra-galactic photon background at lower energies have been computed from BAT [60] and INTEGRAL [61] data and from observations of Type-I supernovae [62]. The expression in Eq. (6.9) is completely general and applicable to any ensemble of particles which decays to photons with energies within the transparency window O(keV) E γ O(TeV). To illustrate how this expression can be evaluated in practice, we consider a concrete example involving a specific set of injection spectra dN i /dE γ for the ensemble constituents. In particular, we consider a scenario in which each constituent χ i decays with a branching fraction of effectively unity to a pair of photons via the process χ i → γ + γ. Under the assumption that the χ i are "cold" (i.e., non-relativistic) by the time t LS , each of these photons has energy E γ ≈ m i /2 in the comoving frame at the moment of injection. We thus have While higher-order processes will generically also give rise to a continuum spectrum which peaks around E γ ∼ O(10 −3 m i ), we focus here on the contribution from the photons with E γ ≈ m i /2, since line-like features arising from monochromatic photon emission are far easier to detect than continuum features. Thus, for the injection spectrum in Eq. (6.13), the total contribution to the differential extra-galactic diffuse photon flux from the decays of the constituents within this ensemble is given by (6.14)
VII. APPLICATIONS: TWO EXAMPLES
In Sects. III-VI, we examined the observational limits on electromagnetic injection from particle decays after the BBN epoch and formulated a set of constraints which can be broadly applied to any generic ensemble of unstable particles. In this section, we examine the implications of these constraints by applying them to a pair of toy ensembles which exhibit a variety of scaling behaviors for the constituent masses m i , extrapolated abundances Ω i , and lifetimes τ i . For simplicity, we shall focus primarily on the "early-universe" regime in which 1 s τ i t LS for all of the χ i . Within this regime, the leading constraints on injection are those from modifications of the abundances of light nuclei and from distortions in the CMB.
We consider two different classes of mass spectra for our decaying ensemble constituents. The first consists of spectra in which the masses of these decaying particles are evenly distributed on a linear scale according to the scaling relation where k = 0, 1, ..., N − 1 and where m 0 and ∆m are taken to be free parameters which describe the mass of the lightest ensemble constituent and the subsequent mass splittings respectively. Mass spectra of this approximate linear form are realized, for example, for the KK excitations of a particle propagating in a flat extra spacetime dimension whose total length is small compared with 1/m 0 . The second class of mass spectra we consider are spectra in which the particle masses are evenly distributed on a logarithmic scale according to the scaling relation where m 0 once again denotes the mass of the lightest ensemble constituent, and where ξ is a dimensionless free parameter. A mass spectrum of this sort is expected, for example, for axion-like particles in axiverse constructions [16]. A variety of scaling relations for the decay widths Γ i of the χ i across the ensemble can likewise be realized in different scenarios. For simplicity, in this paper we focus on the case in which each χ i decays with a branching fraction of effectively unity to a pair of photons via the process χ i → γ + γ. Moreover, we shall assume that the scaling relation for Γ i takes the form where Λ is a parameter with dimensions of mass and where C i is a dimensionless coefficient -a coefficient which in principle can also have a non-trivial dependence on m i . Such a scaling relation arises naturally, for example, in situations in which each of the χ i decays via a non-renormalizable Lagrangian operator of dimension d = 5 in an effective theory for which Λ is the cutoff scale. For purposes of illustration, we focus on the case in which the coefficient C i = C χ is identical for all C i . In this case, the Γ i are specified by a single free parameter: the ratio M * ≡ Λ/ C χ . We emphasize that it is possible -and indeed in most situations even expected -that C χ 1. Thus, unlike Λ itself, the parameter M * may exceed the reduced Planck mass M P without necessarily rendering the theory inconsistent. However, we remark that for values of M * in this regime, the Weak Gravity Conjecture [63] imposes restrictions on the manner in which the decay operator in Eq. (7.3) can arise.
Finally, we consider a range of different possible scaling relations for the extrapolated abundances Ω i across the ensemble. In particular, we consider a family of scaling relations for the Ω i of the form (7.4) in which the abundance Ω 0 of the lightest ensemble constituent and the scaling exponent γ are both taken to be free parameters. In principle, γ can be either positive or negative. However, since τ i ∝ m −3 i and since the leading constraints on injection at timescales t inj t LS are typically less stringent for earlier t inj , the most interesting regime turns out to be that in which γ ≥ 0 and the initial energy density associated with each χ i is either uniform across the ensemble or decreases with increasing τ i .
In summary, then, each of our toy ensembles is characterized by a set of six parameters: {N, m 0 , ∆m, M * , Ω 0 , γ} for the mass spectrum in Eq. (7.1) and {N, m 0 , ξ, M * , Ω 0 , γ} for the mass spectrum in Eq. (7.2). Note that for any particular choice of these parameters, the lifetimes of the individual ensemble constituents span a range τ N −1 ≤ τ i ≤ τ 0 .
A. Results for Uniform Mass Splittings
We begin by examining the results for a toy ensemble with a mass spectrum given by Eq. (7.1). For such a mass spectrum, the density n m of states per unit mass is uniform across the ensemble; however, it is also useful to consider the density n τ of states per unit lifetime, which is more directly related to the injection history. In the continuum limit in which the difference between the lifetimes of sequential states χ i and χ i+1 in the ensemble is small and the lifetime τ may be approximated as a continuous variable, we find that The density of states per unit lifetime is therefore greatest within the most unstable regions of the ensemble. Moreover, we also note that the density of states per unit ln(τ ) also decreases with ln(τ ). These considerations turn out to have important implications for the bounds on such ensembles, as shall become apparent below. In Fig. 9, we show the constraints on an ensemble comprising N = 3 unstable particles in which the lightest ensemble constituent has a mass m 0 = 10 GeV. We consider this ensemble to be "low-density" in that it has a density of states per unit τ which is small throughout the range of parameters shown. (This will later be contrasted with analogous results in Fig. 10 for a "high-density" ensemble.) The contours shown in the left and center panels of Fig. 9 represent upper bounds on the total extrapolated abundance Ω tot as functions of M * for fixed ∆m. The contours shown in the left panel correspond to the choice of ∆m = 15 GeV; those in the center panel correspond to the choice of ∆m = 95 GeV. In the right panel, we show the corresponding contours as functions of ∆m for fixed M * = 10 19.5 GeV. The blue, green, and red curves shown in each panel represent the bounds on Ω tot from constraints on the primordial abundances of D, 7 Li, and 6 Li, respectively. We use a uniform color for the D contour -a contour which reflects the sum of contributions to δY D from both production and destruction processes. The dotted, dashed, and solid curves of each color correspond respectively to the choices γ = {0, 1, 2} for the scaling exponent in Eq. (7.4). The corresponding black curves shown in each panel represent the bounds arising from limits on distortions of the CMBphoton spectrum obtained by taking whichever bound (either that from δµ or that from δy C ) is more stringent at every point.
In interpreting the results shown in Fig. 9, we note that each choice of parameters specifies a particular range of lifetimes τ N −1 ≤ τ i ≤ τ 0 for the ensemble constituents. This range of lifetimes varies across the range of M * shown in the left panel, from 1.03 × 10 4 s ≤ τ i ≤ 6.58 × 10 5 s for M * = 10 16.5 GeV to 1.03 × 10 10 s ≤ τ i ≤ 6.58 × 10 11 s for M * = 10 19.5 GeV. Across this entire range of M * values, the range of τ i values for the ensemble is reasonably narrow, spanning only roughly a single order of magnitude. As a result, the bounds on Ω tot for a decaying ensemble shown in the left panel of Fig. 9 as functions of M * closely resemble the bounds on Ω χ for a single decaying particle species shown in Fig. 6 as a function of τ χ . Moreover, since the τ i for all particles in the ensemble are roughly comparable, the results are not particularly sensitive to γ, which determines how Ω tot is distributed across the ensemble. The most stringent constraint on Ω tot is the bound associated with the primordial D abundance over almost the entire range of M * shown. The only exception occurs in the regime where M * is large and the lifetimes of all of the ensemble constituents are extremely long, in which case the constraint associated with CMB distortions supersedes the D constraint. We note that for values of Ω tot which lie above the D contour, D may be either overabundant or underabundant. We also note that a "funnel" similar to that appearing in Fig. 6 is visible in the D constraint contours for all values of γ shown in the figure, and that this funnel occurs where M * is such that the lifetimes of the ensemble constituents are approximately τ i ≈ 10 6.5 s. By contrast, the results shown in the center panel of Fig. 9 correspond to the regime in which the range of lifetimes spanned by the ensemble is quite broad, extending over several orders of magnitude in τ i . In particular, the range of lifetimes varies across the range of M * shown in this panel from 1.03 × 10 4 s ≤ τ i ≤ 8.22 × 10 7 s for M * = 10 17.5 GeV to 8.22 × 10 7 s ≤ τ i ≤ 6.58 × 10 11 s for M * = 10 19.5 GeV. We note that the range of M * shown in this panel has been truncated relative to the left panel in order to ensure that the characteristic decay timescales for all ensemble constituents occur well after Fig. 6, we see that the bounds on a decaying ensemble differ not only quantitatively but even qualitatively from the bounds on its individual constituents. the BBN epoch. For an ensemble with such a broad range of lifetimes, the effect of distributing Ω tot over multiple particle species is readily apparent. Indeed, the shapes of the constraint contours in the center panel of Fig. 9 bear little resemblance to the shapes of the contours in Fig. 6. Thus, comparing the results shown in the center panel of Fig. 9 with those in Fig. 6, we see that the bounds on a decaying ensemble indeed differ not only quantitatively but even qualitatively from the bounds on its individual constituents. For example, two separate funnels arise in the D constraint contour for γ = 2. Each of these funnels represents not merely a cancellation between the production and destruction contributions to Y D from a single particle, but rather a cancellation among the individual contributions from all constituent particle species in the ensemble. The funnel on the left arises due to a cancellation between the net negative contributions to Y D from χ 1 and χ 2 and the net positive contribution from χ 0 . By contrast, the funnel on the right arises due to a cancellation between a net negative contribution to Y D from χ 2 and positive contributions from χ 0 and χ 1 , with the latter contribution suppressed because τ 1 ≈ 10 6.5 s. The positions of these funnels, then, reflect the collective nature of the ensemble -and they have important consequences. Indeed, within certain regions of parameter space within which the D constraints are weakened by cancellations among δY D contributions from different ensemble constituents, the constraint associated with the 6 Li abundance actually represents the leading bound on Ω tot for the ensemble.
Finally, in the right panel of Fig. 9, we show how the bounds on Ω tot vary as a function of ∆m for fixed M * = 10 19.5 GeV. We choose this benchmark for M * because the CMB-distortion constraints and the leading constraints from light nuclei are comparable for this choice of M * . In interpreting the results displayed in this panel, it is important to note that τ 0 is independent of ∆m. By contrast, the τ i for all other χ i decrease with increasing ∆m. Since, broadly speaking, the impact that a decaying particle has on both CMB distortions and on the primordial abundances of light nuclei tends to decrease as τ i decreases, the bounds on decaying ensembles generally grow weaker as ∆m increases. This weakening of the constraints generally grows more pronounced as γ increases and as a larger fraction of the abundance is carried by the heavier ensemble constituents. For γ = 0, on the other hand, the fraction of Ω tot carried by the lightest ensemble constituent χ 0 is independent of ∆m. As a result, many of the constraint contours shown in the right panel of Fig. 9 become essentially flat for sufficiently large ∆m as the lifetimes of the remaining χ i become so short that the impact of their decays on the corresponding cosmological observables is negligible in comparison with those of χ 0 .
Thus far, we have focused on the regime in which the spacing between the lifetimes τ i of the ensemble constituents is large -in other words, the regime in which the density of states per unit τ is small across the ensemble. However, it is also interesting to consider the opposite regime, in which the density of states per unit τ is sufficiently large over the entire range τ N −1 ≤ τ ≤ τ 0 that the ensemble effectively acts as a continuous source of electromagnetic injection over this interval. This range of lifetimes is completely determined in our toy model by M * , m 0 , and the quantity m N −1 = (N − 1)∆m + m 0 . Thus, we may make a direct comparison between the bounds on such "high-density ensembles" and the bounds on "low-density" ensembles with the same τ i range by varying N and ∆m oppositely while holding M * , m 0 , and m N −1 fixed. Towards this end, in Fig. 10, we show the upper bounds on Ω tot for an ensemble with a higher density of states per unit lifetime. The constraint contours in the left and center panels of the figure represent the same choices of m 0 and m N −1 as in the corresponding panels of Fig. 9, but for N = 20 rather than N = 3. For these parameter choices, the ensemble is effectively in the "high-density" regime over the entire range of M * shown in each panel. In the right panel of Fig. 10, we once again show constraint contours as functions of ∆m for fixed M * .
As one might expect, the primary effect of distributing the abundance of the ensemble more evenly across the same range of lifetimes is that features in the constraint contours associated with particular χ i are in large part smoothed out. While the D funnels in the left panel are still present, this reflects the fact that the range of τ i for the ensemble is sufficiently narrow that distributing the abundance more democratically across this range has little impact on the constraint contours. A single D funnel also appears in the constraint contour for γ = 2 in the center panel of Fig. 10. The presence of this feature reflects the fact that for γ = 2, the vast majority of Ω tot is carried by the most massive -and therefore most unstable -ensemble constituents. Moreover, the density of states per unit lifetime is also much higher for these shorter-lived states than it is for the rest of the ensemble. The funnel can be understood as corresponding to the value of M * for which the lifetimes of these states, which are smaller than but roughly similar to τ N −1 , satisfy τ i ∼ τ N −1 ≈ 10 6.5 s. Once again, however, we note that the location of the funnel is shifted slightly away from this value of M * due to the collective contribution to Y D from the longer-lived constituents in the ensemble. A similar feature is also apparent in the constraint contour for γ = 2 in the right panel of Fig. 10.
We can gain further insight into the results displayed in Fig. 10 by examining the continuum limit in which ∆m → 0.
In this limit, the sums appearing in Eqs. (3.52)-(3.54) and in Eqs. (4.17) and (4.18) can be recast as integrals over the continuous lifetime variable τ -integrals which may be evaluated in a straightforward manner, yielding analytic expressions for δY a , δµ, and δy C . For example, in the case in which the τ i span the entire range from t Aa to t f a for each relevant nucleus, we find that for any γ ≥ 0 the term δY where Γ(n, z) is the incomplete gamma function, where we have defined where A a and B a are defined in Table II and where the limits of integration are where the limits of integration are given by expressions analogous to those appearing in Eq. (7.8).
Expressions for the CMB-distortion parameters δµ and δy C may be obtained in a similar fashion. In particular, for the case in which the τ i span the entire range from t e to t LS , we find that δµ takes the form x min 2µ (7.10) for all γ ≥ 0, except for the special case in which γ = 1. For this special case, the replacement should be made in the fourth line of Eq. (7.10). Likewise, we find that δy C takes the form x min 2y , (7.12) where 2 F 1 (a, b; c; z) denotes the ordinary hypergeometric function, where A a , B a , and α a are defined in Table III, and where the limits of integration are We now consider the results for a toy ensemble with a mass spectrum given by Eq. (7.2). The density of states per unit mass is inversely proportional to m in this case, and the corresponding density of states per unit lifetime is which likewise decreases with τ . However, we note that the density of states per unit ln(τ ) in this case remains constant across the ensemble.
In Fig. 11 we show the bounds on Ω tot for a "lowdensity" ensemble with a mass spectrum given by Eq. (7.2) and the same benchmark parameter choices N = 3 and m 0 = 10 GeV as in Fig. 9. In the left and center panels, the value of ξ has been chosen such that the range of τ i for the ensemble constituents is the same as in Fig. 9 at both ends of the range of M * shown. For the mass spectrum considered here, this range of τ i depends on m 0 , on M * , and on the quantity ξ N −1 . In the right panel of Fig. 11, we show how the bounds on Ω tot vary as a function of ξ for fixed M * = 10 19.5 GeV. The contours shown in Fig. 11 exhibit the same overall scaling behaviors as those in Fig. 9. However, there are salient differences which reflect the fact the the density of states per unit ln(τ ) is uniform in this case across the ensemble.
In Fig. 12, we show the corresponding results for an ensemble with m 0 = 10 GeV and a mass spectrum again given by Eq. (7.2), but with a sufficiently high n τ throughout the range of M * or ξ shown in each panel that the ensemble is effectively always within the "highdensity" regime. We find that taking N = 10 is sufficient to achieve this. The left and center panels of Fig. 12 correspond to the same ranges of M * and the same value of ξ N −1 as the left and center panels of Fig. 11. The right panel once again shows how the bounds on Ω tot vary as a function of ξ for fixed M * = 10 19.5 GeV. Once again, we see that the principal consequence of increasing n τ is that constraint contours become increasingly smooth and featureless.
Once again, as we did for the case of an ensemble with a uniform mass splitting, we may derive the corresponding expressions for δY a , δµ, and δy C in the case of an exponentially rising mass splitting in the continuum limit by direct integration of the expressions in Eqs. (3.52)-(3.54) and in Eqs. (4.17) and (4.18). For the mass spectrum in Eq. (7.2), this limit corresponds to taking ξ → 1. For the case in which the τ i span the entire range from t Aa to t f a for each relevant nucleus, we find that for γ > 0, the term δY Fig. 11, but for "high-density" ensembles with N = 10. In the left and center panels, we plot the upper bound on Ωtot as a function of the coupling-suppression scale M * with fixed ξ = 1.2 and with fixed ξ = 1.39, respectively. In the right panel, we plot the bound as a function of ξ with fixed M * = 10 19.5 GeV. where the limits of integration are given by expressions analogous to those appearing in Eq. (7.8). Once again, as was the case with δY (1) a , the expression for δY (2) a in Eq. (7.19) requires modification for the special case in which γ = 0. In particular, the replacement specified in Eq. (7.17) should likewise be made in the last term in square brackets on the second line of Eq. (7.19). Moreover, for γ = 0, the quantity Ξ(t) is given by Eq. (7.18) rather than by Eq. (7.16).
The corresponding expressions for δµ and δy C for the case in which the τ i span the entire range from t e to t LS can likewise be computed by direct integration. In particular, we find that δµ is given by x min 2µ (7.20) for all γ ≥ 0, except for the special case in which γ = 2. For this special case, the replacement should be made in the fourth line of Eq. (7.20). Likewise, we find that δy C is given by x max 0y x min 0y + B y Ξ(t 1y ) 6x α+ 3−2γ 6 6α µ − 2γ + 3 2 F 1 1, 1 + 3 − 2γ 6α y ; 2 + 3 − 2γ 6α y ; −x αy x max 1y x min 1y + B y Ξ(t 2y ) 3x x min 2y .
(7.22)
The limits of integration in Eqs. (7.20) and (7.22) are defined as in Eq. (7.8), and Ξ(t) is defined as in Eq. (7.16) for γ > 0 and as in Eq. (7.18) for γ = 0. Once again, we note that in cases in which the τ i do not span the relevant range of timescales during which particle decays can affect one of these cosmological observables, the limits of integration in Eqs. (7.15)-(7.22) should be replaced by the values which restrict the overall range of lifetimes appropriately.
In summary, the results shown in Figs. 9-12 illustrate some of the ways in which the constraints on injection from unstable-particle decays can be modified in scenarios in which the injected energy density is distributed across an ensemble of particles with a range of lifetimes. These results also illustrate that a nontrivial interplay between the contributions from different decaying particle species within the ensemble can have unexpected and potentially dramatic effects on the upper bound on Ω tot -as when cancellations among positive and negative contributions to δY D from a broad range of particles within the ensemble result in a significant weakening of this upper bound. Thus, we see that the bounds on a decaying ensemble can exhibit collective properties and behaviors that transcend those associated with the decays of its individual constituents.
VIII. CONCLUSIONS: DISCUSSION AND SUMMARY OF RESULTS
In this paper, we have considered ensembles of unstable particle species and investigated the cosmological constraints which may be placed on such ensembles due to limits on electromagnetic injection since the conclusion of the BBN epoch. Indeed, as we have discussed, such injection has the potential to modify the primordial abundances of light nuclei established during BBN. Such injection can also give rise to spectral distortions in the CMB, alter the ionization history of the universe, and leave observable imprints in the diffuse photon background. For each of these individual considerations, we have presented an approximate analytic formulation for the corresponding constraint which may be applied to generic ensembles of particles with lifetimes spanning a broad range of timescales 10 2 s τ i t now . For ease of reference, these analytic approximations, along with the corresponding equation numbers, are compiled in Table IV. In deriving these results, we have taken advantage of certain linear and uniform-decay approximations. We have also demonstrated how these results can be applied within the context of toy scenarios in which the mass spectrum for the decaying ensemble takes one of two characteristic forms realized in certain commonly-studied
Constraint
Analytic formulation
Primordial abundances of light nuclei
Analytic approximations for δYa given in Eq. (3.51), with the corresponding limits and parameter values given in Table II.
Spectral distortions in the CMB
Analytic approximation for δµ and δy C given in Eqs. (4.17) and (4.18), with the parameter values given in Table III Diffuse extra-galactic photon background Differential signal flux given by Eq. (6.9), with differential background fluxes given in Eqs. (6.10)-(6.12). extensions of the SM.
Several comments are in order. First, it is worth noting that while the values of the parameters in Table II were derived assuming a particular set of initial comoving number densities Y a for the relevant nuclei at the end of the BBN epoch, the corresponding parameter values for different sets of initial comoving number densities Y a may be obtained in a straightforward manner from the values in Table II. Provided that Y a and Y a are not significantly different, the characteristic timescales t Ba , t Ca , and t Xa for each process are to a good approximation unchanged, and a shift in the initial comoving number densities can be compensated by an appropriate rescaling of the relevant normalization parameters A a and B a . For example, a shift in the initial helium mass fraction at the end of BBN can be compensated by a rescaling of the fit parameters A * D and B * D associated with D production and of the parameters A4 He and B4 He associated with 4 He destruction by Y p /Y p , as well as a rescaling of the parameters A * 6 Li and B * 6 Li associated with secondary 6 Li production by (Y p /Y p ) 2 . The quadratic dependence on Y p /Y p exhibited by the last of these rescaling factors reflects the fact that the reactions through which the non-thermal populations of 3 He and T nuclei are initially produced and the reactions through which these nuclei in turn contribute to 6 Li production both involve 4 He in the initial state. Similarly, a shift in the initial 7 Li abundance can be compensated by a rescaling of the fit parameters A7 Li and B7 Li associated with 7 Li destruction and of the parameters A6 Li and B6 Li associated with primary 6 Li production by Y 7 Li /Y7 Li , while a shift in the initial D abundance can be compensated by a rescaling of the parameters A D and B D associated with D destruction by Y D /Y D .
Second, we remark that in this paper we have focused primarily on constraints related to electromagnetic injection. In situations in which a significant fraction of the energy liberated during unstable particle decays is released in the form of hadrons, the limits obtained from bounds on the primordial abundances of light nuclei may differ significantly from those we have derived here. Detailed analyses of the limits on hadronic injection in the case of a single decaying particle species have been performed by a number of authors [3,[64][65][66]. It would be interesting to generalize these results to the case of a decaying ensemble as well.
Third, as we have seen, there are two fundamentally different timescale regimes in which the corresponding physics is subject to very different leading bounds: an "early" regime 10 2 s < ∼ t inj < ∼ 10 12 s, and a "late" regime t inj > ∼ 10 12 s. This distinction is particularly important when considering the implications of our results within the context of the Dynamical Dark Matter framework [18,19]. Within this framework, the more stable particle species within the ensemble provide the dominant contribution to the dark-matter abundance at present time, while the species with shorter lifetimes have largely decayed away. However, the masses, decay widths, abundances, etc., of the constituents of a realistic DDM ensemble are governed by the same underlying set of scaling relations. Thus, in principle, one might hope to establish bounds on the ensemble as a whole within the DDM framework by constraining the properties of those ensemble constituents which decay on timescales τ t LS -i.e., "early" timescales on which the stringent limits on CMB distortions and the primordial abundances of light nuclei can be brought to bear.
In practice, however, it turns out that these considerations have little power to constrain most realistic DDM scenarios. On the one hand, both the lifetimes τ i and extrapolated abundances Ω i of the individual ensemble constituents typically decrease monotonically with i in such scenarios -either exponentially [67] or as a power law [19,68,69]. On the other hand, as indicated in Fig. 8, the constraint associated with the broadening of the surface of last scattering due to ionization from particle decays becomes increasingly stringent as the lifetime of the particle decreases down to around τ i ∼ t LS . This implies that for an ensemble of unstable particles in which the collective abundance of the cosmologically stable constituents is around Ω DM ≈ 0.26, the Ω i for those constituents with lifetimes τ i t LS are constrained to be extremely small. Thus, the constraints associated with CMB distortions and with the primordial abundances of light nuclei in the "early" regime are not particularly relevant for constraining the properties of DDM ensembles in which the abundances scale with lifetimes in this way. Nevertheless, we note that in ensembles in which τ i and Ω i do not increase monotonically with i (such as those in Ref. [69]), the constraints associated with these considerations could indeed be relevant. This is currently under study [70]. | 36,171.8 | 2018-10-24T00:00:00.000 | [
"Physics"
] |
Lateralization of spatial information processing in response monitoring
The current study aims at identifying how lateralized multisensory spatial information processing affects response monitoring and action control. In a previous study, we investigated multimodal sensory integration in response monitoring processes using a Simon task. Behavioral and neurophysiologic results suggested that different aspects of response monitoring are asymmetrically and independently allocated to the hemispheres: while efference-copy-based information on the motor execution of the task is further processed in the hemisphere that originally generated the motor command, proprioception-based spatial information is processed in the hemisphere contralateral to the effector. Hence, crossing hands (entering a “foreign” spatial hemifield) yielded an augmented bilateral activation during response monitoring since these two kinds of information were processed in opposing hemispheres. Because the traditional Simon task does not provide the possibility to investigate which aspect of the spatial configuration leads to the observed hemispheric allocation, we introduced a new “double crossed” condition that allows for the dissociation of internal/physiological and external/physical influences on response monitoring processes. Comparing behavioral and neurophysiologic measures of this new condition to those of the traditional Simon task setup, we could demonstrate that the egocentric representation of the physiological effector's spatial location accounts for the observed lateralization of spatial information in action control. The finding that the location of the physical effector had a very small influence on response monitoring measures suggests that this aspect is either less important and/or processed in different brain areas than egocentric physiological information.
INTRODUCTION
In order to adequately interact with our environment, we constantly monitor our actions so that we can adjust them in case of undesired consequences (Logan, 1985;Stuss and Alexander, 2007;Fukui and Gomi, 2012). Properly doing so is a fairly complex endeavor because for a proper adjustment of the outcome, parameters of movements also need to be integrated in the process of response evaluation.
Given that different features (like speed, spatial position, applied force of the response, etc.) influence our movements, these parameters have to be integrated in the evaluation process (Praamstra et al., 2009;Fukui and Gomi, 2012;Gonzalez and Burke, 2013;Stock et al., 2013). We recently investigated the effects of multimodal sensory integration in response monitoring processes by recording an EEG during a Simon task (see Stock et al., 2013 for details) and demonstrated that both proprioception-based spatial information and efference-copy-based information on the motor execution are integrated in the supplementary motor area (SMA) during response monitoring and evaluation. Among other things, this brain region has been associated with the processing efference copies of motor commands (Neshige et al., 1988;Ikeda et al., 1995;Babiloni et al., 2001;Haggard and Whitford, 2004;Beaulé et al., 2012), egocentric proprioceptive information (Tarkka and Hallett, 1991;Hallett, 1994;Loayza et al., 2011), motor control (Angel, 1976;Wolpert and Flanagan, 2001;Allain et al., 2004;Yordanova et al., 2004;Feldman, 2009;Hoffmann and Falkenstein, 2010;Roger et al., 2010), and error monitoring (Peterburs et al., 2011). However, we obtained an unexpected pattern of hemispheric activation by asking the subjects to either cross their hands or keep them parallel while responding: in parallel hands, only the SMA contralateral to the responding hand showed a negative deflection of event-related potentials (ERPs) around the time of the response while the SMA ipsilateral to the responding hand showed a positivation. By contrast, the simple act of crossing one hand one over another reduced most of the differences in hemispheric activation/ERPs as the activity pattern of the hemisphere ipsilateral to the responding hand approximated that of the contralateral hemisphere. This suggests that in case of an unnatural posture (crossed hands) motor efference copies and motor proprioceptive information were allocated to the hemispheres according to different rules: efference-copy-based motor information seemed to be rather immutably locked to the hemisphere in which the motor command was initially processed. In contrast, the hemispheric allocation of proprioception-based spatial information was based on an external representation of space. As a result of these different lateralization mechanisms, crossing hands www.frontiersin.org (manually entering a "foreign" spatial hemifield) most probably resulted in a conflict, yielding an augmented bilateral activation and higher error rates.
Even though these findings seem to answer the question in which hemisphere the monitoring of motor and spatial information is allocated, the paradigm provided no possibility to determine whether the laterlized allocation of spatial information during response monitoring was influenced by internal (proprioceptive) information about the position of the physiologic effector (hand) or by external (egocentric) information about the position of the physical effector (button).
In the current study, we aimed at answering this question. For this purpose, we modified the Simon task by introducing a "double crossed" condition. While the regular Simon task only encompasses a parallel-hands and a crossed-hands condition, our new double crossed condition required the subjects to also operate crossed levers in half of the trials. As a consequence, the effect site (button) which was pressed when crossing hands in lever responses was in a different hemifield than the responding hand so that the button was the same as during a regular parallel hands button response (see Figure 1 for further elucidation). Based on this dissociation of physiological effector (hand) and physical effect site (button), our question could be tackled: in case the spatial allocation of the hand is the relevant factor to the lateralization of response monitoring processes, parallel and crossed hands should yield comparable ERPs, irrespective of whether buttons or levers are used to respond. If however, the external effect site of the button was the critical feature, parallel-hands button responses should yield results similar to those of crossed-hands lever responses.
SAMPLE
Right-handed participants (N = 21; ♀ = 11, ♂ = 10) were included in the study. The mean age was 23.2 years (min 19, max 32, SEM = 0.73) and none of the participants presented with a history of psychiatric or neurological disease. Handedness was confirmed by the Edinburgh Handedness Inventory (Oldfield, 1971), yielding an average score of 0.81 (min 0.25, max 1.0, SEM = 0.05). All subjects gave written informed consent and were treated in accordance with the declaration of Helsinki. Each participant was reimbursed with 15€. The study was approved by the ethics committee of the medical faculty of the University of Bochum.
SETTINGS AND TASK
Because this study aims to extend previous findings reported by Stock et al. (2013), the settings and task were very similar to that study (see Stock et al., 2013 for details): participants were seated in front of a 17 in CRT computer monitor (at a distance of 57 cm) in a dimly lit and sound-attenuated room. Responses were made with four custom-made buttons mounted on a regular keyboard (see Figure 1 for illustration).
The Simon task originally references the task used by Wascher et al. (2001). Throughout the whole task, a white fixation cross and two horizontally aligned white frame boxes were continuously displayed in the center of a dark blue screen. The two boxes were at the same vertical level as the fixation cross (1.1 • distance between fixation cross and the inner border of the frames). Each trial began with the simultaneous presentation of a target stimulus (a yellow capital letter "A" or "B") and a noise stimulus (three white horizontal bars). Both target and noise stimuli were approximately FIGURE 1 | The four different response conditions resulting from hand position (parallel vs. crossed) and button type (buttons vs. levers). When crossing hands, the participants were instructed to place the left arm ("marked" with two wristbands in the picture) on top of the right arm. In button responses, the physiological effector (hand) is in the same hemifield as the physical effector (button) so that their relevance for the hemispheric allocation of response monitoring processes cannot be determined. In contrast, the levers provide the necessary dissociation because the physical effector (button) is now located in a different hemifield than the physiological effector (hand). For mechanical reasons, buttons had to be pressed while levers had to be lifted. 0.5 • wide and 0.6 • high and presented within the two opposing white boxes. After 200 ms, the stimuli disappeared and the trial was ended by the first (button press) response. If the participants did not respond within the first 500 ms after the onset of the trial, a speed-up sign (containing the German word "Schneller!" which translates to "Faster!") was presented above the stimuli until the end of the trial. In case no response was given, the trial automatically ended 1700 ms after its onset and was coded as a "miss." The response-stimulus intervals (RSIs) varied randomly and ranged between 2000 and 2500 ms.
The experiment consisted of eight blocks, each comprising 160 trials. The four stimuli ("A" on the left side/"A" on the right side/"B" on the left side/"B" on the right side) were randomized and occurred equally often, resulting in 40 trials per condition and block. For all blocks, participants were instructed to respond using the left index finger whenever the target stimulus was an "A" and to respond using the right index finger whenever the target stimulus was a "B" (in both cases irrespective of the target's location on the screen). In blocks 1, 3, 5, and 7 they were asked to respond with parallel hands while they were asked to cross their hands (placing the left arm above the right arm) in blocks 2, 4, 6, and 8. In addition to the setup of our previous study (Stock et al., 2013), participants were requested to respond by pressing the buttons in blocks 1, 2, 5, and 6 while levers had to be used in blocks 3, 4, 7, and 8 (see Figure 1). Hence, there were two blocks for each combination of hand position (parallel/crossed) and button type (buttons/levers). Following from this, there were equal numbers of congruent and incongruent trials (classified depending on whether the responding hand was placed in the same hemifield as the target stimulus).
EEG RECORDING DATA PROCESSING
As for the settings and task, EEG data recording and data processing are very similar to techniques used for our previous publication (see Stock et al., 2013 for details): an EEG was recorded from 65 Ag-AgCl electrodes at standard positions (international 10-20 system) while the participants were performing the task. Electrode impedances were kept below 5 k . During recording, a filter bandwidth of 0-80 Hz was applied and EEG data was recorded with a sampling rate of 1000 samples per second against a reference at electrode FCz. After recording, the data was downsampled to 256 Hz and an IIR filter (notch at 50 Hz; high-pass at 0.5 Hz and low-pass at 18 Hz, using a slope of 48 dB/oct each) was applied. Subsequently, technical artifacts and irregular muscular artifacts (e.g., jaw clenching) were removed during a visual raw data inspection. Uniform artifacts (primarily blinks, eye movements and pulse artifacts) were removed by means of an independent component analysis (ICA) applying the infomax algorithm.
For stimulus-locked event-related lateralizations (ERLs), segments were formed for the different conditions. Epochs started 200 ms before the stimulus presentation (which was set to time point zero) and ended 1200 ms after the response, resulting in a total epoch length of 1400 ms. For the analysis of responselocked event-related potentials (ERPs), segments were formed for the different conditions. Epochs started 1200 ms before the response (which was set to time point zero) and ended 1200 ms after the response, resulting in a total epoch length of 2400 ms.
Independent of the locking point (stimulus or response), only trials that had been correctly answered within the first 1500 ms after the onset of the stimulus presentation were included. Furthermore, an automated artifact rejection removed amplitudes above 100 μV and below −100 μV as well as activity of less than 0.5 μV over a time span of 100 ms or more. Subsequently, a current source density (CSD) transformation was applied to eliminate the reference potential (Perrin et al., 1989;Nunez and Pilgreen, 1991;Nunez et al., 1997).
For the analysis of stimulus-locked ERLs/N2pc, a baseline correction from −200 to 0 ms was run before the segments of the different conditions were averaged. Based on the topographic distribution of the activity and the literature relevant to this task, ERLs were formed for electrodes PO7 and PO8 (Praamstra and Oostenveld, 2003;Wiegand and Wascher, 2005;Verleger et al., 2012;Cespón et al., 2013;Stock et al., 2013). For this purpose, the values of the hemisphere ipsilateral to the target stimulus site were subtracted from the values of the hemisphere contralateral to the target stimulus site (PO7-PO8 for stimuli presented on the right side and PO8-PO7 for stimuli presented on the left side) and averaged for both hands. For statistical analyses, we extracted the mean electrode activity between 180 and 270 ms (the time frame was based on the negative peak and differences between the conditions; see Figure 2).
For the analysis of response-locked ERPs, a baseline correction from −1200 to −800 ms was run before the segments of the different conditions were averaged. Based on our previous study, we decided to quantify the response-locked ERPs at electrodes FC1 and FC2 because these electrodes have been shown to optimally depict response evaluation differences/changes in SMA activity between the different conditions of this task (see Coles, 1989;Leuthold, 2011;Stock et al., 2013 for details). For statistical analyses, we extracted the mean electrode activity between −60 and 60 ms (the time frame was based on the differences between the conditions; see Figure 3).
STATISTICAL ANALYSIS
Behavioral data (RTs and the number of hits/correct responses) were analyzed using repeated-measures analyses of variance (ANOVA). "Button type" (button responses vs. lever responses), "hand position" (parallel hands vs. crossed hands), and "congruency" (congruent vs. incongruent; codes whether the target stimulus was presented on the side where the responding hand was placed) were used as within-subjects factors. The electrophysiological stimulus-locked data was analyzed using repeated-measures ANOVA with the within-subjects factors "button type," "hand position," and "congruency." Because ERLs are based on the difference between the hemisphere contralateral and ipsilateral to the stimulus presentation site, there was no factor for side/hemisphere. The electrophysiological response-locked data was analyzed in similar fashion using a repeated-measures ANOVA with the within-subjects factors "button type," "hand position," "congruency," and"executive hemisphere"(electrode above the hemisphere responsible for the motor execution of the response vs. electrode above the hemisphere irresponsible for the motor execution of www.frontiersin.org the response). Greenhouse-Geisser-correction was used whenever necessary. All p-levels for post hoc t-tests were adjusted using Bonferroni correction. Effect sizes were given as the proportion of variance accounted for (η 2 ). As a measure of variability, the standard error of the mean (SEM) together with the mean values was given.
Summary of behavioral results
Briefly summing up the behavioral results, significant interactions show that the subjects hit rate was differently modulated across congruency: in congruent trials only, button responses had higher hit rates than lever responses while in incongruent trials only, parallel-hand responses had higher hit rates than crossed-hand responses.
Furthermore, a threefold interaction indicated that hit RTs were modulated by button type, congruency, and hand position: while congruency modulated the RT in both button and lever responses (congruent faster than incongruent), only button response RTs were additionally modulated by hand position (parallel faster than crossed).
Stimulus-locked data
Stimulus-locked data at electrodes PO7 and PO8 are depicted in Figure 2.
Response-locked data
Response-locked ERPs at electrodes FC1 and FC2 are depicted in Figure 3.
Summary of neurophysiological results
Briefly summing up the electrophysiological results, the stimuluslocked ERLs of correct responses were modulated by an interaction of congruency and hand position: only in parallel-hand responses, congruent trials evoked significantly more negative ERLs than incongruent trials. Furthermore, the response-locked ERPs of correct responses were modulated by an interaction of button type, hand position, and hemisphere (but not by congruency): in the non-executive hemisphere, button and lever responses differed from each other when hands were crossed (but not when they were parallel). By comparison, the mean amplitudes of the executive hemisphere only differed between parallel and crossed-hand responses.
DISCUSSION
The current study aimed at determining whether the location of an internal/physiologic effector (hand) or the location of an external, physical effector (response button) accounts for the previously observed asymmetric lateralization of spatial aspects of response monitoring processes (Stock et al., 2013).
The results (especially the interaction pattern observed in the response-locked ERP data) suggest that the spatial location of the physiologic effectors accounts for the largest part of the observed changes in the hemispheric allocation of spatial information during response monitoring. In order to elucidate the rationale behind this interpretation, we would like to explain the theoretical background of our experimental manipulation: the basic assumption behind the additional factor"button type"is that"each hemisphere preferentially processes and integrates the contralateral egocentric and allocentric spatial information" (Zhou et al., 2012). Following from this, trials with button responses provide a "baseline" measurement because the hand and button involved in a response are always located in the same spatial hemifield. Differences between the two hand positions (parallel vs. crossed) can be attributed to spatial properties of the effectors, but the effectors (hand vs. button) cannot be told apart. In contrast to this, trials with lever responses provide the measurement of our "experimental manipulation" because in this condition, the responding hand and the button pressed are always located in opposing spatial hemifields. Hence, the influence of the different effectors can be distinguished by comparing baseline and experimental manipulation/button and lever trials: influences exerted by the physiologic effector/the location of the hand should yield identical or at least similar result for both button types (i.e., parallel-hand button responses ≈ parallel-hand lever responses and crossed-hand button responses ≈ crossed-hand lever responses). In contrast to this, influences exerted by the physical effector/the location of the button should yield opposing or at least different results for the two button types (i.e., parallel-hand button responses ≈ crossed-hand lever responses and crossed-hand button responses ≈ parallel-hand lever responses).
The first option is basically what was observed in the responselocked ERPs. Such fronto-central ERPs are known to reflect response monitoring and evaluation processes and are most likely generated within the SMA, anterior cingulate cortex, and adjacent areas (Macar et al., 1999;Luu and Tucker, 2001;Beste et al., 2010aBeste et al., ,b, 2012Roger et al., 2010;Wascher and Beste, 2010). In our previous study, we could demonstrate the response-locked ERPs quantified in this study originate within the SMA and are sensitive to the spatial allocation of the effector (Stock et al., 2013). As described above, we aimed at identifying the effector (physical or physiological) by comparing button and lever response conditions. As can be seen in the top row ("button responses") of Figure 3, placing the effectors in their usual hemifield yields a positivation of the response-locked ERP over the non-executive hemisphere. By contrast, placing the effectors in the "foreign" hemifield yields a negativation of the response-locked ERP over the non-executive hemisphere so that it resembles the course of the ERP curve over the executive hemisphere. Furthermore, it can be noted that the ERP over the non-executive hemisphere is more negative when the effectors are placed in the contralateral hemifield. A repeated-measures ANOVA was run to compare lever responses to button responses. Due to the interactions of factors, the main effects of hand position and hemisphere cannot be subject to interpretation. We would however like to point out that there was no main effect of button type. Hence, there was no basic fundamental difference between buttons and levers which is in favor of assuming the hands to be the relevant effectors. Two interactions are important: first, there was an interaction of hand position and congruency. Because both post hoc tests yielded significant differences between the hand positions (each parallel > crossed), the finding only differed quantitatively between congruent and incongruent trials. Second, there was a threefold interaction of button type, hand position, and hemisphere. This interaction is crucial when trying to answer the question of which effector (hand or button) accounts for lateralization of spatial aspects of response monitoring processes. The button Frontiers in Psychology | Cognition type had no effect on the executive hemisphere that always processes efference-copy-based information of the motor aspects of the response and information on spatial properties of the response in half of the trials. In the non-executive hemisphere, the button type only had an effect in crossed hands (lever responses yielding more positive ERPs than button responses), but not in parallel hands.
Our interpretation is as follows: the fact that the activation of the non-executive hemisphere does not differ in parallel-hand responses suggests that this hemisphere does not contribute to response monitoring/process spatial information in neither button nor lever response trials. This suggests that the location of physiological effectors (the hands which stayed within their "natural" hemifield) accounts for the lateralization of response monitoring processes and that the physical effector (the location of the button) does not. The non-executive hemisphere difference between buttons and levers in crossed hands is not strictly in line with the assumption that only the hands are responsible for the hemispheric allocation of spatial information during response monitoring. Yet, it is unlikely that the physical effector (button) plays a major role in the allocation of response monitoring processes. The reason for this is that based on the explanations above, one would expect a "reversal" of parallel and crossed non-executive hemisphere ERPs across the button types. In case of an allocation based on the location of the physical effector, lever responses should produce a positive peak in crossed hands and a negative peak in parallel hands (crossed > parallel) over the non-executive hemisphere. This criterion is not fulfilled since both in button and in lever responses; parallel hands yield a more positive ERP than crossed hands (see Figure 1). Because of the different polarity of ERP peaks around the time of the response, we based the statistical analysis on mean activity measures. While these measures can depict differences between the epochs over which the ERP data was averaged, they unfortunately cannot account for the course of the curves within these epochs. Yet, we obtained no convincing statistical results in favor of a physical effector approach and the grand averages (Figure 3) further support the assumption that the physiologic effector (hand) determines the allocation of spatial response monitoring processes: despite the detected differences, the course of the ERP curves of crossed-hand lever responses is very similar to that of crossed-hand button responses while both crossedhand conditions markedly differ from the course of parallel-hand responses.
Furthermore, the behavioral results of this study are line with previous findings on this paradigm (e.g., Wiegand and Wascher, 2005;Leuthold, 2011) suggesting that the task was correctly implemented/worked as intended. Both hit rates and RTs were modulated by the hand position as well as the spatial congruency of the stimulus presentation site and the location of the responding hand. In all significant main effects and interactions, parallel-hand responses yielded a better (more accurate/faster) performance than crossed-hand responses and congruent trials yielded better results than incongruent trials. Matching results were obtained for the stimulus-locked ERLs/N2pc. As expected, the ERLs showed an interaction of hand position and congruency (see Praamstra and Oostenveld, 2003;Wiegand and Wascher, 2005;Böckler et al., 2011;Leuthold, 2011;Verleger et al., 2012). For the ERLs, there was no effect of button type whatsoever. Since stimulus-response congruency had been defined with respect to the location of the hand (not the button), this finding clearly indicates that external/physical effectors do not seem to have an influence on congruency and on early attentional processing and/or filtering (Luck and Hillyard, 1994;Böckler et al., 2011;Leuthold, 2011;Verleger et al., 2012).
From this study, it can be concluded that the spatial location of physiologic effectors (in our case, this would be the hands) plays a major role in the asymmetrical allocation of response monitoring processes: whenever the physiologic effectors enter a "foreign" hemifield, the hemisphere contralateral to this hemifield seems to handle information on spatial aspects of the response. By comparison, the location of the physical effector (in our case, this would be the buttons) plays a minor role. Yet, the possibility that it still contributes to response monitoring processes cannot be ruled out completely. Furthermore, these findings allow for the conclusion that potentially different action goals of button and lever responses do not substantially influence the lateralized allocation of response monitoring processes (compare to Buhlmann et al., 2007). Our study extends the established fact that each hand operates "in its own egocentric space" (Haggard et al., 2000) by demonstrating that egocentric space continues to play a role in the subsequent processes of response monitoring and evaluation. Also, our results are in line with the findings that proprioceptive (Allain et al., 2004) and internal sensorimotor information is used for response evaluation (Fukui and Gomi, 2012) and that each hemisphere preferentially processes information from the contralateral hemifield (Zhou et al., 2012). | 5,765.2 | 2014-01-24T00:00:00.000 | [
"Biology"
] |
Dengue virus replication enhances labile zinc pools by modulation of ZIP8
Abstract Zinc‐dependent viral proteins rely on intracellular zinc homeostasis for successful completion of infectious life‐cycle. Here, we report that the intracellular labile zinc levels were elevated at early stages of dengue virus (DENV) infection in hepatic cells and this increase in free zinc was abolished in cells infected with UV‐inactivated virus or with a DENV replication inhibitor implicating a role for zinc homeostasis in viral RNA replication. This change in free zinc was mediated by zinc transporter, ZIP8, as siRNA‐mediated knockdown of ZIP8 resulted in abrogation of increase in free zinc levels leading to significant reduction in DENV titers suggesting a crucial role for ZIP8 in early stages of DENV replication. Furthermore, elevated free zinc levels correlated with high copy numbers of dengue genome in peripheral blood leukocytes obtained from dengue patients compared to healthy controls suggesting a critical role for zinc homeostasis in dengue infection. Take Aways Dengue virus utilises cellular zinc homeostasis during replication of its RNA. ZIP8 upregulates free zinc levels during dengue virus replication. Enhanced viremia associates with elevated intracellular free zinc in dengue.
| INTRODUCTION
Zinc is an essential micronutrient and an important structural constituent of various proteins such as transcription factors, enzymes, growth factors, cytokines and receptors involved in cellular signalling cascade (Fukada, Yamasaki, Nishida, Murakami, & Hirano, 2011). It is the second most abundant transition metal in cellular organism after iron and is an integral part of about 10% of all human proteins (Andreini, Bertini, & Cavallaro, 2011). The total cellular zinc content is in the range of hundreds of micromolar which is mainly constituted by protein-bound form of zinc. The labile (free) zinc pool in cytosol, which is more accessible, is of significant interest as its exchange between the cytosol and subcellular compartments by zinc transporters regulates physiological functions (Frederickson, Koh, & Bush, 2005). This exchange is mediated majorly through two protein families, Zrt, Irt-like proteins (ZIP) and zinc transporters (ZnT). ZIPs are responsible for importing zinc inside the cytosol from extracellular space or from the intracellular compartments while ZnTs transport Aleksha Panwar and Jigme Wangchuk contributed equally to this work. zinc out of the cytosol into intracellular organelles or into the extracellular space (Kambe, Hashimoto, & Fujimoto, 2014;Maret, 2017). Zinc homeostasis is modulated by external stimuli and redistribution of zinc into tissue spaces as part of acute-phase response may also limit or enhance the availability of zinc which is required for the survival and propagation of the invading pathogen depending on its tissue tropism.
Although the cellular level of free zinc is of several orders of magnitude lesser compared to the total zinc or the protein-bound form (Maret, 2013), they represent the zinc pool which is readily available and involved in cellular metabolism.
We further confirmed this in Huh-7 cells by infecting cells with DENV at 3 MOI and 1 hr after virus adsoprtion, cells were washed and serum-free medium containing DMSO or 0.5 μM TPEN was added and cultured for 24 hr. Viral titers in the supernatant were measured by plaque assay at 24 hpi. Consistent with our previous findings, we observed that zinc depletion by TPEN led to a significant reduction in DENV titers indicating the importance of zinc in DENV life-cycle ( Figure 1a). Similarly, we performed zinc supplementation experiments, where cells were infected as above and serum-free medium containing 100 μM of ZnSO4 was added after virus adsorption and cultured for 24 hr and virus titers were measured by plaque assay at 24 hpi. As observed earlier in epithelial cells (Kar et al., 2019;Khan et al., 2020), supplementing media with excess zinc did not have any effect suggesting that unlike other viruses which are susceptible to inhibition by zinc (Suara & Crowe, 2004;Yuasa et al., 2006;Zhang et al., 1991), DENV infection is not perturbed by excess zinc ( Figure 1b). Since DENV replication was sensitive to perturbation in zinc homeostasis and labile zinc pool acts as an indicator of change in zinc homeostasis, we next sought to determine whether DENV infection leads to modulation of zinc homeostasis. Labile zinc levels were visualised by confocal microscopy using zinc fluorophore, fluozin-3 a.m. (FLZ-3 a.m.), in Huh-7 cells infected with DENV at 8 and 16 hpi.
We observed a two-fold increase in labile zinc levels in DENV infection at 16 hpi (Figure 1c(i),(ii)). The labile zinc levels at 24 hr were comparable with the mock-infected samples further suggesting that the transient nature of this induction is most likely to meet the demand of viral RNA replication (Figure 1c(i),(ii). The increase in labile zinc levels were found to be MOI-dependent as we observed $1.5-fold and $3-fold increase at 0.3 and 3 MOI, respectively, which coincided with exponential phase of virus replication as estimated by qRT-PCR and plaque assay respectively (Figure S1a-c). Further, we verified the above observations in primary human hepatocytes which were infected with DENV at 5 MOI and stained with FLZ-3 a.m. at 16 hpi. Similar to Huh-7 cells, primary hepatocytes showed elevated free zinc levels upon DENV infection (Figure 1d(i),(ii)). We observed $80-90% infection under these conditions in these primary cells which was similar to what we observe with 3 MOI in Huh-7 cells ( Figure S1d). These data suggest that DENV infection modulates zinc homeostasis to meet the demand for zinc in viral replication by inducing elevation in cytosolic free zinc levels in cells of hepatic origin. To verify this, we infected cells with UV-inactivated DENV and measured labile zinc levels at 16 hpi. We found that UVinactivated virus did not lead to increase in zinc levels as efficiently as an actively replicating virus (Figure 1e(i), (ii)). We next used fluoxetine hydrochloride (FLX), which we have shown earlier to inhibit DENV replication (Medigeshi, Kumar, Dhamija, Agrawal, & Kar, 2016). We observed no change in zinc levels when cells were treated with FLX post-infection ( Figure 1f). To further validate this observation, we used BHK-21 cells stably expressing a DENV-2 replicon ( Figure S1e) (Boonyasuppayakorn, Reichert, Manzano, Nagarajan, & Padmanabhan, 2014). Mock BHK-21 cells and cells stably expressing DENV-2 replicon were stained with FLZ-3 a.m.
to visualise labile zinc levels. (Figure 1g). Similar to virus infection in Huh-7 cells, we observed a significant increase in labile zinc levels in cells expressing the DENV replicon suggesting that active replication of DENV leads to increase in labile zinc pools in the cytosol.
We next sought to identify the mechanism of increase in labile zinc levels in dengue infection. Free zinc levels in the cytoplasm may increase due to enhanced cellular uptake of zinc or due to redistribution of zinc from intracellular organelles to cytosol or by the release of zinc from metallothioneins which act as zinc stores (Kimura & Kambe, 2016). Huh-7 cells were infected with DENV and at 16 hpi, total cell extracts were subjected to inductively coupled-plasma mass spectrometry (ICP-MS) for the detection of total zinc (Zn) content.
The amount of zinc ions detected by ICP-MS in parts per billion (ppb) was normalised to the total protein content of the cells to account for any difference in cell numbers. We did not observe any change in the total zinc content in DENV infection conditions ( Figure 2a). Flavivirus replication utilises cytosolic membranous compartments derived from the ER and Golgi (Mackenzie, 2005). Therefore, we were interested to examine whether this free zinc pool is localised to flavivirus replication compartments. However, localization of free zinc with viral proteins and viral replication compartment was not possible as permeabilization of cells led to loss of FLZ-3 a.m. signal. Therefore, we transfected plasmid constructs coding for mCherry-tagged ER or Golgi-resident proteins in Huh-7 cells followed by infection with DENV after 24 hr and visualised free zinc levels by confocal microscopy at 16 hpi. We consistently observed higher levels of free zinc in infection conditions which concentrated in the perinuclear region as We speculated that the increase in cytosolic zinc induced by dengue virus infection could be regulated by zinc transporters. To test F I G U R E 1 DENV infection leads to increase in cytosolic free zinc levels. Huh-7 cells were infected with DENV at 3 MOI and after viral adsorption, cells were cultured in either 0.5 μM TPEN (a) or 100 μM ZnSO 4 (b) and viral titers in supernatants was estimated by plaque assay at 24 hpi. (n = 9 for both [a] and [b]) (c) Huh-7 cells were infected with DENV and free zinc levels were determined by immunofluorescence assay at indicated time points. (i) Representative images of cells stained with FLZ-3 a.m., (ii) Sum grey intensity of FLZ-3 stain was evaluated using cellSens software. Data represent relative change in sum grey intensity of FLZ-3 a.m. stain in cells (n = 95-115) counted from eight different fields. (d) Primary hepatocytes were infected with 5 MOI of DENV and free zinc levels were assessed by immunofluorescence assay at 16 hpi. (i) Representative images of cells stained with FLZ-3 a.m., (ii) Data represent relative change in sum grey intensity of FLZ-3 a.m. stain in cells (n = 15-20) counted from seven different fields. (e) Huh-7 cells were either mock-infected or with 3 MOI of DENV or UV-inactivated virus. Cells were stained with FLZ-3 staining at 16 hpi. (i) Representative images (upper panel) and DENV replication was assessed by dsRNA staining using antibody against dsRNA (lower panel), (ii) Data represent relative change in sum grey intensity of FLZ-3 a.m. stain in cells (n = 80-100) counted from seven different fields. (f) Huh-7 cells were infected with DENV and fluoxetine hydrochloride (FLX) was added at a concentration of 4 μM after virus adsorption. Data represent fold change in sum grey intensity of FLZ-3 a.m. stain in cells (n = 80-100) counted from eight different fields. (g) Free zinc levels were determined in BHK-21 cells stably expressing DENV-2 replicon and mock BHK-21 cells using FLZ-3 a.m. stain. Representative images are shown. All the data presented here are from two or three independent experiments. Data are presented as M ± SD. Scale bar is 10 μM. ns, not significant. **p < .01; ***p < .001 (King et al., 1999;Schmid, Diamond, & Harris, 2014;Valero et al., 2014). Based on our in vitro data, we next probed whether blood cells from dengue patients show an increase in labile zinc levels. We recruited 21 paediatric dengue patients who were in the viremic phase (within 7 days of fever) into the study to collect blood samples and analyse for free zinc levels in peripheral blood cells by flow cytometry. The clinical features of these patients have been reported recently as part of another study (Khan et al., 2021). As per WHO classification guidelines, 8 patients had mild dengue, 7 had dengue with warning signs and 6 were with severe dengue. Total RNA was isolated from whole blood and processed for estimation of viral RNA copy numbers by qRT-PCR as described previously (Kar et al., 2017;Khan et al., 2021;Singla et al., 2016). We also
| EXPERIMENTAL PROCEDURES
List of reagents used in the study is provided in Table S1. Screening and enrolment of patients were exactly as described previously (Kar et al., 2017;Khan et al., 2021;Singla et al., 2016). Written informed consent for the study was taken from parents/guardians to collect blood samples at the time of admission. Children aged between 4-14 years with symptoms suggestive of dengue were screened using a
| Detection of free zinc levels by confocal microscopy
FluoZin-3 a.m. (FLZ-3 a.m.) (Invitrogen) was used to determine the intracellular free zinc levels. Medium was removed and cells were washed twice with 1X PBS followed by incubation with FLZ-3 a.m. at a concentration of 10 μM. Cells were incubated for 20 min at 37 C.
After incubation, cells were washed three times with 1X PBS and stained with 4 0 ,6-diamidino-2-phenylindole (DAPI) at a dilution of 1:10,000 in PBS for 5 min. Cells were fixed using 3% paraformaldehyde (PFA) for 10 mins followed by washing with 1X PBS three times and mounted using Prolong Gold antifade reagent (Invitrogen). Images were captured using FV3000 confocal microscope (Olympus). Fluorescence images with FLZ-3 a.m. were captured using an excitation wavelength of 488 nm. The parameters for detection and capturing images were digitally controlled to keep same settings throughout the experiments. For quantitative analysis, sum grey intensity per cell was calculated using CellSens software (Olympus) and plotted as relative change.
For zinc supplementation experiment, 100 μM of ZnSO 4 was added in serum-free media after viral adsorption and viral titers were measured at 24 hpi. To study the effect of fluoxetine hydrochloride on DENVinduced free zinc levels, Huh-7 cells were infected with DENV and fluoxetine hydrochloride was added after viral adsorption at a concentration of 4 μM in 2% DMEM. Cells were incubated for 16 hpi and processed for FLZ-3 a.m. staining. Coverslips were mounted on glass slides, sealed and images were captured at 100X using FV3000 confocal microscope (Olympus).
| Transient transfections
Huh-7 cells were seeded at 60,000 cells per well in a 24-well plate on glass coverslips. 24 hr later, cells were washed with 1X PBS, and transfected with 800 ng of plasmids expressing mCherry-tagged endoplasmic reticulum (Addgene-55,041) and Golgi (Addgene-55,052) markers in serum-free media using Lipofectamine2000 (Invitrogen) as per the manufacturers' instructions. After 6 hr incubation period, the transfection mixtures were removed and replaced with fresh growth medium. After 24 hr post transfection, cells were infected with DENV at 3 MOI. At 16 hpi, cells were stained with FLZ-3 a.m. and processed for confocal microscopy as described in the previous sections.
| Viral dsRNA staining
At 16 hpi, cells were washed with cold PBS and fixed in ice-cold methanol at À20 C for ≥20 min. Cells were washed twice with PBS followed by incubation in 0.2% BSA in IMF buffer (20 mM HEPES pH 7.5, 0.1% Triton X-100, 150 mM NaCl, 5 mM EDTA, 0.02% sodium azide) for 1 hr at room temperature (RT). Cells were then incu- Signals were detected using a gel documentation system (Azure biosystems C400).
| Ultraviolet (UV)-inactivation of DENV
DENV inactivation was carried out using a UV crosslinker (UVP-CL1000S). DENV stock was diluted in a ratio of 1:100 in 1 ml of MEM media and placed in a 35 mm diameter petri dish. DENV stock was treated with a dose of UV light at 9999 Â 10 2 μJ/cm 2 . The supernatant was exposed three times for 10 min with an interval of 3 min.
After the inactivation process, virus stock was aliquoted and stored at À80 C, later titered and used for the experiment. UV-inactivated virus showed no replication as observed by viral dsRNA staining.
| siRNA knockdown
Smartpool siRNAs targeting the human ZIP1 and ZIP8 genes or nontargeting controls (NTC) were purchased from Dharmacon. Transfections were carried out as described previously . All transfections were performed as per manufacturer's instructions.
Briefly, 10 nM concentrations of NTC, ZIP1 and ZIP8 siRNAs were mixed with Opti-MEM (Life Technologies) and 1 μl of Lipofectamine RNAiMax to a total volume of 100 μl in a 24-well plate. Cells were trypsinized and volume made up so as to contain 30,000 cells in 400 μl antibiotic-free medium. After 20 min incubation of the transfection complex, cell suspension was added into each well. Knockdown efficiency was determined by qRT-PCR at 48 hr post transfection.
| Quantitative real time PCR (qRT-PCR)
Huh-7 cells were infected with DENV and at indicated time points, cells were collected in TRIzoL reagent (Takara) and RNA was isolated using manufacturer's instructions. cDNA synthesis was performed using PrimeScript RT reagent kit with gDNA eraser (Takara). 100 ng of cDNA was used to determine genes expression using DyNAmo flash SYBR green quantitative PCR reagent (Thermo Scientific). Reaction conditions used were as follows: (95 C-7 min; 95 C-10 s followed by 60 C for 30 s). GAPDH primer was used as housekeeping control. For DENV RNA detection by reverse-transcription polymerase chain reaction (RT-PCR), total RNA was extracted from cells at the indicated time points using RNAiso Plus (TaKaRa), and 200 ng of RNA was used in multiplex TaqMan one-step RT-PCR with DENV primers, DENV probe and human GAPDH primer-probe mix (Applied Biosystems). At indicated time points, supernatant was collected for estimating viral titers by plaque assay and cells were harvested for positive and negative strand detection PCR as described previously (Kar et al., 2017).
Expression levels of GAPDH was used to calculate fold change and normalisation. Data were analysed using the ΔΔ C T method, where C T is threshold cycle.
| Data analysis
Data were analysed and graphs were prepared using Prism 7 (Version 7.0e) software (GraphPad Software Inc.). All the graphs represent results from two or more independent experiments; values are presented as M ± SD. Statistical significance was estimated by Mann-Whitney test. The data were corrected for multiple comparisons using Bonferroni-Dunn method wherever applicable.
ACKNOWLEDGMENTS
We thank all the members of the CCV lab for their support and critical inputs. We thank all the patients for providing their samples for the study. This work was supported by DBT/Wellcome Trust India Alliance Fellowship [grant number IA/S/14/1/501291] awarded to GRM.
The funders had no role in study design, data collection and analysis, decision to publish or preparation of the manuscript.
CONFLICT OF INTEREST
The authors declare no conflict of interest.
DATA AVAILABILITY STATEMENT
All the data presented in the manuscript is available as part of this manuscript and as supplementary information. | 4,054.4 | 2021-10-07T00:00:00.000 | [
"Biology",
"Medicine",
"Chemistry"
] |
Porous Selective Laser Melted Ti and Ti6Al4V Materials for Medical Applications
This chapter characterises scaffolds manufactured in line with the make-to-order concept according to individual needs of each patient. The clinical data acquired from a patient during computer tomography, nuclear magnetic resonance or using traditional plaster casts is converted by a computer into a virtual solid model of a patient’s loss. The model, through the multiplication of a unit cell, is converted into a porous model on the basis of which an actual object is manufactured with the method of selective laser melting (SLM) from Ti/Ti6Al4V powders. The created scaffold is characterised by good mechanical properties, which is confirmed by the results of the performed tensile and compressive strength tests. The material is additionally subjected to surface treatment consisting of the deposition of atomic layers of titanium dioxide with nanometric thickness.
Introduction
Additive manufacturing technologies are proliferating and becoming increasingly popular across various industries nowadays. Owing to their benefits, they support the development of multiple disciplines in the field of engineering wherever a clear need exists to fabricate elements with a complicated shape, geometry and a distinctive structure. Such elements include individualised implants, representing living tissues, closely accommodated to a given patient based on the results of computer tomography or nuclear magnetic resonance or traditional plaster casts.
Porous structures are much desired in medicine, especially where a porous element is to replace a missing bone. In such a situation, the task of the fabricated element is to stimulate a regeneration process of the adjacent bone tissue through an osteoconductive and osteoinductive activity. Osteoconduction is a process of bone loss regeneration consisting of the in-growth of osteoblasts-which are bone-forming cells originating from the adjoining bone stock-into the porous implant. In osteoinduction, though, the differentiation of mesenchymal cells is stimulated in the environment of osteoblasts. The cells represent a connective embryonic tissue occurring only in the embryonic period, from which all types of connective tissues, bone tissue, cartilage tissue and muscle tissue are created. In order for titanium scaffolds fabricated with SLS technologies to fulfil their role well in a patient's organism, they should be characterised by the appropriate size of pores, appropriate porosity, as well as strength permitting usage in bone implants functioning as scaffolds, which become a substructure and a support for the bone growing into them. Literature data shows [1][2][3][4][5][6][7][8][9][10][11][12] that the size of pores allowing the development of the bone growth process into the created scaffold, varies between the minimum of 50-200 µm and the maximum of 500 µm and the porosity of such a scaffold should not exceed 50%. A satisfactory result of a manufacturing process of porous titanium materials is seen when an element is achieved with open pores, characterised by an appropriate level of porosity and sufficiently good strength properties, which should be similar to the corresponding properties of a living bone tissue. Bone porosity is referred to as the volume fraction of the fluid phase filling the pore space of a bone in a given bone volume unit. The bone fluid phase consists of blood vessels together with blood, nerve fibres, bone cells and extracellular bone fluid. A cortical bone has the density of 1.99 g/cm 3 . It has the longitudinal compressive strength of 131-224 MPa and the longitudinal bending strength of 79-151 MPa. The transverse compressive strength of a cortical bone is 106-133 MPa and bending strength is 51-56 MPa [13,14].
Additive manufacturing enables to create objects with the final shape while allowing to control the manufactured element in each part of its volume. An advantage of AM technologies, as compared to competitive fabrication methods of porous materials with space fillers [10][11][12], is also an ability to fabricate without applying costly and time-intensive casting moulds, as a result of which the product achieved does not contain external admixtures which are often present in cast products. The maximum possible reduction of wastes generated in a manufacturing process as compared to waste-generating machining is ranking AM technologies higher than traditional manufacturing processes applied until now. The entire additive manufacturing process takes place in an atmosphere of inert gas, which prevents the creation of unwanted products of the reactions occurring between a material used for fabrication of elements and air components. Selective laser melting (SLM) technology, due to its advantages, is very well suitable for manufacturing in line with the make-to-order concept of individualised craniofacial implants, including palate implants.
Computer-aided materials design
An additive manufacturing process carried out with the SLM technique begins at the sage of computer design of 3D models of objects planned to be manufactured in reality. The key benefit of the concept is that no constraints exist as to the shape and filling of a virtual model at the design stage; it allows representing an anatomical structure of organs, bones and human and animal tissues very accurately [15][16][17][18][19][20]. One of the specific aspects in this domain, being the subject of a series of experiments performed by the authors of this chapter, is the designing and manufacturing of porous biomimetic implants replacing a patient's palate loss. The loss may result from a genetic defect or a mechanical injury, or a neoplasm. A palate implant designed individually for each patient is a scaffold, the structure of which is made of open pores. A porous structure is to ensure appropriate growth conditions on the surface of pores for living cells by ensuring availability of nutrients. The computer-aided design of virtual models is preceded by a clinical stage at which accurate data is acquired from a patient regarding a 3D shape and dimensions of one's palate loss. The data can be acquired in the course of examinations conducted with computer tomography or nuclear magnetic resonance, or with traditional plaster casts, the shape of which is then transferred to a computer with a 3D scanner. Regardless the option chosen, the outcome of such activities is a virtual solid implant model of a palate which is further processed into a porous model.
A unit cell which, when multiplied many times, will create a porous scaffold structure, is required to create a virtual porous palate implant. A unit cell can be chosen from the avail- able database being part of commercial software or it can be individually designed according to individual preferences. When designing a unit cell, it is significant to model it in such a way that it is symmetrical relative to all the axes of symmetry, which guarantees the correct transformation of a virtual solid model into a porous model, without errors on the created lattice. Unit cells according to a custom design are shown in Figure 1. All the designed unit cells inscribe themselves into a cube with the side of 500 µm. A type A cell (Figure 1a) has the shape of a spatial cross with arms with the dimensions of 100 × 100 µm. A type B cell ( Figure 1b) is a cross with its arms 100 µm thick and has small quadrangle openings in such arms with the side of 100 µm. A type C cell (Figure 1c) is a cube with the side of 300 µm, on each wall of which tongues are arranged symmetrically in the form of smaller cubes with the side of 100 µm. A type D cell (Figure 1d) is a skeleton of a cube with the sides equal to 100 µm. A solid implant model and the applied unit cell are saved in stl files before a virtual solid model of a piece of a palate loss is transformed into a virtual porous model. The extension is needed so that the model is transferred to a device carrying out a selective laser sintering (SLM) process. This ensures the automatic conversion of a virtual scaffold model into a model made of the net of triangles with the standard size of the area of a single triangle creating a net which is predefined directly by CAD software. It is possible to control the size of the area of a single triangle to densify the net, thus to control the accuracy of surface representation of an element produced having a complicated shape. If the field of a single triangle is decreased when converting the model into a net of triangles, accuracy is increased, but also the process is extended, both, at the stage of multiplication as well as at the stage of object fabrication. Highly efficient, latest generation software must therefore be used for very high accuracies. A virtual porous model created after transformation with a specifically designed structure composed of multiplied unit cells, has geometrical dimensions perfectly corresponding to the model from before the transformation (Figure 2).
Selective laser melting of scaffolds
Pristine titanium and Ti6Al4V titanium alloy, which, as per the international standard [21], is dedicated to biomedical applications, is a material used for the fabrication of actual objects which are to act as individualised palate implants. Titanium is classified as a light metal with a density of ρ = 4507 g/cm 3 , a melting point of 1668°C and boiling point of 3260°C. It occurs in two allotropic variants, i.e. α and β. The α titanium variant has a hexagonal lattice (A3), which, at the temperature of 882.5°C, is converted into a high-temperature variant β crystallising in the regular system (A2). Pristine titanium occurs in the single-phase variant α and the titanium alloy Ti6Al4V in the double-phase variant α and β [22][23][24]. Titanium possesses high mechanical strength relative to the mass; however, its essential properties, allowing to employ it in medicine, include high biocompatibility with living tissues and pitting corrosion resistance, intercrystalline corrosion resistance and stress corrosion resistance. An input material in the form of powder must be used for the selective laser melting technology. Pristine titanium powder with the grain size of 0-45 µm and titanium Ti6Al4V alloy powder with the grain size of 15-45 µm, with the chemical composition consistent with the manufacturer's specification shown in Table 1, were used in the course of the experiments carried out to fabricate actual objects. The results of the qualitative chemical composition analysis performed with the EDS method (Figure 3) confirm the presence of the elements stated by a manufacturer in technical and commercial specifications. Powder grains in the both cases are spherically shaped, which is observable in microscopic photos (Figure 4).
Pristine titanium and titanium Ti6Al4V alloy powders are an input material from which porous implants for medical uses are manufactured in a selective laser sintering process. Preparatory measures are taken before the actual manufacturing process, during which, after transferring a 3D virtual scaffold model to an SLM device, the virtual object is placed in a working chamber in the adequate position, i.e. on the relevant edge and under the right angle. The selection of the adequate virtual model position in a working chamber is an important aspect due to the following factors: the amount of powder needed to carry out a manufacturing process once, the mechanical properties of the scaffolds produced [25] and the number of supports needed to generate scaffolds. The purpose of the supports is to support the produced actual object and to secure it against collapse under its own weight. In the next stage, the virtual model is divided into layers parallel to the working platform surface of the device on which it is to be manufactured. The number of layers depends on the parameter set by the operator, i.e. powder layer thickness to be given before each melting. The following manufacturing conditions are also selected prior to commencing a fabrication process of an actual object: laser power, scanning rate, distance between consecutive remelting paths and the laser beam diameter. Before starting the selective laser melting process, titanium powder is heated in a vacuum in the surrounding of shielding gas at an elevated temperature (160-200°C) to remove any moisture from the powder, as necessary.
The actual selective laser melting process serves to produce metallic scaffolds encompasses the selective melting of powders point after point and layer after layer using a high-power laser [30][31][32][33][34] with a device the diagram of which is shown in Figure 5. A process of object manufacturing by SLM was carried out from the bottom, i.e. from the working platform side. Each layer produced is adhering to the preceding one until the process is completed [35,36]. Powder is fed from a magazine holding loose material and is then distributed with a specific quantity with a shaft travelling across a working platform, which is descending by the exact height of the layer being sintered, whose thickness corresponds to one layer of virtual 3D model section. The excess powder is collected with a roll to a second empty magazine. A computer-controlled laser beam is melting the powder (Figure 6) in a specifically predefined manner and in selectively picked points. A powder layer is deposited and melted selectively in an alternate fashion, until the entire, permanently integrated real object is created. The excess powder, removed from a working platform, can be re-used, by collecting it into a separate special magazine, after being finely sifted in subsequent fabrication processes [37,38]. An example of a scaffold observed with a bare eye, whose size matches the palate loss of one of the patients participating in clinical trials, is shown in Figure 7, showing, respectively, view from the top (Figure 7a) and from the bottom (Figure 7b). This scaffold, manufactured using Ti6Al4V titanium alloy powder, was removed mechanically from the working platform and supports were removed. The scaffolds created were subjected to microscopic observations using a scanning electron microscope. It was observed that the surface topography of the created scaffolds shows a porous, regular latticework-shaped structure. It was also found that the pores of the scaffolds produced are open, which was one of the designers' key assumptions due to the fact that this material, acting as an implant, is to grow through a patient's living tissue. Microscope observations of the studied material's surface topography indicate also the presence of singular, spherically shaped powder grains on its surface, which were deposited there due to adhering to the scaffold surface remelted in an SLM process. Figure 8 shows a surface topography of the scaffolds manufactured with Ti6Al4V powder with the SLM method using the pre-produced virtual models comprised of multiplied unit cells created by the author, of, respectively, type A (Figure 8a), type B (Figure 8b), type C ( Figure 8c) and type D (Figure 8d).
Mechanical properties of scaffolds
On the one hand, scaffolds are made of robust materials such as pristine titanium and Ti6Al4V titanium alloy and on the other hand, they have a distinct structure consisting of open pores and possess interesting mechanical properties such as tensile and compressive strength. The tensile strength of material is determined according to the dependency (Eq. (1)) [39], using for calculations the results of strength tests determining the maximum tensile strength and a known field area of the sample cross section. The dependency (Eq. (2)) [39] allows to calculate the compressive strength of material using data such as the maximum compressive strength values obtained during strength tests and the sample cross section. The shape and dimensions of the samples designed to perform tensile and compressive strength tests are shown in Figure 9.
where: The strength properties of scaffolds depend on the size of their pores, laser path curve and unit cell arrangement in the space of a system of coordinates [25]. Table 2 shows the results of experiments made after selecting the optimum conditions of executing a manufacturing process, such as the size of scaffold pores of 250 µm, a laser path with an improved curve and unit cell arrangement at the angle of 45° relative to the axis of abscissa of the system of coordinates.
The average tensile strength value of laser-sintered scaffolds of Ti6Al4V powder is 47 MPa and is over 30% higher than the strength of scaffolds produced in the same conditions using pristine titanium powder, as presented in Figure 10. The characteristic of progression of tensile curves proves that both the porous titanium and porous Ti6Al4V titanium alloy are elastic-plastic materials with a clearly marked elastic strength and yield strength. Similar as in the case of tensile strength, a scaffold made of titanium alloy possesses much higher compressive strength than a titanium scaffold. The difference is much higher, though, because the compressive strength of the material made of Ti6Al4V is 225 MPa and is 80% higher than the compressive strength of a titanium sample (120 MPa). Figure 11 shows charts presenting a dependency between compressive stress and deformation for the porous materials sintered with a laser using, respectively, Ti and Ti6Al4V powders.
Atomic layers of dioxide titanium deposited onto scaffolds' surface
Pristine titanium and its Ti6Al4V alloy, of which scaffolds are made which are to act as implants of palate fragments, are the materials broadly used in medicine as implants due to their low density, a beneficial strength-to-yield stress limit ratio, good corrosive resistance and biocompatibility. A further improvement in those materials' properties such as biotolerance and osteoconduction is possible by employing surface treatment. Thin layers are depos- Figure 11. Dependency between compressive stress and deformation recorded for porous materials sintered with laser from Ti and Ti6Al4V powders. ited permanently onto the surface of implants made of metallic biomaterials intended for long use in a human organism, most often with the following methods [40][41][42][43]: plasma sputtering, electrophoresis, physical vapour deposition (PVD) and chemical vapour deposition (CVD), sputter coating and electrochemical deposition. Where layers are deposited onto the surface of porous biomaterials with complicated shapes, it is very important to be able to accurately control the growth mechanisms enabling to constitute a very thin layer with its thickness measured at a nanoscale, but most of all, it is essential to be able to deposit geometrically complex areas uniformly. One method, i.e. atomic layer deposition (ALD), gives such an opportunity now, as shown in Figure 12a, against optional methods. The ALD method is a variant of the CVD method characterised by the cyclic use of alternate precursor pulses with strong reactivity with a chamber purged with inert gas between such pulses (Figure 12b). By applying strongly reactive precursors, which-after supplying them into a chamber-are reacting immediately with a substrate by forming a monolayer and preventing a further reaction, each cycle increases the layer thickness by a strictly specified value within the range of 0.01-0. The ALD technique enables to deposit a chosen chemical compound more uniformly across the entire surface of the part being treated, also if this part has a porous structure, as is the case with scaffolds. The thickness of the layers deposited by ALD is determined with a spectroscope ellipsometer equipped with special software. The average thickness of layers deposited by ALD technique for the said cases of 550, 1050 and 1550 cycles is 55.95, 98.90 and 148.73 nm, respectively. The difference in the thickness of the deposited TiO 2 layers on the studied area does not exceed 2 nm, which can be analysed in detail by studying layer thickness distribution maps. The best results were obtained for a layer deposited in 1050 cycles. A difference in the thickness of the deposited layer in this case does not exceed 1.1 nm across the entire area of the surface-treated item. Bar charts for each number of cycles, presenting the thickness of the deposited TiO 2 layer in the particular measuring points and the corresponding layer thickness distribution maps, are shown in Figure 13.
Changes in sample colour depending on the number of the executed ALD cycles, hence depending on the thickness of the deposited TiO 2 layer, is an interesting phenomenon observed with a bare eye (Figure 14) and in a light stereoscopic microscope (Figure 15). The uncoated element, with silver-metallic colour, undergoing surface treatment with ALD becomes, successively: brown-gold (550 cycles), dark blue (1050 cycles) and light blue with silver shade (1550 cycles).
The topography of scaffolds' surface coated with layers deposited by ALD is distinct for their irregularities measured at a nanometric scale, the number of which is rising proportionally to the number of the deposited layers. In particular, a layer deposited in 550 cycles has a rather uniform granular structure and the larger clusters of atoms are occurring on it only occasionally. In the case of a layer deposited in 1050 cycles, clusters of atoms with the diameter of about 1 µm occur every several microns. The biggest clusters of atoms, forming 'islands' with the length of up to several microns, exist in the case of a layer deposited in 1550 cycles, as shown in Figure 16.
The nanometric thickness of titanium dioxide layers deposited by ALD results in the fact that the layers deposited can be observed in a scanning electron microscope only for very high
Conclusions
Currently, there is a high social demand for individualised implants, which would significantly improve the quality of life of patients with partial palate losses caused by mechanical injuries, tumorous diseases or cleft palate. The methods currently in use, such as metallic or polymeric prostheses, do not meet users' expectations as they often lack durability, convenience and aesthetics. A porous scaffold with its dimensions and shape perfectly suited to a patient plate loss made of a biocompatible material (Ti or Ti6AlV4) and additionally coated with a nanometric layer of osteoconductive titanium oxide seems to be a breakthrough solution. Modern CAMD software allows converting data acquired at a clinical stage into a 3D solid model of a palate loss piece. The model is then converted into a porous model through the multiplication of a unit cell whose dimensions and shape may be designed according to a patient's individual preferences. The pores existing in the material structure have the diameter of approx. 500 µm and should be open, because a scaffold, in its intended conditions of use, is to grow through a patient's living tissue. The experiments made confirm that the selective laser melting technology allows, following process conditions' optimisation, to produce biomimetic objects with a structure featuring open pores, as this was confirmed in microscopic (SEM) examinations. The objects have sufficiently good mechanical properties such as bending and compressive strength, which are similar to the properties of a cortical bone. The biotolerance and osteoconduction of laser-sintered scaffolds can be additionally improved through surface treatment allowing to cover a complex geometrical surface, such as a porous scaffold structure, uniformly from all sides. The treatment is carried out by the deposition of atomic layers and the deposited layer is 50-150 nm thick depending on the number of cycles. | 5,251.4 | 2017-03-29T00:00:00.000 | [
"Materials Science"
] |
Selective Leaching of LiFePO4 by H2SO4 in the Presence of NaClO3
Abstract: Herein, problems commonly observed for the wet leaching of waste LiFePO4 cathode materials, namely extensive Fe leaching and impurity removal, are mitigated through the use of NaClO3, and the effects of leaching parameters on Fe, P, Li, and Al leaching efficiencies are probed. As a result, optimal leaching conditions are determined as temperature = 90°C, H2SO4 concentration = 1 M, liquid-to-solid ratio = 5:1, leaching time = 1 h, stirring speed = 150 rpm, and NaClO3 dosage = 25 g per 100 g raw material, with the corresponding Fe, P, Li, and Al leaching efficiencies obtained as 0.21, 0.03, 97.23 and 11.87%, respectively.
Introduction
Methods of spent LiFePO4 battery disposal include the cascade utilisation and recovery of valuable metals [1,2]. Cascade utilisation aims to ensure the quality and safety of target products but is only applied in some demonstration plants, as the performance of waste batteries is unpredictable [3].
At present, the hydrometallurgical recovery of Li from waste LiFePO4 batteries is widely employed in the lithium battery recovery industry [4][5][6]. Zhen et al. [7] leached the LiFePO4 cathode material with a mixture of H2SO4 and H2O2 and adjusted the pH of the obtained solution with alkali to precipitate Fe as Fe(OH)3, further adjusting pH to 5.0-8.0 to remove other heavy metal ions and thus obtain a solution of Li2SO4. The latter solution was treated with solid Na2CO3, concentrated, and crystallised to obtain Li2CO3. The above method consumes large amounts of alkali and produces much slag, therefore not complying with the concept of environmental protection and economy.
Yang et al. [8] probed the effects of mechanochemical activation of spent cathode powder and its potential to achieve selective Li recovery, revealing that after mechanochemical activation, ~97.67% Fe and 94.29% Li could be recovered under optimised conditions. The lixivium was evaporated and concentrated to recover FePO4, and the filtrate pH was adjusted to neutral to afford Li3PO4. The above process is lengthy, featuring the drawbacks of high energy/alkali consumption and affording a product with high impurity content. Li et al. [9] found that the use of dilute H2SO4 as a leachant and H2O2 as an oxidant allows Li to be selectively leached into solution, while Fe and P remain in the residue as FePO4, which is different from the traditional process of using excess mineral acid to leach all elements into solution.
Under optimised conditions, Li, Fe, and P leaching efficiencies of 96.85, 0.027, and 1.95%, respectively, were recorded. However, the above study is incomplete, employing an insufficient amount of experimental materials and therefore providing unconvincing results. Moreover, the chemical stability of H2O2 is low, the leaching process is prone to overflow, and the achieved oxidation efficiency is far lower than that observed for NaClO3 and NaClO.
When waste LiFePO4 cathode material is treated with H2SO4, Fe is solubilised as Fe 2+ , and its removal consumes large amounts of alkali liquor and produces much solid waste [10,11]. Importantly, the metal value of LiFePO4 is lower than that of lithium nickel cobalt manganate, which results in low market value and low recovery enthusiasm [12][13][14]. Herein, LiFePO4 was leached with H2SO4 in the Figure 1 shows that the dominant area of FePO4 in the Li-Fe-P-H2O system is characterised by high potential and low pH. During the leaching of LiFePO4 by acid, control of pH and system potential allows Fe and Li to be present as FePO4 and Li + , respectively, thus enabling selective leaching and FePO4 isolation.
In the absence of an oxidant, the reaction of LiFePO4 with H2SO4 can be described as
A beaker filled with H2SO4 of a certain concentration and volume was charged with LiFePO4 (100 g) and put into a water bath held at a specified temperature and equipped with an overhead mixer. Subsequently, NaClO3 was added, and after a fixed retention time, the reaction mixture was filtered, and the Fe, P, Li and Al contents of the dry filter cake were determined. The filtration residue was analysed by diffraction of x-rays (XRD Ultima IV, JP) and scanning electron microscope (SEM Apreo C, US), The filtrate was analysed by inductively coupled plasma spectrometry (ICAP 7000SERIES, UK), and leaching efficiency (ηi) was calculated as where mA is the mass of a given element in the raw material, and ma is the mass of this element in the filter residue. Figure 2 shows the effects of temperature (T) on leaching efficiency at [H2SO4] = 1 M, liquid-tosolid ratio (L/S) = 5:1, leaching time (t) = 1 h, NaClO3 dosage (D) = 25 g, and stirring speed (s) = 150 rpm.
Figure 2. Effect of temperature on leaching efficiency
With increasing T, the leaching efficiency of Li increased, that of Fe declined, and that of P did not significantly change, and that of Al increased first and then decreased. At 90°C, the leaching efficiencies of Fe, P, Li, and Al equalled 0.07, 0.017, 98.12, and 16.72%, respectively. In order to achieve the results of high lithium leaching rate and low iron leaching rate, the above temperature (90 °C) was therefore selected as optimal. Figure 4 shows With increasing t, the leaching efficiency of Li increased and then saturated, while those of Fe and P did not significantly change. After 1-h leaching, the leaching efficiencies of Fe, P, Li, and Al reached 0.025, 0.001, 98.099, and 4.54%, respectively. Thus, t = 1 h was chosen as optimal.
Effect of leaching time on leaching efficiency
The leaching efficiency of P and Fe significantly increased under conventional leaching. In the presence of NaClO3, all Fe 2+ ions were oxidised to Fe 3+ and precipitated together with P as FePO4. Figure 6 shows the effects of D on leaching efficiency (T = 90 °C, [H2SO4] = 1 M, L/S = 5:1, t = 1 h, s = 150 rpm). With increasing D, the leaching efficiency of Li increased, while those of Fe and P decreased. At D = 25 g per 100 g raw material, Fe, P, Li, and Al leaching efficiencies reached 0.13, 0.004, 96.88, and 12.71%, respectively, and the above dosage was therefore chosen as optimal. When s increased from 50 to 100 rpm, the leaching efficiencies of all elements increased. Upon a further increase to 150 rpm, P, Fe, and Al leaching efficiencies did not change significantly, while Li leaching efficiency increased to 96.26%, remaining stable at higher s. This behaviour was ascribed to the promotional effect of high stirring speed on substance diffusion.
Verification test
On the basis of single-factor experiments, optimum leaching conditions were determined as T = 90 °C, [H2SO4] = 1 M, L/S = 5:1, t = 1 h, D = 25 g per 100 g raw material. Under these conditions, the leaching efficiencies of Fe, P, Li, and Al equalled 0.21, 0.03, 97.23, and 11.87%, respectively. The morphological features of leaching residue particles obtained after leaching at optimum leaching conditions are displayed in Figure 8(a,b), It can be observed that the particles of leached residue are of different sizes (Figure 8a), and the surface of leached slag is of flocculent structure (Figure 8b). In order to identify the crystalline structure of leaching residue particles, XRD spectra were recorded and presented in Figure 8c. In order to identify the crystalline structure of leaching residue particles, XRD spectra were recorded and presented in Figure 8c. As can be seen from the above figures, LiFePO4 was leached with H2SO4 in the presence of NaClO3 at controlled pH and system potential，FePO4 is the main component of the leaching residue. This indicates in the technology，we can removal of Fe in the form of insoluble FePO4 to afford a solution of Li2SO4 and thus achieve selective Li leaching.
Conclusions
Herein, we realised selective leaching of Li from spent LiFePO4 cathodes, showing that in the presence of NaClO3, all Fe 2+ ions in solution were oxidised to Fe 3+ and precipitated together with P as FePO4 to afford a P-and Fe-free solution containing Li. 1) Optimum conditions for the selective leaching of LiFePO4 were determined as leaching temperature = 90°C, [H2SO4] = 1 M, L/S = 5:1, leaching time = 1 h, NaClO3 dosage = 25 g per 100 g raw material.
2) FePO4 can be obtained by selective leaching. After purification and modification, pure FePO4 was obtained, which was used to prepare LiFePO4 and realize resource recycling.
The paper, written in English, will publish on the website as .pdf file, ONE COLUMN; it should better between 4 and 12 pages. The article should be composed of the title, author(s), abstract, keywords, introduction, materials and methods, results and discussion, conclusions, and references. | 2,014.6 | 2020-08-04T00:00:00.000 | [
"Materials Science",
"Chemistry"
] |
Magnetic Properties of Co-FeB Amorphous Films Thermomagnetically Treated with Different Field Directions
Co-Fe-B films were prepared by electroless plating. As-deposited films were thermomagnetically treated in the applied magnetic field of 500 Oe with different field directions at 300◦C for 1 hour. The effects of magnetic field direction of thermomagnetic treatment on the structure and magnetic properties of Co-Fe-B thin films were investigated. It is found that two phases existed in annealed Co-Fe-B films: one is weak crystallized CoFe phase, the other being amorphous phase. The surface morphologies of the treated films are found to be affected by the direction of thermomagnetic treatment field. The results also show that the magnetic properties of thermomagnetically treated films are influenced greatly by the treatment field direction.
Introduction
Soft magnetic films have been widely used in magnetic heads, magnetic sensors, and planar inductors at high-frequency range.To play decisive roles in these devices, the magnetic materials must simultaneously possess higher saturation magnetization, lower coercivity, larger magnetic anisotropy, and higher resistivity.Amorphous Co-Fe-B thin films have been found to be good candidates for high-frequency applications, such as magnetic tunnel junction, HDD read-heads, and GHz magnetic film inductors [1][2][3].Varieties of techniques have been used to prepare amorphous Co-Fe-B thin films, such as magnetron sputtering [4,5], electroless deposition [6], and so forth.The electroless deposition method provides an alternative and promising fabrication process with advantages of low-cost, low-energy consumption and adaptation of complicated shapes.
Magnetic field annealing is an efficient way to induce different kinds of anisotropies in some crystalline magnetic materials [7][8][9][10], which closely correspond to magnetic performance.It has been reported that the magnetic field annealing can affect the microstructure, domain structure, and magnetic properties of amorphous materials [8,11].The previous researchers [12,13] have indicated that the nanomagnetic phase or cluster can be formed during annealing treatment even when the annealing temperature is below crystallization temperature.In this case, one can expect that the direction of magnetic field may affect the structure and properties of the amorphous films during thermomagnetic treatment.However, there are few reports on the influence of thermomagnetic treatment, especially under small magnetic field with different directions, on the magnetic properties of Co-Fe-B amorphous films.
In this paper, we prepared Co-Fe-B thin films which were annealed in small magnetic field (500 Oe) with respect to the film normal at different angles (α = 0 ∼ 90 • ) subsequently.The mechanism of structural relaxation in Co-Fe-B amorphous films induced by thermomagnetic treatment was discussed.
Material and Methods
Co-Fe-B films were prepared by electroless plating on microscope glass slides (25 mm × 16 mm) with predeposited 5 nm thick Ni film as an active layer.The electroless plating solution utilized in this study was the mixture of cobalt sulfate heptahydrate (0.029 M), iron sulfate heptahydrate (0.009 M) as metal sources, sodium borohydride (0.13 M) as reducing agent and boron source, sodium tartrate (0.239 M) as complexing agent, ammonium sulfate (0.189 M), and sodium tetraborate (0.01 M) as additive.During the plating, the solution temperature was maintained at 30 ± 1 • C using a constant temperature bath and the pH was maintained constant (12.7 ± 0.1) without agitation.After 30 min deposition, the sample was removed from the plating solution, then rinsed with distilled water several times and dried with blowing air.The as-deposited film was cut into four pieces (12.5 mm × 8 mm) in preparation for the following treatment.Structures and magnetic properties of Co-Fe-B films of varied thickness and composition were studied.Therefore, we focus on the impact of thermomagnetic treatment at different angles on the magnetic properties of Co-Fe-B films with the optimize composition [6] and determined thickness (200 nm).
As-deposited films were then annealed in an applied magnetic field of 500 Oe, oriented differently with respect to the film normal direction at 300 • C for 1 hour.In order to avoid oxidation during the heat treatment, the films were treated in pure argon gas.The longitudinal direction of all films was kept perpendicular to the field, while the film's normal direction varied from parallel to perpendicular to the field direction during the annealing process.We defined α (0∼90 • ) as the annealing angle between the film normal direction and the magnetic field direction (corresponds to the films a-d).
Grazing incidence X-ray diffraction (GIXRD) was carried out with the incident angle of 5 • on a Philips X'Pert diffractometer with Cu Kα radiation generated at 40 kV and 40 mA.Scanning electronic microscopy (SEM) was performed to determine the surface morphology and to estimate the film thickness from cross-sectional views.Vibrating sample magnetometer (VSM) was used to characterize the magnetic properties of the annealed Co-Fe-B films.
Results and Discussion
As shown in Figure 1, amorphous phase is the main component of all the films after being annealed at different magnetic field directions.When the film normal direction and the magnetic field direction approach to be parallel, that is, α 30 • , a weak peak can be found around 44.8 • , which is Figure 2 shows the SEM photographs of the Co-Fe-B thin films after thermomagnetic treatment at different angles.It is clear that some tiny particles can be found in all films.The variety of the films' surface morphology after magnetic heat treatment results from local structure relaxation.Under the condition of magnetic field, the short-range order of the amorphous film is enhanced, and the free volume density of the film decreases, resulting in the fluctuation of the surface morphology of the films.For the amorphous films annealed at different magnetic field direction, the conditions of shortrange order and bond structure are different, leading to the differences on the surface morphology of the films treated at different magnetic field directions.Energy dispersion X-ray (EDX) analysis shows that the atomic ratio of Co and Fe is 2 : 1.
The in-plane magnetic hysteresis curves of the Co-Fe-B thin films thermomagnetically treated with different α angle are shown in Figure 3.When α 30 • , the films display weakly wasp-waisted hysteresis loops.The shapes of loops are approximately square when α 60 • .Figure 3(e) shows the variation of saturation magnetization and the coercivity as a function of annealing angle measured along the in-plane direction.The maximum of the saturation magnetization and the coercivity reached 854 emu/cc and 73 Oe when α = 30 • , respectively.As the annealing angle increased above 30 • , they all descended.Especially, the film annealed at 60 • exhibits lower coercivity and higher saturation magnetization.
As illustrated in Figure 4, the out-of-plane hysteresis loops of the Co-Fe-B thin films thermomagnetically treated with different α angles are very different from those of inplane ones.They also exhibit wasp-waisted shape within the small range of applied magnetic field (±500 Oe).With the annealing angle increases from 0 • to 90 • , the wasp-waisted shape becomes more and more obvious.
Thermal annealing of amorphous materials at the temperature well below the crystallization temperature will lead to the change of structure towards a more relaxed state [14].In the process of structural relaxation, applied magnetic field can affect the anisotropy of the relaxed regions with high magnetization [15].The easy direction of Fe-Co rich region tends to be parallel to the applied magnetic field to reduce the energy.However, the effective magnetic field in the film is affected by the demagnetization effect in the case of thin film state [16].
In the case of thin films, when α angle is small (i.e., α 30 • ) for the thermomagnetic treatment, the in-plane applied magnetic field is small, and out-of-plane magnetic field is small as well because of the demagnetization effect.This leads to weak anisotropy in the relaxed regions.However, the effect of in-plane demagnetization field can be neglected when the applied magnetic field is in-plane during thermomagnetic treatment.The in-plane magnetic field applied in the thermomagnetic treatment increases with α angle increasing.The optimum soft magnetic properties obtained in the film thermomagnetically treated with the incline angle of 60 • is the result of the competition between the demagnetization effect and the driving force of annealing magnetic field.
On the basis of the above analysis, it can be concluded that the large magnetic anisotropy induced by thermomagnetic treatment is mainly in-plane when α angle is larger.However, the out-of plane magnetic anisotropy may be weak at small α angle due to the demagnetization effect.When the hysteresis loops were measured in the in-plane direction, the loops are normal ones with square-like shape.Because large anisotropy is mainly parallel to the measurement direction.The weak wasp-waisted shape in-plane hysteresis loop at low value of α angle may be caused by the weak anisotropy in the out-of-plane direction [17].When the hysteresis loops are measured in the out-of-plane direction, the loops for the film thermomagnetically treated with high α angle exhibits more obvious wasp-waisted shape, because of the large magnetic anisotropy in the in-plane direction.The anisotropy constant of Co-Fe-B with an extremely high Curie temperature (about 1300 K [18]) is big.Therefore, it is hard to form a perpendicular magnetic anisotropy and to get the out-of-plane magnetization saturated (Figures 4(a)-4(d)).The shape of hysteresis loop merely varied under small magnetic field.
Conclusions
Co-Fe-B thin films were prepared by electroless deposition successfully.The deposits were annealed in an applied magnetic field of 500 Oe being oriented differently with respect to the direction of film normal at 300 • C for 1 hour.It is found weak crystallized CoFe phase exist in annealed films.The surface morphologies of annealed films depended strongly on annealing angle.
It is found that magnetic field annealing is an efficient way to obtain different anisotropy in amorphous Co-Fe-B films.In the occasion of small magnetic field annealing, it is found that the optimum soft magnetic properties were not obtained when annealing field was completely parallel to the film plane, but with a little incline (α = 60 • ).
Figure 1 :
Figure 1: GIXRD spectra of the Co-Fe-B thin films thermomagnetically treated with different α angles. | 2,279.2 | 2012-01-01T00:00:00.000 | [
"Materials Science",
"Physics"
] |
Optimising spectroscopic observations of transiting exoplanets
When observing the atmospheres of transiting exoplanets using high-resolution spectroscopy, one aims to detect well-resolved spectral features with high signal-to-noise ratios (SNR) as is possible today with modern spectrographs. However, obtaining such high-quality observations comes with a trade-off: a lower cadence of fewer, longer exposures across the transit collects more photons thanks to reduced overheads, enhancing the SNR of each observation, while a higher cadence of several, shorter exposures minimises spectral feature smearing due to the continuously changing radial velocity of the planet. Considering that maximising SNR and minimising smearing are both beneficial to analysis, there is a need to establish where the optimal compromise lies. In this work, we model real transit events based on targets as they would be observed with VLT/CRIRES+ at Paranal Observatory. Creating four hypothetical scenarios, we simulate each observation across 100 realisations of the same transit event in order to vary the time resolution only. We remove telluric and stellar lines using the SYSREM algorithm and analyse them through cross-correlation with model templates, measuring how successfully each time resolution and case detects the planetary signal. We demonstrate that there is a continuous change in the detection significance based on time resolutions, and that the function of this significance has clear maxima. The strength and location of this maxima varies on e.g. planet system parameters, instrumentation, and no. of removal iterations. We discuss why observers should therefore take several factors into account, using a strategy akin to the 'exposure triangle' from traditional photography where a balance must be struck by considering the full context of the observation. Our method is robust and may be employed by observers to estimate best observational strategies for other targets.
Introduction
As exoplanets orbit their host stars, the chemical and physical structures of their atmospheres can be studied using spectroscopic observations.New exoplanet candidates are continually being discovered by transit-searching surveys, including both ground-based surveys, such as WASP (Pollacco et al. 2006), HATNet (Bakos et al. 2004), HATSouth (Bakos et al. 2013), KELT (Pepper et al. 2007), TRAPPIST (Jehin et al. 2011), and MASCARA (Talens et al. 2017), and space-based surveys, such as Kepler (Borucki et al. 2010), K2 (Howell et al. 2014), and TESS (Ricker et al. 2015).Together, such surveys have resulted in over 5500 confirmed exoplanets on record today, with over 4500 of those being discovered in the last ten years 1 .Thanks to this progress, alongside our increasing ability to study these planets in detail with new instruments and improvements in our methods, the relatively new field of exoplanet atmosphere characterisation is rapidly maturing.
Spectroscopic characterisation can be carried out through transmission observations of the nightsides of exoplanets or through emission and/or reflection observations of their daysides.Most confirmed exoplanetary systems are observed 1 NASA Exoplanet Archive: https://exoplanetarchive.ipac.caltech.edu/edge-on, meaning that planets can be seen regularly transiting across the face of their host stars (a primary eclipse), during which we observe the dusk-dawn terminator on their nightsides and collect light that has been transmitted through the upper layers of the planet atmosphere.Beyond this eclipse, the planets can also be studied as they continue along their orbit and are illuminated by their host star, with this dayside reflecting light from the host star together with whatever emission features come from the irradiated planet.This type of observation is ideally performed immediately prior to or following the moment the planet disappears behind the host (a secondary eclipse), but is possible during a much larger window of opportunity in time and has the advantage that it can be done for non-transiting systems too.
As such, transmission studies face unique challenges in their physical time constraints.The two branches of transit spectroscopy, namely low-resolution from space and high-resolution from the ground, both provide impressive results thanks to their specific and complementary strengths.Low spectral resolution has a drawback in that one cannot distinguish individual spectral lines, but space-based observations benefit from the lack of telluric contamination and flux-calibrated spectra with high photometric precision across a wide spectral range.This means that the contrast between wavelengths with strong absorption (where the exoplanet atmosphere is opaque) and those with little absorption is diminished with low spectral resolution, but the remaining effect can be detected with high photometric precision.This is possible from space, and characterisation has been regularly achieved with space-based telescopes such as the Hubble Space Telescope (as Seager & Sasselov (2000) predicted would be possible) for over 20 years now (e.g.Charbonneau et al. 2002;Vidal-Madjar et al. 2004;Swain et al. 2008;Pont et al. 2008;Sing et al. 2008Sing et al. , 2011;;Kreidberg et al. 2014Kreidberg et al. , 2018;;Stevenson 2016;MacDonald & Madhusudhan 2017;Benneke et al. 2019;Zhou et al. 2022) and more recently with JWST (e.g.Rustamkulov et al. 2023; JWST Transiting Exoplanet Community Early Release Science Team 2023).
Comparatively, the mirror size and associated instrumentation required for high spectral resolution observations across a large wavelength range poses an engineering challenge that is currently unachievable (in part financially) from space, and therefore these observations are restricted to being groundbased.High spectral resolution allows lines to be more easily separated, and for ground-based observations, where spectral resolution can be high across a large wavelength range, the high contrast between the effective blocking area of the planet in different wavelengths provides the ability to look between the cores of strong telluric absorption lines that dominate most of the near-infrared.High-resolution ground-based results have been consistently obtained for over a decade (e.g.Redfield et al. 2008;Snellen et al. 2008Snellen et al. , 2010;;Bean et al. 2010;Astudillo-Defru & Rojo 2013;Di Gloria et al. 2015;Allart et al. 2017;Brogi et al. 2018;Diamond-Lowe et al. 2018;Merritt et al. 2021;Nikolov et al. 2021;Maimone et al. 2022;Maguire et al. 2023), and more recently, even in collaboration with space-based results (e.g.Boucher et al. 2023;Spyratos et al. 2023).
Together, these efforts are contributing to an expanding catalogue of confirmed exoplanets and an evolving understanding of their composition, revealing a large diversity in the architecture of exoplanetary systems.With the advent of highly stable spectrographs on the largest ground-based telescopes, combining wide wavelength range with high spectral resolution, such as HARPS (Mayor et al. 2003), IGRINS (Park et al. 2014), SPIRou (Artigau et al. 2014), MAROON-X (Seifahrt et al. 2020), ESPRESSO (Pepe et al. 2021), and CRIRES+ (Dorn et al. 2023), the number of exoplanets well characterised with high-resolution spectroscopy is steadily increasing.Cross-correlation analysis with realistic templates initially made it possible to perform analyses of the atmospheric composition and basic dynamics of hot Jupiters (Brogi et al. 2016;Hoeijmakers et al. 2018), and today, the quality of observations and the advances of the analysis techniques (for many instruments, including benefits from adaptive optics systems) have reached a point where we can even speculate about more detailed meteorological effects.This is possible because we are starting to identify horizontal and vertical stratification on exoplanets as well as dynamic weather patterns as inferred from the shape of individual lines, which are resolvable with high-quality, high-resolution spectroscopy (e.g.Ehrenreich et al. 2020;Pino et al. 2022;Prinoth et al. 2022;Gandhi et al. 2023;Yan et al. 2023).
This ability to identify spectral features in detail is not only enabling more detailed characterisation, but is also relevant as studies indicate that the occurrence rates of exoplanets are high around cooler stars such as FGK stars (Kunimoto & Matthews 2020) and M-dwarfs (Hardegree-Ullman et al. 2019; Kanodia et al. 2019).As the photospheres of cooler stars allow complex molecular species to exist, these colder planethosting stars have more chemically complicated atmospheres whose absorption features flood the continuum, creating greater challenges for disentangling the planetary and stellar spectra (Wakeford et al. 2019) that may benefit from well-resolved planetary features.M-dwarfs also see an overabundance of low-mass planets with shorter periods in particular (Sabotta et al. 2021), meaning they are especially suitable for transit studies.
Great efforts are going into extending the success so far achieved in characterising gas giants to the domain of low-mass planets in potentially habitable zones.This task remains a challenge for (at least) observations and so in the present paper we explore one important question of high-resolution transit spectroscopy: how to obtain the best possible data by considering the trade-off between signal-to-noise ratio (S/N) and the time resolution of observations.
During the observation of a transit, there are two possible strategies that may be employed: exposures can either be taken at a lower cadence with fewer, longer exposures or at a higher cadence of several shorter exposures.Across the transit, longer exposures collect more photons thanks to reduced overheads, which enhances the S/N of each exposure.Meanwhile, shorter exposures minimise the effect whereby spectral features get smeared due to the continuously changing radial velocity of the planet across the single exposure.Considering that both maximising the S/N per exposure and minimising the effect of smearing will benefit analysis, there is a need to identify the optimal compromise between the two for a given target.
In the present paper, we explore this balance by investigating what range of time resolutions results in the strongest possible planetary signal with traditional cross-correlation analysis (as a measure of what can be considered an 'optimal' time resolution).We demonstrate that there is a continuous change in the significance of the cross-correlation detection based on the trade-off between a small number of higher-S/N exposures and a large number of lower-S/N exposures of a transiting target, and that averaged over a large number of realisations, the function of this significance has a clear maximum between the two.By testing different case studies, we find that the locations of these maxima depend on a number of factors, including target and system parameters.With this work, we are also presenting a robust method to determine the ideal time resolution for observing other given targets using simulated spectra.Using the methods presented here may help observers plan what exposure cadence to use prior to obtaining their data, therefore optimising their own transit spectroscopy observations.In Sect.2, we describe how we explored this question using model spectra (i.e.simulated observations).We describe how these spectra are generated to create a set of data that is representative of the observations obtainable with CRIRES+ while still maintaining full control of external variables for the purpose of comparability and repeatability.In Sect.3, we describe how our simulated observations are then analysed (following the same methodology that would be used on real data), removing the stellar signal and the telluric contamination with the commonly used SYSREM method before performing cross-correlation analysis on the remaining planetary signal.We present these results in Sect.4, followed by some further discussion regarding other considerations in Sect. 5. Final conclusions and recommendations are presented in Sect.6.
Time resolution optimisation problem
A transmission spectrum observed from Earth consists of three components: (i) stellar spectral lines from the exoplanet host A244, page 2 of 17 star, (ii) telluric spectral lines from the Earth's atmosphere, and (iii) the minor contribution of spectral lines from the exoplanet atmosphere.Importantly, the spectral and telluric lines are relatively stationary across the transit in comparison to the lines of the orbiting exoplanet; as the exoplanet moves towards and then away from the observer, its lines will vary from being blueshifted at the beginning of its transit and redshifted towards the end.In analysis, this difference in Doppler shifts between lines of different origins is exploited to help distinguish the planetary component from the stellar and telluric component (see Sect. 3.1).
The planetary component is only found in the spectra of exposures taken during the actual eclipse of the host star by the moving planet.This is limited in time, and so it sets a firm limitation on the possible maximum S/Ns of the data collected with a given instrument.In practice, observations are also taken before and after the transit to assess the stability of the instrument and to extend the baseline for following data trends such as the strong telluric features, but this data does not contain any planetary signal.
Consider the extreme case where a transit observation consists of a single exposure over an entire transit.In this case, light collection and thus S/N is maximised (much like taking a longexposure photograph in the dark on Earth) but all information about radial velocity shifts of the planet is lost as one cannot see the Doppler shift of the planetary component.This would render the data useless because subsequent analysis will not be able to tell apart planetary, telluric, and stellar features.The total transit exposure must therefore be split up into some number of subdivisions in order to track this shift, even if this results in some loss of S/N.
On the other end of extremes, in the case of too many subdivisions, there must also be a lower limit of S/N below which the cross-correlation signal will be dominated by the noise.This is because if one ignores readout noise and exposure overheads, the total S/N is (approximately) preserved while subdividing exposures; but in reality, what sets the limit is the background and the readout noise, and both grow linearly in combined data with the number of subexposures.
Establishing an appropriate number of exposures across the transit is not straightforward.Observing a transit at a lower time resolution, meaning we take a smaller number of longer exposures, increases total light collection per exposure as less time is lost to overheads, which increases S/N.However, this also results in an effect where resolved spectral features captured over a longer period are 'smeared' during the exposure due to the changing radial velocity of the moving target.A schematic illustrating how this smearing effect arises is shown in Fig. 1.
This smearing effect makes analysis and the identification of spectral lines more challenging or even impossible, especially when trying to retain the true line profile shape for more detailed atmospheric characterisation.Conversely, at a higher time resolution, meaning a larger number of shorter exposures, there is less smearing of spectral features but more time is lost in overheads between exposures (decreasing total exposure time) which gives a lower total light collection and lower S/N on the single exposure.As retaining line profiles with a high time resolution and obtaining high S/N with a low time resolution are both beneficial for analysis, there is a need to strike a balance between the two.
Generating model CRIRES+ spectra
Our objective is to analyse observational data of different time resolutions to determine where the balance between high and low exposure cadence lies.In order to analyse an ideal scenario with all uncertainties controlled, we produced simulated model spectra to conduct our analysis on.This is because if one was to attempt an investigation of our time resolution optimisation problem using real observational data only, the immediate obstacle would be that one cannot test different time resolutions simultaneously.Consequently, one would have to settle for observations of the same target using different time resolutions taken across different nights -and all would have different weather conditions, visibility, lunar contamination, etc.As such, using simulated spectra ensures that all other parameters unrelated to our study are fully controlled, and allows us to play with the level of the planetary signal for different targets to follow detection deterioration.
As the results are instrument-and target-specific, the model spectra were designed to be representative of the type and quality of data that can be delivered by CRIRES+ at the VLT at Paranal Observatory in Chile.CRIRES+ is the upgrade project of the previous CRIRES instrument (CRyogenic InfraRed Echelle Spectrograph, in operation until 2014) that was recently completed with first science observations conducted in October 2021.CRIRES+ is a cross-dispersed spectrograph with a nominal spectral resolution of up to approximately R = 100 000 across wavelengths from 0.95 µm to 5.3 µm (YJHKLM bands), covering the near-infrared region and partial mid-infrared region.This range is vital for resolving the spectral lines of several molecular species (such as CO, CO 2 , H 2 O, NH 3 , and CH 4 ), and notably, CRIRES+ is rare in its ability to cover the domain of 3.0 µm to 5.2 µm at high-resolution (Dorn et al. 2023).The range of CRIRES+ is also highly complementary with the spectroscopic A244, page 3 of 17 Boldt-Christmas, L., et al.: A&A, 683, A244 (2024) capabilities of the recently launched JWST, namely the NIR-Spec instrument which covers 0.6 µm to 5.3 µm (Birkmann et al. 2022;Jakobsen et al. 2022).
Each simulated data set corresponds to a real transit event of a hypothetical target planet as visible from Paranal, and comes with realistic time-dependent local conditions such as air mass variation, barycentric velocities for each exposure, and the telluric absorption computed for the zenith distance of the target at the moment of observations.Target and system parameters were taken from a selected case study target, WASP-127 b.This target is a puffy gas giant of radius 1.31 R J and mass 0.165 M J , found at a semi-major axis distance of 0.0484 au from its G5 host star with an orbital period of 4.18 days and a transit duration of 4.35 h (Seidel et al. 2020b) meaning it can reasonably be studied using a single transit.
The transit observations were simulated using a custom script2 that generates spectral data by following the path of the light from the stellar surface to the spectrograph detector.All simulation parameters can be found in Table 1.
Our spectral simulator first sets up the Keplerian orbits of both the planet and host star around the barycentre according to the catalogued ephemerides, and the orbits are used to determine positions and velocities of the two bodies for each exposure.An appropriate stellar template spectrum I ⋆ is then selected from the PHOENIX spectral library (Husser et al. 2013).For the planetary spectrum, we use a reference exoplanet atmosphere transmission spectrum that has been computed using the radiative transfer package petitRADTRANS (Mollière et al. 2019) and the HITEMP line list for CO and H 2 O (Rothman et al. 2010), modelling a theoretical atmosphere based on parameters for a generic hot Jupiter atmosphere (Bouchy et al. 2005;Addison et al. 2019).This planetary spectrum is then imprinted onto the stellar spectrum to create the simulated total transmission spectrum, I tr : where I ⋆ is the template that represents the out-of-transit or baseline stellar spectrum, and I b is the stellar flux that is blocked by the planetary disk and its surrounding atmosphere i.e. the spectrum of exoplanetary absorption features.Spectra are initially in units of energy/wavelength bin (erg cm −2 s −1 cm −1 ) and are converted into units of photons per pixel at a later step.At the end, planetary and stellar components of I tr are Dopplershifted according to the radial velocities of the planet, star, and barycentric velocity.
To calculate I b , the planet can be thought of as a single opaque disk with a wavelength-dependent variable for the blocked radius R b (λ).The area of this radius covers a region of the stellar surface, with a representative spectrum of the blocked stellar region given by I ⋆b (λ, t) for each exposure during the transit.This blocked spectrum varies in intensity and rest frame from the overall stellar spectrum I ⋆ due to the limb darkening and rotation of the stellar surface, with the varying amount of occultation during the ingress and egress also accounted for here.Thus, the radiation that is blocked by the planet and its atmosphere I b can be expressed as: Our spectral simulator makes the following assumptions: (i) the star has no surface inhomogeneities such as stellar spots Mass (a) M * (M ⊙ ) 0.950 ± 0.020 Radius (a) R * (R ⊙ ) 1.333 ± 0.027 Spectral class (b) G5 Magnitude (K-band) (c) m K s 8.641 ± 0.019 Effective temperature (d) T eff (K) 5828.010Orbital eccentricity (a) e 0 Orbital inclination (a) i (deg) 87.84 +0.36 Transit depth (a) δ (%) 1.021 +0.005 −0.029 Assumed eq.temperature (a) T eq (K) 1400 ± 24 Observing (VLT/CRIRES+)
or flares and emits a constant limb-darkened spectrum; (ii) the planet surface and atmosphere are homogeneous and produce a time-independent planetary spectrum; (iii) the observation is photon noise limited, and other sources of noise can be ignored.
Extinction is accounted for according to the air mass at each exposure.Telluric lines are imprinted in the combined spectrum of star and planet using a synthetic model spectrum of the standard atmosphere at Paranal, which was modeled using Molecfit (Smette et al. 2015;Kausch et al. 2015) and then scaled to the airmass value (without accounting for a scaling of H 2 O lines A244, page 4 of 17 Boldt-Christmas, L., et al.: A&A, 683, A244 (2024) with the amount of precipitable water).The model also accounts for the Rossiter-McLaughlin effect as the simulation removes the received stellar flux and Doppler shifts this spectrum according to the projected rotational velocity of the stellar surface behind the planet, before subtracting it from the overall spectrum.All spectra were simulated for observations in the CRIRES+ K-band (2.0 µm to 2.5 µm) using wavelength setting K2148.Instrumental effects acting on the spectra include the instrumental profile, which is assumed to be of Gaussian shape with a FWHM corresponding to a spectral resolution of 100 000 (i.e.approximately 3 pixels), and the spectra are scaled according to the blaze functions measured for CRIRES+.For more technical details regarding the specifications of the CRIRES+ instrument, please consult the publicly available CRIRES+ User Manual from ESO3 .
The expected S/N per pixel for a given observation was estimated using values computed using the adaptive optics (AO) setup with the CRIRES Exposure Time Calculator4 , which were fitted to yield the following relation for the K-band: with the K-band magnitude m K , the extinction coefficient assumed to be κ = 0.05, the airmass AM, and the exposure time t exp in seconds.After scaling the spectra according to the S/N for the desired exposure time, artificial photon noise drawn from a normal distribution (with a width of the S/N) was added.Throughout a transit, the planet's atmospheric lines will be smeared over multiple detector pixels due to the planetary motion.We simulate this effect for longer exposures by computing spectra for several shorter subexposures (assuming no additional readout overhead) that are then co-added with the proper Doppler shifts.
The generated model data differs from real data in that it does not go through the standard CRIRES+ data reduction pipeline.In reality, raw data from CRIRES+ is 2D (in the dispersion and spatial directions) and the reduction pipeline converts these observations into 1D spectra.Our simulations generate 1D spectra and so the 2D extraction step is not needed.Several other effects that are handled by the reduction pipeline -such as correction for dark current, bias, cosmic rays, and other instrumental effects -are not included in the model data.
At each time resolution, 100 realisations of the same transit event were generated in order to create a larger statistical sample to analyse.Between realisations, only random data noise and start time of the observation relative to the transit mid-point (varying by 1 exposure length) changes.Model observations were also generated for imagined variations of this reference case in order to investigate if and how results would vary for other theoretical planets: one case where the planetary signal was halved (fainter target); one case where the transit duration was halved (same semi-major axis, but shorter period); and one case where the target was observed at a lower spectral resolution (R = 50 000 instead of R = 100 000).Observational data was simulated for each of the four cases at nine different time resolutions, ranging from approximately 2-50 min-long exposures, for 100 simulated realisations of the same transit event; in total, this results in 3600 simulations (see Sect. 4).
Analysis
It is important that the simulated observational data is treated as equally as possible to genuine observational data in order to retain realism.As such, the subsequent analysis for the model data follows standard analysis procedures as closely as possible, starting with the removal of stellar and telluric contamination followed by cross-correlation.
SYSREM
The first step is to isolate the exoplanetary atmosphere spectrum.As described in Sect.2.1, the planetary component of the transmission spectrum can be distinguished from the stellar and telluric components by its comparatively large radial velocity shift.This shift is sufficiently significant that the movement of planetary lines should be notable from exposure to exposure across the transit (to an extent that naturally depends on the time resolution) while stellar and telluric lines are systematically present with nearly negligible offset between exposures.For example, in the generated spectra for our reference case, the stellar lines shift by 0.82 km s −1 during the observation (due to barycentric velocity); the telluric lines are completely stationary; and the planet by ±16.45 km s −1 throughout the transit.The difference between these can be used to identify and remove non-planetary lines.
The SYSREM algorithm first described in Tamuz et al. (2005) was originally designed to correct the systematic variations of light curve observations, and it has since been used in a wide range of exoplanet cross-correlation studies for removing stellar and telluric features (e.g.Birkby et al. 2013Birkby et al. , 2017;;Hawker et al. 2018;Sánchez-López et al. 2019).A major advantage of this algorithm is that it does not require any a priori knowledge of the observational features that might influence the measurements.In the visible wavelength regions that are less contaminated by tellurics, alternative approaches to using iterative algorithms like SYSREM generally include methods that model tellurics with radiative transfer codes (such as Molecfit, mentioned in Sect.2.2) and then mask the affected spectral regions before dividing out the stellar signal using baseline measurements from before or after the transit (e.g.Allart et al. 2017;McCloat et al. 2021;Mounzer et al. 2022); however, in the infrared wavelengths, the spectrum is so flooded with tellurics that this is often not a practical approach.
SYSREM achieves its goal by representing the data with a model f that consists of two components: a spectrum S that is constant in time (but dependent on wavelength) and the time variation A (constant in wavelength but dependent on time) as expressed by: Both components S and A are fitted consecutively to the data by minimising the sum of the residuals squared meaning the spectrum S will contain both the stellar and the telluric spectrum, and the time variation A includes several factors like the seeing variation, changes in air mass, etc.The planet atmosphere signal is, however, not included in the model as it changes in time due to the Doppler shift.As such, after removing the SYSREM model f from the observations, one should be left with the planet signal, the residuals of time variability not associated with systematic Doppler shift (e.g.variations of seeing), and the noise.This method is closely related to the method of principal component analysis (PCA) and can be interpreted as removing the N largest components from the observations.A244, page 5 of 17 Boldt-Christmas, L., et al.: A&A, 683, A244 (2024) Usually this algorithm is iterated several times to gradually remove features present at the same wavelengths.The number of iterations must be carefully selected; an insufficient number of iterations can result in unwanted contamination being retained, yet an excess of iterations will eventually remove the planet signal, particularly in the phases where the Doppler-shifted planetary lines fall on the wavelengths of strong telluric features.As the algorithm is not formulated as an optimisation problem, there remains an element of subjectivity in selecting the number of iterations.This choice is not always obvious, so one example of a more robust approach involves injecting an artificial (known) signal and testing how many PCA iterations were required to remove it as tested by Cheverall et al. (2023).In this work, we tested a wide range of SYSREM iterations in order to track the impact of selecting this iteration number.
Cross-correlation
Using cross-correlation is by now a standard practice in the field of characterising exoplanetary atmospheres with high-resolution spectroscopy (e.g.Birkby et al. 2013Birkby et al. , 2017;;Kok et al. 2014;Brogi et al. 2017;Hawker et al. 2018;Giacobbe et al. 2021;Prinoth et al. 2022).After applying SYSREM to remove stellar and telluric signals, one is left with a minute planetary signal buried within the noise in the form of residuals for each exposure.It is these residuals that are then cross-correlated with a template spectrum of the target atmosphere at a range of radial velocity offsets, which effectively combines the signal from all planet atmosphere absorption lines and boosts the S/N of the atmosphere detection.
The template is a simulated transmission spectrum computed using a highly simplified planet atmosphere model and atomic/molecular data of the species expected to be present at the given physical conditions.Provided that the modelled features in the template do in fact appear in the planetary spectrum, a crosscorrelation of the observation and template should confirm their presence even though the predicted relative strength of different lines may be incorrect.For real data, the cross-correlation is carried out between this template and the reduced observational data.In this work, the cross-correlation is carried out between the template and the reduced simulated data, for which the planetary signal is modelled using the very same template (created with petitRADTRANS as detailed in Sect.2.2).This means that in our subsequent analysis, we know that the features definitely exist in our data, giving us the largest possible cross-correlation peak.What we evaluate here is therefore not merely the existence of a cross-correlation detection, but rather its significance relative to the noise of our simulations across the parameter space we intend to explore.
Generally, cross-correlation studies are performed using templates of individual atomic or molecular species at a time, and the strength of each species/template's detection is calculated respectively.Once this is completed, it can be possible to use this information to infer an overall atmospheric composition using a suite of retrievals based on the detection strength of each chemical species, and the fit of this retrieved model can be measured by then cross-correlating the observed spectrum with the retrieved spectrum as shown by e.g.Brogi & Line (2019); Gibson et al. (2020); Lesjak et al. (2023); Prinoth et al. (2023).Effectively, in this work, as we are certain that we know the true global transmission spectrum a priori (i.e.our template), it is this final step of likelihood-fitting that is being simulated and whose performance is measured.
The weighted cross-correlation function (CCF) dependent on velocity v and time t can be expressed as: where i is the pixel index, R is the spectrum of residuals (after SYSREM), M is the template spectrum from petitRADTRANS, and σ R is the uncertainty of R. σ R is usually obtained from the pipeline, but for our model, this is estimated by assuming a constant error of every pixel and subsequently applying the same correction during the steps of normalisation and SYSREM to these as to the data.
For each exposure, the template spectrum is shifted in 1 km s −1 steps across a range of ±200 km s −1 around the systemic velocity v sys of the host star.The top panel of Fig. 2 shows a plot of all exposures across the planet's orbital phases along the vertical axis (i.e.time, where each row is one exposure).At each v sys step along the horizontal axis, the cross-correlation value can be determined with the SYSREM residuals of each exposure, where a high value indicates matching spectral features between those in the observation and those in the template spectrum.Thus, the transit of the planet reveals itself as a diagonal line of high crosscorrelation values: the first, bottom rows of exposures show no significant cross-correlation peak (pre-transit).This is followed by a line of high cross-correlation values, which is slanted due to the orbital motion of the planet being first blueshifted (negative velocities) and then redshifted (positive velocities).Finally, the top rows of exposures again show no significant cross-correlation peak (post-transit).
By shifting every cross-correlation function in this plot to the planetary rest frame, the slanted line becomes vertical.The shift corresponds to the planetary radial velocity semi-amplitude (K p = 126 km s −1 , which can be calculated from orbital parameters of WASP-127 b) and collapsing the plot vertically will then amplify the cross-correlation peak.Without a priori setting the correct K p , one can also try different values of K p from a large range (e.g.0-400 km s −1 ); this gives results that will each be similar to the top panel of Fig. 2 but with different slant angles, and realisations producing more vertical lines will result in narrower and higher peak after collapsing the image vertically.
Stacking the results of different shifts as sorted by K p produces an image similar to the middle panel of Fig. 2.This is a 'detection map' as the planetary atmosphere detection will show up as a bright, central region that is centred at v sys = 0 km s −1 and K p = 126 km s −1 .Overlaying each K p value in a single plot in this way gives an intuitive, visual representation of the detection strength as the brightness indicates where cross-correlation values are high, i.e.where there is a peak in the cross-correlated S/N, as seen at the bottom of Fig. 2.
Measure of optimisation
Considering this work is in the pursuit of 'optimal' observations of an exoplanet atmosphere, we must explicitly describe what we mean by this potentially subjective term.As the field of exoplanetary atmosphere characterisation is presently largely focused on attempts to identify and characterise chemical species present in exoplanetary atmospheres using comparisons to templates, the definition employed in this paper for what constitutes as an 'optimal observation' is: whatever time resolution results in a cross-correlation detection map, i.e. the K p − v sys plot, with the most significant detection.Using this value as a proxy for 'optimal' is valid as it is not only the value arguably most relevant to real observational measurements (as it is the value generally used to determine whether a detection has or has not been made), but also because it will vary with both parameters, as it should improve with both increased S/N per exposure (low time resolution) and with reduced smearing (high time resolution).
This relationship can be understood more explicitly by considering how to calculate planetary S/N.For a close-in exoplanet, assuming negligible telluric interference, the S/N for the planet can be estimated to the first order to depend on: where subscripts ⋆ and p refer to the star and planet respectively; s is signal strength; and N lines is the number of detected (resolvable) lines in the given wavelength range, which accounts for both how many lines are identified and their depths.By detecting multiple spectral lines, the S/N is boosted by the factor of √ N lines since retaining both depth and plurality of spectral lines gives greater confidence in our identification of each unique combination of line patterns as being those of a particular molecule or species (Snellen et al. 2015;Birkby 2018).This is ultimately why S/N is both negatively affected by smearing and positively affected by boosted signal strength: smearing spectral lines reduces the S/N of the detection, as this results in fewer resolvable lines, while an increase in planetary signal strength improves the s p /s ⋆ ratio.
The measure of detection strength in each detection map, denoted by S, is given by the maximum cross-correlation value resulting from dividing the cross-correlation peak by the standard deviation of the detection map (calculated by excluding the central peak, i.e. everything that is further than 50 km s −1 to the left or right in the detection map).Physically, the S-value is therefore an estimate of significance of the peak in each plot in comparison to random fluctuations.In order to confirm that this sampling method is robust against the fact that individual samples in the K p − v sys should not be statistically independent, the performance of this method was compared to that of two other sampling methods (first sampling at only one row, namely the row of the signal's K p , and then sampling at every tenth K p row to avoid small number statistics).These three sampling methods were found to give very similar results with an average standard deviation of σ = 0.078 from each other (ranging from 0.007 ≤ σ ≤ 0.192), indicating that the choice of exact sampling method is of negligible impact.
In order to provide more intuitive results to the reader, the measure of time resolution is given in S/N per exposure (S /N exp ) in this paper.As described, a high time resolution (shorter exposures) gives lower S/N per exposure, and a low time resolution (longer exposures) gives higher S/N per exposure.Considering that different systems require different total exposure times (depending on e.g.transit duration for a given exoplanet), citing a time resolution as an exposure length of n seconds does not automatically evoke an idea of whether this is high or low; instead, we cite total S/N per exposure to facilitate the comparison of relative exposure cadences across different systems.Here, a low total S /N exp represents a high time resolution (short exposures) and a high total S /N exp represents a low time resolution (long exposures).
Results
For the 'target' of each hypothetical case (reference study case and its variations), simulations were generated for nine different time resolutions.These ranged from the highest resolution at 186 exposures of 132 s each (plus 14 s overhead for each exposure) to the lowest resolution at 9 exposures of 2998 s each for the entire observation including baseline out-of-transit exposures.The nine time resolutions resulted in S /N exp ranging between 50 and 250, increasing in increments of S /N = 25 for each set of longer exposures.For each combination, we computed 100 realisations of the same transit event, and the results of these A244, page 7 of 17 realisations were averaged in order to obtain a mean solution as a reference and to assess differences between individual realisations.Each set was then filtered by SYSREM and K p − v sys detection maps were generated after each SYSREM iteration.
An example of what different single-realisation detection maps looked like for different time resolutions can be seen in Fig. 3.An overview of the different parameters for each case is detailed in Table 2.The four cases were constructed as follows: (i) Reference case: This refers to the fiducial planet that is representative of the type of target that is nominally observable with our selected instrument.For the target, we use the system and transit parameters of exoplanet WASP-127 b and its host star.
(ii) Half signal: This refers to the case of halved planetary signal strength, which is achieved by reducing the line depths of the planetary model by 50%.This case effectively simulates a range of circumstances that could result in the reduction of signal strength, such as planet size, host star magnitude, distance, data lost to weather, etc.
(iii) Half transit duration: This refers to the case of the halved transit duration, which is modelled by reducing the orbital period only.The transit duration was nominally 4.35 h, and so for this case, it was reduced to 2.18 h.For this case, the same number of exposures and the same exposure lengths were used; effectively, the change here is that a smaller number of exposures contains planetary signal.
(iv) Half spectral resolution: This refers to the case of the halved spectral resolution, simulating observations taken at R = 50 000 instead of R = 100 000.All other instrumental effects remain the same.
Idealised case: no stellar/telluric signal, no SYSREM
For certain types of observations, removal of stellar and telluric lines is significantly less complicated and does not require algorithms like SYSREM.For example, in the optical regime, telluric lines are much more sparse as the dense bands of molecular absorption do not appear until regions of longer wavelengths.As mentioned, at these spectral ranges, telluric removal can often be done by simply masking out the comparatively small number of telluric lines as needed while the stellar signal can be divided out based on data from out-of-transit exposures.However, in the infrared regime, this is not possible due to the vast breadth of wavelength regions that would have to be masked -and so, having a method for removing stellar and telluric signal in the more targeted way that SYSREM offers is very necessary.
This means that for any observations that are simulated specifically for a spectrograph like CRIRES+, algorithms like SYSREM make up an important step in analysis that is needed to be retained for realism, but this also means that any and all interpretations of the results have to be made through the lens of this method.In order to disentangle our results from the effects of SYSREM, we first explore an idealised case (based on the same parameters of the so-called reference case) where the stellar spectrum is replaced with a simple limb-darkened black body and where no tellurics have been added, also simulated over 100 realisations.This is essentially a case where one can assume that all stellar and telluric signal have been successfully removed Here, all stellar and telluric contamination are assumed to have been removed successfully (or be otherwise absent, e.g. for telluric contamination in space-based observations) through methods other than SYSREM, which is how stellar and telluric signal will be removed for the subsequent non-ideal cases.
through alternative methods (meaning this case also represents those of space-based observations).Figure 4 shows a plot for this idealised scenario, demonstrating that under these conditions, the maximum for a trade-off does indeed exist and it lies between S /N exp = 100-125.
CRIRES+ simulations using SYSREM: averaged realisations
The value of S for each detection map was recorded, and the average S-values over 100 transit realisations for the four cases were plotted on a grid as heatmaps as seen in Fig. 5.The brighter areas indicate higher values of S for that particular combination of time resolution (S /N exp value) along the vertical axis, and the number of SYSREM iterations along the horizontal axis.As the first two SYSREM iterations are not sufficient to remove telluric features, these were not included and only data from iterations 3-10 were kept for the results seen in this section.
For the reference case (top left), i.e. simulations of planet WASP-127 b over 100 realisations, there is a well-defined parametric space of maximum S-values that indicates robust detection located in the range of approximately S /N exp = 50-150 across 4-7 SYSREM iterations.Specifically for S /N exp = 75-100, S-values peak at S ≳ 15.Horizontally, S-values appear to peak around SYSREM ≈ 4-5 iterations, with subsequent iterations seeing a steady decrease across all time resolutions (as per the 'triangular' shape of the heat map).This is an indication that at lower time resolutions (high S /N exp ), even a low number of SYSREM iterations will start to destroy the planetary signal.
For the case of the halved signal (top right), S-values are approximately halved (S ≳ 7) compared to the reference case as one might intuitively expect -but notably, while the numerical values of S have scaled accordingly by approximately half, the location of the parameter space maximum has not moved significantly.In this case, the approximate best range of S/N is also S /N exp = 50-150 with a similar decay following SYSREM iterations beyond ∼4.In this case, the maxima has not shifted in location or breadth, but only in strength.
For the case of the halved transit duration (bottom left), the number of exposures that are taken during the transit is less than for the previous two cases; the total observing time and the exposure length vs. exposure time configuration have not changed, but the number of baseline exposures that are taken in out-of-transit has increased since the transit is shorter in time.Here, the region of maximised S-values is shifted down towards higher time resolutions, i.e. larger number of exposures of S /N ≲ 100, with the highest S-values here being approximately S ∼ 13.This can be explained by the interpretation that as the radial velocity of the planet increases, the smearing effect per exposure also increases, and the need for a larger number of exposures becomes more pertinent.As the planet transit time is reduced, so is the total S/N collected across the transit, which is why the maximum Svalues of this case is less than that of the reference case.For this case, SYSREM performance still tends to decay at higher iterations, but the 'slope' of this triangular shape is flatter than for the previous two cases.
For the case of the halved spectral resolution (bottom right), the overall S-values are globally lower compared to those of the reference case and the halved transit case, with highest values of S ∼ 11.These S-values are still inevitably higher than for the case of the halved signal, but the contrast between this case and the reference case greatly emphasises the benefit of high spectral resolution for ground-based observations.While this case also shows SYSREM performance decay with more iterations at longer exposures, it also shows a much more uniform distribution of S-values across the vertical axis.This reflects the reduced impact of the smearing due to the change of planet line-of-sight velocity in comparison to the spectral resolution of the instrument.In this case, the detection significance is already low, and shorter exposures are not able to recover lost spectral resolution.In a sense, as the contrast of planetary spectral lines is lost due to lower resolution, this loss dominates the detection strength of an observation more significantly than any strength lost or gained due to changes in time resolution.As such, the loss of sensitivity to atmospheric features due to reduced spectral resolution cannot be fully recovered with improved time resolution.
The effect of smearing being more pronounced for faster targets with shorter transits yet less pronounced for observations with lower spectral resolution can also be seen by plotting kernel density estimates (KDE) for our data sets as in Fig. 6.In this plot, the range of S-values is shown for the halved transit duration case and for the halved spectral resolution case compared to the reference case.
Here, the reference case's time resolutions of S /N exp = 75, 100 is shown to indeed return the highest S-values, with S /N exp = 50 faring next best, followed by S /N exp = 125, 150.However, for the halved transit duration case, the range of S-values are much more spread; here, S /N exp = 50-100 all fare similarly well (as confirmed by the heatmaps) but subsequent time resolutions quickly deteriorate with each successive time resolution resulting in a decreased peak of S-values due to increased smearing.Comparatively, for the halved spectral resolution, this is not true; here, S /N exp = 50-100 still results in the highest S-values but S /N exp = 125-175 are not far behind.For all three cases, the time resolutions of S /N exp = 200-250 fares significantly worse as is also seen in the heatmaps of Fig. 5, in part due to the diminished success of SYSREM at higher iterations.
This spread can also be identified by calculating the variance (σ 2 ) of these data sets.A pronounced smearing effect should register as a larger variance across the data set (due to larger differences between performance of shorter versus longer exposures) and a diminished smearing effect should register as a A244, page 9 of 17 In these plots, a large number of short exposures corresponds to a lower S /N exp (bottom of vertical axis) and a small number of long exposures corresponds to a higher S /N exp (top of vertical axis).smaller variance.Two values for variance are given in the titles of the plots of Fig. 6.The first value σ 2 s.i.comes from calculating the variance across all time resolutions in a single SYSREM iteration (effectively the variance of S-values in a single column of the heatmaps of Fig. 5).Here, σ 2 s.i. is calculated for SYSREM = 4 as this iteration consistently performs well for the three respective cases.The second value σ 2 t.r. is a calculation of variance across time resolutions for all SYSREM iterations (effectively a measure of how much the KDE plots for different time resolutions shift).This value represents the amount of variation between each row of the heatmaps -and considering all S /N exp ≥ 200 fare consistently badly due to poor SYSREM performance at higher iterations, σ 2 t.r.excludes S /N exp = 200-250.
For both σ 2 s.i. and σ 2 t.r., the trend seen in the heatmaps holds: the smearing effect is larger (higher σ) for the halved transit case, and smaller (lower σ) for the halved resolution case.
CRIRES+ simulations using SYSREM: comparison between single realisations
In contrast to these S-values based on detection maps averaged over 100 realisations, the S-values of a single realisation can be seen in Fig. 7. Here, grids of three randomly selected realisations out of 100 (realisation number 10, 50, and 80) for each case are shown in order to illustrate how much individual realisations can deviate from the mean.
A244, page 10 of 17 Fig. 6.Kernel density estimates of S-values for all different time resolutions S /N exp = 50-250 across all SYSREM iterations, seen for the reference case (left) compared to the halved transit duration (middle) case and the halved spectral resolution range (right) using the data presented in Fig. 5.The variances are calculated for each case across all time resolutions at a selected iteration of SYSREM = 4 (σ 2 s.i. ) and across all SYSREM iterations for time resolutions S /N exp = 50-175 (σ 2 t.r. ).
Importantly, we find that while the individual realisations' maxima are still confined within the range of maxima seen for the averaged realisations, the exact range of optimal time resolutions do not show up as a clear maxima in a single realisation in the same way.This contrast between the individual realisations and the averaged realisations clearly underlines how small the planetary signal is relative to noise and fluctuations associated with observations, meaning that observations from a given night can -and often will -vary significantly from the mean.This can also be seen in the presence of horizontal stripes of higher S-values in the single realisation plots: as noise was generated anew for each time resolution (vertical axis), each row has its own maxima of peak S-values in a SYSREM range.There are notable variations between each row as each have their own randomly generated noise, impacting each data set independently.
This is an important finding as we can see that two instances of the same exact circumstances -an observation of the same planetary system, with the same instrument, at the same time resolution -can seemingly in the worst case result in a detection or a non-detection, with favourable or unfavourable random variation governing the outcome.For example, the individual realisations of the reference case (first row of plots in Fig. 7) gives a high S-value detection at time resolution S /N exp = 75 at SYSREM = 4-5 (which the averaged realisation confirms is a good choice of parameters) for realisation no. 10, but a detection that is comparatively much more marginal for realisation no.50 and no.80. Similar variations can be seen between all cases' different individual realisations, indicating that this is a persistent, systematic issue.In order to eliminate the possibility that this result could be merely a manifestation of S being calculated from the peak value (which is itself noisy), this was also tested by using a Gaussian fitting technique, but this method still replicated these results for both the averaged realisations in Fig. 5 and the variation seen in the single realisations of Fig. 7.
Discussion
In this paper, we have used observations simulated for a specific instrument and targets.In order to explore how our findings would vary for other cases and conditions, observers may use the same method to simulate spectra for their own specific target, transit event, or instrument to experiment with resulting crosscorrelation S-values as presented here.While probing every category of such variations is certainly outside the scope of this paper, there are certain trends that we expect to hold for any highspectral resolution transit spectroscopy observation, with certain parameters impacting the range of optimal time resolutions.These are discussed below in the context of our results.
Summary of the results
1.The SYSREM + cross-correlation approach tends to work well with relatively low S/N observations.When planning observations, one should favour higher time resolution versus higher S /N exp as long as the data reaches S /N exp of 50 or so.While higher S/N per exposure helps mitigate the large variation of detection significance that we see across the individual realisations, we usually do not have the luxury of combining many transits -as such, striving for higher S/N should not happen at the cost of time resolution beyond this point.Another aspect that could motivate longer (higher S/N) exposures is the overhead time that can be comparable to exposure times for large CCD detectors (see Sect. 5.6). 2. Chances of detecting planetary signal are increased by using the highest spectral resolution available.The reason for this is two-fold: firstly, higher resolution improves contrast and effectively boosts N lines (from Eq. ( 6)).Secondly, high spectral resolution minimises the impact of telluric contamination (for ground-based observations) and opens the possibility of combining multiple transits in a single analysis if one can observe between telluric lines across multiple nights.The downside of high resolution is the difficulty of achieving high S/N and so the compromise must be found for each target and telescope/instrument combination.In our case, the width of the instrumental profile (1 km s −1 ) is a factor of 10 smaller than the variation of the planet line-of-sight velocity during the transit, which increases the potential of reliably detecting atmospheric signal without losing too much light on the slit.3. The use of SYSREM has a clear impact on results, with variation in optimal time resolution seen both between cases that do and do not use it (the ideal case of Sect.4.1 versus the A244, page 11 of 17 reference case of Sect.4.2), as well as between different choices of iterations within the same time resolution and case (along the horizontal axes of Figs. 5 and 7).In the results of Sect.4.2, SYSREM cleans the spectra of telluric and stellar features sufficiently already after 4-6 iterations.Generally, it can therefore be said that a large number of iterations is seemingly detrimental to the recovery of the atmospheric signal (as SYSREM may then be struggling to identify what is linear and not with such a small number of exposures) and should be avoided.In our cases, SYSREM performance is more affected by the time resolution of the data than by its S/N, leading to point 1 above.
Stacking multiple nights
If one can observe and combine data from multiple transits, this would improve the total S/N for a given target by stacking observations from different nights -in particular if done with the same instrumental setup.Such observations have been done for both transmission and emission studies in the past using a range of ground-based instruments (e.g.Giacobbe et al. 2021;Kesseli et al. 2022;Scandariato et al. 2023).It has also been shown that stacking multiple observations will be required for future characterisation of Earth-like and super-Earth exoplanets from space (e.g.Kaltenegger & Traub 2009;Rauer et al. 2011;Wunderlich et al. 2019), and this has indeed been achieved with recent JWST data (Lustig-Yaeger et al. 2023).However, stacking of ground-based observations in the mid-infrared will require more advanced telluric removal algorithms than SYSREM considering that the barycentric velocity change between nights and weather variations will results in large changes of telluric features between transits.As shown in Sect.4, there is a notable difference in the distribution of S-values for a single realisation compared to the distribution of the multi-realisation S-values; however, these effectively show an average of multiple observations, which is not the same as stacking observations to amplify the S/N.Differences in observing conditions (seeing, telluric line strength) may set a limit to the number of transits that can be properly combined with a gain to the planetary signal detection.
If the quality of observations of individual transits is similar, the stacked data set will be effectively similar to 'oversampling' planet transit with more spectra of the same time and spectral resolution.In any case, stacking multiple transits should result in higher values of S provided that SYSREM is modified to be used with such data set.
Variable exposure lengths
In this work, we have explored different time resolutions with simulations ranging from several shorter exposures to fewer longer exposures.However, it may be possible that a variation exists in what should be prioritised during different of the transit event; is the trade-off balance the same during ingress/egress as it is during the middle of the transit?If not, further optimisation could be possible if exposure lengths were varied over time, with either shorter exposures during ingress/egress and longer exposures during mid-transit, or vice versa.
Considering that the background spectrum of the star changes fastest in time close to the limb, this might suggest that shorter exposures during ingress and egress would be beneficial.On the other hand, the specific intensity is much lower across the limbs than in disk centre, which may suggest the opposite.This exact balance -and how it may be affected by further parameters e.g.planetary orbit eccentricity (Van Eylen et al. 2019) could be explored in future work, but because the two effects are working against each other, one could hope that equal exposure times are not far from the optimal strategy.Furthermore, it is reasonable to assume that the effect will nonetheless be significantly smaller compared to other, more dominant factors such as the number of SYSREM iterations.This, alongside the balance of the two opposite influences, discouraged further exploration of variable exposure lengths as we do not expect it to have a significant effect on the detectability of planetary signal for our particular cases.
Exoplanet atmospheric features
Several exoplanet atmospheric features have been found to affect the transmission spectra of exoplanets.Thanks to the progress in the field, we have been increasingly obliged to acknowledge that exoplanetary atmospheres are complicated and threedimensional systems that should not be overly simplified; more chemically complicated species like aerosols are now known to be commonly found in many types of exoplanetary atmospheres, creating clouds and hazes that make characterisation difficult at the lower atmosphere layers (e.g.Helling 2019;Gao et al. 2021), and the formation of other known meteorological phenomena such as precipitation and winds are also being studied in great detail (e.g.Heng & Showman 2015;Seidel et al. 2020a;Fortney et al. 2021;Loftus & Wordsworth 2021).Recent years have also seen an rise in adopting 3D general circulation models, or global climate models (GCM), for use with exoplanets in order to demonstrate the importance of atmospheric dynamics and stratification for interpretation of exoplanets' transmission spectra (e.g.Ding & Wordsworth 2019;Wolf et al. 2019;Way & Del Genio 2020;Fauchez et al. 2022).
The cross-correlation detection map is indeed sensitive to many such atmospheric effects, with evidence of high-altitude winds manifesting as broadening or off-sets.The simulations in this work did not take these effects into account, including e.g.rotational broadening -an effect that would be present most notably on tidally-locked planets with shorter orbital periodswhich would presumably lead to an increase in a type of smearing that arises separately from choice of time resolution.As such, in the case of strong atmospheric features, there may be a supplementary sensitivity in what time resolutions result in the optimal cross-correlation detection but the exact nature of this relationship would need to be explored before further conclusions can be reached.This relationship may be investigated by testing the dependence of inferred input atmospheric parameters on simulated observing choices as per e.g.Savel et al. (in prep.).
Host star activity
As mentioned in Sect. 1, cooler stars such as M-dwarfs are likely hosts of exoplanets and thus commonly observed.M-dwarfs are known to be active stars with magnetic fields that may affect close-in orbiting exoplanets (e.g.Morin et al. 2010;Airapetian et al. 2017;Kochukhov 2021), and this high level of activity means that they are likely to have starspots appear on their photospheres.While such activity of M-dwarf host stars can in cases be exploited to aid exoplanet atmosphere studies (e.g.Diamond-Lowe et al. 2023), there has been concern regarding what the influence may be of starspots present on the surface of host stars during transits -and unfortunately, the fear of such spots creating false spectral features appears to be well-founded especially A244, page 13 of 17 Boldt-Christmas, L., et al.: A&A, 683, A244 (2024) at lower spectral resolutions (Rackham et al. 2018;Apai et al. 2018;Barclay et al. 2021;Moran et al. 2023;Libby-Roberts et al. 2023).Considering the simulated transmission spectra in this work assumes a uniform stellar disk with no features, the risk of contamination due to potential activity of host stars should be taken into account when considering the validity of such an assumption.The situation could be different for stars of later spectral type, where the impact of the combination of larger level of activity and smaller convective cells must be further explored.
Differences between instruments
By now, there are tens of astronomy instruments available globally of world-class quality that are regularly used to take spectroscopic observations of exoplanets.As discussed in Sect. 1, the different traits of the instruments across this wide range all provide advantages and disadvantages regarding high or low spectral resolution, ground-or space-based, mirror size, wavelength coverage, and many more.All of these will combine uniquely to create an exclusive case for each given instrument and target, and so it is possible that the range of optimal time resolutions varies distinctly when using different instruments/telescopes.Furthermore, differences between instruments could be taken advantage of in the case that observers explore the possibility of stacking multiple observations from multiple instruments, as done by e.g.Ridden-Harper et al. (2023).
One important aspect to consider when selecting between different instruments is how large the observing overheads are.CRIRES+ has the benefit of being able to observe with relatively small overheads, requiring only a few seconds between sequential exposures.Other instruments have notably longer overheads of e.g. 100 s for each pair of frames for the Gemini-N/MAROON-X instrument, and either 45 s (fast readout mode) or 68 s (slow readout mode) for the VLT/ESPRESSO instrument.
Accounting for overhead time per exposure (t OH ) is important because large overheads will eat into valuable light collection during the transit, with a larger number of exposures (n exp ) consuming more observing time.The amount of observing time lost to overheads for a given instrument is given by t OH (n exp − 1), which means that there will be a larger difference between how much total light collection is lost for a high n exp and low n exp if the instrument has a high t OH .As such, for instruments with notably large overheads, faint targets requiring a low number of n exp need to be reasonably slow, and fast targets requiring a high number of n exp need to be reasonably bright.For targets that have both short transit times and low signal strengths, observers will benefit from using instruments with relatively small overheads instead.
Alternatives to PCA-based approaches
The major drawback of ground-based observations in the infrared is telluric contamination.Its strength is, ironically, the reason why we observe at these wavelengths as the species present in our atmosphere are often those we search for in our targets.Considering that various techniques of removing telluric features such as SYSREM and other functionally-similar tools have several limitations, their iterative nature makes it difficult to control their impact on the planetary atmosphere features.As such, it is very difficult to assess this impact even in the analysis of the cross-correlation results.
An alternative way to approach this problem of isolating the planetary contribution is to instead model the observations as a whole, treating the components (stellar, telluric, and planetary spectra) as unknown functions.This will lead to an inverse problem that can be solved for a global minimum assuming that the orbital elements and the corresponding Doppler shifts of the components are well known for all phases of the transit.While both algorithmically and computationally more complex, an inverse problem approach has the benefit of producing a single global planet transmission spectrum, rather than a set of residuals for each phase as PCA-based approaches do.Such a technique is currently under development by Piskunov et al. (in prep.), building on preliminary theoretical work by Aronson & Waldén (2015).
Implications for previous studies
With the finding that the detection strength of individual realisations vary significantly from the mean (and from each other), there is a need to assess what this means for current methodologies as the choice of time resolution or number of SYSREM iteration could be the difference between a detection and nondetection.
Many previous studies have acknowledged the lack of consistent treatment and the challenges associated with using iterative detrending methods such as SYSREM, including the potential for false positives (e.g.Hawker et al. 2018;Cabot et al. 2019;Cheverall et al. 2023).Our findings further underline this need for caution, and for all results (new or revisited), care should be taken to avoid the risk of both false positives and false negatives due to noise variation per observing night.One potential mitigation would be to carry out PCA-analysis injection tests (with their proven success record) over multiple realisations to obtain an averaged result similar to our work as this might improve iteration selection -especially for PCA-based approaches that do not use the same noise model for the injection and for the databut further work is needed to develop and test the performance of such a method.
An 'exposure triangle' rule for planning observations
Our results indicate that the concept of a trade-off in the context of transit spectroscopy observations is a multi-dimensional problem.By varying different parameters, the maxima of the 'optimal' time resolution's range can in response vary in location, strength, and size -and only clearly so over a large number of realisations.With so many variables that can influence the result, it is not surprising that changes do not manifest in a linear fashion across the breadth of parameters tested.
It is therefore our finding that while observers will fare best by creating their own simulations and investigating specific scenarios as needed, as a rule of thumb, a strategy akin to the socalled 'exposure triangle' may be recommended.In traditional photography, the rule of the exposure triangle is approximately that three elements must be considered prior to taking a photograph: aperture (how widely the camera lens' diaphragm opens), shutter speed (how long the camera shutter opens for), and film speed (how much light can be registered on the camera's film or digital sensor).These three controls balance to determine how much light enters (through aperture and shutter speed) based on how sensitive the instrument is (film speed).Depending on several conditions -e.g. the subject's motion, or the lighting of the subject's location -photographers may be restrained in one or more of these dimensions, and considering the exposure triangle allows them to determine how to compensate for those restraints through adjusting one or both of the others.For example, if one A244, page 14 of 17 Boldt-Christmas, L., et al.: A&A, 683, A244 (2024) is taking a photograph of a fast-moving car, one needs to reduce the shutter speed to capture it without blurring, but can compensate for that loss of light by increasing the aperture or increasing film speed.
By considering an analogous scenario for capturing transiting exoplanets, a similar rule may be employed: if one is restrained by a particular observational parameter -a short transit, a faint target, or an instrument of low spectral resolution -one may consider the other variables to determine approximately which parameters to prioritise or adjust.As such, there is no single range of time resolutions that will consistently deliver optimal results; like photography, the way to centre your exposure within this set of parameters will depend on several factors as discussed in this section and above.As per our results, certain adjustments are fairly intuitive -for example, that smearing effects are more pronounced for shorter periods but less pronounced at lower spectral resolutions -but others require more consideration or sacrifice.
While exact priorities must therefore be considered for each given case, one can still produce a general recommendation for a single target: smearing is more of a concern when the transit duration is short (because the effect is more pronounced) and less of a concern when resolving power is low (because spectral resolution cannot be regained by shorter exposures).Therefore, observers should first consider the parameters of their target system and chosen spectrograph/instrument. Smearing will be of minimal concern when their target has a relatively long transit duration, and when their instrument has a relatively low spectral resolution; in this case, observers should prioritise a longer exposure for maximum S/N.Inversely, smearing will be of maximal concern when their target has a relatively short transit duration, and their instrument has a relatively high spectral resolution; in this case, observers should prioritise a shorter exposure for minimal smearing.
In the case where neither of these factors is of notable effect, i.e. either because both factors are effectively balanced (long duration with high R, or short duration with low R) or because both are of roughly equal detriment (neither factor stands out as being of particularly great impact), one should instead consider the signal strength.Per our simulations, as long as the S /N exp can reach ∼50-75, longer exposures do not appear to improve detections sufficiently to be worthwhile, especially considering that SYSREM clearly performs more poorly with a lower number of spectra (longer exposures).This means that for systems that can guarantee higher signal strength, where observers can trust that they can receive a relatively high S/N even in a shorter exposure due to e.g.high contrast, they may prioritise a shorter exposure in order to retain as much spectral resolution as possible.However, if the signal strength is expected to be weak, and observers know they can only obtain sufficiently high signal strengths by longer exposures, they are recommended to do so in spite of increased smearing.
This generalised recommendation is outlined schematically in Fig. 8, whose first plot shows a simple flow chart that may be consulted for broader decision-making, with the second plot illustrating the logic of the flow chart structure.What exact values will qualify as being a short/long transit duration, or of high/low spectral resolution, or a high/low signal strength cannot be established explicitly here for every combination of possible targets and instruments; as such, decisions based on this recommendation require a judgement call that must be made by astronomers based on their science goals (e.g.detection or characterisation) and on their global understanding of their upcoming observation.It is this evaluation, where one judges what may or Fig. 8. Generalised recommendation for a single target.Top panel: a flow chart that observers may consult while planning their observations.First, observers should consider their instrument and system parameters.For lower spectral resolutions (R) and longer transit durations (t d ), longer exposures may be prioritised; for the inverse case, shorter exposures are more suitable.For other cases, the choice is determined by what signal strength is possible.Bottom panel: the logic of the top schematic is illustrated here.Smearing is more of a concern for low R and long t d (case A ) and less of a concern for the inverse (case B).For all other cases (C, D, E), a larger concern for signal (i.e.lower signal strength) encourages longer exposures and a lesser concern encourages shorter exposures.
may not be of greatest priority for a particular case, that can be considered as the use of an 'exposure triangle' approach in this context.Again, if resources permit, more specific boundaries on time resolution may be calculated if observers choose to create their own simulations following the methodology outlined in this work, but the recommendation above can still be employed as a short-hand guide.
Conclusions
In this work, we simulated multiple CRIRES+ observations of four hypothetical targets based on fiducial planet WASP-127 b and used the PCA-like SYSREM algorithm -a standard analysis tool in exoplanetary atmosphere studies -to investigate the trade-off between S/N and time cadence when conducting transmission spectroscopy observations.
(1) We demonstrate that time resolution significantly impacts the strength of cross-correlation detection for transiting exoplanets.Across our range of data sets, there is a clear trade-off between observing a transit with a small number of higher-S/N exposures and a large number of lower-S/N exposures, with the A244, page 15 of 17 Boldt-Christmas, L., et al.: A&A, 683, A244 (2024) range of optimal time resolution and overall performance varying based on the characteristics of the system being observed.The exact location, strength, and size of this range varies based on multiple variables, which means that observers need to consider the entire context of the observation when evaluating the best strategy (akin to the 'exposure triangle' rule of traditional photography), and supplementary techniques such as stacking may return different maxima ranges.We caution those planning observations to robustly simulate target systems ahead of time in order to take full advantage of the time limitations imposed both by transit duration and telescope access.
(2) Based on our simulations, variation in target signal strength does not appear to shift the location of these maxima in the parameter space, but the benefits of prioritising a larger number of exposures should be taken into account for targets with shorter transit times.At lower spectral resolution, the smearing effect associated with faster transiting targets appears to be less pronounced, although a higher spectral resolution clearly results in more significant detection at the optimal time resolution.For the four hypothetical cases explored in this paper, as a general rule, time resolutions of S /N exp ≈ 50-100 appear to provide relatively high detection strength for all cases, with no case achieving an optimal resolution of S /N exp > 200.
(3) There also appears to be a minimum threshold of time resolution below which SYSREM may destroy the planetary signal after only a small number of iterations; however, this threshold needs to be determined more precisely for specific targets and instrument configurations.For our cases, time resolutions of S /N exp ≳ 200 appear to be especially sensitive, and SYSREM ≈ 4-6 generally seems to yield the best results.
(4) We also find that night-specific variations in seeing and data noise, for example, have a significant impact on the results of that night.As this work has explored ideal cases, this impact is reasonably assumed to be even greater for real-life circumstances such as variations in weather conditions.Considering that many studies are based on single-night observations, this needs to be accounted for during analysis (including the steps pertaining to stellar and/or telluric removal) and in evaluating the validity of single-night detections.
As the field of exoplanetary atmosphere characterisation makes solid progress in every direction -improved instrumentation quality, analytical methods, data reduction, and modelling methods for generating template spectra for cross-correlationwe can reasonably expect good progress in the relatively near future concerning characterisation of lower-mass exoplanets to match our achievements with higher-mass exoplanets.With such developments making rapid strides, it is important to monitor our tools over time to ensure that they do not become obsolete; and so, by monitoring the variation of optimal time resolution for different observations, we can verify that we are continuously obtaining the best possible data for our scientific progress.
(a) Seidel et al. (2020b),(b) Lam et al. (2017),(c) 2MASS All-Sky Catalog(Cutri et al. 2003),(d) TESS Input Catalog(Stassun et al. 2019),(e) Gaia DR2 (Gaia Collaboration 2018), ( f ) ESO/CRIRES+ User Manual,(g) Dorn et al. ( Fig. 2. Examples of data plots at three different points in the analysis.(a) Top panel: exposures are stacked along the vertical axis in time (where orbital phase ϕ = 1.0 is the middle of the primary eclipse).The dotted white lines indicate before and after the transit.(b) Middle panel: K p − v sys detection map, where all values have been shifted into the planetary rest frame.The white dotted lines indicate rest frame system velocity (v sys = 0 km s −1 ) and the semi-amplitude of the radial velocity of the planet (K p = 126 km s −1 ).(c) Bottom panel: example of what the crosscorrelation function looks like at K p = 126 km s −1 , i.e. at the horizontal white line from (b).
Fig. 3 .
Fig. 3. Example of three K p − v sys detection maps for different time resolutions ('high' resolution for S /N exp = 50, 'medium' for S /N exp = 150, and 'low' for S /N exp = 250) for the WASP-127 b case at SYSREM = 5.For each detection map, its respective measure of detection strength S is cited (see Sect. 3.3 for definition) with a higher S-value implying a stronger cross-correlation detection.
Fig. 4 .
Fig.4.Idealised reference case of no stellar or telluric signal, averaged over 100 realisations.Here, all stellar and telluric contamination are assumed to have been removed successfully (or be otherwise absent, e.g. for telluric contamination in space-based observations) through methods other than SYSREM, which is how stellar and telluric signal will be removed for the subsequent non-ideal cases.
Fig. 5 .
Fig. 5. How the cross-correlation detection strength, S, varies across time resolution and SYSREM iterations for four hypothetical cases averaged across 100 realisations of the same observing night.(i) Top left: reference case of a hypothetical observation (R = 100 000) based on fiducial planet WASP-127 b. (ii) Top right: scenario where the reference case's planetary signal strength is halved.(iii) Bottom left: scenario where the reference case's transit duration is halved.(iv) Bottom right: scenario where the reference case is observed with half the spectral resolution (R = 50 000).In these plots, a large number of short exposures corresponds to a lower S /N exp (bottom of vertical axis) and a small number of long exposures corresponds to a higher S /N exp (top of vertical axis).
Fig. 7 .
Fig. 7. How the cross-correlation detection strength, S, varies across time resolution and SYSREM iterations for individual realisations of the four hypothetical cases.Each row represents (i) the reference case, (ii) the case of the halved planetary signal, (iii) the case of the halved transit duration, and (iv) the case of the halved spectral resolution, following the same format and parameters as in Fig. 5.Note that some values are beyond the colour bar's scale, which has been locked to facilitate comparison to Fig. 5. A244, page 12 of 17
Table 1 .
All parameters used for generating simulated spectra of our reference case study, a fiducial planet (based on exoplanet WASP-127 b) as would be observed with VLT/CRIRES+.
Table 2 .
Differences between the four hypothetical cases, noting what their respective deviations are from the reference case (which uses parameters from Table1). | 18,296.6 | 2023-12-13T00:00:00.000 | [
"Physics",
"Environmental Science"
] |
Hyperconjugation in Carbocations, a BLW Study with DFT approximation
The geometry of ethyl cation is discussed, and the hyperconjugation effect in carbocations is evaluated at the B3LYP/6-311G(d) level. The Block Localized Wavefunction (BLW) method is used for all evaluations of the hyperconjugation, considered as the energy gained by the delocalization onto the C+ atom. This energy is defined as the energy difference between the delocalized (standard) calculation, where the electrons are freely delocalized, and a localized form where the positive charge sits on the carbon center. It is evaluated for 18 carbocations, including conjugated systems. In these cases we were particularly interested in the additional stabilization brought by hyperconjugative effects. Among other effects, the β-silicon effect is computed. Hyperconjugation amounts in several cases to an energy similar to conjugation effects.
INTRODUCTION
Carbocations' stabilization by hyperconjugation is one of the cornerstones of chemistry, and has received a considerable attention, particularly in educational, organic, and theoretical literature (Hehre, 1975). They are involved in numerous reactions, whenever an anionic chemical group leaves a carbon atom, as it is the case in S N 1 reaction for instance, or by positively charged species attachment (White et al., 1999). Olah et al. have boosted their experimental study with exceptionally strong acids (Olah, 1993(Olah, , 2001. However, computational studies are essential for the evaluation of the energetics at work (Lambert and Ciro, 1996;Müller et al., 2005). The recent review by Aue (2011) made a special emphasis on the study of their stability, plus a presentation of carbocations of practical interest, for instance in biological systems.
Carbocations stability is a key in numerous reaction mechanisms, particularly near transition states, where bond breaking/forming processes modify the electronic density of a species. Thus, their stabilization occurs frequently in systems that can be distorted from their equilibrium geometry. Their stability relies particularly on charge delocalization over the whole chemical species, and this can be attained via conjugation and hyperconjugation (Müller et al., 2005;Hadzic et al., 2011;Newhouse and Baran, 2011;Emanuelsson et al., 2013;Zimmerman and Weinhold, 2013). Even when it is small in magnitude, hyperconjugation can determine reactivity, and is of primary importance (Cieplak, 1999;Ingold and DiLabio, 2006;Braïda et al., 2009;Fernandez et al., 2013).
The conformation plays an important role, and sometimes it can be used to switch the delocalization off (deconjugated bond), and evaluate its effects by difference with the conjugated conformation (Wiberg et al., 1990;Gobbi and Frenking, 1994).
Conjugation involves an interaction between π orbitals. It reputedly implies large stabilization energy (or resonance energy) and the effect extends across several bonds (Milian-Medina and Gierschner, 2012). Here, in the carbocation case, the charge delocalization corresponds to the interaction between the empty π orbital of the carbocation center, and at least one filled π orbital expanded on neighboring atoms (Scheme 1A) (Alabugin et al., 2011). We shall use the allyl cation as a model system to evaluate such stabilization. The hyperconjugation (Scheme 1B) involves filled CH orbitals, which are in principle lower in energy. Because orbitals interact better if they are close in energy, the effect is in principle larger for conjugation than for hyperconjugation.
As stated above in the allyl cation, a deconjugation by rotation around a CC bond (Scheme 2) can give an estimation of the resonance energy. However, this gives an underestimation of the energy because hyperconjugative effects lower the rotated structure (Mo, 2004). The value of the resonance energy in allyls has been the subject of some debates, which chiefly concerned the allyl anion (Mo et al., 1996;Mo and Peyerimhoff, 1998;Barbour and Karty, 2004;Linares et al., 2008). As far as the cation is concerned, there are less discrepancies among the authors although electronic correlation is significant and some variations are encountered. We shall retain that, with an Orbital Deletion Procedure (ODP) (Mo, 2006) the resonance energy in the allyl cation was evaluated to 36.6 kcal/mol at the HF level, and to 48.8 with B3LYP/6-311+G(d,p) level using the "Block Localized Wavefunction" (BLW) approach (Mo et al., 2007). These correspond to "Adiabatic Resonance Energies" (ARE), which means that geometrical parameters are relaxed in the localized calculation (Figure 1). The later value is close to the value of 50 kcal/mol, obtained with a corrected Hydride Ion Affinities procedure (HIA) (Aue, 2011). It is also close to the value obtained with our Lewis-based Valence Bond BOND scheme, 55 kcal/mol (Linares et al., 2006), although our value is to be considered as a Vertical Resonance Energy (VRE) because the geometry of the localized structure is constrained to that of the delocalized system. As reminded above, hyperconjugation is frequently considered as a (small) second order conjugation, which is justified by the lower energy of CH bonding orbitals compared to π (Scheme 1). This is particularly true in neutral systems, and the interaction can be small in these cases. However, there are evidences that hyperconjugation can be large, and even of similar magnitude as conjugation (Daudey et al., 1980;Mullins, 2012;Wu and Schleyer, 2013). Large hyperconjugation effects (in silicon substituted species) are reported to lead to very significant rate enhancements (up to 10 12 times larger) (Lambert and Chelius, 1990;Creary and Kochly, 2009). They also have been isolated and an X-ray structure is even available. 1 Several approaches are being used to describe the conjugation and hyperconjugation effects. We reminded in the introduction different publications using isodesmic reactions, based on 1 With two silicons and two tins in β position; see Schormann et al. (2002). In bold is the energy curve of the delocalized wave function, while in plain is the localized. VRE and ARE differ by the geometry used to compute the localized wave function. The optimum geometry with the delocalized wave function is G deloc , and G loc for the localized one.
There is also a rich literature on silicon substituted carbocations, and cross method evaluations have been recently published on these systems (Fernandez and Frenking, 2007;Dabbagh et al., 2012Dabbagh et al., , 2013. Still on carbocations, Schleyer et al. very recently , showed large hyperconjugation effects in various strained systems. However, for simple carbocations, which are our subject here, there have been fewer studies. Particularly, the values published by Mo (2006) with the BLW method at the HF level have not been updated with a care for electronic correlation.
Such an evaluation at the correlated level is certainly desired, and this is an objective of the present paper. We evaluated the energetics of hyperconjugation at the B3LYP/6-311G(d) level, which includes some correlation effects. We used the BLW approach in all the cases. Despite some discussions on its apparent basis set dependency (Mo et al., 2010;Zielinski et al., 2010) this type of calculation is becoming a standard for such an evaluation (Steinmann et al., 2011;Wu et al., 2012;Fernandez et al., 2013). Our BLW results shall update and extend the values published previously at the uncorrelated level (Mo, 2006). We expect that, as it was the case for conjugation, correlated value for hyperconjugation will be somehow larger than Hartree-Fock.
The paper is organized as follows. In the computational considerations, we first define our levels of calculations, programs we used, and we write a short memo on the way we used the "BLW" approach in the specific case of carbocations. We then turn our attention to the conformations of the ethyl cation as a model of all the hyperconjugated cations. The results and discussion part is divided into three subsections. The first one deals with the hyperconjugation in the ethyl cation. The second extends to secondary and tertiary carbocations, with methyl substituents and silicon β-effects. In the last part, we added a conjugated link between the C + atom and the substituent (e.g., a C#C triple bond). We evaluated here the incremental stabilization due to hyperconjugation in already conjugated species.
COMPUTATIONAL CONSIDERATIONS
The computations of the ethyl cation displayed in Table 1 were done with Gaussian 03 (Frisch et al., 2004). The three methods (HF, B3LYP, and CCSD) were used for the geometrical optimization with two basis sets, Pople's 6-311G(d) (Krishnan et al., 1980) and Dunning's cc-pvQZ (McLean and Chandler, 1980). DFT calculations are not very sensitive to the size of the basis set, but CCSD is much more basis set dependant. For the B3LYP calculations we used the default implementation of Gamess, with the original VWN5 correlation functional 2 rather than the defaults Gaussian's implementation (Vosko et al., 1980;Lee et al., 1988;Becke, 1993). 3 As it is also the default in Gamess, 6D orbitals were used throughout. As the basis set dependency was small only the 6-311G(d) results are discussed here. The results obtained with the cc-pvQZ basis set are given in the supplementary materials.
For all the BLW calculations, we used a version of Gamess that was modified by Mo to implement the BLW method (Mo and Peyerimhoff, 1998;Mo et al., 2000;Cembran et al., 2009). This implementation permits to re-optimize the geometry of the cations with the localization constraints. This feature was used to compute the geometrical effects of the localization: we obtained the CC + bond lengthening, and its impact on the resonance energy (hence we obtained both VREs and AREs). However, the ARE will be of little use here. We rather discuss the vertical energies because they can concern directly and unequivocally reactive intermediates in their genuine geometry. Geometrical variations upon localization/delocalization ( d) are discussed though.
With BLW, we used there the standard 6-311G(d) basis set, with no diffuses on heavy atoms and no polarizations on hydrogens. These restrictions intend to preserve the localized calculations' meaning. Diffuse orbitals on first neighbors of the carbocation, as well as π orbitals on H atoms could bring some confusion on the validity of the localization constraints (Galbraith et al., 2013). In these calculations we oriented the systems in such a way that the π system is along the z axis, and two blocks are defined for the localized calculations. One contains zero electrons (it is empty), and is defined over the p z , d xz , d yz atomic orbitals of the C + site. This block ensures an appropriate localization of the positive charge. The other block contains all the electrons and is defined over all the remaining orbitals. For the delocalized calculations, we removed the (empty) block, and added the p z , d xz , d yz atomic orbitals to the other block, so delocalization is now allowed. The analysis of the BLW results concerns both energies and difference between electronic densities. For these densities we used two "cube" files generated by Gaussian 03. One has densities obtained with the orbitals of the localized calculation. The second uses the delocalized orbitals. 4 The density differences at each point of the grid defined in the cube files were drawn with the VMD freeware. 5 We refer to these drawings as Electronic Densities Difference maps (EDD maps). They indicate the flux of electron density (gain/loss) when localization constraints are relaxed.
ETHYL CARBOCATION: ON THE Cs GEOMETRY
It is noteworthy that in the ethyl cation, which is the smallest system useful to describe and evaluate the hyperconjugation effects, the conformation with an hyperconjugation from σ-CH bonding orbital (Scheme 3B) is a minimum at the Hartree-Fock level, but this minimum collapses to a bridged conformation (Scheme 3C) at correlated levels of calculation such as B3LYP, and CCSD for both 6-311G(d) and cc-pvQZ basis sets. The corresponding HF/6-311G(d) optimized geometries are displayed in Figure 2 and both energetics and geometrical values are in Table 1. The results with the cc-pvQZ basis set are given in the supplementary materials.
Because it involves bond breaking/forming, the bridged cation (1c) needs a priori a higher level of computation than the hyperconjugated system (1a) (van Alem et al., 1998). However, all the correlated levels converged to similar energy differences, within a few kcal/mol. The average energy difference between the two conformations is E ac = −4 ± 2 kcal/mol. This is one order of magnitude smaller that the hyperconjugation energies at work (vide infra).
At the HF/6-311G(d) level the (1b) conformation and the bridged one (1c) are two different minima, but the small CCH angle (95 • ) indicates that the proton transfer has already started in (1b), and is effective in (1c). Such 1,2 transfers are related to chemical reactivity (Crone and Kirsch, 2008) (bonds are changing) rather than hyperconjugation itself. However, the limit between reactivity and resonance is somehow difficult to define in hyperconjugation because the orbitals that act as donors are C-H bonding orbitals, hence single bonds are partly broken, which corresponds (partly) to a chemical reaction. For a fair and transferable/comparable evaluation of the hyperconjugation effects, we decided to use the (1a) conformation, even though it is characterized as a transition state. The fact that at the correlated levels (CCSD and B3LYP) conformation (1b) collapses to (1c) has also motivated our choice. The (1a) conformation corresponds to the interaction between a π-CH bonding (filled) orbital and the pure empty p orbital of the carbocation (Scheme 3A). A similar scheme can be drawn for conformation (1b). The hyperconjugation in conformation (1a) is shown by both the CC + distance, which is shorter than a normal single bond, and the out of plane CCH angle which is smaller than normal sp 3 angles (e.g., at the B3LYP/6-311G(d) level d CC + = 1.412Å and CCH = 108 • - Table 1). The results are similar with the cc-pvQZ basis (see Supplementaries, Table S1).
RESULTS AND DISCUSSION
The results for hyperconjugation in simple carbocations are in Table 2 and Figure 3, while Table 3 and Figure 4 concern the evaluation of the hyperconjugation in conjugated carbocations. In the tables, the two first columns correspond to VRE and ARE as defined in Figure 1. In the three last columns is the CC + bond variation when hyperconjugation is activated. The CC + bond shortens when the delocalization is allowed and d is thus always negative. These results were of course expected since the electronic delocalization evidently builds a kind of π bonding between these two atoms. In the discussions that follow, VRE's are used more often than ARE's because their definition is more straightforward.
To have in mind an order of magnitude for our calculations, we shall recall that for the allyl cation, the VRE amounts to 56.0 kcal/mol (Table 2, entry 7). This value is to be considered as large.
HYPERCONJUGATION IN THE ETHYL CATION
The HF geometries and energies obtained for the ethyl cation are similar to those obtained previously by Mo with the ODP procedure (Mo, 2006). With the B3LYP approximation there is a shortening of the CC + bond, from 1.438Å at the HF level, down to 1.412Å. For the energetic values, we expected an increase at the correlated level, just as it was the case for allyl cation. In this case, Mo reported a resonance energy of 36.6 kcal/mol at the HF/6-311+G(d) level (Mo, 2006), and 48.8 kcal/mol with B3LYP/6-311+G(d,p) 6 (Mo et al., 2007), which corresponds to 33% of increase. The relative variation is larger for the ethyl cation: the ARE varies here from 12.3 kcal/mol at the HF level (Table 2 entry 1) to 23.2 with B3LYP (entry 2). It corresponds to 90% of increase. The VRE amounts to nearly 30 kcal/mol. Hyperconjugation is thus smaller than conjugation in the allyl cation, but the order of magnitude is similar, with a ratio ethyl/allyl = 0.53. NBO calculations 7 on the ethyl cation give access to a second order perturbation theory analysis of the Fock matrix, where the hyperconjugation is evaluated to 18.2 kcal/mol for each of the two CH bonds concerned. The total hyperconjugation can thus be evaluated to 36 kcal/mol with this approach, which is slightly larger than our BLW evaluation. The electronic transfer 6 These are Adiabatic Resonance Energies (ARE). The value we report in Table 2. In green is the electron gain, and in translucent red is the electron loss. An isodensity of 4·10 −3 was used throughout.
FIGURE 3 | Electron Density Difference maps (EDD) for the hyperconjugated carbocations reported in
from the CH bonds to the C + atom amounts to 0.27 electron. For comparison, on the allyl cation NBO gives an interaction between the π bond and the C + atom that amounts to 127 kcal/mol. Very logically, this electron transfer concerns 0.5 electron. With the NBO approach the ratio of the interactions is ethyl/allyl = 0.28. It is somehow smaller than with BLW. However, the perturbative evaluation of the interaction energy in the allyl might be subject to some caution due to the large effect we are addressing here perturbatively. Both BLW and NBO evaluations of the hyperconjugation in the ethyl cation give a relatively strong hyperconjugative interaction, and this is consistent with the significant CC + bond shortening, d = −0.10 Å. We shall note that almost the same shortening is obtained in the allyl ( Table 2, d = −0.12 Å).
HYPERCONJUGATION AND SUBSTITUTION EFFECTS
The substitution effects can be studied via two types of systems, depending on whether the substitution takes place on the carbocation atom, leading to secondary and tertiary carbocations, (cases 1, 2, 3, 6 in Table 2) or if it takes place on the atom at the α-position (hence leading to β-substituted primary carbocations) (cases 4 and 5).
The EDD map displayed in Figure 3 shows clearly the electron loss along the two CH bonds and the electron gain, with the shape of a π bond between the two carbon atoms. This corresponds to the idealized picture of hyperconjugation (Scheme 1). These EDD can only be used qualitatively, but very large differences can be visualized. For instance, the delocalization is obviously much larger in the ethyl cation (1) than in the SiH 3 substituted equivalent (6). The computed energetics are consistent with the drawing: the VRE amounts to 29.7 kcal/mol in 1, but is as small as 12.6 kcal/mol in 6.
For cases 1, 2, 3, not surprisingly, secondary and tertiary carbocations have larger and larger delocalization energy, up to VRE = 45 kcal/mol for the tertiary carbocation (CH 3 ) 3 -C + . It is interesting to note that this value is similar to the conjugation in allyl. 8 The effects of the methyl groups are not additive though. The first methyl brings about 30 kcal/mol, 10 for the second, 5 for the third. Hence, the average stabilization is 15 kcal/mol per methyl group. In (CH 3 ) 3 C-C + , the three CC + bond shortenings are accordingly smaller than in the ethyl cation, d = −0.05Å although no direct correlation between bond shortening and hyperconjugation energy can be drawn. Steric effects may also be considered to moderate the bond shortening.
For cases 4 and 5, the substitution with two methyls in αposition (4) gives almost no change in the delocalization energy as compared to ethyl cation. It is larger by only 3 kcal/mol (VRE = 33.0 kcal/mol). This variation is similar to the variation reported using other approaches, for instance by Aue with the Hydride ion affinity (+5 kcal/mol) (Aue, 2011). The disilyl (SiH 3 ) 2 substitution (5) corresponds to a β-substituents, and leads to a significant increase in the resonance energy (by almost +30 kcal/mol). It is much larger than for the dimethyl (4) (CH 3 ) 2 moieties. The delocalization energy, VRE = 61.8 kcal/mol, is larger than the resonance energy in the allyl cation (7) at the same level. Table 2. Table 3. In green is the electron gain, and in translucent red is the electron loss. An isodensity of 4·10 −3 was used throughout.
Our results correspond roughly to Frenking's EDA evaluation of the relative stabilization energies between these two systems (33 kcal/mol) (Fernandez and Frenking, 2007), and similar results were reported by others for such silicon in β-position, for instance by isodesmic reactions 9 (Lambert, 1990;Creary and Kochly, 2009). The delocalization energy differences are similar, but the delocalization energies differ, sometimes significantly. For instance, Frenking's EDA approach gives almost twice larger E π (100 kcal/mol for the di-silyl substitution) (Fernandez and Frenking, 2007). 9 See for instance Lambert (1990).
One shall also note that the bond shortening for the CC + bond is similar for these three primary carbocations, although the resonance energy can be very different. Although it is true that the CC + distance variation reflects hyperconjugation, linear correlations cannot systematically be drawn. 10 In 6, the Si-C + bond distance shortens by about −0.12Å, which is a large difference for such a small energetic effect. These SiC bonds are in principle longer and more flexible than CC bonds. 11 The distance changes upon delocalization ( d) are probably less relevant than the actual bond length, obtained with standard calculations (that is without block localization constraints).
HYPERCONJUGATION IN CONJUGATED SYSTEMS
We already mentioned that conjugation and hyperconjugation might have similar stabilization energies. We include in this part a few examples where conjugation is evaluated in typical systems such as the already discussed allyl cation, the aromatic benzyl cation and the C#C triple bond. These results are extended with some substituted systems to study how additional hyperconjugation operates in already conjugated systems. The results are in Table 3, with some EDD maps in Figure 4.
The hyperconjugation effect on the allyl cation (7) is shown with three substitutions: one on position 2 (8) and the other two on position 3 (9, 10) (Scheme 4). In 8 there is almost no effect: the resonance energy with the methyl substituent (57.4 kcal/mol) is very similar to the unsubstituted case (56.0 kcal/mol), but for a substituent in position 3, the resonance energy increases by about +10 kcal/mol (66.9 kcal/mol in 9) with a methyl, and by +25 for the di-silyl methyl (81.8 kcal/mol in 10). These 10 However, convincing linear regressions have been discussed on the subject, see Fernandez and Frenking (2007). 11 ν CC = 995 cm −1 vs. ν SiC = 700 cm −1 , values extracted from http:// webbook.nist.gov (a) ν CC from Shimanouchi (1971) Shimanouchi (1977). hyperconjugated substituents act exactly as any conjugated sp 3 substituent would act (e.g., −OH, −NH 2 ). In 8 the methyl is conjugated with the double bond, but it is deconjugated from the positive charge on C 1 , hence its effect is negligible when delocalization is forbidden/allowed on C 1 . In 9, the methyl is conjugated with both the double bond and with C 1 , hence the effect that we calculated on C 1 is enhanced. It is even one of the largest resonance effects: it is similar to that of the benzyl cation (11), 68.7 kcal/mol.
The delocalization effects in the triply bonded systems are evaluated in the remaining systems (12-18). It is shown on the unsubstituted case that delocalization effects in the propyne cation (12) are similar to the allyl (7). In both species the resonance energy is evaluated to about 55 ± 1 kcal/mol. Substitutions at the acidic position in these systems increase the stabilization energies by about +10 kcal/mol for most species. The substitution by either a methyl (13) or a silyl (14) gives approximately the same resonance energy (about 63 kcal/mol). Larger substituents such as ter-Butyl (15), tri-Methyl Silyl (TMS) (16) or di-Methyl Silyl (DMS) (17) leads to only slightly larger resonance energies (65-68 kcal/mol). However, large resonance energy is found (again) with di-Silyl-Methyl (DSM) derivative. In that case, the (vertical) resonance energy increases to 78.9 kcal/mol. This value corresponds to an increment of some +25 kcal/mol (compared to the unsubstituted case), as it was the case for the allyl (10).
The large resonance energy corresponds to more efficient σbond delocalization. However, in both DSM and DMS, the same type of SiC (or CSi) σ-bonds interacts with the conjugated carbocation. The stabilization increment is significantly larger for DSM (+25 kcal/mol) than for DMS (+10) we shall attribute it to the rather short distance between the conjugated link and the CSi bond in DSM (which is much smaller for DSM than DMS-Scheme 5). The interaction would finally be favored due to a better overlap.
Similarly to the previous series, EDD maps can be used to visualize main electronic effects in these conjugated systems (Figure 4). Of course, most of the delocalization comes from the conjugated link, but larger hyperconjugations correspond to larger domains of electron loss. This is the case for the cations 10 and 18 but much smaller domains appear for 9 and 17.
CONCLUSION
Using B3LYP we pinned down resonance energies in a variety of carbocations, with a special attention to hyperconjugative effects. Our discussion focused of vertical resonance energies, and we showed here how hyperconjugation could be of similar magnitude as conjugation, but this evaluation necessitates some correlated methods.
The fact that we considered cationic systems enhanced the delocalization effects. Smaller effects are expected (and reported) for neutral systems (Fernandez and Frenking, 2006). Nevertheless, a dimethyl-silyl substituent (DMS), delocalizes a significant amount of electron density from the CSi bonds onto the neighboring C + , and this hyperconjugation corresponds to a stabilization energy as large as 61.8 kcal/mol. This is to be compared to the vinyl delocalization onto the C + , in the allyl cation. It amounts to "only" 56.0 kcal/mol of conjugation; hence hyperconjugative effects on energy can be larger than conjugation. The CC + bond distances are accordingly short, e.g., 1.371Å for the DMS-CH 2 + carbocation.
Long-range hyperconjugative effects travel across an unsaturated linkage (a double or a triple bond here). We show that the energy associated to them can be as large as 25 kcal/mol. They can be logically extended to aryl linkages, for instance in para substituted benzyl cations. | 5,972.4 | 2013-11-21T00:00:00.000 | [
"Chemistry",
"Physics"
] |
PEBA/PDMS Composite Multilayer Hollow Fiber Membranes for the Selective Separation of Butanol by Pervaporation
The growing interest in the production of biofuels has motivated numerous studies on separation techniques that allow the separation/concentration of organics produced by fermentation, improving productivity and performance. In this work, the preparation and characterization of new butanol-selective membranes was reported. The prepared membranes had a hollow fiber configuration and consisted of two dense selective layers: a first layer of PEBA and a second (outer) layer of PDMS. The membranes were tested to evaluate their separation performance in the selective removal of organics from a synthetic ABE solution. Membranes with various thicknesses were prepared in order to evaluate the effect of the PDMS protective layer on permeant fluxes and membrane selectivity. The mass transport phenomena in the pervaporation process were characterized using a resistances-in-series model. The experimental results showed that PEBA as the material of the dense separating layer is the most favorable in terms of selectivity towards butanol with respect to the other components of the feed stream. The addition of a protective layer of PDMS allows the sealing of possible pinholes; however, its thickness should be kept as small as possible since permeation fluxes decrease with increasing thickness of PDMS and this material also has greater selectivity towards acetone compared to other feed components.
Introduction
In recent years, a series of initiatives have been carried out to support research and development that allow the replacement of fossil fuels with biofuels produced from renewable resources [1][2][3]. Among the various biofuels that have attracted attention, biobutanol stands out as it has, compared to ethanol, a higher energy density, lower miscibility with water, and lower vapor pressure. In addition, important advances have taken place in the production of biobutanol from different feedstocks [4,5]. As commented by Iyyappan et al. [6], a process involving cost-effective substrate and efficient biobutanol recovery methods could help with the implementation of the biobutanol production industry. Sugarcane bagasse, algal biomass, crude glycerol, and lignocellulosic biomass are considered potential cost-effective substrates for the production of butanol, which could replace glucose-based substrates.
Although, to a large extent, n-butanol is currently produced by chemical synthesis, there is a growing interest in its production by a biochemical route, mainly through the process known as ABE fermentation, in which, in addition to n-butanol, microorganisms also produce acetone and ethanol [7]. The biological production of butanol is specific to several Clostridia species. Among these, Clostridium acetobutylicum is considered the main species for biobutanol production; although, in recent years, other options have also aroused interest, such as the use of Escherichia coli or genetically modified organisms [6,8]. ABE fermentation by C. acetobutylicum takes place in two phases: an acidogenesis phase wherein the microbes mainly produce acetic acid and butyric acid, followed by a solventogenesis phase wherein the microbes mainly produce ABE compounds [9].
The industrial implementation of the fermentation process has faced several obstacles [10], including the inhibition of microorganisms by the butanol formed during fermentation. This phenomenon causes the ABE fermentation to stop when the concentration of the solvents produced is around 2 wt%, causing low utilization of the substrate and the costly recovery of butanol from diluted solutions. The obstacle due to the low solvent tolerance could be solved with the continuous butanol removal from the fermentation broth. Various separation techniques have been proposed and studied to remove products and increase the efficiency of the fermentation process, including gas-stripping, liquid-liquid extraction, and membrane technologies such as pervaporation and membrane distillation [11][12][13][14][15]. Some technical-economic studies indicate that the in situ product recovery (ISPR) from the ABE fermentation broth not only increases the productivity and the yield of ABE by eliminating the product inhibition, but it also reduces the energy consumption and the separation cost [16].
Among the separation techniques, pervaporation has attracted sustained interest from the scientific community in recent times. Pervaporation is a technique that allows the separation of liquid mixtures by permeating their components at different rates through a dense selective membrane, applying a certain vacuum on the downstream side of the membrane to establish the driving force for mass transfer. A series of advantages are attributed to pervaporation, such as high selectivity, low operating temperature, reasonable performance to cost ratio, possibility of modular design, and the absence of a separating agent that could cause product contamination [17,18]. Particularly, the low/moderate operating temperatures make pervaporation especially useful to work coupled with a bioreactor without harming the activity of microorganisms [7]. In addition, potential drawbacks associated with the use of membranes should be taken into account, such as membrane fouling, high equipment cost, and low/moderate productivity conditioned by permeation fluxes through the membrane [15,16].
For the case that interests us, which involves the selective removal of organic compounds from an aqueous solution, hydrophobic pervaporation membranes are used. Various polymeric materials for membranes have been tested for the separation of alcohols in the studies reported in the literature, among which poly(dimethylsiloxane) (PDMS) is the most used and, to a lesser extent, there are poly(ether-block-amide) (PEBA), poly(octylmethylsiloxane) (POMS), poly[1-(trimethylsilyl)-1-propyne] (PTMSP), polyurethane, and poly(vinylidene fluoride) (PVDF) [19][20][21]. In addition, in recent years, there has been an increasing interest in composite membranes formed with a polymer matrix with the addition of fillers such as zeolites [22,23], ZIF [24], MOF [25], and graphene oxide [26]. From all these polymer materials, PEBA has been especially selective towards butanol [27] due to its affinity for this compound that allows an appreciable solubility of butanol in the polymeric matrix, while the PEBA's hydrophobic character limits the transport of water through the membrane.
Although most of the studies on pervaporation reported in the literature are carried out with flat sheet membranes, there is a growing interest in the development of membranes with a hollow fiber configuration that allow the construction of compact membrane modules with high membrane surface areas [28]. Fibers can be made as isotropic membranes with a uniformly dense structure [29], but they are preferably formed as a microporous structure with a dense selective layer on the outside or inside surface (anisotropic membranes) [30]. The dense surface layer can be either integral with the fiber or a separate layer coated on the porous support fiber. Given that our purpose is to obtain hollow fibers with a thin dense layer of PEBA, it was found in previous studies [31] that a suitable way of doing this is by depositing a thin layer of PEBA by dip-coating on a porous support of a suitable material, such as polypropylene. However, the aim of achieving PEBA dense layer thicknesses equal to or lower than 1 µm makes the presence of pinholes on the membrane surface more likely, causing selectivity losses. One way to repair the presence of pinholes is to cover the membrane with a second thin coating layer of a relatively permeable material such as silicone rubber to seal defects. Thus, a defect-free membrane would be obtained where the silicone rubber protective layer improves selectivity at the cost of a certain decrease in permeability.
Several examples of studies on multilayer membranes, including a protective layer of PDMS, are mentioned in the review papers by Dai et al. [32] and by Kujawski and col. [33]. Yahaya [34] reported a study on the separation of phenol from aqueous streams by PV using PDMS/PEBA two-layer hollow fiber membranes. From the pervaporation experiments, it was found that a significant improvement in the phenol/water separation factor and phenol flux was achieved with two-layer (PDMS/PEBA) membranes compared to that achieved using only the PDMS membrane. These results indicated that two-layer membranes combine the unique features of PDMS (exhibiting high permeability) and PEBA (exhibiting high permselectivity) to achieve this improvement in membrane performance. Jiang and Song [35] prepared polysulfone (outer layer)/Matrimid (inner) dual-layer hollow fiber PV membranes, applying them for tert-butanol dehydration. This work is a useful reference for mass transfer modeling since it uses a resistances-in-series model adapted to the cylindrical configuration of hollow fibers.
In the present work, a study about the preparation of multilayer membranes with hollow fiber configuration is reported. The fibers have two separating dense layers: a first layer of PEBA and a second (outer) layer of PDMS. The membranes were characterized in terms of their morphology and were subsequently tested to evaluate their separation performance in the selective removal of organics from a synthetic ABE solution. Membranes with various thicknesses of PDMS were prepared in order to evaluate the effect of the PDMS protective layer on permeant fluxes and membrane selectivity.
Materials
For the preparation of the composite HF membranes, Celgard X-20 polypropylene (PP) commercial HF membranes (supplied by Celanese) were used as support. These fibers had an internal diameter of 400 µm and a wall thickness of 30 µm, the porosity was equal to 40%, and the pores in the membrane were approximately 0.115 µm in diameter.
The aqueous feed solution for the PV experiments was prepared by mixing n-butanol (Merck), ethanol (Merck), and acetone (Riedel-de Haën) with ultrapure water Milli Q obtained from a Merck-Millipore system (supplied by Merck KGaA, Darmstadt, Germany). The acetone/butanol/ethanol content in the feed solution was 1:2:1 wt%. All of the materials were of analytical grade and were used without further purification.
Fabrication of Multilayer Hollow Fibers
As previously indicated, the purpose of this work was the preparation of multilayer membranes with hollow fiber configuration, incorporating two separating dense layers: a first layer of PEBA and a second (outer) layer of PDMS. Since both PEBA and PDMS polymers are rubbery materials, the deposition of two thin dense layers can be adequately carried out by dip-coating. The fabrication of ultrathin skin layer hollow fiber membranes implies a certain complexity in terms of the operating variables to be taken into account, as pointed out by Chung and col. [36]. Here, the dip-coating procedure consisted of immersing the commercial fibers in the coating solution for few seconds (3-5 s) to allow a thin film formation on the outer phase of the fibers. Covering of the ends of the fibers was previously performed to prevent the dip-coating solution from entering the fiber lumen.
The first layer was placed on the outside of the support by the dip-coating of a polymeric solution containing PEBA, following the procedure described in our previous work [31]. The coating solution used was prepared with a concentration of 2 wt% of PEBA in n-butanol, keeping it under stirring at 70 • C for 24 h. Then, it was left to rest for natural degasification and cooling for 24 h at room temperature. The viscosity was measured at room temperature using a rotational viscometer (Model Alpha Series L, Fungilab S.A., Spain), obtaining a value of 7.5 cP. The procedure that we followed allowed us to obtain a dense layer thickness of about 1.6 µm, as measured by SEM analysis. After depositing the PEBA layer, the fibers were left to stand for 1-2 days at room temperature for drying.
The second dense layer of the HF membranes was obtained by dip-coating with a PDMS coating solution. In all cases, the same PP support was used, and the dense layer thickness of PEBA was 1.6 µm. The PDMS polymer solution prepared from the two-component kit was slowly diluted in n-hexane to prepare various coating solutions, depending on the thickness of the dense layer to be obtained. The viscosity of the mixture was measured at room temperature using a rotational viscometer. Finally, the dip-coating of the hollow fibers was conducted by immersing a fiber for a few seconds (3-5 s) in a specific coating solution according to the intended dense layer thickness, making sure the whole fiber was covered. The fibers were placed such that they did not have contact in between and were left for curing at room temperature for 1-2 days. Thermal accelerated aging treatment was carried out by securely fastening the fibers in an oven and heating it up to 100 • C for 1 h for the complete crosslinking of the polymer. After, they were left for cooling at room temperature for one day. A uniform film of PDMS was formed on top of the fiber.
The fibers were potted into modules for pervaporation tests, for which epoxy adhesive (DP 105, 3M Scotch-Weld) was used to seal the ends of the module. Each module consists of 15 fibers with a length of 15 cm, and the total membrane area was 28.3 cm 2 based on the internal diameter of the HF.
In addition, PDMS dense films were prepared to be used in other characterization tests such as contact angle measurements and to establish an indicative relationship between the viscosity of the PDMS polymer solution and the thickness of the film. A small glass plate was immersed in the same PDMS coating solution used for the second layer formation and followed by the same crosslinking process after which the film thickness was measured by using a digital micrometer (Mitutoyo, Germany).
Membrane Characterization
The thickness of the dense layers and cross-sectional morphologies of the composite membranes were determined by means of scanning electron microscopy (SEM, model Zeiss EVO MA15). The samples were prepared by immersing and fracturing the membranes in liquid nitrogen, followed by gold thin film deposition using a sputter coater.
The static contact angle for a PDMS film was measured by the sessile liquid drop method using a contact angle measurement system (DSA25, Krüss, Germany) in order to obtain information about hydrophobicity/hydrophilicity and the wetting behavior of the membranes prepared. For the goniometric measurements, a flat sheet membrane was prepared as described in the previous section. A 2.0 µL drop of different pure solvents (water, acetone, butanol, and ethanol) and ABE solution were deposited on the material membrane's surface at five different sites. Each value was obtained using the software provided through image recognition. The average value for the contact angle was then considered.
Additionally, the results of TGA and FTIR analyses for the membranes used in this work were included as Supplementary Materials. The thermal stability of the components used in this work (Pebax 2533, PDMS, and PP) was examined by thermogravimetric analysis (TGA), both individually and together on the hollow fiber membrane. The TGA experiments were performed using a DTG-60H Shimadzu thermobalance. Membrane samples with an initial mass between 3 and 15 mg were placed on an alumina cell. The samples were heated up to 700 • C at a heating rate of 5 • C min −1 in nitrogen (25 mL min −1 ). In the specific case of PDMS, the final temperature was 1000 • C because this material has a higher resistance to degradation in a nitrogen atmosphere. On the other hand, an ATR-FTIR analysis of the polymers used for the fabrication of the selective layer of the membranes developed in this work, Pebax 2533 and PDMS (as dense homogeneous films), were carried out using a Perkin Elmer spectrum 65 Fourier Transform Infrared Spectrometer in the region of 400-3900 cm −1 . In addition, this analysis was carried out on one of the hollow fiber membranes composed of a polypropylene support, a dense Pebax layer, and a second (outer) dense PDMS layer.
PV Experiments
The pervaporation experiments were carried out in a laboratory-scale unit supplied by Sulzer Chemtech (Germany) that was previously used by the authors in other pervaporation studies [37,38]. The separation process takes place when the feed mixture goes through the shell side, and the permeate comes out of the lumen side when the vacuum is applied ( Figure 1). Unless otherwise stated, pervaporation tests were performed at 40 • C, since this temperature is within the appropriate temperature range to carry out the ABE fermentation process as indicated in the literature [39]. The permeate side was kept below 10 mbar using a diaphragm vacuum pump (Vacuubrand PC 3004 VARIO), and samples of the feed and permeate were collected every 30 min and were weighed and analyzed by gas chromatography (GC). After condensation, the permeate undergoes a separation of phases into an n-butanol-rich phase and a water-rich phase. After measuring the mass, the permeate was diluted with water to form a single-phase solution and then ABE concentrations were measured by GC. Samples were analyzed by duplicate in a headspace gas chromatograph (GCMS-QP2010, Ultra Shimadzu) equipped with a flame ionization detector (FID). Compounds were separated into a DB-Wax 30 m × 0.25 mm × 0.25 µm column with a detector temperature of 270 • C. Helium was used as a carrier gas at a flow rate of 82 mL min −1 . The oven temperature was initially set at 80 • C and was subsequently increased to 150 • C at 10 • C min −1 . GC calibration was performed with external standards. Each PV experiment lasted for at least 4 h after the stabilization process (stable operating temperature) to be sure that a pseudo steady state was reached.
The PV performance of a membrane was evaluated in terms of permeate flux J, membrane selectivity α, separation factor β, and pervaporation separation index (PSI). The total flux J (kg m −2 h −1 ) across the membrane is obtained by relating the mass of permeate collected with the time interval and the membrane area, as follows: After that, the flux for each component J i is calculated from the total flux and the permeate composition obtained by chromatographic analysis. Membrane selectivity α, separation factor β, and the pervaporation separation index (PSI) are calculated using Equations (2)-(4), respectively: In this work, Equation (5) was adopted to describe the flux of a permeant species, where the driving force for mass transfer is a function of permeant activity and the overall resistance (R i,OV ) includes contributions of resistances from the liquid boundary layer and the membrane itself, as shown in the next section.
Membranes 2022, 12, 1007 6 of 16 resistance (Ri,OV) includes contributions of resistances from the liquid boundary layer and the membrane itself, as shown in the next section.
Membrane Characterization
In order to establish a frame of reference for the subsequent preparation of hollow fibers with PDMS dense protective layers with different thicknesses, we started by preparing a series of homogeneous PDMS films. As previously detailed in Section 2.2, polymer solutions with different PDMS contents were used, whose viscosity were measured using a rotational viscometer. Subsequently, the thickness of the PDMS films obtained was measured using a digital micrometer. Table 1 shows the results obtained, where it can be seen how the thickness of the films ranged from 2 to 80 μm as the concentration of the PDMS polymer solution and its viscosity increased.
Membrane Characterization
In order to establish a frame of reference for the subsequent preparation of hollow fibers with PDMS dense protective layers with different thicknesses, we started by preparing a series of homogeneous PDMS films. As previously detailed in Section 2.2, polymer solutions with different PDMS contents were used, whose viscosity were measured using a rotational viscometer. Subsequently, the thickness of the PDMS films obtained was measured using a digital micrometer. Table 1 shows the results obtained, where it can be seen how the thickness of the films ranged from 2 to 80 µm as the concentration of the PDMS polymer solution and its viscosity increased. Figure 2 shows how the thicknesses of the dense PDMS films obtained by dip-coating depend on the viscosity of the polymer solution at room temperature. This trend is an indicative useful guide for selecting the characteristics of the coating solution when it is intended to obtain dense layers with a thickness less than 5 µm, especially due to the practical difficulty in handling polymer solutions with high viscosity and non-Newtonian behavior [40]. Thus, PDMS polymer solutions in the same range of viscosities shown in Figure 2 were later used as coating solutions to cover the hollow fibers with a thin outer dense layer. 20 2.5 ± 0.06 8 ± 1 40 11 ± 0.2 20 ± 1 60 200 ± 3 50 ± 1 80 560 ± 7.5 70 ± 1.5 100 4900 ± 60 80 ± 1.5 (a) It corresponds to the dilution of the two-component kit in n-hexane. Figure 2 shows how the thicknesses of the dense PDMS films obtained by dip-coating depend on the viscosity of the polymer solution at room temperature. This trend is an indicative useful guide for selecting the characteristics of the coating solution when it is intended to obtain dense layers with a thickness less than 5 μm, especially due to the practical difficulty in handling polymer solutions with high viscosity and non-Newtonian behavior [40]. Thus, PDMS polymer solutions in the same range of viscosities shown in Figure 2 were later used as coating solutions to cover the hollow fibers with a thin outer dense layer. In order to obtain additional information about the characteristics of the polymeric materials used, we measured the contact angle between the PDMS film and different chemical compounds that make up the ABE solution. The measured contact angles for PDMS are shown in Table 2, where the measured values for Pebax 2533 films, which were already reported in a previous work [31], were also included as a reference. The water contact angle measured for PDMS was 101°, highlighting the hydrophobic character of the surface. Data reported in previous studies show some variability, which has been attributed to surface roughness or experimental difficulties, with data ranging from 95° to 120° [41]. As Knozowska et al. [42] mentioned, the contact angle for a given material depends on the degree of crosslinking, and this also corresponds to an increase in roughness. The contact angles for organics were always <90°, while, for ABE mixture, the measured value was very close to that of water. Also included in Table 2 are the surface tension values for pure compounds (from Dortmund Data Bank), showing that the contact angle is influenced by the surface tension but does not depend solely on it but on the affinity between the components of the solution and the surface of the polymeric material. Several In order to obtain additional information about the characteristics of the polymeric materials used, we measured the contact angle between the PDMS film and different chemical compounds that make up the ABE solution. The measured contact angles for PDMS are shown in Table 2, where the measured values for Pebax 2533 films, which were already reported in a previous work [31], were also included as a reference. The water contact angle measured for PDMS was 101 • , highlighting the hydrophobic character of the surface. Data reported in previous studies show some variability, which has been attributed to surface roughness or experimental difficulties, with data ranging from 95 • to 120 • [41]. As Knozowska et al. [42] mentioned, the contact angle for a given material depends on the degree of crosslinking, and this also corresponds to an increase in roughness. The contact angles for organics were always <90 • , while, for ABE mixture, the measured value was very close to that of water. Also included in Table 2 are the surface tension values for pure compounds (from Dortmund Data Bank), showing that the contact angle is influenced by the surface tension but does not depend solely on it but on the affinity between the components of the solution and the surface of the polymeric material. Several authors have reported the relationship between the contact angle and the surface free energy (SFE). In general, wetting and lower contact angles occur when the surface and the liquid have similar surface energies (surface tensions, in the case of the liquids). In the case of water, a lower contact angle value was observed for Pebax 2533-based membranes compared to PDMS membranes, the SFE values being in the opposite order, according to data calculated by Knozowska et al. [42].
Membrane Performance in Pervaporation of ABE Solutions
This section reports the results corresponding to pervaporation tests carried out with membrane modules built with different sets of hollow fibers. It was worked with various thicknesses of the dense layers in order to be able to determine the contribution of each material to the membrane performance.
PV experiments were performed with each membrane module flowing (1:2:1 wt%) ABE solutions at 40 • C. For each experiment, the partial permeation fluxes were related to the driving force (activity gradient) to obtain the overall resistance to mass transfer (see Equation (5)). The activity coefficients for the components in the liquid feed mix were evaluated by the NRTL method using the Aspen Plus software.
It is well known that, in separation operations with selective membranes, the resistance to mass transfer in the fluid phase of the feed adjacent to the membrane can have a notable influence on the separation performance. Usually, hydrophobic membranes are significantly more permeable to dissolved organic compounds than to water, causing a depletion of the former compounds in the liquid boundary layer. This phenomenon is known as concentration polarization. Such effects depend mainly on the hydrodynamic conditions in the liquid phase and are usually investigated by changing the feed flow rate in pervaporation experiments. Therefore, it is essential to be able to quantify the incidence of the concentration polarization phenomenon in our tests in order to later be able to analyze the intrinsic resistance of the membrane. In a previous study [31] working with PEBA thin-film composite hollow fiber membranes, it was reported that the total resistance was fitted (Wilson plot) by the reciprocal of the lineal velocity u (m min −1 ) through the membrane module raised to an exponent of 0.9, a factor frequently adopted for parallel flow in membrane contactors [43,44]. In this work, further analysis of those permeation data for organic compounds was made in order to obtain a correlation that describes the transport parameters in terms of characteristic dimensionless numbers. Given the cylindrical configuration of the hollow fibers and taking into account that, in this study, it was adopted referring the permeation fluxes in all cases to the internal area of the support (PP), the contribution of the individual mass transfer resistances to the overall resistance for the case of fibers with a single selective dense layer of PEBA is given by the following equation: where k i,ov is the mass transfer coefficient for component i; k i,bl is the mass transfer coefficient for component i at the liquid boundary layer; γ i,F is the activity coefficient for component i in the liquid phase; ρ m is the molar density of feed liquid; P i,PEBA is the membrane permeability for component i through the PEBA layer; δ PEBA is the thickness of the PEBA layer; R i,supp is the mass transfer resistance in membrane support; A supp,in is the membrane area based on the internal diameter of the porous support; A fiber,out is the membrane area based on the outer diameter; A LM,PEBA is the logarithmic mean area of the PEBA layer. The mass transfer coefficient at the feed boundary layer (k i,bl ) depends on the circulation configuration of the ABE solution in the membrane module. In this system, the ABE solution circulates through the shell side in order to maximize mass transfer coefficients, increasing the transfer area and improving the hydraulic conditions [45]. Usually, the mass transfer coefficients in the liquid boundary layer are correlated through the Sherwood number (Sh) as a function of the dimensionless Reynolds (Re) and Schmidt (Sc) numbers. Numerous papers have dealt with reviewing and analyzing the correlations proposed for predicting shell-side mass-transfer. Among them, the papers by Lipnizki and Field [46], Shen et al. [47], and the recent work by Estay et al. [45] deserve to be highlighted. For the calculation of the Reynolds number, the definition of equivalent diameter (d eq ) for the shell-side flow proposed by Dahuron and Cussler [48] was adopted, as follows: Thus, the Reynolds number is calculated as follows: The mass transfer coefficient in the liquid boundary layer (shell-side) is related to the Sherwood number as follows: where the equivalent diameter is used as the characteristic length. The values of the diffusion coefficients (D i ) for organic compounds in aqueous solution was estimated using the Wilke-Chang correlation. Thus, an estimation of parameters was carried out to achieve a correlation that links the mass transfer coefficient at the liquid phase with the properties and operating conditions. For this, a set of experimental data was used that was obtained from PV tests with various modules built with hollow fibers coated with dense layers of PEBA with various thicknesses, working with ABE solutions at 40 • C and different flow rates (0.2, 0.3, 0.5, 1.2, 2.0, and 4.5 L min −1 ). Reynolds numbers for the experiments were in the range of 340-7800, and the length of the modules was 15 cm in all cases. The estimation procedure established that the correlation that best describes the mass transfer in the liquid phase circulating through the shell-side in flow parallel to the fibers is the following: Sh = 0.025 (1 − ϕ) Re 0.9 Sc 0.33 (12) where the packing fraction (ϕ) was calculated as follows: The parameter estimation procedure allowed, at the same time, to determine the permeability values of PEBA and resistance in the support for each one of the permeants, whose values are shown in Table 3. In the case of water as permeant, the mass transfer resistance in the liquid phase was assumed negligible. The overall mass transfer resistance values for butanol calculated with the model for three membrane modules were plotted against the experimental data to build the model parity graph (Figure 3), proving that the fit of the model can be taken as satisfactory, with good agreement between the experimental and simulated data. The small difference in the R i,supp and P i,PEBA values reported in Table 3 with respect to those reported in previous work [31] is due to the fact that, in this work, in Equation (6) the ratio of areas in each term was included so that the mass transfer resistances were referred in all cases to the internal area of the support. These data confirm that the order in which organic compounds permeate through PEBA in terms of permeabilities (kmol m −1 s −1 ) is given by: P BuOH > P EtOH > P Acet . The intrinsic selectivity of PEBA membranes towards n-butanol is explained by the preferential sorption of nbutanol over acetone and ethanol, as reported by experimental studies by Liu and Feng [49] and Heitmann et al. [50].
values for butanol calculated with the model for three membrane modules were plotted against the experimental data to build the model parity graph (Figure 3), proving that the fit of the model can be taken as satisfactory, with good agreement between the experimental and simulated data. The small difference in the Ri,supp and Pi,PEBA values reported in Table 3 with respect to those reported in previous work [31] is due to the fact that, in this work, in Equation (6) the ratio of areas in each term was included so that the mass transfer resistances were referred in all cases to the internal area of the support. These data confirm that the order in which organic compounds permeate through PEBA in terms of permeabilities (kmol m −1 s −1 ) is given by: PBuOH > PEtOH > PAcet. The intrinsic selectivity of PEBA membranes towards n-butanol is explained by the preferential sorption of n-butanol over acetone and ethanol, as reported by experimental studies by Liu and Feng [49] and Heitmann et al. [50]. Below are the results obtained in PV tests working with membrane modules made of hollow fibers consisting of two dense layers of selective material: a first dense layer of PEBA deposited on the PP support and a second dense (outer) layer of PDMS. Different thicknesses of PDMS for the outer layer were deposited by dip-coating in the range of 3-80 µm. Thus, the contribution of the individual mass transfer resistances to the overall resistance for the case of fibers with two selective dense layers is given by the following equation: where P i,PDMS is the membrane permeability for component i through the PDMS layer; δ PDMS is the thickness of the PDMS layer Membrane modules were assembled with each type of fiber, which, in all cases, contained 15 fibers with a useful length of 15 cm. The PV experiments were performed at 40 • C, while the feed flowrate was 4.5 L min −1 , which was the maximum possible flow rate for the feed liquid on the shell side in order to minimize the mass transfer resistance in the liquid boundary layer. The data collected in these experiments made it possible to evaluate the PDMS permeability values for the different permeants by means of Equation (14), taking into account that the permeabilities of PEBA were previously determined. Analyzing the PDMS permeability values for the organic compounds reported in Table 3, these follow the order acetone > n-butanol > ethanol when the driving force for mass transfer is the difference in activities for the permeant species across the membrane. Although it is not easy to make a comparison with previous studies reported in the literature due to the influence of different materials and working conditions, the higher permeability through PDMS for acetone over n-butanol and ethanol is consistent with the results presented by Rozicka et al. [51] and van Wyk et al. [52]. In this sense, Rozicka et al. [51] reported the performance of three commercial PDMS-based membranes (Pervatech, Pervap 4060 and PolyAn) in the pervaporation removal of acetone, butanol, and ethanol from binary aqueous mixtures at 25 • C. Using the data reported in that study to evaluate permeabilities (recalculating them to consider the difference in activities for the permeant species across the membrane as the driving force for mass transfer), it turns out that the highest permeability value corresponds to acetone, as well as higher separation factors for acetone over n-butanol and ethanol. Results of the same order were also reported by van Wyk et al. [52] in a study on the separation of ABE model solutions with PDMS membranes (Pervatech) in the range of 30-50 • C. The behavior of PDMS membranes more favorable to acetone permeation compared to PEBA membranes can be largely attributed to the different solvent uptake of both membrane materials. As mentioned above, the experimental studies by Liu and Feng [49] and Heitmann et al. [50] showed that the solvent uptake in PEBA 2533 is considerably higher for n-butanol than for acetone or ethanol. However, swelling studies for PDMS membranes, carried out with both pure solvents [51,53] and aqueous solutions [50], have shown that the acetone uptake is at least similar to or even higher than that of butanol, while the ethanol uptake is lower than that of the other two organics. The higher affinity of PDMS for acetone compared to the other permeants was also reported by Yang et al. [54] in a study that used inverse gas chromatography for the characterization of the solubility thermodynamics and diffusion of solvent-PDMS systems. In order to have a broader view of the variables that influence the separation performance of a PV module with dual-layer hollow fibers, a sensitivity analysis performed with simulation tools is presented below. Since the fiber length is usually around 1.0-1.5 m in membrane modules for industrial applications, we performed a simulation study to evaluate the influence of the PDMS dense layer thickness and the feed flow rate on the separation performance for a hypothetical membrane module 1 m in length. In all cases, fibers that included the porous support of PP and a selective dense layer of PEBA with a thickness of 1.6 µm were considered, on which a dense protective layer of PDMS was deposited, with variable thicknesses in the range of 0 to 20 µm. The equations that describe the mass transfer in the HF membranes (Equations (5)- (14)) together with the material and energy balances were included in a distributed parameter model that was implemented in Aspen Custom Modeler, making use of Aspen Plus subroutines for the estimation of thermodynamic properties (densities, vapor pressures, activity coefficients, enthalpies) and transport properties (viscosity, diffusion coefficients), and then a series of simulations were run. Setting the separation/concentration of n-butanol as a priority, the most favorable results in terms of PSI and butanol content (wt%) in the permeate stream correspond to the hollow fibers with the thinnest thickness of the PDMS dense layer and the highest feed flow rate (4.5 L min −1 ), as shown in Figure 4. It is important to highlight the relevant effect that the fluid dynamic conditions in the liquid phase of the feed can have on the separation performance when selective membranes are used for the removal of organics from dilute solutions. Thus, considering the case of a dual-layer HF membrane with a PDMS layer thickness of 1 µm and by increasing the flow rate from 0.2 to 4.5 L min −1 (which corresponds to a Reynolds number interval of 340-7690), the contribution of the boundary layer resistance to the overall mass transfer resistance decreased from 39% to 3.7% for butanol, while the butanol content in the permeate increased from 21.2 to 28.7 wt%. This phenomenon is probably due to the transition from a laminar to a turbulent flow for the feed liquid on the shell-side.
Conclusions
The experimental results allowed for finding a relationship between the viscosity of the PDMS coating solution and the thickness of the dense layer that can be achieved by the dip-coating procedure. The mass transport phenomena in the pervaporation process were characterized using a resistances-in-series model. Knowing that, under working conditions, fluid dynamic conditions can significantly influence the separation achieved, a correlation for the mass transfer coefficients in the liquid boundary layer as a function of the dimensionless Reynolds (Re) and Schmidt (Sc) numbers was developed. The parameter estimation procedure allowed to determine that the correlation that best describes the mass transfer in the liquid phase circulating through the shell side in the flow parallel to From the simulations, it was observed that, as the PDMS layer becomes thicker, it becomes more difficult to remove the butanol; thus, the boundary layer resistance becomes insignificant, regardless of its magnitude. We can see that, in thicker PDMS layers, there is less variation in the content of n-butanol in the permeate with a change in the feed flowrate due to the lower contribution of the boundary layer resistance to the overall resistance.
Conclusions
The experimental results allowed for finding a relationship between the viscosity of the PDMS coating solution and the thickness of the dense layer that can be achieved by the dip-coating procedure. The mass transport phenomena in the pervaporation process were characterized using a resistances-in-series model. Knowing that, under working conditions, fluid dynamic conditions can significantly influence the separation achieved, a correlation for the mass transfer coefficients in the liquid boundary layer as a function of the dimensionless Reynolds (Re) and Schmidt (Sc) numbers was developed. The parameter estimation procedure allowed to determine that the correlation that best describes the mass transfer in the liquid phase circulating through the shell side in the flow parallel to the fibers is Sh = 0.025 (1 − ϕ) Re 0.9 Sc 0.33 for HF modules with low packing fraction and Reynolds numbers in the range 340-7800. The PV results showed that PEBA, as the material of the dense separating layer, is the most favorable in terms of selectivity towards butanol with respect to the other organics. The addition of a protective layer of PDMS allows the sealing of possible pinholes; however, its thickness should be kept as thin as possible since permeation fluxes decrease with the increasing thickness of PDMS, and this material also shows greater selectivity towards acetone compared to other organics.
Supplementary Materials:
The following supporting information can be downloaded at: https: //www.mdpi.com/article/10.3390/membranes12101007/s1, Figure S1: Influence of the PDMS content in the polymer solution on the film thickness; Figure S2: TGA of polypropylene, Pebax and PDMS polymers, and a hollow fiber membrane with two dense layers (Pebax + PDMS) on a PP support; Figure S3: Differential thermogravimetric (DTG) curves of the studied materials; Figure S4: ATR-FTIR analysis for Pebax, for PDMS and for the hollow fiber composite membrane. Funding: This research was funded by Agencia Estatal de Investigación (PID2019-104369RB-I00/AEI/ 10.13039/501100011033). This work was also partially funded by the European Union through the project "HYLANTIC"-EAPA_204/2016, which is cofinanced by the European Regional Development Fund in the framework of the INTERREG Atlantic program, and by the Project ENERGY PUSH SOE3/P3/E0865, which is cofinanced by the European Regional Development Fund (ERPF) in the framework of the INTERREG SUDOE Programme. The Spanish Ministry of Science and Innovation is also gratefully acknowledged for the FPI research scholarship BES-2017-081708 (C.A.-S.).
Informed Consent Statement: Not applicable.
Data Availability Statement: Data is contained within the article or Supplementary Materials.
Conflicts of Interest:
The authors declare no conflict of interest.
Abbreviations
The following abbreviations are used in this manuscript: | 9,553.6 | 2022-10-01T00:00:00.000 | [
"Engineering",
"Materials Science"
] |
The Stokes complex for Virtual Elements with application to Navier--Stokes flows
In the present paper, we investigate the underlying Stokes complex structure of the Virtual Element Method for Stokes and Navier--Stokes introduced in previous papers by the same authors, restricting our attention to the two dimensional case. We introduce a Virtual Element space $\Phi_h \subset H^2(\Omega)$ and prove that the triad $\{\Phi_h, V_h, Q_h\}$ (with $V_h$ and $Q_h$ denoting the discrete velocity and pressure spaces) is an exact Stokes complex. Furthermore, we show the computability of the associated differential operators in terms of the adopted degrees of freedom and explore also a different discretization of the convective trilinear form. The theoretical findings are supported by numerical tests.
It was soon recognized that the more general construction of VEM, that is not limited to polynomial functions on the elements, may allow for additional interesting features in additional to polytopal meshing. An example can be found in [10,11] where the authors developed a (conforming) Virtual Element Method for the Stokes and Navier-Stokes problems that guarantees a divergence free velocity, a property that yields advantages with respect to standard inf-sup stable schemes (see for instance [33]). And, most importantly, the proposed approach fitted quite naturally in the virtual element setting, so that the ensuing element is not particularly complicated to code or to handle.
Our aim is to further develop the idea in [10,11], also in order to get a deeper understanding of the underlying structure. In [26] the term "Stokes exact complex" was introduced; in that paper the authors underline that, if a given velocity/pressure FE scheme is associated to a discrete Stokes exact complex, than not only the existence of an unique solution is guaranteed, but also the divergence-free character of the discrete velocity. In addition, this allows to construct an equivalent curl formulation of the problem in a potential-like variable. This matter is one of interest in the FEM community, see for instance [36,33], also due to the difficulty in deriving exact Stokes complexes for Finite Elements, that often yield quite "cumbersome" schemes.
In the present paper, we unveil the underlying 2D Stokes complex structure of the VEM in [10,11] by introducing a Virtual Element space Φ h ⊂ H 2 (Ω) and proving that the triad {Φ h , V h , Q h } (with V h and Q h velocity and pressure spaces of [11]) is an exact Stokes complex. Furthermore, we show the computability of the associated differential operators in terms of the adopted degrees of freedom (a key aspect in VEM discretizations) and we explore also a different discretization of the convective trilinear form. As a byproduct of the above exactsequence construction, we obtain a discrete curl formulation of the Navier-Stokes problem (set in the smaller space Φ h ) that yields the same velocity as the original method (while the pressure needs to be recovered by solving a global rectangular system). For completeness, we also briefly present and compare a stream-function formulation approach, that is based on a direct discretization (with C 1 Virtual Elements) of the continuous stream function formulation of the problem. Some numerical tests are developed at the end of the paper, in order to show the performance of the methods, also comparing aspects such as condition number and size of the linear system. We note that a related study was developed in [3], but only for the lowest order case without enhancements (that is, suitable for Stokes but not for Navier-Stokes).
The paper is organized as follows. In Section 2 we review the Navier-Stokes problem in strong and variational form, together with some basic theoretical facts. In Section 4 (after introducing some preliminaries and definitions in Section 3) we review the Virtual scheme in [11], propose a third option for the discretization of the convective term and extend the convergence results also to this case. In Section 5 we introduce the space Φ h together with the associated degrees of freedom, prove the exact Stokes complex property and state the alternative curl formulation for the discrete problem. In Section 6 we present a set of numerical tests, that also compare the proposed method with a direct C 1 discretization of the stream-function problem, briefly described in the Appendix, that is not associated to a Stokes complex.
Throughout the paper, we will follow the usual notation for Sobolev spaces and norms [1]. Hence, for an open bounded domain ω, the norms in the spaces W s p (ω) and L p (ω) are denoted by · W s p (ω) and · L p (ω) respectively. Norm and seminorm in H s (ω) are denoted respectively by · s,ω and |·| s,ω , while (·, ·) ω and · ω denote the L 2 -inner product and the L 2 -norm (the subscript ω may be omitted when ω is the whole computational domain Ω). Moreover with a usual notation, the symbols ∇, ∆ and ∇ 2 denote the gradient, Laplacian and Hessian matrix for scalar functions, while ∆, ∇, and div denote the vector Laplacian, the gradient operator and the divergence for vector fields. Furthermore for a scalar function ϕ and a vector field v := (v 1 , v 2 ) we set
The Navier-Stokes equation
We consider the steady Navier-Stokes equation on a polygonal simply connected domain Ω ⊆ R 2 (for more details, see for instance [31]) where u, p are the velocity and the pressure fields, respectively, ν ∈ R, ν > 0 is the viscosity of the fluid and f ∈ [L 2 (Ω)] 2 represents the external force. For sake of simplicity we here consider Dirichlet homogeneous boundary conditions, different boundary conditions can be treated as well. Problem (1) can be written in the equivalent rotational form Systems (1) and (2) are equivalent in the sense that the velocity solutions u coincide and the rotational pressure solution P of Problem (2), the so-called Bernoulli pressure, and the convective pressure solution p of Problem (1) are jointed by the relation where, for the time being, λ denotes a suitable constant. Let us consider the spaces Let us introduce the bilinear forms It is well known (see, for instance, [31]) that in the diffusion dominated regime, i.e. under the assumption where C denotes the continuity constant of c(·; ·, ·) with respect to the V-norm, Problem (11) is well-posed and the unique solution (u, p) ∈ V × Q satisfies Finally, let us introduce the kernel of bilinear form b(·, ·), Then, Problem (11) can be formulated in the equivalent kernel form: In this case, from (9) and (10) it is straightforward to see that
Curl and Stream Formulation of the Navier-Stokes Equations
If Ω is a simply connected domain, a well known result (see [31] for the details) states that a vector function v ∈ Z if and only if there exists a scalar potential function ϕ ∈ H 2 (Ω), called stream function such that v = curl ϕ .
Clearly the function ϕ is defined up to a constant. Let us consider the space endowed with norm Then, Problem (15) can be formulated in the following curl formulation: A different approach that makes use of the stream functions is the following. Let ψ be the stream function of the velocity solution u of (1), i.e. u = curlψ. Then applying the curl operator to the equation (2), and using simple computations on the differential operators we obtain the equivalent following problem: This elliptic equation, can be reformulated in a variational way as follows, obtaining the socalled stream formulation (we refer again to [31]): where Since the formulations (17) and (19) are equivalent to Problem (15) (in turn equivalent to Problem (11)), the well-posedness of curl and stream formulations follows from assumption (A0). Moreover from (13) follows the stability estimate
Definitions and preliminaries
In the present section we introduce some basic tools and notations useful in the construction and theoretical analysis of Virtual Element Methods. Let { Ω h } h be a sequence of decompositions of Ω into general polygonal elements E with We suppose that for all h, each element E in Ω h fulfils the following assumptions: where is a uniform positive constant. We remark that the hypotheses above, though not too restrictive in many practical cases, can be further relaxed, as investigated in [9]. For any E ∈ Ω h , using standard VEM notations, for n ∈ N let us introduce the spaces: • P n (E) the set of polynomials on E of degree ≤ n (with the extended notation P −1 (E) = ∅), • B n (∂E) := {v ∈ C 0 (∂E) s.t v |e ∈ P n (e) for all edge e ⊂ ∂E}.
Remark 3.1. Note that (23) implies that the operator curl is an isomorphism from x ⊥ P n−1 (E) to the whole P n−1 (E), i.e. for any q n−1 ∈ P n−1 (E) there exists a unique p n−1 ∈ P n−1 (E) such that q n−1 = curl(x ⊥ p n−1 ) .
We also have that where n E is the number of edges (or the number of vertexes) of the polygon E. One core idea in the VEM construction is to define suitable (computable) polynomial projections. For any n ∈ N and E ∈ Ω h we introduce the following polynomial projections: • the L 2 -projection Π 0,E n : L 2 (E) → P n (E), given by for all v ∈ L 2 (E) and for all q n ∈ P n (E), with obvious extension for vector functions Π 0,E n : [L 2 (E)] 2 → [P n (E)] 2 , and tensor func- and for all q n ∈ P n (E), with obvious extension for vector functions Π ∇,E and for all q n ∈ P n (E), Finally, let us recall a classical approximation result for P n polynomials on star-shaped domains, see for instance [17]: with C depending only on n and the shape constant in assumptions (A1) and (A2).
In the following the symbol C will indicate a generic positive constant, independent of the mesh size h, the viscosity ν and the constant γ appearing in (A0), but which may depend on Ω, the integer k (representing the "polynomial" order of the method) and on the shape constant in assumptions (A1) and (A2). Furthermore, C may vary at each occurrence.
Virtual elements velocity-pressure formulation
In the present section we outline a short overview of the Virtual Element discretization of Navier-Stokes Problem (11). We will make use of various tools from the virtual element technology, that will be described briefly; we refer the interested reader to the papers [3,10,44,11]. We recall that in [10] a new family of Virtual Elements for the Stokes Equation has been introduced. The core idea is to define suitable Virtual space of velocities, associated to a Stokes-like variational problem on each element, such that the discrete velocity kernel is pointwise divergence-free. In [44] has been presented an enhanced Virtual space to be used in place of the original one, that, taking the inspiration from [2], allows the explicit knowledge of the L 2 -projection onto the polynomial space P k (being k the order of the method). In [11] a Virtual Element scheme for the Navier-Stokes equation in classical velocity-pressure formulation has been proposed. In the following we give some basic tools and a brief overview of such scheme. We focus particularly on the virtual element discretization of Navier-Stokes equation in rotation form (2) related to the trilinear form c rot (·; ·, ·) defined in (21) that was not treated in [11]. Specifically for the resulting discrete scheme we develop the convergence analysis for both the Bernoulli and the related convective pressure.
Virtual elements spaces
Let k ≥ 2 the polynomial degree of accuracy of the method. We consider on each element E ∈ Ω h the (enlarged) finite dimensional local virtual space Now, we define the Virtual Element space V E h as the restriction of U E h given by (cf. [11]): We here summarize the main properties of the virtual space V E h (we refer [44,11] for a deeper analysis): where n E is the number of vertexes of E; • degrees of freedom: the following linear operators D V , split into four subsets (see Figure 1) constitute a set of DoFs for V E h : -D V 1: the values of v at the vertexes of the polygon E, • projections: the DoFs D V allow us to compute exactly (c.f. (26) and (25)) in the sense that, given any v h ∈ V E h , we are able to compute the polynomials Π ∇,E k v h , Π 0,E k v h and Π 0,E k−1 ∇v h only using, as unique information, the degree of freedom values The global virtual element space is obtained as usual by combining the local spaces V E h accordingly to the local degrees of freedom, as in standard finite elements and considering the homogeneous boundary conditions. We obtain the global space The space V h has an optimal interpolation order of accuracy with respect to k (see Theorem 4.1 in [11]).
where the constant C depends only on the degree k and the shape regularity constant (see assumptions (A1) and (A2) of Section 3).
The pressure space is simply given by the piecewise polynomial functions with the obvious associated set of global degrees of freedom: A crucial observation is that, by definitions (30) and (31), it holds Therefore the discrete kernel is a subspace of the continuous kernel Z defined in (14), i.e.
This leads to a series of important advantages, as explored in [34,33,11,44]. Finally, we remark that the kernel Z h is obtained by gluing the local kernels explicitly defined by where n E is the number of vertexes of E.
Discrete bilinear forms and load term approximation
In this subsection we briefly describe the construction of a discrete version of the bilinear form a(·, ·) given in (4) and trilinear forms c(·; ·, ·) (cf. (6), (7), (8)). We can follow in a rather slavish way the procedure initially introduced in [7] for the laplace problem and further developed in [10,44,11] for flow problems. First, we decompose into local contributions the bilinear form a(·, ·) and the trilinear forms c(·; ·, ·) by considering: Therefore, following a standard procedure in the VEM framework, we define a computable discrete local bilinear form approximating the continuous form a E (·, ·), and defined by with α * and α * positive constants independent of the element E. For instance, a standard choice (26) and properties (39) imply that the discrete form a E h (·, ·) satisfies the consistency and stability properties. The global approximated bilinear form a h (·, ·) : V h × V h → R is defined by simply summing the local contributions: We now define discrete versions of the forms c(·; ·, ·). Referring to (6), (7), (8) we set for all and note that all quantities in the previous formulas are computable. Let c E h (·; ·, ·) be one of the discrete trilinear forms listed above. As usual we define the global approximated trilinear form by adding the local contributions: The forms c h (·; ·, ·) are immediately extendable to the whole V (simply apply the same definition for any w, u, v ∈ V). Moreover, the trilinear forms c h (·; ·, ·) are continuous on V, i.e. there exists a uniform bounded constant C h such that The proof of the continuity for the trilinear forms c conv,h (·; ·, ·) and c skew,h (·; ·, ·) can be found in Proposition 3.3 in [11]. Analogous techniques can be used to prove the continuity of the trilinear form c rot,h (·; ·, ·). For what concerns b(·, ·), as noticed in [10] we do not need to introduce any approximation of the bilinear form, since it can be exactly computed by the DoFs D V . The last step consists in constructing a computable approximation of the right-hand side (f , v) in (11). We define the approximated load term f h as and consider:
The discrete problem
Referring to (30), (31), (40), (44), (5) and (47), the virtual element approximation of the Navier-Stokes equation in the velocity-pressure formulation is given by: with c h (·; ·, ·) given by (41), (42) or (43). Whenever the choice (43) is adopted, the pressure output in (48) approximates the Bernoulli pressure P in (3) instead of the convective pressure p. Recalling the kernel inclusion (34), Problem (48) can be also formulated in the equivalent kernel form The well-posedness of the discrete problems can be stated in the following theorem (cf. [11]).
Theorem 4.2. Under the assumption
We have the following approximation results (see Theorem 4.6 and Remark 4.2 in [11] for the choices (41) and (42)).
for a suitable functions F, H, K independent of h.
Following the same steps as in [11], Theorem 4.3 easily extends also to the choice (43). In such case we preliminary observe that if the velocity solution u ∈ [H s+1 (Ω)] 2 and the convective pressure p ∈ H s (Ω) then the Bernoulli pressure P is in H s (Ω). As a matter of fact, recalling (3) by the Hölder inequality and Sobolev embedding H s+1 (Ω) ⊂ W s 4 (Ω), we recover Now let (u h , P h ) be the solution of the virtual problem (48) with the trilinear form (43) and (u, P ) be the solution of the Navier-Stokes equation (2). Then (53) is substituted by In such case the following computable approximation p h of the convective pressure p is available: where λ h is the mean value of the piecewise polynomial function 1 2 Π 0,E k u h ·Π 0,E k u h . The optimal order of accuracy for the convective pressure can established as follows. Definitions (3) (taking λ as the mean value of 1 2 u · u) and (55) easily imply where in the second inequality we have used the fact that all terms inside the norms are zero averaged. The first term in the right hand side of (56) is bounded by (54). Whereas for the terms µ E the triangular inequality and the Hölder inequality entail Using the Sobolev embedding H 1 (Ω) ⊂ L 4 (Ω), the continuity of the projection Π 0,E k with respect to the L 4 -norm and the H 1 -norm (see, for instance, [11]), and polynomial approximation estimates on star shaped polygons of Lemma 3.1, from (57) we infer Combining bound (58) with the Hölder inequality for sequences, the velocity error estimate (52) and with the stability estimates (13) and (51), it follows for a suitable functions L and I independent of h. Finally, inserting estimates (54) and (59) in (56) we obtain the optimal convergence result for the convective pressure also for choice (43). Remark 4.1. We observe that, due to the divergence-free property, the estimate on the velocity error in Theorem 4.3 does not depend on the continuous pressure, whereas the velocity errors of classical methods have a pressure contribution, see [11] for more details on this aspect. [11]) shows that if the exact velocity solution u ∈ [P k (Ω)] 2 and the trilinear form c conv,h (·; ·, ·) or the trilinear form c rot,h (·; ·, ·) are adopted in (48), the corresponding schemes provide a higher order approximation errors, that are respectively These are to be compared with the error of standard inf-sup stable Finite Elements, that in the same situations would be O(h k ) due to the pressure contribution to the velocity error. Remark 4.3. Another interesting aspect related to method (48) is the so called "reduced" version. Noting that the discrete solution satisfies div u h = 0, one can immediately set to zero all D V 4 degrees of freedom, and correspondingly eliminate also the associated (locally zero average) discrete pressures. The resulting equivalent scheme has much less internal-to-element velocity DoFs and only piecewise constant pressures (we refer to [10,11] for more details).
Virtual elements Stokes complex and curl formulation
In the present part we present the VEM stream function space and the associated Stokes Complex.
Virtual element space of the stream functions
The aim of the present section is to introduce a suitable virtual space Φ h approximating the continuous space of the stream functions Φ defined in (16), such that In particular this will allow to exploit the kernel formulation (49) in order to define an equivalent VEM approximation for the Navier-Stokes equation in curl form (cf. (17)). Note that a related approach, but limited to a lowest order case and suitable only for the Stokes problem, was presented in [3]. In order to construct the space of the virtual stream functions Φ h , we proceed step by step, following the enhanced technique used in Subsection 4.2. On each element E ∈ Ω h we consider the enlarged local virtual space: (61) Then we define the enhanced space of the stream functions (62) It is straightforward to see that P k+1 (E) ⊆ Φ E h . We are now ready to introduce a suitable set of degrees of freedom for the local space of virtual stream functions Φ E h . Given a function ϕ ∈ Φ E h , we take the following linear operators D Φ , split into five subsets (see We note that the linear operators D Φ 1 and D Φ 2 are always needed to enforce the C 1continuity at the vertices. Moreover it is immediate to check that for any stream function ϕ ∈ Φ E h (the same holds for Ψ E h ), the linear operator evaluations of D Φ 1, D Φ 2, D Φ 3, D Φ 4 uniquely determine ϕ and its gradient on ∂E. We now prove the following result.
where as usual n E denotes the number of edges of the polygon E. Proof. We preliminary prove that the linear operators D Φ plus the additional moments of curl ϕ constitute a set of degrees of freedom for Ψ E h . An integration by parts and recalling Remark 3.1 imply that the linear operators D Ψ 5 + D Φ 5 are equivalent to prescribe the moments E ϕ q k−1 dE for all q k−1 ∈ P k−1 (E). Indeed, Remark 3.1 and simple computations give where the boundary integral is computable using the DoFs values. Now the assertion easily follows by a direct application of Proposition 4.1 in [18]. In particular, from (24) it holds that The next step is to prove that the linear operators D Φ are unisolvent for Φ E h . From (24) it holds dim P k−1\k−3 (E) = dim (P k−1 (E)) − dim (P k−3 (E)) = 2 k − 1 .
Hence, neglecting the independence of the additional (2 k − 1) conditions in (62), the dimension of Φ E h is bounded from below by Due to (65), the proof is concluded if we show that a function ϕ ∈ Φ E h such that D Φ (ϕ) = 0 is identically zero. In such case, ϕ = 0 and ∇ϕ = 0 on the skeleton ∂E and this entails curl ϕ = 0 on ∂E. Moreover we note that in this case the Π ∇,E k (curl ϕ) = 0; as a matter of fact, by definition (26), we get The boundary integral is zero being curl ϕ = 0 on the skeleton ∂E. For the internal integral, in the light of (23), let us set where the boundary integral is zero since ϕ = 0 on ∂E, whereas the second term is zero since D Φ 5(ϕ) = 0. In particular we proved that, since Π ∇,E k (curl ϕ) = 0, recalling (62) also the moments D Ψ 5 of ϕ are zero. Since D Φ (ϕ) = 0 by assumption, recalling that ϕ ∈ Φ E h ⊂ Ψ E h and that {D Φ , D Ψ 5} are a set of degrees of freedom for Ψ E h , it follows ϕ = 0.
1. An alternative way to define a unisolvent set of DoFs for the space Φ E h is to provide in the place of D Φ 5 the following operators [18] • D Φ 5 : the moments of ϕ against the polynomial of degree up to degree k − 3 but such choice is less suitable for the exact sequence construction of the present work.
The global virtual space Φ h is obtained by combining the local spaces Φ E h accordingly to the local degrees of freedom, taking into account the boundary conditions: , where n P (resp., n e and n V ) is the number of elements (resp., internal edges and vertexes) in the decomposition Ω h .
Virtual element Stokes complex
The aim of the present subsection is to provide a virtual element counterpart of the continuous Stokes complex [32]: where i denotes the mapping that to every real number r associates the constant function identically equal to r and we recall that a sequence is exact if the image of each operator coincides with the kernel of the following one. The case without boundary conditions is handled analogously. We start by characterizing the space Φ E h as the space of the stream functions associated to the discrete kernel Z E h . Proposition 5.2. For any E ∈ Ω h let Z E h and Φ E h be the spaces defined in (35) and (62), respectively. Then, it holds that Concerning the condition on the skeleton ∂E, we observe that ϕ h|∂E ∈ B k+1 (∂E) and Inside the element, by simple calculations and by definition (62), we infer curl ∆v h = curl ∆(curl ϕ h ) = ∆ 2 ϕ h ∈ P k−1 (E). In the light of Remark 3.1, the previous relation is equivalent to Therefore, there exists p k−1 ∈ P k−1 (E) such that curl(∆v h − x ⊥ p k−1 ) = 0. Since E is simply connected, there exists s such that ∆v h − x ⊥ p k−1 = ∇s. Thus we have shown that Moreover, by definition (62), for all At this point is clear that (69), (70), (71), (72) and definition (35), The proof now follows by a dimensional argument. In fact, from (63) and (36) easily follows that Therefore we can conclude that curl h , from the degrees of freedom values D Φ of ϕ h , we are able to compute the DoFs values D V of curl ϕ h . In particular it holds that and Therefore, for any ϕ h ∈ Φ E h , the DoFs D Φ allow to compute the polynomial projections . As a consequence of Proposition 5.2 we have the following Stokes exact sequence for our discrete VEM spaces and its reduced version (see also Figures 3 and 4). (62) and (28), respectively, and let V E h denote the reduced velocity space, see Remark 4.3. Then, the following sequences are exact Remark 5.3. In terms of degrees of freedom, our lowest order element (when restricted to triangles!) can be compared with the Zienkiewicz element [23,32], all other FEM elements in the literature being either higher order or needing a sub-element partition and more DoFs. The reduced version of our VEM element for k = 2 (see Remark 4.3 and Figure 4) has piecewise constant pressures and no internal degrees of freedom for velocities, and thus in terms of degrees of freedom exactly corresponds to the above finite element (the difference is that we use the VE approach instead of introducing rational basis functions). But note that the element here presented yields O(h 2 ) convergence rate for velocities and also for the local pressure average (full O(h 2 ) pressure convergence can be recovered by a local post-processing), instead of linear convergence as [32]. In addition, we avoid integration of rational functions. Clearly, this comes at the price of having a virtual formulation and thus the absence of an explicit expression of the shape functions. The following results are the global counterpart of Proposition 5.2 and Corollary 5.1.
Proposition 5.3.
Let Z h and Φ h be the spaces defined in (33) and (67), respectively. Then, it holds that Proof. We note that Proposition 5.2 endowed with the boundary condition in the definitions (33) and (67) imply curl Φ h ⊆ Z h . The proof now follows by a dimensional argument using the Euler formula. (67), (30) and (31), respectively, and let V h and Q h denote the reduced velocity space and the piecewise constant pressures, respectively, see Remark 4.3. Then, the following sequences are exact
Corollary 5.2. Let Φ E h , V h and Q h be the spaces defined in
The case without boundary conditions follows analogously.
The discrete problem
In the light of Proposition 5.2, referring to (49) and (67) we can set the virtual element approximation of the Navier-Stokes equation in the curl formulation: (77) Due to Proposition 5.3, Problem (77) is equivalent to (49). We remark that all forms in (77) are exactly computable by the DoFs D Φ . In fact, recalling Remark 5.2, the polynomials are computable on the basis of D Φ , so that, referring to (40), (44) and (47), we infer that are exactly computable from DoFs D Φ , see also The convergence of the discrete solution curl ψ h of (77) to the continuous solution curl ϕ of (17) follows immediately from Theorem 4.3, taking u = curl ϕ and u h = curl ϕ h .
Clearly Problem (77) does not provide any information on the pressure p. Nevertheless, the Stokes complex associated to the proposed scheme turns out to be very helpful if we are interested in computing suitable approximation p h of p. Indeed referring to (30) and (48), starting from the solution ψ h of Problem (77), we infer the following problem ), the previous system, is actually an overdetermined system, i.e. there are more equations than unknowns. Nevertheless the well-posedness of Problem (79) is guaranteed by Theorem 4.2. We refer to Section 6 for a deeper analysis and computational aspects of (79).
We stress that the curl virtual formulation (77) exhibits important differences from the computational point of view compared with the velocity-pressure formulation (48). First of all the linear system associated to Problem (77) has 2(n P − 1) less DoFs than Problem (48), even if considering its equivalent reduced form (see Remark 4.3). Moreover the first iteration of the Newton method applied to the the non-linear virtual stream formulation (77) results in a linear system which is symmetric and positive definite, whereas applied to the virtual element method (48) in velocity-pressure formulation leads to an indefinite linear system. These advantages come at the price of a higher condition number of the involved linear systems. Remark 5.4. Simple integration by parts gives (f , curl ϕ) = (curl f , ϕ). By Remark 5.1, the DoFs D Φ allow us to compute the L 2 -projection Π 0,E k−1 : Φ E h → P k−1 (E), so that we can consider a new computable right hand-side This new formulation of the right-hand side gets the same order of accuracy of the original one.
In particular if the external force is irrotational, i.e. f = ∇f , we improve the error estimate in (52) by removing the dependence of the error by the load. More generally, with the choice (80), we completely remove the influence in the error stemming from the irrotational part in the Helmholtz decomposition of the load. Clearly (80) can be applied only when f is given as an explicit function.
Numerical Tests
In this section we present two sets of numerical experiments to test the practical performance of the proposed virtual element methods (77), also compared with a direct C 1 VEM discretization of the stream formulation (18) described in the Appendix, see equation (87). For the scheme (77), in all tests we investigate the three possible options for the trilinear form in (41), (42), (43). In Test 6.1 we study the convergence of the proposed virtual element schemes (77) and (87) for the discretization of the Navier-Stokes equation in curl formulation and stream formulation respectively. A comparison of (77) (in terms of errors, number of DoFs, condition number of the resulting linear systems) with the equivalent virtual element scheme (48) for the Navier-Stokes equation in velocity-pressure formulation is also performed. In Test 6.2 we consider a benchmark problem for the Navier-Stokes equation (18) with the property of having the velocity and stream solution in the corresponding discrete spaces. It is well known that classical mixed finite element methods lead to significant velocity errors, stemming from the velocity/pressure coupling in the error estimates. This effect is greatly reduced by the presented methods (cf. Theorem 4.3, estimate (52) and Remark 4.2). In order to compute the VEM errors, we consider the computable error quantities: for the velocity-pressure formulation (48) and for the curl and stream formulations (see (77) and (87), respectively). For what concerns the pressures we simply compute the standard L 2 error error(p, L 2 ) := p − p h 0 . For the computation of the discrete pressure for the virtual element scheme (77) we follow (79) and solve the overdetermined system by means of the least squares method. We briefly sketch the construction of the least square formula.
be the canonical basis functions of V h and let us denote with r h the vector with component i.e. r h contains the values of the degrees of freedom associated to the right hand side of (79) with respect to the basis be the canonical basis functions of Q h and for any piecewise polynomial p k−1 ∈ Q h we denote with p h the vector containing the values of the coefficients with respect to the basis {q i } associated to p k−1 . Then the least squares formula associated to (79) is An example of the adopted meshes is shown in Figure 5. For the generation of the Voronoi meshes we use the code Polymesher [42]. The distorted quadrilateral meshes are obtained starting from the uniform square meshes and displacing the internal vertexes with a proportional "distortion amplitude" of 0. 3 We test the virtual element scheme (77). In Figures 7 and 8 we show the results obtained with the sequences of Voronoi meshes V h and quadrilateral meshes Q h , by considering the three possible choices of the trilinear forms. We stress that in all cases considered we compare the discrete convective pressure p h (for the trilinear form in c rot,h (·; ·, ·) we consider the definition (55)).
We notice that the theoretical predictions of Section 5 are confirmed. Moreover, we observe that the virtual element methods obtained with the three different trilinear forms exhibit almost identical results, at least for this example and with the adopted meshes. In Figures 7 (left) and 8 (left) we also depict the error for the direct C 1 discretization of the stream formulation (87), that follows a similar behaviour to (77). Note that we do not compute a pressure error for scheme (87) since the computation of a discrete pressure is a more complex issue in this case, see Remark 7.1. Finally we test the corresponding virtual element method (48) with the same sequences of polygonal meshes V h , Q h . Table 1 shows the results obtained respectively with VEM (48) and (77) obtained considering the trilinear form c rot,h (·; ·, ·). The results are analogous also for the other two proposed trilinear forms (not shown). In Table 2 we compare the number of DoFs and the condition number of the resulting linear systems V h 1/8 3.704032467e-1 3.891840615e-1 3.704032467e-1 3.891840615e-1 1/16 9.153568669e-2 8.875084726e-2 9.153568669e-2 8.875084726e-2 1/32 2.308710367e-2 1.994452869e-2 2.308710367e-2 1.994452869e-2 1/64 5.791512013e-3 4.602515029e-3 5.791512013e-3 4.602515029e-3 Q h 1/10 3.047752518e-1 3.714633884e-1 3.047752518e-1 3.714633884e-1 1/20 8.709526360e-2 8.363888240e-2 8.709526360e-2 8.363888240e-2 1/40 2.188243443e-2 1.945853612e-2 2.188243443e-2 1.945853612e-2 1/80 5.523374104e-3 4.762632907e-3 5.523374104e-3 4.762632907e-3 (stemming from the fist iteration of the Newton method) for both formulations (48) and (77). As observed in Section 5, the scheme (77) has the advantage of having (2 n p − 2) less of unknowns, even when considering the reduced version (see Remark 4.3) for formulation (48). The drawback is that the condition number of the system resulting from the velocity-pressure scheme (48) behaves as h −2 , while the asymptotic rate of the condition number of the linear system resulting from the scheme (77) We here mention only the main properties of the virtual space Φ E h and refer to [18,4] for a deeper description: , that is the same dimension of Φ E h (cf. (63)); • degrees of freedom: the linear operators D Φ : D Φ 1, D Φ 2, D Φ 3, D Φ 4, D Φ 5 (see Remark 5.1) constitute a set of DoFs for Φ E h ; • projections: the DoFs D Φ allow us to compute exactly (c.f. (27) and (25)) in the sense that, given any ϕ h ∈ Φ E h , we are able to compute the polynomials Π ∇ 2 ,E k+1 ϕ h , Π 0,E k−1 ∆ϕ h , Π 0,E k−1 ϕ h , Π 0,E k−1 ∇ϕ h , and Π 0,E k−1 curlϕ h only using, as unique information, the degree of freedom values D Φ of ϕ h .
The global virtual element space is obtained as usual by combining the local spaces Φ E | 8,884.4 | 2018-07-27T00:00:00.000 | [
"Materials Science"
] |
A Concrete Composite 2-Higgs Doublet Model
We consider a Composite Higgs Model (CHM) with two isospin doublet Higgs fields arising as pseudo Nambu-Goldstone bosons from a ${\rm SO}(6)\to {\rm SO}(4)\times {\rm SO}(2)$ breaking. The main focus of this work is to explicitly compute the properties of these Higgses in terms of the fundamental parameters of the composite sector such as masses, Yukawa and gauge couplings of the new spin-1 and spin-1/2 resonances. Concretely, we calculate the Higgs potential at one-loop level through the Coleman-Weinberg mechanism from the explicit breaking of the ${\rm SO(6)}$ global symmetry by the partial compositeness of fermions and gauge bosons. We derive then the phenomenological properties of the Higgs states and highlight the main signatures of this Composite 2-Higgs Doublet Model at the Large Hadron Collider, including modifications to the SM-like Higgs couplings as well as production and decay channels of heavier Higgs bosons. We also consider flavour bounds that are typical of CHMs with more than one Higgs doublet.
Introduction
While the discovered Higgs boson is consistent with the Standard Model (SM) one, this could just be the first manifestation of an Electro-Weak Symmetry Breaking (EWSB) dynamics that is far richer than the minimal one existing in the current prevalent description of Nature. On the one hand, the latter is non-minimal in both its matter (as there are three generations of quarks and leptons) and interaction (as multiple gauge bosons states of different multiplicities exist) content, so that one is well motivated to postulate a non-minimal Higgs sector too. On the other hand, bearing in mind that the discovered Higgs state has a doublet construction, one is well justified in pursuing first, in the quest for some Beyond the SM (BSM) physics, the study of 2-Higgs Doublet Models (2HDMs). In fact, these scenarios always include a neutral scalar Higgs state that can play the role of the detected one. Furthermore, these constructs offer additional (pseudo)scalar states, both neutral and charged, amenable to discovery by the ATLAS and CMS collaborations, which are now substantially engaged in direct searches for new Higgs bosons, in parallel with extracting a possible BSM dynamics indirectly from the precision measurements of the detected one.
However, 2HDMs do not have the ability to solve the so-called hierarchy problem of the SM. An elegant way to overcome it is to presume that the revealed Higgs state and its possible 2HDM companions are not fundamental particles, just like any spin-0 object discovered so far in Nature. In this sense, one would be interpreting these (pseudo)scalar states belonging to a Composite 2HDM (C2HDM) as (fermion) composites, i.e., mesonic states of a new theory of strong interactions not dissimilar from QCD. A phenomenologically viable possibility, wherein the mass of the lightest Higgs state is kept naturally lighter than a new strong scale (of compositeness, f , in the ∼ TeV region) is, in particular, the one of assigning to these QCD-like states a pseudo-Nambu-Goldstone Boson (pNGB) nature, like in Composite Higgs Models (CHMs) arising from the spontaneous symmetry breaking around the TeV scale of the global symmetry of such a new strong sector [1]. The residual symmetry is then explicitly broken by the SM interactions through the partial compositeness paradigm [2,3]. In the minimal CHM [4,5,6,7,8,9,10,11] the lone Higgs state is a pNGB (surrounded by various composite resonances, both spin-1/2 and spin-1, generally heavier). Hence, it is natural to assume that the new (pseudo)scalar Higgs states of a C2HDM can also be created as pNGBs.
Such C2HDMs embedding pNGBs, which arise from a new confining strong dynamics, can be constructed by explicitly imposing a specific symmetry breaking structure. Herein, we will analyse 2HDMs based on the spontaneous global symmetry breaking of a SO(6) → SO(4) × SO(2) symmetry [12]. Within this construct, one can then study both the deviations of C2HDM couplings from those of a generic renormalisable Elementary 2HDM (E2HDM) [13] as well as pursue searches for new non-SM-like Higgs signals.
We explicitly construct here a C2HDM making only a few specific assumptions about the strong sector, namely, the global symmetries, their pattern of spontaneous breaking and the sources of explicit breaking (as intimated, in our approach, they come from the couplings of the new strong sector with the SM fields) and by generalising to the coset SO(6)/SO(4) × SO(2) the 2-site minimal construction developed in [6]. (We will also show in Appendix A the equivalence with the standard prescription by Callan, Coleman, Wess and Zumino (CCWZ) [14,15].) The scalar potential is in the end generated by loop effects a la Coleman-Weinberg (CW) [16] and, at the lowest order, is mainly determined by the free parameters associated to the sole top-quark and the complete gauge sector [12].
Calculations of the Higgs potential in the composite realisation of 2HDMs have been performed in the pioneering work [12] based on the CCWZ technique, in which each coefficient of the potential is expressed in terms of invariants of the global symmetry computed as an expansion in the parameters responsible for the partial compositeness. This approach is quite general, but the undetermined O(1) coefficients associated to each invariant prevent one to exploit the dependence on the masses and couplings of the resonances generated by the new strong sector. Furthermore, the computation of such coefficients is crucial to make a clear connection with the parameters of the strong dynamics which depends on the choice of the model setup.
As mentioned before, we adopt an explicit 2-site model based on [6] originally developed in the context of minimal CHMs governed by the SO(5) symmetry and here extended to SO (6). These models are composed of two sectors, i.e., an elementary one including particles whose quantum numbers under the SU(2) L × U(1) Y gauge symmetry are the same as those of the SM fermions and gauge bosons plus a composite sector having new spin-1 and spin-1/2 resonances introduced as multiplets of the global group. The mixing between states in these two sectors realise the partial compositeness. With this construction, we can evaluate observables in the Higgs sector such as masses and couplings. This analysis also allows to clarify the differences between these observables and those from renormalisable E2HDMs [17]. The aim of the paper is to show that a composite scenario could give rise to a concrete realisation of a 2HDM and also to highlight the phenomenological aspects which could reveal at the Large Hadron Collider (LHC) the composite nature of the Higgs states described by our construction.
The plan is as follows. In Sec. 2 we describe the general features of a C2HDM based on SO(6) → SO(4) × SO(2) and we discuss the corresponding symmetries. In Sec. 3 we present the explicit model on which our analysis is based. In Sec. 4 we compute explicitly the Higgs potential and we discuss in Sec. 5 the Higgs boson couplings to fermions and bosons as well as amongst themselves. In Sec. 6 we present the Higgs spectrum of the model and discuss some phenomenological results which may act as smoking gun signals of the C2HDM. In addition, we comment on the implications from flavour constraints. Finally, we conclude in Sec. 7. We leave to the Appendices the connection with the CCWZ construction and other technical details.
The SO(6)→SO(4)×SO(2) symmetry breaking
In this section we discuss the main aspects of C2HDMs highlighting the general properties that follow by their constructions as effective field theories. The scenarios we consider are schematically characterised by the following structure: where L 2HDM has the same form as the Lagrangian of the E2HDM and contains the kinetic terms, the scalar potential (up to quartic terms) and Yukawa interactions, with H 1 and H 2 being the isospin scalar doublets and The L d≥6 element includes effective operators (starting from dimension 6) that can generate modifications to the Higgs couplings to bosons and fermions, hence effects in specific experimental observables in Higgs and flavour physics as well as global Electro-Weak (EW) precision tests. In general, these effective operators generate effects that are suppressed by v 2 /f 2 (with v being an EW scale parameter connected to the Higgs doublet Vacuum Expectation Values (VEVs) and f the scale of compositeness), however, larger suppressions can be achieved by virtue of some approximate symmetries of the underlying composite dynamics. We will compute here, through an expansion in v 2 /f 2 , the leading contributions to the 2HDM parameters m 2 i (i = 1, ...3) and λ j (j = 1, ...7) originating from the explicit breaking of the global symmetry. We then obtain the phenomenological observables, such as masses and couplings, that were only estimated in [12] on the basis of symmetry arguments and produce explicitly the low energy particle spectrum of the C2HDM.
In order to be concrete we need to choose a coset space and describe how the global symmetries are explicitly broken by the elementary sector. In the remainder of this work we have as a main focus the model G H = SO(6) SO(4) × SO (2) , expanding upon the work presented in [12] and recently discussed in [17]. The NGB fluctuations are described by the matrix U in the vector representation of SO(6) where φ 1,2 are two real fourplets (the two Higgs doublets) that can be rearranged into two SU (2) doublets as Besides the NGBs and the elementary SM fields, the model describes the extra spin-1 and spin-1/2 resonances. While the representations of the spin-1 states are fixed by the gauge symmetry, the model allows for some freedom in the choice of the fermion ones and in the embedding of the elementary fermions in representations of G. We choose: • chiral elementary fermions in the 6 of SO(6), • vector-like composite fermions in the fourplet ψ I and doublet ψ α of SO(4)×SO (2).
The resonances and their interactions with the elementary sector are then fully described by the following Lagrangian The last part describes also the derivative interactions of the NGBs with the composite matter fields and can be parameterised by means of the CCWZ formalism as shown in Appendix A. The interactions of the NGBs are suppressed by 1/f and are H-symmetric while the resonances have an overall mass scale of size m * . The elementary sector contains the gauge and fermionic kinetic terms for the SM-like fields while the mixing term is the crucial ingredient since all the phenomenology strongly depends upon the interactions between the elementary and composite sectors. Concretely, the mixing Lagrangian contains the partial compositeness terms that generate masses for the SM fermions, where i, j are flavour indices that run over the three families and we wrote schematically with a dot all the possible invariants that can be formed (see the next section for the actual implementation). Then, upon integrating out the resonances of the composite sector, we generate several effects that would allow us to match L C2HDM in Eq. (8) to the reference Lagrangian L Composite of Eq. (1). In the spirit of CHMs with partial compositeness, the parameters that enter the two sectors in L Composite , i.e., the usual part L 2HDM and the new one L d≥6 , are related to each other, since all the Higgs interactions and effective operators with SM fields are mainly generated by the explicit breaking of the global symmetry under which the Higgs doublets then behave as pNGBs. In order to set the stage for the discussion of our C2HDM we now quickly recall the main aspects of CHMs with partial compositeness.
Custodial and discrete symmetries
A renormalisable 2HDM never faces custodial breaking effects at tree level. This can be traced back to the presence, when the hypercharge coupling is neglected, of a large SU(2) L ×Sp(4) symmetry in the kinetic terms of the two Higgs doublets. Since in the renormalisable E2HDM there are no terms in the Lagrangian that contribute to theT parameter other than the kinetic terms, no custodial violation is present for any number of Higgs doublets.
In CHMs, the non-linearities of the effective Lagrangian for NGBs contribute with operators of dimension 6 of the following form which do not respect the Sp(4) symmetry and contribute to theT parameter for generic VEVs of the two Higgs doublets. However, the value of the coefficients c,c's is constrained by the symmetry of the subgroup H. This in turn suggests that only models where the unbroken group contains H ⊃ SU(2) L ×Sp(4) are free from tree level violation of custodial symmetry for any form of the Higgs VEVs. This is not the case for SO(4)×SO(2), which does not contain the full symmetry of the renormalisable kinetic terms, therefore, in our case the coefficients in Eq. (10) are non-vanishing and fixed by the symmetries, which then predict aT parameter [12] such that Since custodial breaking is sensitive to the combination Im[ H 1 † H 2 ] there are two approximate symmetries, discussed in the following, that can be used to reduce these effects: i) CP, which is well approximated in the SM; ii) a new symmetry, C 2 , that forbids a VEV for one of the two Higgs doublets.
CP invariance
In this case we realise a scenario where the two Higgs doublets have VEVs aligned in phase as required by the vanishing of the contribution in Eq. (11). Without a very accurate alignment, the bound coming from precision tests can be roughly estimated as δT < 10 −3 , which then constrains the phase misalignment ∆φ = φ 1 − φ 2 , defined through H 1,2 . Such a value can be achieved by assuming an approximate CP symmetry in the scalar potential. Interestingly, the interactions of the NGBs among themselves and with other composite fields respect automatically charge conjugation C since H i → H * i is realised on the real degrees of freedom φ 1,2 encoded in the matrix U as which is an element of SO(4). Because of this argument, we find it rather natural to consider the scenario where CP is a good symmetry of the composite sector and very well approximated in the elementary couplings (needed to comply with flavour constraints).
C 2 invariance
Another possibility to control the deviations in Eq. (11), as extensively discussed in [12], is to make the stronger assumption that one of the two Higgs doublets has a discrete symmetry that forbids any VEV, e.g., which, contrary to C, is not a symmetry of the composite sector in the sense that it is not an element of H 1 . Although this condition is not what is strictly required from Eq. (11), it has an interesting byproduct since it selects a specific pattern of Higgs couplings to fermions, because only one Higgs doublet is coupled to them. As well known, in this case, any tree level mediations of Flavour Changing Neutral Currents (FCNCs) from the scalar sector are absent for a generic flavour structure of the Yukawa couplings.
Flavour structure
When CP is the only discrete symmetry acting on the Higgs doublets, the Yukawa couplings of the renormalisable 2HDM are of the following form: Therefore, only under the assumption that the coefficients a's are the identity in flavour space, the above interactions do not generate Higgs-mediated FCNCs at tree level. Under this assumption, FCNCs are therefore confined to loop effects as in the SM. In a C2HDM the above description is modified by the presence of higher dimension operators that contribute to the Yukawas of Eq. (15) and in general one would expect any kind of operator of the form κ ijk ψψH i H † j H k + h.c. However, thanks to the pNGB nature of the Higgs doublets, the structure of the higher-dimensional operators is highly constrained by the symmetry of the theory and in general [18] the Yukawa terms including all the non-linearities are simply where the functions F u 1,2 are trigonometric invariants of H i and start with a linear term in H 1,2 , respectively. Therefore, the elementary case of Eq. (15) is automatically included as a specific case in Eq. (16) and this shows that the assumption of aligned Yukawa couplings is not a stronger requirement in the composite scenario than in the elementary one.
A difference between the elementary and composite cases arises though when one considers the additional constraints on the theory induced by the alignment of Eq. (16). In other words, while in the elementary case Eq. (15) is the only possible source of flavour violation, in the composite one L Composite contains four-fermion operators generated by integrating out the composite fermions and vectors of the form with ψ being a SM fermion, which can mediate FCNCs at tree level if the flavour coefficients x ijkl are generic, and where we neglected the precise chirality structure of fermions. This shows that the aligned structure of Eq. (16) is not sufficient to avoid tree level effects, since a diagonalisation of the Yukawas may still leave the set of four-fermion operators misaligned with respect to the mass basis. Notice also that, despite the four-fermion operators of Eq. (17) originate from different effects than (pseudo)scalar-mediated FCNCs, they are not less harmful and, more importantly, they are present also in the C 2 symmetric case. A mechanism to generate an approximate alignment between the flavour structures of Eqs (16) and (17) is possible under the working assumption of a flavour universal composite sector. In this case the latter enjoys a symmetry G 5 F that commutes with H while the elementary sector has the usual U(3) 5 flavour symmetry.
Only by the explicit breaking of G 5 F and U(3) 5 down to baryon number the CKM structure of the Yukawa sector can be reproduced: in absence of this breaking the SM fermions are in fact massless. The couplings y,ỹ in Eq. (9) break explicitly this symmetry and, depending on their structure in flavour space, they may lead to misaligned (with respect to the mass basis) flavour interactions.
Several possibilities have been considered in the literature to prevent large tree level flavour violations in composite models [19,20]. In particular, it is worth mentioning the following.
1. Higgs-mediated tree level FCNCs are absent when there is only one flavour structure per SM representation. This means that the invariants in L mixing given in Eq. (9) need to be aligned in flavour space (e.g., y ij L ∝ỹ ij L ). For any form of y ij in flavour space then Higgs-mediated FCNCs are zero at tree level and appear only at loop order.
2. Tree level effects in the four-fermion operators of Eq. (17) are much suppressed when, in addition to the assumption in point 1, one also realises a partial alignment of y ij with the CKM matrix.
We will work under these assumptions in order to realise a flavour symmetric composite sector.
An explicit model
When parameterising a composite sector one is faced with a few practical approximations that are needed to capture its main features. Since we would like to focus on the connection between Higgs phenomenology and the spectrum of heavier resonances, we adopt a description of the composite sector based on a 2-site model, as a generalisation of [6], as already intimated. We consider here a simplified picture that includes the minimal amount of new resonances that allow for a calculable Higgs potential [6]. Here, we focus on the gauge sector. Despite we are interested in the full coset structure the consistent inclusion of spin-1 resonances requires additional (gauged) symmetries as typical of 2 and 3-site models [21]. In principle, we should expect any type of resonances classified accordingly to the unbroken group H, however, since we are mainly interested in deriving the Higgs potential, only spin-1 resonances associated to SO(4) × SO(2) would play a major role 2 .
In the 2-site scenario, the Lagrangian of the gauge sector can be written as where two copies, G 1,2 , of the symmetry group G = SO(6) × U(1) X characterise the two sites. Here, G 2 is a local group and describes the spin-1 resonances through the gauge fields ρ X µ and ρ A µ , with A ∈ Adj(SO(6)). Further, G 1 is global with only the SU(2) L × U(1) Y component lifted to a local subgroup. The corresponding elementary SM gauge fields are conveniently embedded into spurions, A A µ and X µ , of the adjoint of G 1 , and g A , g X are the corresponding gauge couplings. The U 1 field in Eq. (19) is commonly dubbed link due to its transformation properties under both symmetry groups of the two sites it connects, namely, Notice that g 1 is an element of the global G 1 group in the first site while g 2 belongs to the local G 2 in the second site. By virtue of the EW gauging, the SU(2) L × U(1) Y component of the global symmetry on the first site is promoted to a local one. The Σ 2 field defined on the second site transforms, instead, under the local group G 2 . Having specified the transformation properties of the fields introduced in Eq. (19), the covariant derivatives are easily worked out to be where A µ ≡ A A µ T A + X µ T X and ρ µ ≡ ρ A µ T A + ρ X µ T X , with T A and T X being the generators of SO(6) and U(1) X , respectively. The link field U 1 realises the spontaneous symmetry breaking of G 1 × G 2 to the diagonal component G while the VEV of Σ 2 accounts for the breaking to SO(4) × SO(2) × U(1) X . The 2-site construction is schematically depicted in Fig. 1. This breaking pattern provides 24 NGBs, 16 of which are reabsorbed in the longitudinal components of the gauge fields, while the remaining 8 can be identified with Higgs fields. In the unitary gauge, where the physical degrees of freedom are clearly evident, the U 1 and Σ 2 are given by U i = exp i f f 2 i Π and Σ 2 = U 2 Σ 0 U T 2 with Π being the usual NGB matrix given in Eq.
., 6 and f −2 = f −2 1 + f −2 2 , following from the canonical normalisation of the Higgs kinetic term. Before EW gauging, the model possesses an unbroken SO(4) × SO(2) × U(1) X symmetry and the masses of the spin-1 resonances are described by the following relations: where the first two characterise the resonances spanning the unbroken group while the last one is related to the broken sector. The gauging of the EW subgroup explicitly breaks SO(4) × SO(2) × U(1) X and induces a mixing between the elementary and composite fields as well as corrections to the masses defined above. Upon integrating out the spin-1 resonances, we obtain the following effective Lagrangian in momentum space up to quadratic terms: where Σ = U 1 Σ 2 U T 1 and P T µν is the projection operator P T µν = η µν − q µ q ν /q 2 . The form factorsΠ are determined by the parameters of the strong sector, namely, by the masses and couplings of the resonances, with their explicit expressions given in Appendix B. At q 2 = 0, theΠ 1 form factor can be fixed in terms of the SM gauge masses. Indeed, keeping only the CP-even components of the Σ matrix and removing the non-dynamical fields, the W and Z gauge boson masses are given by where θ W is the Weinberg angle and we recall that v 2 = v 2 1 + v 2 2 , with v 1,2 the VEVs of the two CPeven Higgs boson components. The form factor Π W is normalised in order to correctly reproduce the canonical kinetic term of the SM gauge fields. From Eq. (23) we can finally identify where v SM 246 GeV is the SM VEV and g is the SU(2) L gauge coupling. As usual, the corrections, with respect to the SM, will be parameterised by ξ = v 2 SM /f 2 . Differently from the gauge sector, which is model independent and fixed only by the symmetry group of the strong dynamics, the fermion sector is not uniquely determined due to the possibility to choose different group representations for the fermionic fields. In this work we opt for the simplest one in which the SM fermions are embedded into the fundamental of SO (6). Another scenario, for instance, envisages the 20 representation [12].
For the sake of simplicity, we consider only the third generation and focus ourself on the top quark contributions. Indeed, all the other SM quarks provide only sub-leading corrections to the Higgs effective potential. Needless to say, all the other fermions can be included, if necessary, by simply extending the formalism described below.
In order to construct a SO(6) × U(1) X invariant Lagrangian, it is useful to embed the top quark using the spurion method into a complete representation of SO(6) with X = 2/3. More precisely, the 6 of SO(6) decomposes into the (4, 1) ⊕ (1, 2) of SO(4) × SO(2). The L-hand top quark doublet q L has a unique embedding into the (4, 1) 2/3 while for the R-handed component of the top quark t R , described by the (1, 2) 2/3 , an extra angle θ t parameterises the ambiguity of the embedding in the 6, since the fundamental representation contains two SU(2) L singlets. The R-handed component of the bottom quark b R is coupled to the (1, 2) −1/3 of 6 −1/3 and, due to U(1) X invariance, a second embedding for q L (in another 6 −1/3 ) is needed in order to generate the bottom mass. The embedding of the τ lepton follows the same line of reasoning of the bottom quark with X = −1. In particular, the L-hand (doublet) and R-handed (singlet) components are promoted to spurions with α being the SU (2) L index and the spurion VEVs defined as and Υ τ L,R = Υ b L,R with b → τ . The spin-1/2 resonances of the top quark are described by 6-plets Ψ with X = 2/3. As 6 2/3 = (4, 1) 2/3 ⊕(1, 2) 2/3 , the former delivers two SU(2) L doublets with hypercharge 7/6 and 1/6, respectively, while the latter delivers two SU(2) L singlets with hypercharge 2/3. After EWSB we count four top partners with electric charge Q = 2/3, one bottom partner with Q = −1/3 and one exotic fermion with Q = 5/3. For the down type quarks and charged leptons, we can analogously introduce other spin-1/2 resonances with X = −1/3 and −1, respectively. The Lagrangian of the composite fermion sector is where the covariant derivatives of the elementary fermions include the interactions with the elementary gauge bosons while the covariant derivative of the resonance Ψ provides the couplings to the spin-1 resonances introduced above. In the following, we will restrict ourselves to a realisation with I = 1, 2 fermionic resonances. The dimensionful parameters ∆ L and ∆ R induce a mixing between the elementary and composite fermions and, as such, explicitly break the SO(4)×SO(2)×U(1) X symmetry. All the parameters in the above Lagrangian are taken to be real in order to realise a CP invariant scenario. Moreover, we assume the LR structure of the fermionic Lagrangian discussed in [6] which allows us to simplify the parameterisation of the spin-1/2 resonances and to reduce the number of free couplings. This construction is sketched in Fig. 2. This represents the minimal choice able to generate the SM Yukawa interactions and to guarantee the Ultra-Violet (UV) finiteness of the CW potential.
Notice that the LR assumption requires ∆ 2 Upon integrating out the heavy resonances Ψ I , the effective Lagrangian takes the form where the explicit expression of the form factors is given in Appendix B. Due to Σ † = −Σ, with Σ = U 1 Σ 2 U T 1 , the hermeticity of the Lagrangian implies that the form factorsΠ qt 1 andΠ t 1 are purely imaginary whileΠ qt,t 0 andΠ qt,t 2 are real. The form factorsM t 1 andM t 2 can be complex but we will restrict them to real values, as a consequence of the reality of the strong sector parameters.
It is important to notice that the above Lagrangian has non-canonically normalised fields, therefore, when computing a given quantity, one has to take this fact into account. The form factors in Eq. (28) are listed in Appendix B.
The mass spectrum of the top-partners can be extracted from the poles and zeros of the form factors of Eq. (28). For instance, before EWSB the spectrum of the heavy top-partner resonances is given by • two 2 7/6 with masses m 4 ,m 4 from the poles ofΠ qt 0 , • two 2 1/6 with masses m Q ,m Q from the zeros ofΠ qt 0 , • two 1 2/3 with masses m T ,m T from the zeros ofΠ t 0 , • two 1 2/3 with masses m 1 ,m 1 from the poles ofM t 1 , where, e.g., 2 7/6 denotes a SU (2) L doublet with hypercharge 7/6. The masses listed above are functions of the fundamental parameters and, in particular, only m Q ,m Q and m T ,m T get corrections from the elementary/composite mixings ∆ L,R .
The Higgs potential
As already pointed out, the elementary sector is defined by the SM fermions and the gauge fields that linearly couple to operators of the strong sector and explicitly break its symmetry. As a result, the NGB symmetry becomes only approximate and the scalar potential for the Higgs is radiatively generated together with the SM gauge boson and fermion masses. Once the symmetries of the strong sector are fixed and the representations of the fermion fields are chosen, the computation of the effective potential can be carried out without a complete knowledge of all the details of the strong UV dynamics. Indeed, it can be entirely expressed in terms of the form factors introduced in Eqs. (22) and (28) which have been obtained after the integration of the heavy resonances. The CW effective potential, at one-loop order in perturbation theory and up to the fourth power in 1/f , is formally written as show the same structure of the Higgs potential in the renormalisable E2HDM. The normalisation of the parameters has been fixed in Eq. (3) and they read as where their explicit dependence on the form factors is given below. Here we focus only on the leading top quark and gauge contributions as well as on the CP-conserving scenario while we allow, at the same time, for an explicit C 2 breaking (differently from [12] where the C 2 symmetry is enforced). The former is realised if the parameters of the strong sector are real and the embedding of the R-handed top is aligned in the θ t = 0 direction. In this configuration, all the form factors are real and, in particular, Π t,qt 1 = 0. The contribution of the gauge bosons is given by The relation between the form factors appearing in the previous equations and those from Eq. (22) is worked out in Appendix B. The fermion contribution to the parameters of the scalar potential is The quadratic as well as quartic parameters and, therefore, the masses and couplings of the Higgs bosons are completely predicted by the strong sector. Notice also that in our construction
Structure of the scalar potential
In order to understand the relevance of the various terms in the potential we can organise our discussion based on the presence of accidental symmetries and on the amount of breaking of the shift symmetry due to the elementary couplings.
Gauge contribution
From the above discussion we notice that the gauge contribution to the Higgs potential respects the SO(2) symmetry that rotates the two scalar four-plets and is both CP and C 2 conserving for any value of the parameters. Moreover, the explicit expressions ofΠ W,B are proportional to the ratio g 2 /g 2 ρ and, by construction, give a positive contribution to the mass terms of the two Higgs doublets, which are proportional to and grows with the mass of the spin-1 resonances. In isolation, the gauge radiative corrections are not sufficient to break the EW symmetry.
Fermionic contribution
The effect of the fermion sector on the Higgs potential is more complicated due to the fact that the elementary-composite mixings of the L-handed doublet q L and R-handed t R top break different symmetries. In order to treat the fermionic contribution on the same footing as the gauge one, it is convenient to introduce the dimensionless couplings y L,R = ∆ L,R /f . The presence of a C 2 breaking in the composite sector generates a non-vanishing m 2 3 and λ 6,7 . At the approximation we are working, these contributions are proportional to each other. It is important to stress that the effects of C 2 breaking appear at quartic order in the couplings y L,R , as the scaling is given by while the contributions to the mass terms m 2 1,2 start at the quadratic order. Accidentally, we also notice that the fermionic contribution to λ 3 vanishes when the composite sector displays a C 2 symmetry.
Comparison with other studies
The presence of the C 2 breaking terms represents the main difference with respect to the scenario discussed in [12] which, instead, focused on the CP and C 2 invariant configuration. Incidentally, the parities under CP and C 2 of each of the operators that can be generated at one loop, and that have been classified in [12], are the same, with the only exception of one contribution proportional to where δ ij and mn sum, respectively, over the SO(4) and SO (2) indices. This operator provides the possibility to construct a CP invariant model that is not C 2 symmetric and to eventually realise the scenario that we have considered in this work.
Parameters of the model
Before moving to the characterisation of the scalar spectrum, we stress again that the effective potential may be, in general, UV divergent and that an explicit realisation of the strong sector requires, at least, two heavy fermions. Among the possible structures of the parameters controlling the strong dynamics, the LR one presented in [6] and discussed above provides the most economical condition for a calculable potential. In contrast, the gauge contributions do not give rise to UV singularities at one-loop level. Moreover, for the sake of simplicity, we can also require f 1 = f 2 and g ρ = g ρ X , which gives us eight free input parameters to the Higgs sector of the C2HDM, namely, Under these assumptions, all the results presented below have been obtained with parameters scanned in the ranges 600 Before moving to the study of the EWSB dynamics and the characterisation of the spectrum of the composite Higgs bosons, we show in Fig. 3 the parameters of the scalar potential in the general basis as functions of f that represent the main outcome of the constraints imposed by the strong dynamics. Of some relevance here is to establish a contact with the results of Refs. [22], [23] and [24], where the following parameter region of the C2HDM scalar potential was investigated: λ 6,7 λ 1,2,3,4,5 , mimicking the one normally exploited in investigations of E2HDMs with a softly broken Z 2 symmetry, namely λ 6 = λ 7 = 0 and m 2 3 = 0. To this end, we show in Fig. 4 the population of points, extracted from those in Fig. 3, which satisfy the requirements λ 6,7 < 0.1 λ 1,2,3,4,5 , plotted over the plane (m 2 3 /m 2 1 , m 2 3 /m 2 2 ). As the values of the mass parameters were in Refs. [22], [23] and [24] taken within an order of magnitude of each other, over the corresponding region of parameter space of Fig. 4, the same results found therein can be adopted for our C2HDM construct as well.
EWSB and the significance of tan β
Differently from an E2HDM scenario in which all the Lagrangian parameters (m 2 i , λ j ) appearing in Eq. (3) can be taken as free 3 , all the masses and couplings in the Higgs potential of the C2HDM are predicted by the strong dynamics. Therefore, achieving EWSB successfully in the C2HDM is not straightforward and a given amount of tuning is always necessary. Moreover, the potential of a CPconserving 2HDM can allow, in general, two separate minima [25] and one has to make sure that the EW one corresponds to a stable configuration. In our analysis we have explicitly checked that, if this particular configuration is realised, the EW vacuum always corresponds to the global minimum. In addition, we have further demanded to reconstruct the observed Higgs and top masses in the intervals (120, 130) GeV and (165, 175) GeV, respectively. As a result of these constraints and of the implications of the strong dynamics, the distributions of the allowed points imply strong correlations among the physical observables that will be investigated in the following.
The existence of a non-trivial vacuum is secured by a careful solution of the two tadpole equations which provide values for the VEV and tan β. While v SM is fixed to 246 GeV, tan β could potentially be unconstrained if it were not for the requirements on m h and m t , which select tan β ∼ O(1 − 10) among all its possible values. The surviving range of tan β, mapped against f , after imposing the two aforementioned constraints is illustrated in Fig. 5 (left), from where it is clear that tan β never exceeds ∼ 10.
Before moving to the characterisation of the scalar spectrum we comment on the significance of tan β for 2HDMs, both elementary and composite. In general, tan β is not a physical parameter of the 2HDM since it is not basis-independent [26,27]. This is, for instance, the case of an E2HDM in which the Z 2 transformation properties of the two Higgs doublets are not specified. By imposing specific discrete symmetries, one selects a particular basis (the one in which such symmetries are manifest) where tan β can be uniquely identified, thus promoting it to a physical observable. Notable examples are the type-I and type-II E2HDMs [13]. The composite scenario described in this work does not have a C 2 invariance in the strong sector and thus cannot be related to any of the well know Z 2 realisations of the E2HDM. Nevertheless, the requirement of CP conservation (θ t = 0) in the mixing between the composite states and the elementary R-handed quark automatically selects the C 2 invariant embedding of the latter. This choice eventually picks up a special basis, thus a special tan β among all possible basis-dependent definitions. Interestingly, this is not the case for the E2HDM in which the absence of a Z 2 symmetry prevents the identification of a particular tan β. As a final remark, then, it is worth noticing that one should pay attention in comparing C2HDMs vs E2HDMs for fixed values of tan β as the procedure is not meaningful unless the realisation of the 2HDMs is the same. Indeed, even though the two definitions of tan β may be both physical in the two models, if they belong to different bases, the observables they describe are not the same.
The tuning of the Higgs potential
Another important message learnt from Fig. 5 (left) is that the density of points becomes smaller at large f , as naively expected from fine-tuning arguments. Indeed, when f is far from v SM , severe cancellations among the parameters of the strong sector are necessary to satisfy the tadpole conditions. In general, a single tuning of the order 1/ξ = f 2 /v 2 SM is not sufficient to depart from the natural solution of the tadpole equations, v SM ∼ f , and other cancellations, which depend on the fermionic and gauge embeddings in the global symmetry group, must also be advocated. In order to understand what are the most natural regions of the model, we compute the tuning ∆ associated with the EW scale using the measure [28] where the x i 's span the parameters of the strong sector, i.e., those in Eq. (35). As stated above, ∆ ∼ ξ −1 represents only the minimal and unavoidable tuning necessary to trigger EWSB and bounds from below the distribution of points. Indeed, for a given ξ, the actual tuning may vary over different orders of magnitude. This situation, which is also manifest in SO(5)/SO(4) CHMs with fermions in the fundamental representations, is dubbed double tuning and has been extensively discussed in [29], where the role of the fermion sector has been emphasised. It has been shown that the parametrically leading quadratic contributions y 2 L,R , with y L,R = ∆ L,R /f , in the mass terms cannot be made arbitrarily small without reducing at the same time the quartic couplings. This makes the potential not tunable at the order O(y 2 L,R ). The presence of higher order corrections is therefore crucial and one is obliged to firstly demand the leading y 2 L,R contributions to be of the same order of the subleading y 4 L,R ones before invoking the second cancellation which finally generates the order ξ hierarchy between the EW vacuum and the scale of compositeness.
Here we focus, instead, on the impact of the gauge sector which is most of the time overlooked and neglected with respect to the top quark one due to the smallness of the weak gauge couplings. This is not always the case as one can naively see from Fig. 5 (right), where we have considered the limiting case when g, g → 0, i.e., the unphysical regime where there is no gauge contribution. The distribution of tan β is clearly very different and limited to much more small values with respect to that of Fig. 5 (left), where all the gauge and fermionic contributions are correctly taken into account.
In Fig. 6 (left), we notice that the minimum amount of tuning increases in the region where the coupling g ρ (and, thus, the mass of the spin-1 resonances) becomes large. Indeed, a rather model independent gauge contribution to m 2 1,2 is proportional to (9/32π 2 )g 2 m 2 ρ , which is C 2 symmetric and tends to prevent the breaking of the EW symmetry. As such, a larger cancellation between the fermionic and gauge contributions must be advocated. As the gauge contributions become large, the fermionic ones are forced to increase as well and, being intrinsically C 2 breaking, drive the model into a region of the parameter space which deviates substantially from the inert case. This expectation finds confirmation in Fig. 6 (right), where we show a correlation between large values of g ρ and sizeable values of tan β (we remind the reader that the inert case implies tan β = 0).
Higgs boson masses and couplings
We start here by recalling that the physical Higgs states of the C2HDM are the same as those of the renormalisable E2HDM, namely, the two CP-even scalars h and H (in the C2HDM, h is always the SM-like Higgs with mass around 125 GeV), the pseudoscalar A and the charged Higgs H ± . These are easily identified in the Higgs basis (see Appendix C) in which only one of the two doublets provides a VEV and it is obtained from Eq. (3) after a rotation by an angle β. The correspondence between parameters in the Higgs and general basis is worked out, for instance, in [26]. The mass matrix for the CP-even states is where the prefactor v/v SM arises from the canonical normalisation of the kinetic terms and the parameters M 2 ij and Λ i which are given in Appendix C. The diagonalisation of the mass matrix provides the masses of the physical CP-even Higgses Interestingly, the minimum conditions in the Higgs basis are The dependence of θ from f is depicted in Fig. 8. Further details on the behaviour of the masses and mixing angle are discussed in the following sections. The masses of the CP-odd and charged Higgs states are, instead, given by and, as m H , are naturally of order f . We now present explicitly in analytical form the couplings of all Higgs states of the C2HDM to both fermions and gauge bosons of the the SM as well those among themselves which are relevant for LHC phenomenology. We shall do so in three separate sub-sections.
Couplings to fermions
Assuming flavour alignment in order to guarantee the absence of FCNCs at tree level, the leading couplings of the scalars to the fermions are extracted from Eq. (28) at first order in ξ and can be described by the Yukawa Lagrangian where I f = 1/2(−1/2) for f = u (d, l) and the ξ f coefficients are with As mentioned, the parameter θ denotes the mixing between the two CP-even states while ζ f represents the normalised coupling to the fermion f of the CP-even scalar that does not acquire a VEV in the Higgs basis. Since θ is predicted to be small [12,17], ζ f controls the interactions of the Higgs states H, A, H ± at the zeroth order in ξ. At that order, the structure of the Yukawa Lagrangian is the same as the E2HDM one in which the alignment in the flavour sector has been enforced. The crucial difference is that in the E2HDM ζ f is a free parameter while in the C2HDM is fixed by the strong dynamics and correlated to other physical observables. Some of these correlations will be explored in the following sections.
The mass of fermions are also predicted quantities in CHMs. In particular, exploiting the explicit expressions of the form factors in Eq. (28) (and listed in Appendix B), we read the top mass where m Q,T andm Q,T are the physical masses of the top partners coupled to q L and t R , respectively, and, for simplicity, we dropped the superscript t from the parameters of Eq. (27). When the mixing parameters are such that ∆ ∼ M Ψ the subleading corrections can be numerically relevant at small f . The expression of the bottom mass is totally analogue and better approximated even at smaller f . vertex coupling vertex coupling
Couplings to gauge bosons
The trilinear interaction vertices between the physical Higgses and the SM gauge bosons can easily be extracted from the kinetic Lagrangian of the pNGB matrix in Eq. (19) up to the order ξ. As typical in CHMs, due to the non-linearities of the derivative interactions, the kinetic terms of the NGBs must be rescaled in order to be canonical. This introduces corrections of order ξ in the gauge couplings. The relevant interaction terms and corresponding coefficients have been computed in [23] and are listed in Tab. 1 as functions of the mixing angle θ. In particular, the couplings of the SM-like Higgs to the EW gauge bosons, hV V with V = W, Z, get modified by the usual mixing angle θ, as in every realisation of E2HDMs, but also by corrections of order ξ. A convenient way to parameterise these couplings is to recast them in terms of the so called κ i 'modifiers' of Ref. [30] which are the couplings of the SM-like Higgs boson normalised to the corresponding SM prediction where θ → 0 with f → ∞ corresponds to the alignment limit, i.e., the couplings of h to SM particles become the same as those of the SM-like Higgs at tree level.
Higgs boson self-couplings
The scalar trilinear couplings are extracted from the cubic part of the potential, Here we only show the most relevant ones to the phenomenological studies that will be carried out below, i.e., where the quartic couplings Λ i are defined in the Higgs basis and explicitly given in Appendix C and Λ 345 = Λ 3 + Λ 4 + Λ 5 . Finally, due to θ ∼ ξ (see Eq. (40)), the terms ∝ s n θ for n > 1 in Eq. (47) can safely be omitted at the order O(ξ). We note that other terms at the first order in ξ arise from the non-linearities of the derivative terms of the NGB Lagrangian. For the sake of simplicity, these are not shown here but correctly taken into account in all our numerical computations.
Phenomenology of the C2HDM Higgs bosons
We here list the expression for the scalar masses and mixing angle once the parameters are fixed to reproduce the correct EW VEV. In particular, we have the following prediction for the mass of the Higgs states and the rotation angle θ from the Higgs basis to the mass basis By numerical evaluation we can show that indeed the O(ξ) corrections are negligible in the determination of the mass of the heavy Higgses. In the remainder of this section we are going to discuss in more detail the phenomenological impact of the above formulas trying also to correlate (when possible) the Higgs properties with the parameters of the composite sector.
The mass of the heavy Higgs bosons
Up to corrections of order ξ, the mass of the heavy Higgs bosons is mainly set by the m 2 3 parameter, which is associated to the explicit breaking of C 2 . In order to correlate m H , or equivalently m A and As such, the dependence of m 2 3 on the top-partner masses can be recasted in a simple analytic expression obtained from the integral in Eq. (53) and only through m Q ,m Q , m T ,m T . Such a dependence further simplifies when a given hierarchy is established among the fermion masses. In that case, we expect m 2 3 to be approximated by where m l 1,2 are the masses of the lightest and next-to-lightest top-partners, respectively, among m Q ,m Q , m T ,m T . Notice that between the two lightest states there is always a fermion in the 2 1/6 and one in the 1 2/3 . A comparison with the numerical computations is done in the left panel of Fig. 7. By requiring the next-to-lightest top partners to be much lighter than m l 3 we recover the expected behaviour, while the estimate in Eq. (54) is violated as soon as the condition m l 2 m l 3 is not satisfied, as the full dependence on the four top-partner masses becomes important in the determination of m 2 3 . Notice also that we recover the generic result that the terms proportional to quartic powers of the elementary composite mixings are UV finite and calculable even with just two top-partners.
Finally, in order to make a comparison with previous studies, we plot in Fig. 7 (right) the correlation between the two lightest states in the spectrum of the top-partners, respectively, in the 2 1/6 and (4). In this case, the correlation is usually constrained by the mass of the SM-like Higgs as the aforementioned top-partners dominantly contribute to m h (shown with a red line). In the present scenario, however, the presence of a non-vanishing tan β allows for a larger region of parameter space.
Scalar mixing and mass splittings
We now proceed to show some phenomenological results of relevance to Higgs physics at the LHC, concerning both precision measurements of the SM-like Higgs boson and masses, coupling plus decay properties of its C2HDM companions. However, in order to do so, we first ought to extract the viable regions of C2HDM parameter space following the latest experimental constraints.
The points generated from the scan are tested against void experimental searches for extra Higgs bosons, through HiggsBounds [31], and signal strength measurements of the discovered Higgs state, via HiggsSignals [32]. Acceptable regions of the C2HDM parameter space are determined according to the exclusion limits computed at 95% Confidence Level (CL) and then further sifted if the corresponding χ 2 lied within 2σ from the best fit point. The aforementioned tools have been fed with the normalised (with respect to the SM values) Higgs couplings to SM fermions and bosons and with the neutral and charged Higgs Branching Ratios (BRs) without SM equivalents. The Higgs production and decay rates are computed including O(ξ) corrections with the Feynman rules listed in Sec. 5. For the Higgs couplings and masses, as well as for the top quark mass, we used the numerical values predicted, for each data point, by the strong dynamics. For the sake of definiteness, we assumed that the C 2 symmetry is broken with the same strength in all the three sectors of the third generation, namely, the top, the bottom and the τ lepton. This simplified assumption impliesζ b =ζ τ =ζ t or, equivalently, ζ b = ζ τ = ζ t . Even though this particular choice may have an impact on the couplings of the heavy scalars to SM fermions, the predictions for the parameters of the Higgs potential remain unaffected since they are sensitive only to the top sector and thus to ζ t . Besides the extra scalars discussed above, the spectrum of the model is characterised by vector and fermion resonances. Their impact has been taken into account in the computation of the gauge and fermionic form factors that define the scalar potential. Direct searches in di-boson final states, see for instance [33], effectively constrain the masses and couplings of the vector resonances of composite Higgs models. Our scenario can be mapped, for what concerns the heavy spin-1 states, into the model of the SU (2) vector triplet considered in [33]. We will take into account the corresponding exclusion bound at 2σ level which was obtained by assuming a branching ratio into SM gauge bosons of ∼ 50% and the narrow width approximation. As such, these bounds are more than conservative since, as shown in [34,35], the decay modes into SM gauge bosons can be suppressed and the total width substantially enlarged as soon as the heavy fermion decay channels open. In the sector of the heavy fermion resonances, we will require for the lightest state, which usually corresponds to the exotic resonance X 5/3 , M X 5/ 3 1 TeV in order to comply with the exclusion bound at 2σ level extracted in [36] under the assumption BR(X 5/3 → W t) = 1. Nevertheless, we expect that this constraint will be relaxed in our model compared to the minimal scenario examined in [36] as the presence of an extra Higgs doublet also allows for the decay mode X 5/3 → H + t. Hereafter, all the results discussed in the text and figures are taken to satisfy the constraints from direct and indirect searches described above.
The first consequence of these bounds is on the Higgs couplings. In particular we can constrain the mixing angle θ as depicted in the left panel of Fig. 8 in which we show its dependence on the compositeness scale, in agreement with the expectation that θ ∼ ξ for large f , see Eq. (40). The green points are all those satisfying the bounds discussed above while the gray ones represent those failing these. For smaller f , the mixing angle can vary in principle over a wide range of values but is at present bounded by Higgs coupling measurements to be | sin θ| 0.15. (We also notice that, in the C2HDM, sin θ is predicted to be predominantly of positive sign.) The shape makes clear that the more aligned with the SM-like Higgs predictions the LHC measurements are, the larger f ought to be. Indeed, for f 1 TeV, all the sin θ predicted values do pass the indirect and direct constraints. Conversely, if significant deviations from sin θ = 0 are eventually established, this implies that f might well be at the sub-TeV scale, in turn hinting at the existence of other C2HDM states in the LHC regime. The plot on the right of Fig. 8 shows, instead, the Higgs coupling modifier k V introduced in Eq. (46). With respect to the E2HDM, κ V in the C2HDM approaches the alignment limit more slowly, as evident from the negative O(ξ) corrections, as seen in Eq. (46) and exemplified by the upper edge of the distribution presented. However, sin θ also feeds into this distribution, so that its spread seen on the left-hand side of Fig. 8 is responsible for the departures from the (1 − ξ/2) behaviour on the right-hand side. We finally note that values of κ V 0.9 are currently compatible with LHC data at 1σ level [37], hence allowing for several C2HDM solutions at small f . Finally, concerning the ability to distinguish between the C2HDM hypothesis and the E2HDM one, as stressed in [23], once equipped with a measurement of κ V , one can look for differences in the correlation of possible deviations in κ E and κ D , where E and D represent a charged lepton (e.g., a τ ) and a down-type quark (e.g., a b), respectively.
The size of the mass of the CP-odd scalar is shown in Fig. 9(a) for three specific values of tan β and, as expected, grows linearly in f . Indeed, as shown in Eq. (50), the mass of the pseudoscalar, as well as that of H ± and H, is controlled by m 3 which is not constrained to the EW scale by the minimisation conditions of the potential. From the same equation it is also possible to extract the dependence of the mass on tan β. In particular, as m 2 3 grows linearly in tan β, one finds m 2 A ∝ f 2 (1 + tan 2 β). The splitting between the heavy CP-even state and the CP-odd scalar (or, equivalently, the charged Higgs) is shown in Fig. 9(b). The mass difference m H − m A spans from −20 GeV to 60 GeV while a quite definite prediction exists for the splitting between A and H ± , indeed of high degeneracy, since is mainly controlled by the gauge contribution and scales like g 2 /(16π 2 )g 2 ρ . Hence, the ability to establish A → Z * H or H → Z * A signals, respectively, at the LHC will be a strong hint towards a C2HDM dynamics for EWSB, especially if accompanied by the absence of A → W ± * H ∓ and H ± → W ± * A decays. Clearly, also H → W ± * H ∓ or H ± → W ± * H decays would simultaneously be possible in the C2HDM.
However, similar decay patterns may also emerge in the E2HDM 4 . Notwithstanding this, though, an intriguing situation [23] could occur when, e.g., in the presence of an established deviation (of, say, a few percents) from the SM prediction for the hV V (V = W ± , Z) coupling, the E2HDM would require the mixing between the h and H states to be non-zero whereas in the C2HDM compliance with such a measurement could be achieved also for the zero mixing case. Hence, in this situation, the H → W + W − and ZZ decays would be forbidden in the composite case, while still being allowed in the elementary one. (Similarly, Higgs-strahlung and vector-boson-fusion would be nullified in the C2DHM scenario, unlike in the E2HDM, while potentially large differences would also appear in the case of gluon-gluon fusion and associated production with bb pairs.) Clearly, also intermediate situations can be realised. Therefore, a close scrutiny of the aforementioned signatures of a heavy CP-even Higgs boson, H, would be a key to assess the viability of either model. Regarding the CP-odd Higgs state, A, in the case of non-zero(zero) mixing in the E2HDM(C2HDM), again, it is the absence of a decay, i.e., A → Z * h, in the C2HDM that would distinguish it from the E2HDM. In the case of the H ± state, a similar role is played by the H ± → W ± * h decay. Obviously, for both these states too, intermediate situations are again possible 5 , so that one is eventually forced to also investigate the fermionic decays of heavy Higgs states, chiefly, those into top quarks. (We will dwell further on all this in an upcoming section.)
Comments on the exact C 2 symmetry scenario
An interesting limit of our model isζ t = 0 (Y 1 = 0) which corresponds to a restored C 2 symmetry. This scenario realises a composite version of the inert 2HDM. The presence of a C 2 symmetry is consistent with the fact that only one Higgs doublet develops a VEV. By performing a numerical computation of the Higgs potential in the C 2 symmetric case we verified that m 2 2 gives the mass of the physical components of the second Higgs doublet. We also checked the absence of solutions providing the spontaneous breaking of C 2 . The predicted value of the mass of the heavy CP-even Higgs is shown in Fig. 10 where we include all the points generated by the scan without implementing direct and indirect experimental constraints.
In the case where C 2 is also preserved by lighter quarks and leptons, the neutral component of the second Higgs doublet can be a Dark Matter (DM) candidate. For this to happen and also to avoid strong constraints, at least one neutral component should be lighter than H ± (this is always the case in the parameter space explored). The possibility to have DM as the neutral component of an inert Higgs doublet has been thoroughly discussed in the literature, see for instance [38]. In this context we notice that reproducing the DM relic density Ω DM requires a specific value of the couplings λ hHH,hAA for any mass point. The same couplings are also important for direct detection which occurs via tree level Higgs-exchange and loops of W ± 's. The tree level contribution is a direct test of the quartic coupling λ 345 , which then receives an upper bound from direct detection experiments, λ 345 1 for m H,A 200 GeV [38]. As we can see from the size of the coupling λ 345 presented in Fig. 11, the model could allow for a DM candidate, providing the observed value of the relic abundance, while complying with direct detection bounds, for m H,A 800 GeV [38].
Flavour constraints
In this model, even though we assume a flavour symmetric composite sector, we find modifications to rare flavour transitions in the SM from the exchange of pNGBs. As already stressed previously, in CHMs, there are several effects in flavour physics, depending on which composite resonance is integrated out at low energy. In the literature large attention has been given to vector mediators, while in this section we would only consider effects originating from the scalar sector since this is strongly correlated to the phenomenology discussed in this work.
Under the assumption of a flavour symmetric composite sector, the heavy Higgs bosons can only mediate tree level effects in charged current processes and loop effects in the neutral ones. Since, by virtue of the flavour symmetries, the SM-like Higgs and the heavy companions have interactions with the fermions aligned in flavour space, all the flavour constraints are due to a rescaling of the corresponding SM rates. Therefore, the bounds arise because of the relative precision of the SM observables, which roughly ranges from 1 to 10% accuracy. We review in turn the most stringent ones for our construction. • Meson decay M → ν. The charged Higgs H ± can mediate charged current processes aligned in flavour space alongside the W ± mediated decay of pseudoscalar mesons. The expectation value of the scalar operatorūd, between the vacuum and the M (mesonic) state, can be related to that of the divergence of the axial currentūγ µ γ 5 d. By means of this relation, one can then write the relative variation of the BR of the meson to leptons simply as For example, in the case of the B meson decay B → τ ν, the deviation is proportional to ξ d A ξ l A m 2 B /m 2 H + . This shows that tree level charged current processes are sensitive (mainly) to composite parameters that enter the expressions for ξ d,l A and f , since m H + is linear in f . Notice, however, that the ξ d,l A 's are not directly related to the Higgs potential, since they originate from the down sector which contributes negligibly to v and m h , so they can be taken small enough to reduce effects in the charged currents. Furthermore, under the assumption of a flavour symmetric sector, D → τ ν is sensitive to ξ u A and therefore to the parameters of the Higgs potential but still suppressed by the small ratio m 2 D /m 2 H + .
• Transition b → sγ. Among the best measured quantities are B → X s γ transitions. Differently from tree level ∆F = 1 transitions, here, the relevant couplings are the ones of H ± to the top, given that the main contributions arise from the Wilson coefficients C 7,8 in the weak Hamiltonian which are generated by box diagrams with two H ± 's. From loops of heavy (pseudo)scalars we get (57) • Transition B s → µ + µ − . In the SM the leading contributions arise from Z penguin diagrams contributing to the Wilson coefficient C 10 . In this model we predict at the scale of the resonances that We finally depict in Fig. 12 the impact of the flavour bounds discussed above on the parameter space of the C2HDM. The constraints at 2σ level have been extracted from [39,40] and shown by the red and purple shaded regions. The constraint from the measurement of the M → lν meson decay is usually important only for small charged Higgs masses and/or large couplings and, as such, does not affect the range of parameters discussed here even though ζ τ is taken as large as ζ t . The bound from the B → X s γ transition depends on the interplay between the top and the bottom contributions and on the relative size of the ξ t A and ξ b A couplings. In the scenario discussed here, in which C 2 is broken equally in the top and bottom quark sectors (ξ t A = ξ b A ), the corresponding constraint is shown in Fig. 12 by the red shaded excluded region. Needless to say, in different scenarios realising ζ b < ζ t , the bound from the b → sγ transition can be greatly relaxed so that all the points survive the constraint. For example, the same exclusion bound computed for ζ b < 0.1ζ t lies well below the distribution of points (the latter does not sensibly change if the constraints from HiggsBounds and HiggsSignals are enforced for ζ b = 0.1ζ t ). The bound from the measurement of the B s → µ + µ − transition is, instead, more robust as it only depends on ξ t A and not on the particular realisation of C 2 breaking in the bottom quark sector. Moreover, the corresponding excluded region does not overlap with the distribution of points.
Phenomenology of the heavy scalars
The LHC phenomenology of the second Higgs doublet is determined by the couplings to fermions given in Eq. (42) and by the trilinear couplings in Eq. (47) setting the decay to di-Higgs final states. As repeatedly stressed, a key feature of the C2HDM is the strong correlation between the former and the latter, since they are both generated by the breaking of the NGB shift symmetry. This correlation is exemplified in Fig. 13 where we show the parameters ζ t , controlling the couplings of the heavy scalars to the top quark, λ Hhh , which is responsible for the decay channel H → hh, and sin θ, the sine of the mixing angle of the CP-even scalars which sets the size of the H decay into the SM gauge bosons. From our numerical studies, as already mentioned, we have verified that we are always in the region where m H > m h , therefore we focus on the following three scenarios, each addressing a distinctive C2HDM phenomenology for the three heavy physical states of its spectrum, i.e., H, A and H ± , respectively.
• Scalar H. Our numerical analysis allows us to fully compute the relevant observables for the H state and, amongst these, it is instructive to study first the interplay between SM-like decays to di-boson final states and those with third generation fermions. While it is true that H couplings to fermions have several contributions (see Eq. (43)), for f v SM , the leading contribution to the H → tt decay rate is given by ζ t , with which represents the main decay mode above the tt threshold. The other important decay channel is H → hh that, when the tt mode is kinematically closed, can reach a BR of 80%, with the remaining decay space saturated by H → ZZ, W + W − . The corresponding BRs are shown in Fig. 14 (left). In the H → hh case, there are potentially many contributions due to quartic couplings in the scalar potential and higher dimensional operators from the strong sector Lagrangian. The leading contributions are rather simple, though, since, for f larger than v SM , λ Hhh 3/2v SM Λ 6 , where Λ 6 is the scalar quartic coupling previously defined in the Higgs basis. In this regime we find hence, the study of these three decay topologies would enable one direct access to three key parameters of the C2HDM, i.e., ζ t , θ and λ Hhh , as well as to attest their correlations which are shown in Fig. 13.
Furthermore, according to the predicted mass splittings shown in Fig. 9, also non-SM-like, i.e., H → AZ * and H → H + W − * (off-shell) decays are possible but limited to m H 400 GeV, in which regime the mass splitting between H and A or H + is larger. The corresponding BRs are shown in Fig. 14 (right). These two decay modes are controlled by cos 2 θ which delineates the region of the parameter space where the two decay modes can be sizeable, namely, small θ, that also closes the other SM-like decay modes.
The H production cross section is simply dominated by gluon fusion, where H is produced via its coupling to the top. So it is simply obtained from SM gluon fusion Higgs production (calculated at m H ) rescaled by ζ 2 t . We conclude this part by showing some prospects for H phenomenology at the forthcoming runs of the LHC. Here we focus on the H → hh channel and on its bbγγ final state which has been recently addressed in [41] by the CMS collaboration using data from the LHC at the collider energy √ s = 13 TeV and (integrated) luminosity L = 35.9 fb −1 . In particular, we illustrate in Fig. 15 the interplay between direct and indirect searches and the ability of the High-Luminosity LHC (HL-LHC) and High-Energy LHC (HE-LHC) upgrades to investigate the gg → H → hh → bbγγ signal over regions of the C2HDM parameter space projected onto the (m H , ζ t ) plane, even when no deviations are visible in the coupling strength modifiers κ i of the SM-like Higgs state h (red points) at the end of Run 3 at L = 300 fb −1 and at the end of the HL-LHC at L = 3000 fb −1 . The compliance with the coupling modifiers is achieved by asking that |1 − κ i | is less than the The dashed line in Fig. 15 delimits the excluded region from the measurement of the B s → X s γ transition under the assumption ζ b = ζ t which has been employed to compute the Higgs branching ratios discussed in this section. We stress again that other scenarios with ζ b < ζ t would significantly relax the flavour bound from B s → X s γ measurements while will not change, at the same time, the distribution of points allowed by current direct and indirect searches. Fig. 15 shows the interplay between the HE-LHC reach and the impact of present flavour constraints which can be independently exploited to explore same regions of the parameter space. For example, if the gg → H → hh → bbγγ signal will be established at the future upgraded phases of the LHC in a region of parameter space with large and negative ζ t (namely, below the dashed line in Fig. 15), this would clearly point to a scenario with ζ b < ζ t .
• Pseudoscalar A. Since we consider a CP-symmetric composite sector, the phenomenology of the CP-odd scalar is, as far as decays to SM states are concerned, very constrained, since it can basically only decay to SM fermions and Zh. Indeed, according to Fig. 9, the exotic off-shell decays A → HZ * and A → H ± W ∓ * are strongly suppressed by the very tight phase space available. In Fig. 16 ζ b which has been fixed to ζ t in our scan under the reasonable assumption that C 2 is broken with the same strength in all the strong fermionic sectors. In the limit m A m t , the BRs into SM fermions are particularly simple:
). (61)
Searches for A → tt can then be used to constrain the couplings of A to top quarks (as the production cross section is again ∝ ζ 2 t ) and then one can access ζ b and ζ τ . The decay channel A → Zh is, instead, controlled by the square of the coupling (1 − ξ/2) sin θ. Suitable search strategies are, for instance, performed by reconstructing the Z boson from its leptonic decays and the SM-like Higgs from h → bb or h → τ τ .
• Charged H ± . As far as decays into SM objects are concerned, here, the phenomenology is dictated by H + → tb and H + → W + h, while H + → τ + ν τ is found to be negligible with BR(H + → τ + ν τ ) ≈ 4 × 10 −5 ζ 2 τ /ζ 2 t in the large m H + limit. The non-SM-like modes H + → W + * A and H + → W + * H are also suppressed due to the small mass splittings between the heavy scalars, see Fig. 9. The BRs of the two dominant channels are shown in Fig. 16 (right) with H + → tb being the leading one as m H ± > m t . The partial decay width of H + → tb is determined by ζ 2 t (the contribution of ζ b is suppressed by the ratio m b /m t ) while H + → W + h is driven by the square of (1 − ξ/2) sin θ.
As for the H ± production cross section, in the relevant mass range (i.e., m H ± > m t ), this is governed by the bg → tH − + c.c. channel, which is proportional to the same coupling entering H + → tb decays.
In short, if deviations will be established at the LHC in the couplings of the discovered Higgs state to either SM gauge bosons or matter fermions, then, not only a thorough investigation of the 2HDM hypothesis is called for (as one of the simplest non-minimal version of EWSB induced by the Higgs mechanism via doublet states, like the one already discovered) but also a dedicated scrutiny of the production and decay patterns of all potentially accessible heavy Higgs states, including data from the HL-LHC and HE-LHC options of the CERN machine, could enable one to separate the E2HDM from the C2HDM. In this endeavour, key roles will be played by interactions amongst the Higgs bosons themselves (with or without gauge bosons intervening) and with top quarks [23].
Conclusions
In summary, in this paper, we have used compositeness as a possible remedy to the hierarchy problem of the SM, in particular, assuming a pNGB nature of the discovered Higgs state. In this respect, an intriguing setting is the C2HDM, as it builds upon the experimentally established existence of a doublet structure with a SM-like Higgs state h triggering EWSB, and the need for BSM physics. This scenario in fact surpasses the SM by providing one more composite doublet of Higgs states that can be searched for at the LHC, i.e., the familiar H, A and H ± states of a 2HDM, alongside additional composite gauge bosons and fermions. In fact, in order to obtain an acceptable fine-tuning at the EW scale, the compositeness scale f , which drives the masses of these (heavy) non-SM Higgs states, must be in the TeV region. The C2HDM framework advocated here is thus a BSM setting which is both natural and minimal, offering as byproducts Higgs mass and coupling spectra within the LHC reach.
In fact, the entire physical content of the C2HDM, unlike the case of an elementary 2HDM, is actually predicted by a new confining strong dynamics. Our set up is based on the spontaneous symmetry breaking of the global symmetry of a new strong sector, SO(6) → SO(4) × SO (2), with the residual symmetry in turn explicitly broken by the linear mixing between the (elementary) SM and the (composite) strong sector fields via the so-called partial compositeness mechanism.
In this construct, the scalar potential emerges at one-loop level and governs the dynamics of the aforementioned Higgs states, all realised as pNGBs, i.e., it predicts their properties. We have therefore calculated the mass and coupling spectra of the Higgs sector of this C2HDM explicitly and for the first time in literature. In particular, we have truncated the potential to the quartic order in the (pseudo)scalar fields, further assuming that the dynamics discussed here is CP symmetric yet C 2 broken, so that, in order to avoid Higgs-mediated FCNCs, we had to enforce an alignment in the Yukawa couplings. In order to do so, we had to also account for the corresponding gauge and fermionic spectra. While the structure of the former is dictated by the gauge symmetry and the breaking pattern, there is arbitrariness in the choice of the latter. Here, we have placed the fermions in the fundamental representation of SO(6) and we have further required a LR structure, as it guarantees UV finiteness and reduces the number of free parameters in the new strong sector. Ultimately, this implies that the scalar potential, and thus the couplings amongst (pseudo)scalars and fermions, depend on the composite fermion mass spectrum.
We have highlighted here the presence of correlations among the Higgs sector parameters and the strong sector ones, in particular, the dependence of the extra-Higgs mass scale on the extra-fermion masses. In fact, within this framework, we have obtained that the mass m H 2 of the second Higgs doublet is mainly proportional to the m 3 parameter in the scalar potential, which arises from the breaking of C 2 in the composite sector and correlates with the masses of the two lightest top-partners. Then, as m 3 is driven by f , we have highlighted the fact that all heavy Higgs masses decouple for large f from the SM-light Higgs one, which is maintained light as it emerges from the SM VEV. Hence, as f → ∞, the SM is recovered.
Another interesting limit that we discussed is obtained when the above Yukawa alignment is dismissed and one of the two relevant couplings is set to zero (i.e., Y 1 → 0) to prevent tree level FCNCs. This generates a C 2 symmetric case, wherein one VEV is zero (i.e., tan β = 0) and no mixing exists between the first and second doublet (i.e., sin θ = 0). This generates a 2HDM structure with one inert doublet, thus offering a DM candidate (in the form of the lightest between the additional neutral Higgs states, H and A). The proposed scenario thus provides a concrete realisation of models with two doublets (one playing the role of the SM Higgs and the other being active or inert) originating from a strong confining dynamics.
We have then proceded to discuss the phenomenological implications of this C2HDM at the LHC, in relation to both measurements of the discovered SM-like Higgs state and the (potential) discovery of its heavy companions. In this respect, we have assessed that the mass and coupling patterns that emerge from the strong dynamics embedded in the C2HDM are rather prescrictive, so that simultaneous h measurements and detection of heavy Higgs decays into themselves (e.g., H → AZ * and H → H ± W ∓ * or A → HZ * and H ± → HW ± * , depending on the actual mass hierarchy), with absence of any decays involving A and H ± , would be a hallmark signature of the C2HDM. With this in mind, we have finally tested the scope of the standard LHC, HL-LHC and HE-LHC in accessing the C2HDM in both the above self-interacting Higgs channels and others involving (primarily) production and decay modes with top (anti)quarks involved.
As we have produced all these phenomelogical results in presence of up-to-date experimental (notably including limits from both void searches for heavy Higgs bosons and measurements of the 125 GeV discovered Higgs state as well as flavour data) constraints, we are confident to have set the stage for pursuing dedicated analyses aimed at separating the C2HDM from the E2HDM hypothesis, in turn potentially enabling one to distinguish the composite from the fundamental nature of any Higgs boson accessible at the LHC by the end of all its already scheduled and currently discussed future stages.
Outlook
In this work we have studied a scenario based on the coset SO(6)/SO(4) × SO(2) with fermions embedded into the fundamental SO (6) representations. This represents the simplest scenario with maximal subgroup H providing exactly two Higgs doublets as pNGBs, which was the Higgs sector content conceived here as natural next step up from the discovery of a 'doublet' Higgs field. However, despite its simplicity, some of the phenomenological features described by this setup (as, in particular, the scenario with a composite inert Higgs) are also typical of other realisations. More involved patterns can be constructed by allowing for subsequent breakings as in the case of SO(6) → SO(4) in which the two scalar doublets are accompanied by an extra SM singlet state. Focusing on scenarios with only two Higgs doublets, another interesting model is realised by the coset Sp(6)/SU(2) × Sp(4) in which large corrections to the T parameter are automatically avoided by the extended custodial symmetry of the corresponding subgroup. In this setup the left-handed q L and right-handed t R components of the top quark can be embedded in the 1 2/3 and 14 2/3 , respectively. Since two inequivalent embeddings are present for q L , even though only one invariant can be build among the two representations and the Higgs doublets, the absence of dangerous tree-level FCNCs in the scalar sector must be ensured by enforcing the alignment in flavour space as in our case. A scenario that shares even more similarities with the one addressed in this paper is instead described by the same coset SO(6)/SO(4) × SO(2) but with fermions in the 1 2/3 for the t R and in the 20 2/3 , containing two different embeddings, for the q L [12]. Either requiring CP or C 2 unavoidably selects one of the two q L embeddings. This also avoids the presence of Higgs-mediated FCNCs without resorting to the flavour alignment. If C 2 is preserved, the composite inert 2HDM, described in section 6.2.1, is obtained. In contrast, if only CP is required, C 2 still arises at leading order as an accidental symmetry in the scalar potential with an almost-inert Higgs doublet. As such its phenomenology is expected to be vey similar to the one discussed in section 6.2.1. The construction discussed in this paper has the advantage to encompass several scenarios with very different phenomenological features like, for instance, the C 2 symmetric case that provides an inert Higgs doublet which, as discussed above, is also common to other realisations of CHMs, but also much more peculiar scenarios like the CP -invariant and C 2 -broken case that has been extensively studied in this work.
A The C2HDM with the CCWZ formalism
Let us review some of the main topics discussed in the present work by using the CCWZ formalism and focusing only on the fermion sector. This methodology is based on an effective Lagrangian approach and, as such, does not require the specification of an UV completion and only relies on the features of the symmetry breaking pattern. Therefore, it is possible to draw general conclusions without specifying the details of the model in the strong sector, in contrast to what we have done in the sections above, where we provided an explicit realisation of the C2HDM. Needless to say, the two approaches are completely equivalent at low energy.
In order to keep the discussion as general as possible, we introduce two families of resonances, ψ 4 and ψ 2 , the first one transforming in the fundamental of SO(4) and the other one in the fundamental of SO (2). The Lagrangian for the elementary and composite fermions is In order to simplify the discussion, the down-quark d R has not been explicitly written in the previous equation but it will be reintroduced back when needed. The index i runs over the three SM families while k runs over the fermionic resonances for which the multiplicity can be, in principle, different for each of the two families. The notation q 6 L and u 6 R denotes, as usual, the embedding of the SM fields in the fundamental of SO(6), while the subscripts on the pNGB matrix U represent its projections onto the different SO(4) × SO(2) representations. In particular, the subscripts 2 and2 denote how the two invariants 2 · 2 and 2 ∧ 2 are built from the fundamental representation of SO(2), specifically, by exploiting the contraction with δ αβ and αβ , respectively. Restricting to the top-quark sector and considering only two families of resonances, Eq. (62) can be easily mapped onto the Lagrangian given in Eq. (27). After integrating out the heavy resonances, the momentum space Lagrangian for the quark fields reads as where the gauge interactions arising from the fermionic covariant derivates have been neglected as they play no role in the present discussion. The form factors are where Υ L and Υ i R are the spurions with their VEV defined as in Eq. (26). Notice that, differently from its L-handed counterpart, the Υ i R spurion carries a flavour index because each of the three Rhanded fields u i R can be embedded into a 6 of SO (6) in two different inequivalent ways. This freedom is parameterised by an angle θ i . The contractions of the pNGB matrices in the form factors above encode all the dependence on the Higgs fields and are explicitly defined as where i, j and α, β run over the fundamental representations of SO(4) and SO(2), respectively. In order to make a closer contact with the results presented in the previous sections, where the model in the strong sector has been explicitly specified, we notice that the contractions of the pNGB matrix are related to the Σ matrix by Considering only the top-quark and two families of heavy resonances, the effective Lagrangian after the integration of the heavy resonances can be schematically written in the same form as Eq. (28), namely, with k running only on two families of heavy resonances. After integrating out the heavy degrees of freedom, as naively expected, the CCWZ approach leads to an effective Lagrangian with the same structure of the one derived from an explicit model. The only difference appears in the parameterisation of the form factors in terms of masses and couplings. For instance, we immediately notice that the dependence on the proto-Yukawa couplings is the same, thatΠ 0 andΠ 2 are real whileΠ 1 is imaginary and thatΠ 1 = 0 in a CP invariant scenario. The calculation of the effective potential in terms of the form factors remains completely unchanged, so are Eqs. (31) and (32). As we anticipated above, the form factors obtained in the 2-site model can be easily mapped onto the ones in Eq. (68) once the Lagrangian of Eq. (27) is recasted into the basis in which, for ∆ L,R → 0, the resonance fields Ψ t are mass eigenstates. Notice also that, despite the equivalence of the form factors obtained after the integration of the heavy resonances, the effective Lagrangian in Eq. (62), which is used as the starting point of the CCWZ construction, does not capture all the information encoded in Eq. (27) such as, for instance, the interaction among the resonances and the NGBs.
An issue with flavour changing neutral currents
The most general effective Lagrangian in the fermion sector can be easily constructed using the CCWZ formalism and reads as where ψ L,R denote SM quark fields embedded into incomplete G multiplets while the index A spans over all the possible invariants in the subgroup H with P A being the corresponding projector operator.
As described above, we embed the L-and R-handed components of the elementary SM quarks into the fundamental representation of SO(6) using the spurion fields. At this level, all the information on the interactions with the composite resonances, which have been integrated out, is encoded in the coefficients a A ij . With the fermions in the 6 of SO(6), where 6 = 4 ⊕ 2 under SO(4) × SO(2), three different invariants of H can be constructed with δ ij , δ αβ and αβ , respectively, where latin (greek) indices belong to SO(4) (SO (2)). The sum of the first two invariants is trivial such that only two of them are independent. These can be chosen to be, for instance, δ αβ and αβ . The possibility to build several invariants naturally introduces dangerous FCNCs in the Higgs sector, unless one imposes particular conditions on the coefficients a A ij or discrete symmetries to single out only one invariant. One of the two invariants can be removed by imposing a C 2 symmetry in the strong sector. But, even in that case, since the R-handed quarks can be embedded in the 6 of SO(6) into two independent ways (as already mentioned, this freedom is described by the θ i angle), an extra symmetry must be advocated in order to select one of the two embeddings. For instance, the CP symmetry uniquely picks up the θ i = 0 direction. Therefore, Higgs mediated FCNCs are avoided if one considers a C 2 and CP invariant scenario. If the strong sector does not enjoy such symmetries that prevent the appearance of multiple invariants, FCNCs can only be avoided by restricting the structure of the a A ij matrices as in the flavour alignment limit. The absence of leading contributions to FCNCs is thus ensured if the three matrices (in flavour space) appearing in the form factor Π ij LR , namely the ones proportional to the three independent invariants in Eq. (65), are proportional to each other in the small momentum regime. This condition can be satisfied by requiring, e.g., the following: (a) a 1 y ik L4 = a 2 y ik L2 = a 3 y ik L2 ≡ y ik L and b 1 y ik R4 = b 2 y ik R2 = b 3 y ik R2 ≡ y ik R , so that y ik L y jk * R can be factorised, (b) the alignment of the R-handed spurion fields, θ i ≡ θ and (c) the proportionality of the resonance masses c 1 m 4,k = c 2 m 2,k ≡ m k for all k = 1 . . . N ψ , where the number of resonances N ψ has been chosen to be the same for the two families.
Under the flavour alignment assumption, the Yukawa Lagrangian can be finally recast in the following form: where we have reintroduced back the d R quark field. The g u,d (H 1 , H 2 ) are functions of the Higgs fields and the SM Yukawa couplings Y ij u,d are defined by The Yukawa structure in Eq. (70) clearly ensures the absence of Higgs mediated FCNCs.
B Form factors
The form factors characterising the gauge part of the effective Lagrangian in Eq. (22) arẽ and, by choosing for simplicity g ρ X = g ρ , which implies m ρ X = m ρ as shown in Eq. (21), we get the normalised form factors withΠ where r W,B = g 2 W,B /g 2 ρ . The couplings g W,B are implicitly defined by where g and g are the SM SU(2) L and U(1) Y coupling constants. The form factors appearing in Eq. (31) are normalised asΠ W,B = f 2 /q 2 Π W,B . Finally, the form factors extracted after the integration of the heavy spin-1/2 resonances coupled to the top quark are explicitly given bỹ In the previous equations we used the shorthand notation and we required to enforce the LR symmetry.
Notice that, when m 2 | 22,145.2 | 2018-10-15T00:00:00.000 | [
"Physics"
] |
Fluorescence-detected two-quantum and one-quantum-two-quantum 2D electronic spectroscopy of Rhodamine 700
. We demonstrate the simultaneous acquisition of three fourth-order nonlinear signal contributions using a shot-to-shot-modulating pulse shaper and fluorescence detection. Beside the 1Q photon echo, two different species of two-quantum contributions can be isolated without any background via phase cycling.
Introduction
Coherent two-dimensional (2D) spectroscopy is a versatile technique that reveals detailed information about molecular systems via probing their dynamics with ultrashort laser-pulse sequences. Besides well-established one-quantum (1Q) 2D spectroscopy enabling, e.g., the study of exciton transport in light-harvesting systems, two-quantum (2Q) 2D spectroscopy correlates doubly-excited vibrational or electronic states with their constituent singlyexcited (1Q) states [1,2]. It was shown that the obtained coherently-detected signals are suitable for quantifying electron correlation energies [3], which were believed to be unmeasurable for a long time. The core idea is to measure the energy shift of a 2Q state with respect to twice the energy of the 1Q state, reflecting the magnitude of interaction between the two involved electrons [2,3]. Significant challenges are that 2Q signals are typically weak and that there is a strong background from scattering and nonresonant solvent contributions. This may have precluded a more widespread application of the method.
Here we solve these problems by introducing a new approach for 2Q 2D spectroscopy that measures fluorescence instead of coherent four-wave mixing signals. While incoherent detection is already used for 1Q 2D spectroscopy [4][5][6], studies for 2Q signals are scarce. We obtain a background-and scattering-free signal employing only one broadband excitation beam. Utilizing an acousto-optic pulse shaper on a shot-to-shot basis [6,7] and phase cycling [4,8] we extract various signal contributions from the same raw data set. In addition, we experimentally introduce 1Q-2Q spectroscopy [8] with less congested spectra than in the 2Q case.
Coherent two-quantum signals from an incoherent observable
In our approach, we create a final fourth-order excited-state population via phase-coherent pulse sequences and utilize the unique phase dependency of different nonlinear signal contributions in order to extract them via phase cycling [4,8]. It was theoretically shown [8] that by cycling the phases of a three-pulse sequence 16 times, one should be able to recover the photon echo, the 2Q signal and also a so-called 1Q-2Q signal. The key difference between the latter two population-based techniques lies in the different time ordering of the twice-interacting pulse. This can be illustrated by double-sided Feynman diagrams [ Fig. 1(a)], where arrow directions represent interaction phase + (right) or (left). For a three-level system with ground (g), singly excited (e), and doubly excited state (f), three pathways Q1, Q2, and Q3, which are weighted with quantum yields e and f [ Fig. 1(a)] are present. Two coherence times, τ and t, are scanned and then a 2D Fourier transform is performed. We simulate the 2Q [ Fig. 1(b)] and 1Q-2Q [ Fig. 1(c)] spectra for a system with 100 meV correlation energy (Δ) between the first (|g|e, ħωeg = 2.0 eV) and the second (|e|f, ħωfe = 2.1 eV) 1Q transitions using the Lindblad master equation. It is evident that 1Q-2Q spectroscopy scans the |eg| 1Q coherence only [red interval in Fig. 1(a)], while 2Q spectroscopy probes the |eg| and |fe| 1Q coherences [blue t interval in Fig. 1(a)]. Thus, the latter signal may suffer from interference between different 1Q coherences [distortions in Fig. 1(b) as compared to Fig. 1(c)]. All obtained data have in common that, because fluorescence is the observable, nonresonant solvent responses or scattering are absent. This is especially important for 2Q signals because the resulting artefacts would be strong compared to the naturally weak 2Q signals.
Experimental setup and results
We use a commercial Ti:Sa chirped-pulse amplifier (800 nm, 1 kHz, 35 fs) to generate a long-term stable supercontinuum output via focusing a spatially stabilized beam into an argon-filled hollow-core fiber [ Fig. 2(a)]. Second-and third-order dispersion control is granted by a dual grism compressor and the pulse shaper. Phase-coherent and time-delayed pulse sequences are streamed shot-to-shot by an acousto-optic programmable dispersive filter (AOPDF, Dazzler) [6,7]. A 0.1 mM solution of Rhodamine 700 in ethanol is pumped through a flow cuvette, into which we focus the beam. The resulting fluorescence signal is collected via microscope objectives and detected by an avalanche photodiode in a linear range. For data analysis, we remove potential artefacts which may arise from imperfections of the pulse shaper by subtracting normalized time-domain data acquired with low pulse energy (0.7 nJ) from those with high pulse energy (19 nJ).
The extracted real-valued nonlinear contributions from one measurement are shown in Fig. 2 and contain the 1Q photon-echo (b), the 2Q (c), and the 1Q-2Q signals (d). Via phase cycling, we directly obtain the real-valued signals so that no additional phasing is required. Both 2Q-associated spectra [(c) and (d)] look very similar. However, by analyzing the absolute-valued projections onto their respective 1Q axes [ Fig. 2(e)], it is evident that the |e|f transition manifests as a blueshift of the 2Q lineshape. This indicates a low value for f, so that path Q2 contributes significantly [see Fig. 1(a)]. By simulating the 2D spectra with varying Δ, best agreement with experiment is obtained for Δ = 30 meV.
Conclusion
We introduced a single-beam setup for fluorescence-detected two-quantum (2Q) 2D electronic spectroscopy with broadband excitation. Our approach is free of scattering artefacts or nonresonant background and further gives access to 1Q-2Q 2D spectroscopy which is less congested than the 2Q variant. For Rhodamine 700, a comparison between these two spectra revealed a blue-shifted second 1Q transition into a 2Q state with a very low quantum yield, indicating nonradiative relaxation channels into the ground state. | 1,313.2 | 2019-01-01T00:00:00.000 | [
"Physics"
] |
Group linear non-Gaussian component analysis with applications to neuroimaging
Independent component analysis (ICA) is an unsupervised learning method popular in functional magnetic resonance imaging (fMRI). Group ICA has been used to search for biomarkers in neurological disorders including autism spectrum disorder and dementia. However, current methods use a principal component analysis (PCA) step that may remove low-variance features. Linear non-Gaussian component analysis (LNGCA) enables simultaneous dimension reduction and feature estimation including low-variance features in single-subject fMRI. A group LNGCA model is proposed to extract group components shared by more than one subject. Unlike group ICA methods, this novel approach also estimates individual (subject-specific) components orthogonal to the group components. To determine the total number of components in each subject, a parametric resampling test is proposed that samples spatially correlated Gaussian noise to match the spatial dependence observed in data. In simulations, estimated group components achieve higher accuracy compared to group ICA. The method is applied to a resting-state fMRI study on autism spectrum disorder in 342 children (252 typically developing, 90 with autism), where the group signals include resting-state networks. The discovered group components appear to exhibit different levels of temporal engagement in autism versus typically developing children, as revealed using group LNGCA. This novel approach to matrix decomposition is a promising direction for feature detection in neuroimaging.
S.2.1 Robustness to misspecified number of group components
In this section, we provide the results of group LNGCA and group ICA on group components in simulations when the number of group components is misspecified: q G = 4 and q G = 2.
First when q G = 4, both methods perform similar to the case when q G = 3: the three components matched to true group components have similar correlation as when q G = 3, showed in Although the test underestimated the dimensions in many simulations, this was due to possibly missing individual components, while it always retained the group components.
S.2.2 Robustness to the number of time points
To examine the robustness of our method to varying number of time points, we also conduct our simulations with T = 30 and T = 70. We keep all settings fixed except for the number of Gaussian
S.2.3 Decomposition of subject deviations from group signals
We conduct one repetition of our high SVAR simulation setting according to the previous SVAR setting except we add subject-specific deviations in two subjects: one has extra active pixels on the top of the "1" component, and one has extra active pixels on the bottom of the "1" component, as in Fig. S.9. The estimated group signal from group ICA is slightly worse than that from group LNGCA. We plot the individual components from the two subjects that have the highest correlation with the true subject-deviation component. We see group LNGCA successfully captures the individual deviations with relatively high correlation (0.6).
We describe an approach to identify the individual components that represent subject-specific deviations from the group components. We refer to the components from the initial subject-level LNGCA (Step 1 of Algorithm 1) as the separate-subject components. We refer to the individual S.3 Details on resting-state fMRI data example S.3.1 Additional information on the data and preprocessing All children completed a mock scan to acclimate to the scanning environment. Participants were instructed to relax, fixate on a cross-hair, and remain as still as possible. Functional data were preprocessed using SPM12 and custom MATLAB code (https://github.com/KKI-CNIR/CNIR-fmri_ preproc_toolbox). Rs-fMRI scans were slice-time adjusted using the slice acquired in the middle of the TR as a reference, and rigid body realignment parameters were estimated to adjust for motion. The volume collected in the middle of the scan was non-linearly registered to Montreal Neurological Institute (MNI) space using the MNI EPI template. The estimated rigid body and nonlinear spatial transformations were applied to the functional data in one step, producing 2-mm isotropic voxels in MNI space. Voxel time series were linearly detrended. Data were excluded for between-volume translational movements > 3-mm or rotational movements > 3 degrees.
Group ICA and its PCA steps were applied using GIFT. The second stage PCA was implemented using multi-power iterations (Rachakonda et al., 2016).
S.3.2 Dimension estimation
We applied the NG subspace dimension test in Section 2.3 to six participants. We also implemented a sequential test with FOBIasymp and the estimated dimensions are 126, 126, 126, 148, 151 and 153 for participants #1,. . . ,#6 correspondingly. Such large dimension will not help reduce much computation in practice. It also implies FOBIasymp tends to overestimate the number of non-Gaussian components, as discussed in our simulation.
S.3.3 Subject-specific components
Example subject-specific components are depicted in Figure S.10. Figure S.10: Example subject-specific (individual) components from four different participants. These components include artifacts. Activation near the brain edge, as in the first and third rows, is often indicative of a motion artifact. | 1,128.4 | 2021-01-13T00:00:00.000 | [
"Computer Science"
] |
Validation of low-cost pavement monitoring inertial sensor for urban road network
Road networks are monitored to evaluate their decay level and the performances regarding ride comfort, vehicle rolling noise, fuel consumption, etc. In this study, an Inertial Measurement Unit is proposed by using a low-cost three-axis Micro-Electro-Mechanical Systems accelerometer and a GPS instrument, which are connected to a Raspberry Pi Zero W board and embedded inside a vehicle to monitor indirectly the road condition. To assess the level of pavement decay, the comfort index awz defined by the ISO 2631 standard was considered. Considering 21 km of roads, with different levels of pavement decay, validation measures made using the proposed IMU, another preassembled IMU, and a Road Surface Profiler were performed. Therefore, comparisons between awz determined with accelerations measured on the two different IMU are made; in addition, also correlations between awz, International Roughness Index (IRI), and Ride Number (RN) were performed. The results were shown very good correlations between the awz calculated with the proposed IMU and ones in the other IMU. In addition, the correlations between awz and IRI and RN were showed promising results, considering the use and the costs of the proposed IMU as a reliable method to assess the pavements decay in road networks where the use of traditional systems is difficult and/or not cheap.
Introduction
The management of infrastructural assets is a complex process that integrates many multidisciplinary strategies for the maintenance of public infrastructures [1]. Generally, the process interests on the later phases of the infrastructure's life cycle, but it would be better to integrate this process in the design phase [2].
This process aims to organize and implement strategies to maintain infrastructures enhancing their performance and extend their life span [3]. In fact, the infrastructures and in particular the transport ones are fundamental components for maintaining the quality of life in society and the efficiency of the Countries' economy.
Road pavement is a very important transport infrastructure asset that require an accurate assessment of the distresses for understanding how to fix them.
Pavement Management Systems (PMS) were employed by road agencies in the North America since the 1970s to manage their networks; these systems are evolved over the years to become reliable tools for the effective management of pavements for all road networks; since then their use has spread to all countries of the world [4].
The pavement roughness is measured using high-performance equipment (contact or non-contact profilers), which detect road profiles along the pavement [15], and the acquired data are evaluated in terms of globally recognized indexes worldwide [16].
The most popular index used around the world to evaluate pavement roughness starting from the measured profile is the International Roughness Index (IRI) [17]. Many threshold values are available to depend on the length of the profile, the type of pavement (asphalt concrete or Portland Cement Concrete), and other pavement characteristics [18][19][20][21]; other interesting researches proposed different threshold values considering the design speed of the road [22,23], so to accept higher IRI thresholds for the roads where the design speed is lower. The costs associated with sophisticated pavement evaluation equipment such as Mobile Measurement System (MMS) can be significant [24,25] relatively to the bargain budget of road agencies. For these reasons, the Road Surface Profilers (RSPs) are currently used to evaluate pavement roughness in nonurban road networks (roads outside administrative borders of cities with speed limits more than 70-80 km/h).
There are also operative problems that limits the use of RSPs in the urban road networks: these devices provide reliable results only at certain measurement speeds, generally higher than 30-35 km/h, which are not always possible in urban areas for various reasons (the presence of speed limits, the low planimetric radius, the many intersections, etc.). In addition, non-contact profilers need a launch segment free of obstacle that allows them to reach the predetermined survey speed, which further limits their application in an urban context. It should also be considered that the medium level of distress of urban pavements often does not allow the correct operation of these vehicles [14], which, as mentioned, are designed for nonurban roads.
There are alternative systems than can evaluate pavement roughness in indirect way considering indexes; these indexes can be determined starting from pavement profiles (i.e. Ride Number, RN) or considering methods involve the use of an accelerometer mounted in a moving vehicle. These last methods are potentially useful tools for pavement condition assessment in a cost-efficient way, but a preliminary calibration could be required to take into account the dynamic characteristics of the test vehicle and its speed [14].
Whatever system used to evaluate pavement roughness (using direct or indirect method) should be integrated at least with a high-precision GPS receiver to allow the correct localization and positioning of measurements on the road [26][27][28][29][30][31][32].
It is noted that the essential measurement systems necessary for the ride evaluation (three-axis accelerometers and GPS), are already available in the modern smartphones where they are suitably integrated and synchronized [33][34][35][36][37][38][39].
For this reason, many of these solutions were recently proposed over the world with different approaches: starting from accelerations, some apps try to estimate IRI along the surveyed road (divided into constant segments, 20-50-100 m) so to provide a typical evaluation (generally, using IRI) of pavement quality [40]. Other apps propose new indexes to evaluate pavement conditions [41] and other using its self-approach for classification [42].
To the shortcomings highlighted concerning the relevant devices, there is also the further difficulty that consists in the inadequateness of limit and thresholds for the various roughness indexes currently in use and previously described, which have been defined to correspond to particular needs and peculiarities of urban roads; first of all the low speeds, generally below 50 km/h [22,43,44]. Where an attempt has been made to overpass this lack, such as some limits of the IRI index defined according to the type of road or pavement, there is still some doubt to apply them in urban road network [45,46].
In consideration of all these problems, the choice of the system monitoring and the assessment method for urban road pavements could be overcome by using an index that depends on the vertical accelerations measured inside a vehicle in motion considering its characteristics [47][48][49][50].
The solutions based on the survey of vertical accelerations inside the passenger compartment with a low-cost seem to be an interesting alternative to solve these difficult in the monitoring and assessment of urban pavement.
These sensors could arouse interest for those road network managers who do not yet have any continuous monitoring system for their pavements. In fact, they entrust the choice of maintenance strategies and the related interventions to procedures independent of monitoring (time-based maintenance) or in consequence of the occurrence of failures (run-to-failure maintenance) with serious losses of direct and indirect costs for the community.
Instead, it would be desirable to carry out maintenance referring to performancebased systems that allow identifying the appropriate time to perform maintenance interventions with respect to the conditions of the entire network and the available budget.
Considering the accelerations measured at a certain speed onboard a vehicle the whole vibration index called awz according to ISO 2631 [51] can be adopted associating with a level of pavement decay. In fact, the thresholds defined by the ISO 2631 standard in terms of comfort levels can be related, considering vehicle speed, to the different levels of decay pavement, obtaining a substantial correspondence with respect to the IRI and other analogous indexes [43].
In this study, an Inertial Measurement Unit was developed by using a low-cost threeaxis Micro-Electro-Mechanical Systems (MEMS) accelerometer and a GPS instrument, which are connected to a Raspberry Pi Zero W board [52] and embedded inside a vehicle for monitoring indirectly the road condition. To assess the level of pavement decay, the comfort index awz defined by the ISO2631 standard was considered.
Considering 21 km of roads, with different levels of pavement decay, validation measures made using both the proposed sensor, a pre-assembled IMU (Landmark 10 GPSA-150-10-200), and a RSP were performed. Therefore, comparisons between awzs determined with accelerations measured on the proposed sensor and ones of the other more expensive IMU are made; in addition, also correlations between awz, IRI and RN determined using respectively the proposed sensor and the RSP were performed. The results were shown very good correlations between the awz calculated with the sensor proposed and ones in the other IMU. In addition, the correlations between awz and IRI and RN were shown promising results, considering the use of the proposed sensor as a reliable method to assess the pavements decay in road networks where the use of traditional systems is complicated and/or not cheap.
Materials and Methods
In this section, the methods and procedures used for evaluating, indirectly and directly, pavement condition are described.
Whole-Body Vibration-ISO 2631
Starting from the vertical accelerations in the time domain, measured onboard the test vehicle, the root mean square (RMS) accelerations through the evaluation of the PSD can be determined for all the frequency range of interest for the human response to vibrations (between 0.5-80 Hz), and analyzed by a spectrum of 23 one-third octaves bands. This procedure is specified by the technical standards currently in use [53,54] and it is similar to other analysis to transform the signals measured in the time domain in spectrum in the frequency domain.
Once the RMS acceleration one-third octave spectrum is known ( = ( ,1 , ,1 , … , , 23 ), corresponding to the 23 frequencies proposed by ISO2631 (0 (1) where , are the frequency weightings in one-third octaves bands for the sensor position, provided by the standards ISO2631 [51] and , is the vertical RMS acceleration for the i-th one-third octave band. Then, the calculated values can be compared with the threshold values proposed by ISO 2631 for public transport (Table 1), in order to identify the comfort level perceived by users in all roads sections, also considering several speeds of transit. Considering the real characteristics of the acceleration sensor used during the measure and analysis (the analysis time, and output data rate frequency, = 1 Δ ⁄ in Hz, where Δ is the signal sampling), not all the 23 one-third octaves bands could be determined. At any rate, the evaluation of PSD was done using a DFT function in Matlab® and considering the Nyquist-Shannon sampling theorem [55]. In addition, to minimize the effects of performing DFT over a no integer number of cycles, the classic technique of Split-Cosine Bell windowing was used. In the Figure 1 an example of three different spectrum calculated starting from acceleration data are depicted. Wz ISO2631 frequencies weightings The ISO2631 was developed by the International Organization for Standardization (ISO) and is regarded to as a standard model adopted by several countries over the world. This standard provides several comfort levels (Table 1) introducing an overlapping zone between two adjacent ones because many factors (e.g., user age, acoustic noise, temperature, etc.) contribute to determine the degree to which discomfort could be noted or tolerated.
At any rate, the comfort levels proposed by ISO 2631 are adopted in many countries and they may be compared with the RMS values of the frequency-weighted vertical acceleration in the vehicle awz obtained inside a vehicle, giving approximate indications of likely reactions to various magnitudes of overall vibration total values in public transport.
In order to define specific limits to be used by road agencies, it is necessary to link awz values to IRI ones as proposed by many researchers [7,21,43,45,49,50]. In this way, it is possible to relate comfort perception (also influenced by vehicle characteristics) with a parameter that represents the condition and performance of road pavements surfaces.
International Roughness Index (IRI)-ASTM E 1926
The IRI was elaborated from a World Bank study in the 1980s [56] and it is one of the most adopted index used to evaluate the pavement roughness. It is based on a mathematical model called quarter-car and was developed in order to assess the pavement condition relating to all the detrimental effects such as ride quality, increasing dynamic load, tyre rolling noise, fuel consumption, and the road safety.
Many decay curves have been proposed to predict the maintenance plan over time [57] and the consequent service life of the pavement knowing its operating conditions (traffic, climate, etc.) [58].
The calculation of IRI was performed using computer program that implementing the simulation of the mechanical model considering a profile according to the Equation (2): where L is the length of the profile in km, V is the simulated speed set to 80 km/h, is the vertical displacement of the sprung mass in m, and is the vertical displacement of the unsprung mass in m. The final value is expressed in slope units (e.g., m/km or mm/m). In the present work, the algorithm proposed by the ASTM E1926 standard [59] for IRI calculation was used.
As reported in [44], there is a high heterogeneity of IRI thresholds adopted around the world. In fact, IRI limit values mainly depends from several aspects: road surface type (i.e., asphalt or cement concrete pavements), road functional category, average annual daily traffic (AADT), legal speed limit and segment length considered for IRI calculation.
The most common segment length indicated in non-US countries is equal to 100 m [44] but frequently also length of 50 m and 20 m are adopted to better take into account the contribution of the single event bumps respect to distributed unevenness.
Ride Number RN
The Ride Number (RN) is result of a mathematical algorithm obtained using two longitudinal profiles that allows the estimation of the subjective ride quality perceived by road users.
It is quite used over the world and it is correlated to the perceived comfort experimented by user riding on roughness pavement.
The RN index is the result of an international research conducted in the 1980s and sponsored by The National Cooperative Highway Research Program (NCHRP), with the aim of analyzing how the characteristics of road profiles influence the ride comfort perceived by user road [60].
The RN thresholds were obtained determining how characteristics in road profiles were linked to subjective opinion about the road from interviewed users; it is possible evaluate the pavement condition using a 0-to-5 scale where correspond respectively to "impassable" and "perfect" pavement condition (Table 2). The RN calculus requires a pavement survey using a "Class I" profiler of two profiles, and two Profile Indexes (PILeft, PIRight) were calculated adopting the algorithm reported in the ASTM E 1489 -98 [61].
The calculation of RN was performed by means of Equations (3) and (4): With some exceptions, the wavelengths' range of interest for RN is similar to that of IRI, as reported in some researches [17], in consequence good correlations can be find between IRI and RN [12]. In particular, RN presents a higher sensitivity to low wavelengths than IRI, which has a greater sensitivity to wavelengths of 16 m or longer than RN.
General architecture of the proposed sensor
The proposed low-cost and easy-to-operate device has as main aspect its similarity with smartphones regarding sensors configuration, performance, and cost. Thus, the two devices assembled and set up for the described work are composed of the following consumer-grade components: a Raspberry single board microcomputer, a micro-electrical mechanical Inertial Measurement Unit, a mini Global Navigation Satellite System (GNSS) module, a power supply, and a flashcard. This section describes these components as follows and highlights the most important features regarding the described application.
Raspberry Pi Zero W single-board microcomputer
The Raspberry Pi Zero W is a low-cost single-board microcomputer of 6.5 x 3.0 cm developed by Raspberry Pi Foundation for applications such as education and prototyping. This Raspberry model has a 512 RAM, a 1 GHz single-core microprocessor, and a 40pin general-purpose input/output (GPIO) [52]. It also has 802.11 wireless LAN (Wi-Fi) and Bluetooth connectivity, which simplifies re-mote control and data transmission without the need for uninstallation and reinstallation. The Raspberries used in the described tests run the Linux-based Raspbian operating system.
Inertial Measurement Unit
A micro-electrical mechanical (MEMS) based Inertial Measurement Unit (IMU) is a single chip multi-axis sensor that provides estimates at least linear accelerations and angular velocities and, thus, integrates accelerometer and gyroscope. Some versions of MEMS IMU single chips also integrate non-inertial sensors such as magnetometer and barometer. The recent MEMS technology progress focused on mobile gadgets has been yielding very low cost and very small smartphone-grade IMU units with a cost of about cents, size of about square centimetres, and satisfactory performance for non-critical applications. Thus, the main advantages of these inertial sensors when compared with traditional mechanical and solid-state sensors are the size reduction, the low power consumption and the low production cost [62].
For this research, we used the InvenSense MPU-9250, a 10 degrees-of-freedom module of 1,4 x 1,4 cm. This inertial module integrates the three-axis MEMS inertial sensors (accelerometer and gyro-scope) to a magnetometer and a pressure module BMP280 (a barometer plus a thermometer) [63,64]. The voltage readings from the inertial sensors are digitized using on-chip 16-bit resolution Analog-to-Digital Converters (ADC) for each axis, and this digital output is sent to the Raspberry through inter-integrated circuit (I2C) interface. Besides the raw measurements, the MPU-9250 module measures and has a digital motion processor that provides fused output for gesture recognition applications. Table 3 presents the main features of MPU-9250 accelerometer, gyroscope, and magnetometer. The C++/Python library named RTIMULib [65] was used for sensors setup, initial calibration on Raspbian, and conversion of values form hexadecimal to floating-point representation. The following data is obtained: i) three-axial raw linear accelerations (including gravity) in the sensor frame, in g; ii) three-axial raw angular velocities in the sensor frame, in rad/s; iii) three-axial raw magnetic field in the sensor frame, in µT; iv) pressure, in hPa; v) height derived from the barometric calculation, in m; vi) temperature, in °C; vii) sensor attitude (roll, pitch, and yaw, in degrees). Regarding attitude data, angles are obtained by RTIMULib through Extended Kalman Filter (EFK) integrating inertial and magnetic data, a technique that adapts Kalman Filter to a nonlinear problem such as the attitude estimation.
The output data rate was set up at 100 Hz given the optimum performance on preliminary tests, the aimed data analyses, and the usual sample rate for medium-grade smartphones. However, the maximum mean sample rate effectively obtained during operation (> 10 s) was about 83 Hz owing to hardware and software limitations. Table 4. The output rate for the GPS module was set up at 1 Hz regarding the performance during pre-liminary tests, and the sample rate for medium-grade smartphones. The lower sample rate in com-parison with IMU rate requires interpolation of PVT data using the OS timestamp as the key attribute. Furthermore, GPS and IMU data are recorded in separated files since it has the best performance un-der the abovementioned configuration.
The Python library called GPSD [69] allows for the acquisition, on Raspbian environment, of position, velocity, and time (PVT) data through US National Marine Electronics Association (NMEA) protocol. The following GPS data is obtained: geographic coordinates (latitude and longitude) of the acquisition point referred to WGS84 datum (GPS datum), geometric height, UTC time of the acquisition point, velocity, number of visible satellites, and uncertainty-related parameters. Figure 2 shows the core components already assembled. The IMU module and the GPS module were connected to the processing unit and glued to the Raspberry case.
Other components
Each Raspberry operates with a 16 GB micro-SD used to store the operating system and the gathered data. Moreover, power is supplied by a portable rechargeable battery unit with 10,400 mAh capacity through a micro-USB port. Considering storage and power capacities under the aforementioned configuration, the sensor sets presented an autonomy of at least 50 hours during the preliminary test.
LandMark 10 GPSA-150-10-200
In order to validate the results of the measurements made with the proposed sensor on the same test road test also the pre-assembled inertial platform LandMark 10 GPSA-150-10-200 was employed.
The most important product characteristics are summarized in the code name, that reporting the operating range of both gyroscopes (±150°/sec) and accelerometers (±10g's) as well as the product type GPS/AHRS (Attitude Heading Reference System).
The main component parts of this instrument are: • the Inertial Measurement Unit (IMU) (Figure 3a); • the integrated GPS receiver ( Figure 3b); • the power supply to connect to a laptop (Figure 3c). This connection also allows recording the data measured to a comma-separated values (CSV) file. The software named "GLAMR" to acquire the data have to be installed in a standard Notebook.
The Kalman filter is automatically implemented inside the LandMark 10 GPSA-150-10-200; the Kalman filter is an efficient recursive filter that evaluates the state of a dynamic system starting from a series of measurements subject to noise. Its use allows to eliminate part of the background noise that could affect the measurements.
Automatic pavement data-collection vehicle
In order to validate the results of the measurements made with the proposed sensor on the same test roads, a Road Surface profiler was employed. For this purpose, a multifunction Mobile Measurement System was used thanks the support of the Laboratory of Road Materials and Maintenance of the Italian National Road Agency (Centro Sperimentale Strade di Cesano di ANAS S.p.A. Gruppo Ferrovie dello Stato Italiane) This mobile laboratory named "Cartesio" was designed according to the Department of Road Maintenance of Italian National Roads Department (ANAS). The system has been operating since 2018 on the whole road network managed by ANAS.
Mobile Mapping is the term that identifies the techniques of detection from moving vehicles; the data obtained from on-board sensors are georeferenced as a positioning and orientation system is mounted.
The main fields of application of this multifunction road-quality surveying instrument are updating of the road cadastre, road maintenance and the survey of infrastructure elements, networks and services that meet along the road axes.
The main components of this high-performance vehicle are: • the positioning and orientation system so as to georeferenced the data collected from on-board sensors; • on-board sensors (5 high resolution digital cameras; 2 LIDAR Laser Scanner; Laser Crack Measurement System (LCMS); 3 inertial profilometers); • the synchronization system coordinated by a management system.
Other auxiliary systems are: • the data storage system; • the power supply system for equipment and documents. All the components are permanently installed on a Fiat Ducato 290.
Field Tests for system validation
With the aim to validate the proposed IMU, some field tests were carried out using two identical prototypes of the device described in section 3.1; in this paper, these two IMUs can be distinguished with the code "SENSORS#1" and "SENSORS#2".
The accelerometer data recorded by the sensors placed inside test vehicles were processed using program code written in MATLAB® in order to get awz index values every 1 second. For the acceleration signals, an analysis time of 2 seconds was considered, so, for each device, an overlap of the acceleration signals were obtained of 1 second.
The validation test was performed using all the devices at the same time identifying a total of about 21 km of roads ( Figure 4) with flexible pavement located in the northern outskirts of Rome. The route started and ended at the same section; it was articulated on both urban and nonurban roads (Table 5) Field tests were carried out without closing roads to traffic and no change in driving behavior was requested to the drivers of test vehicles in which the sensors were placed, so speed value recorded during the measurements were variable in consequence to the road and traffic condition ( Figure 5).
Figure 5. Recorded speed in a test vehicle
In order to have a correct interpretation of the results obtained from the proposed SENSORS#1 and SENSORS#2, the ultimate aim is to evaluate the possible use of such devices for pavement condition analysis. In addition, during the measurement campaigns also additional instruments were used: • LandMark 10 GPSA-150-10-200, a precision measuring instrument [70] with sampling frequency equal to 100 Hz. Post-processing acceleration data recorded from this IMU was aimed to obtain the frequency-weighted vertical acceleration awz considering analysis times by one second each; • Mobile Mapping System "Cartesio" with 3 inertial profilers (PaveProf System model) able to measure the road profiles in the left and right wheel paths as well as in the center lane ( Figure 6). This system is able to determine IRI and RN considering subsections of 10 meters length each. Table 6. The SENSOR#1 was positioned on the passenger side floor of a Renault Zoe together with the LANDMARK, while the SENSOR#2 was installed on "Cartesio" dashboard. No particular details (foams, rubber layers or similar) were adopted for fixing the device parts to the box support (only screws, bolts and rubber bands) or to the vehicle dashboard (only double-sided tape), because it is foreseen in the future that these instruments should be able to be simply mounted without special provisions. Table 6. Characteristics of the IMUs' position inside the vehicle during the tests.
Results and discussion
A first comparison between the low-cost pavement monitoring SENSOR#1 and the LANDMARK was in terms of speed values measured during the survey by both devices at the same time (Figure 8).
The two IMUs collected speed values with different frequency rate: 100 Hz for the LANDMARK and 1 Hz for the proposed SENSOR#1. The comparison between speed values collected by two IMUs at the same time sample showed a good correlation (Figure 9).
In some isolated position, a maximum of 20% of difference was registered, and, in total, an average total value of only 0.2% between the two speed values was resulted. whole vibration index awz every 1 second of accelerometer signal measured using different IMUs were obtained.
Considering that the vehicle speed in the IMU device tests was variable around 10-16 m/s, consequently it was possible to determine the measurements of the aforementioned indexes every 1 second and, therefore, approximately every 10-16 m. It was not considered useful, as well as difficult, to exactly match the position in which all the indexes (awz obtained with 3 devices, IRI and RN) were available ( Figure 10). For this reason, in this preliminary validation phase, fixed and constant long sub-sections (100m) of road were considered. The average value of the indexes that the positions were included in a generic section were assumed representative for that sub-section. For operational reasons related to the use of the manager's profiler, the considered total section of urban and nonurban road network of 21 km was measured during the morning of a working day with traffic conditions such as not to always allow to the vehicle the minimum speed for the correct survey of IRI and RN measures.
Consequently, not all the collected measurements were considered useful for validation.
In the 100m sub-sections where the speed of the profiler vehicle was greater than the minimum value considered acceptable for the indexes reliability, a subdivision into performance classes with reference to the pavement decay was adopted.
Three different pavement condition categories ("Good", "Fair", and "Poor") derived from related researches [22,43,47,71] were adopted considering the IRI threshold values (Table 7). For this validation phase, it is assumed to consider road sections where the pavement conditions did not vary continuously from a sub-section of 100m to the next or the before. On the other hand, the variability of pavement conditions is quite frequent during a normal survey regardless of whatever index is adopted. For this reason, in the usual practice of the pavement monitoring procedure, it is necessary to identify appropriate homogeneous sections in relation to the deterioration conditions surveyed [72].
On the contrary, during the validation procedure, in the entire 21km road section, 3 sufficiently long sections (at least equal to 400 m, containing 4 sub-sections) respectively in good, fair and poor conditions were identified ( Figure 11 and Table 8). It was calculated the average value of each index per sub-section length of 100 m
Comparison between SENSOR#1-awz and LANDMARK-awz
The first step in results analysis process was to find a relationship between awz values calculated from data collected by SENSOR#1 and awz values based on data collected by LandMark 10 GPSA-150-10-200 ( Figure 12); for clarification purposes it's important to underline that both devices, one next to the other, was inside the same vehicle during the same test. The regression results showed very good correlations between the frequencyweighted vertical accelerations calculated with the proposed IMU (SENSOR#1) and ones in the reference IMU (LANDMARK). The coefficient of determination (R 2 =0.98) indicates a strong concordance between measurements from the SENSOR#1 and the LandMark10, although the former presented a smaller accuracy and an it were obtained awz index values about 10% greater than the ones of reference IMU.
Comparison between SENSOR#1-awz and SENSOR#2-awz
This paragraph focuses on the comparison between the frequency-weighted vertical acceleration values based on data collected from SENSOR#1 and SENSOR#2 respectively ( Figure 13). The dispersion of data points around the regression line and a non-unit slope are in agreement that the two identical prototypes were placed in different positions inside vehicles which in turn differed in the physical and mechanical characteristics and also in the recorded speeds: these factors significantly influence the final values of awz index. For the purpose of this research, it is also important the comparison between awz index, calculated from data collected in the field tests respectively by SENSOR#1 and SEN-SOR#2, and the values of International Roughness Index and Ride Number related to the data acquired by "Cartesio" (Figure 14) Figure 14. Linear regression awz -IRI vs awz -RN As shown in Figure 14, the calculated awz by the proposed system and the indexes determined by RSP are highly correlated in all cases with the coefficient of determination more than 0.83. Considering the awz determined with acceleration data measured by IMU located in the same vehicle where the RN were measured, a higher coefficient of determination was obtained.
Conclusions
This work aimed to verify the feasibility of using a Raspberry-based IMU device for monitoring the road pavement condition in urban areas. Tests were carried out using two identical Raspberry-based prototypes along about 21 km of urban and nonurban roads with flexible pavement located, and the validation was performed employing as reference concomitant measurements by the IMU LandMark10+GPS and the "Cartesio" Mobile Measurement System vehicle.
RN
Considering the comfort index awz in accordance with ISO 2631 standard, the results showed very good correlations between the frequency-weighted vertical accelerations calculated with the proposed IMU (SENSOR#1) and ones in the reference IMU. The coefficient of determination (R 2 =0.98) indicates a strong concordance between measurements from the SENSOR#1 and the LandMark10, although the former presented a smaller accuracy and an awz index value 10 per cent greater than the reference IMU. Besides, the comparison between the two Raspberry-based devices yielded a coefficient of determination (R 2 =0.73), with discrepancy explained by the fact that the sensors were installed in different vehicles and different positions inside each vehicle. This leads to an initial indication of how speed and the vehicle's physical and mechanical characteristics may affect the estimate of the comfort indicator awz.
Furthermore, the evaluation of the correlation between data gathered by "Cartesio" (IRI and RN) and the awz indexes calculated by SENSOR#1 and SENSOR#2 revealed a good consistency between the measurements. The correlation coefficients were in all arrangements greater than 0.82, implying a high correlation between the reference data and the measurements obtained from the proposed devices. It must be emphasized that the greatest correlation (R²=0.90) was, as expected, verified between RN and the frequencyweighted vertical accelerations calculated from the SENSOR#2 located inside the "Cartesio".
It may be concluded that the proposed sensor can be considered a valuable tool for a quick, low-cost road survey if considered that the repeatability of the results is conditioned by the speed and physical-mechanical characteristics of the vehicle. The proposed study is not intended to establish the Raspberry platform on the same level as the other precision devices. Instead, the correct interpretation is to provide an affordable tool that does not require dedicated staff and that can be easily installed in public service vehicles, local public transport vehicles, and even two-wheeled vehicles, widening the range of monitorable pavements (including sidewalks and bike lanes).
Since the described device is a prototype, it could be possible to perform improvements such as its integration with a GSM unit to transmit data directly to a server. In this context, for reasons of repeatability, information such as type of vehicle, position inside the vehicle, the fixing system and, finally, the speed at which the recording was carried out would be mandatory to enable a weighted evaluation of the measurements. Remaining within the scope of instrumentation refinement, it could be envisaged to develop a GIS system for the positioning and cataloguing of measurements in terms of awz index in order to enable better integration with traditional measurement systems. | 7,681.8 | 2021-04-01T00:00:00.000 | [
"Engineering",
"Environmental Science"
] |
A Spatial Compounding Method for Non-Delayed Sequential Beamforming
: We present a new spatial compounding method to improve the contrast of ultrasonic images for non-delayed sequential beamforming (NDSB). Sequential beamforming adopts more than one beamformer to reconstruct B-mode images which has the advantage of simple front-end electronics and fast data transfer rate. Via field pattern analysis, we propose a compounding method where two more sub-images can be reconstructed along with the NDSB sub-image. These sub-images can be seen as being produced with different transmit origins; thus, their summation enhances image contrast. Image quality was analyzed in terms of spatial resolution, contrast ratio (CR), and contrast-to-noise ratio (CNR). The proposed compounding method improves the lateral resolution up to 41%. In vitro results confirm a 13.0-dB CR and 4.0-dB CNR improvement. In vivo results reveal 10.9-dB and 6.0-dB improvement in CR and CNR for cross-section jugular vein and 8.0-dB and 4.5-dB improvement in CR and CNR for the longitudinal-section carotid artery.
Introduction
Synthetic aperture focusing technique (SAFT) was first introduced for the singleelement transducer. A 'virtual source' concept is used by treating the focus point as a point source emitting spherical waves towards and away from the transducer [1,2]. Improved resolution is obtained for areas away from the focus. For linear array, similar algorithms have been introduced [3][4][5][6][7][8].
Although the image quality is significantly improved, the computational cost of synthetic aperture beamformer is heavy. The work of Nguyen [6] showed that the size of the beamformed image is 30 mm × 30 mm. The computation time for a traditional synthetic aperture beamformer is 59 min, and the advanced synthetic aperture beamformer takes 145 min, whereas dynamic focusing takes 42 s with an O(n t n t n s ) computation complexity (n s denotes the number of transducer element and n t denotes the number of time samples). The requirement of real-time imaging demands the beamformer to be computationally efficient while retaining the image quality.
A dual-stage beamformer combining an advanced beamformer and traditional delayand-sum (DAS) beamformer has been proposed to enhance the resolution and contrast [9,10]. A dual-stage beamformer (or termed 'sequential beamforming') has also been addressed in the context of synthetic aperture imaging [11][12][13]. A sequential beamforming algorithm termed non-delayed sequential beamforming (NDSB) is developed to simplify the frontend electronics design [13]. In the first stage, the received signals are directly summed without delay. In the second stage, the high-resolution image is reconstructed based on the two-way travel path. The authors proved this new approach can achieve a comparable imaging quality with a simpler receive electronic front-end than dynamic receive focusing (DRF). Due to the lack of focusing in the first beamformer, NDSB requires a large transmit opening angle to combine more data to generate a high-resolution image in the second stage. This means a large number of active elements. Even so, the contrast of NDSB is still inferior to dynamic receive focusing.
Spatial compounding can enhance image contrast by combining limited-correlating images of the same region-of-interest (ROI). To generate the limited-correlating images, we can mechanically translate the transducer, or electrically activate different transmit elements, or apply different transmit apodization [14,15]. To date, spatial compounding methods have not been applied to sequential beamforming in synthetic aperture imaging.
Here, a new spatial compounding method is proposed for NDSB. Spatiotemporal analysis of the pressure field is performed. By modeling the transmit sub-aperture as a continuous curved aperture, the pressure field can be characterized by three excitation signals with different transmit origins and different time delays. Sub-images can be reconstructed based on the propagation path of the corresponding excitation signals. Because these excitation signals are spatially separated, the reconstructed sub-images carry different spatial frequencies and thus decorrelate from each other. Spatial compounding can be done via summing these sub-images. The proposed compounding method improves contrast ratio and contrast-to-noise ratio in-vitro, and in-vivo.
Non-Delayed Sequential Beamforming
Non-delayed sequential beamforming introduced in [13] proposes a new dual-stage beamformer. A low-resolution image (LRI) is produced by the direct summation of channel data at the first stage. A second beamformer then reconstructs a high-resolution image (HRI) using the LRI. The time-of-flight (TOF) for the NDSB is shown in Figure 1, which is given as pl. Sci. 2021, 11, x FOR PEER REVIEW 3 Figure 1. Wave propagation path for NDSB. ⃗, ⃗, ⃗, and ⃗ denote the positions of the vir source, the receive element, the transmit origin, and the imaging point, respectively. The blu rows indicate the transmit path, and the red arrows indicate the receive path. The active elem are indicated as black.
Pressure Field Analysis
The spatiotemporal impulse response of the transmit sub-aperture is used to ext excitation signals which correspond to different transmit configurations [6]. For simp ity, we consider only the region where the spherical wave assumption is valid.
We make some assumptions: all elements of the transducer share the same elec mechanical impulse response with uniform directivity function, and the discrete trans sub-aperture is modeled as a continuous curved aperture. The pressure field of the im ing area can be obtained through a spatiotemporal impulse response [16]. In the equations, → r VS , → r r , → r e , and → r ip denote the positions of the virtual source, the receive element, the transmit origin, and the imaging point, respectively. c is the speed of sound. To reconstruct (x, z) in the HRI, a time-delay ∆t is applied to the LRI data at the lateral position x f : where z f is the focus depth. The high-resolution image can be written as where HRI NDSB denotes the high-resolution image for NDSB, W is the apodization function, and LRI denotes the low-resolution image.
Pressure Field Analysis
The spatiotemporal impulse response of the transmit sub-aperture is used to extract excitation signals which correspond to different transmit configurations [6]. For simplicity, we consider only the region where the spherical wave assumption is valid.
We make some assumptions: all elements of the transducer share the same electromechanical impulse response with uniform directivity function, and the discrete transmit sub-aperture is modeled as a continuous curved aperture. The pressure field of the imaging area can be obtained through a spatiotemporal impulse response [16].
Assuming a sinusoidal excitation v(t) and a rigid baffle, the pressure field for point x p at time t can be written as where ρ is the equilibrium density of the medium, and h( → x p , t) in the equation can be given by [16].
where R is the distance from the point x p to the aperture, and β(R) is the corresponding angle in the polar coordinates. Integrating the above equation over the aperture surface, the pressure field for the point above the focus in Figure 2a can be expressed as where β 1 , β 2 , and β i are associated with R 1 , R 2 , and R min . For the point below the focus in Figure 2b, the pressure can be expressed as where β i is associated with R max . Equations (7) and (8) show that the transmit wave shape can be characterized by three spherical waves. The total field is a combination of the three pulses with differen transmit origins and different time delays. NDSB selects the excitation signal with the highest energy to do the reconstruction which is denoted as − in Equation (7) Equations (7) and (8) show that the transmit wave shape can be characterized by three spherical waves. The total field is a combination of the three pulses with different transmit origins and different time delays. NDSB selects the excitation signal with the highest energy to do the reconstruction which is denoted as v(t − R min c ) in Equation (7) and v(t − R max c ) in Equation (8). On the other hand, beamforming can also be performed under two other excitation signals from the two ends of the aperture. Because the three excitation signals are from different origins, their corresponding beamformed images are therefore spatially decorrelated. Compounding can be done by summing these images together.
To reconstruct the image originated from the two transmit aperture ends, the wave propagation path needs to be adjusted.
The TOF corresponds to the excitation signal emitted from the ends of the active aperture can be expressed as Figure 3a presents the TOF to generate the image corresponds to the left end of the active aperture, and Figure 3b corresponds to the right end. To produce HRI at (x, z), the delay time ∆t is also adjusted as where x O denotes the lateral position of the left/right end of the active aperture. The high-resolution image is denoted as HRI le f t_end and HRI right_end using Equation (2).
Equations (7) and (8) show that the transmit wave shape can be characteriz three spherical waves. The total field is a combination of the three pulses with dif transmit origins and different time delays. NDSB selects the excitation signal wit highest energy to do the reconstruction which is denoted as − in Equati and − in Equation (8). On the other hand, beamforming can also be perfo under two other excitation signals from the two ends of the aperture. Because the excitation signals are from different origins, their corresponding beamformed imag therefore spatially decorrelated. Compounding can be done by summing these im together.
To reconstruct the image originated from the two transmit aperture ends, the propagation path needs to be adjusted.
The TOF corresponds to the excitation signal emitted from the ends of the activ erture can be expressed as Here we produce two compound images using the three high-resolution images (HRI NDSB , HRI le f t_end and HRI right_end ): summation of HRI le f t_end and HRI right_end gets compound image 1 (denoted as Comp1). Summation of HRI NDSB , HRI le f t_end and HRI right_end gets compound image 2 (denoted as Comp2).
Experiment Setup
The Verasonics Vantage 256 scanner (Redmond, WA, USA) and a L11-4v linear array were used to collect experimental data. The array has a 6.25-MHz central frequency, Appl. Sci. 2021, 11, 9200 5 of 10 128 elements, 67% bandwidth, and 0.3-mm pitch size. The excitation signal is a 1-cycle sinusoid tone burst at the central frequency. Transmit F-number (F #, defined as the ratio between focal length and the aperture diameter) is set to 0.75 for all experiments. During transmission, the number of the active elements is 16, boxcar apodization is used for every transmit event. The focused beam is translated by an element pitch in the lateral direction of 0.3 mm. The transmit focus is always on the centerline of the active aperture. During reception, all 128 elements are active to collect radio-frequency data. The raw data are sampled at 25 Msamples/s, resulting in a 0.078 mm-axial sampling distance. The Verasonics system stores 4096 samples of each receiver signal which covers data ranging from 3 mm to 34 mm in depth. Off-line processing is performed to produce the final images. B-mode images were obtained through Hilbert transform and log-compression. A phantom made of nylon threads of 0.2-mm diameter immersed in water was used for the quantification of the lateral resolution, and a lab-made cyst phantom is used to quantify realistic image contrast. In vivo imaging is performed to image the long-section and cross-section of the carotid artery.
Evaluation
Two proposed compound images are present, denoted as Comp1 and Comp2. To provide a fair comparison, both NDSB and DAS images are shown.
The reconstructed images are evaluated in terms of resolution and image contrast. The resolution is quantified by imaging a point target. We define the resolution as the full-width-at-half-maximum (FWHM) in the lateral and axial direction. The performance of image contrast is evaluated in terms of contrast ratio (CR) which is defined as ), where I in denotes the mean intensities of the measured cyst and I out denotes the mean intensities of the measured background.
The contrast-to-noise ratio (CNR) value is also used which is defined as ).
The unit of CR and CNR is decibel (dB).
Resolution
To measure the ability of the proposed beamformer to resolve a point target, a phantom made of nylon threads of 0.2-mm diameter was immersed in water. The distance between the target to the transducer surface is 15.1 mm. Four beamformers are used and the results are compared. Figure 4 presents the beamformed images with a 30-dB dynamic range. It can be seen that NDSB produces a slightly larger point spread function (PSF) compared with DAS. It can be seen in Figure 4c,d that the side lobes for proposed beamformers are slightly asymmetric. The reason is in Section 2.2 we assume the pressure field contributions from the two aperture ends are equal. But for real discrete transmit sub-aperture, these contributions might not be perfectly the same. Weighting strategy can be used for each contribution, but the design of weighting parameters is beyond the scope of current work. The proposed compound method in Figure 4c,d improves the resolution of NDSB. Figure 5 plots the lateral and axial profile of the beamformed point target. The detailed resolution in the lateral and axial direction is in Table 1. The proposed Comp1 method provides the best resolution in both directions, which is 1.37 mm in the lateral direction and 0.43 mm in the axial direction. The proposed two compounding methods (Comp1, Comp2) outperform DAS and NDSB in terms of resolution. current work. The proposed compound method in Figure 4c,d improves the resolution of NDSB. Figure 5 plots the lateral and axial profile of the beamformed point target. The detailed resolution in the lateral and axial direction is in Table 1. The proposed Comp1 method provides the best resolution in both directions, which is 1.37 mm in the lateral direction and 0.43 mm in the axial direction. The proposed two compounding methods (Comp1, Comp2) outperform DAS and NDSB in terms of resolution.
Contrast
To test the robustness of the beamformers in terms of phase aberration. A lab-made cyst phantom is used to compare image contrast under different beamformers. This phantom has a random background and two anechoic cysts. The left cyst is centered at (x, z) = (−8, 22) mm and the right one is centered at (x, z) = (4, 18) mm. The anechoic cysts are method provides the best resolution in both directions, which is 1.37 mm in the latera direction and 0.43 mm in the axial direction. The proposed two compounding method (Comp1, Comp2) outperform DAS and NDSB in terms of resolution.
Contrast
To test the robustness of the beamformers in terms of phase aberration. A lab-mad cyst phantom is used to compare image contrast under different beamformers. This phan tom has a random background and two anechoic cysts. The left cyst is centered at (x, z) = (−8, 22) mm and the right one is centered at (x, z) = (4, 18) mm. The anechoic cysts ar
Contrast
To test the robustness of the beamformers in terms of phase aberration. A labmade cyst phantom is used to compare image contrast under different beamformers. This phantom has a random background and two anechoic cysts. The left cyst is centered at (x, z) = (−8, 22) mm and the right one is centered at (x, z) = (4, 18) mm. The anechoic cysts are surrounded by randomly placed scatterers, the attenuation coefficient is 0.6 dB/MHz/cm and the speed of sound within the phantom is 1539-1550 m/s.
The beamformed images are shown in Figure 6. For CR and CNR calculation, the measured cyst region is enclosed in the white circle and the background in the white rectangle in Figure 6. The background region with uniform speckle patterns shares the same centerline as the measured cyst and contains no particular bright reflectors. The detailed CR and CNR values are given in Table 2. and the speed of sound within the phantom is 1539 m/s−1550 m/s. The beamformed images are shown in Figure 6. For CR and CNR calculation, the measured cyst region is enclosed in the white circle and the background in the white rectangle in Figure 6. The background region with uniform speckle patterns shares the same centerline as the measured cyst and contains no particular bright reflectors. The detailed CR and CNR values are given in Table 2. From the beamformed images, we can find cyst region enclosed by the white circle gets cleaner after compounding. In Table 2, Comp2 has the better CR performance which is 11.1 dB higher than DAS and 13.0 dB higher than NDSB. CNR value also validates the improvement in image contrast. Comp2 gives the best CNR value which provides 3.9-dB and 4.0-dB improvement compared to DAS and NDSB, respectively.
In Vivo Imaging
To test the robustness of the proposed compounding method in clinical circumstances, we acquired in vivo data from a 26-year-old volunteer using the Verasonics scanner. This data was obtained with appropriate ethical clearance and informed consent. During the experiment, the probe and the human object are kept still to eliminate large motion artifacts. Figure 7 shows the in vivo images. The image contains the transverse view of two vessels in the neck, including the jugular vein and the carotid artery. We choose the region inside the jugular vein to calculate CR and CNR values. The ROI is enclosed in the white circle in Figure 7. The background is depicted in the white rectangular which is within the tissue region in the thyroid gland. After compounding, the contrast for the jugular vein improves. The measured CR and CNR are shown in Table 3.
From Table 3, the proposed Comp2 gives the best CR and CNR performance, Comp1 also gives similar performance. Comp2 outperforms DAS by 9.7 dB and 5.2 dB in terms of CR and CNR. Compared with NDSB, the compounding method provides up to 10.9-dB and 6-dB improvement in CR and CNR, respectively. From the beamformed images, we can find cyst region enclosed by the white circle gets cleaner after compounding. In Table 2, Comp2 has the better CR performance which is 11.1 dB higher than DAS and 13.0 dB higher than NDSB. CNR value also validates the improvement in image contrast. Comp2 gives the best CNR value which provides 3.9-dB and 4.0-dB improvement compared to DAS and NDSB, respectively.
In Vivo Imaging
To test the robustness of the proposed compounding method in clinical circumstances, we acquired in vivo data from a 26-year-old volunteer using the Verasonics scanner. This data was obtained with appropriate ethical clearance and informed consent. During the experiment, the probe and the human object are kept still to eliminate large motion artifacts. Figure 7 shows the in vivo images. The image contains the transverse view of two vessels in the neck, including the jugular vein and the carotid artery. We choose the region inside the jugular vein to calculate CR and CNR values. The ROI is enclosed in the white circle in Figure 7. The background is depicted in the white rectangular which is within the tissue region in the thyroid gland. After compounding, the contrast for the jugular vein improves. The measured CR and CNR are shown in Table 3. The longitudinal section beamformed images of the carotid artery are shown in Figure 8. Visually, the contrast within the carotid artery improves with the proposed method. The ROI is chosen within the carotid artery depicted in the white circle in Figure 8. The From Table 3, the proposed Comp2 gives the best CR and CNR performance, Comp1 also gives similar performance. Comp2 outperforms DAS by 9.7 dB and 5.2 dB in terms of CR and CNR. Compared with NDSB, the compounding method provides up to 10.9-dB and 6-dB improvement in CR and CNR, respectively.
The longitudinal section beamformed images of the carotid artery are shown in Figure 8. Visually, the contrast within the carotid artery improves with the proposed method. The ROI is chosen within the carotid artery depicted in the white circle in Figure 8. The background region is within the white rectangular. The measured CR and CNR are shown in Table 4. The longitudinal section beamformed images of the carotid artery are shown in Figure 8. Visually, the contrast within the carotid artery improves with the proposed method. The ROI is chosen within the carotid artery depicted in the white circle in Figure 8. The background region is within the white rectangular. The measured CR and CNR are shown in Table 4.
From Table 4, the proposed Comp2 gives the best CR and CNR performance, Comp2 outperforms DAS by 6.4 dB and 1.4 dB in terms of CR and CNR. Compared with NDSB, the compounding method provides up to 8.0-dB and 4.5-dB improvement in CR and CNR, respectively.
Discussion
This paper exploits a spatial compounding method for non-delayed sequential beamforming. Our proposed method depends on theoretical modeling of the transmit pressure field. As a result, two beamformed images originated from two ends of the transmit aperture can be produced using the same received signals. From the in vitro results, our proposed compounding images significantly improve the image contrast without sacrificing From Table 4, the proposed Comp2 gives the best CR and CNR performance, Comp2 outperforms DAS by 6.4 dB and 1.4 dB in terms of CR and CNR. Compared with NDSB, the compounding method provides up to 8.0-dB and 4.5-dB improvement in CR and CNR, respectively.
Discussion
This paper exploits a spatial compounding method for non-delayed sequential beamforming. Our proposed method depends on theoretical modeling of the transmit pressure field. As a result, two beamformed images originated from two ends of the transmit aperture can be produced using the same received signals. From the in vitro results, our proposed compounding images significantly improve the image contrast without sacrificing the resolution. In vivo studies also demonstrate the robustness of our proposed compounding methods. Image contrast is greatly enhanced in both the longitudinal and cross-section carotid artery images.
The aim of NDSB is to simplify the front-end architecture and retain the beamformer's quality comparable to DRF. The image quality is highly affected by the transmit F#: smaller F# results in better resolution and contrast. Small F# means larger transmit opening angle. This ensures more data in the low-resolution image can be synthesized to generate a pixel point in the high-resolution image. The optimal F# is set to 0.75 in [13]. With this setting, degrading lateral resolution is found at shallow depth (<40 mm) and inferior contrast (CR, CNR) compared with DRF.
In this work, the transmit F# is also 0.75. NDSB produces inferior image quality (lateral resolution, CR, CNR) compared to DAS. This conclusion is the same in [13]. The proposed compounding improves the resolution and image contrast by summation of limited-correlated images. The effect of transmit F# to the compounding method is illustrated in Figure 9. Two transmit F# are compared: 0.75 and 3. To measure the lateral resolution, the same point target in Figure 4 is used. The result is shown in Figure 9a. The y-label is the ratio of lateral resolution of NDSB, Comp1, and Comp2 to that of DAS in certain transmit F#. When the F# increases, the lateral resolution of the compound method degrades. To address the image contrast, we image the carotid artery same in Figure 7. The ratio of CR and CNR (to that of DAS) in terms of transmit F# is plotted in Figure 9b. We can find the same conclusion: when F# increases, the image contrast also degrades. This is easily understood: a larger F# means a smaller active aperture given the same focus depth. The transmit origins for the contributing signal get closer and the beamformed images carry almost the same spatial frequencies. The summation of these images gives limited improvement compared with a small F# (larger active aperture). The computation time to reconstruct an image of 30 mm × 30 mm is 7.2 s for DAS, 8.5 s for NDSB, 18.5 s for Comp1, and 19.2 s for Comp2.
in Figure 9. Two transmit F# are compared: 0.75 and 3. To measure the lateral r the same point target in Figure 4 is used. The result is shown in Figure 9a. The the ratio of lateral resolution of NDSB, Comp1, and Comp2 to that of DAS in cert mit F#. When the F# increases, the lateral resolution of the compound method To address the image contrast, we image the carotid artery same in Figure 7. Th CR and CNR (to that of DAS) in terms of transmit F# is plotted in Figure 9b. W the same conclusion: when F# increases, the image contrast also degrades. Thi understood: a larger F# means a smaller active aperture given the same focus d transmit origins for the contributing signal get closer and the beamformed ima almost the same spatial frequencies. The summation of these images gives lim provement compared with a small F# (larger active aperture). The computatio reconstruct an image of 30 mm × 30 mm is 7.2 s for DAS, 8.5 s for NDSB, 18.5 s fo and 19.2 s for Comp2.
Conclusions
In this work, we propose two spatial compounding methods for a sequent forming algorithm named non-delay sequential beamforming (NDSB). Pres analysis indicates different sub-images can be beamformed under different tran figurations. The summation of these sub-images produces a compound final vitro and in vivo studies are carried out to quantify the proposed compounding The results show the proposed compounding methods significantly improve detectability: in vitro results confirm a 13.0-dB better CR and 4.0-dB better CNR c with NDSB. In vivo imaging of cross-section jugular vein reveals 10.9-dB and 6 provement in CR and CNR using the proposed method. For the longitudinal-s rotid artery, the compounding method provides up to 8.0-dB and 4.5-dB improv CR and CNR, respectively.
Conclusions
In this work, we propose two spatial compounding methods for a sequential beamforming algorithm named non-delay sequential beamforming (NDSB). Pressure field analysis indicates different sub-images can be beamformed under different transmit configurations. The summation of these sub-images produces a compound final image. In vitro and in vivo studies are carried out to quantify the proposed compounding methods. The results show the proposed compounding methods significantly improve the target detectability: in vitro results confirm a 13.0-dB better CR and 4.0-dB better CNR compared with NDSB. In vivo imaging of cross-section jugular vein reveals 10.9-dB and 6.0-dB improvement in CR and CNR using the proposed method. For the longitudinal-section carotid artery, the compounding method provides up to 8.0-dB and 4.5-dB improvement in CR and CNR, respectively. Informed Consent Statement: Informed consent was obtained from all subjects involved in the study.
Data Availability Statement:
The data that support the findings of this study are available on request from the corresponding author. | 6,231.4 | 2021-10-02T00:00:00.000 | [
"Engineering",
"Physics"
] |
A Highly Segmented Neutron Polarimeter for A1
A new neutron polarimeter for measuring the neutron's electric form factor was designed and constructed to complement the A1 spectrometer setup at the Mainz Microtron (MAMI). The design is based on a previous polarimeter with significant improvements to halve the error of the extracted form factor. A higher granularity of the polarimeter sections and a deeper first section on the one hand, and a faster readout employing Time-over-Threshold methods to measure the signal amplitudes combined with a high-precision FPGA-based TDC on the other hand will allow to achieve this goal. The performance of the new polarimeter during a first measurement campaign in 2019 using liquid hydrogen and deuterium targets will be discussed.
Introduction
Measuring the neutron's electric form factor G E,n requires a sophisticated setup.The detection of neutrons with high momenta (several hundred MeV/c) is notoriously difficult and needs large detector volumes.This leads to a susceptibility for unwanted background processes, especially at high beam currents needed for such a measurement.
The general design concept and measurement principle, which was already used at A1 [1,2], is explained in Sec. 2. The polarimeter's construction and readout and trigger concept are discussed in Sec. 3 to 5. Finally, the performance of the new polarimeter during a beam time with liquid hydrogen and deuterium targets in 2019 is shown in Sec. 6.
Concept and Polarimetry
In order to measure the neutron electric form factor, G E,n , in a model-independent way at the high Q 2 , the same method that previously has been established at MAMI [1,2] will be used.A beam of polarised electrons hits an unpolarised, liquid deuterium target.In the quasielastic reaction D(⃗ e, e ′ ⃗ n)p polarised neutrons are produced.The components of the outgoing neutron polarisation carry the information on its electric and magnetic form factors.For free (e,n) scattering Arnold, Carlson and Gross obtained [3]: 2 ) P y = 0 (1) The coordinate frame (x, y, z) is defined relative to the electron scattering plane and the direction of the momentum transfer: ẑ = ⃗ q/|⃗ q|, ŷ = ⃗ p e i × ⃗ p e f /|⃗ p e i × ⃗ p e f |, x = ŷ × ẑ.The variable τ = Q 2 /4M 2 n measures the negative squared 4momentum transfer, Q 2 , in units of the neutron mass, M n .P e denotes the degree of longitudinal electron beam polarisation, and h = ±1 the electron helicity.For quasielastic scattering off a neutron bound in deuterium, Eqs. 1 are basically maintained [4] in the case of parallel kinematics, where the neutron is detected along the direction of the momentum transfer, ⃗ q.The ratio is linear in G E,n , and independent of the absolute value of P e .The aim of the experiment is to measure this ratio.
The overall setup of the experiment is sketched in Fig. 1.In order to tag the ex- clusive (⃗ e, e ′ ⃗ n) reaction, the scattered electrons and neutrons are detected in coincidence.Energy and angles of the scattered electrons are measured with high resolution in one magnetic spectrometer of the A1 setup [5].This guarantees clean separation of π production events.An overdetermined reconstruction of the D(⃗ e, e ′ ⃗ n)p kinematics is achieved through the additional measurement of the time-offlight of the neutrons and their hit position in a segmented scintillator, first wall in Fig, 1, some 6 m away from the target.The neutron polarisation can be analysed in the detector itself [6].This requires to detect the neutrons a second time in a split wall, second wall in Fig. 1, the two parts of which are positioned above and below the target (electron scattering plane).
The angular distribution of the scattered neutrons follows where, ε(Θ ′ n , Φ ′ n ) is the detection efficiency of the second wall scintillators, as a function of the polar and azimuthal angles of the first wall scattering.L h denotes the helicity specific luminosity and σ 0 the polarisationindependent scattering cross-section.The effective analysing power, A eff , is a priori unknown because different reaction channels contribute in the scintillator material.
Taking advantage of the fact that a reversal of the beam helicity h is equivalent to a shift of the azimuthal angle by 180 • , the asymmetry can be constructed in a way independent of L h , ε, and σ 0 .Here N ± (Φ ′ n ) is the count rate in the second wall of scintillators.
It has been demonstrated that the remaining problem of analysing power calibration of the polarimeter can be avoided by rotating the direction of the neutron spin in an external magnetic field [1].In case of a field perpendicular to the neutron trajectory, the precession angle is proportional to the integrated field strength, The same dipole magnet as in previous A3 [1] and A1 [2] experiments at MAMI is used.
Since the y-component of the magnetic field is absolutely dominant, the spins precess in the x-z plane.The resulting transverse neutron polarisation behind the magnet is given by the superposition P ⊥ (χ) = P x cos χ − P z sin χ =: P 0 sin(χ − χ 0 ) (6) with The transverse polarisation component, P ⊥ (χ), varies with the magnetic field strength, and thus the corresponding asymmetry A ⊥ (χ) in the second detector wall.For the particular case of vanishing transverse polarisation, P ⊥ (χ 0 ) = 0, after spin precession by the angle χ 0 , the neutron electric form factor can be experimentally determined from In the asymmetry ratio, both P e and A eff cancel out.Their absolute values affect only the statistical error of the determination of the zero crossing point χ 0 , but not its absolute value.
The electron beam polarisation has to be monitored, in order to detect possible changes with time.This is done with a Møller polarimeter [7].
Detector Configuration
The neutron polarimeter consists of a total of 287 scintillator bars (cf.Tab. 1) in four different sizes (cf.Tab.3), which are, as discussed in the previous section, arranged in two walls (cf.Fig. 1).The sequence of scintillator types is the same in both walls: first a layer of thin veto bars and then a matrix of thicker signal bars.The task of the veto bars is to detect charged particles entering the respective wall.Therefore, they are kept thin to minimise neutron interactions.Compared to the previous design [8,9,10,11] (cf.Tab. 2), the new polarimeter has a deeper first wall and a finer segmentation of both walls.This will provide a better position resolution and reduce the rates in individual scintillators.Except for the veto bars of the first wall, all bars were made of Eljen Technology 1 EJ-200 because of its long optical attenuation length and fast timing properties.For the veto bars in the first wall, EJ-212 was chosen due to its superior light yield, as this type of bar has a large length-to-thickness ratio and its signal is important to reduce charged background.The second wall veto bars are less important for background reduction and thus are made from EJ-200 for economical reasons.An overview of the different bar dimensions and used scintillator material is given in Tab. 3. Every bar is at both ends connected via a tapered light guide to a photomultiplier tube (PMT).The light guides are glued to the scintillator bars using optical cement2 .Each bar is then wrapped with a thin aluminised mylar film to enhance the light yield and covered with black foil to shield from ambient light.
The PMTs are coupled to the light guides using a silicon elastomer 3 to allow for an easy replacement.Special attention was paid to the coaxial alignment of the PMTs of the first wall, as the tolerances for mounting were tight.
For signal and veto bars of the first wall, the fast 10-stage PMT 9142SB from ET Enterprises 4 with a diameter of 29 mm with the C637ASN-05 passive voltage divider is chosen.All signal bars of the second wall are equipped with ET Enterprises 9822B PMTs with a diameter of 78 mm, together with the passive voltage divider C638SFN-01.The veto bars are read out with Hamamatsu5 R580-17 PMTs with the passive voltage divider E2183-500MOD1.
All PMTs are shielded with µ-metal tubes from the dipole magnet fringe field.
The 150 signal bars of the first wall are arranged vertically in a matrix of 25×6 bars (width × depth) with a veto bar in front of each of the 25 columns.Every second column is shifted back by half a scintillator thickness to allow for enough space for the PMTs.The entire first wall is composed of 13 special support structures or modules (see Fig. 2).12 of these modules contain two columns of six signal and one veto bar each and one module with only one such column.
The second wall is placed 2.5 m behind the first wall.The second wall consists of two frames (cf.Fig. 1) which are located one above the other with a gap of 80 cm around beam level.Each frame consists of 48 signal and eight veto bars arranged horizontally on a stair-like support with an incline of 18 • in eight staggered rows.
Readout
The neutron polarimeter has to determine the time-of-flight of the neutrons and provide a precise location for any interaction within the detector volume to reconstruct the scattering angle.The interaction point is reconstructed by recording the arrival time t 1,2 of the scintillation light on either end of a scintillator bar.The distance d from the centre of a bar is then given by with v e f f being the effective speed of light in a scintillator bar.Moreover, information on the deposited energy will help to distinguish different particle species.
The readout electronics for the neutron polarimeter thus must provide a precise timepickoff and amplitude measurement while coping with high rates (few MHz).Therefore, the readout system is based on the simplest time-pickoff method, leading-edge triggering, and time-over-threshold for amplitude determination which requires no additional delays in contrast to other methods.However, a very precise and fast TDC 6 is needed for such a 6 Time-to-Digital Converter concept to work.Especially, since plastic scintillator signals have a fast rise time and short width.
The frontend electronics is based on the NINO chip [12] developed by CERN.It provides eight input channels with very precise discriminators, a high-rate capability, and time-over-threshold encoding in the differential output signal.The NINO is operated in its single-ended mode.In addition, a programmable attenuator is placed before the NINO input to allow adjustments for each channel to avoid operation outwith the NINO specifications.The frontend board was developed in-house and incorporates four NINO chips to provide 32 input channels.It has two operating modes: a local mode where an FPGA 7 add-on board controls all settings and receives the NINO output signals, and a remote mode where all settings are controlled via an I 2 C connection from a computer.
The NINO output signals are then sent to multi-hit TDCs implemented in FPGAs on TRB3 boards [13,14].To avoid crosstalk by concurrent signals on adjacent lines, which could degrade time-pickoff, we use 3 m long Cat.7 network cables with individually shielded twisted pairs and an additional shielding 8 .
The TRB3 TDC has a typical bin width of just 15 ps and provides 48 channels.Each channel has a ring buffer for storing hit information until a global trigger arrives.Each TRB3 board contains four such TDCs and a central FPGA for slow control management and data transport.Furthermore, the TRB TDC allows to implement certain user-defined trigger conditions.The use of a time window with respect to the global trigger allows to select only the relevant hits from the inter-nal memory to reduce the required bandwidth.The data are sent from each TRB board in parallel to a readout PC, combined there and forwarded to the A1 DAQ system [15].
In total, seven TRB3 boards are used to readout the neutron polarimeter: six boards in TDC configuration and one board as main trigger and slow control unit.Of the six TDC boards four are used to read out the first wall and two for the second wall.
For the actual measurements we expect large interaction rates at least in the first layers of the first wall.Therefore, the rate capability was investigated in a dedicated test experiment.A fully instrumented scintillator bar for the first wall was irradiated with an electron beam (E beam = 855 MeV).The beam intensity was varied by changing the Wehnelt voltage U W of the electron source.The count rate was measured using a scaler implemented on a Zynq FPGA board 9 .Ten 1 s-intervals were measured and the average and standard deviation computed (s.Tab. 4).A reference scintillation counter was installed behind the bar off the beam axis thus reducing the counting rates in this counter to avoid dead time effects.Then the ratio between the scintillator bar and the reference counter was computed and normalized to 1 at the lowest intensity.The error of this normalized ratio was computed according to with R the raw ratio and ∆R its error, R 0 the raw ratio used for normalization and ∆R 0 its error.This measurement was then repeated using the TRB3 readout to exclude any negative impact on the rate capability by this readout 9 Details here... method.Here, the count rate could not be determined as precisely and the error was simply estimated as √ N (cf.Tab. 4).The normalized ratios for both Zynq and TRB readout are shown in Fig. 3 together with the estimated limit due to dead time effects when assuming a signal width of 45 ns to 65 ns.As can be seen, both readout method yield similar results close to the limit and thus are not introducing a bottleneck.
Trigger
The neutron polarimeter must provide a trigger signal to the A1 trigger system to select coincidences with the spectrometer detecting the scattered electrons.
A valid neutron polarimeter event must contain at least one signal in each wall.A signal is defined as a simultaneously detected hit in both PMTs attached to a scintillator bar.The TRB3 TDC provides the capability to generate trigger signals based on its input signals.
A list of channel pairs is written to the TDC configuration to form the basic coincidences at bar level.The coincidence time window is set to 40 ns.An OR is generated from all thus defined channel pairs and sent to the central FPGA on the TRB3 board.There, all four peripheral trigger signals are combined and sent out to the neutron polarimeter trigger logic.There, the signals of the first and second wall are combined first separately and then a coincidence of first and second wall triggers is formed.This final trigger signal is then forwarded to the global A1 trigger system.The A1 trigger signal is then returned to the TRB3 system together with an event number.The trigger signal is forwarded to all TRB3 TDCs and the event number is included in the data sent to the A1 DAQ system.
The veto scintillator information is deliberately omitted from the neutron polarimeter trigger decision to avoid event losses due to poorly calibrated veto bars.As the individual bar trigger rate (≈ 1 MHz) is well below the maximum sustainable rate (≈ 2.5 MHz) at the maximum beam current, this decision does not impact on the overall performance of the neutron polarimeter.
Performance
The new neutron polarimeter was installed at the A1 three-spectrometer setup at MAMI to measure the neutron's electric form factor at a momentum transfer of Q 2 = 0.7 (GeV/c) 2 in quasi-free scattering of polarized electrons with a beam energy E e = 855 MeV off a liquid deuterium target.
The initial calibration of the neutron polarimeter's scintillator bars was performed using cosmics and protons from a liquid hydrogen target and was used to adjust the high voltage of the PMTs to equalise the time-overthreshold spectra.When using the liquid hydrogen target, the elastically scattered electron was detected in coincidence by a spectrometer of the A1 setup.
As a next step, the time difference spectra need to be corrected for any offsets introduced due to different signal propagation delays.The nearly homogeneous hit distribution across the bars in all detector elements is shown in Fig. ference distribution is determined and the corresponding offset computed.An additional first wall signal bar was installed horizontally in front of the first wall at beam level to help with the position calibration.Requiring a hit in this additional bar allowed to validate the computed offsets for the first wall.
In addition to the offsets, the width of these time difference distributions allows the determination of v e f f for each type of bar.The height at which the width is measured is set to match a previous precise measurement of v e f f for a first wall scintillator bar (v e f f = (138.40± 0.28) mm ns ).The extracted values for v e f f are given in Tab. 5. Note that no value could extracted for the vetos of the second wall as their PMT response was not adjusted properly.Further steps in the calibration and analysis require information on the signal amplitude which is correlated to the time-overthreshold.A scintillator's average time-overthreshold ⟨T hit ⟩ is defined as the geometric mean of both signals from its PMTs as This additional information facilitates the separation of different particle types helping to suppress background contributions in the analysis.Fig. 5 shows ⟨T hit ⟩ of a veto bar of the first wall as a function of the time-of-flight.
There is a clear correlation seen corresponding to protons scattered from the liquid hydrogen target and a distinct band of uncorrelated background at lower time-over-threshold values.Furthermore, ⟨T hit ⟩ can be used to improve the time-pickoff as leading edge discrimination suffers from time-walk effects.To study these effects, the additional horizontal bar was used.The difference of the reconstructed hit position d and the geometrical position d ′ of the first wall bars behind hit too can be expressed in terms of the measured time-overthreshold of each PMT signal as with the slope parameter m, assuming that the time-over-threshold is proportional to the signal amplitude and that the signal shape can be approximated by a second-order polynomial up to the set threshold level.the signal bars of the first wall, this value for m is used to correct the time-pickoff for all signal bars in the first wall.However, such a measurement cannot not be done for other scintillator bars of the neutron polarimeter.
The position reconstruction in the first wall is cross checked using data with a liquid hydrogen target.Here, the initial direction and velocity of the proton can be calculated from the scattered electron kinematics.Propagating this track through the dipole magnet, a hit position on the first wall can be computed and compared to the reconstructed position (cf.Fig. 7) showing an excellent agreement.Most hits are within a circle with a radius of 45 mm around the predicted hit position, the tails caused by interactions with the target cell.The events within the circled area are then used to correct time-of-flight offsets for each bar in the first wall.For the second wall, a different approach is required to determine these offsets as there is, by design, no direct line-ofsight to the target.Selecting light relativistic particles, that are scattered within in the first wall, allow to measure the time of flight between the two walls and extract the offset parameters.
For the final extraction of the neutron electric form factor, two different scattering reactions in the neutron polarimeter have to be distinguished: either the incoming neutron is scattered in the first wall and then detected in the second (nn-channel) or the incoming neutron transfers in the first wall its momentum to a proton which is then detected in the second wall (np-channel).
The coarse separation of these two cases is based on the response of the veto bars and the hit multiplicity in each wall.However, it was found that the performance of the veto layers was insufficient and, therefore, the first layer of each wall was included in the veto decision.
For brevity, only the nn-channel is discussed below as the detection of protons is by far easier.The pre-selection of these events requires no hits in the veto and first signal bar layer of both walls.
This selection can be further improved due to the excellent performance of the neutron polarimeter.A clear correlation between the deposited energy ⟨T hit ⟩ and the incoming neutron's velocity β 01 is visible, see Fig. 8.The corresponding cut limits are indicated by the coloured lines.
The same correlation can be plotted for the second wall (s.Fig. 9) with the additional requirement of no hit in the last layer of the first wall to improve the pre-selection.The distinction is less clear than for the first wall but a full analysis reveals that the relevant neutron events are in the upper right corner.Only events between the diagonal lines and with β 01 above 0.6 are selected for nn-channel analysis.This improved selection of the relevant events in the neutron polarimeter allows to reduce the background, which would otherwise dilute the measured azimuthal asymmetry.
With all calibrations and selection cuts in place, the azimuthal asymmetry (cf.Eq. 4) can be extracted for different magnetic field strengths.The asymmetry for the maximum magnet current is shown in Fig. 10.
Conclusions
The newly built neutron polarimeter for A1 spectrometer setup at MAMI includes an improved design for the detector layers with a larger depth and higher segmentation.Furthermore, the readout electronics is designed to cope with high rates in individual channels.To facilitate this approach, time-pickoff is achieved by a leading edge discriminator nad the signal amplitude is measured by a time over threshold approach avoiding unnecessary dead times due to digitization.A TRB3 system provides the required high-precision fast TDC to enable this scheme.
This neutron polarimeter was used in a three week long beam time in 2019 to measure the neutron's electric form factor at Q 2 = 0.7 (GeV/c) 2 .The calibration results shown above demonstrate the excellent performance of the new detector system and corroborate the validity of the design decisions.
Figure 1 :
Figure 1: Sketch of Neutron Polarimeter setup with dipole magnet (not to scale).The scintillator (veto) bars in both walls are indicated in green (red).
Figure 2 :
Figure 2: Image of a first wall module installed in the support frame during assembly.
Figure 3 :
Figure 3: Test of the NDET readout rate capability.The normalized rate is shown as a function of the Wehnelt voltage (see text for explanations).The ratios are shown for both readout methods, Zynq and TRB3 board.The grey band indicates the expected ratio for signals with a Time-over-Threshold between 45 ns and 65 ns as observed during the experiment.
4 .
The left-and right-hand edge of the time dif-
Figure 4 :
Figure 4: Time difference spectra for a veto (a) and a signal bar (b) in the first wall and across a signal bar in the second wall (c).Offset corrections (s.text) have already been applied.
Figure 5 :
Figure 5: Average time-over-threshold (ToT) as function of Time-of-Flight (ToF) for a veto scintillator in the first wall using a hydrogen target.
Fig. 6
clearly shows the time-walk behaviour and allows to determine the slope parameter m = −119.16ns √ ns.After applying the corresponding corrections to the measured timepickoff, this correlation has vanished.Since the horizontal bar has the same properties as (a) (b)
Figure 6 :
Figure 6: Deviation of reconstructed from true position along a signal bar before (a) and after (b) walkcorrections.For details, please see text above.
Figure 7 :
Figure 7: Difference between predicted and reconstructed hit position for protons in the first wall.Data was taken with a liquid hydrogen target.
Figure 8 :
Figure 8: Correlation of time over threshold (ToT) with neutron velocity β 01 for nn-channel in the first wall.Only events between the diagonal lines and with β 01 above 0.6 are selected for nn-channel analysis.
Figure 9 :
Figure 9: Correlation second wall ToF and ToT.Correlation of time over threshold (ToT) with neutron velocity β 12 for nn-channel in the second wall.Only events in the upper right quarter are selected for the nn-channel analysis.
Figure 10 :
Figure 10: Final azimuthal asymmetry for full analysis with magnet currents of ±400 A.
Table 1 :
Overview of scintillator and PMT quantities for the new polarimeter design.
Table 3 :
Bar dimensions and scintillator material for all for types used in the neutron polarimeter.
Table 4 :
Raw rates measured with Zynq and TRB3 readout.For the TRB measurement no error is given due to small systematic offsets.
Table 5 :
Effective speed of light v e f f for different detector elements. | 5,741.4 | 2023-08-11T00:00:00.000 | [
"Physics"
] |
Simultaneous Production of Cellulose Nitrates and Bacterial Cellulose from Lignocellulose of Energy Crop
This study is focused on exploring the feasibility of simultaneously producing the two products, cellulose nitrates (CNs) and bacterial cellulose (BC), from Miscanthus × giganteus. The starting cellulose for them was isolated by successive treatments of the feedstock with HNO3 and NaOH solutions. The cellulose was subjected to enzymatic hydrolysis for 2, 8, and 24 h. The cellulose samples after the hydrolysis were distinct in structure from the starting sample (degree of polymerization (DP) 1770, degree of crystallinity (DC) 64%) and between each other (DP 1510–1760, DC 72–75%). The nitration showed that these samples and the starting cellulose could successfully be nitrated to furnish acetone-soluble CNs. Extending the hydrolysis time from 2 h to 24 h led to an enhanced yield of CNs from 116 to 131%, with the nitrogen content and the viscosity of the CN samples increasing from 11.35 to 11.83% and from 94 to 119 mPa·s, respectively. The SEM analysis demonstrated that CNs retained the fiber shape. The IR spectroscopy confirmed that the synthesized material was specifically CNs, as evidenced by the characteristic frequencies of 1657–1659, 1277, 832–833, 747, and 688–690 cm−1. Nutrient media derived from the hydrolyzates obtained in 8 h and 24 h were of good quality for the synthesis of BC, with yields of 11.1% and 9.6%, respectively. The BC samples had a reticulate structure made of interlaced microfibrils with 65 and 81 nm widths and DPs of 2100 and 2300, respectively. It is for the first time that such an approach for the simultaneous production of CNs and BC has been employed.
Introduction
Cellulose nitrates (CNs), being the initial product from the chemical functionalization of cellulose, have been the subject of active study for nearly 200 years [1].This is associated with the global abundance of natural cellulose and humankind's high need for mold plastics, quality lacquers, printing inks, biomolecular adhesive membranes, and energetic binder.The global market keeps growing, especially due to the demand for CNs as the platform for biosensors in analytical medicine [2].CNs have acquired particular importance in disease diagnostics and treatment due to their microporous structure and strong affinity for interacting with and subsequently absorbing a biomaterial (for example, antibodies) [2][3][4]; antiCOVID-19 masks [5] and composite filter membranes for oligonucleotide extraction [6,7] have emerged.The energetic properties of CNs are in demand as constituents of explosive compositions [8][9][10][11] in the mining industry, road construction in mountainous areas, and focused demolition of obsolete structures because the safety and handling issues associated with CN-based compositions have currently been resolved at a very high level [12].It should be emphasized that CNs themselves have become precursors of more complex chemicals with unique energetic characteristics [13].An overview of the published data on the demand for CNs in the industry allows for the conclusion that there is an increasing need for CNs with a nitrogen content ranging from 10.6% to 12.0%.
The cellulose nitration process itself, as an example of the chemical functionalization of the most naturally abundant, easily renewable biopolymer, is of particular importance for fundamental science [14].Cellulose is a naturally made polymer of β-glucose whose units are bound by a 1-4-β-glycosidic linkage and is produced in the amount of about 1.3 × 10 9 tons a year via photosynthesis [15].In the cellulose nitration process, the hydrogen atom of the (-H-OH) hydroxyl is replaced by the (-NO 2 ) nitro group when cellulose is nitrated by nitrating mixtures.It is possible to synthesize CNs with a wide range of functional properties by varying the nitrating mixture composition and nitration process parameters [1,15,16].
Many researchers compare the properties of alternative cellulose with the requirements for cotton cellulose: the α-cellulose content must be no less than 92%.The above-listed studies on nitration differ in their focus: some researchers only reflect nitration process conditions and specify one of the properties of the resultant nitrocellulose, basically the CN degree of substitution [10,20], while others report extensive information including basic functional, supramolecular, morphological, and energetic characteristics of the biopolymers synthesized from a specific feedstock type [9,19,21,22].
The possibility of producing a new CN type from bacterial cellulose (BC) has been justified alongside the use of non-woody cellulose, and prospects for using this new type have been considered [10,[27][28][29].
BC has a chemical structure similar to plant-based cellulose but does not contain hemicelluloses, lignin, or pectin.BC microfibrils form a three-dimensional network, ensuring high values of mechanical strength, degree of polymerization, crystallinity, and waterholding capacity [30,31].The structure and those properties allow BC to have numerous potential technical applications: environmentally friendly electronic devices [31], flexible organic light-emitting diodes, fuel cells, flexible supercapacitors, headphones, monitors, and materials for electromagnetic wave absorption [32].
All the methods for cellulose production from non-woody sources, including Miscanthus, pursue the aim of conforming to the purity of cotton cellulose: the α-cellulose content no less than 92%, the pentosan content no more than 2%, and minimal lignin; therefore, there is no information on enzymatic hydrolysis as the pretreatment method for a cellulosic product to modify its structural characteristics with subsequent nitration.Even more so, there are no examples of utilizing cellulose to directly produce CNs and concurrently as the source of a glucose nutrient medium for subsequent biosynthesis.
The present study aimed to explore whether the two standalone products, CNs (a cellulose chemical modification product) and BC (an enzymatic hydrolysis product), could concurrently be produced from the energy crop Miscanthus × giganteus.
Materials and Methods
All the reagents and materials used in this study were procured from AO Vekton (Saint-Petersburg, Russia).
Feedstock
In this study, Miscanthus × giganteus was used as the feedstock, having the following chemical compositions: 50.2 wt.% Kürschner cellulose, 19.5 wt.% acid-soluble lignin, 21.2 wt.% pentosans, 1.63 wt.% ash, and 0.5 wt.% extractives [53].The quantitative determination methods for the feedstock components were similar to those used for cellulose components (Section 2.2.1), except for the quantitative determination of Kürschner cellulose and extractives.Kürschner cellulose was determined by the extraction of the Miscanthus sample with 1:4 mixed nitric acid-alcohol [62,63].Extractives were determined by extracting the sample in dichloromethane using a Soxhlet extractor according to the Technical Association of the Pulp and Paper Industry (TAPPI) standard [64].
Preparation and Analysis of Cellulose Samples
Since the grinding size plays a decisive role in cellulose isolation [65,66], Miscanthus × giganteus was ground on a KR-02 fodder grinder (TechnoMash, Miass City, Russia) to a particle size of 2-12 mm prior to use.
Cellulose was isolated by the nitric-acid method involving treatment of the weighed portion of the feedstock with dilute solutions of nitric acid (3-6%) and sodium hydroxide (3-6%) under atmospheric pressure at 90-95
Analysis of Chemical Composition and Cellulose Degree of Polymerization (DP)
The chemical composition (contents of α-cellulose, lignin, ash, and pentosans) and cellulose DP were analyzed by standard chemical and physicochemical methods.The α-cellulose content of the cellulose sample was determined as per the TAPPI standard by treating cellulose with a 17.5 wt.% NaOH solution, followed by the quantification of the undissolved residue after cellulose was washed with a 9.5 wt.% NaOH solution and water and dried [67].Klason lignin (acid-insoluble lignin) was measured as per TAPPI T222 om-83 [68].Pentosans were quantified by transforming the same in a boiling 13 wt.%HCl solution into furfural, which was collected in the distillate and determined on a xylose-calibrated UNICO UV-2804 spectrophotometer (United Products & Instruments, Dayton, NJ, USA) calibrated against xylose (a 630-nm wavelength) using orcinol-ferric chloride [69].The ash content was quantified by cellulose incineration in accordance with TAPPI T211 om-85 [70].The cellulose DP was determined from the outflow time of cellulose solution in cadoxene (cadmium oxide in ethylenediamine) in a VPZh-3 viscometer (OOO Ecroskhim, Moscow, Russia) with a capillary diameter of 0.92 mm [71].
X-ray Diffraction Analysis of Cellulose Samples
X-ray examination of the cellulose sample was performed on a DRON-6 monochromatic diffractometer (Burevestnik company, Nalchik City, Russia) with Fe-Kα radiation at 3 to 145 • scattering angles in reflection and transmission geometries at room temperature [43,72,73].
The degree of crystallinity (DC) was defined as the relation between the integrated scattering intensity from the crystalline phase and the total integrated scattering intensity from the crystalline and amorphous phases in reflection geometry (Equation ( 1)): where I c is the total integrated scattering intensity from the crystalline and amorphous components and I am is the integrated scattering intensity from the amorphous component [43,72].
Enzymatic Hydrolysis of Cellulose Samples
Enzymatic hydrolysis of the cellulose sample was performed with an enzyme cocktail of Ultraflo Core (Novozymes A/S, Bagsvaerd, Denmark) and CelloLux-A (Sibbiopharm Ltd., Berdsk, Russia) at a dosage as follows: Ultraflo Core 46 FPU/g solid and CelloLux-A 40 FPU/g solid.The cellulase activity expressed in FPU was determined by the reported procedure [74].
The enzymatic hydrolysis was carried out in a 0.05 M acetate buffer (pH 4.7): a 45.0 g/L initial solid loading on a dry matter basis, a 0.1 L acetate buffer volume, 46 ± 2 • C temperature, and a 150-rpm stirring rate.The first stage of hydrolysis included the measurement of the cellulose DP during the process.For this, the enzymatic hydrolysis was run in a parallel manner in seven 0.5-L conical flasks, with a process time of 2, 4, 6, 8, 24, 32, and 48 h.The stirring was carried out using an ECROS PE-6410 horizontal heated stirrer (Ecohim, Moscow, Russia).Once the time elapsed, the flask was removed from the stirring device, and the reaction mixture was cooled and filtered.The concentration of reducing sugars (RS) in the hydrolyzate was measured on a Cary 60 UV-Vis (Agilent Technologies, Santa Clara, USA) at a 530-nm wavelength using 3,5-dinitrosalicylic acid (Panreac, Spain) as the reagent [69,75].The RS yield was estimated by Equation (2) [74]: where ηRS is the yield of RS on a substrate weight basis (%).C RS is the final concentration of RS in the hydrolyzate (g/L).C S is the substrate concentration on a dry matter basis (g/L).0.9 is the factor associated with the water molecule addition to anhydroglucose residues of the respective monomeric units as a result of hydrolysis.
The solid residue after filtration was thoroughly washed, dried, and weighed to calculate the weight loss.The sample was then analyzed for the cellulose DP (Section 2.2.1) and DC (Section 2.2.2).
The second hydrolysis stage involved working up cellulose samples for nitration and preparing nutrient media (hydrolyzates) for BC synthesis.The process was performed in conical flasks under the same conditions.The initial solid loading was 45 g/L, with the reaction mass volume increasing.The weight of the substrate to be hydrolyzed was calculated with allowance for weight loss at the corresponding hydrolysis time and minimal cellulose weight for nitration, and was 3 g.Upon the process's completion, the resultant reaction mass was filtered, and the solid residue was washed and dried.The solid residue was further analyzed for cellulose and then nitrated.The liquid phase (hydrolyzate) was used for BC synthesis.
To achieve accurate results, three samples were enzymatically hydrolyzed at a time in each experiment.
Cellulose Sample Analysis after Enzymatic Hydrolysis
The DP, DC, and morphology of the cellulose samples after enzymatic hydrolysis were determined by the same methods as those used for the initial Miscanthus cellulose sample (Section 2.2).
Nitration and Analysis of CN Samples
The cellulose nitrate samples were obtained by the common sulfuric-nitric acid process using a commercial sulfuric-nitric acid mixture.The cellulose samples were nitrated as follows: The initial water content in the mixture was 14 wt.%, the nitration temperature was 25-30 • C, the nitration time was 40 min, and the mass ratio of substrate to mixed acid was 1:40.The nitration was performed in a 500-mL porcelain beaker with continuous stirring using a HS-50A-Set vertical stirring device (Witeg, South Korea).The nitration temperature was maintained using a water bath.
After the nitration was completed, the resultant CN samples were separated from the spent mixed acid by using a Büchner funnel and a vacuum pump, then the remaining reaction mixture was expelled with a dilute 25 wt.% mixed acid and further thoroughly washed and subjected to three-step high-temperature stabilization with continuous stirring as follows: boiling in water at 80-90 • C for 1 h, boiling in a 0.03% sodium carbonate solution at 80-90 • C for 3 h, and boiling in water at 80-90 • C for 1 h.After the stabilization process was completed, the target products were washed with distilled water until neutral wash waters and then dried for 24 h in open air at room temperature and then at 100 ± 5 • C for 1 h in a BINDER ED23 drying oven (BINDER GmbH, Tuttlingen, Germany) and analyzed.
The CNs were analyzed using common procedures.The nitrogen content was quantified by the ferrous sulfate method [76][77][78] that involves the saponification of CN with concentrated sulfuric acid and the reduction of the resultant nitric acid with iron (II) sulfate to nitrogen (II) oxide, in which nitric acid in excess of nitrogen (II) oxide produces a complex compound, Fe(NO)]SO 4 , that colors the solution into a yellowish-pink color.The solubility of CN (1 g) in acetone (50 mL) was determined by filtration of the acetone-insoluble CN residue, followed by drying and weighing on an Explorer Pro EP214C analytical balance (Ohaus, Langacher, Switzerland).The viscosity of the CN samples was determined from the outflow time of a 2% acetone solution from a VPZh-3 capillary glass viscometer.The solubility of the CN samples in mixed alcohol/ester was determined by filtration of the CN residue insoluble in mixed alcohol/ester, followed by drying and weighing on an Explorer Pro EP214C analytical balance.
The yield of the CN samples was calculated by Equation ( 3): where m pr is the weight of the synthesized CN samples, g; and m init is the weight of the initial cellulose sample for nitration, g.
Structural Analysis of Cellulose, CN, and Coupled TGA/DTA
The fiber surface morphology of the cellulose and CN samples was examined by scanning electron microscopy (SEM) on a GSM-840 electron microscope (Jeol, Tokyo, Japan) after sputter-coating a Pt layer of 1−5 nm thick.
The molecular structure of the cellulose and CN samples was examined by Fouriertransform spectroscopy on an Infralum FT-801 spectrometer (OOO NPF Lumex-Sibir, Novosibirsk, Russia) operating at 4000−500 cm −1 .To acquire spectra, the samples were pressed into pellets with potassium bromide in a CN:KBr ratio of 1:150.
The thermal behavior of the cellulose and CN samples was examined by thermogravimetric (TGA) and differential thermogravimetric (DTG) analyses using a TGA/DTG-60 thermal analyzer (Shimadzu, Nakagyo-ku, Japan) as follows: a weighed portion of 0.5 g, a heating rate of 10 • C/min, a maximal temperature of 350 • C, and nitrogen as inert medium.
Synthesis of Bacterial Cellulose
Symbiotic Medusomyces gisevii Sa-12, acquired from the Russian National Collection of Industrial Microorganisms, was used as the microbial producer.The vital activity of Medusomyces gisevii Sa-12 was maintained in a Binder-400 climate chamber (Berlin, Germany) under static conditions at 27 • C for 7 days in a synthetic glucose medium composed of glucose and black tea extractives [36,79].The seed material was inoculated as 10 vol.% of the nutrient medium volume, which is equivalent to the following cell count: the total yeast count of at least 12.9-13.2× 10 6 cells per 1 cm 3 and the total acetobacteria count of at least 1.6-2.2× 10 6 per 1 cm 3 .
The biosynthesis of BC was conducted on the enzymatic hydrolyzate under static culture conditions at a temperature of 27 • C, with an initial glucose concentration of 20 g/L and a black tea extract content of 1.6 g/L.The cultivation was carried out in a climate chamber (Binder, Germany) for 10 days.
After the cultivation was completed, the BC gel-film was removed from the surface of the nutrient medium and washed to remove the nutrient medium components and cells through a stepwise treatment with 2 wt.%NaOH and 0.25 wt.% HCl, followed by washing with distilled water until neutral wash waters.The obtained BC films were freeze-dried in an HR7000-M freeze dryer (Harvest Right LLC, Salt Lake City, UT, USA) to constant weight.
The yield of the dried BC was calculated by Equation ( 4): where m is the weight of the BC sample on an oven-dry basis, g.C is the RS concentration in the medium on a glucose basis, g/L.
V is the volume of the medium, L. 0.9 is the conversion factor due to the water molecule detachment upon the polymerization of glucose into cellulose.
The yeast and acetobacteria cell counts, as well as the concentration of reducing sugars in the nutrient medium after removing the BC film, were measured as described in [38].
As a control, BC synthesis was conducted on a synthetic nutrient medium with a glucose concentration of 20 g/L and an extractive content of 1.6 g/L under similar conditions.
The morphology of the BC samples was investigated using a scanning electron microscope (JSM-840, Tokyo, Japan) equipped with a Link-860 series II X-ray microanalyzer.The microfibril width was calculated using the ImageJ 1.53k software.The DP of BC was determined according to the procedure described in Section 2.2.1.
Properties of Cellulose Sample
In the nitric acid method for cellulose production, the preliminary hydrolysis stage involves breaking the bonds between the main components of the lignocellulosic matrix and partially removing hemicelluloses.The subsequent treatment with a diluted nitric acid solution allows for the almost complete removal of hemicelluloses (2.5%), partial dissolution, oxidation, and nitration of lignin, leading to the formation of nitrolignin.Further alkaline treatment solubilizes the nitrolignin and removes it from the product (0.5%).This cellulose production method allows a high-quality product to be isolated with a cellulose content of 95.6%, a pentosan content of 2%, and total non-hydrolyzables (ash and lignin) of only 0.6%.The resultant sample exhibited a high DP of 1770 and a DC of 64%.
The quality indicators of nitric acid-treated cellulose from Miscanthus × giganteus align with those of cellulose samples from other Miscanthus species obtained using the same method, except for their lower DP (880-1050) compared to the latter [26].
The absence of data on the quality indicators of cellulose extracted from Miscanthus × giganteus using nitric acid or other methods that would allow for the production of high-quality cellulose is due to Miscanthus × giganteus being currently in active use for alternative purposes where highquality cellulose is not required, such as chemical modification.For instance, Miscanthus × giganteus is used in the paper industry, where the requirements for cellulose are considerably lower, with the α-cellulose content not exceeding 86% [54,56].
At the same time, there is significant ongoing research into the nitration of alternative cellulosic raw materials.The quality indicators of cellulose derived from Miscanthus × giganteus closely approach or even surpass those of cellulose samples obtained from acacia pulp [80], Rhizophora, oil palm bunches, and kenaf fibers [18], which have undergone successful nitration.
Enzymatic Hydrolysis of Cellulose
The first stage of the enzymatic hydrolysis involved investigating the change in the cellulose DP and DC during a 48 h enzymatic hydrolysis.Additionally, the RS concentration increment and the weight loss were evaluated (Table 1).Conducting the hydrolysis for more than 48 h was not reasonable due to the significant weight loss (over 66%) and, consequently, the small weight of the cellulose residue after the hydrolysis, which residue would further be used for nitration.Throughout the enzymatic hydrolysis process of Miscanthus cellulose samples, there was a gradual increase in the weight loss, indicating the hydrolysis of the substrate and a reduction in the weight of the solid residue required for subsequent nitration.In the initial 2 h of hydrolysis, the cellulose DP decreased from 1770 to 1490, while the DC increased from 64 to 72%.The changes in the cellulose properties may be attributed to the random cleavage of β-1,4-glucosidic bonds by endoglucanase, occurring in the less organized regions of cellulose, leading to a decrease in DP and an increase in crystallinity [81][82][83].
In the subsequent hours of hydrolysis, the DP started to increase and reached its maximum value after 48 h of hydrolysis.As a result of the experiment, the cellulose DP did not almost change: 1790 after hydrolysis vs. 1770 for the initial cellulose.The DC increased by 12% after a 48 h hydrolysis.According to [59,84], significant structural changes are not typical of the enzymatic hydrolysis of cellulosic materials.It was emphasized that there was no substantial decrease in DP during the hydrolysis process, whereas DC may slightly increase.It was explained by the fact that the cellulase complex attacks the cellulose chains and hydrolyzes each chain to the end.As a result, neither the DP nor the ratio of the crystalline to the amorphous material changes significantly.
During the enzymatic hydrolysis process, the concentration of reducing sugars (RS) reached 30 g/L, corresponding to a 60% yield of RS.That said, reducing sugars were generated at 43% of the maximal yield as early as the initial 2 h of the process, after which the hydrolysis rate was slowing down.
The obtained results (Table 1) were used to determine the hydrolysis length in the second stage of the experiment for obtaining cellulose samples for nitration and nutrient media for BC biosynthesis.At this stage, the substrate weight for hydrolysis was calculated based on the weight loss values for the specific hydrolysis duration and the cellulose weight required for nitration (minimum 3 g).
The hydrolysis of the Miscanthus cellulose samples was conducted for 2, 8, and 24 h.The time point at 2 h was chosen because the maximal reduction in the DP occurred within that time, despite the RS concentration in the hydrolyzate not reaching the required value for BC synthesis (20 g/L).The 8 h and 24 h time points were chosen because the structural characteristics of cellulose underwent changes, with the RS concentration being above 20 g/L in the hydrolyzates.The results of the second stage of the experiment are presented in Table 2.As a result of the enzymatic hydrolysis, three cellulose samples (C2, C8, and C24) were obtained, ranging in mass from 3.0 to 3.4 g.These samples differed in their characteristics from the initial Miscanthus cellulose sample (DP 1770, DC 64%) and differed from each other (DP ranging from 1510 to 1760, DC ranging from 72% to 75%) and were of interest for the subsequent nitration.Due to the lack of information on similar experiments with Miscanthus cellulose, it is challenging to compare the observed changes in DP and DC of the cellulose residues after enzymatic hydrolysis.During the enzymatic hydrolysis, enzymatic hydrolyzates (hydrolysates C2, C8, and C24) differing in RS concentration (ranging from 13.4 to 27.5 g/L) were also obtained and investigated as a nutrient medium for BC biosynthesis.A brief diagram of the experiment is given in the Supplementary Materials.
Nitration
Given the high requirements for cellulose used in the chemical conversion (minimal contents of lignin, hemicellulose, ash content, and other side inclusions [14], the results (Section 3.1) obtained regarding the compositional analysis of cellulose from Miscanthus × giganteus (C) do not exclude the possibility of its successful chemical modification into cellulose nitrate (CN) with satisfactory functional properties.Table 3 presents the key functional properties of the CN samples obtained from Miscanthus cellulose before and after enzymatic hydrolysis.
It also follows from Table 3 that an increase in the duration of enzymatic hydrolysis from 2 h to 24 h resulted in a rise in the nitrogen content of the CN samples from 11.35% to 11.83%, an increase in the viscosity from 94 mPa•s to 119 mPa•s, and an elevation in the yield from 116% to 131%.The increase in the nitrogen content and, consequently, the CN yield might be due to the enhanced reactivity of cellulose as a result of the multiple fragmentation of its units by the enzymes.It should also be emphasized that the viscosity of CN was changing consistently with a change in the DP of the initial cellulose samples, depending on the enzymatic hydrolysis length.Furthermore, it is important to note that regardless of the hydrolysis duration, all the synthesized samples were CN esters, as they had a 100% solubility in acetone [14].It can be concluded from the obtained data listed in Table 3 that the CN sample synthesized from cellulose and subjected to 24 h hydrolysis exhibited satisfactory functional properties: a nitrogen content of 11.83%, a viscosity of 119 mPa•s [14], as well as a high degree of homogeneity, as its solubility was 94% in an alcohol-ester mixture compared to the other CN samples after enzymatic hydrolysis and compared to the CN sample derived from the initial cellulose.
The obtained results have no global analogues since there is no information available on the production of CN based on cellulose after enzymatic hydrolysis.However, CN derived from Miscanthus cellulose after 24 h enzymatic hydrolysis showed similar characteristics to CN derived from another Miscanthus species (11.85% nitrogen content and 97% solubility) [28], except for a significantly lower viscosity (18 mPa•s), which is attributed to the initially lower DP of the cellulose (1020).
In addition, the characteristics of the synthesized CN samples in this study align with those of the CN derived from cellulose from acacia pulp [85], rhizophora, palm oil bunches, and kenaf fibers [18], tobacco stems [20], and oat hulls [25].
Figure 1 shows microphotographs (×200 and ×5000 zoom) of Miscanthus cellulose samples before and after enzymatic hydrolysis, as well as the CNs synthesized based on them.
The scanning electron microscopy (SEM) analysis showed that the cellulose sample (Figure 1a), extracted from Miscanthus × giganteus and not subjected to enzymatic hydrolysis, consists mainly of heterogeneous cellulose fibers with varying shapes and sizes, resembling tubes.Besides, the overall mixture contains individual flattened, wide fibers.The surface of the cellulose fibers exhibits micro-roughness.With an increase in the duration of enzymatic hydrolysis from 2 h to 24 h (Figure 1c,e,g), the microphotographs reveal that the cellulose fibers become shorter and the edges become jagged.With a higher magnification (Figure S2 in Supplementary Materials), irregular-shaped pores appear on the fiber surface, and the number of pores on cellulose fibers increases.
After treating the cellulose samples from Miscanthus × giganteus with a sulfuric-nitric acid mixture, the CN fibers primarily retained the shape of the original cellulose fibers while increasing in volume.The surface of the cellulose-nitrate fibers became smoother.The CN fibers of the sample derived from the original Miscanthus × giganteus cellulose (Figure 1b) represented separate tube-like fibers, in contrast to the CN samples based on cellulose after the 2-24 h enzymatic hydrolysis (Figure 1d,f,h), which consisted of a mixture of fibers varying in size and shape.The scanning electron microscopy (SEM) analysis showed that the cellulose sample (Figure 1a), extracted from Miscanthus × giganteus and not subjected to enzymatic hydrolysis, consists mainly of heterogeneous cellulose fibers with varying shapes and sizes, resembling tubes.Besides, the overall mixture contains individual flattened, wide fibers.The surface of the cellulose fibers exhibits micro-roughness.With an increase in the duration of enzymatic hydrolysis from 2 h to 24 h (Figure 1c,e,g), the microphotographs reveal that the cellulose fibers become shorter and the edges become jagged.With a higher magnification (Figure S2 in Supplementary Materials), irregular-shaped pores appear on the fiber surface, and the number of pores on cellulose fibers increases.
After treating the cellulose samples from Miscanthus × giganteus with a sulfuric-nitric acid mixture, the CN fibers primarily retained the shape of the original cellulose fibers while increasing in volume.The surface of the cellulose-nitrate fibers became smoother.The CN fibers of the sample derived from the original Miscanthus × giganteus cellulose (Figure 1b) represented separate tube-like fibers, in contrast to the CN samples based on cellulose after the 2-24 h enzymatic hydrolysis (Figure 1d,f,h), which consisted of a mixture of fibers varying in size and shape.
Figure 2 presents the Fourier-transform infrared spectroscopy results for the cellulose and CN samples.According to Figure 2a, the FTIR spectra of the original cellulose samples exhibit the main functional groups characteristic of cellulose [9,86], namely 3341-3363 cm −1 , 2898-1901 cm −1 , 1428-1430 cm −1 , 1158-1163 cm −1 , and 1059-1060 cm −1 , which are assigned to the O-H stretching, asymmetric and symmetric stretching of C-H, O-H bending of absorbed water, asymmetric bending vibration of CH2, C-O-C stretching, skeletal stretch of C-O, and vibration of the β-glycosidic linkage of cellulose, respectively .The FTIR spectra showed that the cellulose samples did not exhibit peaks corresponding to the stretch vibrations responsible for impurity components like aromatic structures of lignin at around 1500 cm −1 and hemicelluloses at around 1700 cm −1 , proving once again that the cellulose extracted from Miscanthus × giganteus was high-quality.
The FTIR spectra of the CN samples (Figure 2b) exhibit the main functional groups that indicate the formation of low-substituted nitrocellulose ethers (1657-1659 cm −1 , 1277 cm −1 , 832-833 cm −1 , 747 cm −1 , 688-690 cm −1 ).The intense absorption bands in the range of 1657-1659 cm −1 correspond to the vibrations of νa(NO2) nitrate groups, which are associ- According to Figure 2a, the FTIR spectra of the original cellulose samples exhibit the main functional groups characteristic of cellulose [9,86], namely 3341-3363 cm −1 , 2898-1901 cm −1 , 1428-1430 cm −1 , 1158-1163 cm −1 , and 1059-1060 cm −1 , which are assigned to the O-H stretching, asymmetric and symmetric stretching of C-H, O-H bending of absorbed water, asymmetric bending vibration of CH 2 , C-O-C stretching, skeletal stretch of C-O, and vibration of the β-glycosidic linkage of cellulose, respectively .The FTIR spectra showed that the cellulose samples did not exhibit peaks corresponding to the stretch vibrations responsible for impurity components like aromatic structures of lignin at around 1500 cm −1 and hemicelluloses at around 1700 cm −1 , proving once again that the cellulose extracted from Miscanthus × giganteus was high-quality.
The FTIR spectra of the CN samples (Figure 2b) exhibit the main functional groups that indicate the formation of low-substituted nitrocellulose ethers (1657-1659 cm −1 , 1277 cm −1 , 832-833 cm −1 , 747 cm −1 , 688-690 cm −1 ).The intense absorption bands in the range of 1657-1659 cm −1 correspond to the vibrations of ν a (NO 2 ) nitrate groups, which are associated with the CH 2 groups of the glucopyranose rings in the CN (position C( 6)).The intense absorption bands at 1277 cm −1 can be attributed to the stretching symmetric vibrations of nitrate groups.The absorption bands in the ranges of 832-833 cm −1 , 747 cm −1 , and 688-690 cm −1 correspond to the vibrations of nitrate groups: stretching ν a (NO 2 ), wagging γ w (NO 2 ), and scissoring δ(NO 2 ) vibrations, respectively.In addition to the main absorption bands associated with the stretching vibrations of nitrate groups, there are peaks of stretching vibrations of ν(OH) in the range of 3200-3700 cm −1 , appearing as a broad, complex contour.This indicates the incomplete substitution of the CN.The peaks of stretch vibrations in this region belong to the hydroxyl groups of the CN, which participate in hydrogen bonding and are a characteristic feature of the chemical heterogeneity of the ester.Identical functional groups are observed in the FTIR spectra of CN derived from other alternative plant-based raw materials [9,22,25,87].
Figure 3 shows the TGA/DTG thermograms of the original cellulose samples and their CNs.
According to Figure 3a, the obtained TGA curves for cellulose samples before and after the enazymati hydrolysis can be divided into three distinct regions.The first region encompasses the temperature range from the beginning of the experiment to 100 • C, during which the samples undergo drying, exhibiting a weight loss of 0.2-0.8%accompanied by an endothermic peak.The second region extends from 100 • C to 400 • C, where the samples undergo decomposition with a weight loss of 88.6-90.7% and an associated endothermic transformation.The third region spans the range from 400 • C to 450 • C, where the samples continue to decompose with a minor weight loss of 1.4-1.7%.The temperature range for the onset of intensive sample decomposition was determined to be 339-345 • C.
From the analysis of the literature data, it is well known that higher initial decomposition temperatures correspond to higher thermal stability and purity of the original cellulose [14].The DTA curves of the cellulose samples (Figure 3c) showed that the decomposition endothermic peak corresponds to a temperature range from 357 • C to 371 • C, with a weight loss of the samples up to 88.6-90.7%,confirming their purity.These results are consistent with the findings of a study on cellulose derived from bitter bamboo stems [87], which showed the superior thermal stability of cellulose from Giant Miscanthus.
In the case of the CNs obtained from the cellulose samples before and after enzymatic hydrolysis, as determined by TGA (Figure 3b), it was found that regardless of the hydrolysis duration, the decomposition peak of the CN samples occurred at a temperature around 198-199 • C, and the decomposition continued up to a temperature of approximately 260 • C, with a weight loss of the samples ranging from 70.1 to 82.8%.Further decomposition of the samples occurred with a minor weight loss in the range of 6.9-9.5%.
The DTG curves obtained (Figure 3d) illustrate a single narrow exothermic peak at a temperature around 198-199 • C. Comparing the DTG curves of the CN samples with the curves of the original cellulose samples, it is evident that the temperatures of the exothermic peak in the CN samples decrease from 357-371 • C to 198-199 • C.This destructive behavior is associated with the thermolytic cleavage of the weakest O-NO 2 group, initiating autocatalytic decomposition and leading to the formation of reactive radicals that accelerate the thermal decomposition process of the nitrated polymer chains [9].The above findings indicate that the obtained CN samples are chemically pure, high-energy biopolymers.Comparing the obtained TGA/DTA data for the CN synthesized from Giant Miscanthus cellulose with those for the CNs derived from cotton [87,88], giant reed [9,19], brown algae [22], and bitter bamboo stems [21], indicates their close correspondence.Furthermore, it is demonstrated that all CN samples exhibit high specific decomposition heats ranging from 6.53-8.28kJ/g.Thus, the CN samples obtained from cellulose subjected to enzymatic hydrolysis are low-substituted nitric esters of cellulose with satisfactory functional properties and energetic characteristics.Overall, the synthesized CN samples exhibit properties that indicate the suitability of cellulose samples after enzymatic hydrolysis for chemical functionalization into complex cellulose ethers.It is important to emphasize that this approach to obtaining CN is being used for the first time in global practice.According to Figure 3a, the obtained TGA curves for cellulose samples before and after the enazymati hydrolysis can be divided into three distinct regions.The first region encompasses the temperature range from the beginning of the experiment to 100 °C, during which the samples undergo drying, exhibiting a weight loss of 0.2-0.8%accompanied by an endothermic peak.The second region extends from 100 °C to 400 °C, where the samples undergo decomposition with a weight loss of 88.6-90.7% and an associated endothermic transformation.The third region spans the range from 400 °C to 450 °C, where the samples continue to decompose with a minor weight loss of 1.4-1.7%.The temperature range for the onset of intensive sample decomposition was determined to be 339-345 °C.
From the analysis of the literature data, it is well known that higher initial decomposition temperatures correspond to higher thermal stability and purity of the original cellulose [14].The DTA curves of the cellulose samples (Figure 3c) showed that the decomposition endothermic peak corresponds to a temperature range from 357 °C to 371 °C, with a weight loss of the samples up to 88.6-90.7%,confirming their purity.These results are consistent with the findings of a study on cellulose derived from bitter bamboo stems (a,c) original cellulose (C 0 ), cellulose after 2 h hydrolysis (C2), cellulose after 8 h hydrolysis (C8), cellulose after 24 h hydrolysis (C24); and their cellulose nitrates (b,d): cellulose nitrate from the original cellulose (CN 0 ), cellulose nitrate from cellulose after 2 h hydrolysis (CN2), cellulose nitrate from cellulose after 8 h hydrolysis (CN8), and cellulose nitrate from cellulose after 24 h hydrolysis (CN24).
Synthesis of Bacterial Cellulose
Biosynthesis of BC was conducted on enzymatic hydrolyzates obtained after 2, 8, and 24 h.Enzymatic hydrolyzates C8 and C24 with RS concentrations of 22.8 g/L and 27.5 g/L, respectively, were adjusted to a concentration of 20 g/L through dilution.Enzymatic hydrolyzate C2 with a RS concentration of 13.4 g/L was also used for BC biosynthesis.The results of enzymatic hydrolysis are presented in Figure 4.It can be observed from the presented data that the count of yeast at the end of the biosynthesis process exceeds that of acetobacteria in all cases.This can be attributed to the fact that the utilized producer is a consortium of various yeast and acetobacteria species and genera.According to literature data, yeast synthesizes ethanol to stimulate the growth of acetobacteria, which, in turn, produce BC to protect the yeast from the surrounding environment [89,90].Figure 4 indicates that in the synthetic nutrient medium (control) and in the C8 and C24 hydrolyzates, the count of acetobacteria remains relatively constant, ranging from 8-10 million CFU/mL.A low count of acetobacteria, 1 million CFU/mL, is observed in the nutrient medium of hydrolyzate C2.The low count can be attributed to the RS concentration of 13.4 g/L, which is insufficient for the active growth and vitality of acetobacteria.The low count of acetobacteria resulted in the absence of BC biosynthesis, which, in turn, explains the lack of BC gel film in the nutrient medium of hydrolyzate C2.
Figure 4b shows the residual RS concentration in the culture medium after 10 days of cultivation.The RS concentration after 10 days of cultivation in the synthetic nutrient media (control) was less than 4 g/L, while in the nutrient media of the enzymatic hydrolyzates, it ranged from 8 to 10 g/L.The slight decrease in RS concentration in the nutrient medium of hydrolyzate C2 during the biosynthesis process, from 13.4 g/L to 10 g/L, indicates the absence of active vitality in the acetobacteria responsible for BC production.The high residual RS concentration in nutrient medium C2 compared to C24 is explained by It can be observed from the presented data that the count of yeast at the end of the biosynthesis process exceeds that of acetobacteria in all cases.This can be attributed to the fact that the utilized producer is a consortium of various yeast and acetobacteria species and genera.According to literature data, yeast synthesizes ethanol to stimulate the growth of acetobacteria, which, in turn, produce BC to protect the yeast from the surrounding environment [89,90].Figure 4 indicates that in the synthetic nutrient medium (control) and in the C8 and C24 hydrolyzates, the count of acetobacteria remains relatively constant, ranging from 8-10 million CFU/mL.A low count of acetobacteria, 1 million CFU/mL, is observed in the nutrient medium of hydrolyzate C2.The low count can be attributed to the RS concentration of 13.4 g/L, which is insufficient for the active growth and vitality of acetobacteria.The low count of acetobacteria resulted in the absence of BC biosynthesis, which, in turn, explains the lack of BC gel film in the nutrient medium of hydrolyzate C2.
Figure 4b shows the residual RS concentration in the culture medium after 10 days of cultivation.The RS concentration after 10 days of cultivation in the synthetic nutrient media (control) was less than 4 g/L, while in the nutrient media of the enzymatic hydrolyzates, it ranged from 8 to 10 g/L.The slight decrease in RS concentration in the nutrient medium of hydrolyzate C2 during the biosynthesis process, from 13.4 g/L to 10 g/L, indicates the absence of active vitality in the acetobacteria responsible for BC production.The high residual RS concentration in nutrient medium C2 compared to C24 is explained by It can be observed from the presented data that the count of yeast at the end of the biosynthesis process exceeds that of acetobacteria in all cases.This can be attributed to the fact that the utilized producer is a consortium of various yeast and acetobacteria species and genera.According to literature data, yeast synthesizes ethanol to stimulate the growth of acetobacteria, which, in turn, produce BC to protect the yeast from the surrounding environment [89,90].Figure 4 indicates that in the synthetic nutrient medium (control) and in the C8 and C24 hydrolyzates, the count of acetobacteria remains relatively constant, ranging from 8-10 million CFU/mL.A low count of acetobacteria, 1 million CFU/mL, is observed in the nutrient medium of hydrolyzate C2.The low count can be attributed to the RS concentration of 13.4 g/L, which is insufficient for the active growth and vitality of acetobacteria.The low count of acetobacteria resulted in the absence of BC biosynthesis, which, in turn, explains the lack of BC gel film in the nutrient medium of hydrolyzate C2.
Figure 4b shows the residual RS concentration in the culture medium after 10 days of cultivation.The RS concentration after 10 days of cultivation in the synthetic nutrient media (control) was less than 4 g/L, while in the nutrient media of the enzymatic hydrolyzates, it ranged from 8 to 10 g/L.The slight decrease in RS concentration in the nutrient medium of hydrolyzate C2 during the biosynthesis process, from 13.4 g/L to 10 g/L, indicates the absence of active vitality in the acetobacteria responsible for BC production.The high residual RS concentration in nutrient medium C2 compared to C24 is explained by It can be observed from the presented data that the count of yeast at the end of the biosynthesis process exceeds that of acetobacteria in all cases.This can be attributed to the fact that the utilized producer is a consortium of various yeast and acetobacteria species and genera.According to literature data, yeast synthesizes ethanol to stimulate the growth of acetobacteria, which, in turn, produce BC to protect the yeast from the surrounding environment [89,90].Figure 4 indicates that in the synthetic nutrient medium (control) and in the C8 and C24 hydrolyzates, the count of acetobacteria remains relatively constant, ranging from 8-10 million CFU/mL.A low count of acetobacteria, 1 million CFU/mL, is observed in the nutrient medium of hydrolyzate C2.The low count can be attributed to the RS concentration of 13.4 g/L, which is insufficient for the active growth and vitality of acetobacteria.The low count of acetobacteria resulted in the absence of BC biosynthesis, which, in turn, explains the lack of BC gel film in the nutrient medium of hydrolyzate C2.
Figure 4b shows the residual RS concentration in the culture medium after 10 days of cultivation.The RS concentration after 10 days of cultivation in the synthetic nutrient media (control) was less than 4 g/L, while in the nutrient media of the enzymatic hydrolyzates, it ranged from 8 to 10 g/L.The slight decrease in RS concentration in the nutrient medium of hydrolyzate C2 during the biosynthesis process, from 13.4 g/L to 10 g/L, indicates the absence of active vitality in the acetobacteria responsible for BC production.The high residual RS concentration in nutrient medium C2 compared to C24 is explained by the low concentration of acetobacteria (Figure 4a) and, as a consequence, by the low consumption of RS in nutrient medium C2, indicating the absence of the active viability of acetobacteria.
The BC yield in the nutrient media of hydrolyzates C8 and C24 was 11.1% and 9.6%, respectively.This obtained yield is high and comparable to the control yield of 11.8%.These results indicate the preservation of BC yield when transitioning from a synthetic medium to nutrient media derived from the cellulose hydrolyzates of Miscanthus.A BC yield of 10% is not considered low.For example, when using Kombucha Original Bio as a producer, the BC yield in the synthetic nutrient medium (control) and in apple waste nutrient medium was 1% and 4%, respectively [44], which is 10 to 11 times lower than the BC yield obtained in our study on the hydrolyzates.BC yields ranging from 9.6% to 11.1% highlight the advantage of Medusomyces gisevii Sa-12 over individual strains that can yield BCs at only 2.2-6.5% [36,[91][92][93].
The morphology of the BC samples synthesized on synthetic nutrient medium (control) and enzymatic hydrolyzates was investigated by SEM (Figure 5).The overall morphological structure of the BC samples exhibited an intertwined network of microfibrils with inter-fibrillar spaces, consistent with the structure of BC samples reported in the literature [46,94,95].The width of the microfibrils for the BC sample synthesized on the synthetic nutrient medium (control) ranged from 26.0 nm to 229.0 nm, with an average width of 58.0 nm.The width of the microfibrils for the BC samples synthesized on enzymatic hydrolyzates C8 and C24 ranged from 24.0 nm to 186.0 nm, with an average width of 65.0 nm for C8 and 81.0 nm for C24, indicating values close to the control.The width of microfibrils in samples can depend on the nature of the producer or the composition of the nutrient medium [44,96].Therefore, in our case, the nutrient medium composition does not have a significant influence on this characteristic.The DP of the BC samples synthesized on enzymatic hydrolyzates was determined to be 2100 for C8 and 2300 for C24, compared to 2500 for the control.These values are relatively high and similar to each other [97,98].
Thus, it has been established that enzymatic hydrolyzates C8 and C24 are suitable for The DP of the BC samples synthesized on enzymatic hydrolyzates was determined to be 2100 for C8 and 2300 for C24, compared to 2500 for the control.These values are relatively high and similar to each other [97,98].
Thus, it has been established that enzymatic hydrolyzates C8 and C24 are suitable for obtaining high-quality BC samples.
Conclusions
Research has been conducted on the possibility of simultaneous production of two independent products from Miscanthus × giganteus cellulose: CNs and bacterial cellulose.Precursors for CNs and nutrient media for bacterial cellulose (BC) synthesis were obtained through an incomplete enzymatic hydrolysis of the Miscanthus cellulose sample for 2, 8, and 24 h.The solid residues obtained after hydrolysis, which were cellulose samples, differed in their structural characteristics from each other (DP 1510-1760, DC 72-75%), as well as from the original cellulose sample (degree of polymerization of 1770 and crystallinity of 64%).Nitration of the cellulose samples revealed that all precursors were suitable for chemical functionalization, as evidenced by the complete solubility (100%) of the synthesized CN in acetone.Prolonging the duration of enzymatic hydrolysis from 2 to 24 h resulted in a subsequent 0.48% increase in the nitrogen content of CN and a 15% yield increase.It was found that the maximum duration of enzymatic hydrolysis (24 h) led to the production of CN samples with satisfactory functional properties: nitrogen content of 11.83%, viscosity of 119 mPa•s, and solubility in a mixed alcohol/ester and diethyl ether mixture of 94%.SEM showed that during the nitration process, the fibers of the CN samples became smoother, retained the shape of the original cellulose fibers, and exhibited a slight increase in volume.FTIR spectroscopy demonstrated that the obtained CN were low-substituted nitrate esters of cellulose, as all spectra contained major functional group frequencies associated with nitro groups at 1657-1659 cm −1 and 1277 cm −1 .
The enzymatic hydrolysis of Miscanthus cellulose samples for 2, 8, and 24 h resulted in hydrolyzates with reducing sugar concentrations ranging from 13 to 28 g/L.It was found that nutrient media based on the hydrolyzates obtained after 8 and 24 h were of good quality and provided high BC yields of 11.1% and 9.6%, respectively.Scanning electron microscopy (SEM) revealed that the obtained BC samples had a mesh-like structure composed of nanoscale fibrils.The average width of the microfibrils in the BC samples synthesized using the 8 h hydrolyzate was 65.0 nm, while it was 81.0 nm for the 24 h hydrolyzate, which was close to the synthetic nutrient medium (control) at 58.0 nm.The DP of the BC samples was relatively high, measuring 2100 and 2300, respectively, which was slightly lower than the control at 2500.This approach of simultaneous production of CNs and BC has been applied for the first time and tested on lignocellulose from an energy plant, yielding unprecedented results.
Institutional Review Board Statement: Not applicable.
Figure 1 .
Figure 1.SEM images: (a,b) initial Miscanthus cellulose and CN from it; cellulose after 2 h hydrolysis and CN from it; (e,f) cellulose after 8 h hydrolysis and CN from it; and (g,h) cellulose after 24 h hydrolysis and CN from it.Pores on cellulose fibers after hydrolysis are indicated in the SEM images.
Figure 1 .
Figure 1.SEM images: (a,b) initial Miscanthus cellulose and CN from it; (c,d) cellulose after 2 h hydrolysis and CN from it; (e,f) cellulose after 8 h hydrolysis and CN from it; and (g,h) cellulose after 24 h hydrolysis and CN from it.Pores on cellulose fibers after hydrolysis are indicated in the SEM images.
Figure 2
Figure 2 presents the Fourier-transform infrared spectroscopy results for the cellulose and CN samples.
Figure 4 .
Figure 4. Indicators of BC biosynthesis after 10 days of cultivation in the control and in hydrolyzates C2, C8, and C24: (a) yeast and acetic acid bacteria count in the nutrient medium; (b) RS concentration; (c) BC yield.
Figure 4 .Figure 4 .
Figure 4. Indicators of BC biosynthesis after 10 days of cultivation in the control and in hydrolyzates C2, C8, and C24: (a) Figure 4. Indicators of BC biosynthesis after 10 days of cultivation in the control and in hydrolyzates C2, C8, and C24: (a) yeast and acetic acid bacteria count in the nutrient medium; (b) RS concentration; (c) BC yield.
yeast and Polymers 2024, 16, x FOR PEER REVIEW 15 of 22
Figure 4 .
Figure 4. Indicators of BC biosynthesis after 10 days of cultivation in the control and in hydrolyzates C2, C8, and C24: (a) yeast and acetic acid bacteria count in the nutrient medium; (b) RS concentration; (c) BC yield.
acetic acid bacteria count in the nutrient medium; (b) RS concentration; (c) BC yield.
Table 3 .
Basic functional properties of CN samples.
* Note: The yield calculated after CN was open air-dried. | 11,527.6 | 2023-12-21T00:00:00.000 | [
"Environmental Science",
"Materials Science",
"Engineering"
] |
Search for heavy resonances decaying into a $W$ or $Z$ boson and a Higgs boson in final states with leptons and $b$-jets in 36 fb$^{-1}$ of $\sqrt s = 13$ TeV $pp$ collisions with the ATLAS detector
A search is conducted for new resonances decaying into a $W$ or $Z$ boson and a 125 GeV Higgs boson in the $\nu\bar{\nu}b\bar{b}$, $\ell^{\pm}{\nu}b\bar{b}$, and $\ell^+\ell^-b\bar{b}$ final states, where $\ell^{\pm}= e^{\pm}$ or $\mu^{\pm}$, in $pp$ collisions at $\sqrt{s} = 13$ TeV. The data used correspond to a total integrated luminosity of 36.1 fb$^{-1}$ collected with the ATLAS detector at the Large Hadron Collider during the 2015 and 2016 data-taking periods. The search is conducted by examining the reconstructed invariant or transverse mass distributions of $Wh$ and $Zh$ candidates for evidence of a localised excess in the mass range of 220 GeV up to 5 TeV. No significant excess is observed and the results are interpreted in terms of constraints on the production cross-section times branching fraction of heavy $W^\prime$ and $Z^\prime$ resonances in heavy-vector-triplet models and the CP-odd scalar boson $A$ in two-Higgs-doublet models. Upper limits are placed at the 95% confidence level and range between $9.0\times 10^{-4}$ pb and $8.1\times 10^{-1}$ pb depending on the model and mass of the resonance.
Introduction
The ATLAS [1] and the CMS [2] collaborations discovered a Higgs boson (h) with a mass near 125 GeV and properties consistent with the Standard Model (SM) predictions [3][4][5]. Two of the most important questions that remain are how the Higgs boson mass is protected against large radiative corrections (the naturalness problem [6][7][8]) and whether the Higgs boson is part of an extended scalar sector [9], thus making this particle important for searches for new physics beyond the SM.
Various models with dynamical electroweak symmetry breaking scenarios attempt to solve the naturalness problem by assuming a new strong interaction at a higher energy scale. These models generically predict the existence of new vector resonances that naturally decay into a vector boson and a Higgs boson, for example in Minimal Walking Technicolour [10][11][12], Little Higgs [13], and composite Higgs models [14,15]. The decays into a vector-boson and Higgs boson final state are frequently enhanced in these models.
Another possible extension of the SM includes the addition of a second Higgs doublet [16]. A second Higgs doublet arises in many theories beyond the SM, collectively called two-Higgs-doublet models (2HDMs), such as the minimal supersymmetric SM [17][18][19][20][21], axion models [22], and baryogenesis models [23]. In 2HDMs with a CP-conserving Higgs potential, the scalar sector of the theory consists of five Higgs bosons: two charged (H ± ), two neutral CP-even (h, H) and one neutral CP-odd (A). This paper describes a search for the production of new heavy vector bosons, denoted hereafter by W and Z , that decay into a W or a Z boson and an h boson and a search for a heavy CP-odd scalar boson A that decays into a Z and an h boson. The analyses described here target leptonic decays of the vector bosons (W ± → ± ν, Z → + − /νν; ± = e ± , µ ± ) and decays of the h boson into a b-quark pair. This results in three search channels: W → W ± h → ± νbb, Z /A → Zh → + − bb, and Z /A → Zh → ννbb.
Resonance searches are typically not sensitive to all free parameters of the underlying theory, thus simplified models [24] can be used to parameterise a broad class of models, wherein only the relevant couplings and mass parameters are retained in the Lagrangian. For the interpretation of the results in the context of models with heavy vector triplets (HVT), a simplified model [25,26], based on a phenomenological Lagrangian is used as a benchmark. This model incorporates an SU(2) L triplet of heavy vector bosons, which allows the results to be interpreted in a large class of models. The new heavy vector bosons, W and Z , collectively denoted by V , couple to the Higgs and gauge bosons via a combination of parameters g V c H and to the fermions via the combination (g 2 /g V ) c F , where g is the SU(2) L gauge coupling. The parameter g V represents the strength of the new vector-boson interaction, and c H and c F represent corrections to the coupling strength specific to Higgs bosons and fermions, respectively. Two benchmark models are used in this analysis. In the first model, referred to as Model A, the branching fractions to fermions and gauge bosons are comparable, as in some models with an extended gauge symmetry [27]. For Model B, fermionic couplings are suppressed, as in strong dynamical models such as the minimal composite Higgs model [28]. At low resonance masses and large g V couplings, the HVT models fail to reproduce the SM parameters, thus this search focuses on high masses, from 500 GeV up to 5 TeV.
The results from the A → Zh search are interpreted as exclusion limits on the ratio of the vacuum expectation values of the two Higgs doublets, tan(β), and on cos(β − α), where α is the mixing angle between the two CP-even Higgs bosons. The exclusion limits are evaluated for the Type I, Type II, Lepton-specific, and Flipped 2HDMs. These differ with respect to which doublets couple to the up-type and down-type quarks as well as to the charged leptons [16]. Both the production via gluon-gluon fusion and the production with associated b-quarks (bbA) are considered in this search. The A → Zh decay mode is mostly relevant below the tt production threshold and the cross-section falls steeply with increasing A boson mass. Therefore, this search starts at the Zh threshold of approximately 220 GeV and goes up to 2 TeV.
Previous searches in the same final states have been performed by the ATLAS and the CMS collaborations using data at √ s = 8 TeV and 13 TeV. The ATLAS searches for W → Wh (Z → Zh) exclude, at 95% confidence level (CL), W (Z ) resonances with masses below 1.75 (1.49) TeV assuming the HVT benchmark Model A (g V = 1) and below 2.22 (1.58) TeV assuming Model B (g V = 3) [29,30]. Searches by the CMS Collaboration exclude resonances with masses less than 2.0 TeV at 95% CL assuming the HVT benchmark Model B (g V = 3) [31]. Searches using the fully hadronic final state (W/Zh → qq bb) have also been performed by CMS and ATLAS and exclude W (Z ) resonances below 3.15 TeV (2.6 TeV) assuming the HVT benchmark Model B (g V = 3) [32-34]. Previous searches for a CP-odd scalar boson A in the Zh decay mode are reported in Refs. [35][36][37][38][39].
The search presented in this paper is performed by looking for a localised excess in the distribution of the reconstructed mass of the ννbb, ± νbb, and + − bb systems. The mass range covered by the search, from 220 GeV to 5 TeV, probes a wide range of Higgs boson transverse momenta. Thus, two methods are used to reconstruct Higgs boson candidates. At low transverse momenta, the decay products of the Higgs boson are reconstructed as individual jets. At high transverse momenta, the decay products start to merge and are reconstructed as a single jet. The signal yield and background normalisations are determined from a binned maximum-likelihood fit to the data distribution for each of the V and A boson models (W , Z , gluon-gluon fusion A, bbA) and are used to set upper limits on the production cross-section times decay branching fraction. A combined fit using all three lepton channels sets bounds on the HVT model in the case where the V bosons are degenerate in mass. This paper is structured as follows. Sections 2 and 3 provide a brief description of the ATLAS experiment and the data and simulated event samples. The event reconstruction and selections are discussed in Sections 4 and 5. The background estimation and systematic uncertainties are described in Sections 6 and 7. Finally, Sections 8 and 9 detail the statistical analysis and provide a discussion of the results and concluding remarks.
ATLAS detector
The ATLAS detector [40] at the LHC covers nearly the entire solid angle 1 around the collision point. It consists of an inner tracking detector (ID) surrounded by a thin superconducting solenoid, electromagnetic and hadronic calorimeters, and a muon spectrometer incorporating three large superconducting toroid magnets.
The ID is immersed in a 2 T axial magnetic field and provides charged-particle tracking in the range |η| < 2.5. It consists of silicon pixel, silicon microstrip, and transition radiation tracking detectors. Prior to the data-taking at the increased centre-of-mass energy of 13 TeV, the ID was enhanced by adding a new layer of pixel detectors (the IBL [41]) inside the existing pixel detector layers in the barrel region (at a radius of approximately 33 mm). The upgraded detector typically provides four three-dimensional 1 ATLAS uses a right-handed coordinate system with its origin at the nominal interaction point (IP) in the centre of the detector and the z-axis along the beam pipe. The x-axis points from the IP to the centre of the LHC ring, and the y-axis points upwards. Cylindrical coordinates (r, φ) are used in the transverse plane, φ being the azimuthal angle around the z-axis. The pseudorapidity is defined in terms of the polar angle θ as η = − ln tan(θ/2). Angular distance is measured in units of ∆R ≡ (∆η) 2 + (∆φ) 2 . measurements for tracks originating from the luminous region. The silicon microstrip tracker provides four two-dimensional measurement points per track. The transition radiation tracker enables track reconstruction at large radii up to |η| = 2.0 and provides electron identification information based on the number of hits above the threshold for transition radiation.
The calorimeter system covers the pseudorapidity range |η| < 4.9. Within the region |η| < 3.2, electromagnetic calorimetry is provided by barrel and endcap high-granularity lead/liquid-argon (LAr) electromagnetic calorimeters. An additional thin LAr presampler, covering |η| < 1.8, is used to correct for energy loss in material upstream of the calorimeters. Hadronic calorimetry is provided by a steel/scintillatortile calorimeter, segmented into three barrel structures within |η| < 1.7, and two copper/LAr hadronic endcap calorimeters. The solid-angle coverage is completed with forward copper/LAr and tungsten/LAr calorimeter modules optimised for electromagnetic and hadronic measurements, respectively.
The muon spectrometer is composed of separate trigger and high-precision tracking chambers, measuring the deflection of muons in a magnetic field generated by superconducting air-core toroids. The precision chamber system covers the region |η| < 2.7 with three layers of monitored drift tubes, complemented by cathode strip chambers in the forward region, where the particle flux is highest. The muon trigger system covers the range |η| < 2.4 with resistive plate chambers in the barrel, and thin-gap chambers in the endcap regions.
A two-level trigger system is used to select interesting events [42]. The level-1 trigger is implemented in hardware and uses a subset of detector information to reduce the event rate to a design value of at most 100 kHz. This is followed by the software-based trigger level, the high-level trigger, which reduces the event rate further to about 1 kHz.
Data and simulated event samples
The data used in this analysis were recorded with the ATLAS detector during the 2015 and 2016 ppcollision runs at √ s = 13 TeV and correspond to a total integrated luminosity of 36.1 fb −1 . The data are required to satisfy a number of criteria that ensure that the ATLAS detector was in good operating condition. A number of Monte Carlo (MC) simulation samples are used to model the background and signal processes for this search.
For the W and Z processes, simulated events were generated with MadGraph5_aMC@NLO (MG5_aMC) 2.2.2 [43] at leading-order (LO) accuracy using the NNPDF 2.3 LO parton density function (PDF) set [44]. The parton shower and hadronisation were simulated with Pythia 8.186 [45] using the A14 set [46] of tuned parameters ("tune") together with the NNPDF 2.3 LO PDF set. Events were generated for a range of resonance masses from 500 to 5000 GeV, assuming a zero natural width. Higgs boson decays into bb and cc pairs were simulated, with a relative branching fraction B(h → cc)/B(h → bb) = 0.05 fixed to the SM prediction [47].
Events for the gluon-gluon fusion production of A bosons were generated at LO accuracy using the same set-up as for the W and Z samples. The b-quark associated production of A bosons was simulated with MG5_aMC 2.2.3 using next-to-leading-order (NLO) matrix elements with massive b-quarks and the CT10F4 NLO PDF set [48]. The parton shower and hadronisation were simulated with Pythia 8.210 [49]. Events were generated for a range of A boson masses from 220 to 2000 GeV assuming a zero natural width. For the A boson signals, only decays of the Higgs boson into a bb pair were generated.
For the interpretation of the A → Zh search in the context of 2HDMs, the masses of the H ± and H bosons are assumed to be equal to the mass of the A boson. The cross-sections were calculated using up to next-to-next-to-leading-order (NNLO) QCD corrections for gluon-gluon fusion and b-quark associated production in the five-flavour scheme as implemented in Sushi [50][51][52][53]. For the b-quark associated production, a cross-section in the four-flavour scheme was also calculated as described in Refs. [54,55] and the results were combined with the five-flavour scheme calculation following Ref. [56]. The A boson width and the branching fractions for A → Zh and h → bb have been calculated using 2HDMC [57,58]. The procedure for the calculation of the cross-section and branching fractions as well as the choice of the 2HDM parameters follows Ref. [9].
The production of W and Z bosons in association with jets was simulated with Sherpa 2.2.1 [59] using the NNPDF 3.0 NNLO PDF set [60] for both the matrix element calculation and the dedicated parton-shower tuning developed by the Sherpa authors. The event generation utilises Comix [61] and OpenLoops [62], for the matrix element calculation, matched to the Sherpa parton shower using the ME+PS@NLO prescription [63]. The matrix elements were calculated for up to two additional partons at NLO and for three and four partons at LO in QCD. The cross-sections for W/Z+jets were calculated at NNLO accuracy [64].
The production of single top quarks (t-channel, s-channel, and Wt) was simulated using the Powheg-Box event generator with the CT10 PDF set. The shower and hadronisation was simulated using the same event generator and set-up as for the tt process. The cross-section for the t-and s-channel single-topquark production was calculated at NLO accuracy using Hathor v2.1 [79,80], while for the Wt process an approximate NNLO calculation was used [81]. The top mass is set fixed to 172.5 GeV in the tt and single-top-quark samples.
Diboson events (WW, WZ, ZZ) were simulated using the Sherpa 2.1.1 event generator using the CT10 PDFs. Matrix elements were calculated for up to one (ZZ) or no (WW, WZ) additional partons at NLO and up to three additional partons at LO, and the cross-sections were calculated at NLO accuracy.
Finally, the SM processes Vh(h → bb), tth, ttW, and ttZ are included in the total background estimation. The qq → Zh and qq → Wh processes were simulated at LO with Pythia 8.186 using the NNPDF 2.3 LO PDF set and the A14 tune. The gg → Zh process was simulated at NLO using the Powheg-Box v2 event generator with the CT10 PDF set. The modelling of the shower, hadronisation and underlying event was provided by Pythia 8.186 using the AZNLO tune [82] with the same PDF set as for the matrix element calculation. The cross-sections for the Wh and Zh processes were taken from Ref. [9]. The tth and ttV samples were generated at NLO accuracy with MG5_aMC 2.3.2 interfaced to Pythia 8.210. The NNPDF 3.0 NLO PDF set was used in the matrix element calculation while for the parton shower the A14 tune was used with the NNPDF 2.3 LO PDF set.
A summary of event generators used for the simulation of signal and background processes is shown in Table 1.
All simulated event samples include the effect of multiple pp interactions in the same and neighbouring bunch crossings (pile-up) by overlaying simulated minimum-bias events on each generated signal or background event. The minimum-bias events were simulated with the single-, double-and non-diffractive pp processes of Pythia 8.186 using the A2 tune [83] and the MSTW2008 LO PDF [84]. For all Mad-Graph and Powheg samples, the EvtGen v1.2.0 program [85] was used for the bottom and charm hadron decays. The generated samples were processed using the Geant-based ATLAS detector simulation [86,87] and the same event reconstruction algorithms were used as for the data.
Event reconstruction
This search makes use of the reconstruction of multi-particle vertices, the identification and the kinematic properties of reconstructed electrons, muons, τ leptons, jets, and the determination of missing transverse momentum.
Collision vertices are reconstructed from at least two ID tracks with transverse momentum p T > 400 MeV. The primary vertex is selected as the one with the highest p 2 T , calculated considering all associated tracks.
Electrons are reconstructed from ID tracks that are matched to energy clusters in the electromagnetic calorimeter. The clusters are reconstructed using the standard ATLAS sliding-window algorithm, which clusters calorimeter cells within fixed-size η/φ rectangles [88]. Electron candidates are required to satisfy requirements for the electromagnetic shower shape, track quality, and track-cluster matching; these requirements are applied using a likelihood-based approach. The "Loose" and "Tight" working points defined in Ref. [89] are used.
Muons are identified by matching tracks found in the ID to either full tracks or track segments reconstructed in the muon spectrometer ("combined muons"), or by stand-alone tracks in the muon spectrometer [90]. Muons are required to pass identification requirements based on quality requirements applied to the ID and muon spectrometer tracks. The "Loose" and "Medium" identification working points defined in Ref. [90] are used in this analysis. The "Loose" working point includes muons reconstructed with the muon spectrometer alone to extend the acceptance to |η| = 2.7.
Electron and muon candidates are required to have a minimum p T of 7 GeV and to lie within a region where there is good reconstruction and identification efficiency (|η| < 2.7 for muons and |η| < 2.47 for electrons). The "Loose" lepton identification criterion has to be fulfilled and all candidates have to originate from the primary vertex. The last condition is satisfied by requiring that the significance of the transverse impact parameter |d 0 |/σ(d 0 ) is less than 5.0 for electrons (< 3.0 for muons) and |z 0 sin(θ)| is less than 0.5 mm, where z 0 is the longitudinal impact parameter and θ is the polar angle defined in Section 2. The lepton candidates are required to be isolated using requirements on the sum of the p T of the tracks lying in a cone around the lepton direction whose size, ∆R, decreases as a function of the lepton p T [91]. The efficiency of the isolation selection is tuned to be larger than 99% in a sample of Z → + − decays [88,90,92]. The identification and isolation efficiencies of both the electrons and muons are calibrated using a tag-and-probe method in Z → + − data events [88,90].
Two types of calorimeter-based jets, "small-R" and "large-R" jets, are used to reconstruct Higgs boson candidates over a wide momentum spectrum. Small-R jets are reconstructed from noise-suppressed topological clusters in the calorimeter [93] using the anti-k t [94] algorithm implemented in the FastJet package [95] with a radius parameter R = 0.4 and are required to have a p T > 20 GeV for |η| < 2.5 (central jets) or p T > 30 GeV for 2.5 < |η| < 4.5 (forward jets). To reduce the number of small-R jets originating from pile-up interactions, these jets are required to pass the jet vertex tagger [96] selection, with an efficiency of about 90%, if they are in the range p T < 60 GeV and |η| < 2.4.
Large-R jets are used to reconstruct Higgs boson candidates with high momenta for which the b-quarks are emitted close to each other. They are constructed using the anti-k t algorithm with a radius parameter of R = 1.0 and are trimmed [97] to remove the energy of clusters that originate from initial-state radiation, pile-up interactions or the underlying event. This is done by reclustering the constituents of the initial jet, using the k t algorithm [98,99], into smaller R sub = 0.2 subjets and then removing any subjet that has a p T less than 5% of the p T of the parent jet [100]. The jet mass resolution is improved at high momentum using tracking in addition to calorimeter information [101]. Large-R jets are required to have p T > 250 GeV and |η| < 2.0.
The momenta of both the large-R and small-R jets are corrected for energy losses in passive material and for the non-compensating response of the calorimeter. Small-R jets are also corrected for the average additional energy due to pile-up interactions [102,103].
A third type of jet, built from tracks (hereafter referred to as a track-jet), is used in this analysis for the identification of b-jets from decays of boosted Higgs bosons. The jets are built with the anti-k t algorithm with R = 0.2 from at least two ID tracks with p T > 400 MeV associated with the primary vertex, or with a longitudinal impact parameter |z 0 sin(θ)| < 3 mm [104]. Track-jets are required to have p T > 10 GeV and |η| < 2.5, and are matched to the large-R jets via ghost-association [105].
Small-R jets and track-jets containing b-hadrons are identified with the multivariate MV2c10 b-tagging algorithm [106,107], which makes use of information about the jet kinematics, the properties of tracks within jets, and the presence of displaced secondary vertices. The algorithm is used at the 70% efficiency working point and provides a factor of 380 (120) in rejecting small-R jets (track-jets) from gluons and light quarks, and a factor of 12 (7) in rejecting small-R jets (track-jets) from c-quarks. Jets satisfying these requirements are referred to as "b-tagged jets".
To improve the mass resolution of the Higgs boson candidate, dedicated energy corrections are applied for b-tagged small-R and large-R jets to account for the semileptonic decays of the b-hadrons. The momentum of the closest muon in ∆R with p T larger than 5 GeV inside the jet cone is added to the jet momentum after removing the energy deposited by the muon in the calorimeter (the muon-in-jet correction) [104]. For this correction, muons are not required to pass the isolation requirements. For small-R jets only, an additional p T -dependent correction, denoted "PtReco", is applied to the jet four-momentum to account for biases in the response of b-jets, improving the resolution of the dijet mass. This correction is determined from Vh(h → bb) simulated events by calculating the ratio of the p T of the true b-jets from the Higgs boson decay to the p T of the reconstructed b-tagged jets after the muon-in-jet correction. The resolution of the dijet mass, m jj (m J ), in this process is improved by 18% (22%) for the resolved (merged) Higgs boson reconstruction after these corrections, as shown in Figure 1. Hadronically decaying τ-lepton candidates (τ had ) are identified using small-R jets with p T > 20 GeV and |η| < 2.5, outside the transition region between the barrel and endcap calorimeters (1.37 < |η| < 1.52). These τ had candidates must have either one or three associated tracks and must satisfy the "Medium" identification criterion [109]. They are used in the ννbb channel to reject backgrounds with real hadronic τ-leptons.
ATLAS
The presence of neutrinos in the ννbb and ± νbb final states can be inferred from a momentum imbalance in the transverse plane. The missing transverse momentum ( E miss T ) is calculated as the negative vectorial sum of the transverse momenta of all the muons, electrons, small-R jets, and ID tracks associated with the primary vertex but not associated with any of those leptons and jets [110,111]. To suppress non-collision and multijet backgrounds in the ννbb channel, an additional track-based missing transverse momentum estimator, p miss T , is built independently as the negative vectorial sum of the transverse momenta of all tracks from the primary vertex.
An overlap-removal algorithm is applied to prevent double counting of the leptons and jets used for the resonance reconstruction. A τ-lepton is removed if the ∆R between the τ-lepton and an electron or a muon is below 0.2. In the case of a muon, the τ-lepton is not removed if the τ-lepton has p T above 50 GeV and the reconstructed muon is not a combined muon. If a reconstructed muon and electron share the same ID track then the electron is removed. Small-R jets are removed if they are within a cone of size ∆R = 0.2 around an electron or muon that has passed the isolation requirements. To account for semi-muonic b-jet decays, the jet is only removed if it has fewer than three associated tracks, or if more than 70% of the sum of the p T of its associated tracks comes from the muon and p is the p T of the jet (muon). Next, electrons and muons within a cone of size ∆R = 0.4 around a surviving small-R jet are discarded if their distance from the jet direction is smaller than ∆R = (0.04 + 10 GeV/p T ). The shrinking cone size ensures a high efficiency for boosted topologies. Small-R jets are also removed if they are within ∆R = 0.2 of the axis of a τ had candidate. Finally, large-R jets within ∆R = 1.2 of any surviving electron are removed.
Analysis strategy and event selection
The search for the Z and A bosons in the Zh → ννbb and Zh → + − bb decay modes uses event samples wherein the number of reconstructed charged leptons is exactly zero or two (0-lepton and 2-lepton channels). For the W search in the Wh → ± νbb channel, events with exactly zero or one charged lepton are used (0-lepton or 1-lepton channels). The lepton selection requirements described in the previous section are applied, using the "Loose" identification working point. The selections outlined below define regions sensitive to the different models.
For the 0-lepton channel, an E miss T trigger with a threshold of 70 GeV was used to record the data in 2015 runs; the threshold varied between 90 and 110 GeV in 2016 runs due to the increasing instantaneous luminosity. Events are required to have E miss T > 150 GeV, where E miss T is reconstructed with fully calibrated leptons and jets. The efficiency of the trigger selection exceeds 80% above 150 GeV. In the 2-lepton channel, events were recorded using a combination of single-lepton triggers with isolation requirements. In 2015, the lowest p T threshold was 24 GeV; in 2016, it ranged from 24 to 26 GeV. Additional triggers without an isolation requirement are used to recover efficiency for leptons with p T > 60 GeV. In the single-electron channel, the same single-electron triggers as in the 2-lepton channel are applied. In the single-muon channel, the same E miss T triggers as in the 0-lepton channel are used because they are more efficient than the single-muon triggers for this analysis. For events selected by the lepton triggers, the lepton that satisfied the trigger is required to match a reconstructed electron (muon) with p T > 27 GeV and |η| < 2.47 (|η| < 2.5).
The wide range of resonance masses probed by this search implies that the resonance decay products can be produced with a wide range of transverse momenta. When the Higgs boson has relatively low p T , the b-quarks from its decay can be reconstructed as two small-R jets. As the momentum of the Higgs boson increases, the two b-quarks become more collimated and a selection using a single large-R jet becomes more efficient. Two different methods are used for the reconstruction of the Higgs boson candidate: a "resolved" category in which two small-R jets are used to build the Higgs boson candidate, and a "merged" category where the highest-p T ("leading") large-R jet is selected as the Higgs boson candidate.
For the resolved signal region, two small-R jets are required to have an invariant mass (m jj ) in the range 110 -140 GeV for the 0-and 1-lepton channels and in the range 100 -145 GeV for the 2-lepton channel. The latter selection is relaxed to take advantage of the smaller backgrounds in this channel. This dijet candidate is defined by the two leading b-tagged small-R jets when two or more b-tagged jets are present in the event. In the case where only one b-tagged jet is present, the dijet pair is defined by the b-tagged jet and the leading small-R jet in the remaining set. The leading jet in the pair must have p T > 45 GeV. For the merged signal region, a large-R jet is required with mass (m J ) in the range 75 to 145 GeV and at least one associated b-tagged track-jet.
Events which satisfy the selection requirements of both the resolved and merged categories, are assigned to the resolved one, since its better dijet mass resolution and lower background contamination increases the expected sensitivity. Events failing to satisfy both the resolved and merged signal region requirements are assigned to control regions defined in Section 6, with priority given to the resolved category. This procedure provides a higher sensitivity for resonances of mass near 1 TeV compared to a procedure in which the merged category is prioritised.
Higgs boson candidates with one or two b-tagged jets define the "1 b-tag" or "2 b-tag" categories, respectively. For the merged selection, only one or two leading track-jets associated with the large-R jet are considered in this counting. For the 0-and 2-lepton channels, resolved events with more than two b-tagged jets or merged events with additional b-tagged track-jets not associated with the large-R jet are used to define signal regions sensitive to bbA production. These are labeled as "3+ b-tag" in the resolved category, and "1 b-tag additional b-tag" or "2 b-tag additional b-tag" in the merged category. In the 2-lepton channel, the latter two are merged and labeled as "1+2 b-tag additional b-tag".
The calculation of the reconstructed resonance mass depends on the decay channel. In the 0-lepton channel, where it is not possible to reconstruct the Zh system fully due to the presence of two neutrinos from the Z boson decay, the transverse mass defined as is used as the final discriminant. In order to reconstruct the invariant mass of the Wh → ± νbb system in the 1-lepton channel, the momentum of the neutrino in the z-direction, p z , is obtained by imposing a W boson mass constraint on the lepton-E miss T system. In the resulting quadratic equation, the neutrino p z is taken as the real component in the case of complex solutions, or as the smaller of the two solutions if both solutions are real. The mass resolution of the Vh system is improved in the resolved signal regions of all channels by rescaling the four-momentum of the dijet system by 125 GeV/ m jj . In the 2-lepton channel, the four-momentum of the dimuon system is scaled by 91.2 GeV / m µµ as well in all signal regions. This helps to address the worse momentum resolution of high-momentum muons which are measured solely by the tracking detectors.
Additional selections are applied for each lepton channel, as outlined below, to reduce the main backgrounds and enhance the signal sensitivity. These selections are summarised in Table 2.
For the resolved and merged categories in the 0-lepton channel, the following selections are applied to reduce multijet and non-collision backgrounds to a negligible level: • p miss T > 30 GeV (not applied in the resolved 2 and 3+ b-tag categories); • the azimuthal angle between E miss T and p miss T , ∆φ( E miss T , p miss T ) < π/2; • the azimuthal angle between E miss T and the Higgs boson candidate momentum direction, ∆φ( E miss T , h) > 2π/3; • the azimuthal angle between E miss T and the nearest small-R jet momentum direction, min[∆φ( E miss T , small-R jet)] > π/9 (for the resolved category with four or more jets, > π/6 is used). For the resolved category, min[∆φ( E miss T , small-R jet)] is calculated using the small-R jets that constitute the Higgs boson candidate and an additional small-R jet, which is the third leading b-tagged jet (if the event contains at least three b-tagged jets), the leading central jet which is not b-tagged (if the event contains only two b-tagged jets) and the leading forward jet (if the event contains only two central small-R jets). For the merged category, all central and forward small-R jets are used in the min[∆φ( E miss T , small-R jet)] calculation. For the Z /A search, the tt and W+jets backgrounds are further reduced by rejecting events with at least one identified τ had candidate. This veto is not applied when searching for the W boson or in the HVT combined search, because it leads to a loss of signal events in the Wh → τ ± νbb final state.
For the 0-lepton resolved category, two additional selections are applied: • the scalar sum of the p T of the three leading central small-R jets, p jet i T , is greater than 150 GeV. In the case where there are only two central small-R jets, the sum of the p T of these two jets and of the leading forward small-R jet, if any, is required to be greater than 120 GeV; • the azimuthal angle between the two jets used to reconstruct the Higgs boson candidate, ∆φ(j, j), is required to be less than 7π/9.
Finally, for the merged category, the missing transverse momentum must be larger than 200 GeV.
For the 1-lepton channel, a selection on the transverse momentum of the W boson candidate (p T,W ), which increases as a function of the reconstructed resonance mass, is applied to reduce the contribution of W+jets: p T,W > max[150, 710 − 3.3 × 10 5 GeV/m Vh ] GeV for the resolved category, while p T,W > max[150, 394 · ln(m Vh /(1 GeV)) − 2350] GeV for the merged category. This selection is optimised taking advantage of the larger transverse momentum of W bosons expected to be produced in the decays of high-mass resonances. The tt background is reduced in the resolved category by requiring fewer than four central jets in the event and in the merged category by rejecting events with additional b-tagged track-jets not associated with the large-R jet. For all categories, the transverse mass of the W candidate (m T,W ), calculated from the transverse components of the lepton and E miss T momentum vectors, is required to be less than 300 GeV.
In the 1-lepton channel, a significant contribution of multijet events arises mainly from non-prompt leptons from hadron decays and from jets misidentified as electrons. This background is significantly reduced by applying tighter selection requirements on the lepton isolation and identification, as well as on E miss T . Muons must satisfy the "Medium" identification and electrons must satisfy the "Tight" identification requirements. Stringent lepton isolation requirements are applied: the scalar sum of the p T of tracks within a variable-size cone around the lepton (excluding its own track) must be less than 6% of the lepton p T . In addition, in the case of electrons the sum of the transverse energy of the calorimeter energy clusters in a cone of ∆R = 0.2 around the electron must be less than 6% of the electron p T [90,92]. Finally, the E miss T value is required to be greater than 100 GeV for the merged category and greater than 80 (40) GeV for the resolved category in the electron (muon) channel.
In the 2-lepton channel, same-flavour leptons (ee or µµ) are used. For both the resolved and merged categories, three kinematic selections are optimised as a function of the resonance mass to reduce the tt and Z+jets backgrounds. Selections on the mass of the dilepton system, max[40 GeV, 87 GeV − 0.030 · m Vh ] < m < 97 GeV +0.013·m Vh , and on E miss T / √ (1 GeV) · H T < 1.15 +8×10 −3 ·m Vh / (1 GeV) are relaxed > π/9 (2 or 3 jets), > π/6 (≥ 4 jets) for higher-mass resonances to account for resolution effects and smaller backgrounds. The variable H T is calculated as the scalar sum of the p T of the leptons and small-R jets in the event. The momentum of the dilepton system (p T, ) is required to be greater than 20 GeV + 9 GeV · √ m Vh /(1 GeV) − 320 for m Vh greater than 320 GeV. In the resolved dimuon category, an opposite-charge requirement is applied since the probability to mis-reconstruct the charge of individual muons is extremely low. Additionally, in this category the leading muon is required to have |η| less than 2.5. Finally, for the merged category, the sub-leading lepton is required to have p T > 25 GeV and for muons |η| is restricted to be less than 2.5.
Background estimation
The background contamination in the signal regions is different for each of the three channels studied. In the 0-lepton channel, the dominant background sources are Z+jets and tt events with a significant contribution from W+jets. In the 1-lepton channel, the largest backgrounds are tt, single-top-quark and W+jets production. In the 2-lepton channel, Z+jets production is the predominant background followed by the tt background. The contribution from diboson, SM Vh, tth, and ttV production is small in all three channels. The multijet background, due to semileptonic hadron decays or misidentified jets, is found to be negligible in the 0-and 2-lepton channels after applying the event selections described in Section 5. In the 1-lepton channel, the multijet background remains significant only in the resolved 1 b-tag category. All background distribution shapes except those for multijet are estimated from the samples of simulated events with normalisations of the main backgrounds estimated from the data; the multijet shape and normalisation is determined using data.
The W/Z+jets simulated event samples are split into different components. In the resolved category, the samples are split according to the true flavour of the two small-R jets forming the Higgs boson candidate. In the merged category, they are split according to the true flavour of the one or two leading track-jets associated with the large-R jet. The true jet flavour is determined by counting true heavy-flavour hadrons with p T > 5 GeV within the cone of the reconstructed jet. If a true b-hadron is found, the jet is labelled as a b-jet, otherwise if a true c-hadron is found the jet is labelled as a c-jet. If neither a true b-hadron nor a true c-hadron is associated with the reconstructed jet, it is labelled as a light jet. For large-R jets with only one track-jet, the true hadrons are counted within this track-jet. Based on this association scheme, the W/Z+jets simulated event samples are split into six components: W/Z+bb, W/Z+bc, W/Z+bl, W/Z+cc, W/Z+cl and W/Z+ll; in this notation l refers to a light jet. In the statistical analysis described in Section 8, the components W/Z+bb, W/Z+bc, and W/Z+cc are treated as a single component denoted by W/Z+(bb, bc, cc). The combination of W/Z+bl and W/Z+cl is denoted by W/Z+(bl, cl). For the HVT, Z , and A boson interpretations, the normalisations of the largest components Z+(bb, bc, cc) and Z+(bl, cl) are determined from data. In the A boson interpretation, the Z+(bb, bc, cc) background normalisation in the 3+ b-tag region is determined from this region independently. The normalisations of W+(bb, bc, cc) and W+(bl, cl) are determined from data for the W and HVT interpretations.
The normalisation of the tt background is determined from the fits to data separately for the 0-, 1-, and 2lepton channels. In the 0-lepton channel, only the signal regions are used in the fit. In the 1-and 2-lepton channels, dedicated control regions enhanced in tt events are used in addition to the signal regions. In the 1-lepton channel, resolved events in the sidebands of the m jj distribution between 50 GeV and 200 GeV (excluding the signal region with 110 < m jj < 140 GeV) are primarily composed of tt and W+jets events. These control regions are included in the fit for the 1 and 2 b-tag categories. In the 2-lepton channel, a tt control region is defined using resolved events with different-flavour (eµ), oppositely charged leptons, and without the E miss T / √ H T requirement. The tt purity of this selection is greater than 90%. This region combining the 1 and 2 b-tag events is used in the A, Z , and HVT interpretations; for the A interpretation, a control region with eµ events and 3+ b-tags is also included in the fit to provide an independent constraint on tt production with associated heavy-flavour jets.
The shape of the multijet background in the 1-lepton channel is estimated from a sample of data events orthogonal to the signal regions, the anti-isolated lepton region. In the muon channel, this region is defined by events where the sum of the transverse momentum of tracks in a cone of ∆R = 0.2 around the muon is between 6% and 15% of the muon p T . In the electron channel, this region is defined by events where the sum of the calorimeter energy deposits in a cone of ∆R = 0.2 around the electron is larger than 6% of the electron p T ; this region is defined after applying the track isolation requirement described in Section 4. A template shape for the multijet background is extracted from the anti-isolated lepton region after removing the contribution from the simulated electroweak and top-quark backgrounds. In this subtraction, the normalisation of the simulated electroweak and top-quark backgrounds is estimated Table 3: Relative systematic uncertainties in the normalisation, cross-region extrapolation, and shape of the signal and background processes included in the fits described in the text. An "S" indicates a shape variation is included for the sources listed, "/" indicates a ratio of two regions, and "norm." is the sum of cross-section and acceptance variations. A range of values means the value depends on the lepton channel. Parentheses indicate when the uncertainty applies only to a given fit or a given region. > 200 GeV where the contribution from multijet events is negligible. In the signal and control regions used in the statistical analysis, the multijet normalisation is determined by fitting the E miss T multijet template and the E miss T combined template of the electroweak and top-quark backgrounds to data in the 1 and 2 b-tag categories separately. Using this method, the multijet contribution is estimated to be less than 6% in all signal and control regions and is included in the statistical analysis.
Systematic uncertainties
Two types of systematic uncertainties, experimental and modelling, affect the reconstruction of the m Vh and m T,Vh observables. Experimental uncertainties arise due to the trigger selection, the reconstruction, identification, energy/momentum, mass, and resolution for the leptons, jets and missing transverse momentum. Modelling uncertainties result in shape and normalisation uncertainties of the different MC samples used to model the signal and backgrounds. These stem from uncertainties in the matrix element calculation, the choice of parton shower and hadronisation models and their free parameters, the PDF set and the choice of renormalisation and factorisation scales.
The largest experimental systematic uncertainties are associated with the calibration and resolution of the small-R and large-R jet energy, the calibration and resolution of the large-R jet mass, and the determination of the jet b-tagging efficiency and misidentification rate. The uncertainties in the small-R jet energy scale have contributions from in situ calibration studies, from the dependency on the pile-up activity and on the flavour composition of jets [103,112]. The small-R jet uncertainties are propagated to the E miss T measurement. The uncertainty in the scale and resolution of large-R jet energy and mass is estimated by comparing the ratio of calorimeter-based to track-based measurements in dijet data and simulation [100, 101]. The flavour tagging efficiency and its uncertainty for b-jets and c-jets is estimated in tt and W + cjet events, respectively, while the light-jet misidentification rate and uncertainty is determined using dijet events [106, 107,113,114]. Other experimental systematic uncertainties with a smaller impact are those in the lepton energy and momentum scales, in lepton reconstruction and identification efficiency, and in the efficiency of the triggers. Finally, a global normalisation uncertainty of 3.2% is assigned due to the luminosity measurement from a preliminary calibration of the luminosity scale using x-y beamseparation scans performed in August 2015 and May 2016, following a methodology similar to that detailed in Ref. [115]. Experimental uncertainties have an impact on the shape of the mass distributions and account for possible migration of events across the different regions.
Modelling uncertainties are assigned to each signal and background process and lead to variations in the normalisation and in the case of main backgrounds also in the shape of the templates in the different regions. In addition, for all MC samples, the statistical uncertainty arising from the number of simulated events is considered by introducing shape variations determined from the uncertainty in each bin of the m Vh or m T,Vh distributions. The modelling uncertainties considered are shown in Table 3 and described below.
For the signal processes, the uncertainties in the acceptance were derived by considering the following variations: the renormalisation and factorisation scales were varied by a factor of two, the nominal PDF set was replaced by the MSTW2008 LO PDF set and the tuned parameters were varied according to the variations derived from the eigentune method [46]. For both the A and V signals, the total variations are less than 3% at resonance masses above 500 GeV. The variations increase to 7% for the A boson masses below 500 GeV.
The modelling uncertainties affecting tt and single-top-quark processes are derived as follows [116]. A variation of the parton shower, hadronisation, and the underlying-event model is obtained by replacing Pythia 6.428 by Herwig++ (version 2.7.1) [117] with the UE-EE-5 tune and the CTEQ6L1 PDF set [70]. To assess potential differences in the matrix element calculation, a comparison is made to a sample where Powheg is replaced by MG5_aMC [43]. A comparison is also made to samples with smaller and larger amount of initial-and final-state radiation (ISR/FSR) by changing the renormalisation and factorisation scales by a factor of two and switching to the corresponding low-and high-radiation Perugia 2012 tunes. Finally, the difference between the nominal and corrected distributions due to the top-quark and tt p T reweighting described in Section 3 is included as a symmetrised shape uncertainty.
Similarly, for W/Z+jets backgrounds, the following comparisons have been performed. The PDF set in the nominal samples was replaced by the alternative PDF sets: the hundred NNPDF 3.0 NNLO replicas, including the sets resulting from variations of α S [60], the MMHT2014 NNLO set, and the CT14 NNLO PDF set [118]. The scale uncertainties are estimated by comparing samples where the renormalisation and factorisation scales were modified by a factor of two. Finally, a comparison was made to a sample generated using MG5_aMC v2.2.2 interfaced to Pythia 8.186 and using the A14 tune together with the NNPDF 2.3 LO PDF set [49,119,120].
For the tt, single-top-quark, and W/Z+jets backgrounds, the acceptance differences that affect the relative normalisation across regions with a common background normalisation are estimated by summing in Table 4: A list of the signal and control regions (separated by commas below) included in the statistical analysis of the A and HVT model hypotheses. The notation 1+2 b-tag indicates the 1 and 2 b-tag regions are combined, and add. b-tag indicates the regions with additional b-tags not associated with the large-R jet.
Fit
Channel Resolved Merged Resolved signal regions signal regions control regions A 0-lepton 1, 2, 3+ b-tag 1, 2 b-tag, and 1, 2 b-tag add. b-tag -2- lepton 1, 2, 3+ b-tag 1, 2 b-tag, and 1+2 quadrature the relative yield variations between the different regions. These uncertainties are assigned to all regions used in the fit as shown in Table 4 and across the different lepton channels as shown in Table 3.
For the multijet background included in the 1-lepton channel, a 50% uncertainty in the normalisation is estimated from the fit to the E miss T distribution described in Section 6. Also, a shape variation is included to account for uncertainties in the determination of the template in the anti-isolated lepton region, arising from differences in the trigger scheme between isolated and anti-isolated regions and uncertainties in the normalisation of the top-quark and electroweak backgrounds in this region.
Finally, for the remaining small backgrounds only a normalisation uncertainty is assigned. For the diboson backgrounds a normalisation uncertainty of 11% is applied [121]. For the SM Vh, ttV, and tth production, a 50% uncertainty is assigned which covers the uncertainty in the cross-sections.
Results
In order to test for the presence of a massive resonance, the m T,Vh and m Vh templates obtained from the signal and background simulated event samples are fit to data using a binned maximum-likelihood approach based on the RooStats framework [122][123][124]. A total of five different fits are performed according to the signal interpretation: Z , W , HVT, A in gluon-gluon fusion, and A in b-quark associated production. The list of channels and regions used for the different fits is shown in Table 4.
The fits are performed on the m T,Vh distribution in the 0-lepton channel and the m Vh distribution in the 1and 2-lepton channels using a binning of the distributions chosen to optimise the search sensitivity while minimising statistical fluctuations. As described in Section 6, the normalisations of the tt, Z+(bb, bc, cc), and Z+(bl, cl) backgrounds are free parameters in all fits, as are the normalisations of W+(bb, bc, cc) and W+(bl, cl) in the W and HVT fits. The systematic uncertainties described in Section 7 are incorporated in the fit as nuisance parameters with correlations across regions and processes taken into account. The signal normalisation is a free parameter in the fit. In order to account for migrations of signal events across different channels due to lepton reconstruction and selection inefficiencies, the Zh → + − bb (Wh → ± νbb) signal samples are included in the 1(0)-lepton categories.
The total uncertainty in the signal yield is dominated by different sources of systematic uncertainty depending on the mass of the resonance used in the fit. The uncertainties in the W/Z+(bb, bc, cc) shape and normalisation, tt normalisation, and in the flavour tagging efficiencies constitute the dominant sources of systematic uncertainty for low-mass resonances. For all interpretations, the statistical uncertainty dominates for resonances above 1 TeV. The uncertainties in the large-R jet mass resolution and in the track-jet b-tagging efficiency constitute the dominant systematic uncertainties at high masses.
The expected and observed event yields after the HVT fit are shown in Table 5. The m T,Vh and m Vh distributions after the HVT fit are shown in Figures 2 and 3. Similar distributions are obtained from the W , Z and A fits with background yields consistent within the uncertainties. The mass distributions for the resolved 3+ b-tag category and the merged categories with additional b-tagged jets, used in the A boson fits, are shown in Figure 4.
As no significant excess over the background prediction is observed, upper limits at the 95% CL are set on the production cross-section times the branching fraction for each model. The limits are evaluated using a modified frequentist method known as CL s [125] and the profile-likelihood-ratio test statistic [126] using the asymptotic approximation.
The 95% CL upper limits on the production cross-section multiplied by the branching fraction for W → Wh and Z → Zh and the sum of branching fractions B(h → bb) + B(h → cc), which is fixed to 60.6% [47], are shown in Figure 5(a) and Figure 5(b) as a function of the resonance mass. The existence of W and Z bosons with masses m W < 2.67 TeV and m Z < 2.65 TeV, respectively, are excluded for the HVT benchmark Model A with coupling constant g V = 1 [25]. For Model B with coupling constant g V = 3 [25], the corresponding excluded masses are m W < 2.82 TeV and m Z < 2.83 TeV.
To study the scenario in which the masses of charged and neutral resonances are degenerate, a likelihood fit over all the signal regions and control regions is performed. The 95% CL upper limit on the combined signal strength for the processes W → Wh and Z → Zh, assuming m W = m Z , relative to the HVT model predictions, is shown in Figure 5 Figure 5(d) where all three lepton channels are combined, taking into account the branching fractions to Wh and Zh from the HVT model prediction. Here, the parameter c F is assumed to be the same for quarks and leptons, including third-generation fermions, and other parameters involving more than one heavy vector boson, g V c VVV , g 2 V c VVhh and c VVW , are assumed to have negligible contributions to the overall cross-sections for the processes of interest.
Figures 6(a) and 6(b) show the 95% CL upper limits on the production cross-section of the A boson times its branching fraction to Zh and the branching fraction of h → bb as a function of the resonance mass. Upper limits are placed separately for a signal arising from pure gluon-gluon fusion production ( Figure 6(a)) and from pure b-quark associated production ( Figure 6(b)). In the search for the A boson with b-quark associated production, a mild excess of events is observed around 440 GeV, mainly driven by the dimuon channel in the resolved category with 3+ b-tags. The local significance of this excess with respect to the background-only hypothesis is estimated to be 3.6 σ, and the global significance, accounting for the look-elsewhere effect [127] is estimated to be 2.4 σ.
The data are also interpreted in terms of limits at 95% CL on the 2HDM parameters tan(β) and cos(β − α). The admixture of gluon-gluon fusion and b-quark associated production, and the variation of the A and h boson widths and branching fractions are taken into account according to the predictions of the different models. In this interpretation, the m T,Vh and m Vh distributions of the simulated signal events are smeared according to a Breit-Wigner function with a width predicted by the parameters of the model. This procedure has been verified to produce the same line-shape as the one including non-resonant and interference effects for widths Γ A /m A < 10%. Figure 7 shows the excluded parameter space for a resonance mass of m A = 300 GeV in four 2HDM types: I, II, Lepton-Specific, and Flipped. Greater sensitivity is observed at high tan(β) for the Type-II and Flipped models, due to an increased cross-section for b-quark associated production. The narrow regions with no exclusion power in Type-I and Type-II at low tan(β) that are far from cos(β − α)=0 are caused by the vanishing branching fraction of h → bb. Figure 8 shows the parameter exclusion for the four models in the tan(β)-m A plane for cos(β − α) = 0.1. For the interpretation in Type-II and Flipped 2HDMs, the b-quark associated production is included in addition to the gluon-gluon fusion production. The shape of the expected exclusions is determined by the interplay of the expected cross-section limit, which decreases as a function of m A , and the signal production cross-section times the A → Zh branching fraction at a given m A and tan(β). This branching fraction decreases significantly at m A = 350 GeV due to the opening of the A → tt channel, but increases again at higher m A , maintaining similar sensitivity into this m A region. The variable tan(β) controls the admixture of the gluon-gluon fusion and b-quark associated production thereby affecting the rate at which the signal cross-section falls as a function of m A , which leads to a varying sensitivity as a function of tan(β). The excesses or deficits in the data visible in Figure 6 are also reflected in Figure 8. Table 5: The predicted and observed event yields in the signal regions defined in the text. The yields in the 1 and 2 b-tag regions correspond to the HVT fit for a signal of mass 1.5 TeV. In the 3+ b-tag and 1 and 2 b-tag with additional b-tags regions, the yields are from the fit using the A boson produced in association with b-quarks as signal with a mass of 1.5 TeV. The quoted uncertainties are the statistical and systematic uncertainties combined in quadrature after the fit. The uncertainties in the individual background predictions are larger than the total background uncertainty due to correlations in the normalisation parameters in the fit.
0-lepton
Resolved Merged Figure 5: Upper limits as a function of the resonance mass at the 95% CL for (a) the production cross-section of Z times its branching fraction to Zh and the branching fraction B(h → bb, cc) and (b) the production crosssection of W times its branching fraction to Wh and the branching fraction B(h → bb, cc). (c) Upper limits at the 95% CL for the scaling factor of the production cross-section for V times its branching fraction to Wh/Zh in = 13 TeV, 36.1 fb s 95% CL limit (b) Pure b-quark associated production Figure 6: Upper limits at the 95% CL on the product of the production cross-section for pp → A and the branching fractions for A → Zh and h → bb evaluated by combining the 0-lepton and 2-lepton channels. The possible signal components of the data are interpreted assuming (a) pure gluon-gluon fusion production, and (b) pure b-quark associated production.
Conclusion
A search for W and Z bosons and for a CP-odd Higgs boson A in the ννbb, ± νbb and + − bb final states is performed using 36.1 fb −1 of 13 TeV pp collision data collected with the ATLAS detector at the LHC. No significant excess of events is observed above the SM predictions in all three channels.
Upper limits are placed at the 95% CL on the cross-section times branching fraction, σ(pp → V → Vh) × B(h → bb, cc), ranging between 1.1 × 10 −3 and 2.8 × 10 −1 pb for the W boson and between 9.0 × 10 −4 and 1.3 × 10 −1 pb for the Z boson in the mass range of 500 GeV to 5 TeV. | 13,918.8 | 2017-12-18T00:00:00.000 | [
"Physics"
] |
Paracrine Regulation and Immune System Pathways in the Inflammatory Tumor Microenvironment of Lung Cancer: Insights into Oncogenesis and Immunotherapeutic Strategies
Simple Summary Despite massive strides taken across the board in oncology, there remain gaps in understanding the relationship between cancer cells and the body’s immune system, tissues, and signaling pathways. This review explores some of the recent steps made toward filling these gaps and understanding how certain cytokine signals create a tumor microenvironment that facilitates the growth and survival of cancer cells. New treatment approaches targeting these facilitators have been developed as potential disruptors of tumor growth. Lastly, a discussion of current gaps in the research can help navigate new directions to continue life-saving research for lung cancer treatments. Abstract The intricate interplay between inflammatory processes and the tumor microenvironment (TME) in lung cancer has garnered increasing attention due to its implications for both oncogenesis and therapeutic strategies. In this review, we explore recent advances in understanding the paracrine regulation and immune system pathways within the inflammatory TME of lung cancer. We delve into the molecular mechanisms underpinning oncogenesis, highlighting the role of immune cell populations, cancer-associated fibroblasts, and endothelial cells, as well as their interactions through immune system pathways regulated in a paracrine pattern. Additionally, we discuss emerging immunotherapeutic strategies with a specific focus on the potential of leveraging the inflammatory TME through these pathways to enhance treatment efficacy in lung cancer.
Introduction
Inflammation, a crucial component of the immune response, has dual roles in lung cancer-fighting acute infections and contributing to chronic conditions conducive to carcinogenesis.The tumor microenvironment (TME) encompasses immune cells, cancerassociated fibroblasts (CAFs), cancer cells, and endothelial cells that orchestrate paracrine signaling through growth factors, cytokines, and chemokines.Understanding the intricate interplay between inflammation, the immune response, and the TME is essential for unraveling the complexities of lung cancer development and devising effective therapeutic interventions.
The development of new blood vessels (angiogenesis) is crucial for supplying nutrients and oxygen to growing tumors [1].Endothelial cells within the TME engage in paracrine signaling with cancer cells and other stromal components [1].Factors such as vascular endothelial growth factor (VEGF) released by cancer cells stimulate endothelial cells to sprout and form new blood vessels [1,5].Additionally, VEGF suppresses T cell activity, increases the recruitment of Tregs and MDSCs, and inhibits DCs, contributing to antitumor immunity [16].This process is a hallmark of tumor progression and can be influenced by immune cells as well, further highlighting the complexity of paracrine interactions in the inflammatory TME [1].
Immune System Pathways
The immune system's intricate involvement in the TME of lung cancer is a dynamic interplay that shapes the tumor's fate, progression, and response to therapy.Immune cell populations, cytokines, chemokines, and immune checkpoints collaborate in a complex network of pathways that dictate whether the tumor is eliminated, contained, or allowed to thrive [3,17].
Chemokines guide immune cell trafficking to specific locations within the TME, promote EMT, and influence the distribution and behavior of immune cells [28,29].Specifically, chemokines are involved in the recruitment of tumor-associated neutrophils (TANs) to the TME [30].The polarization from the more antitumor N1 phenotype to the protumor N2 is likely mediated by cancer and immune cell cytokines like TGF-β [30,31].Moreover, they are involved in this complex crosstalk between the tumor stroma and cancer cells that make up the TME [28].
DCs represent one of the principal antigen-presenting cells in the immune system, typically involved in adaptive immunity T cell activation.However, in an inflammatory TME rich in IL-6, the subsequent activation of STAT3 pathways downregulates DC maturation.Other paracrine signals like IL-10 and VEGF also show inhibitory effects on DC function [32].Like Tregs, MDSCs represent another cell responsible for immunosuppression in the TME through their reduction in CD8+ T cells and promotion of T cell apoptosis through their production of TGF-β [33].Additionally, they play a role in suppressing the immune response against invading tumor cells, promoting EMT and stimulating angiogenesis, creating an environment susceptible to metastasis [33][34][35].NK cells represent the opposite end of the immune spectrum, more focused on innate immunity-mediated tumor cell death through the release of cytolytic enzymes or the expression of the apoptosis-promoting Fas ligand.However, metastasis-initiating cancer cells can auto-induce quiescence through Dickkopf WNT signaling pathway inhibitor (DKK)-1.This dormancy allows these metastasizing cancer cells to evade NK-cell killing and re-establish a growth state afterward [36].
Molecular Mechanisms of Oncogenesis
Exploring the interactions within the TME is critical to creating a thorough understanding of the tumor growth cycle needed in the current age of oncology.Exploring each cell in isolation helps identify unique or common pathways and molecules that they utilize to affect the cells around them and, ultimately, contribute to carcinogenesis.Various mechanisms that result in the chronic inflammation, hypoxia, EMT, immunosuppression, angiogenesis, proliferation, migration, and invasion of cancer cells are important to the understanding of carcinogenesis that goes beyond a unilateral image that focuses solely on genetic mutations.Moreover, understanding the involvement of paracrine signaling on the primary killer of tumor cells, the immune system, is crucial for identifying the gaps that allow cancer cells to grow unregulated and unchecked.
Chronic inflammation in the context of the TME of lung cancer is a dynamic process that contributes to genetic and epigenetic alterations in cancer cells, thereby facilitating carcinogenesis.Chronic inflammation creates an environment rich in proinflammatory cytokines and chemokines, which can enhance tumor progression and have direct effects on promoting cancer cell growth and survival (Table 1).
COX-2/PGE2
Cyclooxygenase-2 (COX-2) is an enzyme responsible for the formation of prostaglandins (PGs) and has been found to be a factor of proliferation and survival in lung cancer and can even be used as a marker for metastasis [37].Uniquely, COX-2 was found to be expressed at elevated levels in various lung cancer types and precursors and specifically in NSCLC [37,38].Understanding the correlation between COX-2 expression and cancer is multifaceted as it encompasses various mechanisms of survival: apoptosis inhibition, angiogenesis, immunosuppression, EMT, and invasion [37,39].An elevated COX-2 expression is linked with inflammation, IL-1β, TGF-β, EGF, mutant KRAS or TP53, hypoxia, the loss of IL-10 receptor expression, and constitutively localized STAT-6 [40].
PGE2 and its prostanoid receptor EP4 have specifically been identified to increase the matrix metalloproteinase (MMP)-2 and CD44 expressions and subsequent NSCLC tumor invasion and angiogenesis [39].These effects were found both in autocrine and paracrine forms as the overexpression of COX-2 enzymes consequentially results in increased tumor cell EP4 expression [39].Another mechanism of PGE2-mediated invasion in NSCLC includes the suppression of the E-cadherin expression through the induction of the transcriptional repressors ZEB1 and Snail, thereby reducing intercellular adhesion and enhancing tumor invasion and metastasis, contributing to its poor prognostic indication [41].
COX-2 and PGE2 upregulation are also involved in another important mechanism of survival: immunosuppression [42,43].Primarily through the increase in FOXP3 in both CD4+CD25+ (Treg) and CD4+ CD25-T lymphocytes that causes the latter to switch into a regulatory phenotype, PGE2 at high levels stimulates an immunosuppressive shift in the TME [42,44].Moreover, COX-2 is involved in the suppression of DC function as well, with a reduction in various surface molecule expressions that hinder their abilities to present antigens, induce alloreactivity, or secrete IL-12 [45].
COX-2 plays a role in increasing the apoptotic threshold of tumor cells in NSCLC through its stabilization and subsequently reduced ubiquitination of the molecule survivin, a spindle microtubule binding protein involved in apoptotic avoidance independent of the cell cycle [46].Moreover, COX-2 and PGE2 play roles in increasing the expressions of chemokines CXCL8 (IL-8) and CXCL5 via nuclear translocation NF-κB [43].Through the upregulation of this process, COX-2 has been found to increase tumor growth and angiogenesis in vivo [43].
TGF-β
Transforming growth factor (TGF-β) is part of a polypeptide family including 3 TGF-β isoforms, Nodal, bone morphogenetic proteins (BMPs), activins, and a few others found to be impaired in many cancer types including lung cancer, where it has antiproliferative functions [47].TGF-β is secreted from most cells in the TME, including fibroblasts, immune cells, and epithelial cells.Functioning through two transmembrane receptors, TGFRI and TGFRII, TGF-β induces downstream signal cascades through Smad2 and Smad3, which then phosphorylates Smad4, which translocates to the nucleus [48].This process undergoes feedback inhibition through Smad6 and Smad7, which recruit Smurf to disrupt the signaling at the receptor level [49].Non-canonically, the transduction may utilize NF-κB, PI3K, or MAPK, making the signaling stream complex and widespread in effect [50,51].TGF-β has been identified to play two roles in tumors, one of tumor suppression early on and later, a more tumorigenic role in mediating EMT and other aspects of the TME [48,50,51].This EMT is also potentially subject to crosstalk with TNF-α, which enhances the EMT by causing cancer cells to switch to a cytokine-and chemokine-secreting phenotype, increasing the capacity for invasion [52].Moreover, CAFs also play an integral role in this process through their secretion of IL-6 under TGF-β-mediated fibrosis [53,54].
The cytostatic function of TGF-β is mediated by its induction of the CDK inhibitor (i.e., p15, p21) expression, reduction in c-Myc expression, and several other mechanisms downstream of Smad, which together help arrest the cell cycle in the G1 phase.In terms of its protumor function, TGF-β plays an important role in the function of TANs, inducing the immunosuppressive N2 polarization, which inhibits NK cells, recruiting Tregs and antiinflammatory M2 macrophages, and releasing angiogenic MMP9 [30,31].In the absence of this signal, the N1 polarization can target tumor cells directly or through their influence on other proinflammatory immune cells [31].Moreover, more recent research has focused on its role in the protumor metabolic transitions of the TME including CAFs, immune cells, and cancer cells themselves.For example, by promoting "reverse Warburg effects" in CAFs, whereby, a high rate of aerobic glycolysis utilized to transport metabolites to cancer cells demonstrated a key function of TGF-β and a potential explanation for the survival of cancer cells during metastasis before angiogenesis can take place [51,55].
BMPs, members of the same family, have also been identified to be involved in the tumorigenic process.Namely, BMP-2, BMP-4, and BMP-7 are correlated with poor prognosis in lung cancer through their induction of angiogenesis and tumor growth [56].
EGF
Epidermal growth factor receptor (EGFR)-mediated activations of MAPK, PI3K, and STAT3 signaling pathways demonstrate important mechanisms by which various ligands (i.e., EGF, TGF-α, amphiregulin, etc.) influence tumor growth, survival, and metastasis [57][58][59].Notably, upregulated EGF and amphiregulin are associated with poor prognosis in NSCLC including malignant metastasis and resistance to targeted inhibitors as part of an intricate autocrine EGFR growth loop [57,60,61].In terms of apoptotic evasion in NSCLC, amphiregulin inactivates the proapoptotic BAX, while the inhibition of EGFR expression and activity see an upregulation in the proapoptotic markers and decreases in the tumorigenic markers [60,62,63].Specifically, EGF has an ability to stimulate the production of angiogenic factors like VEGF, bFGF, and hypoxia-induced factor 1 (HIF-1) downstream of its MAPK or PI3K cascades [56].
Moreover, EGF's involvement in tumorigenesis extends to tumor migration as well.While the focus is generally on TGF-β mediated Smad signaling, there is some involvement of EGF-mediated MAPK signaling to instigate migration; however, in A549 lung adenocarcinoma cells, EGF had no effect on the EMT marker's MMP2 expression, and EGF was not responsible for significantly increasing the invasive potential of these cells, highlighting the importance of TGF-β to EMT and invasion as discussed [64].However, EGFR may participate in immune evasion in NSCLC through a potential regulation of the B7-H5 expression [62].Additionally, the activation of this pathway is involved in inducing PD-L1 expression as EGFR can stimulate the IL-6/JAK/STAT3 pathway in NSCLC cells [65].
FGF
Another important growth factor to be discussed in its role in lung cancer is fibroblast growth factor (FGF).One such member of this family specifically identified in high levels in NSCLC tumors is the basic fibroblast growth factor (bFGF/FGF2) [66].Utilizing its receptor FGFR, FGF signals to several key downstream pathways, including MAPK, PI3K, PLC, and STAT, promoting several tumorigenic mechanisms.While several mechanisms of the involvement of FGF exist across different cells of the TME, in lung cancer, FGF may play a potential role in inducing IFN-γ-mediated PD-L1 expression, diminished T cell infiltration, Treg generation, and T cell depletion [67].Moreover, the upregulation in FGFR signaling plays an important role in tumor growth by promoting the survival of MDSCs, recruitment of TAMs, EMT, and angiogenesis [67,68].The role of FGF is particularly important in the understanding of CAFs in the TME, which overexpresses FGF9 and bFGF and secrete bFGF in significant amounts that contribute to tumor cell growth in lung adenocarcinoma cell models [68].FGF may also play a role in angiogenesis as an upstream stimulator of VEGF and PDGF as well as a regulator of newly formed vessels whereby they have the potential to activate Bcl-2 and inhibit apoptosis in endothelial cells or MAPKs [69].
Factors of Angiogenesis: VEGFA, HIF-1α, CSF, and PDGF
Angiogenesis is a crucial process for tumor cell growth, nutrient acquisition, metastasis promotion, and inflammation and is thus an important aspect to understand [56].Several stromal cells are deeply involved in this process, including macrophages, which secrete angiogenic growth factors and degrade the perivascular ECM, neutrophils, which are important for triggering angiogenesis, lymphocytes, which can be anti-or pro-angiogenic, and CAFs, which produce the new ECM and more angiogenic factors.Overall, the TME may also take on different metabolic properties in order to provide energy to fuel angiogenesis [70].
One important mechanism for this process in lung cancer utilizes overexpressed vascular endothelial growth factor (VEGF) to drive the formation of blood vessels in and around the tumor [56].The main angiogenic VEGF, VEGF-A, utilizes two tyrosine kinase receptors, VEGFR-1 and VEGFR-2, and through their signaling, VEGF can stimulate angiogenic signals in endothelial cells as well as the secretion of von Willebrand factor via VEGFR-2 and a weaker pro-angiogenic activity from VEGFR-1 [56,71].VEGFR-3 and VEGF-C/D may also be involved in tumor angiogenesis [71].Like VEGF, platelet-derived growth factor (PDGF) receptors have also been found to be overexpressed in lung cancer cells and linked to a poor prognosis [56].The regulation of VEGF through other mechanisms has also been explored in NSCLC as p53 and Bcl2 were significantly associated with VEGF expression; however, a recent study demonstrated no difference in survival regarding the Bcl2 and VEGF statuses for patients with advanced NSCLC despite the identified correlation [72,73].
As the solid tumor expands, the hypoxic conditions, due to a lack of blood supply, promote cancer cell survival and angiogenesis by activating HIF-1α, inducing anaerobic glycolysis, inhibiting apoptosis, and stimulating the expressions of pro-angiogenic factors like VEGF.This alters the TME to support tumor growth and immune evasion [74].Notably, elevated levels of HIF-1α, VEGF, and CCL28 are linked to a greater infiltration of Treg cells in lung adenocarcinoma samples, and high levels of HIF-1α and HIF-2α are associated with poor prognoses in both SCLC and NSCLC [75,76].Additionally, hypoxia's impact on immune chemokine secretion in SCLC tumor cells is known to attract immunosuppressive cells [75].
Another group of cytokines, colony-stimulating factors (CSFs), which typically focus their effects on white blood cell proliferation and differentiation, also aid in angiogenesis.The four main types, G-CSF, GM-CSF, M-CSF, and IL-3, function on different cell types, and each may play a role in lung cancer angiogenesis.G-CSF and GM-CSF are particularly associated with aggressive and angiogenic lung tumors [56,77].Other direct effects of G-CSF include promoting survival, proliferation, and migration in cancer cells, stimulating M2 polarization, and increasing MDSC and Treg phenotypes [77].Interestingly, G-CSF has also been found to promote metastasis through its mobilization of Ly6G+Ly6C+ granulocytes to other sites prior to tumor cell arrival.These granulocytes produce Bv8 protein, which may contribute to premetastatic angiogenesis [78].
Other Cytokines
Under the influence of chronic inflammation, proinflammatory cytokines and chemokines such as IL-6, IL-17, and IL-23 secreted by immune cells stimulate fibroblasts within the TME, promoting cancer cell growth and survival [79].These CAFs can then actively participate in paracrine signaling.Fibroblasts secrete growth factors, such as FGF and TGF-β [80].These growth factors have the capacity to then stimulate cancer cell proliferation, promote EMT, and enhance tumor invasiveness as discussed.Further, the presence of integrin α11β1, a receptor for collagen XIII on CAFs, governs ECM stiffness, IGF-2 secretion, and, consequently, metastasis and NSCLC tumor growth.Integrin α11β1 influences lysyl oxidase-like 1 (LOXL1), an ECM cross-linking enzyme critical for tumor growth and invasion [81].
Various other cytokines exist that, depending on their secretion and target cells, can produce varying cancer-promoting effects.One such cytokine is IL-1β, a ligand for the IL-1R that produces in vivo and in vitro carcinogenic changes downstream of MAPK and NF-κB that promotes immunosuppression, angiogenesis, and the invasion of cancer [3,82,83].Mainly expressed in innate immune cells (i.e., monocytes, macrophages, and DCs), IL-1β demonstrates a significant facilitator in lung cancer metastasis and growth through these paracrine interactions [3,83].For example, IL-1β promotes EMT in cancer cells, PG production and leukocyte adhesion on endothelial cells, proteinase production and COX-2-mediated angiogenesis in stromal cells, immunosuppression through induced MDSCs, as well as protumor cytokine production, like that of IL-22 [82,84].Additionally, IL-1β can stimulate VEGF, COX-2, and TGF-β to drive tumorigenesis through the angiogenic, apoptosis-resistant, and immunosuppressive mechanisms discussed for each of these factors [82].However, IL-1β upregulation may have the additional effect of inducing mutations in TP53 that are associated with NSCLC and elevated IL-1β expressions [82,85].
IL-4, released by Th2 lymphocytes, is involved in regulating the immune response and has been found in high expressions in lung cancer [86].IL-8, also known as CXCL8, is another proinflammatory cytokine that dominates inflammation signaling via NF-κB whereby it recruits MDSCs and immunosuppressive neutrophils and can also stimulate EMT and promote angiogenesis [86].Finally, TNF-α is a key activator of NF-κB, and its effects on apoptotic evasion and cell survival will be discussed below.
Interferons are cytokines that can be divided into three types: type 1, which includes IFN-α, IFN-β, IFN-ε, IFN-κ, IFN-τ, and IFN-ω; type 2, which is IFN-γ; and type 3, which contains IFN-λ1, IFN-λ2, IFN-λ3, and IFN-λ4 [87].Type 2, IFN-γ, has been found to be associated with tumor progression.Antigen-specific T cells that accumulate in the TME cause an increase in the IFN-γ concentration in the tissue.The IFN-γ can cause an increased expression of PD-L1 in lymphatic endothelium, which further limits cytotoxic T cell access to the TME [88].IFN-γ is also associated with IDO upregulation.IDO is an enzyme that can suppress T and NK cells and is proposed to be freed up by an IFN-γ-stimulated macrophage degradation in tryptophan, allowing it to inhibit T and NK cells [89,90].
Key Intracellular Signals
One of the key transcription factors activated by chronic inflammation in the TME is NF-κB.The activation of this factor can involve two different pathways: canonical and non-canonical.The canonical pathway of NF-κB is active during responses involving inflammation and immune responses.It is particularly critical in modulating innate immunity [102].On the other hand, the non-canonical pathway for activating NF-κB is required for lymphoid organ development as well as adaptive immunity [103].The canonical pathway is initiated by triggers like proinflammatory cytokines TNF-α and IL-1, as well as bacterial substances such as lipopolysaccharide (LPS).These triggers activate IκB kinases (IKKs), which subsequently phosphorylate the primary inhibitor of NF-κB: IκBα.This phosphorylation step results in the ubiquitination and subsequent degradation in IκBα by the proteasome.The NF-κB complex then moves to the nucleus, where it binds to κB enhancers in the regulatory regions of various genes, initiating transcription [104].NF-κB target genes have diverse roles encompassing functions not limited to proliferation, survival, or angiogenesis [105].
With regard to NF-κB's effect on cell proliferation and survival in lung cancer, NF-κB signaling activates cyclin D1 by binding to its promoter, promoting cell proliferation [97].Cyclin-dependent kinases (CDKs) are key in regulating the cell cycle, as they interact with cyclin proteins [106].Cyclin D1 binds to CDK6 and CDK4, causing Rb protein phosphorylation.This prevents Rb from inhibiting the E2F family transcription factors, facilitating the transcription of numerous genes required for the G1-to-S phase transition, and ultimately promotes cellular proliferation [107].
The Bcl-2 protein family, a group of proto-oncogenes, negatively regulate apoptosis and are often dysfunctional in various cancer types [98].The human Bcl-2 promoter and inhibitors of apoptosis (IAPs) have been found to contain an NF-κB binding site, providing resistance to apoptosis induced by TNF-α [108].Consequently, NF-κB activation in cancer cells during chemotherapy or radiation therapy is primarily associated with apoptosis resistance, a significant obstacle to effective cancer treatment [109].In cancer cells, the NF-κB signaling pathway promotes angiogenesis, which involves regulating key pro-angiogenic factors such as VEGF and the proinflammatory cytokine IL-8 [99].
The STAT proteins regulate many aspects of growth, survival, and differentiation in cells.Ref. [110] STAT3 signaling is frequently overactive in most human cancers and serves as a recognized intrinsic pathway that promotes inflammation, cellular transformation, survival, proliferation, invasion, angiogenesis, metastasis, and immune evasion in cancer [100,101,111].In addition, IL-6 activates STAT3 and is associated with advanced-stage disease and reduced survival in cancer [112].There are also collaboration and crosstalk that occur between NF-κB and STAT3 in cancer progression, and efforts to target these pathways have yielded promising outcomes in cancer treatment [100,110,113].
The activation of STAT3 in macrophages and neutrophils is essential for safeguarding against chronic inflammation [114].Mutant mice lacking this activation exhibited heightened susceptibility to endotoxin shock, resulting in elevated levels of inflammatory cytokines, including TNF-α, IL-6, IL-1β, and IFN-δ in their serum [110].
Immunotherapeutic Strategies
Lung cancer can be subdivided into certain types.The most common division seen in the literature is NSCLC and SCLC.NSCLC has been found to have several common oncogenes, such as KRAS, c-MET, ALK, RET, BRAF, ROS1, NTRK, TP53, and ERBB2.Certain genomic alterations will allow for T cell-specific adaptive immune responses.The genomic alterations will cause a change in the APC-mediated antigen presentation.Immunotherapies focus on assisting the immune system in fighting the cancer cells.An important aspect of immunotherapies relating to cancer is the TME, which features a combination of ECM, fibroblasts, mesenchymal cells, and immune cells, among others.
Immune Checkpoint Inhibitors
TMEs contain immune checkpoints, which are receptor/ligand interactions that suppress the T cell response.Immune checkpoints are generally found in healthy cells; however, certain tumors express the checkpoint, preventing a T cell-mediated response.An important therapy that has emerged for NSCLC is the immune checkpoint inhibitor, which would stop the interaction of the T cells with immune checkpoints, including CTLA-4 and PD-1/PDL-1 [115].Multiple studies have shown the therapeutic value of anti-CTLA-4, anti-PD-1/L1 and their combination, with or without chemotherapy in metastatic or advanced NSCLC (Table 2) [116][117][118][119][120][121][122][123].
CAR T Cell Therapy
Chimeric Antigen Receptors (CARs) are synthetic receptors that are transduced onto T cells.These receptors have activation sites that allow for increased T cell responses, often with costimulatory binding sites to modulate responses.EGFR-targeting CAR T cells are undergoing multiple clinical trials.In an ongoing phase 1 trial (NCT0415379; Table 2), 11 patients with EGFR-positive tumors are being administered anti-EGFR-modified CAR T cells.In a recent phase 2 trial (NCT01869166), patients with advanced NSCLC with an over 50% EGFR expression received the anti-EFGR CAR T cell treatment, and the patients were able to tolerate the therapy without severe toxicity for 3-5 days at a time [124].Similarly, other tumor-specific molecules can be primed to the CAR T cells, allowing for patientspecific treatment.Further examples of such molecules include CEA, MUC1, MSLN, PD-L1, ROR1, and HER2 [124].
CAFs, TAMs, and TANs
CAFs, TAMs, and TANs all play important roles in tumorigenesis and subsequent immunotherapy.TAM will secrete cytokines that are involved in angiogenesis and tumor invasion along with immunosuppression [125].CAFs have been shown to suppress the immune system through CTLA4 upregulation, inhibiting CD8+ T cells [126].TANs have been found to secrete PGE2, which promotes NSCLC cell proliferation, causing increased tumor growth.Another important aspect of neutrophil involvement is ROS damage causing pre-disposed cells to go through an oncogenic transformation [127].
Several immunotherapeutic strategies have been proposed to use these cells as a means of curtailing tumor growth.IFN therapies have shown an increase in neutrophil antitumor therapy by way of ICAM1 upregulation.Increasing ICAM1 would increase immune activity [128].A phase 1 clinical trial (NCT02001974) has shown the benefit of inhibiting the CXCR-1 and CXCR-2 chemokines in patients with triple-negative breast cancer using Reparixin in tandem with paclitaxel [129].These chemokines recruit neutrophils to the tumor, creating the TAN and its subsequent protumor effects.An ongoing phase 2 study (NCT02370238) is evaluating the progression-free survival (PFS) of patients with TNBC treated with Repaxirin.There are currently no ongoing trials on anti-TAN therapy for patients with lung cancer.However, there are promising RNA vaccine therapies being explored that can use TAN-specific molecules as targets [130].
Some of the approaches taken to use CAF inhibition to curtail cancer growth include using CTLA-4 antibodies to block the effects of CAF and using NOX4 inhibition to prevent myofibroblast activation by inhibiting TGF-β activity.A phase 3 trial that evaluated antibodies that targeted TGF-β and PD-L1 in lung cancer was discontinued.Another important approach is the inhibition of FAP, or a fibroblast-activating protein using primed CAR T cells.A recent phase 1 clinical trial (NCT01722149) evaluated four patients with metastatic pleural mesothelioma and found that intrapleural injections of FAP CAR T cells increases proinflammatory cytokines in the sera with minimal upper respiratory infections and thromboembolic events [131].
TGF-β is involved in the differentiation of CAF, and Galunisertib is a TGF-β inhibitor [131].A phase 1b/2 clinical trial (NCT02423343) featured 41 participants with advanced solid tumors and recurrent NSCLC or HCC.Researchers found that the MTD of galunisertib was 300 mg in phase 1b.In phase 2, the results showed that no patients were found to have anti-nivolumab antibodies after being administered galunisertib.The ORR was 24% in NSCLC patients, and the median PFS was found to be 5.26 months [132].
Several therapies target TAMs, such as sitravatinib, cabozantinib, bemcentinib, BA30011, and INCB081776 [133].Sitravitinib targets receptor tyrosine kinases, including those that are found on the TAMs.In a phase 1 trial, sitravitinib was found to shift TAMs to an immunostimulatory state, increasing the ratio of M1 to M2 macrophages found in the TME [134].In a phase 2 study on the same drug along with nivolumab in NSCLC patients with prior CPI therapy, the drugs were found to be clinically active together [135].However, in the subsequent phase 3 trial, SAPPHIRE (NCT03906071), the combination of sitravatinib plus nivolumab did not improve survival when compared to docetaxel in patients with previously treated advanced nonsquamous NSCLC [136].
Cabozantinib is also a tyrosine kinase inhibitor that has been implicated in various cancers such as medullary thyroid cancer, hepatocellular carcinoma, and RCC [133].
The phase 3 CONTACT-01 trial compared cabozantinib with atezolizumab against docetaxel in NSCLC patients who had received platinum-based chemotherapy and CPI (NCT04471428) [133,137].However, the study did not meet its primary endpoint of overall survival at the final analysis [138].
Oncolytic Viruses
Genetically modified viruses can be used for oncolytic purposes, resulting in inhibitory effects on tumor development [139].The viruses will have a tropism for specific targets within cancer cells such as PSA or COX2, or even surface markers such as EGFR and CD20 [140].There have been few clinical trials conducted on human subjects.The lysogenic adenovirus was found to result in extended disease progression in a phase 2 study (NCT01574729) that looked at rAd-p53 gene therapy combined with surgery on NSCLC.The study comprised patients in stage III or IV NSCLC who received either a combination of the virus targeting the p53 gene (intervention arm) and chemotherapy through the bronchial artery or chemotherapy alone (control arm).The intervention arm was found to have an extended disease progression period, with an MS of 7.7 months as opposed to 5.5 months, p = 0.018.Two of the patients exhibited complete responses to the combinatorial treatment, both of whom had stage III NSCLC [141].
Tumor-Infiltrating Lymphocytes (TILs)
TILs are lymphocytes that are combined with specific T cell clones of tumor antigens, allowing for specialized tropism to the tumor.The approach consists of taking lymphocytes from the tumor and causing ex vivo proliferation using IL2 [142].There have been several studies conducted on TILs in patients with lung cancer.An early study on TILs was performed on 131 stage II and III patients that had undergone a resection of NSCLC [143].Ex vivo recombinant IL-2 was used to proliferate the extracted lymphocytes.The intervention arm of the study was found to have a higher MS than the control arm, 22.4 months as opposed to 14.1 months.A 3-year survival was found to be better for patients who underwent the TIL therapy (p < 0.05).TIL therapy has also been found to be effective in PD-1-resistant lung tumors in a phase 1 clinical trial (NCT03215810).A total of 20 patients with advanced NSCLC were administered autologous TIL along with nivolumab, and 11 patients were found to have a reduction in tumor burden (NCT03215810) [144].
IL-1β
IL-1β has been found to increase metastasis in lung cancer through angiogenesis, tumor epithelial-to-mesenchymal transition, adhesion, growth invasion, and cytokine production [3,[145][146][147].IL-1β has also been found to polarize M2 macrophages, increasing immune suppression and angiogenesis [148,149].Currently, the major clinical program looking at IL-1β therapy in lung cancer is CANOPY (Canakinumab Outcomes in Patients with NSCLC Study).Canakinumbad is an FDA-approved human monoclonal antibody that targets IL-1β for acute systematic juvenile arthritis and periodic fevers [3].The CANOPY program consists of six different trials.Half of the trials study the combination of anti-PD-1 and anti-IL-1β, which has shown an increase in CD8+ cytotoxic T cell infiltrate in the tumor [84,150].Although the currently completed CANOPY trials did not yield positive results in the primary endpoints, they did demonstrate significant improvement in patient biomarker-based subgroups [3].
NF-KB
One of the major therapies that have been approved by the FDA that specifically targets VEGF by way of NF-KB is bevacizumab [102].In the E4599 phase 3 trial, 15 mg/kg of bevacizumab in addition to carboplatin/paclitaxel was found to improve the median OS in patients with NSCLC, as opposed to chemotherapy alone.The trial yielded a hazard ration of 0.79 (p = 0.003), and a median OS of 12.3 months as opposed to 10.3 months with chemotherapy alone [151].In the AVAiL trial, another large stage III randomized study, bevacizumab at 7.5 and 15 mg/kg was evaluated in addition to cisplatin/gemcitabine against chemotherapy alone in patients with NSCLC.The trial was found to increase PFS from 6.1 to 6.7 months in the 7.5 mg/kg dosage group (HR 0.75, p = 0.003) and to 6.5 months in the 15 mg/kg group (HR = 0.82, p = 0.03) [152].Another important therapy that targets NF-KB is bortezomib.A phase 2 clinical trial (NCT00075751) examined the effects of bortezomib along with gemcitabine/carboplatin together in patients with stage IIIB/IV NSCLC.The study found that that median OS was 11 months, (95% CI: 8.2-13.4months), and the median PFS was 5 months (95% CI: 3.5-5.3months).Out of the 113 patients evaluated for safety, 3 patients had grade 3 hemorrhages, and 1 patient had febrile neutropenia [153].
IL-6
There are two current trials that are evaluating the efficacy of tocilizumab in lung cancer.Tocilizumab, an anti-IL-6 receptor antibody, has been found to improve cachexia in lung cancers that overexpress Il-6 in mice [154].The first trial, NCT04940299, is a phase 2 clinical trial that is studying the effects of tocilizumab, nivolumab, and ipilimumab in patients with NSCLC, urothelial carcinoma, and melanoma.The other trial, NCT04691817, is a phase 1/2 trial examining tocilizumab in combination with atezolizumab in patients with NSCLC that is either locally advanced or that has metastasized and has not responded to treatment (Table 2).
STAT3
A phase 1 trial of OPB-51602 (NCT01184807) was found to achieve partial responses in two patients with NSCLC [155].However, there are no substantive human trials currently that have assessed the relationship between STAT3 inhibition and lung cancer.A promising trial that is currently underway is a phase 2 trial looking at danvatirsen and durvalumab in patients with advanced and refractory pancreatic cancer, NSCLC, and colorectal cancer.Danvatirsen (AZD9150) is an inhibitor of STAT3 (NCT02983578).
TNF-α
Certozilumab is an important inhibitor of TNF-α that underwent a phase 1 trial for patients with stage IV lung adenocarcinoma.The study involved chemotherapy combined with certolizumab to evaluate the toxic effects of the drug (NCT02120807, Table 2).The trial found that the standard dose of 400 mg of certolizumab was well tolerated and had potential for further study [156].
IL-8
There is one ongoing trial that is examining the effect of administering nivolumab with anti-IL-8 versus a CCR2/5 inhibitor both before surgery and after surgery in patients with NSCLC or hepatocellular carcinoma (NCT04123379, Table 2).
IL-10
Pegilodecakin, or recombinant IL-10, has undergone two major phase 2 randomized controlled trials to test its effectiveness in relation to NSCLC.In CYPRESS 1 (NCT03382899, Table 2), the control arms were administered pembrolizumab, and in CYPRESS 2 (NCT03382912, Table 2), the control arms were administered nivolumab, while the experimental arms were administered pegilodecakin with the respective checkpoint inhibitors.CYPRESS 1 featured a median PFS increase from 6.1 to 6.3 (hazard ratio of 0.937), and a median OS of 16.3 months as opposed to not reached (hazard ratio = 1.507).CYPRESS 2, however, had a median PFS of 1.9 in both arms, and a median OS of 6.7 months in the experimental arm vs. 10.7 in the control arm [157].
Challenges and Future Directions
Despite the various mechanisms of immunotherapy at play in trials on lung cancer, there remain many challenges to address on the road to the overwhelming goal of curing cancer.In particular, the patient-specific differences in response to the same drug create a lack of efficacy, as well as the various resistance mechanisms at play and the difficulty in bringing scientific success into the clinic.Going forward, these challenges need to be addressed across all fronts.An expansion of known biomarkers will help understand case-by-case variations in responses, while a stronger understanding of the immune system will open the door to how it can best be utilized and modified.Lastly, a stronger focus on combinational therapy and more improved treatment directives will help create more durable clinical successes and improve the standardization of future research.
Case-Specific Variations in Efficacy
One of the most pressing challenges in the treatment of many cancer types, particularly lung cancers, is that the current approaches developed have not been able to yield a broadly applicable treatment.Within this challenge exists various proposed causes: variations, previous treatment history, inherent immunosuppressive features of malignancy, type and stage of growth, TME heterogeneity, and pathways for malignancy [158].
Because of how specific current research is on the direct inhibition of single molecules, the efficacy of these treatments is limited to a small population, and even within that population, there is no guarantee of success.As mentioned, ICBs are common tools in the repertoire of oncologists treating cancer; however, their success has not been universal, given the heterogeneity of immune regulation systems and the difficulty in targeting them [115,159].Their lack of use as first-line therapy in many cancers also represents difficulties when discussing their efficacy as they are commonly administered after chemotherapy in patients with impaired immune responses [158,159].Moreover, as the literature continues to shift toward the discussion of various combinational therapies, it will become increasingly difficult to identify which of these is the most recommended for a particular patient.The validity of immunotherapy is quite strong, particularly in NSCLC with brain metastases, where immune checkpoint inhibitor treatments produced longer PFSs and OSs, and these results are improved only when combined with chemotherapy [160].However, it is important to note that across the spectrum of lung cancer and particularly NSCLC, the number of patients qualifying for each therapy or responding strongly to it is highly variable.
Resistance Mechanisms
Many of the mechanisms discussed as part of proliferative or survival pathways are potentially active in resistance against immunotherapeutics.For example, the creation of subclonal tumor cells that do not express neoantigens dampens the possible response to immunotherapies dependent on T cell cytotoxicity [161].Further, heterogeneity in the mechanisms responsible for growth, antigens expressed, deficiency in antigen presentation, low tumor mutation burden (TMB), and PD-L1 expression are all possible mechanisms of intrinsic resistance.As identified in Figure 1, there are also various TME-related immune modulations that can all pose possible extrinsic resistance mechanisms.For example, host immunosuppressive cells being recruited or activated, T cell exhaustion, cytokine or chemokine alterations, and increases in immune surveillance avoidance mechanisms (i.e., increased PD-1 expression) all contribute to the heterogeneity of resistance against immunotherapy [157,161].Along these lines, SCLC has been particularly difficult to treat with immunotherapy.Despite data from clinical trials of checkpoint inhibitors providing hints at improved long-term survival, the immunosuppressive TME, with low PD-L1 and MHC antigen expressions as well as avascularity that restricts immune reach, has created an environment resistant to immune modulation [162,163].Trying to understand these changes and interactions should be at the forefront of the oncological approach, as without an adequate understanding of these mechanisms, the same pitfalls against treatment will continue to dampen the efficacy of immunotherapy.
to treat with immunotherapy.Despite data from clinical trials of checkpoint inhibitors providing hints at improved long-term survival, the immunosuppressive TME, with low PD-L1 and MHC antigen expressions as well as avascularity that restricts immune reach, has created an environment resistant to immune modulation [162,163].Trying to understand these changes and interactions should be at the forefront of the oncological approach, as without an adequate understanding of these mechanisms, the same pitfalls against treatment will continue to dampen the efficacy of immunotherapy.
Bringing Scientific Success into the Clinic Going Forward
Currently, the PD-L1 expression is the most widely used biomarker in the clinic; however, its value as a predictive marker suffers at the hands of tumor heterogeneity as well as differences in defining those variable expression levels [162,164].Several potential biomarkers in various stages of development propose a possible answer to this issue.As possible alternatives to the standard immunohistochemical methods of analyzing PD-L1 expression, turning more toward gene expression-based, TMB, CBC, peripheral blood mononuclear cell (PBMC), TIL, extracellular vesicle, imaging, and microbiome biomarkers may provide a more complete picture of the tumor and its relationship in immunotherapeutic treatment.
Another promising approach that is now at the forefront of many clinical trials and studies includes combinations of immunotherapy with a plethora of other treatment types.For example, the double barricade approach of combining checkpoint inhibitors like anti-PD-L1 and anti-CTLA4 demonstrates a potential future for immunotherapy as each inhibitor's unique abilities to modulate the TME propose a more durable protection against TME-mediated resistance [165].Other combinations like those with DNA repair targeting agents, targeted agents (i.e., G12C inhibitors, EGFR inhibitors, etc.), chemoradiation, chemotherapy, and novel checkpoint inhibitors that are currently undergoing trials also represent a new frontier in immunotherapeutics aiming for a more complete treatment approach [162].In any of these approaches, there exists a need for improved clinical trial designs that take into account the likelihood of a delayed immune response, baseline immune statuses, and the difficulty in expanding qualified cohorts [159].
Conclusions
The inflammatory TME plays a central role in lung cancer development, fostering an environment conducive to oncogenesis while also shaping the efficacy of immunotherapies.Recent advances in our understanding of paracrine regulation, immune pathways, and molecular mechanisms offer valuable insights for devising novel strategies to combat lung cancer.Despite the current difficulties in maintaining efficacious treatments across a larger population and addressing the hypervariable resistance mechanisms that come with them, the future of immunotherapy is anything but bleak.By developing a more thorough understanding of the intricate interplay between inflammation, immune response, and the TME, we can potentially enhance the effectiveness of immunotherapeutic interventions, paving the way for improved outcomes in lung cancer treatment.
Figure 1 .Figure 1 .
Figure 1.Paracrine regulation and immune system pathways in the TME.APCs (e.g., DCs, macrophages) sample antigens from cancerous cells and present the antigen on MHC class I to CD8+ T cells and MHC II class 2 to CD4+ T cells.After activation by the APC, CTLs and Th cells migrate to the site of inflammation and induce a cytotoxic response directly (CD4+ and CD8+) or indirectly through other effector cells (CD4+).CAFs in the TME stroma have various functions including the Figure 1.Paracrine regulation and immune system pathways in the TME.APCs (e.g., DCs, macrophages) sample antigens from cancerous cells and present the antigen on MHC class I to CD8+ T cells and MHC II class 2 to CD4+ T cells.After activation by the APC, CTLs and Th cells migrate to the site of inflammation and induce a cytotoxic response directly (CD4+ and CD8+) or indirectly through other effector cells (CD4+).CAFs in the TME stroma have various functions including the recruitment of immunosuppressive cells (e.g., Tregs and MDSCs), increasing the expression of immune checkpoint receptors, facilitating EMT, remodeling the ECM, and angiogenesis.The induction of immuno-suppressive cell phenotypes through cancer cellmediated pathways results in a reduction in anti-inflammatory immune function.Cancer cells also secrete VEGF, which stimulates proliferative angiogenesis as it interacts with ECs.TAMs are distinguished into two main subclasses: M1 TAMs are involved in proinflammatory antitumor reactions through cytokines like IL-12, whereas M2, is associated with tissue repair and thus is anti-inflammatory through IL-10.M2 polarization through IL-4 commonly occurs in lung cancers and contributes to various mechanisms of tumor growth.Cancer cells themselves exhibit changes in expressions including an increase in ICMs like PD-L1 and a reduction in MHC class I expression, therefore diminishing CTL recognition.Chemokines are involved in immune cell trafficking within the TME and, specifically, with the recruitment of TANs.Subpopulations of TANs include the antitumor N1 and protumor N2, which is favored in a proliferative TME with the production of TGF-β.Several other cytokines and chemokines produced by cancer cells are involved in the inflammatory environment as well.Red arrows denote cancer cell function.Black arrows denote stromal cell functions.Arrowheads represent stimulation or increases, while flat-heads represent inhibition or reductions.(Created using BioRender.com).Abbreviations used: APC = antigenpresenting cell; Th = helper T cell; CTL = cytotoxic T lymphocyte; Treg = regulatory T cell; MDSC = myeloidderived suppressor cell; NK = natural killer cell; ICM = immune checkpoint molecules; EMT = epithelialmesenchymal transition; ECM = extracellular matrix; CAF = cancer-associated fibroblast; TAN = tumorassociated neutrophil; TAM = tumor-associated macrophage; IL-# = interleukin-#; TGF-β = transforming growth factor-beta; VEGF = vascular endothelial growth factor; GM-CSF = granulocyte macrophage colonystimulating factor; CXCL = chemokine (C-X-C motif) ligand; MCP = monocyte chemotactic protein.
Table 2 .
Ongoing clinical trials exploring regulation of TME. | 9,706.2 | 2024-03-01T00:00:00.000 | [
"Medicine",
"Biology",
"Environmental Science"
] |
Speed-Accuracy Tradeoffs in Tagging with Variable-Order CRFs and Structured Sparsity
We propose a method for learning the structure of variable-order CRFs, a more flexible variant of higher-order linear-chain CRFs. Variable-order CRFs achieve faster inference by including features for only some of the tag n - grams. Our learning method discovers the useful higher-order features at the same time as it trains their weights, by maximizing an objective that combines log-likelihood with a structured-sparsity regularizer. An active-set outer loop allows the feature set to grow as far as needed. On part-of-speech tagging in 5 randomly chosen languages from the Universal Dependencies dataset, our method of shrinking the model achieved a 2 – 6 x speedup over a baseline, with no significant drop in accuracy.
Introduction
Conditional Random Fields (CRFs) (Lafferty et al., 2001) are a convenient formalism for sequence labeling tasks common in NLP. A CRF defines a featurerich conditional distribution over tag sequences (output) given an observed word sequence (input).
The key advantage of the CRF framework is the flexibility to consider arbitrary features of the input, as well as enough features over the output structure to encourage it to be well-formed and consistent. However, inference in CRFs is fast only if the features over the output structure are limited. For example, an order-k CRF (or "k-CRF" for short, with k > 1 being "higher-order") allows expressive features over a window of k+1 adjacent tags (as well as the input), and then inference takes time O(n·|Y | k+1 ), where Y is the set of tags and n is the length of the input.
How large does k need to be? Typically k = 2 works well, with big gains from 0 → 1 and modest * Equal contribution Figure 1: Speed-accuracy tradeoff curves on test data for the 5 languages. Large dark circles represent the k-CRFs of ascending orders along x-axis (marked on for Slovenian). Smaller triangles each represent a VoCRF discovered by sweeping the speed parameters γ. We find faster models at similar accuracy to the best k-CRFs ( §5).
gains from 1 → 2 (Fig. 1). Small k may be sufficient when there is enough training data to allow the model to attend to many fine-grained features of the input (Toutanova et al., 2003;Liang et al., 2008). For example, when predicting POS tags in morphologicallyrich languages, certain words are easily tagged based on their spelling without considering the context (k = 0). In fact, such languages tend to have a more free word order, making tag context less useful. We investigate a hybrid approach that gives the accuracy of higher-order models while reducing runtime. We build on variable-order CRFs (Ye et al., 2009) (VoCRF), which support features on tag subsequences of mixed orders. Since only modest gains are obtained from moving to higher-order models, we posit that only a small fraction of the higher-order features are necessary. We introduce a hyperparameter γ that discourages the model from using many higher-order features (= faster inference) and a hyperparameter λ that encourages generalization. Thus, sweeping a range of values for γ and λ gives rise to a number of operating points along the speed-accuracy curve (triangle points in Fig. 1).
We present three contributions: (1) A simplified exposition of VoCRFs, including an algorithm for computing gradients that is asymptotically more efficient than prior art (Cuong et al., 2014). (2) We develop a structure learning algorithm for discovering the essential set of higher-order dependencies so that inference is fast and accurate. (3) We investigate the effectiveness of our approach on POS tagging in five diverse languages. We find that the amount of required context for accurate prediction is highly language-dependent. In all languages, however, our approach meets the accuracy of fixed-order models at a fraction of the runtime.
Variable-Order CRFs
An order-k CRF (k-CRF, for short) is a conditional probability distribution of the form where n is the length of the input x, θ ∈ R d is the model parameter, and f is an arbitrary user-defined function that computes a vector in R d of features of the tag substring s = y t−k . . . y t when it appears at position t of input x. We define y i to be a distinguished boundary tag # when i / ∈ [1, n]. A variable-order CRF or VoCRF is a refinement of the k-CRF, in which f may not always depend on all k + 1 of the tags that it has access to. The features of a particular tag substring s may sometimes be determined by a shorter suffix of s.
To be precise, a VoCRF specifies a finite set W ⊂ Y * that is sufficient for feature computation (where Y * denotes the set of all tag sequences). 1 The VoCRF's featurization function f (x, t, s) is then defined as f (x, t, w(s)) where f can be any function and w(s) ∈ Y * is the longest suffix of s that appears in W (or ε if none exists). The full power of a k-CRF can be obtained by specifying W = Y k+1 , but smaller W will in general allow speedups.
To support our algorithms, we define W to be the closure of W under prefixes and last-character substitution. Formally, W is the smallest nonempty superset of W such that if hy ∈ W for some h ∈ Y * Algorithm 1 FORWARD: Compute log Z θ (x). α(·, ·) = 0; α(0, #) = 1 initialization for t = 1 to n + 1 : and y ∈ Y , then h ∈ W and also hy ∈ W for all y ∈ Y . This implies that we can factor W as H × Y , where H ⊂ Y * is called the set of histories.
We now define NEXT(h, y) to return the longest suffix of hy that is in H (which may be hy itself, or even ε). We may regard NEXT as the transition function of a deterministic finite-state automaton (DFA) with state set H and alphabet Y . If this DFA is used to read any tag sequence y ∈ Y * , then the arc that reads y t comes from a state h such that hy t is the longest suffix of s = y t−k . . . y t that appears in W-and thus w(hy t ) = w(s) ∈ W and provides sufficient information to compute f (x, t, s). 2 For a given x of length n and given parameters θ, the log-normalizer log Z θ (x)-which will be needed to compute the log-probability in eq. (1) below-can be found in time O(|W| n) by dynamic programming. Concise pseudocode is in Alg. 1. In effect, this runs the forward algorithm on the lattice of taggings given by length-n paths through the DFA.
For finding the parameters θ that minimize eq. (1) below, we want the gradient ∇ θ log Z θ (x). By applying algorithmic differentiation to Alg. 1, we obtain Alg. 2, which uses back-propagation to compute the gradient (asymptotically) as fast as Alg. 1 and |H| times faster than Cuong et al. (2014)'s algorithm-a significant speedup since |H| is often quite large (up to 300 in our experiments). Algs. 1-2 together effectively run the forward-backward algorithm on the lattice of taggings. 3 It is straightforward to modify Alg. 1 to obtain a Viterbi decoder that finds the most-likely tag sequence under p θ (· | x). It is also straightforward to modify Alg. 2 to compute the marginal probabilities of tag substrings occurring at particular positions.
Structured Sparsity and Active Sets
We begin with a k-CRF model whose feature vector . . y t and is 0 otherwise. 4 To obtain the advantages of a VoCRF, we merely have to choose a sparse weight vector θ. The set W can then be defined to be the set of strings in Y * whose features have nonzero weight. Prior work (Cuong et al., 2014) has left the construction of W to domain experts or "one size fits all" strategies (e.g., k-CRF). Our goal is to choose θ-and thus W-so that inference is accurate and fast.
Our approach is to modify the usual L 2regularized log-likelihood training criterion with a carefully defined runtime penalty scaled by a parameter γ to balance competing objectives: likelihood on Recall that the runtime of inference on a given sentence is proportional to the size of W, the closure of W under prefixes and last-character replacement.
(Any tag strings in W\W can get nonzero weight without increasing runtime.) Thus, R(θ) would ideally measure |W|, or proportionately, |H|. Experimentally, we find that |W| has > 99% Pearson correlation with wallclock time, making it an excellent proxy for wallclock time while being more replicable. We relax this regularizer to a convex functiona tree-structured group lasso objective (Yuan and Lin, 2006;Nelakanti et al., 2013). For each string h ∈ Y * , we have a group G h consisting of the indicator features (in f (2) ) for all strings w ∈ W that have h as a proper prefix. Fig. 2 gives a visual depiction. We now define R(θ) = h∈Y * ||θ G h || 2 . This penalty encourages each group of weights to remain all at zero (thereby conserving runtime, in our setting, because it means that h does not need to be added to H). Once a single weight in a group becomes nonzero, the "initial inertia" induced by the group lasso penalty is overcome, and other features in the group can be more cheaply adjusted away from zero.
Although eq. (1) is now convex, directly optimizing it would be expensive for large k, since θ then contains very many parameters. We thus use a heuristic optimization algorithm, the active set method (Schmidt, 2010), which starts with a low-dimensional θ and incrementally adds features to the model. This also frees us from needing to specify a limit k. Rather, W grows until further extensions are unhelpful, and then implicitly k = max w∈W |w| − 1.
The method defines f (2) to include indicator features for all tag sequences w in an active set W active . Thus, θ (2) is always a vector of |W active | real numbers. Initially, we take W active = Y and θ = 0. At each active set iteration, we fully optimize eq. (1) to obtain a sparse θ and a set W = {w ∈ W active | θ (2) w = 0} of features that are known to be "useful." 5 We then update W active to {wy | w ∈ W, y ∈ Y }, so that it includes single-tag extensions of these useful features; this expands θ to consider additional features that plausibly might prove useful. Finally, we complete the iteration by updating W active to its closure W active , simply because this further expansion of the feature set will not slow down our algorithms. When eq. (1) is re-optimized at the next iteration, some of these newly added features in W active may acquire nonzero weights and thus enter W, allowing further extensions. We can halt once W no longer changes.
As a final step, we follow common practice by running "debiasing" (Martins et al., 2011a), where we fix our f (2) feature set to be given by the final W, and retrain θ without the group lasso penalty term.
In practice, we optimized eq. (1) using the online proximal gradient algorithm SPOM (Martins et al., 2011b) and Adagrad (Duchi et al., 2011) with η = 0.01 and 15 inner epochs. We limited to 3 active set iterations, and as a result, our final W contained at most tag trigrams.
Related Work
Our paper can be seen as transferring methods of Cotterell and Eisner (2015) to the CRF setting. They too used tree-structured group lasso and active set to select variable-order n-gram features W for globally-normalized sequence models (in their case, to rapidly and accurately approximate beliefs during message-passing inference). Similarly, Nelakanti et al. (2013) used tree-structured group lasso to regularize a variable-order language model (though their focus was training speed). Here we apply these techniques to conditional models for tagging.
Our work directly builds on the variable-order CRF of Cuong et al. (2014), with a speedup in Alg. 2, but our approach also learns the VoCRF structure. Our method is also related to the generative variable-order tagger of Schütze and Singer (1994).
Our static feature selection chooses a single model that permits fast exact marginal inference, similar to learning a low-treewidth graphical model (Bach and Jordan, 2001;Elidan and Gould, 2008). This contrasts with recent papers that learn to do approximate 1-best inference using a sequence of models, whether by dynamic feature selection within a greedy inference algorithm (Strubell et al., 2015), or by gradually increasing the feature set of a 1-best global inference algorithm and pruning its hypothesis space after each increase (Weiss and Taskar, 2010;He et al., 2013). Schmidt (2010) explores the use of group lasso penalties and the active set method for learning the structure of a graphical model, but does not consider learning repeated structures (in our setting, W defines a structure that is reused at each position). Steinhardt and Liang (2015) jointly modeled the amount of context to use in a variable-order model that dynamically determines how much context to use in a beam search decoder.
Experiments 6
Data: We conduct experiments on multilingual POS tagging. The task is to label each word in a sentence with one of |Y | = 17 labels. We train on five typologically-diverse languages from the Universal Dependencies (UD) corpora (Petrov et al., 2012): Basque, Bulgarian, Hindi, Norwegian and Slovenian. For each language, we start with the original train / dev / test split in the UD dataset, then move random sentences from train into dev until the dev set has 3000 sentences. This ensures more stable hyperparameter tuning. We use these new splits below.
Eval: We train models with (λ, γ) ∈ {10 −4 · m, 10 −3 ·m, 10 −2 ·m}×{0, 0.1·m, 0.2·m, . . . , m}, where m is the number of training sentences. To tag a dev or test sentence, we choose its most probable tag sequence. For each of several model sizes, Table 1 selects the model of that size that achieved the highest per-token tagging accuracy on the dev set, and reports that model's accuracy on the test set.
Features: Recall from §3 that our features include non-stationary zeroth-order features f (1) as well as the stationary features based on W. For f (1) (x, t, y t ) we consider the following language-agnostic properties of (x, t): • The identities of the tokens x t−3 , ..., x t+3 , and the token bigrams (x t+1 , x t ), (x t , x t−1 ), Each row's best results are in boldface, where ties in accuracy are broken in favor of faster models. Superscript k indicates that the accuracy is significantly different from the k-CRF (paired permutation test, p < 0.05) and this superscript is in blue/red if the accuracy is higher/lower than the k-CRF. In all cases, we find a VoCRF (underlined) that is about as accurate as the 2-CRF (i.e., not significantly less accurate) and far faster, since the 2-CRF has |W| = 4913. Fig. 1 plots the Pareto frontiers.
(x t−1 , x t+1 ). We use special boundary symbols for tokens at positions beyond the start or end of the sentence.
• Prefixes and suffixes of x t , up to 4 characters long, that occur ≥ 5 times in the training data. • Indicators for whether x t is all caps, is lowercase, or has a digit. • Word shape of x t , which maps the token string into the following character classes (uppercase, lowercase, number) with punctuation unmodified (e.g., VoCRF-like ⇒ AaAAA-aaaa, $5,432.10 ⇒ $8,888.88). For efficiency, we hash these properties into 2 22 bins. The f (1) features are obtained by conjoining these bins with y t (Weinberger et al., 2009): e.g., there is a feature that returns 0 unless y t = NOUN, in which case it counts the number of bin 1234567's properties that (x, t) has. (The f (2) features are not hashed.) Results: Our results are presented in Fig. 1 and Table 1. We highlight two key points: (i) Across all languages we learned a tagger about as accurate as a 2-CRF, but much faster. (ii) The size of the set W required is highly language-dependent. For many languages, learning a full k-CRF is wasteful; our method resolves this problem.
In each language, the fastest "good" VoCRF is rather faster than the fastest "good" k-CRF (where "good" means statistically indistinguishable from the 2-CRF). These two systems are underlined; the underlined VoCRF systems are smaller than the underlined k-CRF systems (for the 5 languages respectively) by factors of 1.9, 6.4, 3.4, 1.9, and 2.9. In every language, we learn a VoCRF with |W| ≤ 850 that is not significantly worse than a 2-CRF with |W| = 17 3 = 4913.
We also notice an interesting language-dependent effect, whereby certain languages require a small number of tag strings in order to perform well. For example, Hindi has a competitive model that ignores the previous tag y t−1 unless it is in {NOUN, VERB, ADP, PROPN}: thus the stationary features are 17 unigrams plus 4 × 17 bigrams, for a total of |W| = 85. At the other extreme, the Slavic languages Slovenian and Bulgarian seem to require more expressive models over the tag space, remembering as many as 98 useful left-context histories (unigrams and bigrams) for the current tag. An interesting direction for future research would be to determine which morpho-syntactic properties of a language tend to increase the complexity of tagging.
Conclusion
We presented a structured sparsity approach for structure learning in VoCRFs, which achieves the accuracy of higher-order CRFs at a fraction of the runtime. Additionally, we derive an asymptotically faster algorithm for the gradients necessary to train a VoCRF than prior work. Our method provides an effective speed-accuracy tradeoff for POS tagging across five languages-confirming that significant speed-ups are possible with little-to-no loss in accuracy. | 4,271.4 | 2016-11-01T00:00:00.000 | [
"Computer Science"
] |
Quantum computing dataset of maximum independent set problem on king lattice of over hundred Rydberg atoms
Finding the maximum independent set (MIS) of a large-size graph is a nondeterministic polynomial-time (NP)-complete problem not efficiently solvable with classical computations. Here, we present a set of quantum adiabatic computing data of Rydberg-atom experiments performed to solve the MIS problem of up to 141 atoms randomly arranged on the king lattice. A total of 582,916 events of Rydberg-atom measurements are collected for experimental MIS solutions of 733,853 different graphs. We provide the raw image data along with the entire binary determinations of the measured many-body ground states and the classified graph data, to offer bench-mark testing and advanced data-driven analyses for validation of the performance and system improvements of the Rydberg-atom approach.
BACKGROUND & SUMMARY
The maximum independent set (MIS) problem belongs to the computational class of nondeterministic polynomial (NP)-complete problems, the hardest computational problems of no known classical algorithms that are efficient [1].For a given graph G(V, E), the MIS problem aims to find the maximum independent set, the largest set among the independent sets, where the independent set I ⊂ V is defined as a set of unedged vertices, i.e., v i , v j ∈ I and (v i , v j ) / ∈ E. Rydberg quantum simulators are currently one of the biggest quantum computing physical platforms, capable of utilizing up to a few hundred qubits [2,3].In particular, the constraint of the independent set is implementable intrinsically with the Rydberg blockade effect that forbids two atoms proximate within a certain distance from being simultaneously excited to the same Rydberg-atom state [4][5][6][7].Therefore, a set of atoms arranged to a graph and simultaneously pumped to Rydberg atoms results in a non-adjacent arrangement of Rydberg atoms fulfilling the independent set constraint.In addition, by tuning the Hamiltonian of the atoms to maximize the number of Rydberg atoms, the MIS is achieved by the set of Rydberg atoms in the many-body ground state.In that regards, Rydberg-atom systems could perform adiabatic quantum computation (AQC) for the MIS problem in such a way that a target many-body ground state is prepared adiabatically from an easily preparable initial ground state.There are several recent experiments computing the solution of the MIS problem on the Rydberg-atom system by AQC [8][9][10].
Here we provide a set of experimental AQC data of the MIS problem performed on the Rydberg-atom system.We first prepare an 11-by-18 array of optical tweezers.This lattice is identical to the union-jack-like king's graph [11] recently experimented by Ebadi et al., [8] in *<EMAIL_ADDRESS>which its NP completeness on MIS problem has been addressed, as in Fig. 1.The MIS problem embedded on this graph has also been researched theoretically about its possibilities on quantum speedup [12,13].On 198 optical tweezer traps, atoms are stochasitcally loaded with about probability of about 50%, and resulting random graphs are used.The size of experimented atom arrays is maximally 141 and on average about 104.Atoms then go through the initial state preparation stage, and the global laser field is applied to drive the total system for AQC.Laser intensity (Rabi frequency) and frequency (detuning) is slowly sweeped from the initial condition, and the initially prepared ground states are turned into the target final state before being measured with the imaging of the post AQC atomic array.There are 45 different experiment sets with different parameters including adiabatic sweep time, initial and final detunings, and for each set of data about 5,000 to 30,000 experiments were repeated.Besides the main data, we analyze major error sources from the measurement, Rydberg state decay, and the atom loss due to the finite temperature of the atoms.The pure effect of the control error scales with respect to the AQC sweeping time of an order of ∼ τ −0.54 (4) , comparable with the earlier reported ∼ τ −0.48 (2) in Ebadi et al [8].This data could be harnessed for the analysis of adiabatic computing behavior, the exploration of quantum phase transitions (QPT) in the transverse Ising model, and as a reference for conducting benchmark tests of the Rydberg atom approach to optimization problems.
Atom-array preparation
Rubidium atoms ( 87 Rb) are loaded on to an array of 18 by 11 optical tweezers, with the nearest atom distance of d = 6.0 µm.The initial trap phase was generated with the Gerchberg-Saxton weighted (GSW) algorithm with 100 iterations starting from the uniform trap phase guess [14].For a calibration of uniform traps, we measure the intensity of each 198 traps by (1) first image the trap beam itself with the charged coupled device (CCD) (2) resonance scan the amount of a.c.Stark shift of the each trap site with the push-out beam (5S 1/2 , F = 2 → 5P 3/2 , F ′ = 3).Meausred intensity is used to determine the weight given for each trap while iterating the GSW algorithm as [15], where I i is the intensity of the i-th trap.We achieve the standard deviation below 3% based on the atomic resonance.Initially inside the vacuum chamber, the 87 Rb atoms are cooled and loaded inside the magneto optical trap (MOT) as an ensemble initially by turning on the anti-Helmholtz coil and the Doppler cooling beam.Optical tweezer beam from the Ti:Sapphire oscillator (TiC of Avesta) with wavelength 813 nm is turned on and atoms are loaded and by polarization gradient cooling (PGC) cooled down to ∼ 30 µK inside the optical tweezer.We use objective lens with high numerical aperture (NA) of 0.5 (G Plan Apo 50X of Mitutoyo).Each trap has a beam radius of 1.1 µm, therefore each tweezer has about 1 mK depth in average.The trap depth data of all sites are recorded (see Data Records).The temperature is measured using the release and recapture method [16] before the main experiment.The temperature change due to laser conditions of the MOT beam causes different loss rate, which is considered in Technical Validation.Atoms are stochas-tically loaded in the tweezers with the per-site loading probability ∼ 0.5, because of the collisional blockade [17], resulting in a random geometry to be experimented for each repetition.Atom lifetime inside the trap is ∼ 40 s and the experiment is operated with a repetition cycle of 1 s.
AQC programming of MIS problem
Atoms are prepared in a random graph G(V, E), in which a vertice in V corresponds to each atom and an edge in E corresponds to the van der Waals interaction between a strongly interacting atom pair.To be specific, we regard the atoms within the Rydberg blockade radius r b = (C 6 /Ω) 1/6 ∼ 10 µm are edged.The nearest neighbor distance is 6 µm, and the next nearest neighbor distance (of diagonal edges) is 6 √ 2 ∼ 8.5 µm, so the atoms are edged according to the union-jack-like king's graph as shown in Fig. 1.Solving the MIS problem on the given graph is the same as finding the many-body ground state of the atom array in the MIS phase.The Hamitonian of the given Rydberg system consists is a transverse anti-ferromagnetic (AF) Ising Hamiltonian, given by ĤRyd = Ω(t) 2 where ℏ = 1, Ω is Rabi frequency, ∆ is detuning, −6 , and ni = (σ z i + 1)/2 defined for a psedo-spinor system of |g⟩ = The Rydberg operation is performed, and the tweezer is turned on, then results are imaged.(c) Adiabatic quantum computing sequence.Rabi frequencies are turned on and off linearly with time tΩ with the maximum value Ω0, and the detuning is swept from ∆i to ∆ f linearly during the time t∆.
The hamiltonian is driven and controlled with the twophoton transition of 780 nm and 480 nm beams.For Ω = 0, ∆ < 0, the many-body ground state is the paramagnetic |g⟩ ⊗N .When Ω = 0, 0 < ∆ < U diag , the many-body ground state is in the MIS phase [10].
The experimental sequence diagram, which includes AQC, is depicted in Fig. 2(b).AQC takes place when the optical tweezers are turned off, and photographs are taken of each randomly configured graph along with the final results both before and after the AQC process.We program our AQC to sweep the atom array from the paramagnetic |g⟩ ⊗N phase to the MIS phase.Initially, we prepare the initial ground state |g⟩ ⊗N by optical pumping.Then we take three passages, as in Fig. 2 for t Ω while turning off the trap.For our experiment, we fix t Ω = 0.3 µs and Ω 0 is position dependent along the yaxis of the lattice because the beam is Gaussian, and the values of Ω 0 for different positions are calculated from Rabi freuqency measurements as in Table .I (see Technical Validations).We then turn off the trap for 6.5 µs for an AQC experiment.There are 45 distinct experiments characterized by varying t ∆ , ∆ i , and ∆ f , each having data of 4,000 to 30,000 repetitions.The experiments are summarized in Table.II.For the all 45 experiments, the final detuning ∆ f is larger than 0 and smaller than the diagonal interaction U diag = C 6 /(8.5 µm) 6 = 2π×2.7 MHz, satisfying the MIS conditions.To obtain the information of the initial atom array and the AQC result, two images of the atom array embedded in the tweezers Fig. 3(a) are captured in each experimental iteration using the electron-multiplied charged coupled device (EMCCD, iXon Ultra 888 by Andor).The first photo, e.g., in Fig 3(b), is taken following array preparation, where atoms are loaded randomly and the second photo, e.g., in Fig 3(c), is taken after the AQC process.During each image taking, the fluorescence of each trap site is analyzed to determine whether an atom is trapped inside the optical tweezer.Discriminating between the photon counts of trapped and non-trapped atoms provides digitized data for the experiment, as in Fig 3(d).Notably, for the second photo taken after AQC, the trap is activated before the photograph, caus-ing atoms in the Rydberg state to be anti-trapped from the optical tweezers.Therefore, the remaining atoms in the second photo are considered to be in the ground state |g⟩, while the disappeared atoms are regarded as being in the Rydberg state |r⟩.
DATA RECORDS
The compilation of 45 distinct experimental parameters can be found in The primary data is provided in the form of '.tif' and '.mat' files, including raw image files, fluorescence data for each atom, and digitized data for each atom.Raw image files are in TIF format, constituting a list of image files structured as a three-dimensional array.The first two dimensions denote the image, while the third dimension represents the stack of images across experimental repetitions.Due to the substantial file size, raw data is segmented into multiple files for the same experiments, named as 'Exp{Exp#} X{#}.tif',where Exp# corresponds to the experiment index in Table II, and # is assigned in sequential order.Fluorescence data and digitized data are stored in MATLAB data files, featuring the three-dimensional array variables 'fluoAreshape' and 'fluodigreshape,' alongside the variable 'threshold,' which establishes the threshold for fluorescence data digitization following the outlined methods.The 3D array data in 'flouAreshape' and 'floudigreshape' is organized as follows: the first dimension denotes each tweezer, the second dimension signifies the two photographs of each repetition, and the third dimension corresponds to the number of experiment repetitions.For instance, in MAT-LAB language, extracting the 2D array of the experiment for the first repetition from 'fluodigreshape' is achieved by using 'squeeze(fluodigreshape(:,:,1))'.The first column of this array presents digitized data from the first photo, while the second column represents data from the second photo.A value of '1' indicates that the atom is trapped, while a value of '0' signifies that the atom is not trapped.Consequently, in the second column, a '1' implies measurement in the ground state |g⟩, whereas a '0' signifies measurement in the Rydberg state |r⟩.The data for the |g⟩ → |r⟩ SPAM error is labeled as 'SPAM{p rg,SPAM #}', encompassing both the raw image file in '.tif' format and the data file in '.mat' format.The image file contains experiments related to release and recapture data, while the MATLAB data file incorporates variables such as 'fluoAreshape', 'fluodigreshape', along with two one-dimensional arrays, namely 'Prg' and 'counts'.Within the MATLAB data, 'Prg' captures the measurement error of |g⟩ → |r⟩ for each atom, as determined by the release and recapture method (see Tech- We organize both the exact solution and the experimentally obtained solutions for each randomly configured (isomorphic) graph.Initially, the graphs are categorized into connected components, as illustrated in Fig. 3(e).Subsequently, for each connected graph, the exact solutions are determined using the 'MISSolver.py'code (see Code Availability).The resulting data is structured in a table implemented as a linked list class named 'Graph-Table ', and stored within the Python script 'Graph-Table .py'(Fig. 3(f), see Code Availability).The instances of the generated 'GraphTable' are then saved by the pickle module of Python under the file name of 'ExpExp#.pkl'.Each 'GraphTable' instance represents a Python list of the linked list class 'GraphLinkedList' (saved in 'GraphLinkedList.py',see Code Availability), arranged according to the graph size |G|.Within each 'GraphLinkedList', there exists a linked list pertaining to the same graph size |G|.Each instance of the node class ('GraphNode') encapsulates the data of each isomorphic graph, comprising the graph structure, exact solutions, and a Python list detailing the experimented solutions.
TECHNICAL VALIDATION
In Fig. 4(a), the histogram illustrates log events organized by both the graph size G and the corresponding errors δ G for each graph G in Experiment 10 as an example.The error δ G for a given graph G is calculated as the difference between the exact MIS solution and the experimental MIS solution (Rydberg excitation number).The major error sources in the AQC include the state-preparation-and-measurement (SPAM) error and the control error.In the technical validation section, we assess the SPAM error's magnitude and establish the validity of our data by analyzing the behavior of the control error while varying the sweep speed and final detuning of the AQC.SPAM errors are single-site bit-flip errors following a binomial distribution.Consequently, their effect on the MIS solution of the graph G(V, E) is, on average, proportional to the graph size |G| itself.In the quasi-adiabatic regime, the control error adheres to the quantum Kibble-Zurek mechanism (QKZM) [2,[18][19][20][21], which is also proportional to the graph size |G|.Thus, the overall error in the graph G(V, E) can be modeled on average as ⟨δ G ⟩ = (αp gr − (1 − α)p rg )|G|, where α represents the average solution ratio, and p gr and p rg denote the error rates for transitions from |r⟩ → |g⟩ and |g⟩ → |r⟩, respectively.
Rabi frequency for each atoms
Due to the finite radius of the Gaussian Rydberg beam, the Rabi frequencies of individual atoms exhibit variations.To see the difference of Rabi frequencies among atoms, we measure the Rabi frequency on 25 atoms, each positioned sufficiently far apart to ensure independence (with a minimum distance of 17 µm, where r B = 10 µm).The Rabi frequency for each y axis position is determined through fitting based on this data, allowing us to calculate the Rabi frequency for the 198 atoms in the main experiment as in Fig 4(b).Additionally, we provide the average calculated Rabi frequency for each of the 18 rows of the experiment in Table I for reference.In the subsequent discussion on control error, we do not account for the differences in control error resulting from varying Rabi frequencies but instead focus on the average control error.
SPAM errors
SPAM errors originating from |r⟩ → |g⟩ (p gr,SPAM ) and |g⟩ → |r⟩ (p rg,SPAM ) stem from distinct sources.The SPAM error from |r⟩ → |g⟩ (p gr,SPAM ) is predominantly attributed to the decay of the Rydberg state.Conversely, the error from |g⟩ → |r⟩ (p rg,SPAM ) primarily results from atom loss in the trap, stemming from factors such as the finite temperature of atoms and non-resonant scattering, among others [22].To characterize these errors, we initially examine the overlap of the fluorescence data at each atom site.Subsequently, we measure the Rydberg decay time, and finally, we directly measure the loss rate by releasing the atom and subsequently recapturing it.
For the measurement of the Rydberg decay time T 1 , we initiate a π pulse, followed by a wait period of t, and then apply another π pulse.By varying the duration t, we observe the decay of the measurement, as depicted in Fig. 4(c).Fitting the data with the model e −t/T1 + c, we obtain T 1 = 49(18) µs.While ideal theoretical considerations, accounting for black body radiation, suggest a lifetime of approximately 150µs [23], realistic errors have led to similar values, with other groups reporting T 1 = 50(8)µs [24].For our experiment, where the trap is turned off for 6.5 µs, we can calculate p gr,SPAM = 1 − e −6.5/49 (18) = 0.12 +0.06 −0.03 .To directly measure the atom loss rate stemming from factors like finite temperature and non-resonant laser scattering, the release and recapture method is employed [16].After trapping the atom, we turn the trap off for a duration of 6.5 µs, aligning with the trap turningoff time in the main experiment.Subsequently, the atom is recaptured by reactivating the trap laser.The recapture rate is measured, and the loss rate p rg,SPAM is determined as 1 minus the recapture rate.Three distinct p rg,SPAM measurements (p rg,SPAM # in Table II) are presented, reflecting variations in cooling laser power conditions leading to different Polarization Gradient Cooling (PGC) rates and, consequently, distinct loss rates.In each instance, the mean loss rates are 0.0730, 0.0877, and 0.0966, respectively.The data for these three different loss rate measurements are also provided for each atom under the name 'SPAM#.mat'(see Data Records).To characterize the potential nonuniformity in the total optical tweezer array, which could result in a nonuniform atom loss rate for each trap site, the trap depth is measured for each tweezer site.The trap depth data is available in the 'trapdepth.mat'file, presented in frequency units (MHz).
Control errors
Early stage control error measurement for tΩ and ∆i.
To validate that our data exhibits minimal errors during the initial sweep of the measurement, we conducted experiments as depicted in Fig. 4(d).During this pro- cess, the laser is both turned on (ramping Ω from Ω 0 to 0) and turned off (ramping Ω from 0 to Ω 0 ) while maintaining a fixed detuning at the initial value ∆ i .In the absence of control errors during this stage, there should be no atom loss as the system returns to its initial condition.We varied the detuning from −4.4MHz to 6.6 MHz with increments of 1 MHz and adjusted the ramping time t Ω from 0.1 µs to 1 µs with a step size of 0.1 µs.
The loss probability is calculated as one minus the ratio of remaining atoms compared to the initially captured atoms.For the main experiment, we selected t Ω = 0.3 µs and ∆ i = −4.4MHz and −4 MHz, as these parameters yielded the lowest loss probabilities among the considered parameters.Therefore, we can affirm that our main experiment is robust to early-stage control errors.
Control error estimates from the main experiment data
Control errors are investigated by conducting experiments with varying sweep rates in the sequence.The control errors in our experiment align with the universal scaling predicted by the Quantum Kibble-Zurek Mechanism (QKZM) in relation to the sweep speed.The sweep speed s is defined as s = (∆ f −∆ i )/t ∆ , where t ∆ is the detuning sweep time.Experiments 1 to 25, 26 to 35, and 36 to 45 can be grouped into experimental sets to examine control errors by altering specific parameters (s or ∆ f ) while keeping others constant.The average total error is modeled as previously, ⟨δ G ⟩ = (αp gr − (1 − α)p rg )|G|.This linear behavior can be observed in the initial segment of Fig. 4 where various sweep rates s were explored.In this context, it is clear that when ∆ f is held constant, the control error in our data follows the anticipated pattern: the control error diminishes for smaller sweep speeds s or extended annealing times s −1 , while it escalates for larger sweep speeds s or abbreviated annealing times s −1 .Furthermore, we can fit the control error as a function of the sweep rate s due to QKZM, yielding CE ∝ s µ with µ = 0.54(4) for the fitting range s = 7.5MHz/µs to 60 MHz/µs.This value is in line with earlier research, which reported µ = 0.48(2) using the same lattice (supplementary of Ref. [8]).We note that the slope values ⟨δ G ⟩/|G| and the ratio α are obtained by using |G| < 50 data.
USAGE NOTES
The dataset is accessible using MATLAB.Python users can access the data using the scipy.io.loadmat function within the scipy library.Analysis of randomly generated graphs can be performed using the 'networkx' library in Python.The data can be grouped from 1 to 18, 19 to 25, 26 to 35, and 36 to 45, considering that only one parameter differs within each group.
FIG. 1 .
FIG. 1. Image of 198 optical tweezers with its index.Right side of the figure shows the connection between the atoms.The nearest atoms (6 µm) and the next nearest atoms (6 √ 2 µm) are connected (Rydberg blockade radius r b ∼ 10 µm), configuring king's graph.
FIG. 2 .
FIG. 2. (a) Experimental setup.(b) Experimental sequence.Loaded atoms are imaged initially, then the tweezer is turned off.The Rydberg operation is performed, and the tweezer is turned on, then results are imaged.(c) Adiabatic quantum computing sequence.Rabi frequencies are turned on and off linearly with time tΩ with the maximum value Ω0, and the detuning is swept from ∆i to ∆ f linearly during the time t∆.
FIG. 3 .
FIG. 3. Data processing of quantum computing data (a) Initial 198 optical tweezer sites (b) Atoms are randomly loaded into tweezer with a probability of ∼ 0.5 (c) Atoms after the AQC.Rydberg atoms are anti-trapped.(d) Digitize the prepared array (circles) and the Rydberg atoms (filled circles) with the images from (b) and (c).(e) Classify the graph into connected components (colored if connected) (f) List all connected graphs, same sizes linked by the linked list structure, where each node contains a graph, exact MIS solution, experimented MIS solutions (by the list), and the graph hash.
1 FIG. 4 .
FIG. 4. Figures for technical validation(a) Plot of log events with regards to the graph size |G| and the error δG for 'Exp10'.Here the error δG is the difference between the exact MIS solution and the experimented MIS solution.(b) Rabi frequency for each y pixel.Average values for each row are given in Table I.(c) Measurement of Rydberg state decay time.Rydberg probability after π pulse and a time t is measured.(d) Calibration for initial detuning ∆i.From the sequence of Fig. 3(c), the sequence (2) is deleted, and ∆ f = ∆i for the calibration experiment, is expected to have no loss.Shaded region of ∆i and tΩ = 0.3 µs are used for our main experiment, where loss probability is small.(e) Control error parameter CE as a function of sweep rate s.Fitting for s > 7.5 MHz/µs gives µ = 0.54(4) comparable with the value from other group 0.48(2)[8].
(a) (red line).We can now express the equation as follows:CE ≡ αp gr,control − (1 − α)p rg,control = ⟨δ G ⟩ |G| + ((1 − α)p rg,SPAM − αp gr,SPAM )(2)We introduce the control error parameter CE to represent the second term of the equation, which characterizes the control error value in the experiment.The value CE can be computed using the last term of the equation, which is accessible through experimental data.The average solution rate, denoted as α, can be calculated from the exact solutions for each graph obtained in the experiment, with a fixed value of α ≃ 0.43 regardless of the sweep speed.The value ⟨δ G ⟩/|G| represents the slope of the red line in Fig 4(a) and varies with the sweep speed.Fig.4(e) illustrates the plot of CE for all experiments,
TABLE I .
Average y pixel and Rabi frequency for each of 11 rows.
TABLE II .
Initial, final detunings, detuning sweep time, prg number, and repetition number of the experiment. | 5,635 | 2023-11-23T00:00:00.000 | [
"Physics",
"Computer Science"
] |
An adaptive geometric search algorithm for macromolecular scaffold selection
Abstract A wide variety of protein and peptidomimetic design tasks require matching functional 3D motifs to potential oligomeric scaffolds. For example, during enzyme design, one aims to graft active-site patterns—typically consisting of 3–15 residues—onto new protein surfaces. Identifying protein scaffolds suitable for such active-site engraftment requires costly searches for protein folds that provide the correct side chain positioning to host the desired active site. Other examples of biodesign tasks that require similar fast exact geometric searches of potential side chain positioning include mimicking binding hotspots, design of metal binding clusters and the design of modular hydrogen binding networks for specificity. In these applications, the speed and scaling of geometric searches limits the scope of downstream design to small patterns. Here, we present an adaptive algorithm capable of searching for side chain take-off angles, which is compatible with an arbitrarily specified functional pattern and which enjoys substantive performance improvements over previous methods. We demonstrate this method in both genetically encoded (protein) and synthetic (peptidomimetic) design scenarios. Examples of using this method with the Rosetta framework for protein design are provided. Our implementation is compatible with multiple protein design frameworks and is freely available as a set of python scripts (https://github.com/JiangTian/adaptive-geometric-search-for-protein-design).
Introduction
In the past 15 years, protein design has advanced considerably in scale, accuracy and the variety of design tasks carried out by practitioners. Early successes in protein design focused on protein fold design (including novel folds) and hyperstabilization of proteins . The redesign of protein-protein (Boyken et al., 2016) and protein-DNA (Ashworth et al., 2006) interfaces is a step towards functional rewiring of biological networks. More recently, protein engineers have turned toward the redesign of protein active sites and smaller functional patterns that demand sub-angstrom accuracy in the positioning of key side chains. Such works include both the engraftment of known active sites onto new scaffolds (Jiang et al., 2008) as well as the engraftment of novel active sites (derived from quantum mechanical modeling of desired reactions) (Röthlisberger et al., 2008) onto new scaffold proteins. In these enzyme design applications, active site patterns can become quite large-as residues involved in substrate binding, reaction mechanism and the surrounding environment may be considered. Enzyme design and related design tasks involving functional site or hotspot transplantation depend, in part, upon methods for matching a spatial pattern of chemical functional groups onto large libraries of potential scaffolds (proteins, nucleic acids or synthetic peptidomimetics, for example).
The earliest geometric matching applications in bioinformatics were aimed at matching whole substructures that indicated a likelihood of shared protein function or distant homology (Holm and Laakso, 2016). In many cases, these algorithms searched for contiguous regions and essentially functioned as the structural analog of sequence alignment algorithms (both gapped and ungapped). Applications included protein function prediction, analysis of protein structure prediction and evaluation of new algorithms (Nussinov and Wolfson, 1991;Fischer et al., 1992;Bonneau et al., 2002). Related work included innovative geometric hashing to extract 3D functional motifs from protein structures (Wallace et al., 2008). In this work, we focus on geometric searches for biodesigns rather than prospecting or annotation.
Geometric searches developed for similar design tasks have used combinations of geometric hashing, side chain conformation libraries and other heuristics that have typically limited the number of functional elements in any given search pattern. Fleishman et al. computationally designed a protein to bind hemaglutinin (HA), targeting a conserved region on the stem (Fleishman et al., 2011). They first identified the spatial positions of possible high-affinity (hotspot) residues by docking single amino acids onto the HA stem region and calculating a binding energy. Next, for residues predicted to have sufficient binding energies to HA, they built inverse rotamer libraries (i.e. rotamer distributions rooted at the side chain functional group rather than the backbone)-which served as anchor sites on which to dock protein scaffolds. The protein scaffolds themselves were selected from proteins not known to bind HA and were filtered for high shape complementarity with the HA target region. A lowresolution docking procedure was used to simultaneously optimize the HA scaffold binding energy as well as the scaffold's ability to accommodate anchor residues. Scaffolds that showed geometric complementarity with the satisfied hotspot residues were used as the starting point for a second round of docking and design to optimize scaffold side chain positions surrounding the hotspot residues.
There are additional examples of geometric search-driven design on synthetic oligomeric foldamers and short peptidomimetic scaffolds. The objectives of the peptidomimetic design task may vary considerably: e.g. active-site mimicry, interface binding, metal binding or surface adhesion (Pacella et al., 2013). The set of oligomeric scaffolds can provide protein-like side chain spatial armaments is quite diverse; examples include linear peptoids (Zuckermann et al., 1992), oligooxopiperazines (OOPs) (Tošovská et al., 2010), HBS helices (Chapman et al., 2004), cyclic peptides (Bhardwaj et al., 2016) and peptoids (Yoo et al., 2010), β-peptides (Molski et al., 2013) and hybrids thereof. A frequent aim is to mimic protein-protein interfacial hotspots in which a small number of side chains scaffolded by a single secondary structure element comprise a significant fraction of the binding energy (Watkins and Arora, 2015). In these cases, moving such side chain groups to a new, nonprotein, scaffold with synthetically restricted backbone degrees of freedom and reduced atomic mass may be a viable route to inhibiting protein-protein interactions (PPIs). Lao et al. (2014) showed that by grafting four side chains from a restricted segment of sequence onto a four-subunit OOP scaffold creates low nanomolar inhibitors of two important PPIs (p53-MDM2 and p300-Hif1α). The first step in this work used a geometric search to dock the OOP scaffold into the binding site, such that side chain take-off angles were compatible with those the three hotspot residues (predicted to comprise the majority of the binding energy) in the experimental structure. After the geometric search instantiated a starting pose, the Rosetta design procedure (with modifications for both NCAA side chains and the OOP backbone) was used to optimize binding-resulting in low nanomolar inhibitors of both complexes. In both cases, the geometric match steps were based on expensive inverse rotamers searches, which limits the procedure to only small peptidomimetics. Drew et al. (2013) previously demonstrated the incorporation of several non-peptidic backbone chemistries in the macromolecular modeling suite, Rosetta. There are many additional abiotic foldamer and peptidomimetic backbones (Guichard and Huc, 2011) that are amenable to such treatment. Determining the foldamer backbone (or hybrid chemistry) most compatible with a given interface is a potential bottleneck as the number of synthetically accessible scaffolds for biomimicry continues to increase.
Here, we describe a new method combining octrees (a data structure that maps regions of 3-dimensional space to nodes in a tree) and a novel adaptive search that grants a significant performance gain for the applications described above. Key innovations include the ability to weight interaction/pattern components by energy and the adaptive nature of the search, which both increases efficiency and allows for specification of error tolerance (per component of the template pattern) and number of mismatches. We pose the problem by describing a typical setup. We then describe our core algorithm. Finally, we describe applications to protein and peptidomimetic design tasks.
Problem setup
Given a library of molecular scaffolds, our method will find a suitable set of scaffolds to cause a set of target functional groups to be fixed in space relative to one another. We use the term functional group to indicate the terminal atoms of a side chain, i.e. those atoms whose position will remain fixed relative to one another during the rotation of the χ angles of the side chain. Examples would not only include the phenyl, imidozol and guanadinium groups of phenylalanine, histidine and arginine, respectively, but also the four terminal carbons of leucine (Cβ, Cγ, Cδ1, Cδ2) and the hydrogens that branch from them. A molecular scaffold is defined generally as any molecule from which designable side groups could branch.
A given scaffold will typically have varying degrees of freedom and these degrees of freedom will therefore define that scaffold's ability to accommodate fixed functional groups. Practically, different scaffolds will have different degrees of flexibility at different positions and this will drive our definition of allowable error of matching. For a peptide, the predominant degrees of freedom are the φ and ψ angles of the backbone and χ angles in the side chains. Peptidomimetic scaffolds will have different degrees of freedom. For example, in peptoids we must also consider the cis/trans state of the preceding-ω angle, which potentially allows for greater diversity of side chain Cα-Cβ bond vectors for a given sequence. Alternatively, an OOP scaffold, which has cyclic constraints between neighboring residues, is theoretically much more restricted in its ability to accommodate fixed functional groups but also has a reduced entropic cost upon binding a target.
Our approach to interface design is a two-step process. In the first step, we consider the most influential energies and conduct an efficient geometric search to eliminate all the impossible designs. In a second step, designs that passed the quick initial screening are further refined using the Rosetta suite (Leaver-Fay et al., 2011), potentially introducing additional mutations. This two-step process efficiently saves all the time that the majority impossible designs would take to be evaluated by Rosetta.
In the first step, since we consider only the most optimal bond angles, the problem is reduced to the following abstract math problem. We consider the binding interface configuration as a polygon whose vertices are the binding nodes. For each side chain, all the possible positions of one binding node (often the end node of the side chain) form a manifold in 3D obtained by fixing all the bond angles to the optimal ones and sample possible rotations of dihedral angles (Fig. 1). Thus, the task is to find (at least) one potential binding node from each manifold such that they form a 'desirable configuration' should such a 'desirable configuration' exist (and later possibly check that all dependent nodes do not physically collide with every other node).
Let P P P P , , , n 1 2 〈 = ⋯ 〉 be the target polygon. In this paper, all polygons are denoted by putting angle brackets around their ordered vertices. We define the error ε of a configuration or a polygon S S S S , , , n 1 2 〈 = … 〉 by its distance to the target configuration P defined as Following the standard notation, we use capital letters to denote points in space and we write AB to denote the length of the line segment joining the points A and B. Let T ε be the maximum error we allow to account for the smaller energies we are ignoring and errors due to the discretization of the manifolds. If T ε ε < , we call the configuration or polygon S a 'desirable configuration'. Therefore, the problem of the peptoid design is to select the best side chain and backbone constitutions such that there exists a desirable binding configuration while maintaining a low-energy state.
Adaptive geometric search algorithm
We employ octrees as the core data structure for our algorithm (Berg et al., 2008). A cubic volume, with sides of length l, centered on a point p, can be subdivided into eight cubes with sides of length l/2, that share p as a vertex. Each of these eight cubes can be further subdivided into eight more cubes each with side of length l/4, and so on. This decomposition of 3D space lends itself to a tree-like representation called an octree. Thus, octrees are tree structures whose nodes correspond to 3D cubes embedded in a hierarchically subdivided overall 3D space and each deeper level of the tree describes a successively smaller volume of space. Each node has eight 'children' nodes obtained by subdividing each side of the cube by the middle in the x, y and z dimensions. All the 3D objects, in our case, points in 3D, are stored in the leaf nodes. Octrees have various stopping criteria to prevent the tree from splitting into forever smaller cubes, including thresholding based on the number of 3D objects in a node, i.e. the octree splits only if the nodes contain more than a certain number of 3D objects. For our problem, these 3D objects are simply points in the 3D space and the stopping criterion is the minimum cube length l s . That is, the octree splits a node only if its corresponding cube has sides of length at least 2l s . Moreover, all empty nodes, i.e. nodes whose corresponding cubes contain no points, are discarded.
To find desirable configurations, first we sample all possible positions of a functional group, or binding node by fixing all the bond angles to the optimal ones and sample possible rotations of dihedral angles (Fig. 1). This process forms a manifold for each functional group. Then the algorithm builds octrees using sampled points from each manifold and the tree stops branching at the leaf cubes of length at least l s . Next, the algorithm compares every two octrees at a time by testing the necessary and sufficient conditions on their corresponding nodes and searching adaptively only down the pairs of nodes that pass the necessary condition (see below). We call a pair of nodes (and the corresponding cubes) that pass the necessary condition a 'possible pair'. The algorithm finds all the possible cube pairs at each tree level until it ends up with the set of all possible pairs of leaf cubes. Then it tests the sufficient condition on the possible pairs of leaf cubes to determine whether to accept or reject all the pairs of points inside them. At the end, all the pairwise desirable cubes are combined through a matrix product to identify desirable n-tuples or 'desirable configurations'.
Establishing necessary and sufficient conditions for matching
Our overall strategy is to enumerate all possible residue positions (when there is a choice on the particular scaffold) and amino acid assignments to these residues and then to use the adaptive geometric algorithm to determine whether the resulting functional groups, or binding nodes, at those positions have the proper geometry. Thus, the adaptive geometric algorithm is the 'inner loop' of the computation with the 'outer loop' being all possible residue positions and amino acid assignments. For this inner loop to be efficient, it must swiftly filter away impossible geometries (Theorem 1 below) and identify promising ones (Theorem 2 below).
Mathematically, the adaptive geometric algorithm efficiently searches for a certain n-polygon among n sets of points in 3D space given an error tolerance and an approximation margin. This general scheme is required for all the applications introduced above and evaluated in the Results section. Given a target polygon P= P P P , , , , be two non-empty cubes with size l and the distance between their centers d, where i j n i j , 1, 2 , , , ∈ { … } ≠ . Then we have the following theorems that help us determine which cubes could possibly contains pairs of points whose line segment matches that edge. That is, the theorems provide acceptance and rejection criteria for pairs of cubes from different trees (which correspond to different manifolds where each manifold corresponds to, for example, a take-off residue from a backbone). The first theorem provides a rejection criterion.
Theorem 1 suggests a 'necessary condition' for any two cubic regions on the same level of the trees to contain any desirable pairs of points (at distance P P i j ). We are going to refer to the condition defined in Theorem 1 as 'Necessary Condition 1' in the sequel. If two cubes do not satisfy the conditions of this theorem, no pairs of points from them could possibly match the edge P P , i j ( ) and will be rejected. That is why we consider this to be a rejection condition for pairs of cubes. By contrast, we have the following 'Sufficient Condition 2' for all pairs of points from two leaf cubes to be desirable (an acceptance condition).
Notice that the condition of Theorem 2 can hold only when P P l d P P l 3 3 . Because the leaf cubes of the octrees must have length l l 2 T s Let t i be the octree generated from manifold A i n for 1, 2, , i = … . Algorithm 1 gives the pseudocode of the adaptive geometric search algorithm. Figure 2 illustrates the algorithm graphically.
Algorithmic complexity
The adaptive geometric search algorithm has three parts, building the octrees, adaptively searching every two octrees and the graph search. Let N be the number of sample points from each manifold. For convenience, we build all octrees with the same initial cube length l 0 . The time complexity of building an octree with initial cube length l 0 and minimum cube length l s is O l l N log / s 2 0 ( ( ) ). Next, we compute the time complexity of the adaptive search between any two octrees (without loss of generality) called t 1 , t 2 . Let the corresponding polygon edge length be l ⁎ . The last part of the algorithm is a graph search. Let s ij be the number of possible leaf cube pairs that also passed Sufficient Condition 2 between octrees t t , i j for i j n i j , 1, 2 , , , ∈ [ … ] < . We view the leaf cubes as vertices and possible pairs of them as undirected edges in the graph. If we want to produce all the desirable n-tuple cubes, then by induction it is easy to see that the upper bound on the time complexity is C S O s i j n ij 1 ( ) = (∏ ) ≤ < ≤ . In practice, we can probably do much better. Consider building a directed graph by giving directions to the edges to form an n-cycle of groups of cubes from t t t , , , n . For space reasons, we skip the algorithm details here.
In summary, we state the total time complexity of the algorithm as follows. In practice, we usually search for a triangle or a four-sided polygon as the target polygon, i.e. n 3 = or 4. When n 3 = , depending on the parameters , T η ε and N the computation time varies but all three terms in the complexity formula (above) are typically of the same order. When there are large numbers of possible pairs s i 's and/ or n 4 = , the term C S ( ) in the last term of the complexity formula (4.1) becomes the dominating term. The number of results s ij 's can be further reduced when we take optimal dihedral angles instead of uniform sampling from 0, 2π [ ] . Full proofs of the above theroms and lemmas can be found in the supporting information.
Results
Our algorithm can be applied to many different problems in macromolecular modeling and design. It efficiently solves the problem of searching for a certain n-polygon among n sets of points in 3D with error tolerance T ε and an approximation margin η. We present three use cases where our algorithm's improved efficiency (run times that are in some cases many thousands of times faster than previous approaches) improves the scaling of the overall task, enabling the use of larger template/target structural patterns.
Scaffold matching: designing OOPs to inhibit MDM2-p53 interface
PPIs mediate many cellular functions and a small number of residues that make significant contributions to the binding affinity of the PPI (deemed 'hotspot' residues) in turn underlay these protein interfaces. Design tasks aimed at protein interfaces abound, as discussed above Fleishman et al. designed an influenza HA binder. Interest in using smaller, easy to synthesize, non-protolyzable macromolecules (called foldamers) as potential therapeutic candidates continues to rise as these systems become more synthetically (and computationally) accessible to a broader community. Foldamer backbone chemistries abound and finding a foldamer backbone type that is well matched to a particular set of interface hotspot residues interface will prove to be a future challenge. Here, we recapitulate an OOP foldamer designed by Drew and coworkers that mimics P53 and can disrupt the P53/MDM2 interaction (Fig. 3), which relies on three hotspot residues on P53 that constitute the majority of the binding affinity for MDM2 (Fig. 3A).
There are two parts of the algorithm. In Step 1, we search through all possible backbones for a matching triangle to the target triangle. In Step 2, for every match result from Step 1, the connecting atom's bond angles are checked against the optimal bond angle. If a match passes Step 2, it is returned as a final result. Otherwise, we continue the iteration in Step 1.
The target triangle is made up of Cβ's of the hotspot residues (Fig. 3C). The algorithm simply searches through the possible takeoff position combinations, four triangles in this example (Fig. 3B), from every backbone for a match in shape within the error bound. Notice that in this case, all Cβ atoms are fixed due to the short lengths of hotspot residues. With longer hotspot residues, there will be a manifold of all the possible Cβ atoms for each hotspot residue. For every possible triplet of take-off positions, there are eight possible D and L-enantiomers. So, for each of these 32 possibilities, we apply adaptive geometric search to find all matches.
Once we have the matching shapes, we calculate the corresponding matrices R's of rotation and translation such that after applying these transformations R's backbones are connected onto the hotspot residues at atoms Cβ's. Finally, we check if the bond angles at the connecting atoms (e.g. N, Cα and Cβ for leucine) are within some error bound to the optimal bond angles (indicated by the CHECKANGLE function in Algorithm 2).
Let A i be the manifold of possible positions of the connecting atom on the ith hotspot residue. For example, in Fig. 3C points in Fig. 2 An illustration of the adaptive search between two octrees. Dotted lines point out the possible cube pairs on each level that pass the Necessary Condition 1. Solid lines link the desirable leaf cube (gray nodes) pairs that pass the Sufficient Condition 2.
colors are sampled from manifolds A A , 1 2 and A 3 respectively. Let P j be the jth polygon of the backbone take-off position combination and for example, there are 4 × 17 of them in Fig. 3B. Let B P denotes the atoms' position matrix corresponding to the backbone where the target polygon P comes from. Let S i denotes the atoms' position matrix for the ith residue. Let δ be the distance error bound and A δ be the angle error bound. Then we describe in pseudocode Algorithm 2. In the adaptive geometric search part of Algorithm 2, the possible candidate pairs are screened out at least exponentially fast as we search down the octrees (Fig. 4). Let C denotes the time complexity for adaptive geometric search. Recall that m is the number of target polygons from backbone take-off site combinations. Then the time complexity of the scaffold matching algorithm is O Cm .
( )
In the search process, we scored all the possible matches by the rootmean-square deviation (RMSD) values for both shape match and angle match in Fig. 5. Our algorithm picked the candidate at the origin (this being identical to the correct conformation that led Lao et al. to low nanomolar inhibitors of this interface). In Fig. 3D, we show this best design for the OOP backbone of the hotspot residues. The run time for the initial geometric search (step on in this design protocol) is 0.02 to 0.12 s, whereas running the same design and producing the same results using the previously described Rosetta codes (the scripts from Lao et al.) takes~18 min (a speedup of >9 000-fold).
Peptoid design: design of new metal binding sites
Proteins and other macromolecules often coordinate metal ions to aid conformational stability or carry out chemical reactions. Proteins that bind Zn 2+ ions often use four residues (most often histidine, cysteine or aspartic acid) to coordinate the zinc ion in a tetrahedral arrangement (Hsin et al., 2008). We next tested our algorithm by designing a peptoid for capturing zinc ions. The binding sites we target in this example are three sulfur atoms lying on the vertices of an equilateral triangle. The search space includes 6-mer, 8-mer and 9-mer scaffolds (peptoid data bank codes 07AA1-6-C, 07AA2-8-C (Shin et al., 2007) and 12AC2-9-C (Butterfoss et al., 2012), respectively) as the backbone and 3-aminopropyl-1-thiol groups as side chains of residues (Fig. 6). Low-energy matches were identified for each scaffold and commonly found to be comprised of alternating residue positions or sequential positions on the narrow end of the macrocycle.
We sampled eight dihedral angles per residue with different lengths of side chains (n = number of carbon atoms), different error values. We recorded the run time to find the first valid target polygon on Intel Core 3.5 GHz (Table I).
Loop modeling
Accurately modeling protein loops is a difficult problem due to their flexibility due to their and lack of regular structure (Mandell et al., 2009). Computational modeling of loops generally involves defining the loop region (the residues that are flexible), anchor or pivot positons residues beyond which the protein structure remains fixed, and a cut point position at one of the residues in the loop region that splits the loop into two parts. A structural perturbation is made to one side of the loop-resulting in a break in the loop at the cut point-and the loop modeling protocol modifies the conformation of the other side of the loop in an attempt to close the break (Stein and Kortemme, 2013). The abstraction of the problem can be described as follows. Given two fixed points in 3D called pivots and two vectors (the take-off vectors), construct the loop from pivot 1 to pivot 2 with k residues with the type N-Cα-C such that the loop has a low energy and it fits in the designated space (Fig. 7A). Biologically relevant loops can vary greatly in length; the H3 CDR loop in human antibodies, for example, can vary from 5 to 26 residues (North et al., 2011). The difficulties of the problem using a direct computation stem from exponential growth in the number of possible loop conformations as a function of loop length, k. We divide the loop into two semiloops by the midpoint or the closest point to the midpoint between two residues. The designated space where the loop resides within can be discretized into cubes of a certain size. We precompute all conformations of a single residue and store the resulting angles and x y z , , coordinates after discretization and encoded as a unique integer. Then we compute and store the table where two residue conformations can connect appropriately, that is, the end atom of one residue and the beginning atom of the other residue lie in the same cube and the two bonds form an angle within the error bound from the optimal bond angle. Now using the precomputed residue conformations and matching table, we develop Fig. 4 Averaging over 60 runs on different octree pairs, we show the number of candidate point pairs goes down at least exponentially as we select only those pairs of cubes that pass the Necessary Condition 1. The best exponential fit to the data is shown. Fig. 5 RMSD of all possible OOP backbones matches with the hotspot residues' side chain positions. The candidate at the origin is a perfect match for both (shape and angle) to the hotspot residues we aim to minimize (use as a template for design) and is analogous to a template used in previously reported successful experimental designs.
the two semi-loops. Let the number of residue conformations be M r and the number of cubes in the lattice space M c . After developing each residue, we collapse the end positions that fall into the same cube and sharing the same last bond angle and store all intermediate results for the purpose of producing final results in backtracking. After the two semi-loops are developed, we have the end atoms of both sides and their spatial intersections. The angles are checked to eliminate from the intersection cubes those that deviate outside the error bound from the optimal bond angle there. Starting from the matched cubes in the middle of the loop, now we backtrack in both sides to the pivots and produce as many results as desired (effectively allowing for efficient sampling of a large number of constraint- compliant loop designs). In the first experiment, we computed a 12residue loop, developing 1000 conformations for each residue and 121 by 121 by 121 cubes in the designated space, setting cube length to 0.1 and maximum bond angle errors to within 0.2 rad. On a 1.3-GHz Intel Core M with 8 GB memory, our algorithm ran a total of 3.6 min to produce the first result (Fig. 7C). The development of each semi-loop took 82 s and the matching in the middle took 20 s. Keeping the number of conformations per residue, error bounds and the cube size, we enlarge the number of cubes to 171 by 171 by 171 to compute for 17-residue loops. On a 2.0-GHz Intel Xeon E5-2620 CPU with 128 GB memory, our algorithm ran a total of 35.5 min to produce the first result (Fig. 7D). The development of each semiloop took 11 min and the matching in the middle took 28 s.
Discussion
We have presented an adaptive method for finding matches between target geometric patterns (that represent protein and peptidomimetic design goals) and scaffolds (which can serve as the biosynthetic or organic synthesis method for positioning side chains in the desired/ target geometry). In the protein, enzyme, and peptidomimetic design communities, such geometric search problems are increasingly becoming limiting steps in design processes. This trend will increase as we scale to larger target patterns and as we compare to growing databases of proteins, peptidomimetic structures and other scaffolds. Thus, improving the speed of geometric function scaffold search algorithms makes a substantive contribution to biomimetic design. | 7,109.2 | 2017-01-11T00:00:00.000 | [
"Biology"
] |
A new genus of coprophagous water scavenger beetle from Africa (Coleoptera, Hydrophilidae, Sphaeridiinae, Megasternini) with a discussion on the Cercyon subgenus Acycreon
A new genus of coprophagous beetle, Evanesternum gen. n. (Hydrophilidae: Sphaeridiinae: Megasternini), is described in order to accommodate Cercyon (Acycreon) pulsatus d’Orchymont, 1937 from the Republic of South Africa and the Democratic Republic of Congo. A detailed description is provided along with habitus photographs, line drawings and SEM micrographs of relevant diagnostic characters. The new genus possesses the tribal synapomorphies of Megasternini but bears several unique morphological characters which are discussed in detail. The morphology of the remaining three species classified in the subgenus Acycreon d’Orchymont, 1942 (i.e. C. punctiger Knisch, 1921, C. collarti d’Orchymont, 1942 and C. apiciflavus Hebauer, 2002), is illustrated in order to provide evidence that Acycreon is an assemblage of morphologically dissimilar and likely not related species. An identification key to the Megasternini genera and subgenera known from the Republic of South Africa is presented.
Introduction
Water scavenger beetles (Hydrophilidae) are mainly known as species associated with a wide variety of aquatic habitats; the majority of the species (ca.65%) inhabit aquatic and semi-aquatic environments (Short andFikáček 2011, Bloom et al. 2014).Aquatic species form the vast majority of the subfamilies Hydrophilinae, Chaetarthriinae, Enochrinae and Acidocerinae (Short and Fikáček 2013).However, more than a third of known hydrophilid beetles, mostly belonging to the subfamilies Cylominae and Sphaeridiinae, have colonised terrestrial habitats, typically those with large amounts of decaying organic matter (e.g.tropical forest leaf litter and rotten plant debris).Several groups are also associated with vertebrate dung (usually excrement from large herbivorous mammals) and many taxa are exclusively coprophagous.Dung provides an abundant and rich source of nutrients to a wide range of arthropods, with beetles being amongst them, along with Diptera, the most conspicuous and diverse.Hydrophilids, together with scarabs, are amongst the most important coprophagous beetles (Holter 2004).
The African continent is well known for its abundance of large mammal species, which in turn serve as a source of dung to be exploited.The diversity and abundance of large mammals (especially herbivores) likely promoted the diversification of beetle groups like scarab dung beetles and terrestrial hydrophilids of the subfamily Sphaeridiinae, which are both abundant and diverse in Africa (Davis and Scholtz 2001;Hebauer 2006).The tribe Megasternini is the most diverse group of hydrophilid beetles associated with terrestrial environments, comprising moreover the vast majority of obligatory coprophagous hydrophilid species.Seventeen genera are known to occur in Africa, of which Cercyon Leach, 1817, Pachysternum Motschulsky, 1863 and Cryptopleurum Mulsant, 1844 are especially species-rich.Nine small genera are endemic for the Afrotropical region: Cercillum Knisch, 1921 (Central and Southern Africa), Cyrtonion Hansen, 1989 (Central Africa), Delimetrium Hansen, 1999 (Republic of South Africa), Parastromus Balfour-Browne, 1948 (Central and Southern Africa), Pelocyon Balfour-Browne, 1950 (Central andSouthern Africa), Pseucyon d'Orchymont, 1948 (Ethiopia), Quadristernum Balfour-Browne, 1950 (Rwanda-Burundi), Acaryon Hebauer, 2003 (Madagascar) and Colerus Hansen, 1999 (Madagascar).All of these are rare in collections and poorly known.They have rarely been mentioned after their description (Balfour-Browne 1948, 1950;Knisch 1921;d'Orchymont 1948) or have been described very recently (Hansen 1999b, Hebauer 2003) and few of them (Colerus, Quadristernum) are known only from one or a few specimens.
During the recent field work in the Cape region of the Republic of South Africa, the authors discovered a morphologically aberrant tiny representative of the Megasternini which represents an undescribed genus.The review of previously known South African species revealed that the species is already described, but misclassified as part of the subgenus Acycreon d'Orchymont, 1942 of the genus Cercyon.In this paper, the generic assignment of this species is re-evaluated, a new genus described for it, the morphology of the remaining species assigned at the moment to Cercyon (Acycreon) is reviewed and the taxonomic composition of this subgenus is discussed.
Materials and methods
This study is based on the specimens deposited in the following entomological collections:
MNRJ
Museu Nacional, UFRJ, Rio de Janeiro, Brazil (B.During the field work, about 15 kg of relatively fresh horse and cow dung was collected in a thick 60 litre plastic bag.The bag was closed and an air buffer was left above the excrement.The bottom of the bag was perforated with 1 cm holes using a knife.This bag was enclosed in another intact bag and hung above the ground in a shaded area.The beetles accumulated overnight in the second bag and were collected in 96% ethanol, without needing to check the excrement by hand.
Part of the specimens examined was dissected, with genitalia embedded in a drop of ethanol-soluble Euparal resin on a small piece of glass glued to cardboard attached below the respective specimen.
Habitus photographs were taken using a Canon D-550 digital camera with attached Canon MP-E65 mm f/2.8 1-5 macro lens.Pictures of genitalia were taken using a Canon D1100 digital camera attached to an Olympus BX41 compound microscope; pictures of different focus were combined in Helicon Focus software.Scanning electron micrographs were taken using Hitachi S-3700N environmental electron microscope at the Department of Paleontology, National Museum in Prague.Pictures used for plates were adapted in Adobe Photoshop CS6.All original pictures including additional views, not presented in this paper, are published and freely available on Flickr (https://www.flickr.com/photos/142655814@N07/sets/72157681650620964) and submitted to Zenodo repository (https://zenodo.org/) under https://doi.org/10.5281/zenodo.806765.
Mesothorax.Mesoventrite completely fused with an episternum; anterior collar of mesothorax narrow.Median portion of mesoventrite elevated as a mesoventral plate slightly overlapping anterior margin of metaventrite; plate well defined posteriorly as a broad, semi-elliptical tablet abruptly vanished in anterior half leaving only a narrow median ridge.Grooves for reception of procoxae well defined by a conspicuous carina, short, transverse (Fig. 4g).Mesepimeron very narrow, widening laterad.Mesocoxal cavities moderately narrowly separated.Scutellar shield small, semi-elliptical, 1.7× as long as wide.Elytra (Fig. 1a, b) weakly convex, narrowly explanate laterally, each elytron bearing 10 series, series 1-9 consisting of setiferous punctures as large as interval punctures but surrounded by a foveolate depression; punctures in series situated in longitudinal impressed sulci; series 1-4 and 9 reaching apex, series 5 and 8 enclosing series 6-7 subapically, series 10 reduced both anteriorly and posteriorly; epipleuron horizontal, weakly gradually narrowing posteriad, accentuating about half the length of metaventrite, vanishing shortly after the posterior margin of the metaventrite, bearing sparse short setae (Fig. 3).
Metathorax.Metaventrite (Fig. 4h) with anterior rim narrow, widening on anterolateral corners; mesal elevate area flat and very wide, almost reaching lateral margins; anterior and posterolateral corners distinctly rugose, with short setae.Femoral lines and anterolateral ridges absent.Metanepisternum ca.5× as long as wide, with anterior oblique ridge, metepimeron with minute ventral portion.Metafurca well developed.Metathoracic wings well developed, with transverse vein r4 arising from basal portion of radial cell, RP rather long, reaching ca.halfway to wing base, basal cubito-anal cell small, closed, wedge cell absent, transverse vein mp-cua joining to MP 3+4 +CuA 1+2 ; anal lobe not defined.
Etymology.The generic name is derived from evanescere (Latin, "to vanish") and sternum (Greek, "chest") which refers to the anteriorly vanishing mesoventral plate.The gender is neutrum.
Metathorax.Metaventrite (Fig. 4h) with raised area very wide, almost reaching lateral margins, about 1.5× as wide as long, rather roughly punctate, punctures transverse, resembling those on dorsal surfaces, with very small fine setae, surface finely squamose.
Abdomen with five ventrites.Ventrite 1 with median longitudinal carina present, slightly narrowing posteriad, briefly projecting posteriorly in both sexes (Fig. 4i); ventrite 5 with rounded apex in both sexes, with a group of longer setae on apex.
Genitalia.Median projection of sternite 9 (Fig. 1f ) subtruncate apically, without subapical setae, median portion narrowing posteriorly, with posterior end distinctly acute, lateral struts joined at mid length of the median projection.Phallobase (Fig. 1c) slightly longer than parameres, asymmetrically narrowing basally, base slightly curved.Parameres weakly narrowing apically, subsinuate and briefly widened near apex.Median lobe (Fig. 1d) narrow, parallel-sided throughout, apex acuminate, with small parallel apical projection, gonopore moderately large, situated subapically.Biology.Recently collected specimens from Western Cape were extracted from cow and horse dung in a small farm close to a river (Fig. 5b).Other hydrophilids from the same dung collected in abundance were Sphaeridium caffrum Castelnau, S. abbreviatum Boheman, Pachysternum capense (Mulsant) and three small Cercyon species (one belonging to the Cercyon nigriceps group and other two unidentified species).Other specimens examined were collected in unspecified type of dung while specimens from the Democratic Republic of Congo were collected from elephant excrement.
Key to the Megasternini genera and subgenera known from the Republic of South Africa
The key includes the genera recorded from the Republic of South Africa by Hansen (1999a) and Hebauer (2006).Due to unclear limits of Cercyon and Parastromus, the subgenera of Cercyon recorded from RSA are also included into the key to allow the identification of as many species as possible.Re-description.2.0-2.5 mm long, 1.8-1.9×as long as wide, 2.9-3.0× as long as high.Integument shining (Fig. 6a).Colouration reddish-brown with pale palpi.Eyes without thickened ridge at posterior margin (Fig. 7a.).Mentum subtrapezoid, with anterior mar- gin almost straight-lined (Fig. 7b).Pronotum with lateral margins moderately impressed (Fig. 7e).Pronotum and elytra deeply punctate without conspicuous microsculpture (Fig. 7c-d).Mesoventral elevation fusiform, short, not reaching anterior margin, procoxal rests not defined by transversal carinae (Fig. 7f).Raised portion of metaventrite about as long as wide, distinctly deeply punctate (Fig. 7h).Median lobe of aedeagus with parameres moderately broad, phallobase about 1.3× as long as parameres (Fig. 6b).Sternite 9 with median projection strongly narrowing anteriorly, lateral struts almost reaching base (Fig. 6c).
Distribution.Only known from the Democratic Republic of Congo.Redescription.1.8 mm long, 1.4× as long as wide, 2.4× as long as high.Integument shining (Fig. 9a).Colouration of head, pronotum, ventral surfaces and legs dark reddish-brown with pale palpi and antennae, elytra black with apex testaceous.Eyes without thickened ridge at posterior margin.Mentum subtrapezoid, with anterior margin straight-lined (Fig. 9b).Pronotum with lateral margins narrowly impressed (Fig. 9e).Pronotum and elytra deeply punctate without conspicuous microsculpture (Figs 9d-e).Mesoventral elevation fusiform, short, not reaching anterior margin, procoxal rests defined by oblique carinae, with a deep setose pore on each side of the mesoventral elevation (Fig. 9g).Raised portion of metaventrite about as long as wide, distinctly deeply punctate (Fig. 9h).
Cercyon (Acycreon) apiciflavus
Distribution.Only known from the type locality in Nepal.Comments.The species differs from the remaining two Acycreon species in the rather globular and widely rounded body.In these aspects, as well as in the morphology of the ventral parts of thorax, it is very similar to several undescribed species from the Chinese provinces, Yunnan and Sichuan (S.Ryndevich, in prep.).The deep, rounded setose pores on the mesoventrite on the side of the central elevation have not been recorded in any other Cercyon species and, along with the other morphological features of this species, suggest its distant relation to the type species of Acycreon subgenus.
Evanesternum pulsatum comb.n. differs from C. punctiger (type species of Acycreon) and the other two species in the following characters: (1) male 9 th sternite with lateral struts attached ca. at midlength (Fig. 1f ) (basally or sub-basally in C. punctiger and C. collarti (Fig. 6c, f ), not known for C. apiciflavus); (2) rugose-reticulate dorsal integument of head and pronotum (Fig. 4d-e) (without any microsculpture in Acycreon (Figs 7c-d, 8c-d); (3) medial raised part of metaventrite nearly reaching lateral margin (Fig. 4h) (present only mesally and not reaching laterally in Acycreon (Figs 7h, 8g, 9h); (4) thickened ridge running along posterior ventral margin of eye (Fig. 4a) (without such ridge in Acycreon (Figs 7a, 8a); (5) procoxal rests of mesoventrite defined by sharp transverse ridges (Fig. 4g) (in Acycreon without such ridges (Figs 7g, 8f ) or with oblique ridges (Fig. 9g)); (6) pronotum deeply impressed along lateral margin (Fig. 4e) (at most weakly impressed in Acycreon; Evanesternum pulsatum comb.n. was mentioned by d' Orchymont (1942) as having the lateral portion of the pronotum similar to that of C. punctiger, but the margin is much deeper and broader in Evanesternum (compare Fig. 3h with Figs 7e and 9e).The above-mentioned morphological differences between Evanesternum pulsatum comb.n. and the members of Acycreon are substantial and it is believed that E. pulsatum is not congeneric with them.Evanesternum gen.n. may, at first sight, resemble the small-bodied African megasternine genera Delimetrium, Pelocyon and Pseucyon or the smaller-sized species of the genus Cercyon (e.g. C. minax Balfour-Browne).The simple prosternum (i.e.not elevated medially) easily distinguishes it from all these genera except Cercyon.Evanesternum would be keyed out asCercyon in the key to genera by Hansen (1991), which remains to date the most comprehensive resource for identification of Megasternini.The new genus has all the diagnostic characters of Megasternini: antenna with compact club, maxilla of male with sucking disc on galea; prosternum with antennal grooves, first abdominal ventrite carinate medially.However, it can be distinguished from Cercyon (as well as other megasternine genera) by the following combination of characters: (1) the mesoventral plate well-defined only posteriorly, not properly demarcated anteriorly (always well-demarcated as a complete plate in Cercyon); (2) mesal bare portion of metaventrite extending far laterally, covering the majority of the metaventral surface (always confined to median portion of metaventrite in Cercyon); (3) pronotum with deep groove along the lateral margin (absent in Cercyon); and (4) the morphology of male sternite 9 with very short lateral struts attaching at mid-length of the medial sclerite (lateral struts long and attaching basally in Cercyon and all genera of the Cercyon-group of genera).The morphology of the male terminalia and surrounding structures (sternite 9, articulation of the median lobe and parameres) seems to be phylogenetically informative in Megasternini, corresponding to the clades recognised by molecular phylogenetic analyses (Short and Fikáček 2013, Arriaga-Varela unpubl. data).Hence, the very unusual morphology of male sternite 9 of Evanesternum may indicate its rather isolated position in the Cercyongroup of the Megasternini.Preliminary analyses of molecular data (Arriaga-Varela, in prep.)seem to support this hypothesis.
Despite the mesal portion of prosternum being flat in Evanesternum, it bears a very faint demarcation of the mesal part with respect to the lateral parts by irregular diagonal sulci.These sulci 'define' the mesal portion which can be also inferred from the differences in the sculpture on the posterior margin and from the long setae present on the anterior margin.The prosternum as found in Evanesternum may possibly represent an intermediate condition between the flat prosternum (as present, for example, in Cercyon) and the mesally demarcated and elevated one (as present in genera Cryptopleurum, Cyrtonion, Delimetrium, Pachysternum, Pelocyon and Pseucyon in Africa) or a highly reduced version of a mesally demarcated prosternum.
Of the valid subgenera, Cercyon s. str.contains the largest number of species (slightly over 200) which are morphologically very diverse, indicating that the subgenus is likely to be an artificial assemblage of species rather than a monophyletic group.The remaining ten subgenera were created to accommodate some morphologically aberrant Cercyon species.They contain many less species and since they were mostly defined by some unique characters of the meso-or metaventrite, they more likely represent monophyletic groups.Still, the delimitation of some of them is unstable, resulting in frequent changes in subgeneric assignments of some species.This confusion concerns especially the subgenera Clinocercyon d'Orchymont, 1942 defined by the oblique, rath-er than horizontal epipleura, which contains a very diverse assemblage of species from the Old World, with some species repeatedly moved in and in and out (e.g.Ryndevich 2007) and Conocercyon Hebauer, 2003 defined by a shape of the postcoxal ridge of the metaventrite (originally described for a few species from Madagascar and Seychelles, with two eastern Palaearctic species assigned to it later: Hebauer 2003, Ryndevich 2007, Hoshina 2008).Three subgenera, Himalcercyon Hebauer, 2002, Oedocercyon d'Orchymont, 1942and Prostercyon Smetana, 1978, are monotypic (containing only one species).The results of this review of the subgenus Acycreon corresponds to the situation observed in these subgenera, i.e. it seems that the subgenus is defined by a single (and moreover weakly defined) character which picks up superficially similar but likely not closely related species.It is however surprising that this problem also concerns Acycreon, i.e. one of the smallest subgenera of Cercyon and clearly indicates that even the small subgenera need to be carefully tested for monophyly in future molecular phylogenetic studies.
After the transfer of C. pulsatus to Evanesternum, Acycreon contains three species, C. (A.) punctiger, C. (A.) collarti and C. (A.) apiciflavus.The mesoventrite of all of them forms a well-defined fusiform plate with a well-marked acute anterior tip, the plate being however rather short, ca.half as long as the length of the mesoventrite.The similarity of all three species hence concerns the relative length of the mesoventral plate, rather than the absence of the plate-like elevation mentioned by previous authors.In all other aspects, the Acycreon species are not very similar to each other in terms of external and genital morphology (compare Figs 6a-f, 7a-h, 8a-g) which indicates that they are likely not closely related and potentially not forming a monophyletic group.The distribution of the species (Oriental region versus tropical Africa) is also congruent with this view.Hence, even after excluding C. pulsatus, Acycreon consists of quite dissimilar species and may still be an artificial rather than natural assemblage.The concept of Acycreon needs to follow the morphology of its type species (C.punctiger) and can be only adapted after the phylogenetic position of that species has been resolved.
Figure 1 .
Figure 1.Evanesternum pulsatum (d'Orchymont) a dorsal habitus b lateral habitus c tegmen of aedeagus d median lobe of aedeagus e detail of apex of median lobe f 9 th sternite.
Figure 4 .
Figure 4. Evanesternum pulsatum (d'Orchymont) a ventral view of head b labium c prosternum d detail of head surface e detail of pronotal surface and lateral margin f detail of elytral surface g mesoventral plate h metaventrite i ventral view of abdomen.
Figure 5 .
Figure 5. Evanesternum pulsatum (d'Orchymont) a distribution map b habitat collecting locality: Summerset Getaway Farm, Western Cape, Republic of South Africa c beetles collected from dung in Summerset Getaway Farm, Western Cape, Republic of South Africa.Red arrow is indicating an Evanesternum pulsatum (d'Orchymont) specimen.
Figure 7 .
Figure 7. Cercyon (Acycreon) punctiger Knisch a ventral view of eye b labium c detail of elytral surface d detail of head surface e detail of pronotal surface and lateral margin f prosternum g mesoventral plate h metaventrite.
Figure 8 .
Figure 8. Cercyon (Acycreon) collarti d'Orchymont a ventral view of eye b labium c detail of elytral surface d detail of head surface e prosternum f mesoventral plate g metaventrite.
Figure 9 .
Figure 9. Cercyon (Acycreon) apiciflavus Hebauer a dorsal habitus b labium c detail of elytral surface d detail of head surface e detail of pronotal surface and lateral margin f prosternum g mesoventral plate (arrow pointing to the setose pore) h metaventrite. | 4,289.6 | 2018-11-01T00:00:00.000 | [
"Biology"
] |
HCM: Hardware-Aware Complexity Metric for Neural Network Architectures
Convolutional Neural Networks (CNNs) have become common in many fields including computer vision, speech recognition, and natural language processing. Although CNN hardware accelerators are already included as part of many SoC architectures, the task of achieving high accuracy on resource-restricted devices is still considered challenging, mainly due to the vast number of design parameters that need to be balanced to achieve an efficient solution. Quantization techniques, when applied to the network parameters, lead to a reduction of power and area and may also change the ratio between communication and computation. As a result, some algorithmic solutions may suffer from lack of memory bandwidth or computational resources and fail to achieve the expected performance due to hardware constraints. Thus, the system designer and the micro-architect need to understand at early development stages the impact of their high-level decisions (e.g., the architecture of the CNN and the amount of bits used to represent its parameters) on the final product (e.g., the expected power saving, area, and accuracy). Unfortunately, existing tools fall short of supporting such decisions. This paper introduces a hardware-aware complexity metric that aims to assist the system designer of the neural network architectures, through the entire project lifetime (especially at its early stages) by predicting the impact of architectural and micro-architectural decisions on the final product. We demonstrate how the proposed metric can help evaluate different design alternatives of neural network models on resource-restricted devices such as real-time embedded systems, and to avoid making design mistakes at early stages.
INTRODUCTION
Domain-specific systems were found to be very efficient, in general, and when developing constrained devices such as IoT, in particular. A system architect of such devices must consider hardware limitations (e.g., bandwidth and local memory capacity), algorithmic factors (e.g., accuracy and representation of data), and system aspects (e.g., cost, power * Equal contribution. Figure 1: Our 3×3 kernel 8-bit processing engine (PE) layout using the TSMC 28nm technology. The carry-save adder can fit 12-bit numbers, which is large enough to store the output of the convolution.
envelop, battery life, and more). Many IoT and other resourceconstrained devices provide support for applications that use convolutional neural networks (CNNs). Such algorithms can achieve spectacular performance in various tasks covering a wide range of domains such as computer vision, medicine, autonomous vehicles, etc. Notwithstanding, CNNs contain a vast number of parameters and require a significant amount of computation during inference, thus monopolizing hardware resources and demanding massively parallel computation engines; see teh example shown in Fig. 1.
These requirements have led to great interest in using custom-designed hardware for efficient inference of CNNs that would allow the promise of neural networks to be used in real-life applications by deploying them on low-power edge devices. Developing such systems requires a new set of design tools due to the tight entanglement between the algorithmic aspects, the chip architecture and the constraints the end product needs to meet. In particular, great efforts were made to develop low-resource CNN architectures [14,24,27,33]. One example of such architectural changes is the splitting of the regular 3 × 3 convolutions into a channel-wise 3 × 3 convolution followed by a 1 × 1 one. Another way to reduce the computational burden is to quantize the CNN parameters, weights and activations, employing low-bit integer representation of the data instead of the expensive floating point representation. Recent quantization-aware training schemes [8,10,16,34,35] achieve near-baseline accuracy for as low as 2-bit quantization. The benefit of quantizing the CNN is twofold: both the number of gates required for each multiplyaccumulate (MAC) operation and the amount of routing are reduced. The decision regarding which algorithm to choose may depend on the architecture (e.g., FPGA or ASIC), the accuracy requirements, and their impact on performance and power. Thus, the architect needs to make these fundamental decisions early in the developing process and no existing tool can help predict these design factors ahead of time.
The impact of the high-level structure of the accelerator, e.g., the type of CNN levels and the representation of the operands, on the power, the area and the performance of the final product needs to be defined and predicted at an early stage of the project. Recent research has shown that ASIC-based architectures are the most efficient solution for CNN accelerators both in datacenters [6,17,22] and in realtime platforms [5,11,25]. Accordingly, we demonstrate the proposed metric and design tool on an implementation of a streaming [23] ASIC-based convolutional engine. Nevertheless, our methodology can be applied for the evaluation of other types of architectures, such as FPGA-based accelerators [1,2,29]. In both cases, the development process includes an important trade-off between the logical gates area, their routing on the silicon and the performance of the resulting system. Unfortunately, all these parameters also depend on the representation of the data, and its impact on both communication and computation. To date, there is no quantitative metric for this trade-off available at the design stage of the CNN accelerator and no tool exists that can assist the architect to predict the impact of high level decisions on the important design parameters. Ideally, the designer would like to have an early estimation of the chip resources required by the accelerator as well as the performance, accuracy and power it can achieve.
A critical difficulty in trying to predict the design parameters for CNN-based systems is the lack of a proper complexity metric. Currently, the most common metric for calculating the computational complexity of CNN algorithms is the number of MAC operations denoted as OPS (or FLOPS in case of floating-point operations). This metric, however, does not take into account the data format or additional operations performed during the inference, such as memory accesses and communication. For that reason, the number of FLOPS does not necessarily correlate with runtime [18] or the required amount of computational resources. This paper proposes to use a different metric for assessing the complexity of CNNbased architectures: the number of bit operations (BOPS) as presented by Baskin et al. [3]. We show that BOPS is well-suited to the task of comparing different architectures with different weight and activation bitwidths.
Contribution.
This paper makes the following contributions: Firstly, we study the impact of CNN quantization on the hardware implementation in terms of computational resources and memory bandwidth considerations. Specifically, we study a single layer in the neural network.
Secondly, we extend the previously proposed computation complexity for quantized CNNs, termed BOPS [3], with a communication complexity analysis to identify the performance bottlenecks that may arise from the data movement.
Thirdly, we extend the roofline model [32] to accommodate this new notion of complexity. We also demonstrate how this tool can be used to assist architecture-level decisions at the early design stages.
Lastly, we implement a basic quantized convolution block with various bitwidths on 28nm processes to demonstrate an accurate estimation of the power/area of the hardware accelerator. This allows changing high-level design decisions at early stages and saves the designer from major mistakes that otherwise would be discovered too late. We compare our estimations with previous approaches and show significant improvement in accuracy of translation between algorithmic complexity to hardware resource utilization.
The rest of the paper is organized as follows: Section 2 reviews the related work; Section 3 describes a proposed hardware-aware complexity metric; Section 4 provides roofline analysis of CNN layer design using the proposed metric; Section 5 provides the experimental results using common CNN architecture and Section 6 concludes the paper.
RELATED WORK
In this section, we provide an overview of prior work that proposed metrics for estimating the complexity and power/energy consumption of different workloads, focusing on neural networks. The most commonly used metric for evaluating computational complexity is FLOPS [19]: the amount of floating point operations required to perform the computation. In the case of integer operations, the obvious generalization of FLOPS is OPS, which is just the number of operations. A fundamental limitation of these metrics is the assumption that the same data representation is used for all operations; otherwise, the calculated complexity does not reflect the real one. Wang et al. [31] claim that FLOPS is an inappropriate metric for estimating the performance of workloads executed in datacenters and proposed a basic operations metric that uses a roofline-based model, taking into account the computational and communication bottlenecks for more accurate estimation of the total performance.
In addition to general-purpose metrics, other metrics were developed specifically for evaluation of neural network complexity. Mishra et al. [20] define the "compute cost" as the product of the number of fused multiplyâȂŞadd (FMA) operations and the sum of the width of the activation and weight operands, without distinguishing between floating-and fixedpoint operations. Using this metric, the authors claimed to have reached a 32× "compute cost" reduction by switching from FP32 to binary representation. Still, as we show further in our paper, this is a rather poor estimate for the hardware resources/area needed to implement the computational element. Jiang et al. [15] notes that a single met-ric cannot comprehensively reflect the performance of deep learning (DL) accelerators. They investigate the impact of various frequently-used hardware optimizations on a typical DL accelerator and quantify their effects on accuracy and throughout under-representative DL inference workloads. Their major conclusion is that high hardware throughput is not necessarily highly correlated with the end-to-end high inference throughput of data feeding between host CPUs and AI accelerators. Finally, Baskin et al. [3] propose to generalize FLOPS and OPS by taking into account the bitwidth of each operand as well as the operation type. The resulting metric, named BOPS (binary operations), allows area estimation of quantized neural networks including cases of mixed quantization.
The aforementioned metrics do not provide any insight on the amount of silicon resources needed to implement them. Our work, accordingly, functions as a bridge between the CNN workload complexity and the real power/area estimation.
COMPLEXITY METRIC
In this section, we describe our hardware-aware complexity metric (HCM), which takes into account the CNN topology, and define the design rules of efficient implementation of quantized neural networks. The HCM metric assesses two elements: the computation complexity, which quantifies the hardware resources needed to implement the CNN on silicon, and the communication complexity, which defines the memory access pattern and bandwidth. We describe the changes resulting from switching from a floating-point representation to a fixed-point one, and then present our computation and communication complexity metrics. All results for the fixedpoint multiplication presented in this section are based on the Synopsys standard library multiplier using TSMC's 28nm process.
The impact of quantization on hardware implementation
Currently, the most common representation of weights and activations for training and inference of CNNs is either 32bit or 16-bit floating-point numbers. The fixed-point MAC operation, however, requires significantly fewer hardware resources, even for the same input bitwidth. To illustrate this fact, we generated two multipliers: one for 32-bit floatingpoint 1 and the other for 32-bit fixed-point operands. The results in Table 1 show that a fixed-point multiplier uses approximately eight time less area, gates, and power than the floating-point counterpart. Next, we generated a convolution with a k × k kernel, a basic operation in CNNs consisting of k 2 MAC operations per output value. After switching from floating-point to fixed-point, we explored the area of a single processing engine (PE) with variable bitwidth. Note that accumulator size depends on the network architecture: the maximal bitwidth of the output value is b w b a + log 2 (k 2 ) + log 2 (n), where n is number of input features. Since the extreme values are very rare, however, it is often possible to reduce the accumulator width without harming the accuracy of the network [6]. 1 FPU100 from https://opencores.org/projects/fpu100 2 shows the silicon area of the PE as a function of the bitwidth. We performed a polynomial regression and observed a quadratic dependence of the PE area on the bitwidth, with the coefficient of determination R 2 = 0.9999877. This nonlinear dependency demonstrates that quantization impact a network hardware resources is quadratic: reducing bitwidth of the operands by half reduces area and, by proxy, power approximately by a factor of four (contrary to what is assumed by, e.g., Mishra et al. [20]).
Computation
We now present the BOPS metric defined in Baskin et al. [3] as our computation complexity metric. In particular, we show that BOPS can be used as an estimator for the area of the accelerator. The area, in turn, is found to be linearly related to the power in case of the PEs.
The computation complexity metric describes the amount of arithmetic "work" needed to calculate the entire network or a single layer. BOPS is defined as the number of bit operations required to perform the calculation: the multiplication of n-bit number by m-bit number requires n · m bit operations, while addition requires max(n, m) bit operations. In particular, Baskin et al. [3] show that a k × k convolutional layer with b a -bit activations and b w -bit weights requires bit operations, where n and m are, respectively, the number of input and output features of the layer. The formula takes into account the width of the accumulator required to accommodate the intermediate calculations, which depends on n.
The BOPS of an entire network is calculated as a sum of the BOPS of the individual layers. Creating larger accelerators that can process more layers in parallel involves simply replicating the same individual PE design. In Fig. 3, we calculated BOPS values for the PEs from Fig. 2 and plotted them against the area. We conclude that for a single PE with variable bitwidth, BOPS can be used to predict the PE area with high accuracy. Next, we tested the predictive power of BOPS scaling with the size of the design. We generated several designs with variable bitwidths, b w = b a ∈ {4, 6, 8}, and variable numbers of PEs n = m ∈ {4, 8, 16} used to accommodate multidimensional inputs and outputs that typically arise in real CNN layers. Fig. 4 shows that the area depends linearly on the BOPS for the range of two orders of magnitude of total area with goodness of fit R 2 = 0.9980. We conclude that since BOPS provides a high-accuracy approximation of the area and power required by the hardware, it can be used as an early estimator. While the area of the accelerator depends on the particular design of the PE, this only affects the slope of the linear fit, since the area is still linearly dependent on the amount of PEs. An architect dealing with algorithms only can use definitions such as the number of input features and output features, kernel size etc. and get an early estimation how much power is needed to solve the network, without having any knowledge about VLSI constraints in advance. Using information such as a number of input/output features and kernel size, it is possible to immediately assess the amount of area the PEs occupy on the silicon.
Communication
Another important aspect of hardware implementation of CNN accelerators is memory communication. The transmission of data from the memory and back is often overlooked by hardware implementation papers [1,5,28] that focus on the raw calculation ability to determine the performance of their hardware. In many cases, there is a difference between the calculated performance and real-life performance, since reallife implementations of accelerators are often memory-bound [17,21,30].
For each layer, the total memory bandwidth is the sum of the activation and weight sizes read and written from memory. In typical CNNs used, e.g., in vision tasks, the first layers consume most of their bandwidth for activations, whereas in deeper layers that have smaller but higher-dimensional feature maps (and, consequently, a bigger number of kernels), weights are the main source of memory communication.
We assume that each PE can calculate one convolution result per clock cycle and the resulting partial sum is saved in the cache. In Fig. 5, we show typical memory access progress at the beginning of the convolutional layer calculation. At first stage, the weights and the first k rows of the activations are read from memory at maximal possible speed to start the calculations as soon as possible. After the initial data are loaded, the unit reaches a "steady state", in which it needs to read from the memory only one new input value per clock cycle (other values are already in the cache). We assume the processed signals to be two-dimensional (images), which additionally requires k new values to be loaded in the beginning of each new row.
Note that until the weights and the first activations are loaded, no calculations are performed. The overhead bandwidth of the pre-fetch stage can be mitigated by doing work in greater batch sizes, loading the weights once and reading several inputs for the same weights. By doing this, we minimize the penalty for reading the weights compared to reading the actual input data to perform the calculation. In the case of real-time processing, however, larger batches are not possible We focus on the latter real-time streaming regime in this paper because of its great importance in a range of applications including automotive, security, and finance. The memory access pattern depicted in Fig. 5 must be kept in mind when designing the hardware, since it may limit the performance of the accelerator and decrease its power efficiency.
ROOFLINE ANALYSIS
So far, we discussed the use of BOPS for the prediction of the physical parameters of the final product, such as the expected power and area. In this section, we extend the BOPS model to a system level, by introducing the OPS-based roofline model. The traditional roofline model, as introduced by Williams et al. [32], suggests depicting the dependencies between the performance (e.g., GFLOPS/second) and the operation density (the average number of operations per information unit transferred over the memory bus). Now, for each machine we can draw "roofs": the horizontal line that represents its computational bounds and the diagonal line that represents its maximal memory bandwidth. An example of the roofline for three applications assuming infinite compute resources and memory bandwidth is shown in Fig. 6. The maximum performance a machine can achieve for any application is visualized by the area below both bounds, shaded in green.
Since, as indicated in Section 3.1, FLOPS cannot be used for efficient estimation of the complexity of quantized CNNs, we introduce a new model that is based on the BOPS metric presented in Section 3.2. This model, to which we refer as the OPS-based roofline model, replaces the GFLOPS/s axis of the roofline plot with a performance metric more adequate for neural networks, e.g., number of operations per second (OPS/s), and the second metric that measures the computational complexity with operations per bit (OPS/bit). Using generic operations and bits allows plotting quantized accelerators with different bitwidths on the same plot.
As an example of the proposed approach, we use two different ResNet-18 layers (a deep layer, which is computationallyintensive, and an early one, which is memory-intensive) on four different accelerator designs: 32-bit floating-point, 32bit fixed-point, and quantized 8-bit and 4-bit fixed-point. The accelerators were implemented using standard ASIC design tools, as detailed in Section 5 and were built using the TSMC 28nm technology, using standard 2.4GHz DDR-4 memory Figure 6: Roofline example. In the case of App1, memory bandwidth prevents the program from achieving its expected performance. In the case of App2, the same happens due to limited computational resources. Finally, App3 represents a program that could achieve its maximum performance on a given system. with a 64-bit data bus.
The first example employs an accelerator with a silicon area of 1mm 2 and 800MHz clock speed. The task is the 11 th layer of ResNet-18 that has a 3 × 3 kernel and 256 input and output features of dimension 14 × 14 each. Looking at Table 1, it is possibly to fit only 85 32-bit floating-point multipliers in 1mm 2 . That allows installation of 9 PEs (without taking into account the area required for the accumulators of the partial sums) and calculation of convolutions with the 3 × 3 × 3 × 3 kernel in a single clock. Using the known areas of 4-bit, 8-bit and 16-bit PEs, we extrapolate the area of the 32-bit fixed point PE to be 16676µm. From these data, we can place 60 PEs with 7 × 7 × 3 × 3 kernels, 220 PEs with 14 × 14 × 3 × 3 kernels and 683 PEs with 26 × 26 × 3 × 3 kernels, for 32-bit, 16-bit and 8-bit fixed-point PEs, respectively, on the given area.
To calculate the amount of OPS/s required by the layer, under the assumption that a full single pixel is produced every clock, we need to calculate the amount of MAC operations required to calculate one output pixel (n × m × (k 2 + 1)) and multiply it by the accelerator frequency. To calculate the OPS/bit for each design, we divide the amount of MAC operations in the layer by the total number of bits transferred over the memory bus, which includes the weights, the input and the output activations. The layer requires 524.288 TOPS /s to be calculated without stalling for memory access and computation. The available performance of the accelerators is summarized in Table 2 and visualised using the proposed OPS-based roofline analysis in Fig. 7.
In this example, the application's requirements are out of the scope of the product definition. On one hand, all accelerators are computationally bound (all horizontal lines are below the application's requirements), indicating that we do not have enough PEs to calculate the layer in one run. On the other hand, even if we decide to increase the computational density by using stronger quantization or by increasing the silicon area (and the cost of the accelerator), we would still hit the memory bound (represented by the diagonal line). In this case, the solution should be found at the algorithmic level or by changing the product's targets; e.g., we can calculate the layer in parts, increase the silicon area of while decreasing the frequency in order not to hit memory wall, or decide to use another algorithm. Our second example explores the feasibility of implementing the second layer of ResNet-18 that has a 3 × 3 kernel and 64 input and output features of dimension 56 × 56. For this example, we increase the silicon area to 6mm 2 and lower the frequency to 100MHz, as proposed earlier, and add a 4-bit quantized accelerator for comparison purposes. The layer requires 4.1 GOPS /s. The accelerators results are summarized in Table 3 and visualised with the OPS-based roofline analysis in Fig. 8.
From Fig. 8 we can see that our 32-bit and 16-bit accelerators are still computationally bound, while the 8-bit and 4-bit quantized accelerators meet the demands of the layer. In particular, the 8-bit accelerator is located at the border of computational ability, meaning this solution has nearly optimal resource allocation, since the hardware is fully utilized. Still, the final choice of the configuration depends on other parameters such as the accuracy of the CNN.
Both examples demonstrate that decisions made at early stages have a critical impact on the quality of the final product. For example, applying an aggressive quantization to the network or increasing the silicon size may not improve the overall performance of the chip if its performance is memory-bound. From the architect's point of view, it is important to balance between the computation and data transfer. Nonetheless, this balance can be achieved in different ways: at the micro-architecture level, at the algorithmic level or by changing the data representation. The architect may also consider (1) changing the hardware to provide faster communication (requires more power and is more expensive), (2) appling communication bandwidth compression algorithms [4,7], (3) using fewer number of bits to represent weights and activations (using 3-or 4-bit representation may solve the communication problem, at the cost of reducing the expected accuracy), or (4) changing the algorithm to transfer data slower (even though that solves the bandwidth issue, the possible drawback is a reduced throughput of the whole system). The proposed OPS-based roofline model helps the architect to choose alternative. After making major architectural decisions we can use BOPS to get an estimation of the impact of different design choices on the final product, such as the expected area, power, optimal operational point, etc. The next section will examine these design processes from the system design point of view.
HCM METRIC EVALUATION
After introducing the use of BOPS as a metric for the hardware complexity of CNN-based algorithms and the use of the OPS-based roofline model to help the architect understand how the decisions at the algorithmic level may impact the characterizations of the final product, this section aims to provide a holistic view of the design process of systems with CNN accelerators. We conducted an extensive evaluation of the design and the implementation of a commonly used CNN architecture for ImageNet [26] classification, ResNet-18 [12]. We also compared our metric to prior art [20] in terms of correspondence between complexity score to hardware utilization for CNN parameters with various bitwidths.
Experimental methodology
We start the evaluation of the HCM metric with a comprehensive review of the use of BOPS as part of the design and implementation process of a CNN accelerator. This section shows the trade-offs involved in the process and verifies the accuracy of the proposed model. It focuses on the implementation of a single PE since PEs are directly affected by the quantization process. The area of an individual PE depends on the chosen bitwidth, while the change in the amount of input and output features changes both the required number of PEs and the size of the accumulator. The leading example we use implemented an all-to-all CNN accelerator that can calculate n input features and m output features in parallel, as depicted in Fig. 9. For simplicity, we choose an equal number of input and output features. In this architecture, all the input features are routed to each of the m blocks of PEs, each calculating a single output feature. The implementation input features PEs output features Figure 9: All-to-all topology with n × m processing elements.
was done for an ASIC using the TSMC 28nm technology library, 800MHz system clock and in the nominal corner of V DD = 0.81V. For the power analysis, input activity factor, and sequential activity factor, we used the value of 0.2. The tool versions are listed in Table 4. For brevity, we present only the results of experiments at 800 MHz clock frequency. We performed additional experiments at 600 MHz and 400 MHz (obviously, neither BOPS nor the area of an accelerator depends on the chip frequency), but do not show these results. As shown in Section 4, lowering the frequency of the design can help to avoid the memory bound, but incurs the penalty of slower solution time.
Our results show a high correlation between the area of the design and BOPS. The choice of an all-to-all topology shown in Fig. 9 was made because of an intuitive understanding of how the accelerator calculates the outputs of the network. This choice, however, has a greater impact on the layout's routing difficulty, with various alternatives such as broadcast or systolic topologies [6]. For example, a systolic topology, a popular choice for high-end NN accelerators [17], eases the routing complexity by using a mesh architecture. Although it reduces the routing effort and improves the flexibility of the input/output feature count, it requires a more complex control for the data movement to the PEs.
To verify the applicability of BOPS to different topologies, we also implemented a systolic array shown in Fig. 10, where each 1 × 1 PE is connected to 4 neighbors with the ability to bypass any input to any output without calculations. The input feature accumulator is located at the input of the PE. This topology generates natural 4 × 1 PEs, but with proper control, it is possible to create flexible accelerators. In the
System-level design using HCM
In this section, we analyze the acceleration of ResNet-18 using the proposed metrics and show the workflow for early estimation of the hardware cost when designing an accelerator. We start the discussion by targeting an ASIC that runs at 800MHz, with 16 × 16 PEs and the same 2.4GHz DDR-4 memory with a 64-bit data bus as used in Section 4. The impact of changing these constraints is discussed at the end of the section. For the first layer, we replace the 7 × 7 convolution with three 3 × 3 convolutions, as proposed by He et al. [13]. This allows us to simplify the analysis by employing universal 3 × 3 PEs for all layers.
We start the design process by comparing different alternatives using the new proposed OPS-based-roofline analysis since it helps to explore the design trade-offs between the multiple solutions. We calculate the amount of OPS/s provided by 16 × 16 PEs at 800MHz and the requirements of each layer. To acquire the roofline, we need to calculate the OPS/bit, which depend on the quantization level. For ResNet-18, the current art [9] achieves 69.56% top-1 accuracy on ImageNet for 4-bit weights and activations, which is only 0.34% less than 32-bit floating-point baseline (69.9%). Thus we decided to focus on 2-, 3-and 4-bit quantization both for weights and activations, which can achieve 65.17%, 68.66%, and 69.56% top-1 accuracy, correspondingly.
For a given bitwidth, OPS/bit is calculated by dividing the total number of operations by the total number of bits transferred over the memory bus, consisting of reading weights and input activations and writing output activations. Fig. 12 presents OPS-based roofline for each quantization bitwidth. Please note that for each layer we provided two points: the red dots are the performance required by the layer, and the green dots are the equivalent performance using partial-sum computation. Fig. 12 clearly indicates that this accelerator is severely limited by both computational resources and lack of enough bandwidth; the system is computationally bounded, which could be inferred from the fact that it does not have enough PEs to calculate all the features simultaneously. Nevertheless, the system is also memory-bound for any quantization level, meaning that adding more PE resources would not solve the problem. It is crucial to make this observation at the early stages of the design since it means that micro-architecture changes would not be sufficient to solve the problem.
One possible solution, as presented in Section 4, is to divide the channels of the input and output feature maps into smaller groups, and use more than one clock cycle to calculate each pixel. In this way, the effective amount of the OPS/s required for the layer is reduced. In the case that the number of feature maps is divisible by the number of available PEs, the layer will fully utilize the computational resources, which is the case for every layer except the first one. Reducing the number of PEs, however, also reduces the data efficiency, and thus the OPS/bit also decreases, shifting the points to the left on the roofline plot. Thus, some layers still require more bandwidth from the memory than what the latter can supply. In particular, in the case of 4-bit quantization, most of the layers are memory-bound. The only option that properly utilizes the hardware is 2-bit quantization, for which all the layers except one are within the memory bound of the accelerator. Another option for solving the problem is to either change the neural network topology being used, or add a data compression scheme on the way to and from the memory [4,7]. Adding compression will reduce the effective memory bandwidth requirement and allow adding more PEs in order to meet the performance requirements -at the expense of cost and power.
At this point, BOPS can be used to estimate the power and the area of each alternative for implementing the the accelerator using the PE micro-design. In addition, we can explore other trade-offs, such as the influence of modifying some parameters that were fixed at the beginning: lowering the ASIC frequency will decrease the computational bound, which reduces the cost and only hurts the performance if the network is not memory-bounded. An equivalent alternative is to decrease the number of PEs. Both procedures will reduce the power consumption of the accelerator as well the computational performance. The system architect may also consider changing the parameters of the algorithm being used, e.g., change the feature sizes, use different quantization for the weights and for the activations, include pruning, and more.
It is also possible to reverse design order: start with a BOPS estimate of the number of PEs that can fit into a given area, and then calculate the ASIC frequency and memory bandwidth that would allow full utilization the accelerator. This can be especially useful if the designer has a specific area or power goal.
To summarize this section, from the architecture point of view it is extremely important to be able to predict, at the Figure 13: Comparison of BOPS and "compute cost" [20] predictive power. BOPS -5% error, "compute cost" -15%. early stages of the design, if the proposed (micro)architecture is going to meet the project targets. At the project exploration stage, the system architect has plenty of alternatives to choose from to make the right trade-offs (or even negotiate to change the product definition and requirements. Introducing such alternatives later may be painful or even near to impossible.
Comparison with prior metrics
In this section, we compare the BOPS [3] metric to another complexity metric, introduced by Mishra et al. [20]. A good complexity metric should have a number of properties. First, it should reflect the real cost of the design. Second, it should be possible to calculate it from micro-designs or prior design results, without needing to generate complete designs. Last, it should generalize well, providing meaningful predictions for a wide spectrum of possible design parameter values. We compare our choice of computational complexity assessment, BOPS, with the "compute cost" proposed by Mishra et al. [20]. To analyze the metrics, we use our real accelerator area results from Section 5 and error bands of linear extrapolation of the measured values. To remind the reader, BOPS and "compute cost" are defined as follows: The error of predicting a new point with "compute cost" is 15% within 2 orders of magnitude, whereas using BOPS, is only 5%. As shown in Fig. 13, "compute cost" introduces a systematic error: each of the distinguishable groups of three points corresponding to a single value of the number of input and output features creates a separate prediction line. This may lead to higher errors in case of extrapolation from a single value of the number of input and output features or a wide range of the considered bitwidth.
DISCUSSION AND CONCLUSIONS
CNN accelerators are commonly used in different systems, starting from IoT and other resource-constrained devices, and ending in datacenters and high-performance computers. Designing accelerators that meet tight constraints is still a challenging task, since the current EDA and design tools do not provide enough information to the architect. To make the right choice, the architects need to understand at the early stages of the design the impact of high-level decisions they make on the final product, and to be able to make a fair comparison between different design alternatives.
In this paper, we showed that one of the fundamental shortcomings of the current design methodologies and tools is the use of GFLOPS as a metric for estimating the complexity of existing hardware solutions. The first contribution of this paper is the definition of the HCM as a metric for hardware complexity. We demonstrated its application to the prediction of such product characteristics as power, performance, etc.
The second contribution of the paper is the introduction of the OPS-based roofline model as a supporting tool for the architect at the very early stages of the development. We showed that this model allows the comparison of different alternatives of the design and the determination of the optimality and feasibility of the solution.
Lastly, we provided several examples of realistic designs, using an actual implementation with standard design tools and a mainstream process technology. By applying the proposed metric, we could build a better system and indicate to the system architect that certain CNN architectures may better fit the constraints of a specific platform. In particular, our metric confirmed that CNN accelerators are more likely to be memory, rather that computationally bound [17,30].
Although this paper is mainly focused on ASIC-based architectures, the same methodology can be applied to many other systems, including FPGA-based implementations and other system-specific domains that allow trading off accuracy and data representation with different physical parameters such as power, performance, and area. | 8,629.6 | 2020-04-19T00:00:00.000 | [
"Computer Science"
] |
Spatial and temporal coherence properties of single free-electron laser pulses
The experimental characterization of the spatial and temporal coherence properties of the free-electron laser in Hamburg (FLASH) at a wavelength of 8.0 nm is presented. Double pinhole diffraction patterns of single femtosecond pulses focused to a size of about 10 microns by 10 microns were measured. A transverse coherence length of 6.2 microns in the horizontal and 8.7 microns in the vertical direction was determined from the most coherent pulses. Using a split and delay unit the coherence time of the pulses produced in the same operation conditions of FLASH was measured to be 1.75 fs. From our experiment we estimated the degeneracy parameter of the FLASH beam to be on the order of $10^{10}$ to $10^{11}$, which exceeds the values of this parameter at any other source in the same energy range by many orders of magnitude.
that the full beam was split in the middle and overlapped again meaning that parts of the center of the beam were overlapped with parts of the edge of the beam (see [27] for details). This corresponds to a large pinhole separation in a Young's double pinhole experiment yielding reduced values of |γ 12 (0)|. The beam was not spatially filtered with the apertures of the PG2 beamline, which would increase the contrast. Additionally, Ce:YAG crystals are known to saturate at high intensities [40], which can result in a degradation of fringe visibility. 40. D.P. Bernstein, Y. Acremann, A. Scherz, M. Burkhardt, J. Sthr, M. Beye, W. F. Schlotter, T. Beeck, F. Sorgenfrei, A. Pietzsch, W. Wurth, A. F¨hlisch, "Near edge x-ray absorption fine structure spectroscopy with x-ray freeelectron lasers," Appl. Phys. Lett. 95, 134102 (2009). 41. We used the following FLASH operation parameters [42, 43] K = 1.23, λ w = 27.3 mm, I = 2200 ± 300 A, γ = 1741, σ ⊥ = 95 ± 35 µm, and A JJ = 0.83 to calculate the FEL parameter ρ in equation (11). The error is derived by applying the Gaussian error propagation law. 42. http://flash.desy.de/accelerator/, access December 2011. 43. B. Faatz, private communication.
Introduction
Free-electron lasers (FELs) based on the self-amplified spontaneous emission (SASE) principle produce extremely brilliant, highly coherent radiation in the extreme ultraviolet (XUV) [1] and hard x-ray [2] range. Utilizing the high photon flux, the femtosecond pulse duration and the high degree of coherence, techniques like coherent x-ray diffraction imaging (CXDI) [3,4,5,6] x-ray holography [7] and, recently, nano-crystallography [8] promise important new insights in biology [9,10], condensed matter physics [11] and atomic physics [12]. Some of these methods can be implemented only if the radiation is sufficiently coherent, both spatially and temporally. This means that the knowledge of the coherence is mandatory for the success of these experiments. Moreover, it was shown recently that small deviations from perfect coherence can be taken into account in the CXDI method if the degree of coherence is known [13,14]. Both, temporal and transverse coherence effects also play a role in the now accessible field of non-linear excitations of atoms and molecules [15].
Free-electron laser in Hamburg (FLASH) started its operation in 2005 as the first FEL in the XUV wavelength range. It delivered radiation of unprecedented brightness at a wavelength down to 6.5 nm (first reached in October 2007) and generated pulses as short as 10-15 fs Full Width at Half Maximum (FWHM) [1]. A variety of ground breaking experiments [16], including the first demonstration of single pulse femtosecond coherent imaging [5] and studies of the photoelectric effect at ultra-high intensities [17], were performed at FLASH. In 2009 FLASH went through a major upgrade [18], where, along with another electron energy upgrade yielding photon wavelengths down to about 4 nm, a 3rd harmonic accelerator module was installed, in order to improve the capabilities for longitudinal electron bunch compression and in this way enhance the machine stability. This led to a higher stability of the FEL operation at the expense of longer pulse durations, typically around 100 fs. The aim of our experiment was to characterize the coherence properties of the FLASH beam after this upgrade.
Measurements of transverse coherence properties of FEL sources have been reported earlier [19,20,21]. In these experiments a number of shots were averaged and an average transverse coherence length was determined. Contrary to a visible laser, where a resonator typically permits the growth of only a single transverse "TEM 00 " mode, in a SASE FEL a variety of modes can be amplified [22]. The signal at the end of the undulator depends on the shot noise in the electron bunch entering the undulator. As such, the radiation properties, including the transverse coherence, may change from shot to shot. Although no significant shot to shot variations of the coherence properties were observed at the Linac Coherent Light Source (LCLS) [23], at FLASH, with its lower electron and photon energies, we may expect these variations to be more important. In this paper we present measurements of the transverse coherence properties of FLASH and investigate how they fluctuate from shot to shot. To address this important question, we employed the single-shot methodology developed in [23] and conducted Young's double pinhole experiment [24,25] with single FEL pulses.
The other important statistical characteristic of the FEL radiation is its temporal coherence. Due to the same electron bunch instabilities, which lead to partial spatial coherence, the FEL pulses are partially coherent in the time domain. The photon pulse is not merely a single longitudinal mode but rather consists of several spikes with a width of about the coherence time. These longitudinal modes can be correlated and interfere with each other, which affects the coherence time. This complicated phenomena was recently observed experimentally [26]. To analyze the temporal coherence properties at FLASH we used an autocorrelation setup [27]. The single FLASH pulse was split into two parts on a wavefront dividing split mirror. These two parts were brought to an overlap on the detector with an adjustable time delay. The visibility of the interference fringes in the overlap region contains information on the degree of correlation of the time delayed pulses, and therefore on the temporal coherence properties of the radiation.
A number of measurements of the transverse [19,20,21] and temporal [21,28,26] coherence of FLASH operating before the upgrade have been reported. Here we combine spatial and temporal coherence measurements using Young's double pinholes and a split and delay unit at the same operation conditions of FLASH after its upgrade in 2009.
Theory
The concept of optical coherence is associated with interference phenomena, where the Mutual Coherence Function (MCF) [24,25] plays the major role. It describes the correlation between two complex values of the electric field E * (r 1 ,t) and E(r 2 ,t + τ) at different points r 1 and r 2 in space, separated by the time interval τ. The brackets · · · T denote the time average over a time interval T . To experimentally characterize the MCF, the spatial and temporal properties are measured. The former can be accessed by Young's double pinhole experiment [24,25], whereas the latter may be studied using a split and delay line [26,28]. In these interference experiments the complex field E(r,t) is divided into two parts, E 1 (r 1 ,t) and E 2 (r 2 ,t), by the double pinhole or the split mirror. Here r 1 and r 2 are the positions in the plane of the apertures or the split mirror. The interference between the two propagated fields E 1 (u,t) and E 2 (u,t + τ) is observed in the overlap region on the detector and the degree of coherence of the incident radiation field can be determined from the contrast of the interference fringes. The coordinate u lies in the observation plane and τ is the time delay, which is introduced through the difference between the propagation path lengths of the two beams. For narrowband light with an average wavelength λ the intensity distribution measured in the observation plane may be expressed as [24] I(u) = I 1 (u) + I 2 (u) where I 1,2 (u) are the intensities of the individual fields and is the complex degree of coherence in the plane of the apertures or the split mirror. Here, according to the definition (1) I 1 ≡ Γ 11 (0), I 2 ≡ Γ 22 (0) are the intensities incident on the double pinhole or on the split mirror. In equation (2) δ (u) is the rapidly changing phase of γ 12 (τ), which gives rise to interference fringes in the observation plane. The slowly varying phase of γ 12 (τ) is denoted by α 12 (τ).
In a Young's double pinhole experiment the fast varying phase term δ (u) can be expressed as [24] δ (u) = 2πu · d/(λ z). Here z is the double pinhole to detector distance, and d = (d x , d y ) is the pinhole separation in the horizontal d x and vertical d y direction. The intensities I 1,2 (u) in equation (2) are given by I 1,2 (u) = I 1,2 I D (u), where I D (u) is the Airy diffraction pattern from a round pinhole of diameter D [24]. The incident intensity is assumed to be constant across each pinhole. For sufficiently big pinhole to detector distances the Airy patterns from different pinholes overlap on the detector [24] and equation (2) can be simplified to where the effective complex degree of coherence is given by From the definition of γ eff 12 (τ) in equation (5) it is immediately seen that if the intensities incident on individual pinholes differ in magnitude, the contrast observed in the experiment is reduced as compared with the degree of coherence of the incident radiation. If the time delay τ associated with the path length difference between the pinhole one and two is smaller than the coherence time τ c , then γ eff 12 (τ) ≈ γ eff 12 (0) and α 12 (τ) ≈ α 12 (0) are good approximations. Typically, the double pinhole experiment is conducted for different pinhole separations d and the transverse coherence length is defined as a characteristic width of |γ 12 (0)|. To characterize the transverse coherence it is convenient to introduce the normalized degree of coherence as [29,30] This quantity approaches unity for highly coherent and zero for incoherent radiation.
In the split and delay line the interfering fields are generated at the split mirror and are propagated through different arms of the instrument. Both beams are brought to full overlap in the observation plane with approximately equal intensities I 1 (u) ≈ I 2 (u). The interference fringes occur due to the recombination angle θ , which also is the angle between the wavefronts of both beams. The fast varying phase term in equation (2) in this case is given by [21] δ (u) = 2πu · n ⊥ tan(θ )/λ , where n ⊥ is the unit vector along the direction perpendicular to the interference fringes. The analysis of the visibility of fringes can be conveniently performed using the Fourier transform. Assuming that the modulus of the complex coherence function |γ 12 (τ)| and α 12 (τ) are constant within the interference pattern, the Fourier transform of equation (2) yields [31] whereĨ(f) andĨ 1 (f) denote the Fourier transforms of I(u) and I 1 (u), δ is the Dirac delta function, ⊗ is the convolution operator, f is the frequency coordinate, and the fringe frequency is given by f s = n ⊥ λ / tan(θ ). Equation (7) contains one central term and two 'sideband' terms. The central term is the sum of the autocorrelation functions of the two beams. The 'sideband' terms contain the information on the degree of cross correlation between the interfering fields. According to equation (7) the modulus of the complex coherence function can be determined as For a fixed overlap the visibility of the resulting fringes is analyzed as a function of the time delay τ c caused by the two different propagation distances in the split and delay unit. From these measurements one can deduce the coherence time of the radiation, τ c , which we defined according to [24] as where In FEL theory the coherence time can be approximated by [22,32] where λ is the resonance frequency, ρ is the FEL parameter and c is the speed of light. The FEL parameter in equation (11) can be expressed as ρ where I is the electron beam current, K is the undulator parameter, λ u is the undulator period, I A = 17 kA is the Alfvén current, γ is the relativistic (Lorentz) factor, and σ ⊥ is the transverse Root Mean Square (rms) size of the electron bunch. The coupling factor A JJ for a planar undulator is
Experiment
The transverse and temporal coherence of FLASH were measured at a lasing wavelength of 8.0 nm. The FEL generated single pulses at a repetition rate of 10 Hz with an average energy of 180 µJ per pulse. FLASH was operated with a bunch charge of 0.8 nC and an electron energy of around 891 MeV. Six modules of the undulator with a total length of 30 m were used. The FEL was expected to lase in the saturation regime at these operation conditions. The pulse duration was estimated to be about 100 fs (see below). The outline of both experiments for the measurements of the transverse and longitudinal coherence is shown in Figure 1.
(a) Transverse coherence measurements
The spatial coherence measurements were carried out at the BL2 beamline at FLASH (see Figure 1(a)). The beam delivery system consisted of two flat distribution mirrors and an elliptical mirror, which focused the beam to a size of about 10 × 10 µm 2 (FWHM) 70 m downstream from the undulator exit. The acceptance of the mirrors was sufficiently large in both directions, therefore we assume that the beam was not cut by the mirrors. The double pinhole apertures were positioned in the focus inside a dedicated vacuum chamber (HORST [33]). Double pinholes with varying separations between the pinholes have been manufactured to measure the degree of coherence at different relative distances. The pinhole separation varied between 4 µm and 11 µm in the horizontal direction and between 2 µm and 15 µm in the vertical direction. With increasing pinhole separation less intense parts of the beam were probed. To record a similar signal from all apertures we varied the pinhole diameter between 340 nm for the smallest pinhole separation and 500 nm for the largest pinhole separation.
Due to the extremely high power densities in the focus, each aperture set was destroyed during the interaction with a single FEL pulse. About ten identical apertures were manufactured for each pinhole separation to improve statistics of the measurements. All apertures were fabricated The double pinhole diffraction patterns were recorded with an in-vacuum Charge Coupled Device (CCD) (LOT/Andor DODX436-BN) with 2048 × 2048 pixels, each (13.5 µm) 2 in size. A 3 mm linear beamstop manufactured out of B 4 C was oriented perpendicular to the interference fringes and protected the CCD from the direct FEL beam (see Figure 1(a)). A 200 nm thick Pd foil supported by 100 nm parylene-N was mounted 29 mm upstream from the camera to absorb the visible light generated during the damage process of the apertures. A sample to detector distance of 0.34 m provided a sufficient sampling of the fringes (13 pixels per fringe for a 15 µm pinhole separation), which is necessary for the evaluation of the double pinhole interference patterns.
(b) Temporal coherence measurements
The temporal coherence measurements were carried out directly after the transverse coherence measurements at the plane grating monochromator beamline PG2 [34]. There was no further tuning of the machine involved. A typical single pulse and average spectra are shown in Figure 2. From the average spectrum we estimate a FWHM bandwidth of about 1.4% for the FEL radiation. Analyzing individual single shot spectra we estimate an average pulse duration to be on the order of 100 fs. This number is retrieved by counting the individual spikes in the single shot spectra and multiplying the average number of spikes per pulse with the measured coherence time (see below). The plane grating monochromator was used in zero order, and the longitudinal coherence was measured by using the permanently installed split and delay unit [27] (see Figure 1(b)) of the PG2 beamline. This device is able to geometrically split each pulse delivered by FLASH into two pulses and delay them up to 5.1 ps in time with respect to each other with less than 100 as accuracy. Afterwards, the two pulses can be overlapped in space on a Ce:YAG screen 3.5 m downstream in the beamline creating interference fringes when the pulses overlap in time. The images of the overlapped pulses were recorded at the machine pulse repetition rate of 10 Hz using a Basler scA1300-32fm FireWire CCD camera outside of the vacuum chamber. The camera has in total 1296 × 966 pixels with a pixel size of (3.75 µm) 2 . As optics we used a Sill Optics S5LPJ0635 lens with a magnification of 1.102 yielding sufficient resolution to resolve the interference fringes. The complete setup was mounted on an x-y positioning stage allowing for a fast alignment of the camera to the beam position. The measurements were not done in the focus of the beam but rather between the two intermediate foci for the horizontal and the vertical direction.
(a) Transverse coherence measurements
Typical recorded and dark field corrected single shot diffraction patterns are shown in Figure 3(a,b) as a function of the momentum transfer q for a pinhole separation of 4 µm. In Figure 3(a) the double pinhole was oriented horizontally and the vertical fringes originate from the interference between the field scattered at different pinholes. In Figure 3(b) a similar diffraction pattern measured with a vertically oriented double pinhole is presented. The first minimum of the Airy distribution is visible at |q| ≈ 25 µm −1 in both figures. The line scans through the measured diffraction patterns (Figure 3(c,d)) show a high contrast level for the small pinhole separation for both, the horizontal and vertical directions. The contrast decreases for larger separations (see Figure 3(e,f)), which indicates a smaller magnitude of the complex degree of coherence at these length scales.
On most of the diffraction patterns additional noise was observed. It consisted of a constant background and a few hot pixels randomly distributed over the whole diffraction pattern ('salt and pepper noise'). We attribute the appearance of this noise to the light generated during the damage process of the pinholes. Since the Pd foil was not attached to the detector but was positioned a few centimeters upstream, light could leak between the foil and detector and can be a source of this noise.
To determine the modulus of the effective complex degree of coherence |γ eff 12 | for each measured single shot interference pattern, equation (4) was fit to the data [35]. We added a constant A to equation (4) to accommodate for the presence of the constant background noise. The hot pixels were removed from the diffraction patterns and were not considered in the analysis procedure. In particular, the two-dimensional area marked with a white rectangle in Figure 3(a,b) was analyzed. Eight fit parameters including γ eff 12 , (I 1 + I 2 ), D, d, α 12 , A and the position of the optical axis in the horizontal and vertical directions were found. The quality of fit was characterized by an R-factor R = ∑ i (I th where I exp is the background corrected measured data, I th is the fit and summation is made over all points in the fitted area. All fits with R > 0.01 were excluded from the further analysis (less than 50% from the total number of the diffraction patterns in each direction). For each shot a confidence interval of |γ eff 12 | was determined as a value for which R was twice as large as the minimum value, while all other fit parameters were fixed. Typical fit results are shown in Figure 3(c-f).
As a result of the data analysis, the modulus of the effective complex degree of coherence |γ eff 12 | as a function of the pinhole separation is shown in Figure 4(a,b) for the horizontal and vertical directions. We approximated the highest values of |γ eff 12 | for each pinhole separation (shown by black squares) with the Gaussian function exp[−d 2 /(2l 2 c )] shown by black line in Figure 4(a,b). This yields an upper bound estimate of the transverse coherence length in each direction. In this way we determined the transverse coherence length (rms) to be l H c = 6.2 ± 0.9 µm in the horizontal and l V c = 8.7 ± 1.0 µm in the vertical direction. During this fitting procedure we fixed the value of |γ eff 11 | at zero pinhole separation to one according to its definition (1,3,5) [36].
According to equation (3) the information about the intensity profile of the beam is required to characterize the MCF Γ 12 (τ) completely. To measure the beam profile in the plane of the apertures we analyzed PMMA imprints produced by single FEL pulses with a varying degree of attenuation of the beam [37]. Three sets of PMMA imprints with one order of magnitude difference in attenuation of the incoming beam were analyzed. Using the Gaussian beam approximation a beam size of (10 ± 2) × (10 ± 2) µm 2 FWHM was determined. In the horizontal direction additional features on the sides of the beam were observed. For the strongly attenuated beam, however, round craters, 15 µm in diameter, indicate that the central part of the beam is round.
As follows from our analysis (see Figure 4) the values of |γ eff 12 | vary significantly from shot to shot for the same pinhole separation. We attribute this variation mainly to the beam position instabilities in the plane of the sample. If the center of the beam does not hit the center of the double pinhole the intensities incident on individual pinholes will necessarily be different. This difference yields a reduced value of |γ eff 12 | as compared with |γ 12 | (see equation (5)). We estimated the deviation of the beam center relative to the sample by analysing the PMMA imprints measurements. The positions of thirty two craters in the PMMA were found and compared with the nominal positions expected from the stage movement. A maximum offset between the position of the apertures and the incident beam was determined to be ±12 µm in the horizontal and ±8 µm in the vertical direction [38]. Using these values as the offset of a Gaussian beam with a size of 10 × 10 µm 2 , we can calculate the difference of the intensity incident on pinhole one and two. The error imposed by this uncertainty in position compared to the Gaussian fit through the highest values of |γ eff 12 | (black solid line in Figure 4(a,b)) is shown by the blue dashed line in Figure 4(a,b). Most of the measured values lie between the black and the blue line, which indicates that the positional uncertainty is the dominant cause for the apparent variations in |γ eff 12 |. However, as this error is quite significant, we cannot definitively exclude shot to shot variations in the degree of coherence |γ 12 |.
Combining the measured beam size with the highest values of the complex coherence function we determined the degree of transverse coherence ζ (see equation (6)) of the radiation at FLASH. Utilizing the Gaussian Schell-model [25] we estimated a value of ζ x = 0.59 ± 0.10 in the horizontal and ζ y = 0.72 ± 0.08 in the vertical directions. That gives for the total degree of coherence obtained in our experiment a value of ζ = ζ x ζ y = 0.42 ± 0.09. The decomposition of the radiation field into a sum of coherent modes [25,30] has shown that 2 modes in each direction are sufficient to describe the coherence properties in the measured area of the beam. This means (see for derivation [23]) that about 62 ± 11% of the total radiation power is concentrated in the dominant transverse mode.
The coherence measurements presented here indicate a significantly higher degree of transverse coherence of the FLASH beam than previously reported values [19]. We attribute this to the implementation of the 3rd harmonic cavity to the FLASH accelerator complex [18]. A comparison with values reported for the LCLS [23] shows, that both machines, though operating at significantly different wavelengths and different pulse energies, provide similar values of the degree of coherence (in the measurements at LCLS [23] the wavelength was 1.6 nm, the average energy per pulse 1 mJ and the degree of coherence about 75%).
(b) Temporal coherence measurements
The visibility of the interference pattern was measured as a function of the time delay between the two pulses. From these measurements one can deduce the mean electric field autocorrelation function of the FEL radiation. The analysis of the data was done by performing a twodimensional Fourier transform of the recorded fringe patterns (see equation (7) and reference [26] for details). In comparison with the experiment reported in [26] the detector unit has been significantly improved by isolation of the detection system from the background light (light from the experimental hall) and magnification of the FEL beam size on the detector. Since FLASH was operating in single bunch mode, each image contains the interference pattern of a single pulse on the Ce:YAG crystal. Therefore, the presented data was not affected by blurring of the interference patterns and thus reduction of the contrast. This blurring can appear for long delays when microbunches in a bunch train have a slightly different wavelength which leads to a change of the fringe spacing [26].
A typical single shot interferogram for the time delay τ = 0 is shown in Figure 5(a). From the recorded interference patterns we subtracted the background, which was calculated for each image by averaging a region in the corner of the image. After this background subtraction the occasionally occurring negative values were set to zero. A two dimensional Fourier transform (see Figure 5(b)) of the processed image was calculated and also background corrected. The modulus of the complex coherence function |γ 12 (τ)| was found for each single shot according to equation (8). An average of the Fourier transform of the data in the regions marked by AC (autocorrelation) and XC (cross correlation) (see Figure 5(b)) were used in this analysis. A number of single shot measurements for each time delay τ was averaged to improve the statistics and to determine the average complex coherence function as a function of the time delay [39].
In order to determine the coherence time the modulus of the complex coherence function was normalized according to equation (10). Figure 6 shows the normalized value |g(τ)| as a function of the time delay. The measured coherence time determined by equation (9) has For the FLASH parameters used in our experiment [41] we estimated the FEL parameter to be ρ = 1.9 × 10 −3 . That gave theoretical estimate of the coherence time using equation (11) τ th c = 1.1 ± 0.3 fs, that concords well with the measured coherence time.
Close inspection of Figure 6 shows, that the modulus of the complex coherence function contains several time scales, which has been also reported in [26]. To determine these time scales the data were fitted with three Gaussian functions. The result of this fit is presented in Table 1. We interprete these three contributions in the following way. The shortest peak 1 with a width of 1.54 ± 0.01 fs (57.8 ± 0.8 wavecycles) is the contribution of the coherence time of the single FEL spikes within one pulse. The longest peak 3 with a width of 42 ± 1 fs (1575 ± 37 wavecycles) describes the decay of the degree of correlation between individual spikes in the FEL pulse. In our experiment, the normalized temporal coherence reaches a value close to zero for delays of about 1300 wavecycles, whereas the measurement of the temporal coherence reported in [26] at 9.6 nm has reached zero already at about 300 wavecycles. In this experiment, the electron bunch charge was set to 0.76 nC, which is close to the 0.8 nC, used during our measurements. However, before the upgrade, FLASH was operated with a non symmetrical electron bunch shape yielding a leading high current peak and a long tail [18]. After the implementation of the 3rd harmonic cavity the pulses tend to be longer. Short pulses were not accessible during our measurements immediately following the restart of the FLASH after the upgrade.
The asymmetry, which can be seen in Figure 6 for short delays ( 5 fs) can be attributed to an asymmetry in the two photon beams and is governed by peak 2 with a width of 4.93 ± 0.06 fs (185 ± 2 wavecycles). It may be a result of a non-constant non-linear chirp in the transverse direction of the photon beam, as well as a tilt of the wave front or longitudinal pulse double structures. Since we split the incoming beam by means of wavefront division, this can be back-translated to a non-linear chirp along the transverse direction of the photon beam. These asymmetries have also been measured in several other experiments using different techniques [26,28,15].
Discussion
The statistical properties of the radiation at FLASH are described by the full mutual coherence function Γ(r 1 , r 2 ; τ). We have characterized the MCF as a function of the space coordinates and as a function of the time delay. Combining the results from these measurements we determined the magnitude of the complete MCF of the radiation at FLASH, assuming the radiation is cross spectrally pure [24]. According to its definition (1) the MCF is a function of two coordinates in space and one coordinate in time. For visualization purposes we show a 3D representation of the MCF |Γ V (y 1 , y 2 ; τ)| in the vertical direction in Figure 7. A similar result is obtained for the MCF |Γ H (x 1 , x 2 ; τ)| in the horizontal direction. Taking into account the pulse duration we can estimate the degeneracy parameter δ [25,22] of the FLASH beam. It describes the number of photons found in the same quantum state or a single mode of radiation. The average total number of photons per single pulse was about 7 × 10 12 . From our transverse coherence measurements we have approximated that about 65% of the total power is concentrated in the dominant transverse mode. Using the determined coherence time of about 2 fs and the average pulse duration of 100 fs we estimate that about 1% of the total power is concentrated in a single longitudinal and transverse mode. This yields an estimate of the degeneracy parameter δ ∼ 10 10 to 10 11 . This number is significantly higher than at any other source at these photon energies. For instance at synchrotron sources the degeneracy parameter is typically δ ≤ 1. The measured degeneracy parameter of FLASH concords well with the theoretical predictions [22] based on detailed SASE simulations. It is also similar to the degeneracy parameter of optical light lasers [25]. This high value of the degeneracy could lead to new applications of FEL sources in the field of quantum and non-linear optics in the XUV and x-ray regime.
Conclusions
We have experimentally characterized the transverse and longitudinal coherence properties of the XUV free-electron laser FLASH. Young's double pinhole experiment was conducted to find the transverse coherence length of the focused FEL beam. The transverse coherence length of the focused FLASH beam was determined to be 6.2±0.9 µm in the horizontal and 8.7±1.0 µm in the vertical direction. Additionally, the intensity beam profile was measured to be (10 ± 2) × (10 ± 2) µm 2 . From our measurements we conclude that the focused FLASH beam is highly coherent with a total degree of transverse coherence of about 40%. A mode decomposition has shown that about 60% of the total power is concentrated in the fundamental mode. These high values indicate that almost the full transverse photon flux is coherent and can be used for coherence-based applications.
The temporal coherence was measured to be 1.75 ± 0.01 fs, which is in good agreement with the expected theoretical value of 1.1 ± 0.3 fs. While the main coherence peak fits well with the previous measurements, the broad component is about a factor of four larger. We attribute this effect to the longer pulses after the implementation of the 3rd harmonic cavity.
We have also estimated the degeneracy parameter of FLASH radiation to be in the range of 10 10 to 10 11 . This number is significantly larger than at any other existing sources operating at this photon energy range and is comparable with the degeneracy parameter of conventional optical lasers.
Acknowledgements
We acknowledge the help of K. Hagemann in post processing the double pinhole samples for Young's experiments and B. Faatz for his assistance in the theoretical estimates of the FEL properties. Part of this work was supported by BMBF Proposal 05K10CHG "Coherent Diffraction Imaging and Scattering of Ultrashort Coherent Pulses with Matter" in the framework of the German-Russian collaboration "Development and Use of Accelerator-Based Photon Sources". We acknowledge the BMBF funding through the Virtual Institute VH-VI-403 from the Helmholtz society. The KIT/Heidelberg group and the University of Hamburg acknowledges funding from BMBF (05K10VH4 and 05K10GU3, respectively within FSP 301 FLASH). | 7,921.2 | 2012-06-05T00:00:00.000 | [
"Physics"
] |
Language and Science: The Importance of English Language Learning for Students of the Physics Education Study Program
. English language learning is fundamental in students' physics education since it provides access to worldwide scientific literature, facilitates international collaboration, and allows for effective communication among scientists. This literature study emphasizes integrating English learning within the physics curriculum. By building bridges between language and science, students of physics education study programs can improve communication skills and share knowledge effectively at the international level. Through innovative approaches to learning, students can enhance their language skills and develop an in-depth understanding of physics concepts, preparing them for success in the academic and professional world
BACKGROUND
English, the language of international communication, has become increasingly important in various fields, including science.Disciplines such as physics, the cornerstone for many technological advances and cross-disciplinary research, are also not spared from the need for good English proficiency.Students of the physics education study program should be able to access scientific literature, communicate with peers from various countries, and participate in international collaborative projects.Thus, building bridges between language and science and strengthening the importance of English learning in the physics curriculum is essential.
English has become the dominant language in modern science.Baugh (1935: 6), an expert in the history of world language, said that the importance of a language is not alone a matter of numbers or territory.The importance of a language in the worldview is closely related to the people who own and speak the language and the influence of its people on the world.As an international language, English has become a significant tool for communication among scientists, researchers, and academics from all parts of the world.Leading scientific literature, journals, and conferences generally use English as a communication medium.English proficiency is critical to accessing the latest knowledge and participating in global scholarly discussions.
For the students of the physics education study program, English language skills are an additional advantage and an urgent need.However, they often face challenges learning English, especially since physics has technical terminology and complex concepts.Understanding scientific terms in English and expressing physics concepts requires strong language proficiency.The challenge is even more significant for students with different language backgrounds or insufficient exposure to English from an early age.
To overcome these challenges, it is essential to integrate English learning specifically into the physics curriculum.English Language Learning should not be viewed as an end but as a tool to improve understanding of physics concepts and scientific communication.By strengthening the English language skills of physics education students, they will be better able to contribute to the global scientific community, keep up with the latest developments in the field of physics, and collaborate with researchers from various parts of the world.
Besides improving their communication skills, English opens doors to more career opportunities.Many international technology companies and research institutions prioritize candidates with good English proficiency, allowing them to work in a multicultural environment and collaborate with global teams.In addition, strong English proficiency also provides a competitive advantage for students of physics education study programs who wish to pursue graduate studies or participate in international research projects.
An innovative approach is needed to optimize English learning for physics education study programs that adapt to students' needs and interests.In addition to conventional classroom lessons, project-based learning, scholarly discussions, and cross-subject collaboration can effectively improve language skills and understanding of physics concepts.
Using technology in learning, such as e-learning platforms and mobile apps, can also help students learn independently and improve their language skills in a fun and interactive way.
THEORETICAL STUDIES
The integration of English language learning and physics learning has increasingly gained attention in global education.Research and practice on this integration show that combining these two subjects can provide various benefits for students in terms of understanding physics concepts and developing English language skills.Indrasari, N. (2016) e- Hal. 01
METHOD OF THE STUDY
The method used in this study is library research.The collection of data or materials needed to complete the research comes from the library, including books, encyclopedias, dictionaries, journals, documents, magazines, and so on (Walker, 2005).This literature review research collects and analyzes existing knowledge to form a more comprehensive understanding."It is necessary to understand the latest developments and trends in the field of research.This method gathers primary data, which refers to original and firsthand information obtained directly from the source material within the library setting (George, 2008).
RESULT AND DISCUSSION
Mackenzie (2014) estimated that one-third of the world's population, or about two billion people, speak English, making English a global lingua franca.It has been explained that English is indeed set to be a global language because it has the most influence from its people.
Integrating English learning into the physics curriculum is crucial in an increasingly globally connected world.This allows physics students to participate in scientific discussions, access current literature, and collaborate with researchers worldwide, strengthening their position in the global scientific community.
The Role of English in the World of Physical Sciences
The success of a scientist depends on the production of scientific papers and the impact factor of the journal in which they publish since most major scientific journals are published in English.98% of scientific publications are written in English, including researchers from English as a Foreign Language (EFL) countries (Ramı ŕez-Castañeda, 2021).English has become the dominant international language in science, facilitating communication and knowledge exchange between physicists worldwide.English has become the lingua franca in physical science, enabling collaboration across borders and knowledge sharing more effectively.
English proficiency has become an inevitable necessity for physicists.Accessing and participating in global scholarly discussions through English is a must.Access to global scientific literature is one of the critical benefits of English proficiency in the physical sciences.
With the ability to read and understand scientific publications in English, physicists can stay current with the latest research and developments in their field.The ability to read scientific publications in English opens the door to a broader understanding of various approaches and methods in contemporary physics.
Besides being a tool for accessing information, English also plays a crucial role in crossborder scientific collaboration.International cooperation in physics often requires smooth communication between scientists from different countries, and English is often the language used in the process.In addition, English is also the dominant language in international scientific conferences.Scientific conferences allow physical scientists to meet, share ideas, and present their research to the global scientific community.English is the most common language used in presentations and discussions at such conferences.
English is essential in oral communication and in writing and publishing scientific papers on physics.A good understanding of English is indispensable in writing clear and precise scientific papers on physics.In addition, the ability to compile scientific manuscripts in English also opens the door for researchers to publish their work in leading international journals.Dehnad et al., 2010) state that ESP learners in their research site expressed that their first required skill is writing, followed by reading, speaking, and listening.
However, while English's importance in physical science is undeniable, experts also highlight non-native physicists' challenges in learning and using the language.The main challenge for non-native physicists in learning English is mastering the language's technical vocabulary and scientific concepts.To overcome these challenges, a structured approach that integrates English language learning with physics learning can help non-native students acquire the necessary English proficiency while improving their understanding of physics concepts.
Overall, the role of English in the world of physical science cannot be underestimated.
The language serves as a means of communication and publication and is key to accessing global scholarly literature, participating in cross-border scholarly collaborations, and e-ISSN : enhancing academic and professional mobility.Therefore, physicists must continue developing their English proficiency to remain competitive and contribute to an increasingly globally connected world of science.
Challenges for Students in Understanding English
The challenges physics education study program students face in understanding English can be significant.Here are some of the main challenges they may face: a. Technical Vocabulary: English in scientific contexts often has a complex technical vocabulary.
Students may have difficulty understanding physics-specific terms used in scientific literature.
b. Sentence Structure: The sentence structure in English may differ from the student's first language.This can make understanding the text difficult, especially if the text has long and complicated sentences.f.Cultural Influence: English culture, different from the student's mother tongue, can affect their understanding of the texts they read.This can include humour, idioms, and certain social conventions.
Students of physics education study programs often face challenges in understanding English.As one of the main courses, English has a significant role in opening access to scientific literature, participating in scientific discussions, and collaborating with colleagues from various countries.However, physics students often have difficulty understanding technical terms and concepts in English, hindering their ability to learn effectively and contribute to the global scientific community.
One of the main challenges physics education students face in understanding English is the lack of exposure to the language from an early age.Many students come from backgrounds where English is not a mother tongue, so they may not have sufficient English proficiency to attend lectures taught in that language.Non-native physics students often have difficulty understanding technical terms and physics concepts in English, affecting their ability to learn and participate in scientific discussions.In addition, English also has a complex structure and vocabulary, which can be an obstacle to understanding physics students.In physics, many technical terms and complex concepts require deep understanding.Physics students need to understand these terms clearly to be able to attend lectures and read scientific literature in English.The need for a deep understanding of technical vocabulary and physics concepts in English is often a challenge for non-native physics students.
The scientific writing style in English can also challenge students in physics education study programs.Scholarly writing in English often follows certain conventions, such as the use of formal language, the organized structure of the essay, and appropriate references and citations.Physics students must master this writing style to compile research reports, papers, and other assignments well.Physics students must learn to write and compose scientific manuscripts well in English to communicate effectively in the global scientific community.
The Importance of Integrating English in the Physics Curriculum
Theoretically, EAP is related to the research and teaching of English required by those using the language to perform academic duties (Charles, 2013).One of the main reasons why integrating English language learning into the physics curriculum is essential is to allow more comprehensive access to global scientific literature.By understanding English, physics students can access the latest scientific publications, international journals, and textbooks relevant to the field of physics.Integrating English into the physics curriculum allows students to access global scientific literature and participate in international scientific discussions.
Integrating English learning is also essential in preparing physics students to collaborate with colleagues from various countries.Communicating well in English is essential in an increasingly connected academic and research environment.English is the primary language in modern science, so physics students need to master it to thrive in this field.
Integrating English learning into the physics curriculum also helps improve physics students' communication and writing skills.Students can compile research reports, papers, and presentations more effectively and precisely by understanding English.These skills are essential in conveying the results of their research to the global scientific community.The integration of English learning in the physics curriculum helps improve students' ability to write and compile scientific manuscripts well.
Not only that, the integration of English learning can also help physics students develop critical and analytical thinking skills.In English language learning, students are assumed to understand and analyze complex texts.This ability is particularly relevant in physics, where students are often exposed to abstract and complex concepts.Students can also deepen their e- understanding of complex physics concepts by deepening their understanding of English.
Although integrating English learning into the physics curriculum has many benefits, challenges may also arise.One of them is the difficulty non-native physics students face in learning English.To overcome this challenge, educational institutions can provide English language support programs specifically designed for physics students.This allows students to get additional help in improving their English language skills.
Integration of English learning in the physics curriculum is an urgent need in the current era of globalization.English is essential for accessing global scientific literature, collaborating with peers from different countries, improving communication and writing skills, and developing critical thinking and analytical skills.With proper integration, physics students can prepare themselves to become successful professionals in an increasingly globally connected field of physics.
Benefits of English Language Learning for Students
The idea of the importance of learning English was started (Munby, 1978) with the term "communication needs processor".It is followed by West (1994) with the term "analysis of needs" until the publication of the book Introducing Need Analysis and English for Specific Purposes (Brown, 2016).English language learning provides a variety of significant benefits for students of physics education study programs.Here are some of those benefits: a. Access to Global Scientific Literature: English is the primary language in global scientific publications.Through English language learning, students have more comprehensive access to scientific physics literature from various countries.Physics students who speak English have a better chance of accessing the latest research and leading minds in physics.
b. International Collaboration: English language proficiency allows students to collaborate with peers from different countries.English is the primary language of communication in the global scientific community, so physics students need to master it to collaborate effectively with scientists from different parts of the world.
c. Improved Communication Skills: English Language Learning helps improve students' communication skills, both orally and in writing.Communicating well in English allows students to convey their ideas clearly and effectively in the global scientific community.The main reason for poor speaking skills in students is their unwillingness to communicate due to many factors.Overcoming language barriers and building a strong will to communicate allows students to be fluent speakers and not afraid to communicate under any circumstances (Kitchenko and M.P, 2017).Thus, English language learning greatly benefits students of physics education study programs, ranging from access to global scientific literature to improving their communication and writing skills.Therefore, integrating English language learning in the physics curriculum is an urgent need in preparing students to become successful professionals in the increasingly globally connected field of physics.
Innovative Approach to English Language Learning for the Students
An innovative approach to learning English for physics education students can provide a more exciting and practical learning experience.Here are some innovative approaches that can be applied: a. Project-Based Learning: Students can engage in collaborative projects that allow them to learn English while exploring physics concepts.For example, they can work in groups to put together presentations or make physics experiment videos in English.
b. Game-Based Learning: Educational games can strengthen language skills and understanding of physics concepts.For example, board games or role-playing games about physics concepts can be modified to use English as a communication medium.
c. Technology-Based Learning: Using mobile apps, e-learning platforms, and other digital resources can increase student engagement in English language learning.For example, students can use the app for vocabulary practice or to participate in online discussion forums.
d. Digital Project-Based Learning: Students may be asked to create digital projects such as blogs, podcasts, or video presentations in English on specific physics topics.This approach improves their language skills and allows them to share their knowledge with a broader audience.
e. Collaborative Learning: Collaboration between students in a supportive learning environment can encourage exchanging ideas and problem-solving in English.For example, students can work in groups to complete project assignments involving discussion, negotiation, and presentation in English.
Through these innovative approaches, educational institutions can create a stimulating and supportive learning environment for physics education students to learn English better. e-
CONCLUSION AND RECOMMENDATION
Building bridges between languages and science is essential in preparing physics students to become successful professionals in an increasingly globally connected world.
Learning English is key in this process, as English is not only the language of international communication but the primary language in the physical scientific literature and the global scientific community.
The importance of learning English for students of the physics education study program is reflected in several aspects.First, English language learning provides greater access to global scientific literature, allowing students to keep abreast of the latest developments in physics.
Second, English language proficiency allows students to collaborate with peers from different countries, opening up opportunities for mutual exchange of ideas and research.Third, English language learning helps improve students' communication and writing skills, which are essential in conveying their ideas and research results to the global scientific community.
Thus, integrating English language learning into the physics curriculum is an urgent need in preparing students to become successful professionals in the increasingly globally connected field of physics.By mastering English well, students can develop their careers in physics and contribute to advancing knowledge and innovation globally.Therefore, learning English is not only an investment in students' personal development but also an investment in the future of science and technology.
To further explore, it is suggested that research be conducted to investigate various language teaching methods and approaches that are most effective for students in physics education study programs.Compare the outcomes of traditional language learning methods with innovative approaches such as immersive learning, technology-enhanced instruction, or content and language-integrated learning (CLIL) to assess which methods yield physics students' highest language proficiency and retention levels.
Language and Science: The Importance of English Language Learning for Students of the Physics Education Study Program 2 Sintaksis : Publikasi Para ahli Bahasa dan Sastra Inggri -Volume.2,No.3 Mei 2024 c. Pronunciation and Intonation: Pronunciation and intonation in English can be challenging to master.Students may have difficulty understanding speech from lecturers or communicating with classmates.d.Writing Skills: The ability to write in English can also be a challenge.Students may find it difficult to compile essays or reports in English correctly and clearly.e. Listening Skills: Listening skills are also necessary, especially when attending lectures or seminars in English.Students may find it challenging to follow well if unfamiliar with the language.
Language and Science: The Importance of English Language Learning for Students of the Physics Education Study Program 4Sintaksis : Publikasi Para ahli Bahasa dan Sastra Inggri -Volume.2,No.3 Mei 2024 Language and Science: The Importance of English Language Learning for Students of the Physics Education Study Program 8 Sintaksis : Publikasi Para ahli Bahasa dan Sastra Inggri -Volume.2,No.3 Mei 2024 d.Writing Skills Development: Students proficient in English have better skills in compiling research reports, papers, and scientific essays.Good writing skills in English help students convey their research results accurately and convincingly.e. Enrichment of Academic Experience: English Language Learning also enriches students' academic experience by enabling them to attend international seminars, conferences, and workshops.Students proficient in English can use this opportunity to share their knowledge and experience with peers from different countries. | 4,351.2 | 2024-04-03T00:00:00.000 | [
"Physics",
"Education"
] |
dUTPase inhibition augments replication defects of 5-Fluorouracil
The antimetabolite 5-Fluorouracil (5-FU) is used in the treatment of various forms of cancer and has a complex mode of action. Despite 6 decades in clinical application the contribution of 5-FdUTP and dUTP [(5-F)dUTP] and 5-FUTP misincorporation into DNA and RNA respectively, for 5-FU-induced toxicity is still under debate. This study investigates DNA replication defects induced by 5-FU treatment and how (5-F)dUTP accumulation contributes to this effect. We reveal that 5-FU treatment leads to extensive problems in DNA replication fork progression, causing accumulation of cells in S-phase, DNA damage and ultimately cell death. Interestingly, these effects can be reinforced by either depletion or inhibition of the deoxyuridine triphosphatase (dUTPase, also known as DUT), highlighting the importance of (5-F)dUTP accumulation for cytotoxicity. With this study, we not only extend the current understanding of the mechanism of action of 5-FU, but also contribute to the characterization of dUTPase inhibitors. We demonstrate that pharmacological inhibition of dUTPase is a promising approach that may improve the efficacy of 5-FU treatment in the clinic.
The 5-fluoro-substituted uracil analogs are metabolized to 5-FdUMP, which binds and thereby occupies the TS-substrate pocket [4]. Inhibition of TS leads to depletion of thymidine but also accumulation of the substrate dUMP, which is phosphorylated to dUTP. In addition, conversion of 5-FU to 5-FdUTP further elevates uracil levels. The increased dUTP/dTTP and 5-FdUTP/ dTTP ratios promote uracil misincorporation into DNA by DNA-polymerases [6]. Subsequent attempts of futile DNA repair eventually lead to cell death [6][7][8][9][10]. Besides DNAassociated toxicity, incorporation of the 5-FU metabolite 5-FUTP into RNA has been shown to contribute to cell death [11][12][13][14]. However, the metabolism and working mechanism of fluoropyrimidines are complex and the contribution of each of these components for toxicity is often debated.
Deoxyuridine triphosphatase (dUTPase, also known as DUT) circumvents high levels of uracil in www.impactjournals.com/oncotarget/ Oncotarget, 2017, Vol. 8, (No. 14), pp: 23713-23726 Research Paper the biosynthetic pool by hydrolyzing dUTP to dUMP and pyrophosphate. This reaction additionally supplies TS with its substrate dUMP [15]. Despite the selective binding pocket of dUTPase, the 5-FU metabolite 5-FdUTP has been shown to be a substrate for this enzyme [16]. The physiological function of dUTPase is to reduce dUTP accumulation and prevent misincorporation of the non-canonical nucleotide into DNA. However, from a treatment perspective, this activity could hamper the therapeutic success of 5-FU.
Several studies have shown that dUTPase levels significantly influence TS-based treatment response. Ectopical overexpression of E. coli dUTPase induced resistance to FUdR in human cells [17]. In contrast, depletion of dUTPase increased response to FUdR and Pemetrexed [18,19]. dUTPase expression also inversely correlated with sensitivity to TS inhibitor ZD9331 [20]. Moreover, in patient samples, high nuclear dUTPase expression was associated with both resistance to 5-FU therapy [21] and metastasis [22]. Interestingly, a dUTPase inhibitor was reported to sensitize cancer cells to 5-FU treatment in a xenograft setting [23].
Despite adjusting treatment regimens and improving TS-based therapies, a large number of patients still exhibit intrinsic or acquired treatment resistance [2]. Further clarification of the 5-FU mechanism of action in combination with dUTPase inhibitors is required to improve the treatment outcome. Here, we demonstrate that 5-FU treatment induces DNA replication defects. Pharmacological inhibition and knockdown of dUTPase further augment 5-FU induced perturbations at the replication fork, DNA damage and cell death, highlighting the importance of 5-FdUTP and dUTP [(5-F)dUTP] and dUTPase for 5-FU-induced cytotoxicity.
RESULTS dUTPase depletion increases cytotoxicity of 5-FU in SW620 colorectal cancer cells
To understand the importance and consequences of (5-F)dUTP accumulation during 5-FU treatment we depleted dUTPase in SW620 colorectal cancer cells using siRNA-mediated knockdown. Transfection with a dUTPase specific siRNA (sidUTPase) depleted protein levels after 48 hours ( Figure 1A and Supplementary Figure 7A). A non-targeting siRNA (siNon-t) control was compared to untransfected cells to rule out non-dUTPase related effects from the siRNA transfection.
dUTPase depleted and control cells were exposed to 5-FU for 48 hours and re-seeded to assess their ability to form colonies. Whereas dUTPase depletion by itself had no effect on cell survival, it significantly increased the cytotoxic effect of 5-FU, when compared to the untransfected or siNon-t transfected cells ( Figure 1B).
To further understand the mechanism of toxicity, dUTPase-depleted and control cells were treated for 48 hours with 5-FU and the cell cycle was analyzed by FACS. While 5-FU treatment of up to 25 μM accumulated cells in S-phase, it had only minimal cytotoxic effects, indicated by a minor increase in the subG1 population ( Figure 1C-1D). dUTPase depletion, upon 5-FU treatment, increased the subG1 population already at the lowest dose of 5-FU tested from 2 to 24% (6.25 μM of 5-FU). Notably, depletion of dUTPase by itself resulted in a small increase of subG1, S-and G2-phase cells and a reduction in the G1 population. No difference in the subG1 population was observed between the untransfected and siNon-t transfected cells (Supplementary Figure 1).
dUTPase depletion increases 5-FU-induced S-phase arrest of the cell cycle
To determine the number of S-phase cells in the cell cycle, we next measured EdU incorporation into DNA. As expected, 5-FU treatment alone increased the amount of cells in S-phase, as demonstrated by more incorporation of EdU into DNA (Figure 2A-2B). Interestingly, dUTPase depletion during the 5-FU treatment led to reduced amount of EdU being incorporated.
We further analyzed DNA replication using the alkaline DNA unwinding (ADU) technique ( Figure 2C) [24]. In this assay, replicating forks are pulse labeled by incorporation of 3 H thymidine, followed by a fresh media treatment. At increasing time points, DNA is unwound for about 60 kb by addition of an alkaline solution. The genome is subsequently fragmented into 3 kb pieces using ultrasonic treatment. This treatment creates a fraction of single stranded DNA (ssDNA) close to the replication fork and double stranded DNA (dsDNA) away from the fork. The radioactive label shifts from the ssDNA to the dsDNA fraction as the fork moves forward. The comparison of radioactivity in the ssDNA compared to the dsDNA fraction is therefore a measure of replication fork progression.
In line with the EdU data, 48 hours of 5-FU treatment led to increased incorporation of total radioactive thymidine at time zero compared to untreated cells, which can be explained by the increased amount of cells in S-phase ( Figure 2D). Furthermore, dUTPase depletion reduced the total amount of thymidine incorporated, supporting the FACS analysis. Since 3 H thymidine was used it would not require dUTPase activity to be introduced into DNA. Therefore, we conclude a true reduced fork rate following dUTPase treatment.
As time progressed and the replication fork proceeded, the radioactive signal moved from the ssDNA to the dsDNA fraction. When cells were treated with increasing concentrations of 5-FU, the ssDNA to dsDNA exchange was delayed in a dose-dependent manner, indicating reduced replication fork speed ( Figure 2E). In www.impactjournals.com/oncotarget dUTPase depleted cells, an even slower exchange was observed, indicating on average slower DNA replication, compared to the siNon-t control ( Figure 2F).
dUTPase depletion increases 5-FU-induced replication defects
While the ADU technique evaluates the average replication speed in a defined population of cells, the EdU technique averages the EdU incorporation per cell. Nevertheless, these average values of replication speed could both indicate a reduced number of fired replication origins (with similar replication speed) or a reduction in fork progression. The speed of single replication forks can be analyzed by using the DNA fiber assay, in which successive incorporations of the thymidine analogues CldU and IdU into DNA is visualized by immunostaining ( Figure 3A). Using this technique, we demonstrate that 5-FU treatment reduces the replication fork speed, explaining the accumulation of cells in S-phase ( Figure 3A-3D). Depletion of dUTPase further decreases the DNA fiber lengths, demonstrating that individual replication forks are severely affected by lack of the dUMP substrate for TS.
Characterization of the dUTPase inhibitors 1 and 2
In order to study the effects of pharmacological dUTPase inhibition, two dUTPase inhibitors (compounds 1 and 2, Figure 4A and 4B respectively) were synthesized as described in the Supplementary Materials and Methods (Supplementary Figure 2) [25][26][27].
The potency of these inhibitors was assessed using an in vitro activity assay, in which dUTPase catalyzed the hydrolysis of dUTP to dUMP and pyrophosphate (PPi). The conversion of inorganic phosphate (Pi) from PPi was analyzed using the malachite green reagent. For this purpose, dUTPase was expressed and purified from bacterial lysates (Supplementary Figure 3A) and its activity was assessed with dUTP and 5-FdUTP (6.6 and 5.5 μM formed PPi per second per μM enzyme, similar to previously reported data) (Supplementary Figure 3B) [28]. Compound 1 shows an IC 50 of 740 nM while compound 2 exhibits an approximately 30-fold higher efficacy with an IC 50 of 25 nM ( Figure 4C). In addition, these compounds showed high selectivity in a pannel of various nucleoside triphosphate pyrophosphatase or phosphohydrolase enzymes tested (Supplementary Figure 4).
Computational docking predicts the putative binding modes for compounds 1 and 2 in the substrate binding pocket of dUTPase ( Figure 4D and Supplementary Materials and Methods). For both compounds, the docking with the lowest Glide SP scores (-8.23 and -7.13 kcal/ mol, respectively) had their uracil moieties inserted deep into the uracil binding pocket and displayed the same H-bonding patterns as described for other uracil-based ligands, interacting with Gly99, Gly110 and a conserved water molecule that bridges uracil, Gly97 and Val112. The flexible side-chains of both ligands had adopted U-shaped conformations with one of the aryl group folding back over the uracil moiety. The amide linker of inhibitor 1 and sulfonamide linker of compound 2 are facing the solvent, while the terminal cyclopentyl moiety of compound 2 partially occupies the same region as the second terminal phenyl group of compound 1. The benzylic α-methyl group of inhibitor 2 occupied the same region as the proline ring of compound 1, providing a hypothesis for the observed stereochemical preference displayed by structurally closely related representatives of these two chemical series [29,30].
dUTPase inhibitors sensitize colorectal cancer cells to 5-FU treatment
We next analyzed whether pharmacological inhibition of dUTPase, using compounds 1 or 2, is a potential strategy to increase the efficacy of 5-FU. Cell survival was assessed after 72 hours of co-treatment using the resazurin assay. Inhibition of dUTPase significantly increased the cytotoxicity of the 5-FU treatment ( Figure 5A and Supplementary Figure 5A for compounds 2 and 1, respectively). In line with protein depletion, dUTPase inhibition alone did not induce cellular toxicity at the concentrations and time points tested. Importantly, the toxicity induced by 2 upon 5-FU treatment was rescued by addition of thymidine ( Figure 5B). In addition, the cervix cancer cell line HeLa showed increased sensitivity to 5-FU by addition of compounds 1 or 2 and to a minor extend a slight effect is observed with the cell line TOV-112D (ovary origin) ( Figure 5C). On the contrary, the osteosarcoma cell line U2OS showed no increase in 5-FU toxicity when dUTPase was inhibited. The sensitivity did only partially correlate with dUTPase expression levels, as both HeLa and SW620 cells exhibit increased sensitivity to 5-FU upon addition of compounds 1 or 2, but only Hela cells show high expression levels of dUTPase ( Figure 5D and Supplementary Figure 7B). These data demonstrate a variability in potentiation of 5-FU toxicity in different cancer cell lines.
dUTPase inhibitors increase 5-FU induced replication defects and DNA damage
We then studied the effects of the dUTPase inhibitors on 5-FU-induced S-phase arrest by co-treating cells for 48 hours with inhibitor 2 and 5-FU and subsequent labelling with EdU. In cells treated with 3.1 μM of 5-FU, inhibition of dUTPase by compound 2 further reduced the amount of incorporated EdU in a dose dependent manner ( Figure 6A-6B). DNA fiber experiments revealed that the reduced EdU incorporation also correlated with decreased replication fork progression, supporting the data previously obtained by dUTPase knockdown experiments ( Figure 6C-6E and Supplementary Figure 5B-5D).
Staining of phosphorylated histone H2A.X (γH2AX) is commonly used to visualize DNA damage in association with replication fork stress [31]. Here, we determine γH2AX foci formation by automated microscopy following treatment of cells for 72 hours with 5-FU, and demonstrate that addition of compound 1 or 2 to the 5-FU treatment further increased DNA damage (Supplementary Figure 6). No increase in γH2AX foci could be detected in cells treated with the dUTPase inhibitors alone. dUTPase inhibition likely leads to accumulation of dUTP and 5-FdUTP and subsequent misincorporation into DNA. To test this hypothesis, we analyzed dU and 5-FdU levels in DNA using mass spectrometry. While 5-FU treatment alone had only minimal effects on the dU levels in DNA, simultaneous dUTPase inhibition significantly raised the amount of dU incorporation into DNA ( Figure 6F). Following co-treatment of compound 2 and 5-FU, some low levels of 5-FdU in DNA were observed but were too close to detection limit to make any firm conclusion (data not shown). No 5-FdU in DNA was detected in DNA from single-treated cells (data not shown).
DISCUSSION
Even 60 years after the first synthesis of antifolates and fluoropyrimidines, the complex mechanism of action is still debated [32,33]. Initially, depletion of thymidine was thought to be the main cause of 5-FU induced toxicity [34,35]. Many studies have in addition highlighted the importance of elevated levels of both uracil and 5-fluorouracil and their misincorporation into DNA [6,36,37]. More recently, incorporation of 5-FUTP into RNA and its associated transcription defects have been considered as the main cause of cell death [38]. Studying the mechanism of action of 5-FU is necessary to understand and overcome frequently observed drug resistance and ultimately improve patient care.
Here, we investigated the DNA replication defects induced by 5-FU treatment and the importance of 5-FdUTP accumulation for this effect. By analyzing EdU incorporation into DNA, we observed that 5-FU treatment leads to accumulation of cells in the S-phase of the cell cycle. ADU experiments and DNA fiber analyses demonstrated slower DNA replication fork progression upon 5-FU treatment, which are in agreement with low dTTP levels generated by TS inhibition. Hence, our conclusion is that reduced replication fork speed by combination treatment of dUTPase inhibitors/siRNA and 5-FU is a result of even lower levels of dTTP, caused by low substrate dUMP levels (by dUTPase loss) and low TS activity (by 5-FU treatment). Uracil analogues (EdU, CldU, IdU) were used for the DNA fiber and cell cycle experiments. Since these uracil analogues are already modified on the 5'-position they likely do not need dUTPase activity to be incorporated into DNA as also suggested by the fact that we observed no decrease in the intensity of fibers following dUTPase inhibition or siRNA treatments.
Interestingly, protein depletion and pharmacological inhibition of the nucleotide triphosphatase dUTPase further augmented the amount of uracil in DNA, DNA replication defects, DNA damage and cytotoxicity of 5-FU, highlighting the importance of (5-F)dUTP accumulation for cytotoxicity. However, one should keep in mind that the 5-FU metabolism involves various enzymes and intermediate species and that the mode of toxicity is most likely multifaceted and dependent on the molecular makeup of the cell.
Despite this complexity, a number of studies have shown that dUTPase levels significantly influence the efficacy of 5-FU and other TS-based therapies [17-20, 22, 39]. These studies have often used siRNA mediated dUTPase depletion. However, in certain situations a discrepancy between protein inhibition and depletion can be observed. Here, we showed that inhibiting dUTPase with small molecules leads to comparable effects as protein depletion by siRNA.
Furthermore, we show that inhibiting dUTPase activity, both by siRNA and pharmacological inhibition, does not lead to severe toxicity when used as a monotreatment. A favorable safety profile was also confirmed by the phase I clinical trial of the dUTPase inhibitor TAS-114 [40].
Importantly, tumors were found to have dysregulated dUTPase expression and high nuclear dUTPase expression correlated with therapy resistance, shorter time to progression and shorter overall survival [21]. With this study, we further elucidate the mechanism of 5-FUinduced toxicity by investigating DNA replication defects. Inhibiting dUTPase activity by siRNA or inhibitors significantly augmented 5-FU induced replication defects and toxicity, highlighting the contribution of (5-F)dUTP to toxicity. These results demonstrate the high potential of dUTPase inhibitors to improve current TS therapies.
MATERIALS AND METHODS
Cell culture SW620, HeLa, TOV-112D and U2OS cells were cultured in 37 °C with 5% CO 2 using DMEM (Life Technologies), supplemented with fetal calf serum (10%), penicillin (50 U/mL) and streptomycin (50 μg/ mL). Mycoplasma contamination was assessed using the MycoAlert TM Mycoplasma Detection Kit (Lonza). Thymidine (Sigma-Aldrich) was diluted in H 2 O, 5-FU (Sigma-Aldrich) was diluted in DMSO to 200 mM, while compounds 1 and 2 were dissolved in DMSO to 10 mM. DMSO concentrations were adjusted to equal levels in all treatments.
RNAi transfection
siRNA was transfected using INTERFERin ® as suggested by the manufacturer's instructions (polyplus transfection TM ). Oligonucleotides targeting all three isoforms of dUTPase (sense strand: 5'CGGACAUUCAGAUAGCGCUTT-3'; antisense strand: 5'-AGCGCUAUCUGAAUGUCCGTT-3'; referred to as sidUTPase) and the All-stars negative control (referred to as siNon-t) were obtained from Qiagen and transfected to a final concentration of 10 nM. Cells were siRNA transfected for 48 hours, re-seeded and incubated overnight to achieve attachment. Additional treatment was performed as indicated in the different sections.
Clonogenic survival assay
After siRNA transfection, the indicated concentration of 5-FU was added for 48 hours, followed by a 24 hours recovery period with fresh media. 200 cells were re-seeded onto petri dishes in triplicate and incubated for 10 days. Colonies were fixed and visualized using methylene blue (4 g/L) in methanol and then assessed by eye. Surviving fractions were calculated by averaging the triplicate values and normalizing these against the untransfected DMSO control.
Propidium iodine (PI) FACS analysis
Following the indicated treatment, the cells and the media were collected. Samples were washed and fixed by freezing cells in 70% ethanol. After two PBS washes, 0.5 mL PI solution (25 μg/mL PI (Sigma) and 100 μg/mL RNaseA (Thermo Fisher Scientific) in PBS) was added for 20 min. PI intensity was measured on a FACSCalibur (Becton Dickinson) and the cell cycle was analyzed using WinMDI 2.9.
5-ethynyl-2'-deoxyuridine (EdU) and To Pro FACS analyses
To assess replication, cells were pulse labeled using 1 μM EdU for 30 min. The Click-iT ® EdU Alexa Fluor ® 488 Imaging Kit (Molecular Probes) was used as described in the manufacturer's manual. DNA was counterstained with 1 μg/ mL To Pro (Molecular Probes). EdU and To Pro intensities were measured on a FACSCalibur (Becton Dickinson) and analyzed using WinMDI 2.9 and Cytobank.
Alkaline DNA unwinding (ADU) technique
The method was performed as described by Johansson et al. [42]. Briefly, cells were pulse-labeled with 3 H-thymidine (7.4 kBq/mL; GE Healthcare) for 30 min. Cells were washed and incubated in media plus the indicated treatment for the specified time-points. Icecold 0.03 M of NaOH in 0.15 M of NaCl was added for 30 min incubation on ice and in darkness. Addition of 1 mL of 0.02 M NaH 2 PO 4 stopped the unwinding. The DNA was fragmented by ultrasonic treatment for 15 seconds (B-12 sonifier with micro-tip; Branson). SDS was added to a final concentration of 0.25% and samples were frozen overnight. After a 1:1 dilution with distilled water, the samples were added to hydroxyapatite columns mounted in an aluminum block maintained at 60 °C. The columns were washed with 0.5 M potassium phosphate before the single stranded and then double stranded DNA fractions were respectively eluted with 0.1 M and 0.25 M potassium phosphate buffer. Radioactivity was assessed on a RackBeta scintillation counter. The amount of single stranded DNA was compared to the total labeled DNA.
DNA fiber technique
The DNA fiber technique was similarly performed as described by Groth et al. [43]. Cells were treated as indicated before, 5-chloro-2'-deoxyuridine (CldU) (25 μM; Sigma) was added for 40 min followed by 40 min incubation with 5-iodo-2'-deoxyuridine (IdU) (250 μM; Sigma), with the indicated treatment present. Cells were scraped in icecold PBS. Unlabeled and labeled cells were mixed in equal proportions. 2.5 μL of the cell suspension were mixed with 7.5 μL spreading buffer (200 mM Tris-HCl, pH 7.4, 50 mM EDTA and 0.5% SDS) on microscopy slides (SuperFrost ® , Menzel Gläser, VWR). After 8 min, the slides were tilted to spread the DNA and then fixed by incubation in MeOH/ AcOH (3:1) overnight at 4 °C. Samples were denatured in 2.5 M HCl for 1 hour and unspecific binding was blocked using PBS containing 1% BSA and 0.1% Tween20. For immunodetection of CldU and IdU, the slides were incubated with monoclonal rat anti-BrdU Ab (Clone BU1/75 (ICR1); Oxford Biotechnologies) and monoclonal mouse anti-BrdU Ab (Clone B44; Becton Dickinson, 347580). Anti-rat Alexa Fluor ® 555 and anti-mouse Alexa Fluor ® 488 (1:500 in blocking solution; Life Technologies) were used as secondary antibodies. Images of coded samples were taken on a Zeiss LSM 510 or 780 inverted confocal microscope. Fiber length was measured using the ImageJ software. 1 μm was converted to 2.59 kilo base pairs. At least 100 forks were analyzed per sample.
Resazurin survival assay
2000 cells were seeded in 50 μL medium per well into 96-well plates. 24 hours later, 40 μL of compound 1 or 2 was added to reach a final concentration of the indicated dose (after addition of 10 μL 5-FU stock). After 2 hours, 10 μL of the 5-FU stock was added to each well to reach the indicated concentration. After 72 hours, resazurin was added to a final concentration of 10 μg/mL and the cells were incubated 3 hours. Fluorescence intensity was measured at 544/590 nm (Ex/Em). Relative survival of the cells was calculated by subtracting the background fluorescence, averaging duplicate measurements and normalizing the value to the untreated well.
Quantification of modified bases in genomic DNA
DNA for nucleoside quantification was isolated by phenol:chloroform:isoamyl alcohol extraction as previously described [44]. Cells were lysed by passing through 21G and 23G syringe needles and subsequent incubation at 37 °C for 1 h with 1000 RPM shaking in a buffer containing 10 mM Tris-HCl (pH 8.0), 10 mM NaCl, 1% SDS, 100 mM DTT, 0.1 mg/mL proteinase K (Worthington Biochemical), 0.1 mg/mL RNase A (Sigma-Aldrich), 50 μM tetrahydrouridine (THU, Merck Millipore). DNA was subsequently extracted from the lysates with 25:24:1 phenol:chloroform:isoamyl alcohol, followed by two washes with 24:1 chloroform:isoamyl alcohol and isopropanol precipitation using 10 M ammonium acetate to precipitate the DNA. RNA and free nucleotides were then removed from the DNA samples by treatment with 4 μg RNase A in 10 mM ammonium bicarbonate (pH 7.0), 10 mM MgCl 2 for 30 min at 37 °C, followed by a subsequent isopropanol precipitation.
Next, the DNA samples were hydrolyzed and dephosphorylated to single nucleosides as previously described [44]. DNA was hydrolyzed to nucleosides by treatment with 0.8 U Nuclease P1 (Sigma-Aldrich), 80 U Benzonase (Santa Cruz Biotechnology), and 7.5 U Antarctic Phosphatase (New England Biolabs) in 50 μL reactions containing 10 mM ammonium acetate (pH 5.5), 1 mM MgCl 2 , 0.1 mM ZnCl 2 and 240 μM THU for 60 min at 37 °C. Enzymes were then precipitated and removed from the reactions by adding three volumes of ice-cold acetonitrile to the reactions, incubating for 10 min on ice, centrifugation at 16,100 rcf for 30 min at 4 °C. The supernatants were transferred to new tubes and lyophilized until dry. For 5-fluoro-2´-deoxyuridine (5FdU), 0.2 U alkaline phosphatase (Sigma-Aldrich) and 240 μM Deferoxamine mesylate (Santa Cruz Biotechnology) were used instead of Antarctic Phosphatase and THU.
To separate dU from dC, the samples were redissolved in water and fractionated on an Agilent 1100 HPLC system (with a UV detector set to 260 nm to identify the canonical nucleosides) and a mixed mode Primesep 200 column (2.1 mm x 150 mm, 5 μm, SieLC) kept at 30 °C using a flow rate of 0.4 mL/min and water and acetonitrile as mobile phase, each containing 0.1% formic acid, as the mobile phase. The 12-min-long HPLC gradient was as follows: 5% acetonitrile for 30 s, ramp to 35% acetonitrile by 1.5 min to 2.5 min, and return to 5% acetonitrile by 2.51 min. The dU-containing fractions were collected from 1.6-1.7 min and vacuum centrifuged until dry. Samples were not pre-fractionated for 5-FdU analysis. The pellets were redissolved in water and analysed by LCMS/MS using a reverse phase column (2.1 mm x 150 mm, 1.8 μm, EclipsePlusC18 RRHD, Agilent Technologies) kept at 25 °C with a flow rate of 0.3 mL/ min on a 1290 Infinity II HPLC coupled to a 6495 Triple Quadrupole mass spectrometer with an electrospray ion source (Agilent Technologies). Water and methanol were used as the mobile phase, each containing 0.1% formic acid. The 13-min-long HPLC gradient was as follows: 5% methanol for 3 min, ramp to 13% methanol by 3.5 min, ramp to 17% methanol by 5.5 min to 7 min, and return to 5% methanol by 8 min. Analysis was performed in positive ionization multiple reaction monitoring mode, using the mass transitions 229.08 → 113.0, 232.08 → 116.0, and 247.1 → 131.0 for 2´-deoxyuridine (dU), 13 C 15 N 2 -dUrd, 5-FdU, respectively.
The Supplementary Materials and Methods contain additional information regarding the expression and purification of human recombinant dUTPase, the dUTPase activity and inhibition assay, the detailed synthetic route for dUTPase inhibitors 1 and 2, as well as their molecular dockings with dUTPase, and the analysis of phosphorylated H2A.X. | 5,915 | 2017-02-28T00:00:00.000 | [
"Biology",
"Chemistry",
"Medicine"
] |
Design of a MW-scale thermo-chemical energy storage reactor
The reversible exothermic reaction of CaO with water is considered one of the most promising reactions for high temperature thermal energy storage. In this paper, a novel technical design of a MW-scale thermochemical energy storage reactor for this reaction is presented. The aim is to provide an easy, modular and scalable reactor, suitable for industrial scale application. The reactor concept features a bubbling fluidized bed with a continuous, guided solid flow and immersed heat exchanger tubes. To investigate the reactor design, a model is build using clustered CSTRs. The technical feasibility of the concept is proven in experimental tests, which are also used to identify key parameters of the model. Fluidization of the fine CaO/Ca(OH) 2 powder was found to be challenging, but problems were overcome using mild calcination conditions and a special gas distributor plate. Using the model, it is found, that a thermalpowerof15MWcanbeexpectedfromareactorvolumeof100m 3 .Tostudyinfluencesofdifferent parameters on the reactor model performance, a sensitivity analysis is carried out and heat transfer between the reactor and the immersed heat exchangers is found to have by far the largest influence and the reaction system performance. Future research should therefore focus more on heat transfer. © 2018 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).
Introduction
In recent years, the mitigation of climate change and thus the de-carbonization of energy supply systems has become a major goal worldwide (United Nations, 2015). Yet, most sources of renewable energy are fluctuating, resulting in an increasing need for energy storage systems. Alongside electricity storage, heat storage is going to play an important role in the transformation of today's energy system as it is in many cases more efficient and cheaper than electricity storage (Mathiesen et al., 2015;Connolly et al., 2014;Lund et al., 2016).
Thermal energy storage can be subdivided into sensible, latent and chemical storage (Dinçer and Rosen, 2011 • is scalable into multi-MW-scale and • enables a separation of storage power and capacity.
There are many possible applications for efficient and cheap high temperature energy storage with different requirements and specifications. The most prominent case is probably solar thermal power generation (Tian and Zhao, 2013;. Flexibilization of conventional power plants and power-to-heat are two additional ones. Even domestic applications were proposed (Schmidt and Linder, 2016). In the author's opinion, power-to-heat may be the most promising application; therefore, charging using electrical heating of the system is also discussed in this paper.
Reaction (1) was first proposed for energy storage purposes in the 1970s (Wentworth and Chen, 1976;Ervin, 1977). Yet up to now no commercial systems are available and the technology is still in a development stage (Pardo et al., 2014b;Cot-Gores et al., 2012). During the last years, research has been intensified and various lab scale reactors and corresponding models were developed. An overview thereof is presented in Table 1. A review on the topic was recently published by Pan and Zhao (Pan and Zhao, 2017).
In general there are four types of reactors available for gas-solid reactions: • Fixed bed reactors • Moving bed reactors • Fluidized bed reactors (FBR) • Entrained flow reactors Fixed bed reactors are the most common setup in lab scale test rigs (e.g. Schaube et al., 2013a;Yan and Zhao, 2016;Linder et al., 2014;Schmidt et al., 2014;Fujii et al., 1985, compare Table 1), how ether, they come with the disadvantage of discontinuous operation while the other concepts allow for continuous operation. Entrained flow reactors are unfeasible, as they require rather large amounts of carrier gas and a larger reactor volume compared to the other types. A moving bed reactor was proposed by Schmidt et al. (2015). This concept has some advantages, e.g. no carrier gas is necessary, but the operation is difficult as the unmodified fine powders have a very low flowability. Up to now, successful operation of a moving bed reactor over a multitude of cycles has not been reported.
FBRs have the advantage of better heat and mass transfer compared to moving bed reactors and require significantly lower gas velocities compared to entrained flow reactors, therefore they have been widely proposed for thermochemical energy storage (e.g. Criado et al., 2017;Flegkas et al., 2018;Rougé S et al., 2017;Criado et al., 2014a). It is reported, that the fluidization of the fine CaO/Ca(OH) 2 powder is difficult, yet successful fluidization of CaO/Ca(OH) 2 using additives or modified particles has been reported Rougé S et al., 2017;Pardo et al., 2014a).
As mentioned, up to now only lab scale reactors with outputs <10 kW ch have been set up and to the best of our knowledge, no detailed concepts for large-scale reactors have been proposed. On the modeling side, detailed models of fixed bed reactors have been developed (Ströhle et al., 2014;Nagel et al., 2014). For FBRs, Ostermeier et al. developed a model for the gas distributor plate used in the system presented here with detailed hydrodynamics but without heat transfer and chemical reaction and validated it against cold model experiments (Ostermeier et al., 2017). Meanwhile Criado set up simple models to recalculate experiments in two kW-scale test rigs at CEA Rougé S et al., 2017;Pardo et al., 2014a). These models feature a K-L-approach for bubbling fluidized beds according to Kunii and Levenspiel (1991), but do not solve a heat balance. Instead, temperature is given as an input from the measurements (Criado et al., , 2014a. A very sophisticated model for a FBR with the thermochemical reaction system MgO/Mg(OH) 2 was recently published by Flegkas et al. (2018). This model features bundled tube heat exchangers in combination with a K-L modeling approach.
Alongside the development of reactors, the reaction kinetics (Criado et al., 2014b;Schaube et al., 2012) and cycling stability (Nagel et al., 2014;Schaube et al., 2012;Lin et al., 2009) have , Linder et al. (2014), Schmidt et al. (2014) and Fujii et al. (1985) Nagel et al. (2014) Moving bed Indirect heat transfer Schmidt et al. (2015) -Fluidized bed Indirect heat transfer Criado et al. (2017), Rougé S et al. (2017) and Pardo et al. (2014a, Flegkas et al. (2018), Criado et al. (2014a) and Ostermeier et al. (2017) been investigated. The chemical reaction for both hydration and dehydration is rather fast and the material shows complete conversion over a multitude of cycles. However, it is reported, that with cycling, the compressive strength of the particles decreased (Lin et al., 2009) and refinement of the particle size occurred (Schaube et al., 2012). FBR designs with a strong focus on heat transfer between the bed and heat exchangers have previously been developed for other applications. Closest related to the design presented here are applications for indirect thermal storage using sand (Steiner et al., 2016;Schwaiger et al., 2014;Haider et al., 2012) and drying of moist lignite (Hoehne et al., 2009;Schreiber et al., 2011). For both applications, large scale fluidized beds with bundled tube heat exchangers are used. The fluidized bed lignite dryer at the Niederaußem power plant reaches a heat transfer of 75 MW (Hoehne et al., 2009) while the sandTES pilot facility reaches a power in-or output of 280 kW (Steiner et al., 2016).
Related to these concepts, this article presents a FBR design for large-scale thermochemical energy storage alongside a simple modeling approach to identify key influence parameters on the reactor performance. The model is taking into account chemical kinetics as well as heat transfer and different flow schemes within the reactor design and is backed by experimental investigations to identify model parameters.
Section 2 details the proposed reactor design of an industrial scale thermochemical energy storage FBR. In Section 3, the modeling of the reactor is explained. As fluidization of pure, unmodified industrial grade CaO, as used in the reactor concept proposed in this paper, has not yet been reported in literature, Section 4 briefly summarizes some experimental pretests to demonstrate the technical feasibility of the concept and identify key parameters of the model. The model is the used in Section 5 to investigate geometry and flow setup of the reactor as well as to identify crucial parameters and bottlenecks of the system.
Conceptual design of a MW-scale reactor
The main task of the reactor concept is to bring the solid CaO/Ca(OH) 2 particles in contact with the gaseous reaction partner. The second task is to efficiently supply or remove heat at an appropriate temperature level for the chemical reaction from the bed and transfer it to a heat transfer medium that is not in direct contact with the bed.
Reactor type
Looking at the overall investment costs of the storage system, it is clear, that the storage material which is available on industrial scale at ∼0.15 e/kWh plays a minor role. The majority of the costs is due to the reactor and heat transfer system. To keep costs for applications with high storage capacity low, it is mandatory to separate the expensive heat exchangers from the cheap storage capacity. This can be accomplished by continuously passing the storage material through a reactor as shown in Fig. 1. This is not possible in fixed bed reactors, leaving moving-and fluidized-bed reactors as well as entrained flow reactors as possible concepts. Due to the advantages in heat and mass transfer and the superior material handling characteristics, a fluidized bed is chosen in this work as the most promising reactor concept for large scale applications.
Reactor and heat exchanger design
The conceptual design of the reactor is shown in Fig. 2. The fluidized bed is operated as a dense bubbling bed. A special gas distributor plate with a high number of nozzles is used to distribute the fluidization gas uniformly over the reactor cross section. The nozzles are inducing the gas with high velocity and turbulence in horizontal direction, reducing channeling and agglomeration. A more detailed description of the gas distributor plate can be found in Ostermeier et al. (2017). The overall fluidization velocity is moderate and should be reduced as far as possible to minimize the parasitic influence of pressure losses.
To keep the fine storage material within the reactor and have a rather clean fluidization gas at the exit, a high temperature filter system is used. Temperatures in the reactor are estimated in a range between 400 and 700 • C. Therefore a sintered metal filter system is suitable. These filters can be cleaned periodically by reverse flow pressure impulses, keeping even the fine fraction of the storage material within the hot reactor volume.
To improve the residence time distribution, baffles are integrated in the bed leading to a guided flow of solids from inlet to outlet. As proposed by Steiner et al. (2016), additional ''gas cushions'' can be used to assist the directed solid flow through the reactor. The material input can be done by screw feeders (as demonstrated in Schmidt et al., 2015), the outlet by overflow. The reactor is separated from the storage silo by gas-locks.
Up to now, fluidization in lab scale setups was achieved in a mixture of steam and air/nitrogen (Criado et al., , 2014a. During charging operation, steam is released due to reaction (1). This steam contains roughly 40% of the energy required for charging, mostly in the form of condensation heat of this steam. It is thus clear, that this energy needs to be utilized to set up an efficient system integration (as shown e.g. in Angerer et al., 2018;Schmidt and Linder, 2016). If this steam is found in a mixture with air/nitrogen, condensation is not possible at constant temperatures and, depending on steam content and overall pressure, is shifted towards low temperatures. To make efficient use of the condensation energy, it is therefore mandatory to operate the fluidized bed (including filters and gas-locks) in pure steam atmosphere as explained in Angerer et al. (2018). The bed can even be operated at elevated pressures to increase the condensation temperature level, if this is useful for the system integration.
The most efficient way to transfer heat from (discharging) or into (charging) the bed is by bundled tube heat exchangers. The heat transfer coefficient for surfaces inside the bed is approximately four times as high as for surfaces in the freeboard above the bed (Kunii and Levenspiel, 1991). In fluidized bed combustion, this is often not possible due to abrasive ash components. As no abrasion is expected with the fine CaO/Ca(OH) 2 particles, heat exchangers in the form of bundled tubes are placed directly inside the bed. The heat exchanger tubes are packed into the bed as dense as possible, as long as fluidization is not negatively affected. Fig. 2 shows an in-line configuration of the heat exchanger tubes with multiple passes and in-and outlets through the reactor sides. It is possible that other configurations like staggered arrangement or u-elbows with in-and outlet from the reactor top are more advantageous, but these questions are beyond the scope of this work.
The most prominent application for the discharging operation is live steam generation. Therefore, in discharging mode, the heat exchanger is used as a steam generator. This steam generator is operated as a once-through boiler. This sets higher requirements to feed water quality, but ensures a much more advantageous part load behavior. With a circulation boiler, part load could only be achieved by reducing bed height, while in a once-through boiler it can be controlled by the feed water mass flow rate.
As it can be seen from Fig. 2, the system is easily scalable. Length, width and number of baffles can be adapted to find a suitable storage design. The reactor can also be used in a modular setup. Reactors can be clustered in parallel or even serial setup, using the same fluidization medium (FM) in multiple reactor stages.
Modeling
To investigate the reactor design, a model is developed. The aim of the model is not to investigate characteristics of the fluidized bed in detail, but to provide a simple and transparent approach to: • Identify bottlenecks and crucial parameters of the system design and thus guide further research • Provide a reactor model for the investigation of the integration into larger energy systems For bubbling fluidized beds, so called K-L-models are popular Flegkas et al., 2018;Kunii and Levenspiel, 1991). These models split the reactor into two homogeneous phases: a lean phase (bubbles) and a dense phase (emulsion). The dense and the lean phase exchange heat and mass with each other.
As the K-L-model is a mass transfer model, it is particularly necessary, if the composition of the gas phase can be different in bubbles and emulsion (e.g. if due to reaction (1) water vapor is removed inside the emulsion leaving only inert carrier gas behind, reducing the water partial pressure and thus the reaction rate). Recent system integration studies have shown that it is highly recommended to use pure steam as fluidization medium to reach higher system efficiencies . Therefore, the reactor concept presented here is operated with a homogeneous steam atmosphere and no difference in water partial pressure is expected within the reactor. Thus, the only impact of a K-L-model would be the possibility of different temperatures in bubbles and emulsion. In well fluidized beds, it is often assumed, that gas and particle temperature are homogeneous. Applying this assumption, the K-L-model can be reduced to a continuously stirred tank (CSTR) model without any disadvantages in model quality. In conclusion, the main assumptions of the model are: • The solid holdup in the reactor is homogeneous (no local discretization, perfect mixing) • Gas and solids in the reactor have the same temperature (no temperature distribution) • The model is stationary (no startup or shutdown procedures, no time discretization)
CSTR single reactor model
A scheme of the reactor model is shown in Fig. 3. Two different cases are distinguished; case A) with heat exchangers in the bed and a working fluid (WF, indexed with W) flowing through these heat exchangers and case B) with electrical heaters with a fixed surface temperature T H inside the bed.
The overall mass conservation of the reactor can be written as: Withṁ FM being the mass flow rate of the FM andṁ S the mass flow rate of the solids consisting of CaO and Ca(OH) 2 : The conversion X is defined as the molar share of Ca(OH) 2 in the reactor: Consequently the mass fraction of Ca(OH) 2 in the reactor is defined as: The reaction rate can be calculated from the kinetic rate equation dX /dt as: With n S,R being the molar holdup of solids in the reactor volume and dX /dt being the experimentally determined kinetic rate equation. The molar holdup is calculated from the bed volume using the bed density in fluidized state ρ R . It is assumed, that the reactor has a square cross section with the length l R .
H B is the bed height in fluidized state and M S,R the average molar mass of the solids in the reactor. The bed density ρ R is calculated from the bulk densities ρ Bu,i of CaO and Ca(OH) 2 and the measured bed expansion H B /H Bu , which was found to be similar for pure CaO and Ca(OH) 2 fluidized beds.
The kinetic rate equation dX /dt depends on several parameters: As the reactor is operated in pure steam, the steam partial pressure is the same as the reactor pressure. It is assumed, that the pressure losses in the reactor are mainly due to the gas distributor plate, which is situated at the inlet of control volume covered by the model. Thus the steam partial pressure in the reactor is: Due to the filters downstream the reactor an additional pressure loss is induced resulting in a lower pressure at the reactor exit: The mass balance for e.g. CaO can be written as: The energy balance of the reactor can be written as: The enthalpies at the inlet depend on the temperatures of the inlet mass flows: While the enthalpies at the outlet depend on the homogeneous reactor temperature: All enthalpies and heat capacities are calculated using data from the NIST chemistry webbook (NIST, 0000). The heat of reaction ∆H r is calculated according to Kirchhoff's law, using the same dataset. The mass flow of the FM at the inlet is calculated from the superficial gas velocity: The temperature difference between the reactor and the heat exchanger surface is calculated as: If the reactor is heated by an electric heater with constant temperature T H , the temperature difference can be simplified: Finally, the energy balance for the WF inside the heat exchanger can be written as: In case of electrical heating, this can again be simplified to: Overall, in case of heat transfer from a WF, the independent Eq. (12), (13) and (19) are found to describe the three independent variables X , T R and T WF ,out . In case of electrical heating, there are only two independent Eqs. (12) and (13) describing X and T R . However, the result is a nonlinear system of equations that is solved using MATLAB's (Mathworks Inc, 0000) lsqnonlin-solver.
Clustering of CSTRs and model of the complete reactor design
In Fig. 2, several baffles are shown to create a distinct flow of the solids through the FBR. To account for this in the reactor model, each reactor section between two baffles is represented by one CSTR model. Fig. 4 shows a setup of two reactors with three baffles (as shown in Fig. 2) in a countercurrent setup of solids and WF modeled using 8 CSTRs. The FM of the first reactor (first row of CSTR) is collected and used again in the second reactor stage. In this case, the flow scheme between solids and FM is cross-cocurrent. Of course, a cross-countercurrent setup is also possible.
In addition to fluidization, a notable amount of steam is needed to clean the filters at the top of the reactor via reversed flow pressure impulses. In case of the dehydration this sums up to 50 % of the inlet mass flow for fluidization. In case of hydration 25% are needed, as the inlet velocity for hydration is double the velocity for dehydration to compensate for the gas consumption of the chemical reaction. In the model, this is accounted for by adding the FM for the filter cleaning after the reactor. The filter cleaning in downstream reactor stages is also conducted with FM at global inlet conditions (FM in ) not with FM at reactor inlet conditions.
In case of a (cross-) cocurrent setup, the calculation can be performed straightforward using the result of one reactor as input of the next reactor. When (cross-)countercurrent flow is applied, an iterative scheme has to be applied. The calculation is done in direction of the solid flow and the results of the previous calculation are used as inputs for the WF and/or FM side. The iterative procedure is repeated until convergence is reached and the sum of the absolute enthalpy differences between the current and the last iteration is below 0.5 kJ/kg (corresponds to approx. 0.02% of the total value).
Experimental pretests and identification of key model parameters
As mentioned before, fluidization has already been successfully demonstrated using a large share of easy-to-fluidize particles (Pardo et al., 2014a) or particles calcined with high temperatures resulting in reduced reactivity . The aim of the concept presented here is to use pure commercial CaO calcined under moderate conditions, thus providing higher reactivity. To identify key parameters for the simulations described in chapter 3, a number of pretests are conducted.
Characterization of the storage material
Throughout this study, as storage material, commercial Ca(OH) 2 (Weißfeinkalkhydrat CL-90 S according to DIN EN 459-1) and industrial grade CaCO 3 provided by Märker Kalk GmbH from a mine Table 2 Characteristic particle diameters (on particle volume base) of the industrial grade CaCO 3 and CaO gained from that material . in Harburg, Germany is used. The Ca(OH) 2 is used for the investigation of the reaction kinetics. The CaCO 3 is calcined under moderate conditions in a lab furnace without purge gas at 800 • C for 48 h, reaching full technical conversion to CaO. The material is cycled in a fixed bed reactor to investigate the mechanical stability . Cycling is done close to equilibrium conditions in pure steam atmosphere at 1.5 bar absolute pressure by alternating the steam feed temperature between 400 • C (hydration) and 600 • C (dehydration). Table 2 shows characteristic values of the particle size distribution. The moderate reaction conditions can also be expected in fluidized bed systems and lead to a high mechanical cycling stability of the particles. The decreasing value of d 10 indicates that fines are formed during cycling, jet this seems to happen to a much lower extent compared to harsh reaction conditions further from equilibrium as reported in literature (Criado et al., 2014b;Schaube et al., 2012;Lin et al., 2009). On the other hand, the increasing value of d 90 proves that agglomeration takes place as well under the mentioned reaction conditions. Further investigation of the matter is necessary especially concerning attrition during fluidized bed cycling, but the particles show a substantially higher mechanical cycling stability than reported in Criado et al.
Chemical equilibrium and reaction kinetics
Several authors have investigated equilibrium and kinetics of reaction (1) 21) and (22) in comparison with literature equilibrium curves from Barin (2008) and Schaube et al., (2012Schaube et al., ( ). al., 1985. However, most of the measurements used in these publications were conducted in mixed steam/nitrogen atmosphere, resulting in absolute steam pressures below 1 bar. As the operation of the reactor presented in this work is expected at steam pressures ≥1 bar, it is necessary to provide suitable kinetic data for this range. Therefore, own measurements in a TGA/DSC with 5 mg material in pure steam atmosphere between 0.5 and 5 bar were conducted.
A theoretical equilibrium curve can be calculated from thermochemical data from Barin (2008). Literature is not conclusive about whether a hysteresis can be expected for reaction (1). Samms and Evans (1968) as well as Halstead and Moore (1957) proposed no hysteresis, looking at their measurement data, a slight hysteresis especially for low steam pressures might be visible. Matsuda et al. (1985) found a strong hysteresis in their measurements (Matsuda et al., 1985). In the data of Schaube et al. (2012), a hysteresis is visible, jet it is neglected in the discussion and an inaccuracy of the measurement is assumed. The results of this work concerning the chemical equilibrium are shown in Fig. 5. It is not the aim of this paper to discuss the question of a hysteresis of reaction (1). The measurement data indicate, that at least the apparent onset points of the reaction show a certain hysteresis, especially for lower steam pressures. Therefore it was decided, to express the onset of reaction by two separate equations: for the hydration. These curves are not to be understood as a chemical equilibrium with a hysteresis but more like reaction onset points used to describe the apparent reaction kinetics.
The determined rate equations correspond to the second cycle of standard two-cycle measurements, which were conducted for three heating rates (1, 4 and 8 K/min) and at six different vapor pressures (0.5, 1.0, 1.25, 1.5, 3.0 and 5.0 bar). The equations were derived from the measurement data using a genetic algorithm. The equations represent the measurement data with an R 2 -value of 95.6% for the Hydration and 98.5% for the dehydration.
The equation gained for the dehydration is: And for the hydration:
Fluidization and heat transfer
Fluidization of the commercial Ca(OH) 2 and Ca(OH) 2 produced from industrial grade CaCO 3 is tested in a cold model, thoroughly described in Ostermeier et al. (2018). Fig. 6 shows the setup used. It consists of a simple glass cylinder with an inner diameter of 140 mm and an interchangeable gas distributor plate. The heat transfer probe, thoroughly described in Ostermeier et al. (2018), can be removed for pure fluidization tests. Experiments are conducted at atmospheric conditions; the FM is nitrogen at ambient temperature. The key findings of the experiments are: • Fluidization is possible to some extent even for fine powders using a proprietary gas distributor plate design supplied by SCHWING Technologies GmbH (more difficult with usual perforated plates or steel composite cloth) • Fluidization works very well with the in-house calcined material described in Table 2 and a bubbling fluidized bed is formed (as can be seen from a video in the supplementary material of the article) • The bed expansion of the bubbling bed at 0.2 m/s is roughly 125% Footage of the fluidization tests presented at Becker et al. (2018) can be found in the supplementary material of the article.
As the viscosity of hot gases exceeds the viscosity of cold gases, fluidization can be expected to be even better under reactive conditions (Yang, 2003). Hot fluidization under reactive conditions was successfully tested at the facilities of Schwing Technologies GmbH and a 10 kW pilot scale reactor is currently under commissioning at TUM.
The measured heat transfer coefficients for the Ca(OH) 2 powders range from 50 to 350 W/m 2 K , depending on the particle size distribution and thus fluidization quality. For the in-house calcined CaCO 3 , values of ∼ 300 W/m 2 K were measured for fluidization velocities between 0.2 and 0.3 m/s , thus this value is used in the simulations presented here.
Results and discussion
The modeling results shown here account for a fluidized bed volume of 40 m 3 . As the freeboard above the bed requires approximately the same volume, the total reactor volume is roughly 100 m 3 . Of course, the reactor design and thus the thermal output can easily be scaled by changing the length or width of the reactor or by applying a modular reactor setup. The discharging (hydration) case is discussed with heat exchanger and a WF, while the charging (dehydration) is discussed with electrical heating. Of course, charging with thermal energy transferred from a WF is possible as well, but is not be discussed here for reasons of simplicity.
First the model is used to assess the general reactor performance, then degrees of freedom of the reactor design, like flow setup and number of baffles are investigated. Finally the influence of key parameters on the reactor model performance is investigated using a sensitivity analysis
Base case and input parameters
The most relevant input parameters are shown in Table 3. They were partly identified in the experiments, described in the previous chapter. For the dehydration a smaller FM inlet velocity is used as additional steam is produced by the reaction and the outlet velocity is higher than the inlet velocity. This results in a lower pressure loss over the gas distributor and a higher relative share of FM necessary for filter cleaning. The electrical heater temperature was chosen at 650 • C so that a specific load of the electrical heaters of 2 W/cm 2 is not exceeded. A bed height of 1.5 m is chosen based on experience of SCHWING Technologies GmbH. The heat transfer surface area results from packing heat exchangers tubes with an outer diameter of 50 mm as close as possible into the bed while leaving 60% of the horizontal and vertical cross section clear for the fluidization. The 40 m 3 of bed volume include both, the volume of the heat exchanger and the bed volume. As a base case, a reactor with 5 baffles (6 reactor compartments) and two reactor stages, further denoted as 2 × 6-configuration, is used. Fig. 7 shows the results of the reference case simulation of the discharging/hydration. The flow setup is countercurrent for the solids and the WF and cross-cocurrent for the solids and the FM, just as shown in Fig. 4. Fig. 7(A) shows the heat-temperaturediagram for the reactor. It can be seen that approximately 15 MW of heat are transferred in the reference case and the WF reaches an outlet temperature of 450 • C. (B) shows the same temperatures, this time over the 12 reactors in the solid flow direction. The solid feed has a temperature of 600 • C and is rapidly cooled to the reaction temperature of <500 • C, which is slightly decreasing over reactors 2-12. A step in the temperature can be seen between reactors 6 and 7, marking the beginning of the second reactor stage with lower pressure (due to pressure losses in the filters of stage one and the GDP of stage two).
At the exit of the reactor, conversion is almost complete (as can be seen in (C)) and thus the solids are slightly cooled down in reactor 12 reaching a solid outlet temperature of 390 • C. By operating the CaO storage at elevated temperatures (in this case 600 • C) and the Ca(OH) 2 storage at lower temperatures (in this case 350 • C), the temperature gap is used as a sensitive energy storage and increases the energy density in the material by 20%. Of course, this can only be applied for limited cycle durations (<1 week). For long-term storage, both silos should be operated at ambient temperature and the remaining heat in the product at the reactor exit should be transferred to the feed by additional heat exchangers. Fig. 7(D) shows the superficial velocities of the FM at the reactor in-and outlet. In the experimental pretests it was found, that fluidization is possible between 0.1 and 0.6 m/s. As steam is consumed by the reaction in case of the hydration, it can be seen that the velocity at the inlet has to be above 0.2 m/s, as the outlet velocity is lower than the inlet velocity. In the first three reactors, this is not the case due the rather low reaction rate and the higher temperatures in these reactors (compare Fig. 7(C) and (B)). Towards the end of the reactor, when temperatures of the solid and thus the fluidization gas drop while the reaction rate is still high, the superficial velocity at the outlet gets critically low. As the flow direction of the solids is from reactor 1 to 12, this can be expected to have a positive effect on the material transport through the baffles as a slight pressure gradient can be expected to be induced by this behavior.
Influence of the reactor design
For discharging operation, the influence of the reactor design on the heat transferred to the WF is shown in Fig. 8. Three conclusions can be drawn from that figure: • The cocurrent setup for the solids and the WF has clear disadvantages compared to countercurrent setups. However, the cross-countercurrent setup with regard to the solids and the FM (indicated as 'Double countercurrent') is not advantageous to the cross-cocurrent setup. Furthermore, the cross-cocurrent setup in this case is simpler to implement technically, as the solids do not have to be transported against pressure gradients. Thus, the optimal reactor setup is countercurrent for WF and solids and cross-cocurrent for FM and solids.
• The number of baffles (and thus the number of reactors in a row in the model) has a positive influence on the reactor efficiency. On the other hand, an increasing number of baffles may result in solid transport problems as reported by Steiner et al. (2016). To keep the reactor simple and yet efficient, a number of 3-5 baffles per reactor stage is recommended.
• Finally, the influence of the number of stages is investigated using three different setups of 12 reactors. From Fig. 8, it can be seen that increasing the number of stages causes disadvantages in transferred heat. This can be attributed to the decreasing operation pressure over the reactor stages, resulting in lower reaction temperatures in higher reactor stages and thus lower temperature difference between reactor and WF. How ether, this is only one measure for the reactor performance. Inducing more reactor stages significantly reduces the amount of FM needed, as the FM from lower stages is ''reused'' in upper stages. In the two stage setup, only 66% and in the three stage setup only 55% of the FM compared to the one stage setup is needed. Depending on the system integration, this may result in a significant reduction of parasitic losses from the compression of the FM. Thus, a setup with two stages is considered a good compromise.
Similar conclusions were found for other cases investigated (electrical and thermal charging), hence the recommendations concerning the reactor design are applicable for these as well. Figures for the electrical dehydration can be found in the supplementary material of the article.
Sensitivity analysis
To investigate the influence of various model and design parameters on the reactor performance, a sensitivity analysis is carried out. To this end, the parameters are varied with ±25% in steps of 5%, the effect on reactor performance is analyzed and related design trade-offs are discussed. The results for the discharging (Hydration) are shown in Fig. 9. Several conclusions can be drawn: • The predominant influence is the heat transfer represented by the kA-value. To maximize the performance, the reactor should thus be designed and operated to reach maximum heat transfer. This implies a maximum dense packing of the heat exchanger pipes and optimized fluidization for high heat transfer coefficients. Optimizing design and parameters concerning heat transfer should be a key target of future research.
• In the same context, changes in the pressures of the FM and the WF can be seen. An increase in FM pressure shifts the chemical equilibrium towards higher temperatures and Fig. 9. Sensitivity analysis of the reactor model in discharging mode. All values are modified by ±25% in 5% steps relative to the reference case given in Table 3. thus increases the reactor temperature and the heat transferred to the WF. The same effect can be observed by decreasing the pressure of the WF. This results in decreasing evaporation temperatures thus increasing the logarithmic temperature difference and the heat transfer.
• Fig. 9, the inlet velocity of the FM should be chosen as low as possible. Anyway, the model does not cover side effects of the FM velocity on the heat transfer. The heat transfer reaches a maximum between 0.2 and 0.3 m/s . Thus the fluidization velocity should be tuned to reach optimized heat transfer coefficients.
• The reactor volume (and thus the residence time) has a very limited effect on the reactor performance (under the assumption of a constant heat transfer surface area). This leads to the conclusion, that the chemical reaction is sufficiently fast and kinetics have a very limited influence on the reactor performance.
• The FM inlet temperature as well as the bed height have a limited influence on the overall reactor performance. Fig. 10 shows the sensitivity analysis for the electrically heated charging process. Again, the reaction is comparably fast, so there is almost no influence from the reactor volume. The same accounts for the fluidization medium velocity and temperature and the bed height. Again, side effects of this quantities on the heat transfer coefficient that might occur in reality are not covered by the model. The pressure at which the reaction takes place (FM pressure) has a certain influence as it shifts the temperature of the reaction and thus the driving temperature difference according to Eq. (18). As for the hydration, the heat transfer coefficient has a large influence on the system performance, indication that the system is controlled by heat transfer.
The predominant effect is caused by the heater temperature. This is due to its effect on the temperature difference in Eqs. (13) and (18). As the temperature of the reactor stays rather constant, the heater temperature directly affects the temperature difference. E.g. an increase of 5% in the heater temperature (from 650 to 682.5 • C) increases the temperature difference by 32.5 • C which is approximately 25% of its value. Thus the heater temperature is very important. Technically up to 850 • C are possible using immersion heaters. The heater temperature, that can actually be reached, depends on the fluidization quality. If the fluidization is inhomogeneous, heaters are likely to overheat and fail in zones of low heat transfer. To improve fluidization quality and homogeneity is thus one way to ensure efficient electrical heating. Another way is to have a detailed control system for the heater surface temperatures to adaptively decrease heating power and prevent destruction of the heaters in zones of low heat transfer. Both topics should be in the focus of future research.
Conclusion and outlook
In this work, a novel concept for a MW-scale fluidized bed thermochemical energy storage reactor using the reaction of CaO with steam is presented. During preliminary experimental tests, the fluidization of pure commercial CaO/Ca(OH) 2 powder proved to be challenging, however this was overcome by using a special, proprietary gas distributor plate and in house calcined CaO. Further pretests, including TGA/DSC measurements to identify reaction kinetics, heat transfer measurements, hot and cold fluidization tests, were carried out to identify key parameters of the system. Based on these findings, a reactor concept for a large-scale storage reactor is derived and a clustered-CSTR model of this reactor design is implemented in MATLAB.
Using this model, parameter-and sensitivity studies are performed and recommendations and trade-offs to be considered during reactor design are elaborated. It is found that with the given assumptions, a 40 m 3 bed (corresponding to approx. 100 m 3 of total reactor volume) can deliver a power output of approximately 15 MW, generating steam at 100 bar/450 • C in discharging. The heat transfer between the fluidized bed and the WF is found to have by far the largest influence on the reactor performance, indicating that heat transfer controls the reaction in an industrial scale reactor, not chemical kinetics, or mass transfer as suggested in literature (e.g. Criado et al., 2017) for lab scale setups. For electrical charging, a 40 m 3 bed can be used as well to reach approximately the same flow of CaO/Ca(OH) 2 . In this case, heat transfer and the heater temperature are the main influence parameters. Fig. 10. Sensitivity analysis of the reactor model in charging mode. All values are modified by ±25% in 5% steps relative to the reference case given in Table 3.
Results herein indicate that heat transfer is dominating an industrial scale reactor design. It is recommended to focus further research on heat transfer between the fine powder and heat exchangers emerged in the fluidized bed, as this topic is not broadly discussed for the CaO/Ca(OH) 2 thermochemical storage system in literature (e.g. Criado et al., 2017;Rougé et al., 2017;Pardo et al., 2014a). Alongside the heat transfer, further research is necessary to identify ideal fluidization conditions to maximize heat transfer while minimizing parasitic losses, and improve the storage materials cycling stability and physical properties required for fluidization.
The reactor design and modeling approach presented here is also suitable for other reaction systems like MgO/Mg(OH) 2 if these are operated in pure steam atmosphere. The model itself can be used to analyze the integration of the storage into larger energy systems to identify possible applications in industry and power generation. | 9,609.6 | 2018-11-01T00:00:00.000 | [
"Engineering",
"Chemistry",
"Environmental Science"
] |
Operando Li metal plating diagnostics via MHz band electromagnetics
A nondestructive detection method for internal Li-metal plating in lithium-ion batteries is essential to improve their lifetime. Here, we demonstrate a direct Li-metal detection technology that focuses on electromagnetic behaviour. Through an interdisciplinary approach combining the ionic behaviour of electrochemical reactions at the negative electrode and the electromagnetic behaviour of electrons based on Maxwell’s equations, we find that internal Li-metal plating can be detected by the decrease in real part of the impedance at high-frequency. This finding enables simpler diagnostics when compared to data-driven analysis because we can correlate a direct response from the electronic behaviour to the metallic material property rather changes in the ionic behaviour. We test this response using commercial Li-ion batteries subject to extremely fast charging conditions to induce Li-metal plating. From this, we develop a battery sensor that detects and monitors the cycle-by-cycle growth of Li-metal plating. This work not only contributes to advancing future Li-ion battery development but may also serve as a tool for Li-metal plating monitoring in real-field applications to increase the useable lifetime of Li-ion batteries and to prevent detrimental Li-metal plating.
Supplementary Discussion 1: High-frequency current distribution on the boundary between two different conductivity materials
A cylindrical (radius = R, height = Z) coordinate system is used to analyse the high-frequency current density characteristics at the boundary of different conductive materials.Equation 1 is derived from Ohm's law: if the current density = ( , , ) , electric field strength = ( , , ) and electrical conductivity σ(z) are functions that fluctuate only in the Z direction, × = × (). (2) Equation 2 can be described as Eqs.(5) Current continuity ( • = ) simplifies Eq. 5 to Eqs. 6 and 7: Finally, the behaviour of AC is described using Eq. 8 and = : Using Eq. 8, we estimate the current distribution on the σ interface.In the wide cylinder model, aspect R >> Z allows 1 ≈ 0, and symmetry allows = 0.
If () is a constant independent of z, then Eq. 8 can be represented as follows: where The well-known skin effect (Eq. 11) can be derived from Eq. 10. (11) In the battery model, () in the layered materials is not constant and induces additional current distribution along the r axis.Based on the above equations, we observed a trend in at each () boundary.In each region, attempts to distribute according to Eq. 12.In the boundary zone, flows and varies continuously at the same time.The current continuity ( • = ) changes the relation between and : where is set to 0. When each region has ( 1 ) = 1 and ( 2 ) = 2 , the approximated current = (, ) and = (, ) can be calculated in the neighborhood of the boundary where () denotes area-specific constant amplitudes.The total current in the z direction = ; hence, satisfy Eq. 15, The trend of the boundary current can be derived by reformulating Eqs. 13 and 15, When 1 is the conductivity of the active material and 2 is the conductivity of the electrolytic solution, we can estimate the vertical current of the electrolytic solution layer as ( 2 , ) = = with 2 ≪ 1 .Then, Eq. 17 is simplified as follows: where ( 1 ) can be described by B in Eq. 15: Equation 18 summarizes the trend of the current distribution in the surface direction (r) on the higher conductivity surface of different conductivity materials.Moreover, this current is increased monotonically by 1 .Consequently, the Li-metal plating that increases the digits of 1 significantly changes the Hz of a battery.
Frequency
Verification of the ECM for EIS.a-c The ECM of electrolyte Zelec at high frequency.a A cell configuration for verifying the high-frequency response of the Zelec.The cell is designed as an electric double-layer capacitor that uses the same electrolyte and separator as the laminate cell shown in the Supplementary Fig. 3.The thickness of the electrolyte is changed by stacking the separator.b,c Frequency characteristics of the measured real part of the impedance Re[Z] and estimated equivalent circuit of Zelec.Relec represents the electrolyte resistance, and Celec represents the geometrical stray capacitance between the collector plates.The Relec is proportional to the thickness of the electrolyte up to 4 MHz.However, Re[Z] converges to the same value from 70 MHz with increasing value by the skin effect and the proximity effect.This behaviour can be modelled by the RC model shown in (c), which uses the conductivity and relative dielectric constant as σelec= 0.01 [S/m] and εelec = 81, respectively, which are used in other simulations, such as the Supplementary Fig. 2. The estimated RC model has a cut-off of approximately 10 MHz, and Re[Zelec] can converge to zero by increasing the frequency.In addition, Relec can be sufficiently small by increasing the cross-sectional area of Selec in a large-capacity battery.Therefore, the high-frequency measurement can deal with ionic degradation as negligible, and even Zelec might change its resistance by degradation.d,e Example of a low-frequency EIS result in the 18650-type battery ID [#Z] shown in Fig. 4, which has a Li-metal plate by degradation.d The Cole-Cole plot from 0.1Hz to 10 kHz.e Re[Z] vs. frequency.The resistance is broadly increased by Li-metal plating mixed with other degradation factors.Laminate cell modelling and visualization of high-frequency electromagnetic behaviour by computational analysis.a Overall negative-facing cell model in a high-frequency multiphysics simulation (COMSOL6.0).The cell size is referenced in the laminated pouch cell used in the experimental result shown in the Supplementary Fig. 3. b Mesh design.A squared mesh was manipulated as the gradationed size in the edges for monitoring the high-frequency current concentration.c Design of the lamination slices and parameters.The graphite layer was sliced into 10 layers, and special conductivity can be applied to the top two graphite layers to emulate Li-metal plating.d Simulation result of the overall current flow at 10 MHz without Li-metal plating.The current spread in the top and bottom collector layers and down straight in the other layers.e Simulation result of the overall current flow at 10 MHz with the Li-metal layer.There is an additional surface current in the middle of the battery, where the conductivity changes from that of graphite to that of Li metal.f,g,h Overall simulation results.∆Z is calculated by the difference between (d) and (e).
Experimental verification of the negative correlation between Li metal and highfrequency impedance (analytical inspection).a Photographs of the laminate-type pouch cell.The cells were connected to the PCB with SMA connectors.b Table summarising the results of the degradation tests conducted on the laminate-type pouch cells, including images of the anode, degradation conditions and capacity loss of the Li-deposited cells.[I] Initial state of the battery used as a reference.Li metal is deposited in [II][IV][V] batteries via cycle tests.Following a cycle test, [III] is exposed to a storage test at a high temperature.At high temperatures, some of the precipitated Li metal transformed into SEI, resulting in the formation of dark brown precipitates.c Conventional ECM valid for low-frequency 10 .d Proposed ECM applicable to the high-frequency region.The presence of high-frequency impedance Zhf, shown as impedance behaviour in the MHz band, including Li-metal plating, deviates from the conventional ECM. e Cole-Cole plot of the measured impedance curves (Zmes) and fitting curves (Zmodel) by the conventional ECM.These curves were fitted in the range of 100 mHz to 750 kHz and showed good agreement with the measured values at low frequencies, regardless of the degradation condition.f Comparison of the measured and fitted values of the real component of the impedance at high frequency.The measured and fitted values were consistent up to 1 MHz but from 2MHz to 3 MHz.The measured values started to increase, whereas the fitted values monotonically decreased.This difference corresponded to Zhf. g Relationship between the Li metal and Re[Zmes − Zmodel].The negative correlation between the volume of the Li metal and Re[Zmes − Zmodel] can be confirmed.In the electromagnetic simulation, the distinction between [I] and [II]-[V] is equivalent to the Re[ΔZ] in Fig. 3.
Inductance R 0 : Geometry and Solution Resistance R 1-3 : Carrier Transfer Resistance Q 1-3 : Constant Phase Element Z hf : High-frequency electrode impedance that is described Fig.1b
Supplementary Fig. 4 |
Overall high-frequency measurement result of the 18650-type battery used in Fig. 4 (1500 mAh, LFP). a Frequency vs. impedance from 0.1MHz to 100 MHz.Impedance |Z| and Re[Z] are proportional to the frequency.Compared to the laminated pouch cell shown in Supplementary Fig.3, an 18650-type battery behaves as an inductor in this frequency range.In the 10 MHz frequency band, both the real and imaginary components have a convex profile.This local behaviour suggests a second potential for this high-frequency diagnosis that directly monitors the status of health (SOH), as summarised in the Supplementary Fig.5.b ∆Z vs. frequency with dependency against the state of charge (SOC).Re[∆Z] is stable against the SOC status in the MHz range used for Li-metal detection.This feature is important for field use since it does not require battery conditioning equipment such as precise charger systems for SOC adjustment.
Fig. 5 |
Further experimental results for 18650-type batteries.a-d Geometrical features and initial characteristics of the evaluated 18650-type batteries.Battery (a) is LFP/1500 mAh with Re[Z1MHz] = 300 mΩ as the initial value.Battery (b) is NCA/3350 mAh that has Re[Z1MHz] = 290 mΩ as the initial value.Battery (c) is NCM/2500 mAh and has Re[Z1MHz] = 780 mΩ as the initial value.Battery cells (a) and (b) have the cathode terminal in the center of the aluminium collector.The battery cell (c) has the cathode terminal on the edge of the aluminium collector.Only battery cell (b) has an anode collector (copper) in the outer surface, whereas the others have dielectric films between its housing.In the battery shown in (c), only excessive rapid charge degradation results are measured due to a lack of sample cells.e-g Measured impedance change at 1MHz.The results show that our method is not affected by the material but is affected by the structural features.e,h The results of the LFP-type (Repost of Fig.4) and the NCM-type.Since battery (c) has the longest terminal-to-terminal edge, the negative collation between SOH and Re[∆Z1MHz] in (f) can be observed to stronger than that in other batteries.f The result of the NCA-type is scattered compared to the LFP-type (d) or NCM-type (f).The reason for this scattering is assumed to be geometrical noise.The unstable outer contact clearly affects the high-frequency impedance by interfering with the anode collector's current flow.h-j Measured impedance change at 20 MHz in each battery.In contrast to the 1 MHz results, both blue and red dots are blended and aligned along the proportionate lines against the SOH.Although a scientific review is needed, it might be applied as a technique for rapid capacity estimation. | 2,489.2 | 2023-11-10T00:00:00.000 | [
"Engineering",
"Materials Science",
"Physics"
] |
An ISM Based Approach for Product Innovation Using a Synthesized Process
Internet shopping has become a global business activity; it also increases the workload of delivery services. The Transnet logistics can easily deliver and collect cargos from everywhere, with difficulty of having the greatest efficiency. This study is composed of four parts; the first step of identifies opportunities in which proposing an auxiliary vehicle for Transnet truck to support service in narrow or crowded road. A weight calculation system determined the relative value of different items and applied the interpretive structural model (ISM) to modularize SET for constructing innovation from integrated new product development (iNPD). From findingGAPs in opportunities of search results are explored an innovative light electric carrier as an auxiliary vehicle is determined. The second part consists of quantifying 21 items based upon 7 specific attributes for user orientation.The third part describes a new kind of light electric vehicle which is developed and mounted on the rear side of transport trucks; it can be used to deliver goods along narrow roads. Finally, more specific details of a collapsible electric motor carrier with a battery that is recharged using truck’s power system after using it are presented. The results of this study establish a research and product design to help Transnet drivers by improving carrying efficiency in narrow and crowded roads.
Introduction
Nowadays it has become very popular to shop online rather than face-to-face in real stores.The Mainland China site "Taobao net" claims to have a huge business channel from which you can buy anything you want.Although virtual transactions can be made on the internet, the distribution of goods still requires a physical delivery method.Therefore, logistical and distribution industries need to change their systems so as to satisfy the needs of online shoppers.The development of the Transnet service has been a step forward, but it still must improve if it wants to be the most diverse and convenient method.In 1976, the main company of Transnet service, "Takkyubin (Transnet), " evolved from a standard home delivery system.It was pioneered by Daiwa transportation in Japan, and its main business feature is that one can call in and then immediately have goods, even small goods; shipping begins to a specified location for delivery on the next day.In late 2000, this system was first introduced by "TECO" and the "President Group" into Taiwan.Popular carriers such as "Taiwan Pelican" and "President Transnet Corp" vigorously promote the domestic Transnet market through the mass media by whipping up big waves of consumption behavior because the business model is convenient for consumers.
Transnet services gradually expanded into many related peripheral businesses.They quickly transport frozen and perishable cargos, among others, to and from places like convenience stores, using systems like cash on delivery and credit card payment.It has steadily become a more flexible and diverse service, filling a niche in consumption needs.Although this diverse delivery service seems to have a complicated distribution approach, it can deal with a variety of business logistics, although it still needs to improve.Delivering goods anywhere as quickly as possible is an essential requirement for an online business.Each of the goods must reach the consumer's hands in the most effortless way, reducing energy and unnecessary waste.This needs to be a primary goal of the industry.Existing delivery vehicles are mostly large and heavy, preventing delivery to some areas.Therefore, there is a need for a supplementary transportation device, such as a scooter, to collect or send goods to scattered customers' locations in the city.There is still no solution for enabling Transnet services to reach some of the more difficult to reach locations, which means that there is a need for an innovative product or mechanism to solve this problem.
Figure 1 shows that an integrated product innovation process (iNPD) requires the combination of three areas: market research, engineering and product design, and new product development.It is not just a set of methods that can be plugged into an existing company structure.It is a way of thinking that combines three key elements: (1) a truly horizontal and interdisciplinary structure, (2) a commitment to maintain a focus on what customers and other stakeholders value, and (3) a system that begins with an emphasis on qualitative methods of discovery and development and evolves toward quantitative methods of real methods of refinement and manufacture.It is a good way to transfer and combine the marketing research, engineering, and design of products to explore the complex and comprehensive research [1,2].Finally, the process of iNPD can be explored for complex innovative issues, converting an excursive hypothesis into a specific proposal and combining multifaceted fields in a concise approachable way which puts lessons learned from a project-based approach into a form for teaching new cognitive product development to multidisciplinary student teams [3].A proposal has been made to create green Wi-Fi equipment to communicate regarding outdoor activities from iNPD [4].There has been an investigation into developing a highly rational industrial design using the iNPD process of high-tech industrialization by two actual projects through the use of these methods [5].
Outline of the Research Model Development Procedure
This synthesized process creates a combination of design and analysis to construct a product innovation procedure.
Figure 1 shows that the overlap in disciplines of design, engineering, and marketing leads to value of user centered which define the usability, usefulness, and desirability of the product to the customer, the value that also leads to success in the market and profit for the company.Figure 2 shows that there are four phases in this iNPD flow chart: (1) identifying opportunities, (2) understanding opportunities, (3) conceptualizing opportunities, and (4) realizing opportunities [4,[6][7][8].We believe that using this combination for product innovation will produce better solution.
The basic procedure of iNPD has processed an example of a product hybrid that successfully filled a gap was the first Apple iMac.Integrating the monitor and CPU, using translucent plastic combined with a variety of bright candy colors, made the iMac easier and more fun to use than other computers.The iMac evolved with and continued to define the aesthetics of offices and homes, which look sharp with an iMac on the desk.Setup was a breeze, and cable-management issues virtually disappeared.The Apple desktop has continued to use the integration concept and is now an elegant thin, soft-cornered rectangle with a minimal aluminum base [1], It is found that this result has got a usable, useful, and desirable product for customer.
Identifying Opportunities.
There are three phases in this stage.
(1) Selecting and Evaluating SET Factors.This phase introduces a way to collect market information from the essential SET factors (social change: social and cultural trends and drivers, reviving historical trends; economic trends: state of economy, shift in focus on where to spend money, level of disposable income, and technological innovation: stateof-the-art and emerging technology, revaluating existing technology) for locating a better position from which to plan an innovative opportunity.According to research, users' age is related to average daily use of the internet in general and use of the internet for information access [9].In order to collect the factors of all issues of SET, this study uses eight participants whose ages are from 20 to 24 for brainstorming; two participants are graduate students, other two are industrial designers, and other four are the related company drivers.In order to evaluate each SET item, a weighted formula is used to calculate the average value of a particular set of numbers with different levels of relevance.The relevance of each number is called its weight, which is represented as a percentage of the total relevancy.Firstly, each weighted item is given a value 1; the "5" is the most relative to researching topic, gradually reducing number that it represents to decrease the correlation with topic; the weight value is decided by discussion with all participants.This can be seen by the drawn related line between each item of SET; all related lines also have been decided by all participants.Finally, the weighted Avg x is used in formula (1), and the weighted results are shown in Table 1.The greater values of weighted Avg can be used to construct POGs (product opportunity gaps), which are then generalized into fuzzy idea of opportunity.The brain storming and drawing related lines concluded the high scores to build opportunity gaps on qualitative and quantitative methods.The SET factors can identify POGs for a targeted user group.Consider the following: (2) Filling Product Opportunity Gaps.Identifying product opportunities should be the core force driving the companies that manufacture products, supply services, and process information.A product opportunity exists when there is a gap between what is currently on the market and the possibility for new or significantly improved products that result from emerging trend.Figure 4 shows a product that successfully fills product opportunity gaps (POGs) when it meets the conscious and unconscious expectations of customers and is perceived as useful, useable, and desirable [1,2].
(3) Processing Interpretive Structural Model (ISM).Generally speaking, when people study complex and divisive issues and conduct problem analysis and needs assessments, they usually make intuitive judgments based on prior experience rather than rationally assessing the situation.Therefore, they need more effective methods, such as ISM, DEMATEL (Decision-Making Trial Evaluation Laboratory), and the KJ method, which are used for ideal planning [10].ISM is an especially effective method because all elements can be processed with a simple matrix.Interpretive structural modeling (ISM) is a system proposed and developed by Warfield, starting from 1973 [11][12][13][14].It is often used to provide fundamental understanding of complex situations and to put together a course of action for solving a problem.The mathematical foundations of the methodology can be found in various reference works [15].The philosophical basis for the development of the ISM approach is presented in Warfield [11,12], and the conceptual and analytical details of the ISM process were later outlined in greater detail [12][13][14].Interpretive structural modelling (ISM) is a well-established methodology for identifying relationships among specific items.It is useful for defining a specific problem or an issue [16,17].It is a suitable modeling technique for analyzing the influence of one variable on other variables [18], which helps groups of people in structuring their collective knowledge [19].It is used here to refer to the systematic application of some elementary notions of graph theory in such a way that theoretical, conceptual, and computational leverage are efficiently exploited to construct a graph [19,20].
One simple way to describe the relation of the 8 categories within the matrix in this case is to assign weights for the th element according to its relative position in an individual hierarchy.By summing up the individuals, a collective score can be assigned to each element, which can then be used to construct a matrix [] to present the related relationship of each element.Consider the following: Malone [20] terms this the adjacency matrix of , which is constructed by setting = 1, wherever there is an arc in directed from element to element , and by setting = 0 elsewhere.Element is said to be reachable from element if a path can be traced on from to .By convention, an element is said to be reachable from itself by a path of length 0. The reachability matrix of a digraph is defined as a binary matrix in which the entries are 1 if element is reachable from element ; otherwise = 0.It can be shown that the reachability matrix may be obtained operationally from the adjacency matrix by adding the identity matrix and then raising the resulting matrix to successive powers until no new entries are obtained.That is, where is determined such that Some articles published recently on decision making [21,22], design planning [23][24][25], productivity issues [26], and so forth have provided the adequate ground to begin with.
Understanding Opportunities.
To create a breakthrough product, we must know who our customer is and how to place that knowledge in the perspective of marketing our product competitively.In order to understand the possible opportunity of POGs of SET from the first phase, we must continue to verify the opportunity for possible products.As awareness begins to grow, the team produces models of the experience and starts to develop an understanding of the value opportunities for the product.From this, the factors of importance that will make the product useful, usable, and desirable start to emerge.There are seven composite attributes: emotion, ergonomics, aesthetics, identity, impact, core technology, and quality, which are each surveyed for the possible design direction.These attributes are divided into twenty-two value opportunities to process value opportunities analysis (VOA) and to make clear descriptions of POGs through a quantitative survey.There are four steps that can be divided into two phases.(1) Processing of value opportunities analysis (VOA): (A) value opportunities modification: increase or decrease the numbers of value opportunities to fit a fuzzy product idea.(B) Quantitative surveys identify the significant values that have a semantic survey of products from the questionnaire based on 10 segments of a Likert scale.(C) Recognize the importance of value opportunities: when the average scores of descriptive statistics of VOA are less than 4, the importance is deemed as "low." A level of 4 to 8 is deemed as "moderate" requirements, and more than 8 is deemed as "high." Some higher important value opportunities will be focused upon to describe related design concepts.(2) Filling the POGs for caring Transnet drivers to find the fuzzy front end: the high important ones and POGs are integrated as the directions for conceptualizing ideas which can be used as opportunities.
Conceptualizing Opportunities.
In this phase, pictures simulate the preferred orientation of significant value opportunities for target groups which (1) focusing on the LEF (lifestyle, ergonomics, and feature).The user life-style is used to conceptualize the product design direction; integration of style and technology is used to orient an approach from the standpoint of users and is centered on caring for humans and the earth.(2) Moving to the right and upper quadrant we see idea opportunities which must move to a high value quadrant of style, tech, and other values, positioned appropriately for an anticipated product.
Realizing Opportunities.
In order to realize opportunities in this phase, the designer must insist on focusing on an idea opportunity to decrease manpower on Transnet popularity.The proposal will be presented by sketching the concept as a product design.The "care" is a state of mental suffering or of engrossment: to care is to be in a burden mental state, one of anxiety, fear, or solicitude about something or someone [27].To make effort to do something correctly, safely, or without causing damage.The care and design has been constructed as a design issue; it seems to begin from green design, considering humans and the environment to combine motivation [28].In fact, it can let the design get a chance to be reinforced to fit for living creatures; this design concept can be constructed between carers and by-carers to improve product value and create valuable issues for common topics.
Result
After the above mentioned iNPD process, the results can be determined from those four phases, they also have been explained as below description.
Identifying Opportunities.
In order to construct product opportunity gaps (POGs), we have valuated SET and used ISM as effective method for ideal planning.
High Weighted Value.
In order to recognize the value of the opportunity, we selected six people to serve as focus group members to discuss the topic of Transnet from three major related categories: social change, economic trends, and technological innovation.Figure 3 shows all related lines that also have been decided by all participants.Analysis of the discussion results gave us seven social change items, six economic trends items, and five technological innovation items (Table 1).The number of "" is the connection of an item with items from other categories, the "" is the relative weight (1-5) of each item, and, finally, the "weight value" is the score of "" multiplied by "", as in formula (1), which means that we can select the higher importance weight of SET evaluation as samples for ISM.Following the completion of this process, the high scores on behalf of this evaluation have a high value opportunities possibility.All project details are shown in Table 1.The selected high score items from the SET weighted matrix evaluation left us with eight categories: (1) social changes: logistics services (25), internet shopping (24), and Otaku culture (20); (2) economic trends: services commercialization (25) and virtual currency ( 16); (3) technology innovation: mobile network (25), cloud technology (20), and electric vehicles (15).
Filling Product Opportunity Gaps.
The high score items, from the SET weighted matrix evaluation, left us to find a GAP. Figure 4 shows that the value of identifying opportunity is based on experience and discussion to identify the fuzzy front end of the opportunity; we can find a customer-driven product design to use a carrier located on large transport vehicles, it can provide better caring for one person driving in Transnet service.
Processing of the Interpretive Structural Model.
Based on the correlation between elements in a system, customer need assessments are conducted using matrix arithmetic.Directional hierarchical relationship graphics have been generated to formulate execution policies and develop problem-solving strategies.Interpretive structural model (ISM) can compute a model to show our thinking processes in order to deconstruct problems for this study.A few follow the procedure which Hsiao et al. [25] proposed for ISM, illustrated below.
(1) Construct a correlation matrix: to conduct logical operations and analyze the resulting hierarchical structures, we must arrange the high weight value elements (entries) of SET in the form of a matrix to select and sample the structure elements and compare the relationships between the elements of SET (Table 1).To compare the relationships, a directional correlation matrix [] (formula (2)) is formed using the relationship ( ) (formula (3)) between one element and another.Figure 5(a) represents the incidence matrix (original matrix) of an example system containing eight components and displays the incidence relationships.( 2) Generate a reachability matrix: the reachability matrix [] (from formulas ( 4) and ( 5)) is deducted from the incidence matrix [A] if a Boolean -multiple product of [] + [] uniquely converges to for all integers > 0 , where 0 is an appropriate positive integer, [] is a Boolean unity matrix, and + is an addition in the Boolean sense [29].Matrix [] represents all direct and indirect linkages between components.Relation transitivity is a basic assumption in ISM. Figure 5(b) represents the reachability matrix [] derived from matrix [], in which an entry = 1 if component is reachable by , although the path length may be one or more."Reachability" in graph theory is the ability to move from one vertex in a directed graph to some other vertices (formula ( 4)).This is sufficient to find the connected components in the graph.(3) Generate a rearranged matrix: cluster elements that affect one another in the output matrix of the reachability matrix.Figure 5(c) reveals three clusters in the system, namely, {1, 2, 3, 8}, {4, 5, 6}, and {7}, and the clustered components are integrated and treated as a single entity.(4) Illustrate the hierarchical relationships of elements: the hierarchy graph is then obtained by identifying a set of components in matrix [] that cannot reach or be reached by other components outside the set itself.The oriented links then connect the nodes from source to sink based on the incidence matrix.Notably, the rounded rectangles in Figure 6 indicate clusters within the retrieved group, in which the information flow forms a loop.In this step, the elements' hierarchical relationships are illustrated according to the rearranged matrix, providing decision makers with the procedures and hierarchical structures to use in the deconstruction of a problem.(5) Draw the + − element distribution graph: based upon the reachability matrix [], the user must add up the scores of the elements in each row to generate and the elements in each column to generate .After this, calculate the values of + and − to generate a reachability matrix determinant (see Table 2) and then demonstrate + − on a binary scale to interpret both the problem and the target areas.This system can also be employed to analyze independent elements according to their hierarchical relationships.After the element level has been determined, draw the + − element distribution graph to find out the main problem and main target (see Figure 7).Once this has been accomplished, the power-driven and easy to use vehicles can be loaded on a large vehicle which can provide better quality of Transnet for driver from ISM process.This analysis can make the selected SET topics to
Understanding Opportunities. Value opportunities (VO)
of proposed direction can be classified into specific attributes that contribute to the usefulness, usability, and desirability of a product, connecting the features of the target product to evaluate those values from specific attributes.There are seven attributes: emotion, ergonomic, aesthetic, identity, impact, core tech, and quality.
Processing of Value Opportunities Analysis (VOA).
First, the original opportunities of the attribute must be modified to fit the real user needs, and each of the seven attributes must contribute to the overall experience of the product.In doing this, the 22 original value opportunities will be adjusted to 21 items.Of these 21 items, 6 have emotional attributes, 2 have ergonomic attributes, 4 have aesthetic attributes, 3 have identity attributes, 2 have impact attributes, 2 are related to core technology attributes, and 2 are related to quality attributes.Then, to process the quantitative survey section Foldable, easy to carry cargo, small electric vehicles of the questionnaire, the question items are ranked using a 10-segment Likert scale.This questionnaire survey was taken from May 23 to 29, 2013, using a website.Twelve expert examinees participated in this internet survey, 41.7% of the examinees were male, and 58.3% were female.Figure 8 shows that the 21 items have average scores.If the scores are over 8, they are considered high level items and are selected to conceptualize the new product development.
After calculating, the high level scores are selected as shown below (Figure 8).Items related to "comfort, " "point in time, " and "confidence" rate over 8.4; "reputation" is 8.2; items of "security" and "social" are 8.1; items of "environment, " "sense of place, " and "independence" each score 8.0.
The POGs of Creative Design for Caring Transnet
Drivers.In Table 3, we induct those high value items to find the detailed decision for fuzzy front end: (1) Transnet truck needs to pass the narrow streets or alleys, (2) it is easy to drive and needs to be able to charge at any time, and (3) it is collapsible, easy to carry cargo, small electric vehicles, which are some design value opportunities for supporting logistic Transnet system.
Conceptualizing Opportunities.
Here we try to conceptualize the proposed design direction and value opportunities and select the empirical data and major items to describe the fuzzy front end of opportunities: imagination of the user scenario and design approach the upper and right quadrant to explain the location of design opportunity.
Focusing on the LEF-Scenario of User Working and
Product Environment.Figure 9 shows the popularity of smart devices used for special tasks combined with the commercialization of online shopping for service.Together, these suggest the need to achieve a mechanism for Transnet service.The scenario of the user's working and product environment will be with narrow alleys and crowded cars tide where the user drives a truck for desirable, usable, and useful Transnet service.A new product is needed to solve the problems arising from those situations.It also needs to depend on LEF to construct more clear design direction as following: (1) lifestyle impact: a customized carrier for Transnet drivers, (2) ergonomics: an uninterrupted power vehicle with electrical storage, and (3) feature that must be durable and useful for users and the feature becomes minor issue.other quadrants shows us the high style, high-tech, and high values for target products to approach the first quadrant, as in Figure 10.Finally, a design idea can be chosen in which a foldable vehicle is used to access urban narrow alleys, with a lock at the side of truck for convenience.
Realizing Opportunities.
From opportunity to conceptualization, value is based on experience and statistics.Using these, we can recognize and identify the fuzzy front end of the design opportunity.We find that a power-driven and easily usable light vehicle which can be loaded on large vehicles will provide better quality Transnet for drivers and customers.
3.4.1.
Reconfirming the Design Opportunity from LEF.We can reconfirm this design process for delivery shuttle trucks.
It is difficult for drivers to deliver goods to customers because there are still many small and narrow lanes in the city.Although delivery personnel can use a small feeder carrier, they cannot work efficiently.As discussed by the LEF (lifestyle, ergonomics, and feature), a power-driven vehicle can be adapted to drive short distances in urban areas, one which can be charged from a truck on which it can be placed on the side or rear.Such a vehicle can be used for frequent short electric-powered driving without a large rechargeable battery.
Designing a Light
Vehicle for Supporting Truck Driver on Transnet Service.In order to reduce the volume of the vehicle, the design must be collapsible, the first idea is as Figure 11 and the final proposal is as Figure 12(a), and it is with extendable bar to enlarge the cargo volume, Figure 13(a) shows it is with an auxiliary electric wheel hub motor with battery, Figure 13(b) shows that it can charge from truck.This would be convenient and save space and can be located on the side or rear of Transnet truck where it could be easily picked up and used by the truck's drivers.They could add another carrier to carry more goods as shown in Figure 14.
Conclusion
This analysis can make the selected SET topics understand their relationship and what the main problem, main target, and connection lines are.To make effort to do some modification, thinking more safe and more space to care driver.The care and design have been constructed as a design issue; it seems to begin from green design, considering humans and the environment to combine motivation [28].In fact, it can let the design get a chance to be reinforced to fit for living creatures; this design concept can be constructed between carers and by-carers to improve product value and create valuable issues for common topic.This study proposes an application of ISM to make design direction in the new integrated product development from determining the interrelated strengths on each topic.This synthesized process based on ISM to make it more methodical and logical is easier and less time to achieve a complete design procedure which is possible to be done by designers themselves.The methodology of this study can be applied to the beginning stage of product innovation, particularly in the design process of diverse environmental factors.In addition to developing a suitable product for special needs, it can also be applied to establish new methods.Though a caring design is taken to care for logistic Transnet, the proposed method can generally be applied to overlap of disciplines leading to value for user centered design research.Finally, it is responsible for the process of iNPD to propose a light electric vehicle for city logistic Transnet.An electric folding vehicle has excellent maneuverability in narrow alleys in the downtown area, combined with the concept for logistics services.Such a vehicle will increase service efficiency and decrease manpower cost.A power-driven vehicle can be adapted to drive short distances in urban areas, one which can be charged from a truck on which it can be placed on the side or rear.It owns elastic extending platform to increase space carrying light and big cargo.There is an auxiliary electric wheel hub motor with battery under platform and standup space for riding if the platform space is broader enough, therefore, we can imagine this light electric folding vehicle has excellent maneuverability in narrow alleys, the electric power and broader carrier vehicle can increase service efficiency and decrease manpower cost.
Figure 2 :
Figure 2: Flow chart of research and design for Transnet industry.
Figure 3 :•
Figure 3: The gaps concluded from all connections between each SET item.
Original incidence matrix [] • [R] t = 8 (c) Rearranged matrix and retrieval of clusters
Figure 8 :
Figure 8: The scores of value opportunity analysis (VOA).
Figure 9 :
Figure 9: The scenario of desirability, usability, and usefulness of Transnet service.
Figure 10 :
Figure 10: The design idea must approach right and upper quadrant and other reference products.
Figure 11 :
Figure 11: (a) An initial design idea.(b) Two places for locating carrier from side or rear for final design.
Figure 13 :
Figure 13: (a) Explaining electric components for light carrier.(b) The charging socket.
Table 1 :
The high value elements from weighted calculation.
Table 3 :
The fuzzy front end of creative design for caring Transnet drivers. | 6,679.4 | 2014-09-14T00:00:00.000 | [
"Engineering",
"Business"
] |
The reproductive index from SEIR model of Covid-19 epidemic in Asean
As we calculate analytic to link the coefficient of third-order polynomial equations from raw data of an Asean to the SEIR model. The Reproductive index depending on the average incubation period and the average infection period and the coefficient polynomial equations fitted from raw are derived . We also consider the difference of the average incubation period as 5 days and 3 days with the average infection period as 10 day of an Asean. We find that the value of are Indonesia (7.97), Singapore (6.22), Malaysia (3.86), Thailand (2.48), respectively. And we also find that Singapore has 2 values of as 1.54 (16 Feb to 37 March) and 6.22 (31 March-4 April).The peak of infection rate are not found for Singapore and Indonesia at the time of consideration. The model of external stimulus is added into raw data of Singapore and Indonesia to find the maximum rate of infection. We find that Singapore need more magnitude of external stimulus than Indonesia. And the external stimulus for 14 days can stimulate to occur the peak of infected daily case of both country.
Introduction
An Asean (The Association of Southeast Asian Nations) is consist of 10 ten member states as Indonesia, Malaysia, Philippines, Singapore, Thailand, Brunei Darussalam, Viet Nam, Lao PDR, Myanmar, and Cambodia. And we try to predict the number of coronavirus (Covid-19) victims as number of persons who caught the infection and got sick only in this area. The complicated mathematical models are necessary for long-time predictions. The SIR (Susceptible-Infectious-Removed) model are used to obtain the prediction values of the model parameters using the statistical approach for predication the number of infected, susceptible and removed persons [1,2]. The Susceptible-Infectious-Recovered/Death (SIRD) Model was used to formulate an optimal control problem with an expanded epidemic model to compute (Non-pharmaceutical) implementation strategy. [3] A modify SIR called SEIRUS model (Susceptible -Exposed -Infectious -Removed -Undetectable -Susceptible) is generated for evaluate the new deterministic pandemic Covid-19 endemic that originally developed for the control of the prevalence of HIV/AIDS in Africa. [4] The Susceptible-Exposed-Infectious-Removed (SEIR) model was adopt to be SEIRNDC [5] that the total population size N with two extra classes "D" mimicking the public perception of risk regarding the number of severe and critical cases and deaths; and "C" representing the number of cumulative cases. This model proposed the compartmental model that sustained human-to-human transmission of Covid-19 after December 2019 of Wuhan, China.
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted April 29, 2020. .
In this research, we use the daily cases and total cases of Covid-19 infection population of an Asean from website : Worldometer [6]: https://www.worldometers.
info/coronavirus/ between 15/02/2020 to 15/04/2020 . The raw data are fitted with the regression method of third-order polynomial formula. We calculate analytic to link the coefficient of polynomial to the Reproductive index (R0) of the SEIR model (Susceptible-Exposed-Infected-Removed).
Model
We apply the well-known SEIR compartmental model [7] (Susceptible-Exposed-Infected-Removed) for the prediction properties of how a disease spread. The variable description is ( ) is number of susceptible populations, E(t) is the number of exposed populations, ( ) is number of infected populations, ( ) is number of infected populations quarantined and expecting recovery at time. There is no emigration from the total population and there is no immigration into the population. A negligible proportion of individuals move in and out of at a given time that The people susceptible are able to get infected when they contact infectious people. Once infected, they move into the infectious compartment. People recovered are assumed to be immune within our study horizon. Then The dynamics non-linear differential equations are below . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted April 29, 2020.
) ( ) ( Here, the , and denoted the infection rate, the onset rate, and the removal rate. The In our model, we assume that the rate of infection is in the form of third-order polynomial formula as 3 4 Integrate with respect to time, we can get the total accumulation of infective population that 4 3 2 ) ( . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted April 29, 2020. .
Here 0
a is the total cases found in the day before. After take some calculate on solving the first order differential equation on Eqs. (1- ) 20 12 6 2 ( Substitution Eq.(6), Eq. (7) and Eq.(8) in to Eq.(2) , we get By setting t=0, and , the reproductive index in the parameters of raw data is derived as . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
(which was not certified by peer review)
The copyright holder for this preprint this version posted April 29, 2020.
The reproductive rate or the reproductive index, R0 is the course of an epidemic shape. It represents the number of further cases each new case will give rise to. For high value of R0, the number of newly infected people climbs more quickly to a maximum than low value of 0 R . The higher R0, the higher infectious population are.
And base on WHO, we set . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
(which was not certified by peer review)
The copyright holder for this preprint this version posted April 29, 2020. (12) Here 0 a is the accumulation infected population before the date of calculation. 2 1 a and a are constant values and the coefficient of term "t" of dt t di ) ( , respectively.
Results and Discussion
When the spreading of viruses in a population occurred, the number of new cases rises rapidly, peaks, and then declines called the epidemiological curve. The spreading curve should be the flatten as the spreading the infections out of time. In this paper, we model the shape of spreading rate to be the third-order polynomial that we called "bellshape ". According to the total accumulate infected cases, the shape of curve should be gotten the saturation value so that the rate of infectious should be the bell-shape at the end of disease spreading. After analysis the raw data, we find that the daily new cases of Asean are divided into 2 cases; the bell-shape and no bell-shape. The bell-shape cases mean that the infection is in the state of the beginning of the saturated status as we can find the maximum value and we cannot find the maximum value for no bell-shape case.
However, both cases can be solve to find the 0 R .
An Asean is consist of 10 ten member states as Indonesia, Malaysia, Philippines, . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
(which was not certified by peer review)
The copyright holder for this preprint this version posted April 29, 2020. As there are only 4 countries that have enough data to fit well with third-order polynomial; Malaysia, Thailand, Singapore, and Indonesia. There for, Philippines also has enough data but raw data show the swing-type behavior so the third-order polynomial equation cannot fit well, while the other country have not enough data to conform our . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
(which was not certified by peer review)
The copyright holder for this preprint this version posted April 29, 2020. We take the extrapolate on equations to reach the end of time so "Bell-shape" occurred.
To compare with the equal peak, we normalized the results with the maximum of each country and shown in Figure 1. . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
(which was not certified by peer review)
The copyright holder for this preprint this version posted April 29, 2020. This result is agree with more R0 of Malaysia than R0 of Thailand as Table 1.
The raw data of Indonesia and Singapore have shown that these countries take more time than the other country to get a peak. So, their curves do not show the Bellshape at the time of consideration. However, we can calculate the R0 of them that are 6.22 and 7.97 for Singapore and Indonesia, respectively. The high value of the R0 mean that they will take more time to get a peak and the infective population will be abundant if government do nothing.
In this situation, the external stimulus is need for stimulate the system to exhibit the new character such as the FitzHugh -Nagumo equations [12,13]. The Eq.(5) is similar to the FitzHugh -Nagumo equations then we add the external stimulus , The external stimulus will reset the variable of system to become new value. If the external stimulus exceeds a certain threshold value, the system will generate a new behavior. In our model, we need the external stimulus to reduce the exponential increase . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
(which was not certified by peer review)
The copyright holder for this preprint this version posted April 29, 2020. And this constant is equal to the latest value of daily infection per day.
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
Conclusions
The Covid-19 raw data of infected rate of Asean are fitted with the regression method of third-order polynomial formula under the scope of SEIR model. We derived the reproductive index and set into 2 equations as is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted April 29, 2020. . https://doi.org/10.1101/2020.04.24.20078287 doi: medRxiv preprint [14][15][16][17]. However, the peak of infected rate is not found for Singapore and Indonesia at the time of consideration. The model of external stimulus on raw data are added into polynomial equation of Singapore and Indonesia to control the Covid-19. We find that Singapore need more the magnitude of external stimulus than Indonesia. However, the external stimulus for 14 days can stimulate to occur the peak of infective daily case of both country. | 2,623.6 | 2020-04-29T00:00:00.000 | [
"Mathematics",
"Environmental Science"
] |
Seismic Imaging of the North American Midcontinent Rift Using S‐to‐P Receiver Functions
Abstract North America's ~1.1‐Ga failed Midcontinent Rift (MCR) is a striking feature of gravity and magnetic anomaly maps across the continent. However, how the rift affected the underlying lithosphere is not well understood. With data from the Superior Province Rifting Earthscope Experiment and the USArray Transportable Array, we constrain three‐dimensional seismic velocity discontinuity structure in the lithosphere beneath the southwestward arm of the MCR using S‐to‐P receiver functions. We image a velocity increase with depth associated with the Moho at depths of 33–40 ± 4 km, generally deepening toward the east. The Moho amplitude decreases beneath the rift axis in Minnesota and Wisconsin, where the velocity gradient is more gradual, possibly due to crustal underplating. We see hints of a deeper velocity increase at 61 ± 4‐km depth that may be the base of underplating. Beneath the rift axis further south in Iowa, we image two distinct positive phases at 34–39 ± 4 and 62–65 ± 4 km likely related to an altered Moho and an underplated layer. We image velocity decreases with depth at depths of 90–190 ± 7 km in some locations that do not geographically correlate with the rift. These include a discontinuity at depths of 90–120 ± 7 km with a northerly dip in the south that abruptly deepens to 150–190 ± 7 km across the Spirit Lake Tectonic Zone provincial suture. The negative phases may represent a patchy, frozen‐in midlithosphere discontinuity feature that likely predates the MCR and/or be related to lithospheric thickness.
Introduction
The shallow, dense igneous rocks of North America's 1.1-Ga failed Midcontinent Rift (MCR) System create a striking magnetic anomaly and gravity high that demarcate the rift axis (Chase & Gilmer, 1973;Hinze et al., 1992;King & Zietz, 1971). The system consists of two arms that converge in the Lake Superior region-one extending southeast through Michigan and one extending southwest through Minnesota, Wisconsin, and Iowa to Kansas-overall spanning a distance of~3,000 km . Despite its size and prominence in geophysical surveys, the formation, evolution, and failure are still being debated and particularly at a lithospheric scale are not well understood.
A leading question for the MCR is whether rifting was initiated via passive or active rifting. Passive rifting attributes rift initiation to lithospheric tensional stresses that cause extension and thinning of the lithosphere, generating passive decompression upwelling of mantle material. Active rifting attributes impingement of anomalous large-scale hot mantle upwelling and magmatism as the cause, driving thermal erosion and lithospheric weakening that leads to isostatic uplift, which causes tensional stresses (White & McKenzie, detected by the Superior Province Rifting Earthscope Experiment (SPREE; Wolin et al., 2015), the Earthscope Transportable Array, and the US backbone network on the rift and surrounding area. Here we use S-to-P (Sp) receiver functions to illuminate the depth and character of seismic velocity discontinuities throughout the lithosphere and how they vary laterally over the rift and its flanks along the southwestward arm. Constraining variations in the crust-mantle boundary and other lithospheric features aids the evaluation of the extent and role of magmatism during the 1.1-Ga rifting event, in that evidence of lithospheric alteration from past events of magmatism may still remain. Finding the presence or not of these signatures and constraining them at a fine scale are important for a better understanding of rifting dynamics and how a rift may initiate and subsequently cease in the confines of a continent.
Background
Along the arms of the MCR, away from the Lake Superior region, volcanic rocks associated with the MCR rifting event are mostly buried by younger sedimentary deposits (Allen et al., 2006). There are outcrops and drill samples collected that reveal dense mafic rocks overlain by less dense sedimentary sequences, although these samples are few (Ojakangas et al., 2001). Thus, in order to study the nature of the rift along the rift arms, studies must rely on interpretation of seismological, gravitational, and magnetic data and extrapolation from studies of sparse outcrops. Seismic profiling of Lake Superior and the rift's arms and extensive geochemical and isotopic studies of exposed and well-preserved Lake Superior rocks have already constrained a complex history of the MCR rifting events, including crustal extension Hinze et al., 1992), volcanism that filled the rift basin with large volumes of flood basalts (Paces & Miller, 1993;Vervoort et al., 2007), and thermal subsidence and sedimentation (Cannon, 1992). Seismic profiles also reveal thrust faults that bound the rift axis in Lake Superior and the southwestern arm, which are inverted normal faults that primarily accommodated extension (Chandler et al., 1989;Hinze et al., 1992). Raising of the central graben of the rift along these reverse faults is thought to have begun~10-20 Myr after extension had ceased and is a result of compression in the 1.3-0.98-Ga Grenville Orogeny (Cannon, 1994). Crustal shortening lasted~20 Myr and amounted to a shortening of~30 km in the southwestern arm of the rift, based on seismic reflection profiling that recognizes marker horizons between flood-basalt sequences and overlying sedimentary rocks (Cannon, 1994).
The prerift crust is likely to have undergone extensional crustal thinning to less than one third of its original thickness Hutchinson et al., 1990)-however, given its complex history, the resulting MCR crust and mantle structure cannot be simply characterized by rift-related thinning. Instead, seismic imaging suggests the existence of multiple layers of crust and upper mantle within the MCR. Beneath the Lake Superior portion of the rift, active-source reflection seismic surveys reveal shallow crustal additions up to 30 km thick, interpreted to consist of extruded flood basalts during rifting and mainly clastic sediments deposited in the thermally subsiding rift basin . Furthermore, underplating of the crust has been inferred by the presence of intermediate density and seismic velocity between the crust and mantle beneath the rift axis in Lake Superior based on gravity modeling and reflection seismic images (Hutchinson et al., 1990). Receiver function studies have also observed weaker or incoherent phases that are suggestive of a more complex transition from rift crust to mantle beneath the southwestern arm of the MCR (Moidaki et al., 2013;Shen et al., 2013;Zhang et al., 2016), where three-dimensional density modeling also requires intermediate density at the base of the rift crust to satisfy the gravity and topography variations (Levandowski et al., 2015). P-to-S (Ps) waveform fitting and H-κ stacking reveal a crust-mantle transitional layer that is inferred as crustal underplating, up to 25 km thick, beneath SPREE stations located over the rift axis beneath the northern segment (Wisconsin-Minnesota border) and middle segment (Minnesota) of the rift, which is bounded by two weaker discontinuities, the lower of which extends to depths of up to 60 km (Zhang et al., 2016). Further south beneath the southern segment of the rift (Iowa), using a more sparsely distributed, across-rift station line, H-κ stacking has constrained phases related to a velocity increase with depth that are weaker at two stations on the rift axis than the rift flanks and that occur at depths of up to 53 km (Moidaki et al., 2013).
Data
Two data sets are used: teleseismic earthquakes recorded from January 2011 to December 2013 by 82 seismic broadband stations from SPREE (Wolin et al., 2015); and teleseismic earthquakes recorded from 1996 to 2014 by the US backbone station network and the Earthscope Transportable Array stations deployed at roughly 70-km spacing in the region, both together we refer to as the backbone array. The SPREE stations were mainly located over the rift axis, with 66 of the stations distributed along the rift axis and in two riftperpendicular lines (Figure 1)-the other 16 stations are located in Ontario north of Lake Superior. Events located at epicentral distances of 55-80°from the stations are used (Rychert et al., 2007;Wilson et al., 2006). Events with M w ≥ 5.0 are considered from SPREE, and events with M w ≥ 5.8 are considered from the backbone array, forming a data set of 25,401 event-station pairs: 15,901 event-station pairs from the backbone array and 9,500 from SPREE.
S-to-P Receiver Functions With Extended-Time Multitaper Deconvolution
Each event is rotated into theoretical P and S components using a free-surface transformation matrix (Bostock, 1998). We divided the region into areas that are thickly sedimented (approximately ≥1 km thick) and those that are not based on waveform fitting of Ps receiver functions beneath SPREE stations by Zhang et al. (2016). In locations with thick sediment, we assumed surface velocities V p = 4.00 km/s and V s = 2.00 km/s; otherwise, V p = 5.90 km/s and V s = 3.41 km/s. The parent S wave is manually picked, thereby eliminating any unclear S wave arrivals. After this elimination, we were left with 7,964 event-station pairs in the study region (5,582 backbone array and 2,382 SPREE). Each parent S wave is deconvolved from the daughter signal using extended-time multitaper deconvolution (Helffrich, 2006;Rychert et al., 2012) to calculate the receiver functions. Receiver functions that from 0 to 70 s exhibit same-frequency, similar amplitude signals were deemed unstable and are eliminated by inspection, leaving 5,279 Sp receiver functions (3,162 States are labeled: MN, Minnesota; WI, Wisconsin; IA, Iowa. Gray area in the northeast is Lake Superior. Bouguer gravity anomaly data used from Kucks (1999). Red box in inset (bottom right) shows study region. backbone array and 2,217 SPREE). We multiply the receiver functions by À1 so that a positive amplitude indicates a velocity increase with depth and a negative amplitude indicates a velocity decrease with depth, consistent with Ps receiver function studies.
A band-pass filter of 0.02 to 0.5 Hz is applied to the deconvolved waveforms, which are then migrated to depth along the ray paths of the respective Sp phase and stacked into a 0.25°× 0.25°× 1-km grid (Angus et al., 2009;Rychert et al., 2012). Grid points (bins) with less than five hits are discarded, and the grids are subsequently smoothed according to the Fresnel zone (Fowler, 2005) of each waveform, with a minimum width of 20 km. Hit counts in each bin at depths of 36, 105, and 150 km are presented in Figure 2.
Migration Model
The one-dimensional migration model for each receiver function is obtained by tracing the approximate Sp ray path through a three-dimensional Earth model that is based upon previously determined rift and near-rift Earth structure (Zhang et al., 2016) and US-CrustVs-2015 Moho depths in the outer bounds of the grid. For the rifted region, Moho beneath the flanks, the base of an underplate layer beneath the rift axis, and the crustal V p /V s values are defined, which we base on H-κ stacking and waveform fitting results beneath the SPREE stations (Zhang et al., 2016). We assumed these values extend both further north and also south along the rift axis to fill in the larger area considered in our study. The grid then undergoes interpolation to create a grid with 5-km spacing and Gaussian smoothing over a 10-km length scale. The Earth model also includes a sediment layer over the rift. The sediment extent is based on the extent of the Bouguer anomaly map (Kucks, 1999) that illuminates the rift-the high Bouguer anomaly is assumed to be the rift-and sediment depth is based on Zhang et al. (2016). The final crustal thickness used for migration is determined by the Moho piercing point of each waveform in the above model. The crustal P wave velocity structure is designed to be a linear gradient centered on the average P wave velocity of the North American crust (Zhang et al., 2016): V p = 5.90 km/s at surface and V p = 6.90 km/s at the crustal thickness of the Earth model described above. In areas of sediment, we assume a linear gradient through the sediment and the crust with V p = 4.00 km/s at the surface, V p = 5.90 km/s at the base of the sediment, and the same linear (Kucks, 1999). Thinner black lines indicate state boundaries.
10.1029/2018JB015771
Journal of Geophysical Research: Solid Earth gradient to V p = 6.90 km/s at the crustal thickness in the Earth model. For the mantle, we assume values from IASP91 (Kennet, 1991).
Error
Uncertainties for discontinuity depths are defined by changing the V p /V s ratio in the model used for migration by 0.1 via changes to V s , which encompasses extreme mantle lithosphere values predicted by compositional variations (Hacker & Abers, 2004) and most crustal variations. Increasing/decreasing the V p /V s ratio by 0.1 results in midlithospheric discontinuities that are 7 km shallower/deeper on average. The Moho phase is 4 km shallower/deeper on average when this change is applied.
Uncertainties for the amplitudes of the observed phases are defined by the 95% confidence limits, assuming Gaussian statistics, based on the standard error of the mean of the stacked receiver function amplitudes in the associated bins.
Synthetic Waveform Modeling
We perform synthetic waveform modeling of discontinuity structures representative of different locations in the study region. For each location, we forward model a shear velocity-depth profile for a onedimensional slice of our gridded and stacked receiver functions using one-dimensional reflectivity synthetic waveforms (Shearer & Orcutt, 1987). Only phases that are significant from zero according to 95% confidence limits are modeled. Processing of the synthetic waveforms is equivalent or similar to that used in producing the Sp receiver functions-a band-pass filter of 0.02 to 0.5 Hz is applied, and the migration model used is based on the location of each synthetic waveform. We use a ray parameter of 0.1058 s/km, which is the average of the values used in the study using the entire epicentral distance range of 55-80°. In a test to constrain complex structure beneath the rift using a smaller epicentral distance range of 55-60°(section 4.3), a ray parameter of 0.1178 s/km is used. In the synthetic models, V p /V s ratios of 1.72 and 1.80 are used to define the velocity-depth profiles for the crust and mantle, respectively. We use an upper crustal density of 2.80 g/cm 3 , a lower crustal density of 2.95 g/cm 3 , and a mantle density of 3.32 g/cm 3 . The synthetic waveforms and required shear-wave velocity-depth profiles are presented in Figure 3.
Positive Velocity Discontinuities
The Sp receiver functions illuminate a positive polarity phase (seismic velocity increase with depth) that persists over the study region at depths in the range of 33-40 ± 4 km ( Figure 4), consistent with Moho depths in the region (Zhang et al., 2016). Little depth variation from rift flank to rift axis exists outside of our 4-km error bounds. At a distance of~200-300 km east of the rift axis, there is possibly slightly thicker crust throughout most of the region with Moho depths of 37-40 ± 4 km. There appears to be a strong spatial correlation between the rift axis and a weak Moho phase (amplitudes less than 0.06). Away from the rift axis, amplitudes are stronger with values of 0.07-0.12 ± 0.04 in the east and northwest of the study region and values of 0.06-0.08 ± 0.04 in the west and south.
In the northern segment of the rift (north of 44.75°N), the amplitude of the Moho phase beneath the rift is largest, 0.06 ± 0.04, in the north of the segment decreasing to 0.04 ± 0.03 toward the middle segment. Amplitudes of phases lower than 0.04 in the south of the segment are so low that they are typically not significant from zero. The amplitude generally remains as low as 0.07 ± 0.04 up to 100 km from the rift axis on both flanks. Although we present the result in terms of an absolute shear velocity as an example, our only constraint is on the magnitude of relative velocity contrasts in depth.
10.1029/2018JB015771
Journal of Geophysical Research: Solid Earth
Negative Velocity Discontinuities
We image negative polarity receiver function phases (seismic velocity decrease with depth) that are less spatially persistent over the study region than the positive polarity phase we see at Moho depths ( Figure 7). Amplitudes of the negative phases across the study are typically 0.03-0.04 ± 0.02. We plot its depth for all bins with five or more waveforms and with barred hatching indicate bins that are not significant from zero.
A bimodal depth distribution of the negative phases is apparent across the study region. Depths of 90-120 ± 7 km predominate the southern segment and extend into the middle segment, and this phase includes a feature that dips to the north from~100-km depth to~115 km. Depths of 150-190 ± 7 km predominate the northern segment and also extend into the middle segment, although the phase at these depths is patchier.
Synthetic Waveforms
In the northern segment, we perform synthetic waveform modeling on two receiver function stacks related to the rift: (1) A one-dimensional depth profile from our stacked and smoothed grid at a location on the rift axis, which requires a 4% increase in shear velocity with depth at 32.0 km, followed by an increase of 4.5% at 40.0 km (Figure 3a, red). Additionally, we include an increase of 1.5% at 60.0 km and an increase of 1.5% at 70.0 km-these are not required to fit the data and are included to reach a normal mantle shear velocity of 4.44 km/s. (2) In an attempt to resolve a deeper positive phase in the northern and middle segments detected by Ps waveform modeling (Zhang et al., 2016), we stack waveforms whose ray paths pierce the deeper structure beneath the rift axis and test for earthquake back-azimuth and epicentral distance Positive amplitudes correspond to a velocity increase with depth. Each block corresponds to a 0.25°× 0.25°bin that into which receiver functions are stacked. Gray boxes within the study region represent locations that are not significant from zero based on 95% confidence limits. Inverted triangles show the location of the seismic stations: blue, Superior Province Rifting Earthscope Experiment (SPREE); green, Earthscope Transportable Array and US backbone. Diamonds show locations of synthetics in Figure 3. Cross sections, A-E, across the rift presented in Figures 5 and 6 are shown. Gray dashed lines at 44.75°N and 43.25°N separate the northern, middle, and southern segments that are used for descriptive purposes in text. Solid black lines describe the rift axis (Kucks, 1999). Thinner black lines indicate state boundaries.
10.1029/2018JB015771
Journal of Geophysical Research: Solid Earth dependencies. We find no dependency on the back-azimuth. We do observe a double positive phase in the northern segment if we stack receiver functions of this selection from earthquakes located at epicentral distances of 55-60°, possibly due to larger expected conversion transmission coefficients for smaller epicentral distances (Rychert et al., 2007). Synthetic waveform modeling of this stack requires an 8% increase in shear velocity with depth at 31.5 km, followed by an increase of 3.5% at 38.5 km and an increase of 6.5% at 60.5 km (Figure 3b, green).
Synthetic waveform modeling of the rift axis in the middle segment requires a 2.5% increase in shear velocity with depth at 32.0 km, followed by a 3% increase at 38.0 km (Figure 3c, purple). We include an increase of 1.5% at 58.0 km and an increase of 1.5% at 68.0 km-again these are not required to fit the data and are included to reach normal mantle shear velocities.
Synthetic waveform modeling of the rift axis in the southern segment requires a 3.5% increase in shear velocity with depth at 28.5 km, followed by a 7% increase at 35.5 km and a 5.5% increase at 63.0 km (Figure 3d, Figures 1, 2, and 4. The colors indicate the polarity of the seismic discontinuities from receiver functions: red, positive polarity (seismic velocity increase with depth); blue, negative polarity (seismic velocity decrease with depth). Amplitude color bars are shown below each cross section. Circles plotted at depth have a 100-km lateral spacing and correspond to circles along the lines in the reference map and Figures 1, 2, and 4. Gray boxes signify bins that have less than five hits. Bouguer anomaly is plotted above each cross section in units of mGal (Kucks, 1999), where proximal stations are also indicated as inverted triangles-blue, Superior Province Rifting Earthscope Experiment (SPREE); green, Earthscope Transportable Array and US backbone. Bottom right: reference map. Solid black lines describe the rift axis (Kucks, 1999
10.1029/2018JB015771
Journal of Geophysical Research: Solid Earth dark blue). Modeling of the flank requires a 2% increase at 28.2 km, followed by a 6.5% increase at 35.2 km and 2.5% from 39.2 km to 44.2 km. A shear velocity decrease with depth of 5.5% over 6 km centered at 103.7 km is required for the deeper negative phase in this stack (Figure 3e, cyan).
Positive Velocity Discontinuities
Sp Moho depths on the rift flanks in the northern segment generally agree with depths from Ps waveform fitting and H-κ stacking (Zhang et al., 2016; Figure 8a). Sp Moho depths on the rift flanks in the middle segment sometimes agree and are also shallower than those from Ps (Zhang et al., 2016;Figure 8b). Beneath the rift axis in the northern segment, depths of the Sp Moho phase (36-40 ± 4 km) agree with the depths from the H-κ stacking of Zhang et al. (2016). The Sp-binned grid does not resolve a deeper (up to 60 km) discontinuity found using waveform fitting of the Ps receiver functions and interpreted as the base of crustal Figures 1, 2, 4, and 7. The colors indicate the polarity of the seismic discontinuities from receiver functions: red, positive polarity (seismic velocity increase with depth); blue, negative polarity (seismic velocity decrease with depth). Amplitude color bars are shown below each cross section. Circles plotted at depth have a 100-km lateral spacing and correspond to circles along the lines in reference map and Figures 1, 2, 4, and 7. Gray boxes signify bins that have less than five hits. Bouguer anomaly is plotted above each cross section in units of mGal (Kucks, 1999), where proximal stations are also indicated as inverted triangles-blue, Superior Province Rifting Earthscope Experiment (SPREE); green, Earthscope Transportable Array and US backbone. Top right: reference map. Solid black lines describe the rift axis (Kucks, 1999
10.1029/2018JB015771
Journal of Geophysical Research: Solid Earth underplating (Zhang et al., 2016). Similarly, Sp results from the middle rift segment exhibit Moho phase depths of 35-37 ± 4 km in agreement with the H-κ stacking depths but not the base of crustal underplating at up to 60-km depths from waveform modeling (Zhang et al., 2016). However, Sp does image two distinct phases beneath the rift at 34-39 ± 4 and 62-65 ± 4 km in the southern segment similar to the interpreted Ps structure in the northern and middle segments, although the Ps study did not extend as far south as the location of the double Sp discontinuity.
The weaker Sp Moho amplitudes beneath the rift axis in the northern and middle segments (Figure 4 and cross sections A-A 0 and C-C 0 , Figure 5) suggest a gradational seismic velocity increase with depth (Figures 3a and 3c, red and purple). This is generally consistent with the weaker and less coherent Ps phases reported for SPREE stations located on the rift (Zhang et al., 2016). We suggest that the single gradational velocity increase that we image here represents the boundary between prerift crust and deeper crustal underplating. Geophysical interpretations of an active-source seismic profile also show an underplated layer down to 55 km beneath the Lake Superior portion of the rift Shay & Trehu, 1993). The Lake Superior portion of the rift is not included in this study; however, similar crustal features have been inferred along the southwestern arm of the MCR with gravity and magnetic surveys (Allen et al., 2006), and therefore, the two portions of the rift may be comparable. Magmatic underplating of the crust is able to 6 Ga), the Yavapai Province (age 1.7-1.8 Ga), and the Penokean Orogeny (age 1.8-1.9 Ga). Cyan diamond shows location of off-rift synthetics in Figure 3e. All panels: Each block corresponds to a 0.25°× 0.25°bin that receiver functions are stacked in. Barred grid points signify locations where the shown negative phase is not significant from zero according to 95% confidence limits of the amplitude. Cross sections, D and E, across the rift presented in Figure 6 are shown. Gray dashed lines at 44.75°N and 43.25°N separate the northern, middle, and southern segments. Solid black lines describe the rift axis (Kucks, 1999). State borders are marked by thinner black lines.
10.1029/2018JB015771
Journal of Geophysical Research: Solid Earth explain a deeper discontinuity (such as that found by Zhang et al., 2016, at 55-60-km depth and in our stack of Sp receiver functions from 55°to 60°epicentral distances) and is supported by high magmatic activity during rifting (Hutchinson et al., 1990;White, 1997). A possible cause for the general lack in our Sp results of a deeper phase at~60 km in the northern and middle rift segments is that lower inherent frequencies of S waves set a limit on how close two discontinuities can be to be resolved as two phases (Rychert et al., 2005(Rychert et al., , 2007. However, synthetic tests show that two discontinuities placed~30 km apart in depth, as may be expected here, produce two distinct phases (see Figures 3b and 3d, green and dark blue). More likely is that the structure of the crustal underplating is complex in these sections of the rift, possibly with strong lateral variations and steep dips, so that we do not observe a coherent phase that would define the base of crustal underplating when stacking all the data here. In particular, the steeply dipping edges of the transitional layer imaged and interpreted as a crustal underplate by Zhang et al. (2016; Figure 8) are not (Kucks, 1999) in bottom right of each plot-inverted blue triangles are the SPREE stations. Blue, Sp: circles are depths beneath each SPREE station; dashed line is depth across the whole cross section; solid line that represents depth that hints of a deeper positive phase are seen using earthquakes from 55°to 60°in the northern segment. Orange, Ps: circles are depths illuminated by waveform fitting techniques; dashed line is moving average of those; squares are depths illuminated by H-κ stacking techniques. Above each plot Bouguer gravity anomaly along the cross section (gray) and the amplitude of Sp positive phase (red). Gaps in lines are where the associated receiver functions are not significant from zero based on 95% confidence limits.
10.1029/2018JB015771
Journal of Geophysical Research: Solid Earth likely to be resolvable with Sp receiver functions, based upon modeling of wave propagation and Sp conversions through synthetic models of laterally varying velocity structure (Lekić & Fischer, 2017).
The two relatively strong positive phases beneath the rift in the southern segment suggest one distinct layer, delineated by relatively sharp discontinuities at its top and bottom, possibly describing a similar layer of crustal underplating beneath prerift crust. A possible higher volume of magma in the southern segment, as modeled with gravity data (Merino et al., 2013), may have manifested in a more laterally extensive underplated body than beneath the northern segment, allowing our Sp receiver functions to produce coherent phases that define the top and bottom of the layer of crustal underplating. Alternatively, variations in the character of the crustal underplating may be due to the different segments of the rift having formed in different geological terranes that could have affected the mobility of the melt through the crust. Topography of the boundary of the underplated layer could also have been affected by differential uplift dependent on the orientation of the compressional forces related to the Grenville Orogeny after the rifting event (Zhang et al., 2016).
Negative Velocity Discontinuities
Depths dominant in the southern segment (90-120 ± 7 km) are consistent with a midlithospheric discontinuity (MLD; Ford et al., 2010) within the North American continental lithosphere Hansen et al., 2015;Rychert & Shearer, 2009;Selway et al., 2015). The cause of this sharp velocity decrease is not well understood. Proposed causes include the following: elastically accommodated grain boundary sliding (Karato et al., 2015); an anisotropic boundary between a highly depleted chemical lid and a less depleted thermal layer underneath ; and metasomatism of lithospheric mantle rocks to produce seismically slower hydrous minerals, creating a frozen-in layer of volatile-rich melt (Hopper & Fischer, 2015;Selway et al., 2015).
Other Sp analyses in the region also finds a velocity decrease with depth at 70-110 km, interpreted as the MLD (Foster et al., 2014;Hansen et al., 2015), and a deeper phase at 200-240 km, interpreted as the lithosphere-asthenosphere boundary (LAB; Foster et al., 2014). The depth range, 150-190 ± 7 km, predominant in the north half of our study region where we image a negative phase agrees with the lithospheric thickness estimates from Rayleigh waves and receiver functions for the Superior Province (Darbyshire et al., 2007) and also that from SS precursors (Tharimena et al., 2017). The deeper phases in our 150-190 ± 7-km depth range may agree with the LAB of Foster et al. (2014), although these phases are few and the coverage is patchy (Figure 7). The Sp receiver function analysis of the US using Earthscope Transportable Array, SPREE, and other permanent stations also observes very few phases in this depth range (Hopper & Fischer, 2018). The negative phase that we observe at a depth of around 150 ± 7 km is much shallower than LAB depths observed in other Sp studies (Foster et al., 2014), determined by depth constraints of azimuthal anisotropy across the continent , and estimated by teleseismic P wave tomography with the MCR on the edge of the study (Frederiksen et al., 2013).
The depth variability of the negative velocity discontinuities illuminated by Sp receiver functions here does not correspond spatially to the MCR. Instead, it seems more related to the Spirit Lake Tectonic Zone, which separates the Yavapai Province (1.8-1.7 Ga) from the Minnesota River Valley Subprovince (3.6-3.4 Ga) and the Penokean Orogeny (1.9-1.8 Ga; Holm et al., 2007;Shen et al., 2013; Figure 7). Particularly in the center of the middle segment, the overlap of the two negative phases coincides with the Spirit Lake Tectonic Zone (cross section E-E 0 ; Figure 6). The main negative phase at 90-120 ± 7-km depth in the southern and middle segments is also north dipping. The dipping feature may be related to the relict subduction zone from the accretion of the Yavapai Province to the Superior Province~1.7 Ga (Hopper & Fischer, 2015;Thurner et al., 2015). These are therefore likely features that predate the MCR and furthermore do not seem to exhibit riftrelated modification in what we image. Using Earthscope long period magnetotelluric data, Yang et al. (2015) also find no clear electrical resistivity anomalies associated with the MCR in the mantle. Lack of rift-related alteration in our negative discontinuities may be explained by lithospheric healing or compression since the rifting event. Alternatively, rifting may have been accommodated by a narrow magmatic plumbing system that we are unable to resolve in our study. Vervoort et al., 2007;White, 1997). This suggests an active rifting environment for the MCR, with a plume possibly having been centered beneath Lake Superior with a radius of up to 600 km . The present East African Rift is considered to have formed in an active rifting regime (Rychert et al., 2012) and has features that are similar to the MCR. As well as its arm structure, the East African Rift shows crustal thinning beneath its extending arms (Simiyu & Keller, 1997), similar to what the MCR is believed to have experienced during rifting prior to its crustal rethickening. Furthermore, in a passive rifting regime, greater degrees of lithospheric thinning are expected (Huismans & Beaumont, 2011), even over the short lifetime of the MCR of~20 Myr (Vervoort et al., 2007). Instead, we image negative velocity discontinuities in the mantle lithosphere that likely predate the MCR with a lack of observed perturbations spatially related to the rift, particularly in the north-dipping negative discontinuity (cross section E-E 0 ; Figure 6) that is likely related to Yavapai accretion (Hopper & Fischer, 2015;Thurner et al., 2015). Additionally, our constraints on an altered Moho and addition of crustal underplating over a relatively focused, small lateral area beneath the rift axis supports focused magmatism at~60-km depth, which is more consistent with an active component of upwelling.
Conclusions
Here we use S-to-P receiver functions from SPREE, the Earthscope Transportable Array, and the US backbone network to constrain Moho and lithospheric discontinuity structure beneath the southwestern arm of the 1.1-Ga MCR and its flanks. We illuminate a relatively flat positive seismic velocity increase with depth over the region at depths of 33-40 ± 4 km associated with the Moho, with little or no rift-flank depth variation. Beneath the rift in the northern half of the study region, receiver function amplitudes are generally weak for the phase at Moho depths, indicating a more gradual velocity increase with depth, or a Moho with considerable relief. A deeper positive discontinuity at 61 ± 4 km beneath the rift in the north is present in a subset of the data, possibly because it is weaker, more steeply undulating, and/or not laterally coherent over a wide area. A second velocity increase at 62-65 ± 4 km also exists beneath the rift in the southern half of the study region, coupled with a generally sharper velocity discontinuity at Moho depths than in the north. The ability to resolve a layer of crustal underplating in the south suggests that it is more laterally extensive than in the north, which could be due to more magmatism during rifting in the south, or that more melt mobility was permitted in the north. We also image two velocity decreases with depth in the mantle lithosphere. In the southern half of the study region, a north-dipping negative discontinuity exists at depths of 90-120 ± 7 km, which we attribute to old provincial suturing that predates the MCR and is consistent with depths of a MLD revealed in other studies. The northern half of the study region hosts a less spatially consistent negative discontinuity at a depth range of 150-190 ± 7 km that might be the LAB. We do not observe signs of lithospheric mantle alteration related to the MCR, suggesting that the lithosphere has since healed, been compressed post-rift, or that we are unable to resolve potentially narrow signatures of past upwelling mantle material. | 8,003.8 | 2018-09-01T00:00:00.000 | [
"Geology"
] |
Ultra-High and Near-Zero Refractive Indices of Magnetron Sputtered Thin-Film Metamaterials Based on Ti x O y
Metamaterials based onTi x O y with ultra-high andnear-zero refractive indiceswere obtained byDCmagnetron sputtering.Thedata on refractive indices, extinction coefficients, film thickness, and band gaps, obtained by spectroscopic ellipsometry, showed very high potential of these materials as metamaterials. Phase analysis performed by XRD revealed the presence of titanium phases with lower titanium oxidation states resulting from high concentration of oxygen vacancies, which are crucial for such extraordinary jumps and drops of refractive indices. Numerous band gaps for direct and indirect electron transitions additionally confirmed unique properties of these materials.
Introduction
Thin titanium oxide films can be prepared by various routes, but direct current magnetron (DC) reactive sputtering is one of the most promising methods, because it allows control of the deposited film stoichiometry.High density, excellent adhesion, high hardness, and good thickness uniformity in large-area thin films can be also achieved [1][2][3].A metal target can be used and a high deposition rate is achievable as well.This technique is widely applied to deposit hard, wearresistant, low friction, corrosion resistant, and decorative coatings, as well as coatings with specific optical or electrical properties [4][5][6].It is especially useful for deposition of high-quality titanium dioxide (TiO 2 ) films at relatively low temperature [7,8].Additionally, as-prepared TiO 2 films show excellent electrical and optical properties such as high refractive index, dielectric constant, and visible light transmission, as well as strong mechanical and chemical stability, and good insulating properties [9,10].High refractive indices result from special phase composition and structure of the films deposited by this method, while the higher optical band gap values can be attributed to the lattice distortion.This high refractive index in the visible and the near-infrared region makes them suitable for electrooptical devices, sensors, and optical coatings and as photocatalysts for oxidation of organic compounds [3,11].The thin Ti O films obtained by DC magnetron sputtering, presented in this paper, showed extraordinary optical properties (ultra-high and near-zero refractive index), which classify them as metamaterials.The obtained values are two times higher or several times lower (depending on the wavelength) than the typical ones for this system found in the literature.Physical properties of these materials make them suitable for a wide range of various applications such as in microstrip technology, beam self-collimation, strong field enhancement, optical links in lumped nanophotonic circuits, biomolecular sensors, fabrication of adaptive selective lenses, tunable mirrors, isolators, converters, and optical polarizers [12][13][14].
Materials and Methods
Ti O thin films were deposited on glass slides (2.5 × 2.5 mm) using a JCK-500A DC magnetron sputtering system.Previously, the substrate surface was prepared by etching of SiO 2 layer in an ECR plasma etcher using CF 4 as an etching gas at ambient temperature.The etching conditions were 0.02 Pa gas pressure, 400 W microwave power, 500 V accelerating electrode voltage, and 200 mm gap distance between the electrodes.Prior to the introduction of the sputtering gas, vacuum in the chamber was adjusted to 4 × 10 −4 mbar.Sputtering pressure was kept at 6 × 10 −1 mbar.The flow rates of Ar (99.999%) and O 2 (99.999%) were kept constant at 44.9 and 10 sccm (standard cubic centimeter per minute), respectively.The discharges were generated at a constant power of 300 W. The substrate temperature was maintained at 200 ∘ C and the deposition rate at about 4 nm/min.Thin films of samples 1 and 2 were deposited for 45 minutes and those of sample 3 for 7 minutes.Samples 2 and 3 were additionally thermally treated at 400 ∘ C.
Phase composition of the samples (their appearance is shown in Appendix A) was analyzed by XRD (Philips PW 1051 Powder Diffractometer using Ni-filtered Cu K radiation) and FTIR (Nicolet IS 50 FTIR Spectrometer).The morphology of thin films was investigated by SEM (SEM JEOL 5300).EDS (energy dispersive analysis) measurements were performed in order to determine the ratio between titanium and oxygen.
Spectroscopic ellipsometry was used for an additional characterization of Ti O thin films deposited onto glass substrate [13].For these measurements a HORIBA Jobin Yvon UVISEL iHR320 was used with the monochromator wavelength ranging from 200 to 2200 nm (0.6 up to 4.8 eV).Refractive index and extinction coefficient were determined as a function of wavelength.The incident angle was set at 70.1 ∘ and the light spot was 1 mm in diameter.All optical spectra were recorded ex situ, with an energy step of 0.1 eV.Further analysis of the obtained optical spectra was performed by fitting the experimental data using a commercial software package DeltaPsi 2 (see Appendix B).
Phase Analysis.
As shown in Figure 1, the sample without additional thermal treatment 1 shows planes characteristic of TiO: (011), (031), and (−211) at 23.35, 37.24, and 37.43 ∘ , respectively.Also, anatase can be identified by a small peak corresponding to (101) plane observed at 24.87 ∘ and a peak at 37.82 ∘ corresponding to the (004) plane.For sample 2, (101), (004), and (113) planes at 25.36, 37.82, and 38.67 ∘ , respectively, can be attributed to anatase, while the (101) plane at 35.99 ∘ corresponds to rutile and the (−211) plane at 37.14 ∘ to TiO.Finally, sample 3 shows planes characteristic of anatase: (101), (004), and (112) at 25.27, 37.72, and 38.58 ∘ , respectively, as well as of rutile: (101) and (200) at 36.08 and 39.15 ∘ , respectively.Minor rutile phase is noticed in sample 2 while in sample 3 it is the major phase, confirming that the process of recrystallization inside of the system took place during additional thermal treatment of samples.As it is obvious from Figure 1, shapes and maxima of the broad peaks at the diffraction angle 2 between 15 and 30 ∘ are more pronounced for samples 1, 2, and 3 than for amorphous SiO 2 substrate, indicating that the amorphous phase of Ti O is the prevailing titanium phase.The crystallite sizes, calculated using the Scherer equation, are in the range 7-25 nm.
For a better insight into the structure of amorphous Ti O , the first coordination sphere radius was calculated using the well-known formula 2 sin = 1.23,where is the scattering angle and is the X-ray diffraction wavelength [14].The obtained value of 0.224 nm was used for determination of the bond length and comparison with the theoretical bond length of anatase [15,16].From these data, the chemical formula of the amorphous Ti O phase was found to be TiO 1.8 .This indicates that the amorphous phase consists of ordered clusters with at least five interconnected subclusters of TiO 1.8 .
The quantitative phase composition of Ti O thin films, determined by spectroscopic ellipsometry and using DeltaPsi 2 software package (see Appendix B), is shown in Table 1.The obtained data are in fair agreement with the data obtained by XRD.
EDS measurements (typical spot is shown in Appendix C) confirmed findings of XRD and spectroscopic ellipsometry.It was found that the ratio between titanium and oxygen is from 1 : 1.75 to 1 : 1.9.
FTIR Analysis.
FTIR spectra of Ti O films obtained by magnetron sputtering are shown in Figure 2. The band at 835 cm −1 (sample 1) corresponds to the O-O stretching vibration inside TiOOH.The bands between 660 and 600 cm −1 for samples 1-3 tail up to almost 900 cm −1 and can be assigned to Ti-O stretching vibrations of Ti atoms in octahedral (near 600 cm −1 ) and tetrahedral (near 800-900 cm −1 ) surroundings or to stretching vibrations of Ti-O-Ti bonds in polytitanates.The peaks at 665-657 cm −1 can be assigned also to surface phonon splitting of the vibration corresponding to small particle size anatase phase.The peaks at 520-579 cm −1 and broad and strongly pronounced peaks at 745-735 cm −1 belong to Ti-O and Ti-O-Ti stretching vibrations of anatase (sample 3).The pronounced splitting of the band at about 745-731 cm −1 may be induced by the various oxidative states of Ti in TiO 2 and TiO.The bands at 523-517 (band at 517 cm −1 corresponds to rutile phase) are connected with the absorption of Ti-O bonds present in stoichiometric titanium dioxide.Ratio between surfaces of bands at 523-502 cm −1 and bands at 765-751 cm −1 shows ration between Ti-O and T=O, oxides and suboxides.Besides, all peaks between 506 and 565 cm −1 correspond to octahedral surroundings of the titanium atoms in the coatings, while the bands at 492-468 cm −1 may correspond to the TP phonon frequency of TiO 2 in the rutile phase, and the bands at 468-453 cm −1 can be assigned to vibrations of TiO 6 octahedron, related to Ti-O stretching bonds directed either to the interlayer space or to the outer surface formed nanotube, while 434-406 cm −1 bands (samples 2 and 3) correspond to the TP phonon frequency of nanocrystalline TiO 2 phase anatase or rutile phase.
Film Morphology.
At a magnification of 200,000x, SEM images show almost flat surface with barely visible surface patterns (Figure 3).This indicates the epitaxial growth of titanium oxide film formed layer by layer on the substrate Oxygen vacancies, uniformly distributed, are the main defects in the as-grown films.If present in higher concentrations they can coalesce to form pores, but the pores have not been noticed even at very high magnifications and thus the porosity of the films can be neglected.
Band Gap Energy.
The energy gaps for direct and indirect electron transitions from the valence to the conduction band were determined using the Tauc equation: where is the absorption coefficient (derived from the extinction coefficient data: = 4/), is a constant, and = 1 for direct and = 4 for indirect transition [17].The energy gap values, obtained by extrapolation of the linear part of plots (ℎ]) 2/ versus ℎ] (see Appendix D), are given in Table 2.
Refractive Indices.
It can be noticed in Figure 4 that there are differences in refractive index values ranging from ultrahigh to near-zero ones, which is of great importance for the potential application of these materials.The highest values of refractive indices are 5.1 and 5.0 at 399.9 and 688.8 nm for sample 2 and 4.2 at 590.4 nm for sample 1, while the lowest, near-zero ones are 0.28 at 590.4 nm for sample 2 and 0.41 at 1377.6 nm for sample 1.These values indicate that at certain wavelengths this material shows the properties typical of metamaterials.Opposite to samples 1 and 2, sample 3 shows refractive indices in the range expected for anatase/rutile due to lower concentration of oxygen vacancies, that is, higher oxidation states of Ti.The extinction coefficient, as a measure of attenuation during the electromagnetic wave propagation through the material, shows slightly different behavior than refractive indices, with shifted maximums and minimums towards lower wavelength values.
Discussion
On the basis of phase analysis it is obvious that titanium is present in different oxidative states: Ti 2+ (TiO), Ti 3+ (Ti 2 O 3 ), mixed Ti 4+ and Ti 3+ in ratio 1 : 2 (Ti 3 O 5 ), and Ti 4+ (anatase, rutile).In all Ti O , Ti is in octahedral coordination, surrounded by 6O atoms, while each O atom is surrounded by 3Ti atoms (except of TiO).The site symmetry of Ti and O atoms in all of them differs [18].
In the rutile and anatase the site symmetry of Ti is D 2h .The elongation of O-Ti-O bonds along the direction of the coordination polyhedron and deviation of the angles from 90 ∘ lead to a decrease in symmetry from octahedral to tetragonal D 2h symmetry.In Ti 2 O 3 the site symmetry of Ti is C 3i , which leads to the distortion of the coordination polyhedra from octahedral to trigonal symmetry.In Ti 3 O 5 the symmetry of both Ti and O atoms is C s [18].
For each of these phases, the distortion of molecular orbitals occurs due to different symmetry sites of Ti and O atoms.This influences the splitting of Ti d orbitals into triply degenerated t 2g band consisting of , , and orbitals (higher energy state) and doubly degenerated e g band consisting of 2 − 2 and 2 orbitals (lower energy state).e g group has the proper symmetry to interact with the oxygen orbitals, forming sigma bonding () and antibonding ( * ) molecular orbitals (MO): 3e g ( * ) in interaction with O 2p orbitals and 2e g () and 1e g () in interaction with O 2s orbitals.t 2g orbitals form pi bonding () and antibonding ( * ) orbitals: 2t 2g ( * ) and 1t 2g () in interaction with O 2p orbitals.Besides, in the interaction of Ti 4s and O 2s orbitals, 2a 1g () and 1a 1g () MO are formed and 3a 1g ( * ) MO is formed in interaction of Ti 4s and O 2p orbitals.In the interaction of Ti 4p and O 2s orbitals, 2t 1u () and 1t 1u () MO are formed, while in the interaction of Ti 4p and O 2p orbitals 4t 1u (, * ) and nonbonding t 1g ( 0 ) and t 2u ( 0 ) MO orbitals are formed.All antibonding orbitals are empty except 2t 2g ( * ) on the bottom of conduction band, which are partially filled in the case of Ti 2 O 3 and Ti 3 O 5 (having 1 electron) and TiO (having 2 electrons).O 2p states with nonbonding orbitals (t 1g ( 0 ) and t 2u ( 0 )) at the highest energy level occupy the top of the valence band [18][19][20].
As a result of influence of distorted octahedral crystal fields in various Ti O , splitting and rearrangement of energy sublevels in e g and t 2g groups of orbitals in the conduction band occur.For TiO, Ti 2 O 3 , and Ti 3 O 5 , high concentration of oxygen vacancies results in an additional splitting of t 2g orbitals in valence band (near the Fermi level).Therefore, the lowest values of band gaps (0.85 eV to 1.97 eV) probably belong to the transition from the highest occupied molecular orbitals (HOMO), mainly O 2p orbitals, corresponding to the valence band to the lowest unoccupied Ti 3d orbitals corresponding to the conduction band [21].These low values of band gaps show that these phases exhibit metal (TiO) and semiconductive (Ti 2 O 3 and Ti 3 O 5 ) character and consequently their metamaterials behavior.The band gap values 2.61 and 2.68 probably belong to phases with Ti oxidative states between Ti 3+ and Ti 4+ .For oxidative state Ti 4+ , it exhibits insulator characteristics (3.00-4.39eV) [22].
Ultra-high and near-zero refractive indices of Ti O samples 1 and 2 (characteristic for metamaterials) are the result of the previously explained electronic structure of metallic and semiconductive phases present in Ti O films.Such high defect structure induces specific interactions between light electromagnetic waves (LEW) and plasmonic waves of free electrons in the conduction band and weakly bound electrons in valence band coupled with the Ti O crystalline phases embedded in the matrix of amorphous phase.In a simplified model, Ti O nanoparticles can be considered as a network of positive nuclei surrounded by freely oscillating electrons.In interaction with LEW, these electrons accumulate on one side of the nanoparticle surface, while the other remains positive.As-formed nanodipoles have the direction opposite to the electric field of light [23][24][25].They then radiate their own waves, which combine with the incoming waves.The phase between the incoming and the induced waves determines whether the refractive index will be extremely high or close to zero.If the incoming wave oscillates relatively slowly, the electrons can follow the wave: they are in phase, because their oscillation frequency around some equilibrium position is in resonance with the incoming light wave and the refractive index is extremely high (4.2 for sample 1 and 5.1 for sample 2).In the case when the incoming wave frequency exceeds that of the free electrons or particularly electrons coupled with lattices of nanocrystallites and subnanocrystallites, the electrons can no longer follow the wave, and they get out of phase.Then, the waves radiated by the electrons are out of phase too, and the refractive index is close to zero (0.41 for sample 1 and 0.28 for sample 2) [26][27][28][29][30].
Considering high number of various direct and indirect electron transitions between the valence and the conduction band and inside the conduction band of Ti and O in samples 1 and 2 it is obvious that during the interaction of the samples with LEW an overlapping of the conduction and valence zones occurs, forming consequently a continuous range of energy states of electrons.Some of them will be in resonance with incident electromagnetic waves at lower wavelengths (showing extremely high refractive index), while the others will be in the opposite phase at higher light wavelengths (showing close to zero values of the refractive index).
Higher values of extinction coefficients are mainly caused by intraband transitions.They can also cause a significant change in phase angle of the plasmonic wave during the interference with incident LEW.Extinction coefficient is negative in the wavelength range of 619-1033 nm for sample 1 and 688-1240 nm for sample 2. It is influenced by transfer of electrons from ground state to metastable excited state due to the energy of xenon light source during the ellipsometry measurements.Estimated amount of the energy added to the samples from the source, in area where the extinction coefficient is negative, is 1.1-1.9eV for sample 1 and 0.9-1.7 eV for sample 2. This energy induces inverse occupation, that is, transfer of electrons to excited metastable state.Characteristics of the xenon source show that its highest intensity of radiation is in the range of 750-1000 nm.Due to this, the number of electrons in excited metastable state is maximal in that range.
Conclusions
Ti O thin films, thickness from 29 to 176 nm, were obtained by DC magnetron sputtering.The films were mainly constituted of an amorphous phase and crystal phases of anatase, rutile, and TiO.Phase composition of the films proved by XRD and ellipsometry indicated high concentration of oxygen vacancies, which caused ultra-high and close to zero values of refractive indices.The highest value of the refractive index was 5.1 and the lowest 0.28.The band gaps for direct and indirect transitions between the valence and the conduction band were in a wide range from 0.85 to 4.39 eV.The highest refractive index corresponds to the resonance of plasmonic waves of free electrons and incident electromagnetic waves, while the near-zero refractive index corresponds to the interference of plasmonic waves of electrons coupled with lattices of nanocrystals and subnanocrystals and incident electromagnetic waves.
A. Appearance of the Films
Ti O films of various thicknesses show different colors, depending on the film thickness and concentration of oxygen vacancies (Figure 5).where Ψ is amplitude change of reflected wave, while Δ is a phase change of the same reflected light beam.Software package DeltaPsi 2 gives an explicit form depending on the refractive index and extinction coefficient of the wavelength.Fitting the model to the experimental curve was estimated using the Levenberg-Marquardt algorithm, with the analytical parameters which are simultaneously determined through the iterative process change until the chi-squared function (B.2) obtains the minimum value: In the above expression, is the total number of recorded points (Ψ eksp , Δ eksp ), while (Ψ teor , Δ teor ) are experimental values of ellipsometric angles and corresponding values calculated from the model using a parameter Γ Δ, for calculation of experimental error.All calculations were performed using DeltaPsi 2 software package, which is an integral part of the device.
Data interpretation is very difficult from the absolute values of ellipsometric (Ψ, Δ) angles and it is necessary to create an appropriate optical model.In the case of thin layers of TiO 2 fitting of the experimental data was carried out using a twolayer model: one homogeneous layer of TiO 2 and a surface layer of several nanometers, with the composition of 50% TiO 2 + 50% voids, which describes the roughness of the sample.
To describe the properties of thin films of TiO 2 new amorphous theoretical model was used, which is a modified Tauc-Lorentz model.This model is generally used to describe the dielectric that partially absorbs light (which absorbs light
Figure 1 :
Figure 1: XRD patterns for Ti O films obtained by DC magnetron sputtering (A: anatase, R: rutile, s: substrate, 1: sample deposited for 45 min and not further thermally treated, 2: sample deposited for 45 min and thermally treated at 400 ∘ C, and 3: sample deposited for 7 min and thermally treated at 400 ∘ C).
Figure 2 :
Figure 2: FTIR spectra of Ti O films obtained by DC magnetron sputtering (1: sample deposited during 45 min without additional thermal treatment, 2: sample 1 thermally treated at 400 ∘ C, and 3: sample deposited during 7 min and thermally treated at 400 ∘ C).
Table 1 :
Phase composition of the deposited thin Ti O films.
Table 2 :
The band gap energies of samples and the corresponding wavelengths.
surface.Slight surface roughness observed in SEM micrographs probably results from partially shifted neighbor layers induced by the similar roughness of the substrate.The epitaxial growth of the prevailing amorphous Ti O phase is supported by the amorphous silica phase of substrate because for epitaxial growth of high-quality films, it is desirable to have neglected lattice mismatch between film phases and the substrate. | 4,961.6 | 2016-04-18T00:00:00.000 | [
"Physics",
"Materials Science"
] |
Do Female Enrolment Rates Cause Economic Growth in Pakistan ?
There has been much discussion on the relationship between education and economic growth. A few studies have examined the increasing trend of female enrolment in educational institutions and economic growth. The objective of this paper is to empirically investigate four alternatives but equally plausible hypotheses. These are: i) GDP cause female enrolment proxies (the conventional view), ii) Female enrolment proxies cause GDP, iii) There is a bi-directional causality between the two variables and iv) Both variables are causality independent. In order to find the relationship between the two variables set, a time series Co-integration and Granger Causality Tests have been employed separately. Secondary data pertaining to Pakistan from 1966 2008 has been used for analysis. The empirical results moderately support the conventional view that GDP has significant long-run casual effect on the female enrolment proxies in Pakistan. The present study supports the unidirectional causality relationship between the GDP and female enrolment in the specific context of Pakistan.
Introduction
The Millennium Development Goals that emerged from the UN Millennium Declaration of September 2000 are specific measurable targets, including the one for reducing the extreme poverty that still grips more than 1 billion of the world's people by 2015.Central to this promise are the MDGs related to educational outcomes: (1) Ensure that all children complete primary education by 2015.(2) Eliminate gender disparities in primary and secondary education.By 2006, most countries have already fallen well behind the necessary targets to meet these goals (Millennium Development Goal, 2006).
Human capital is considered as an important determinant of economic growth which is effective vehicle for reducing income inequality and absolute poverty (World Bank, 2008).According to Barnet (1990), human capital investment in the form of higher education is recognized as capital investment, while, Ozsoy (2008) says that human capital is an engine of development for the new world economy.According to Abbas (undated), education was considered as a tool for human development in the past but now it is considered as a tool of development in broad meanings such as economic, social and also human resource development.
The number of studies has been examined the role of human capital as an important determinant of economic growth (e.g., Romer, 1986;Lucas, 1988).This is supported by number of empirical studies that human capital (such as years of schooling, school enrolment rates, or literacy rates) have statistically significant and positive effects on economic growth (e.g., Romer, 1990;Barro, 1991;Barro and Sala-i-Martin, 1992;Mankiw, Romer and Weil, 1992;Barro and Lee, 1994).Few studies considered, variables reflecting health status (such as life expectancy) are also significant in cross-country growth regressions (e.g., Wheeler, 1980;Barro and Lee, 1994;Knowles andOwen, 1995, 1997).
Most of the cross-country empirical literature on the effect of human capital on growth is "gender-neutral"'.It usually focuses on levels of education (or health) averaged over the whole (working-age) population.However, female and male education affects growth in quite different ways.Female education along with male education can improve productivity directly when better-educated females participate in the paid workforce and contribute to conventionally measured output.However, female participation rates are generally lower than for males and vary widely across countries.Conventional measures of output, which ignore women's role in non-market home production activities, have long been recognized as understating women's economic contribution relative to men.There is a growing literature, especially in the context of developing countries, female education produces social gains by reducing fertility, infant and child mortality, improving family and child health, increasing life expectancy, and raising the educational attainment of children (e.g., Schultz, 1988;Behrman and Deolalikar, 1988;Bellew et al, 1992;Subbarao and Raney, 1995).
Hence, even if female participation rates are lower than for males, the effects of improved female education on general levels of education, health status and fertility can boost measured productivity growth indirectly.Schultz (1995) concludes that female school enrolment has an apparently greater positive impact on economic growth than male school enrolment.However, he does not control for the influence of any of the other variables generally accepted as affecting growth.Hill and King (1993) suggest that both the level of female education and the gap between the levels of male and female education are significant determinants of economic growth.They imply that failure to improve female education to the same (or higher) average level as that of males acts as a brake on development.This contrasts with the results of one of the most influential recent empirical growth studies (Barro and Lee, 1994).Barro and Lee (hereafter BL) argued that, whereas growth is positively related to male schooling, it is negatively related to female schooling.Stokey (1994) gives an explanation of this puzzling result in response to BL's paper.She argues that the female education variable acts as a dummy variable for geographic regions or ethnic groups that educate women differently from men, especially the (fast-growing) East Asian ``Tigers'' (Hong Kong, Korea, Singapore and Taiwan).She suggests that the female education variable should be dropped from the BL growth equations and that, given the high correlation between the female and male education variables.This is itself of interest given that, some studies have started questioning the role and significance of educational attainment variables in growth equations, e.g., Pritchett (1996) and Bils and Klenow (1998).Also, Knowles andOwen (1995, 1997) argue that education is not statistically significant in a range of models that include life expectancy and base-period output per worker.
In a Pakistan's perspective, United Nations, State Bank of Pakistan Reports and Economic Survey of Pakistan has given the light on Pakistan's' educational scenario.According to Human Development report (2005), "The rank of Pakistan is 135 th among 177 countries indicating low life expectancy at birth, low educational attainment and low income.The report also indicates the adult literacy rate of age (15 years and above) as 35.2% of female as compared to 61.7% of male.In the same report, the Gender related development index (GDI) rank of Pakistan is 107 th among 177 countries.This explains as how the Human Development gap has been further aggravated by substantial gender disparities".
In annual report of State Bank of Pakistan (2004-05) discussed the current situation of education in Pakistan.The report states that, "Unfortunately, Pakistan's track record in literacy and education has not been satisfactory.The education system in the country is characterized by highly illiteracy rate, low gross and net enrolment at all level of education, high dropout rates from schools, a wide disparity at gender and regional level, and a poor quality of education" (Annual report of State Bank of Pakistan, 2004-05).
In this connection, Economic Survey of Pakistan, 2009-10 states that, "The overall literacy rate (age 10 years and above) is 57% (69% for male and 45% for female) compared to 56% (69% for male and 44% for female) for 2007-08.The data shows that literacy remains higher in urban areas (74%) than in rural areas (48%), and is more prevalent for men (69%) compared to women (45%).However, it is evident from the data that overall female literacy is rising over time, but progress is uneven across the provinces.Similarly, the overall school attendance, as measured by the Net Enrolment Rate (NER), for 2008-09 was 57% as compared to 55% in 2007-08.Nationally, the Gross Enrolment Rate (GER), sometimes referred to as the participation rate, which is the number of children attending primary school (irrespective of age) divided by the number of children who ought to be attending, in case of both male and female saw no change and remained at 91% between 2007-08 and 2008-09".
The above discussion shows the strong connection between female education and economic growth.In this paper an analysis has been carried out to find a statistical relationship between female enrolment proxies and economic growth in Pakistan using secondary data from 1966 to 2009.The objectives of this paper are to empirically investigate: Whether the statistical relationship between the female enrolment and the economic growth in Pakistan is unidirectional (female education affect/cause economic growth or growth affect / cause female education ); Whether the statistical relationship between the female enrolment and the economic growth in Pakistan is bi-directional (female education affect/cause growth and growth affect / cause female education ); The two variables (female education and growth) do not influence each other (causality independent).
The paper is organized as follows: after introduction which is provided in Section 1 above, data sources and methodological framework is explained in Section 2. The estimation and interpretation of results is mentioned in Section 3. Section 4 concludes the paper.
Data Source and Methodological Frame Work
The study uses annual observations for the period of 1966-2008.The data is obtained from various issues of Economic Survey of Pakistan, International Financial Statistics, and World Bank Development Indicators data base (2009).To examine the impact of female enrolment in educational institutes on economic growth, the present study used seven different proxies i.e., female primary school enrolment (FPSE), female middle school enrolment (FMSE), female high school enrolment (FHSE), female secondary vocational enrolment (FSVE), female arts & science colleges enrolment (FASCE), female professional colleges enrolment (FPCE) and female universities level enrolment (FULE)] as a dependent variables separately regress on economic growth (GDP) which covering the period of 1966-2008.We have estimated a simple non-linear growth-enrolment model which has been specified as follows: There exists a linear combination between these two series that is stationary at levels i.e., ). 0 Thus, the first step for Cointegration is to test whether each of the series is stationary or not.If they both are stationary say at first difference i.e. they are I(1), then we may go to the second step to verify the long run relationship between them.
Augmented Dickey Fuller (ADF) test is usually applied to test stationarity.It tests the null hypothesis that a series ( t Y ) is non-stationary by calculating a t-statistics for 0 in the following equation:
Error Correction Model (ECM)
If time series are I(1), then regressions could be run in their first differences.However, by taking first differences, the long-run relationship will be lost that is stored in the data.This implies to use variables in levels as well.Advantage of the Error Correction Model (ECM) is that it incorporates variables both in their levels and first differences.By doing this, ECM captures the short-run disequilibrium situations as well as the long-run equilibrium adjustments between variables.ECM term having negative sign and value between "0 and 1" indicates convergence of model towards long run equilibrium and shows how much percentage adjustment takes place every year.
Granger Causality Test
If a pair of series is cointegrated then there must be Granger-causality in at least one direction, which reflects the direction of influence between series.Theoretically, if the current or lagged terms of a time series variables, say t X , determine another time-series variable, say t Y , then there exists a Granger-causality relationship between t X and t Y , in which t Y is granger caused by t X .
Estimation and Interpretation of Results
Economic time-series data are often found to be non-stationary, containing a unit root.Ordinary Least Squares (OLS) estimates are efficient if variables included in the model are stationary of the same order.Therefore, first we need to check the stationarity of all variables i.e.FPSE, FMSE, FHSE, FSVE, FASCE, FPCE, FULE and GDP used in the study.For this purpose we apply Augmented Dickey-Fuller (ADF) test.Table 1 gives the results of ADF tests.
Based on the ADF tests, all variables appear to be non-stationary at levels but stationary at first difference.Thus, we may conclude that these variables are integrated of order one i.e.I (1).The Figure below shows the plots of female enrolment proxies and GDP in first difference forms, which sets the analytical framework as regarding the long-term relationship of enrolment and nominal growth.
Figure 1 shows the plots of female enrolment proxies and GDP in first difference forms, which sets the analytical framework as regarding the long-term relationship of enrolment and nominal growth.The cointegration test between female enrolment variables and GDP are carried out separately as mentioned below:
Cointegration Test for FPSE and GDP
Cointegration test for the first female enrolment variable (FPSE) and economic growth (GDP) would help us to clarify if relationship between these two variables exists.Results of regression and ADF test for the residual are presented in Table 2 and Table 3 respectively.The finding reveals that the residual is stationary at level i.e., it is integrated of order zero.This authenticates our intention that FPSE and GDP are indeed cointegrated that is a long run relationship between them holds.In order to ensure stability of long run relationship between FPSE and GDP, an Error Correction Model (ECM) has been used.The results are presented in Table 4.
The finding reveals that the short run effect and the long run adjustment impact of FPSE and GDP is significant at 5 % level.The adjustment parameter (p) appears with negative value signifying the long run convergence.The ECM estimation reveals that 14.4% of the disequilibrium in FPSE produced by GDP would be adjusted in every year.The conclusion is that there is a stable long run relationship between FPSE and GDP.
To confirm the causal relationship between the FPSE and GDP, a Granger-Causality test has been applied using lag length up to four periods.The following four hypotheses are tested.
The results are provided in Table 5.It shows the hypothesis that "FPSE does not Granger cause GDP" is rejected.This, of course, accords with the conventional hypothesis 1.But in the same table the null hypothesis that "GDP variables do not Granger causes FPSE" is accepted even at four lags.It validates the hypothesis 2. These results, taken together, does not support hypothesis 3 and 4.So a unidirectional relationship between the FPSE and GDP is established.This finding additionally implies that any investigation of the impact of rural poverty on commercial energy consumption should be performed within a simultaneous equation model.
Cointegration Test for FMSE, FHSE, FSVE, FASCE, FPCE, FULE and GDP
The cointegration test for female enrolment variables i.e., FMSE, FHSE, FSVE, FASCE, FPCE and FULE is carried out separately over GDP.Results of regression and ADF test for the residual are presented in Table 6 and Table 7 respectively.The finding reveals that the residual is stationary at level i.e. it is integrated of order zero.This validates the proposition that FMSE, FHSE, FSVE, FASCE, FPCE, FULE and GDP are indeed cointegrated i.e. a long run relationship between them holds.In order to check steadiness of long run relationship between FMSE, FHSE, FSVE, FASCE, FPCE, FULE and GDP, an Error Correction Model (ECM) is applied.
The results are presented in Table 8.
The short run effect of GDP on FMSE, FSVE and FPCE is insignificant, while the long run adjustment impact of GDP on FPCE is insignificant.The remaining proxies for short-run effect on GDP is significant i.e., FHSE, FASCE and FULE.While the long-run adjustment impact of GDP on FMSE, FHSE, FSVE, FASCE and FULE is significant with negative value indicating the long run convergence i.e., 3.7%, 4.5%, 11.8%, 26.3% and 19.2% respectively would be adjusted in every year.The conclusion is that there is a stable long run relationship between them.
To confirm the causal relationship between the FMSE-GDP, FHSE-GDP, FSVE-GDP, FASCE-GDP, FPCE-GDP, FULE-GDP, a Granger-Causality test has been applied using lag length up to four periods.The following four hypotheses are tested.The results are provided in Table 9.It shows that the hypothesis that "FMSE does not Granger cause GDP" is rejected.This does accord with the conventional hypothesis 1.But the null hypothesis that "GDP does not Granger cause to FMSE" is accepted.These results, taken together, does not support hypothesis 3 and 4 and suggest that there is a unidirectional relationship between the FMSE and GDP.The other results shows that FHSE, FSVE, FPCE to GDP and GDP to FHSE, FSVE, FPCE both are independent in nature and support our 4 th hypothesis.The final outcome shows that GDP Granger cause FASCE and FULE in one direction, therefore, there has a uni-directional relationship exist.However, FASCE and FULE both do not Granger cause their alternative hypothesis.This supports our 2 nd hypothesis.
Conclusion
The causality approach was used to analyze the relationship between different female enrolment proxies in educational institutions and economic growth over a period of 1966-2008.The short-run effect of female enrolment proxies on GDP is significant i.e., FHSE, FASCE and FULE.While the long-run adjustment impact of GDP on FMSE, FHSE, FSVE, FASCE and FULE is significant with negative value indicating the long run convergence i.e., 3.7%, 4.5%, 11.8%, 26.3% and 19.2% respectively, which would be adjusted every year.The conclusion is that there is a stable long run relationship between the female enrolment in educational institutions and economic growth.
The empirical results moderately support the conventional view that GDP has significant long-run casual effect on female enrolment proxies in Pakistan.This present study supports the unidirectional causality relationship between the GDP and female enrolment in the specific context of Pakistan.This also suggests that only single equation method is insufficient to assess the strong relationship.Therefore it is important to establish simultaneous equations for long-run relationship.It is recommended that government should upgrade the priority of education by raising public expenditure on education to at least 4 per cent of GDP, as recommended by UNESCO.This study provides evidence for the Government of Pakistan's focus to female enrolment in educational institutions which can contribute towards the prosperous future of Pakistan.
By finding the long run relationship among the variable, Engle and Granger Cointegration test has been applied. When the variables are found cointegrated, an Error -Correction Model (ECM) has been applied to determine the short run dynamics of the system.
Table 1 .
Augmented Dickey-Fuller (ADF) Test on the levels and on the First Difference of the VariablesThe null hypothesis is that the series is non-stationary, or contains a unit root.The rejection of the null hypothesis is based on MacKinnon critical values.The lag length are selected based on SIC criteria, this ranges from lag zero to lag one.
Table 3 .
Augmented Dickey-Fuller Test for the Residuals -FPSE | 4,158 | 2010-10-18T00:00:00.000 | [
"Economics"
] |
Taxonomy-Aware Prototypical Network for Few-Shot Relation Extraction
: Relation extraction aims to predict the relation triple between the tail entity and head entity in a given text. A large body of works adopt meta-learning to address the few-shot issue faced by relation extraction, where each relation category only contains few labeled data for demonstration. Despite promising results achieved by existing meta-learning methods, these methods still struggle to distinguish the subtle differences between different relations with similar expressions. We argue this is largely owing to that these methods cannot capture unbiased and discriminative features in the very few-shot scenario. For alleviating the above problems, we propose a taxonomy-aware prototype network, which consists of a category-aware calibration module and a task-aware training strategy module. The former implicitly and explicitly calibrates the representation of prototype to become sufficiently unbiased and discriminative. The latter balances the weight between easy and hard instances, which enables our proposal to focus on data with more information during the training stage. Finally, comprehensive experiments are conducted on four typical meta tasks. Furthermore, our proposal presents superiority over the competitive baselines with an improvement of 3.30% in terms of average accuracy.
Introduction
Relation extraction (RE) is designed to extract the relation between two entities in a given text [1], and has been widely applied in downstream tasks of Nature Language Processing, e.g., knowledge base population and question answering [2]. Traditional deep neural network methods [3] for RE are typically challenged by the need to gather large amounts of high-quality annotation data, which is expensive and laborious. Therefore, few-shot relation extraction is feasible for realistic applications [4]. Furthermore, metalearning methods are proposed to address such a low-resource dilemma [5]. The core of meta-learning (ML) is to optimize methods via diverse meta-tasks, each with several labeled instances, so that the methods can rapidly learn to identify new relations with only few instances. Figure 1 illustrates an instance of two-way one-shot for few-shot RE.
These ML approaches can be broadly classified into three categories, namely model-, optimization-and metric-based ML methods [6]. As a popular solution, the metric-based ML methods focus on designing a metric function in order to identify the distance between instances in the query set and the categories (illustrated with a few instances) appearing in the support set. Prototypical network [7], a simple and effective metric-based ML method, approximately represents each category via a prototype, which is achieved through averaging the embeddings of these instances that belong to the class. A great deal of works are devoted to improving the representation of prototypes, e.g., Gao et al. [8] modifies the representation of prototypes by highlighting the crucial instances and features, and Wen et al. [5] integrates the transformer model into prototype nets for greater expressiveness. In addition, some recent works have utilized external knowledge to provide more clues to the representation of prototypes, e.g., Qu et al. [9] optimizes the posterior distribution of a prototype via a global relation graph as the initial prior of the prototype, Yang et al. [10] employs the text descriptions of relations and entities to enhance representations of a prototype, and Yang et al. [11] fuses the entity concept to constrain the representations of a prototype. However, there are two main limitations to these methods. First, the prototype representation of the above models bears some bias and the discriminative ability is insufficient in few-shot scenarios, which restricts the performance of these ML methods. Additionally, these improved methods usually design complex structures and introduce excessive parameters, which increases the computational burden and also easily leads to overfitting in the few-shot schema. Second, current ML methods treat all training instances equally [6] or pay more attention to very hard instances [12,13], which prevents these methods from extracting useful information from the training instances. Intuitively, on the one hand, tasks that are overly simple provide no valuable information; on the other hand, even humans can only extract critical information from moderately hard instances and struggle with very hard instances, let alone neural network models.
With the aim of alleviating the above problems, we propose a taxonomy-aware prototypical network (TAPN) method, consisting of two modules: a category-aware calibration module and a task-aware training strategy module. Specifically, the category-aware calibration module leverages relation description to explicitly calibrate the prototype distribution in order to obtain unbiased representations and applies prototype-aware contrastive learning to implicitly calibrate the prototype representations to be more discriminative. The task-aware training strategy module leverages the task-aware difficulty to balance the weights of easy and hard instances, which also dynamically adapt different meta tasks.
We evaluate our proposal on four classic meta tasks, and the broad results of the experiment indicate that TAPN is markedly superior to baselines. Additionally, ablation research further validates the effectiveness of these two modules and an error analysis shows the interpretability of TAPN's good performance.
In summary, our major contributions can be summarized as follows: (1) To the best of our knowledge, we are the first to explicitly and implicitly calibrate the prototype representation simultaneously without introducing extra or even harmful parameters. (2) We design a category-aware calibration module to enable the representation of an unbiased and more discriminative prototype by relation description and prototypeaware contrastive learning, respectively. (3) We propose a task-aware training strategy module to extract beneficial knowledge by exploring hard task and sample instances. (4) The experimental findings confirm the validity of our model in terms of accuracy against the competing baselines.
The remainder of the paper is organized as follows: We review the related work for few-shot RE in Section 2, detail our approach in Section 3, design our experiments in Section 4, analyze the results of our proposal in Section 5 and conclude our work in Section 6.
Relation Extraction
RE is designed to determine the relations between entities in a given sentence. Most traditional RE models extract the relations under supervised settings [14], which can be classified into three categories: neural-, kernel-and feature-based methods [1].
Typically, the aforementioned approaches work well based on numerous labeled data. However, it is time-consuming [9] and impractical to collect such massive annotated data in some professional domains. We focus on extracting relation triple in the few-shot scenario.
Few-Shot Relation Extraction
Meta-learning methods [28] have been extensively applied to the few-shot RE. The ML models are trained in various meta-tasks with few instances as demonstrations, then can be generalized to new meta tasks. In general, these ML methods are divided into three category [29]: metric-, optimization-and model-based methods [17].
Model-based methods [30] emphasize on designing the architecture of the model to address the few-shot task. To be specific, MANN [31] designs a memory-enhanced neural network to quickly absorb new data and proposes an effective strategy for accessing the external memory, which provides the ability to quickly predict new relations. Optimizationbased methods [32,33] try to initialize the parameters well. For instance, Finn et al. [32] optimize parameters with few training data so that they can be adapted to novel tasks with a limited number of gradient descent steps. The metric-based approaches focus on learning a metric function to determine the similarity between support sentences and query sentences. For instance, relation networks [34] learn a deep distance metric on the basis of the neural network instead of the fixed Euclidean distance or dot product. The prototypical network [7] predicts relation labels through computing the similarity between the prototype of each class and query sentences, which is derived from averaging the representations of all the examples belonging to a particular class. In addition, a great deal of works are designed to improve the prototypical network: Gao et al. [8] present hybrid attention-based prototypical networks to deal with the diversity and noise of text, Han et al. [12] introduce external relation description and combine global and local features as hybrid prototypes, that learns better characterization through utilizing relational label information.
However, these improved prototypical networks almost introduce extra parameters, e.g., parameters of the attention mechanism, which require sufficient data for optimization and is not realistic in the few-shot scenario. In addition, the prototype representations are usually biased and insufficiently discriminative. In this paper, compared to vanilla prototypical networks, we calibrate the prototype representation without introducing additional parameters.
Contrastive Learning
Contrastive Learning achieves success in computer vision (CV) [35] through pulling together positive instances and pushing negative instances away simultaneously. Different from positives produced by cropping, flipping, distortion and rotation in CV, methods to construct positives for discrete text sequences present a critical problem. Moreover, there are quantities of works dedicated to solving the above problem. For example, to design proper positives, Wu et al. [36] and Meng et al. [37] design word deletion, reordering and substitution techniques, Yan et al. [38] propose four new data-augmentation techniques (adversarial attack, token shuffling, cutoff and dropout), Gao et al. [39] apply random dropout as noise for sentence text and Jiang et al. [40] introduce different templates to express the same sentence text.
However, the aforementioned works usually construct positives and negatives at the instance-level and ignore the connection between instances and categories. Inspired by this, we design a prototype-aware contrastive learning prototype at the category-level, which drives the representations of categories to become more discriminating.
Approaches
In this section, we will display the details of our proposed technique. As exhibited in Figure 2, the structure of our proposal includes two modules: the category-aware calibration module and the task-aware training strategy module. In detail, the featureencoder first transforms the input sentences and relation descriptions into corresponding embedding. Next, the category-aware calibration module can obtain the unbiased and discriminative prototype representations according to the embeddings of sentences and relation description. We can predict the label of each query sentence based on the similarity between all category prototypes and this query sentence. Finally, the task-aware training strategy module can balance the weight between simple and hard data and then ensure propagation of the correct information. In the next section, we first describe the formulation of the problem in Section 3.1. We then detail the category-aware calibration module in Section 3.2 and the task-aware training strategy module in Section 3.3.
Task Definition
Relation Extraction. Given an L x -word text x with a head entity e h and a tail entity e t , i.e., x = {w 1 , · · · , e h , · · · , e t , · · · , w L x }, the RE task can be formulated as training a model to predict the relation label r between e h and e t , where r belongs to a pre-defined relation label set R. It is worth noting that the entity span may consist of multiple words. Few-shot Relation Extraction. Few-shot relation extraction aims to identify the emerging novel relation labels without sufficient labeled data. Therefore, the predefined relation label set R is divided into the base categories R b and novel categories R n for the training and test stage, respectively. This setting simulates the test environment, where R b ∪ R n = R and R b ∩ R n = . Next, quantities of meta tasks are constructed for few-shot RE. Specifically, a meta task T consists of a support set S and a query set Q: T = (S, Q). Following the typical N-way K-shot setting of ML learning, the support set S = x j r ; r = 1, · · · , N, j = 1, · · · , K contains N categories, each category with K labeled instances. The query set Q includes the same N relation categories as S. The few-shot RE methods are trained on meta tasks sampled from the base categories R b , learn general knowledge, and are tested on other meta tasks sampled from the novel categories R n . External Knowledge. The relation description d r = {w 1 , · · · , w L d } for each relation r is also given, where L d denotes the word length of d r .
Category-Aware Calibration Module
In this section, we first calibrate the representation of each category, and then predict the label of the query sentence by a metric function, which calculates the distance between the query sentence and these categories.
Feature Encoder
We employ E to denote the text feature encoder. We use BERT base as E, as shown in Figure 3. We can then obtain the contextual semantic representation h x of an instance x: where Additionally, we can gain the relation description representation for each relation r: where h d r ∈ R 2d , E(d r )[CLS] ∈ R d demonstrates embedding of the start token of the relation description text, and E(d r )[w i ] ∈ R d illustrates the embedding of word w i in the relation description text.
Category Distribution Calibration
Following vanilla prototypical network [7], we average all the instance embeddings in the support set for each relation as vanilla prototype: where we treat prototype c x r as the representation of category r. However, c x r is vulnerable to outliers in the few-shot scenario where there are only very few instances for demonstration, leading to semantic distribution and discrimination bias. Fortunately, the relation description summarizes the semantic characteristics, which elaborates the real meaning. Therefore, we leverage the relation description to calibrate the distribution of the corresponding category representation: compared to c x r , c r is an unbiased prototype and much closer to the real distribution of relation r. In other words, we calibrate the category distribution without introducing supernumerary parameters. We can then predict the category of a query instance by the following metric function: where p(y = r|q) means the probability of query q belonging to relation r. h q , yielded from Equation (1), is a representation of the query sentence. d(·, ·) is the distance function based on dot production. Subsequently, we apply cross-entropy loss to optimize prototype representation: where I r is an indication function, I r = 1 when query q belongs to relation r, otherwise I r = 0.
Category Discrimination Calibration
As for semantic discrimination bias, we apply contrastive learning to discriminate the representations of instances for each relation. In detail, instances should be close to the prototype belonging to the same category and far away from other prototypes as follows: where τ is a temperature hyper-parameter, h x j r denotes the representation of j-th instance in relation r. Thus, here we can obtain discriminative prototypes based on Equation (7).
Task-Aware Training Strategy Module
For a sentence with true relation label r, we predict that it belongs to r with a confidence of p(y = r|q) by Equation (5). We define the easily classified sentence as a very simple instance when p(y = r|q) → 1. Conversely, the extremely difficult classified sentence is the very hard instance with very low confidence. The task-aware training strategy module is designed to optimize our proposal with moderately hard data.
Hard Instances
Intuitively, the models will benefit if they focus more on hard instances instead of treating all instances equally [13]. Therefore, we apply the focal loss function [13] on hard instance to modify the cross entropy loss L f ocal : where ϕ > 0 is a hyper-parameter [13] and reduces the relative loss contributed by very simple instances. Furthermore, Equation (8) is cross-entropy loss when ϕ = 0.
Hard Meta Tasks
However, the harder the sentence, the higher the L f ocal weight assigned to this sentence, which may lead to TAPN failing to learn knowledge since L f ocal focuses excessively on very hard sentences. Therefore, we design an inverse focal loss function at the meta-task level, which pays less attention to the very hard task consisting of very hard classified sentences. We can observe that the greater the inter-class similarity in a meta task, the harder this meta task becomes. We then use the inter-class similarity matrix M ∈ R N×N to measure the difficulty (hardness) of meta task: where · represents the Euclidean norm. Next, we use a scalar m to determine the hard magnitude of a specific meta task in the current mini-batch as follows: where B is the batch size in training stage, m b is the difficulty of b-th meta task in current batch, and · F is the Frobenius norm. The task-aware loss is then defined as follows: L u f f pays less attention to a hard meta task but focuses on hard instances in the meta task. Namely, L u f f balances the weight between easy and hard data; therefore, it can learn useful knowledge from moderately hard data. Lastly, the final objective loss is designed as follows:
Experiments
In this section, we first discuss several research questions in Section 4.1. We then introduce the dataset and baselines to compare with those in Sections 4.2 and 4.3, respectively. Finally, we provide some implementation details in Section 4.4.
Research Questions
We design the following research questions to guide our experiments and examine the effectiveness of our proposal.
Datasets
We conduct our experiment on FewRel [4]. There are 64, 20 and 16 relations for training, validation and testing, respectively. Since the 20 test relations are not reported, we re-split the original published relations into 50, 14 and 16 for training, validation and testing, respectively, according to existing methods [10,11]. In addition, the statistics of FewRel are listed in Table 1. Moreover, the test relation descriptions are listed in Table 2. Table 1. Statistics of FewRel. "#Rel" and "#Instance" denote the number of relations and instances, respectively. "Length" means the average token length of instances.
Id Relation Name Relation Description
"follows" follows immediately prior item in a series of which the subject is a part. P177 crosses obstacle (body of water, road, . . .) which this bridge crosses over or this tunnel goes under. P206 located in or next to body of water located in or next to "body of water", "sea, lake or river". constellation the area of the celestial sphere of which the subject is a part (from a scientific standpoint, not an astrological one). P641 sport sport in which the subject participates or belongs to. P921 main subject primary topic of a work.
Model Summary
We introduce two group competitive baselines for the few-shot RE task to be compared with. We first illustrate the basic ML methods: -Snail [41] applies the temporal convolutions to aggregate information from past experience and designs a soft attention mechanism to pinpoint specific pieces of information.
-GNN [42] defines a graph neural network architecture to propagate label information from labeled data to unlabeled data. -Siamese [43] uses two twin networks with shared weights to calculate the similarity of two inputs and then determines whether they belong to the same category. -Proto [7] predicts the relation labels by calculating the similarity between the query sentences and the prototype of each category, which is obtained by averaging the representations of all instances belonging to a specific category. -BERT-PAIR [44] concatenates the query instance with all supporting instances with a particular label as a series of sequences, and then calculates the similarity of two pairs of instances for predicting the relation of query instance.
We then list some improved prototypical networks by introducing external knowledge through carefully designed complex modules: -KEFDA [45] designs a knowledge-enhanced prototypical network to conduct instance matching and a relation-meta learning network for implicit relation matching. -ConceptFERE (We use the ConceptFERE(simple) version here to allow for computation overheads. Ref. [11] develops a self-attention-based fusion module to incorporate sentence embedding and entity concept embedding, which is valuable for the relation classifier. -HCRP [12] introduces external relation description and combines global and local features as hybrid prototypes, which better learn representations by exploiting relation label information.
Finally, we present the model proposed in this paper: -TAPN leverages the relation description to calibrate the prototype representation without introducing extra parameters and designs an effective training strategy to optimize the model.
Implementation Details
The model configurations are kept the same across all models discussed, including our proposal and the selected baselines. In detail, following [4,46], we assess the performance of DRK on four classic meta-tasks: 5-way 1-shot, 5-way 5-shot, 10-way 1-shot and 10way 5-shot. We apply BERT base as the feature encoder and use ADAM to optimize all the models. In addition, we follow the parameter setting of FewRel [4] and tune other hyperparameters through performing a grid search on a validation set. Furthermore, we present the parameter settings in Table 3. It is worth noting that we set τ to 0.4 on the 10-way 1-shot meta task through conducting a grid search on a validation set.
Overall Evaluation
For answering RQ1, we assess the RE performance of TAPN along with eight competing baselines on four meta-tasks. The overall results in terms of accuracy are listed in Table 4.
Generally, for meta tasks with the same shot number, the performance of all models deteriorates as the number of relational categories (ways) increases. In addition, for the same way-number meta tasks, all the models achieve better performance as the shotnumber increases. The above phenomenon indicates that the difficulty of the relationextraction task increases as the number of shots reduces and the number of ways increases. This can contribute to under-fitting for test tasks, which suffers from a lack of data. Table 4. Overall performance of our proposal and baselines in terms of average accuracy(%) on four typical meta-tasks. The results of the best baseline and the best performer in each column are underlined and boldfaced, respectively. Statistical significance of pairwise differences of the best baseline against our proposed TAPN is determined via a t-test ( for p < 0.05). † marks the results quoted from the original published papers.
Model
Avg 5-Way 1-Shot 5-Way 5-Shot 10-Way 1-Shot 10-Way 5-Shot Subsequently, we focus on the baseline. For the first group methods, BERT-PAIR achieves the best results due to the carefully designed model structure. For the second group baselines with external knowledge, most models achieve better performance than first group models. In addition, HCRP is the best baseline on four meta tasks. This demonstrates that external knowledge provides rich information to alleviate the few-shot dilemma.
Next, we focus on the performance of our proposal on four meta tasks. Generally, our suggested TAPN is superior to all baselines on all meta-tasks and gains a 3.30% improvement in average accuracy, which confirms the validity of the TAPN. In detail, TAPN exhibits 1.38%, 4.16%, 2.98% and 4.68% improvements in the accuracy of HCRP on 5-way 5-shot, 5-way 1-shot, 10-way 5-shot and 10-way 1-shot meta-tasks, respectively, and the performance growth of our proposal increases as the way-number grows and the shotnumber reduces. This demonstrates that TAPN can capture unbiased and discriminative features in the harsh few-shot scenario. In addition, we also evaluate the performance precision of our proposed method and the state-of-the-art baseline HCRP in Table 5. We can observe that our proposed method still outperforms HCRP by 2% improvement in terms of average precision.
Sentence Length
As for RQ2, we study the influence of sentence length on the behavior of all models, in accordance with the sentence length L s . In detail, considering the distribution of testing data, we group the sentences into four groups, i.e., L s ∈ (0, 15), [15,30), [30,45), [45, +∞). The results are plotted in Figure 4.
Generally, almost all the performances of the models drop with an increase in sentence length, which can be clearly observed in Figure 4c. This phenomenon may be caused by the model failing to capture key information as the sentence length increases. In addition, long sentences are more likely to introduce noise.
Next, we compare the results of our proposed method against the baselines. Furthermore, we take the worst-performing 10-way 1-shot meta task for instance to analyze the results. We find that our proposal obtains the best results at every sentence length on all four meta tasks. Furthermore, our proposal is less sensitive to the length of the input sentence than other baselines. For instance, compared to the best baseline HCRP degrades by 24.98% from 91.97% at L s ∈ (0, 15) to 66.99% at L s ∈ [45, +∞), our proposed TAPN only decreases by 15.93% from 97.36% at L s ∈ (0, 15) to 82.01% at L s ∈ [45, +∞). In addition, the improvement magnitude of TAPN consistently increases along with an increasing length of the input sentence, e.g., TAPN outperforms the best baseline HCRP by an improvement of 5.96%, 13.25%, 14.64% and 15.02% at L s ∈ (0, 15), [15,30), [30,45), [45, +∞), respectively. This demonstrates that our proposal can capture discriminative features to alleviate the noise caused by long and tedious sentences. Similar results can be observed for the 10-way 5-shot, 5-way 5-shot, and 5-way 1-shot meta tasks. . Effect on the performance of our proposed method and baselines affected by sentence length on four typical meta tasks: 5-way 1-shot, 5-way 5-shot, 10-way 1-shot and 10-way 5-shot.
Ablation Study
For RQ3, we perform an ablation study to understand the contribution of the various components of our proposal. In the ablation study, we replace or remove some specific components to measure their influence on TAPN, which is marked with the notation"wo". Specifically, "wo/rel" and "wo/cons" denote removal of the category distribution calibration in Section 3.2.2 and category dicriminative calibration in Section 3.2.3, respectively. The "wo/task" and "wo/instance" refer to removal of the hard meta-task finding component in Section 3.3.2 and hard instance-finding component in Section 3.3.1, respectively. It is worth noting that we only conduct an ablation study on 5-way 1-shot and 5-way 5-shot meta tasks given the high computation cost on 10-way 1-shot and 10-way 5-shot meta tasks. Furthermore, the results are presented in Table 6.
As displayed in Table 6, the removal of components leads to model degeneration, proving the efficacy of each component. Additionally, "wo/rel" leads to the biggest drop among the four components as marked in Table 6. The "wo/rel" plays the most important role, which verifies that the previous prototype representation is vulnerable in the fewshot scenario, which affects subsequent classification accuracy. Furthermore, the category distribution calibration module calibrates prototype representation to be unbiased and discriminative without introducing extra parameters.
Error Analysis
To answer RQ4, we first analyze the accuracy of each test relation, and then determine the error sources via error analysis.
First, we present the accuracy of the best baseline HCRP and our proposal on each test relation in Figure 5. Specifically, following Brody et al. [47], we use the parameters of 10-way 5-shot to evaluate the performance on test data by relation. Specifically, for each test relation, we randomly select 5 examples (that is, K = 5) and 50 examples of that relation and place them into the support set and query set, respectively. As displayed in Figure 5, we can observe that the performance of our model is more stable than HCRP. The good performance of HCRP is contributed to by some easily distinguished relations but fails on some difficulty relations, e.g., the accuracy of HCRP is under 40% on relation P206, P26 and P641. Fortunately, our proposed method performs well across all relation categories, and the accuracy on every relation is over 40%.
Next, we conduct an error analysis on the relation "follows" to determine the error sources and the findings are summarized in Table 7. Generally, TAPN outperforms HCRP by 4% improvement in accuracy on the relation "follows". On the one hand, TAPN reduces the error source, e.g., relation "constellation". This may contribute to the calibration based on the relation description shifting the prototype away from an irrelevant relation category. On the other hand, TAPN decreases the error probability on the relation "part of", meaning that TAPN can capture the discriminative features of each relation.
Conclusions and Future Work
In this paper, we propose a taxonomy-aware prototypical network to solve the few-shot relation extraction. Specifically, we design a category-aware calibration module that utilizes the relation description and contrastive learning to calibrate prototype representation to become sufficiently unbiased and discriminative. Furthermore, we develop a task-aware training strategy module, which dynamically balances the weight of easy and hard tasks. In addition, we conduct extensive experimentation on FewRel for four typical meta tasks.
The results demonstrate that our proposal exceeds the state-of-the-art baseline in average accuracy.
However, our proposal may be limited to addressing the cross-domain relation extraction task, where the testing and training data originate from various domains. Therefore, regarding the feature work, on the one hand, we plan to examine the generalization of TAPN in the cross-domain few-shot scenario [48]. On the other hand, we would like to introduce prompt learning for the true few-shot [49] scenario, where both training and validation data are scarce. For example, we can design a template to close the gap between relation extraction and the pre-trained language model, which can exploit common knowledge learned from pre-trained language models.
Data Availability Statement:
The data presented in this study are available on request from the first author.
Conflicts of Interest:
The authors declare no conflict of interest. | 6,788.6 | 2022-11-21T00:00:00.000 | [
"Computer Science"
] |
Anti-EBOV GP IgGs Lacking α1-3-Galactose and Neu5Gc Prolong Survival and Decrease Blood Viral Load in EBOV-Infected Guinea Pigs.
Polyclonal xenogenic IgGs, although having been used in the prevention and cure of severe infectious diseases, are highly immunogenic, which may restrict their usage in new applications such as Ebola hemorrhagic fever. IgG glycans display powerful xenogeneic antigens in humans, for example α1–3 Galactose and the glycolyl form of neuraminic acid Neu5Gc, and IgGs deprived of these key sugar epitopes may represent an advantage for passive immunotherapy. In this paper, we explored whether low immunogenicity IgGs had a protective effect on a guinea pig model of Ebola virus (EBOV) infection. For this purpose, a double knock-out pig lacking α1–3 Galactose and Neu5Gc was immunized against virus-like particles displaying surface EBOV glycoprotein GP. Following purification from serum, hyper-immune polyclonal IgGs were obtained, exhibiting an anti-EBOV GP titer of 1:100,000 and a virus neutralizing titer of 1:100. Guinea pigs were injected intramuscularly with purified IgGs on day 0 and day 3 post-EBOV infection. Compared to control animals treated with IgGs from non-immunized double KO pigs, the anti-EBOV IgGs-treated animals exhibited a significantly prolonged survival and a decreased virus load in blood on day 3. The data obtained indicated that IgGs lacking α1–3 Galactose and Neu5Gc, two highly immunogenic epitopes in humans, have a protective effect upon EBOV infection.
Introduction
The use of polyclonal antibodies has been the first breakthrough event in the treatment of lifethreatening infectious diseases, including plague, diphtheria and cholera [1,2]. Despite the emergence of monoclonal antibodies, polyclonal antibodies from animal sources are still popularly used to treat toxin intoxication or as immunosuppressive agents in transplant recipients [3] or patients with autoimmune diseases [4]. Although animal-derived polyclonal antibodies have potential clinical advantages [5], an important limitation lies in their antigenicity, which results in the rapid, neutralizing immune response of the recipient towards the foreign IgG antigens. Indeed, all patients receiving animal polyclonal IgGs without other immunosuppression (IS) develop severe symptoms of immune-complex disease (serum sickness disease) [6]. The occurrence of these symptoms decreases with the strength of additional IS [6][7][8]. Thus, it is likely that the injection of high doses of animal IgGs will also result in severe serum sickness disease and the neutralization of their biological effects in the context of the prevention or treatment of severe infectious diseases. Furthermore, serum sickness disease symptoms, which include fever, arthralgia, pseudo-meningitis and skin eruptions, may mimic the symptoms of the severe infectious disease that is being prevented or cured.
The antigenicity of foreign IgGs arise from a combination of peptide and sugar antigens, which involve both the Fc and Fab parts of the IgGs in a polyclonal preparation [9,10]. In contrast, human antibodies do not express αGal nor Neu5Gc. Several attempts have aimed to reduce the immunogenicity of animal polyclonal IgGs, including the enzymatic removal of the Fc [11], the "humanization" of the Ig peptide backbone [12], or, as in this paper, the modification of the IgG glycans via knocking out the genes responsible for the expression of two key sugars that are recognized as major xeno-antigens by the human immune system (α1-3 Galactose, referred to as αGal [13], and the glycolyl form of neuraminic acid, referred to as Neu5Gc [14]).
EBOV belongs to the Filoviridae family, which comprises a group of enveloped negativestrand RNA viruses responsible for severe hemorrhagic fever in humans [15]. The EBOV genome is~19 kb and encodes seven proteins that make up the virion: nucleoprotein (NP), virion proteins (VP) VP40, VP35, VP30, VP24, RNA-dependent RNA polymerase L and spike glycoprotein (GP). Surface GP is expressed as the result of transcriptional RNA editing [16] and is a highly N-and O-glycosylated type 1 glycoprotein composed of disulfide-linked subunits GP1 and GP2 generated by proteolytic cleavage of the GP precursor by the cellular protease furin [17]. EBOV GP is responsible for virus entry and is the target of virus-neutralizing antibodies [15]. Several publications have reported contrasting protective effects of convalescent serum [18][19][20] or monoclonal antibody cocktails [21] in curing or preventing EBOV infection, suggesting that animal-derived hyper-immune anti-EBOV polyclonal IgGs may also be useful [22]. By simultaneously targeting multiple epitopes, anti-EBOV polyclonal IgGs are also expected to prevent the generation of EBOV escape variants, a phenomenon already documented for this virus [23][24][25]. Several small animal models exist for EBOV infection, including mouse, guinea pigs and hamsters. Guinea pig infection with a well-characterized, adapted variant of EBOV induces a rapid and lethal disease state [26][27][28]. Therefore, this model has advantages compared to other rodent models and is useful for obtaining a proof of concept before the more ethically demanding primate model.
In this article, we aimed to provide a proof of concept that an anti-EBOV GP polyclonal IgG lacking αGal and Neu5Gc, and thus with a lower expected immunogenic potential in humans, can modify the course of an EBOV infection. Here, we show that double KO porcine IgGs lacking αGal and Neu5Gc prolong the survival of EBOV-infected guinea pigs and decrease EBOV replication in treated animals.
IB4 lectin and anti-Neu5Gc IgY binding on DKO IgGs
For the detection of αGal, ELISA plates were coated with DKO IgGs overnight at 4°C and were then blocked with PBS -Tween 0.1%-OVA 1% (Sigma-Aldrich, Saint Louis, MO, USA) for two hours at 37°C. After washing, the plates were incubated with Isolectin B4, peroxydase conjugated (IB4, 1/100 in PBSTO, Sigma Aldrich) for one hour at 37°C. After washing, the plates were developed using TMB substrate (Sigma-Aldrich), the reaction was stopped with H 2 SO 4 0.5 M, and the plate was read at 450 nm (reference filter 630 nm). For the detection of Neu5Gc epitopes, the plates were coated with DKO IgGs overnight at 4°C and blocked with PBS-Tween 0.1%-OVA 1%. After three washings, the plates were incubated with chicken IgY anti-Neu5Gc (1/1000 in PBSTO, Biolegend, San Diego, CA, USA) for one hour at room temperature. After washing, the plates were incubated with a goat anti-IgY-HRP (Abcam, Cambridge, UK) for one hour at room temperature. The results were developed as described above for anti-αGal.
Mass spectrometry
The analysis of Neu5Gc, Neu5Ac, and αGal moieties on porcine IgGs was performed using mass spectrometry (MS), as previously described [29]. Briefly, double KO porcine IgGs (100μg) were reduced with dithiothreitol (Sigma-Aldrich), alkylated with iodoacetamide (Sigma-Aldrich), and digested with trypsin (4 μg, Promega, Madison, WI, USA). The digestion mixture was separated by reverse-phase HPLC. The pooled glycopeptide fractions were further digested with β-galactosidase from bovine testes (Prozyme, Hayward, CA, USA). All MS analyses were performed on an UltrafleXtreme mass spectrometer (Bruker Daltonics, Billerica, MA, USA) equipped with LID-LIFT TM technology for tandem MS experiments. Dihydroxy-benzoic acid was used as the matrix.
DKO pig immunization (S1 Table) A 14-month old male DKO pig weighing 102 kgs was obtained by cloning as described in [30], and was immunized with the VLPs at a dose of 700 μg in a 2 ml mix (v/v, 1:1) with Alhydro-gel1 adjuvant 2% (Invivogen, San Diego, CA, USA). Five intramuscular (IM) injections were performed in three different locations on days 0, 15, 29, 44 and 79. A 10 ml volume of blood was harvested on day 0, and after each immunization on days 15, 30, 57 and 83 to assess the antibody titers. On day 91, 100 ml of blood was taken for immunoglobulin extraction. All animal procedures were approved by the local Ethics Committee of the Laboratory of Reproductive Technologies, Avantea srl, and were carried out in accordance to the Italian regulation DGL 116/92.
IgG purification
IgGs were purified on a Protein-A column (high performance Sepharose™, GE Healthcare, Little Chalfont, UK) using a low pressure chromatography and a 280 nm UV, pH and conductivity recorder. The immunoglobulins were eluted with a solution of 0.1 M citric acid pH 3, followed by an immediate pH neutralization of the eluate to pH 7-7.4 with a solution of 1 M TRIS pH 8. The IgGs were then dialyzed against PBS 1X and their amounts were assessed by spectrometry at 280 nm.
Anti-EBOV IgG titers (ELISA) and neutralization assays HEK 293T cells were transfected with phCMV GP. Transfected cells treated with 1% Triton were used as an antigen source for ELISA. Polysorp plates (Nunc, Thermo Fisher Scientific) were coated overnight at 4°C with the antigen (1:500 dilution), and incubated for 1 hour at 37°C in PBS containing 5% skimmed milk and 0.1% Tween 20 prior to incubation with dilutions of pig serum (in PBS containing 1% skimmed milk and 0.1% Tween 20). The presence of anti-EBOV antibodies was developed using an anti-pig HRP (Sigma-Aldrich) and TMB substrate (KPL, Gaithersburg, MD, USA).
The EBOV neutralization assay was based on the neutralization of a recombinant vesicular stomatitis virus (VSV) in which the gene coding for the VSV surface glycoprotein G was replaced with the EBOV GP gene. 200 plaque forming units (PFUs) of infectious VSV-E-BOV-GP were incubated with consecutive two fold dilutions of the sera in 0.1 ml of DMEM for 30 min and were then used to inoculate Vero E6 cell cultures in eight replicates. After 1 hour, 5% FCS DMEM was added and the cells were incubated for 3 days prior to observation of the cytopathic effects by counter-staining with crystal violet. A 50% neutralizing titer was determined graphically as the sera dilution that neutralized the virus in 50% of the wells.
Animal experiments: Assessment of DKO serum toxicity, guinea pigs groups and procedures (Table 1) In order to test the safety of purified pig anti-EBOV IgGs obtained from double KO pigs (a pre-requisite from the ethical committee of the P4 facility to avoid unexpected toxicity of DKO IgGs for rodents), naïve male C57BL/6 mice of 8 weeks of age (Janvier Labs, Le Genest Saint Isle, France) were injected with the IgGs as following: two groups of 5 mice each received 2 mg of IgGs per day (corresponding to 400 mg/kg) intraperitoneally (IP) for 4 consecutive days. The mice were followed for changes in weight and general appearance for 1 month post-injection (S1 Fig).
The therapeutic efficacy of the anti-EBOV IgGs was tested in Hartley guinea pigs (150-200 g female guinea pigs, Harlan Laboratories, Netherlands) infected with recombinant guinea pigadapted EBOV (EBOV 8MC passage 2) [26,27]. All experiments were carried out in the BSL-4 animal facility at the Inserm-Jean Mérieux BSL-4 laboratory in strict accordance with European directive 2010/63 and French regulations. The protocol was approved by the ethics committee for animal experimentation (Comité d'Evaluation Commun au Centre Léon Bérard, à l'Animalerie de transit de l'ENS, au PBES et au laboratoire P4 (CECCAPP) N°C2015; permit N°2015090209307871). Before handling, the animals were anesthetized in an induction box using isoflurane 3% under an air flow of 1 L/min. The animals were challenged through the IP route with 1000 TCID50 suspended in 0.3 ml DMEM. The animals were divided into 4 groups of 5 animals, and were treated as indicated in Table 1. The "Mock-PBS" group (Group 1) received 0.3 ml DMEM (IP) on day 0, and 1.5 ml PBS (IM in three different injection points) on days 0 and 3. The animals of the group 2 ("EBOV + non-immune IgGs from DKO pig") were infected with EBOV on day 0. Three animals received 65mg of IgGs from a non-immunized DKO pig, one received 55.25mg, and the last animal received 1.5 ml of PBS. This difference was due to the limited amount of non-immune DKO IgGs available (n = 4 animals only on the group of 5 animals originally planned). In the group 3 ("EBOV + anti-Ebola IgGs from DKO pig (day 0)" group), all animals were infected at day 0 and received 65 mg of anti-EBOV IgGs the same day. All animals of the group 4 ("EBOV + anti-Ebola IgGs from DKO pig (day 0 and day 3)") were infected on day 0; four animals were treated IM with 68 mg of anti-EBOV IgGs on day 0 and 18.36 mg of anti-EBOV IgGs on day 3, and one animal was treated with 68 mg on day 0 and 6.8 mg on day 3. The differences between the doses at day 0 and day 3 were due to the limited total amount of hyper-immune purified IgGs: the highest dose was given at the first injection on day 0, in order to have comparable doses in groups 3 and 4 on day 0. The animals were monitored for symptoms, weight and temperature every day from day 3 postinfection until death or euthanasia at the end point corresponding to a 20% weight loss. Animal sacrifice was performed under isoflurane anesthesia by intra-cardiac injection of pentobarbital.
Blood samples were obtained on days 0 and 3 from all animals for viral load analysis. The possible toxicity of the anti-EBOV IgGs, as a potential confounding factor of DKO IgG activity, was ruled out in C57/B6 mice injected IP with a dose of 400 mg/kg.
Viral RNA level assessment
Viral RNA was extracted from the sera of EBOV-infected guinea pigs (Qiagen viral RNA extraction kit, Hilden, Germany) and analyzed using a one-step SYBR green RTqPCR kit (Eurobio, Les Ulis, France) and primer pairs: EBOV Forward 5'CGGAGGCTTTAACCCAA ATA (L polymerase, position 14870) and EBOV Reverse 5'TCATACATGGGAGTGTGGCT (L polymerase, position 14987). The analyses were performed on a Roche LC96 real time PCR apparatus. Quantification was performed using a range of dilutions of EBOV DNA and expressed as relative copy number per ml (detection limit of the technique is 180 relative copies per ml).
Statistical analysis
Survival in the different groups was compared using Kaplan-Meier analysis and was analyzed using a log rank test. Weight loss curves were analyzed using a repeated measures two-way ANOVA test. In the group with two anti-EBOV IgG injections, the virus loads on day 3 (before the second injection of IgGs) were compared using a non-parametric Mann Whitney test. The correlation between virus load and survival time was analyzed using Kendall's rank correlation coefficient.
Results
Absence of αGal and Neu5Gc on DKO IgG
Anti-EBOV titers and virus neutralization test
Over an 83 day-period, a single double KO pig was immunized five times with 700 μg EBOV VLPs generated upon the co-expression of the viral proteins VP40, VP24, and NP and the surface glycoprotein GP. An analysis of immunization profile showed that each injection was beneficial in terms of an increase in anti-EBOV antibody titers, with a maximum ELISA titer of 1:100,000 on day 83 (Fig 1). The titer was determined as the last dilution with a signal over the value of the mock control plus two times the standard deviation. Anti-EBOV neutralization tests were performed with the sera collected on days 57 and 83 using an SN50 assay with a recombinant VSV carrying EBOV GP (see the Materials and Methods section). Neutralization titers of the sera were 1:40 and 1:100, respectively.
Protective effect of anti-EBOV IgGs
The design of the experimental procedure is given in Table 1. Fig 2 shows that the weights of the guinea pigs that received the injections of anti-EBOV IgGs on day 0 only or on days 0 and 3 were significantly different (p<0.05) from the weights of the animals that received IgGs from the non-immunized double KO pig, suggesting that the course of the disease was less vigorous in the treated animals. However, the weight curves were similar in the groups that received one dose on day 0 or two doses on days 0 and 3. Fig 3 shows the Kaplan-Meier survival curve of the animals in the various experimental groups. Three animals of group 3 were found dead on days 7, 10 and 11 post-injection, and one from group 4 was found dead on day 7. These animals displayed Ebola virus disease symptoms, notably a high fever, the day before death but had not yet reached the end point criteria at that time. No samples were collected from dead bodies. There was a significant difference (p = 0.042, log rank test) in the survival time of the animals that received the anti-EBOV IgGs compared to the untreated group. However, no further improvement was observed in the animal receiving the two anti-EBOV IgG injections on days 0 and 3. No animal survived the EBOV infection after day 12.
α1-3Gal and Neu5Gc KO Pig IgGs against Ebola Virus
In order to evaluate the toxicity of the hyper-immune IgGs, mice were challenged with a high dose of DKO IgGs (400 mg/kg). All animals had a stable weight (S1 Fig) and normal activities without clinical signs of toxicity up until the end of the follow-up at one month post-injection, when the animals were euthanized. These data indicate that the deaths in guinea pigs that received the virus and the anti-EBOV IgGs were not due to IgG toxicity. Fig 4A shows the level of EBOV transcripts in circulating blood in the various groups of animals. Values under the detection limit of the test (180 relative genome copies/ml) were considered as negative values. There was a low but borderline significance (Mann Whitney test, p = 0.055) of the viral genome load in the anti-EBOV IgG treated guinea pigs compared to the untreated animals, with a 3-log median value decrease obtained on day 3 post-infection (i.e., as assessed in 10 animals, in the serum harvested before the second injection of anti-EBOV IgGs in group 4). Furthermore, as shown in Fig 4B, there was a significant correlation between the level of virus load and the survival time of the treated animals (Kendall's rank correlation coefficient, p = 0.0003). Altogether, the data show that the anti-EBOV IgGs from the DKO pigs could delay the death of the treated guinea pigs and significantly decrease the circulating virus loads at an early time point following infection.
Discussion
Polyclonal IgGs offer several theoretical advantages compared to monoclonal antibodies by displaying an extended repertoire and functional capacity [1,5,31]. Due to the high diversity of the epitopes recognized by polyclonal IgGs and the possibility to build platforms of animal immunization against viral variants, polyclonal IgGs may also alleviate the problem in which viral mutants can escape the effects of individual antibodies. Data on the efficacy of polyclonal preparations to prevent or cure Ebola disease in humans or animal models are conflicting. The treatment of laboratory-acquired infection using convalescent phase serum [18] and early reports of a survival of 7 out of 8 patients after blood transfusions from convalescent patients [19] spurred further experimental studies, although this last study did not reach statistical significance, and was limited by a late treatment of the patients following exposure to the virus [32]. Nevertheless, blood transfusion from convalescent-phase monkeys immunized against EBOV failed to prolong the survival of infected animals [33]. However, hyper-immune horse sera with in vitro neutralizing capacities have shown its capability of reducing viremia in an experimental model of the disease [34], and in some studies, purified polyclonal antibodies from convalescent primates totally protected infested animals following infusion performed immediately or 48 hours after infection [35]. Furthermore, ovine anti-EBOV hyper-immune IgGs with a high anti-EBOV neutralization titer, induced full protection against EBOV in guinea-pigs [22]. Although this study differs from ours, with a possibly less virulent EBOV infection model (as suggested by an average survival time of infected guinea pigs of 11 days versus 9 days using another guinea pigs adapted virus [36]), these data strongly suggest that a passive protection strategy may be successful. Our study also differs by a different virus (IP) and antibody (IM) injection route, aiming to decrease immediate interference between the two moieties which may artificially improve the efficiency of the drug, although an IV route might have been more pertinent when considering the systemic disease model used here. However, this approach using unmodified ovine IgGs is also likely to generate a strong immune response against αGal and Neu5Gc epitopes. Other attempts to passively protect animals from an experimental disease using polyclonal antibodies from convalescent animals have also yielded conflicting results [23,37,38]. Recent data regarding the use of passive plasma-therapy in humans have not been definitive [20].
In this article, we aimed to study the efficacy of polyclonal IgGs obtained from genetically modified pigs, and directed against VLPs displaying several EBOV proteins, on the survival of guinea pigs infected with EBOV. These "humanized" polyclonal IgGs lack αGal and Neu5Gc, two major xenoantigens for the human immune system. These two sugars are not expressed specifically in human beings (and some new world monkeys), due to a loss mutation, however they are naturally expressed in guinea pigs [39]. Therefore, using such IgGs in a guinea pig model does not impact their potential immunogenicity, and our model of infection does not highlight at full degree the importance of "glycan-humanized" porcine IgGs for disease treatment. Rather, we aimed to explore here whether the modifications in these IgG molecules did not alter their in vivo efficacy in an experimental EBOV disease.
In the context of EBOV disease development in humans, these modified IgGs would not be recognized by preexisting anti-αGal [13] (including of IgE isotype [40]) or anti-Neu5Gc antibodies, which are present in most normal individuals as a result of immunization by the gut flora (for αGal) [13,41], or by diet (for Neu5Gc) [42,43].
As the infusion of unmodified animal polyclonal IgGs in almost all cases results in serum sickness disease [6], there have been several attempts to decrease the expected immunogenicity of animal hyperimmune IgGs against life-threatening viruses, including EBOV. For instances, partially "humanized" bovine IgG protein backbones have been obtained from genetically engineered animals [12,44]. Moreover, truncated IgG fragments lacking Fc [11] have been proposed. However, these "humanized" bovine IgGs still display substantial stretches of bovine peptide sequences that do express Neu5Gc and αGal. Truncated polyclonal animal preparations lack the major functions of IgGs related to complement activation and Fc-gIII-R binding, which are likely important in the protection against the virus. In addition, the F(ab)'2 preparation is not deprived of the glycans displayed by hyper-variable regions in approximately 20-40% of the polyclonal molecules [9,10].
The DKO pig mounted a vigorous humoral immune response against the VLPs, although the antigen preparation may not be optimal compared to, for instance, VSV-EBOV GP (an antigen which was not used due to its possible pathological hazards in pigs [45]). The "orphan" situation of SIGLECs, the natural ligands of Neu5Gc [39], results from the CMAH KO and may be favorable for improving the pig immune response against viral antigens since the genome copies/ml, and all values under this threshold were considered as negative data (as represented on the graph). B: Correlation between viral load on day 3 and guinea pig survival. Kendall's rank correlation coefficient showed a significant negative correlation (p = 0.0003) between EBOV viral load at day 3 and survival following infection, when considering all pooled data. Non-treated animal's values are displayed as circles whereas treated animal's values are displayed as squares.
doi:10.1371/journal.pone.0156775.g004 α1-3Gal and Neu5Gc KO Pig IgGs against Ebola Virus SIGLEC/Neu5Gc interaction is a strong inhibitor of B cell activation. The neutralization titers of our anti-Ebola serum remained modest, although the different methods of quantification used for the measurement of antibody titers are difficult to compare. Refinements of the immunization procedure and the antigen preparation may improve efficiency and allow better results. However, the significant effect of the DKO IgGs on the survival and virus load in the guinea pig, despite a low titer of neutralizing antibodies, suggests the possible effect of an increased complement binding compared to wild-type pig IgGs ( [46], Salama A et al., manuscript in preparation). This is also consistent with the fact that the neutralization tested on the pig serum was found to be complement-dependent, and disappeared after heat inactivation.
In this study, an in vivo assessment of the anti-EBOV IgGs therapeutic effect was performed using a guinea pig model [26,47] and a well characterized guinea pig-adapted EBOV [27]. This model allowed for the IM injection of the anti-EBOV IgGs in a different site from that of the virus injection. This is in contrast to reports in the mouse model, in which both the EBOV and the anti-EBOV preparation were injected IP [48]. First, we showed that fatality and weight loss in the infected guinea pigs injected with the genetically modified anti-EBOV IgGs did not result from a possible toxicity of the injected IgG preparation. In a separate experiment using mice we demonstrated that 400mg/kg of anti-EBOV IgGs neither affected the survival nor the weight of the animals. Moreover we showed that the injection of anti-VLP IgGs could significantly prolong the survival times of infected guinea pigs, although no animal ultimately survived the virus injection. A second injection of anti-EBOV IgGs did not result in any further significant prolongation of survival, which could be partly explained by the lower dose of IgGs given at day 3.
Viremia in patients with EBOV diseases is considered to be a strong predictor of death [49][50][51]. The blood virus load measurements showed both a 3-log median value decrease in virus transcript levels (Fig 4A) on day 3 (i.e., before the second IgG injection; as assessed in 10 animals) but also a significant difference between the animals in the circulating EBOV. Apparently the 3-log drop in circulating EBOV was not sufficient to allow survival. Most likely continuous virus replication and, importantly, the release into the blood of soluble viral glycoproteins capable of neutralizing at least some of anti-EBOV IgGs [52,53] could explain why this current design to cure an infection showed only a modest success. However strong viral load diminution induced by the polyclonal IgGs suggests that this treatment may offer an early anti-viral synergy when associated with higher dose of anti-EBOV IgGs, replication inhibitors or a vaccine.
Although our conclusions need to be confirmed using a greater number of animals and increased neutralizing titers, our experiments altogether suggest that IgGs lacking Neu5Gc and αGal can modify the survival and virus load following a lethal injection of EBOV in guinea pigs. These data encourage us to use these genetically modified animals to prepare hyperimmune anti-EBOV IgGs of higher virus neutralizing titers for future studies in guinea pig and cynomolgus models before considering a clinical application. mode prior to treatment of the sample with β-galactosidase; C: positive ionization, reflector detection mode after β-galactosidase treatment. The effect of β-galactosidase on the different glycoforms is shown by the arrows. All terminal galactosyl residues are removed by the enzyme, indicating the absence of galactosyl residues linked in α1,3. (DOCX) S1 Table. Double KO pig immunization protocol. A DKO pig was immunized with five doses of 700μg VLPs each. Injections were done intramuscularly on days 0, 15, 29, 44 and 79. A 10 ml volume of blood was harvested on day 0 and on days 15, 30, 57 and 83 after each immunization assess the anti-EBOV antibody titers. On day 91, 100 ml of blood were taken for IgG extraction. | 6,227.2 | 2016-06-09T00:00:00.000 | [
"Biology"
] |
Graphene for Thermal Storage Applications: Characterization, Simulation and Modelling
In recent years, interest in the thermal properties of graphene constituents has seen rapid growth in the fields of science and engineering. The removal of heat in the continuous processes in the electronics industry has had major issues in thermal transmission in lower-dimensional assemblies. It has also shown fascinating topographies as the carbon allotropes and their derivative compounds expel heat. Numerous research articles reported within the past 15 years have demonstrated enhanced electron flexibility, exceptional thermal conductivity and mechanical behaviour, as well as excellent optical properties of graphene as a single atomic layer. This review article tries to provide a detailed summary of the heat exchange properties of graphene structures and graphene-based materials such as nanoribbons with few-layered graphene. Thermal and energy storage management systems have played a major role in the increase in marketable products in recent times. The purpose of this review is to summarize the current research on thermal properties with regard to the management and energy storage of graphene materials, focusing on characteristic properties, industrialization, modelling and simulation, and their applications in specific thermal storage systems.
Introduction
A typical problem faced by large energy storage and heat exchange system industries is the dissipation of thermal energy. Management of thermal energy is difficult because the concentrated heat density in electronic systems is not experimental. 1 The great challenge of heat dissipation systems in electronic industries is that the high performance in integrated circuits significantly increases the power consumption. In recent times, Moore's law has not been in use, as the physical conditions and limitations imposed on quantum effects is minimized. 2 Graphene materials are emerging as the latest methodology to store energy, and the use of transistors is decreasing, as it is likely to impose tension on the thermal management systems without the inclusion of non-uniform heating between chips. 3 The two-dimensional (2D) materials are layered structures with firm thickness which bond to a hexagonal carbon lattice plane and display good operational, electric, current and mechanical properties, and as a result, these materials have become one of the foremost materials in research fields to employ it in numerous applications of science and engineering. 4 Figure 1 shows the measurement of heat conduction by utilizing a graphene structure. Graphene was heated with a laser light concentrated in the centre of the suspended part. The temperature increase is determined by changing the position of the G peak in the Raman spectrum of graphene. The exceptional properties of graphene and 2D materials have made it possible to employ them in various electronic systems such as batteries, sports equipment, and solar cells, and they have great potential impact in future applications as well. The enhanced thermal conductivity of graphene and 2D-based materials has made them uniquely suited for the direct thermal management and control of electronic systems. 5 The extensive wavelength phonon transport in 2D crystal lattices makes graphene-based materials the strongest with an exceptionally long free path limited by sample size despite deviations of heat transfer. 7 Widespread studies have been carried out using by-products of graphene such as graphene oxide (GO), graphene film, graphene fiber, graphene foam, graphene laminate, and graphene thermal interface materials (TIMs) for heat administration applications along with the exceptional thermal properties of graphene structures and graphene-based materials. 8 Graphene also serves as a strong thermal alloy filler substance rather than carbon nanotubes (CNT) owing to its exceptional thermal requisite property and its inexpensive conditions. The patterned graphene layer films, fabricated from graphene flake layers, display outstanding thermal functioning and enhanced efficiency as heat dissipaters. 9 This mini review article explores the recent developments in thermal management systems using graphene and graphene-based materials along with other 2D materials which include hexagonal boron nitride (HBN). 10,11 Overall, the variation in thermal modelling methods applicable to these graphene materials are obtainable with the view of modelling and simulation in various thermal management systems and energy storage devices. They are summarized and described with parameters and accuracy, and lastly, the major problems and potential uses of graphene materials and other 2D materials for thermal and energy storage systems are discussed in detail.
Stimulation of Graphene Materials in Thermal Engineering
Graphite is a form of naturally occurring carbon. It is, opaque, black to steel grey, and has a soft lubricating texture. Graphite shows two crystalline structures, hexagonal and rhombohedral. Both crystalline structures have high anisotropy levels that determine the characteristics of graphite, particularly electrical and mechanical. Graphite is recognized as a superior heat and electricity driver, for its maximum natural strength and rigidity over other minerals, its resistance to chemical attacks and its resistance to temperatures higher than 3600 °C (6500 °F). Graphene is a pure type of carbon in which each particle can be accessed from different sides for fabrication response. Particles on the edges of a sheet of graphene are unique in chemicals. It has the highest percentage of edge atoms. Impurities in a graphene sheet increase the reactivity of the substance. Its thermal conductivity and mechanical strength result in the good in-plan characteristics of graphite (independently around 3000 W m −1 K −1 and 1060 GPa), and its breakdown quality is comparable to that of carbon nanotubes, for virtually the same kind of defects. Graphene can produce nextgeneration electronics, biomedical products, composites and coating, sensors, and membranes that are currently limited, and manufacture them faster with a wide range of outstanding qualities. The production of enhanced source density batteries such as lithium-ion (Li-ion) batteries has made progress in the usage of cell phones, personal electronics and motorized industries, as the temperature increase outside the usual operating range has a negative impact on the effectiveness of the Li-ion batteries, and if overheated, the battery causes the cell to split or burst. 12 The thermal phase change material (PCM) can predict heat control of high-density ion battery packs, which in turn minimizes the temperature increase in the battery owing to concealed heat storage and segment variations over a minimum temperature range. 13 Typically, PCMs have much lower thermal conductivity, in the range of 0.17-0.35 W m −1 K −1 at room temperature, compared with the heat conduction of silicon (Si) and copper (Cu) at room temperature (RT), which lie in the range of ~145 W m −1 K −1 and ~381 W m −1 K −1 , respectively. 14 Figure 2 shows graphene-based material used as a thermal device under coupling quantities. Thermal devices are classified according to their physical dimensions, which can be mainly divided into four categories: uncoupled thermal devices, thermoacoustic coupling devices, thermoelectric and thermo-optic coupling devices, and various functional devices. The predicted PCMs store heat from the battery as a replacement for heat transfer to the battery pack. They are also employed in battery cells to shield the battery-operated cell from large ambient temperature variations which completely differs from the heat regulation of processor coded chips. 16 In order to minimize the temperature range, the device uses chip surge, thermal interface material (TIM) or heat propagators to allow the transfer of heat from the defect to the heat sink. 17 The heat conduction of TIM is in the range of about 1-10 W m −1 K −1 , while the thermal diffusion of solid graphite-based materials is in the range of 1000 W m −1 K −1 . 18 The graphene material serves as a filler
3
for thermal conductivity by introducing in the current analysis a hybrid PCM of these two separate thermal management methods. 19 The properties that make graphene an extraordinary filler material are its enhanced internal heat conduction and durable bonding to numerous matrix materials. 20
Thermal Conductivity of Graphene
Heat transfer is slowed by the impact of microscopic elements along with the amount of heat carrier charges in the area. The thermal performance is controlled by Fourier's law, which is expressed as: q, the limited temperature flux has units of W m −2 , the heat conduction of the substance has units of W m −1 K −1 and the limited temperature increase has units of K m −1 , which is the heat conduction of the substance. 21 Fourier's law elucidates the heat flow from a high-temperature section to a lower-temperature section, and it can be precisely calculated in a substance. However, in electronic systems the thermal conduction of heat has to be supported over various materials by exchange interfaces, and thermal resistance is widely used to regulate the ability of heat carrier transport as it is stabilized and relaxed. The thermal resistance is: 22 ∆T is the temperature variance (K) and Q is the heat energy (W) of the outer binary surfaces. 23 The entire heat resistance (R) comprises the heat resistance of the materials laterally. The temperature conduction path is determined by the heat conduction and depth of the materials along with the exchange at the interface of two dissimilar materials, based on numerous features including the bonding strength of materials, surface roughness, and surface cleanliness, among many others. 24 In optical performance, the ΔP heater is furnished with an optical laser light focussed on a deferred graphene layer associated with the heat sink at both corners. The temperature increase ΔT, in response to the degenerated power ΔP, is studied by Raman spectroscopy. 25 The Raman G peak in the graphene spectrum revealed the enhanced temperature, T, dependence. Standardization of the spectral position of the G peak at temperature must be done by altering the heat using a lower laser power source in order to avoid limited heating. 26 The deferred graphene layer is heated by growing laser power through thermal conductivity dimensions. The limited temperature increase of graphene is given by ∆T = ∆ωG/φG, where φG is the heat coefficient of the Raman G peak in the suitable temperature range. 27
Thermal Conductivity of Graphene Materials
The unusual thermal conductivity characteristics of graphene have led to further investigations of graphene materials and few-layer graphene (FLG) in TIM, thermal compounds and coverings. 28 In the first study of graphene alloys, smaller freight segments of arbitrary graphene fillers increased the thermal conductivity of epoxy alloys. 29 Various types of graphene thermal alloy fabrication methods are derived from variations in matrix material, graphene efficiency, lateral size and thickness of the graphene filler, and other types of thermal conductivity. 30 Most of the preliminary studies of graphene filler thermal alloys were restricted to the lower loading fractions of the filler, f < 10 vol.%. These conditions recently changed when large graphene loading compounds became available due to technical developments and substantial cost reductions. 31 The thermal properties of alloys with a high loading fraction of graphene or FLG fillers are of interest from the point of view of basic science and practical applications. Higher loading results in thermal percolation in alloys. 32 Thermal percolation is a process which is not well understood as electrical holes. An electrical hole is designated by the scaling law, where the electrical conduction of the alloy, the freight capacity portion of the filler, the loading fraction of the filler and the critical exponential are within the electrical charge limit. 33 Figure 3 depicts the heat dissipation of a graphene-based epoxy composite. In most cases, as opposed to electrical conductivity, thermal conductivity of alloys does not show any visible fluctuations. To overcome the complex of nanofillers in the polymer, their applications improve the thermal conductivity of the mixture, in which case graphene is enhanced by the chemical vapour deposition on the foam. Further production of graphene thermal composites face issues such as hexagonal boron nitride (HBN) control of electrical and thermal separation in graphene alloys with filler optimization and combination of graphene and other electronic 2D insulation fillers. 34 In addition, graphene composite materials have shown that they are good prospects as heat sink materials. 35 Possible heat sinks are made of metal, such as copper or aluminium, with wings to raise the surface area. However, because of its low weight, anisotropy and higher thermal conductivity, carbon-based heat sink structures are of much interest. 36 Graphite has a long history and is known to be a material for heat sinks. In recent years, patents have been filed on the basis of graphene-improved heat sinks due to the benefits of lower weight and higher heat efficiency. 37 In addition, graphene or graphite-based materials for heat sinks can control thermal conductivity in different directions, allowing for preferential heat transfer. 38,39 Some of the reported works about graphene and graphene polymer composites include electromagnetic pollution, which is damaging to human health and interferes with the normal operation of other electronic equipment, is also becoming increasingly serious as miniaturization, densification, and high-power electronic products are rapidly evolving. By absorbing electromagnetic energy and converting into heat energy, it can be effectively reduced, leading to another problem when temperatures increase. There is therefore urgent theoretical significance and potential practical application for the design of polymer composites with a high thermal conductivity (λ) coefficient and outstanding electromagnetic interference shielding effectiveness (EMI SE). The λ value of epoxy nanocomposites reaches a maximum 2.3 times higher than pure epoxy resin. 40 The resulting problem has gained prominence over thermal dissipation, which has greatly affected the stability and service life of the products produced. Preparation of materials with a high coefficient of thermal conductivity (λ) is an important solution. 41,42 An efficient and cost-effective method is to introduce highly thermally conductive composite compounds in order to improve the final λ values of polymers. Excessively high loads of thermally conductive fillers are often in demand, however, to achieve relatively high λ, posing significant problems such as poor treatment, degraded mechanical characteristics, increased density and cost. 43 This is followed by a combined method of in situ polymerization, spinning and hot pressing, to produce the corresponding thermally superior nanocomposite polyimide. In view of the filler or matrix interfaces, filler dispersion and alignment, an improved thermal drive model is also proposed and established. 44
Graphene Foam
Graphene foam comprises graphene structures packaged in a permeable macroscale foam structure. The foam absorbency substantially decreases the actual thermal conductivity of graphene foam, with thermal conductivity varying from 0.26 W m −1 K −1 to 1.7 W m −1 K −1 at 0.45 vol.% of solid density. 46 However, thermal conductivity of graphene foams is similar to the volume of metal, which has a high absorbency. In accumulation, graphene foam has the highest grade of compression, making it desirable for TIM applications. 47 Graphene foam is principally fabricated by graphene chemical vapour deposition (CVD) in Ni foam and then leaves the free-graphene structure by engraving the prototype. Similar construction may also be achieved by the use of a restriction moulding or a hydrothermal decrease of the GO suspension. 48 As stand-alone assemblies, both graphene material foam or graphene CNT air gel are shown for TIM applications with a heat conductivity of about 88 W K −1 for low graphene foam and lower thermal interface resistance at low pressure. Analogous assemblies have been confirmed using h-BN with cross-plane heat conductivity up to 62 W m −1 K −1 for flattened h-BN foam. 49 Graphene and h-BN foam penetrate to form polymer composites. Another vertically oriented graphene foam epoxy complex is formed with a flat-to-thermal conductivity of 35.5 W m −1 K −1 at 19% graphene loading fraction, which is much higher than the randomly distributed graphene improved composites. Thus, graphene-carbon hybrid foam is demonstrated to be a good candidate as a thermal basin. 50
Perpendicular Arrangement of Graphene Sheets
Graphene slip materials have an exceptional level of thermal conductivity; nevertheless, they are usually unfit for heat dissipation applications owing to their poor level of thermal conductivity. 51 A possible resolution for this limitation is to stack several graphene sheet structures to make a weight substantial enough that it can be employed for TIM and further thermal applications. Liang et al. studied the idea of producing a material with isothermal conductivity of 112 W m −1 K −1 . Graphene film is soldered and combined with joints or silicone, and then the thermal conduction axis is applied diagonally to thin slices. 52 This principle was improved at 615 W m −1 K −1 and 1379 W m −1 K −1 , with thermal conductivity, proposed by Wong et al., which eliminates typical TIM defects with unusually high thermal conductivity compared to traditional TIMs and a thicker bond line than the heat sink material [reference]. Figure 4 explains the stepwise process involved in a 3D graphene film composite. An aqueous mixture of PS nanospheres and CMG sheets was filtered to obtain porous chemically modified graphene (CMG) film with a uniform pore size of 2 μm, followed by removal of polystyrene (PS). As a replacement for conventional Thermal interface materials (TIMs), the thermal resistance between the TIM and the bonding exteriors limits the thermal conductivity factors. 53 As Wong et al. and others have noted, overall performance depends primarily on contact, and retaining performance rivals, good ductility and thickness by bonding to a thin indium layer with thin solder joints is critical for slit satisfying applications. 54
Graphene Hybrid Assemblies
The unique properties of 2D graphene materials can be combined and exploited in conjunction with other hybrid materials. In general, the thermal conductivity of graphene structures through the plane is a limitation. 55 One way to improve Z-directional thermal conductivity is to add further heat paths over the covalent bonding nature of the graphene material layers using intermediate content. Figure 5 shows the self-assembly of graphene-based hybrid film coating the surface., with graphene singlephase self-assembly on the copper heating surface. At the time of heating, the graphene oxide sheet is dispersed in the solution and reduces conventionally and accumulates on the heating surface by the nucleation of the vapour bubbles and creates a well-ordered graphene network. Carbon nanorods and recently used SiC nanorods have been shown to be vertically aligned in situ. 56 With this method, the thermal conductivity in the TIM application is increased from 4 W m −1 K −1 to 17 W m −1 K −1 . Hybrid graphene structures or CNT assemblies with increased CNT arrangements on graphene film can be employed for effective thermal degeneracy or new TIM assemblies 57 (Table I).
Modelling of Heat Transfer in Graphene
As the most abundant material, carbon is frequently used in fields such as electrical equipment and energy storage systems, as graphite is considered as the basic brick material of graphene modelling. 64 The advanced graphene modelling techniques diminish the conventional methods and computing theory density of binding models which includes structures for quantum description, electronic properties with their by-products and chemical mixtures which are inadequate to zero temperature range and they encompass in small dimensions in consumption. 65 The late emerging materials such as fullerenes and graphene-based materials and the nanophysics assemblies are considered with their thermal properties of graphene by utilizing molecular kinetic methods. The molecular subtleness is a scientific approach method of atomic motion based on a non-biased clarification of energy domain and the intersonic powers designates the foremost heat transporters. 66 The thermal regulation of nanodevices is achieved efficaciously by accurate amalgamation of graphene-based materials and alloys and their thermal properties are studied when they are in direct contact with the surface and at the same time when contact is active in nature. 67
Modelling-Based Optimization of Thermal Organization with Graphene-Improved PCM
It provides useful knowledge for computer modelling materials and device optimization of battery pack thermal control systems. We have described our experimental findings from a theoretical resolution as well-defined by the thermal dispersal comparison for temporary heat transmission in the parametric data of solid and transparent battery-operated models. 68 The limitations, laterally with the restrained factual temperature and thermal conductivity properties of the adhesive compounds, were utilized in the simulation of conductive heat fluxes in Lithium-ion battery packs using the COMSOL processor simulation package kit. 69 The built-in three-dimensional (3D) model leads to the study of sixcylinder battery-operated packs in dissimilar media. Six batteries, 18.4 mm in diameter, are consistently disseminated in a solid, 70 mm in diameter, all of which are housed within a 1 mm thick aluminium covering. 70 Owing to the basic nature of the battery pack, an additional dense mesh of free tetrahedral was utilized for stable cylinders with a heat transfer intermediate linked to the rigid cylinders and the aluminium sheath. 71 The temporary heat transmission comparison has been resolved to regulate the temperature increase within the external battery pack. In all simulations, only the physical properties of the intermediate run change the interplanetary between the battery cylinders. 68 Various media including air, traditional PCM paraffin and graphene-based materials are enhanced with PCM. For traditional PCM paraffin oil deprived of graphene structure, the usage of thermal conductivity at K = 0.25 W m −1 K −1 , mass density of about 900 kg m −3 and thermal efficiency ranging at 2500 J kg −1 K −1 . The simulation effects provide continuous temperature and time measurement for a single point within a 3D modelling battery-operated packet at any given period. 72 The exact positions of the statistics were evaluated according to the corporal location of the thermo column in the empiric test rig. These simulation findings were compared to the empiric evidence of thermocouples. 73 The maximum temperature ranges of the cylinder decrease to 320 K with the traditional paraffin PCM medium. The best efficiency of these battery cylinders, however, is the amalgam graphene PCM material medium heat control with a determined temperature of about 310-315 K. is the more stable temperature outline. 74 It must be remembered here that even at small graphene loading fractions, a major reduction in the temperature of the battery-operated cylinder is accomplished. 75
Molecular Dynamics Mimics the Interfacial Thermal Resistance Associated with Graphene
Considering and managing interfacial heat resistance with certain graphene structures and graphene layers is critical for the creation and display of FLG-based graphene layer films in heat organization. Many factors control the interfacial thermal resistance of graphene, which is dependent on the number of layers, the surface properties, and the interfacial connection. 76 Figure 6 shows the tensile test performance of graphene using the FLG model. FLG was prepared by applying graphene layers by maintaining a distance of 3.4 Å between them. In this section, we discuss the development of graphene-based interfacial thermal resistance in molecular dynamics, with the goal of delivering a good representation of how these main variables influence the thermal properties of the interface. 77
Influence of Number of Graphene Layers
The number of graphene layers has a direct influence on the internal layer thermal resistance of some graphene layers as well as the resistance between the FLG layers and the substrate. 78 Owing to the investigational imprecision, it is problematic to calculate the numeral persuaded layer heterogeneity of heat resistance unwavering from traditional tentative methods. Providentially, supercomputer simulations such as molecular subtleties offer a precise and quick approach to address these challenges. 79
Inter-layer Thermal Resistance Depending on the Number of Layers in Graphene
Using non-equilibrium molecular dynamics (NEMD) simulation software, Wei et al. measured the layered based interfacial thermal resistance between binary coupled graphene layers where R inter thermal resistance was attained. 81 L is the depth of the layer (0.34 nm), N is the number of layers and K c is the operative cross-plane thermal conductivity of multilayer graphene. The rate of the interfacial thermal resistance for N = 48 is an order of magnitude less than N = 6. It is a hot and cold reservoir that limits the average free path of the phonon, which induces intercontinental thermal resistance and simply decreases with layer number. 82 Ni et al. used a method based on the molecular dynamics of equilibrium (EMD) simulations to eliminate the effect of size in NEMD. The thermal resistance between two structures with a temperature difference between T can be determined using the following equation suggested by R int Volz et al.
If K B is the Boltzmann constant, then n 1 and n 2 apply to the number of degrees of freedom between the two subsystems. This procedure is non limited to heat sinks and can be used to measure the thermal resistance between two identical coatings of graphene. 83 Resistance measured in different ways confirms that as the amount of FLG increases, the resistance to interference decreases. Lastly, the interfacial resistance exceeds the graphite boundary with a large number of layers. 84
Coating Layer Number-Based Interfacial Heat Resistance Between Some Layered Graphene and One Surface
The heat system relationship between the graphene materials and the surface is characterized by the contact angles of van Der Waals interactions, while a significant number of studies have been performed on single-layer graphene (SLG) substrate and FLG-substrate interaction resistance. SLGsubstrate contact resistance is often believed to be independent of layer number because there is no association between resistance and coating number owing to investigational imprecision. 85 Though, molecular dynamic balance controls indicate that the touch resistance between FLG and SiO 2 reduces with the number of layers and congregates for six graphene structure layers, suggesting that the results are in reasonable arrangement with the investigational standards. 86 For FLG with one to four coating layers, with more than four layers, the calculation provides a variety of resistance owing to minimal untried precision, whereas the mean rate is 2 * 10 −8 m 2 KW −1 similar to 1.7 * 10 −8 m 2 KW −1 .
Outcome of Intraocular Coupling and Surface Properties
Intracellular thermal resistance between the graphene or FLG graphene layer and substrate is a major limitation of their thermal efficiency in devices. 87 The covalent molecule has been shown to successfully facilitate the flow of heat between the interfaces by adding additional heat pathways across the molecule. Research on the thermal efficiency of a graphene film is aided by the inclusion of silicone-functional molecules. 88 The calculations of their molecular subtleties demonstrate that the equivalent thermal conductivity and the thickness of the silane-based particles decrease. 89 Wang et al. stated that interglacial heat transfer engineering can be achieved by interaction with the host organ, such as the organ that governs interfacial thermal conductivity, and by electronically isolating the assisted graphene layer. 90 Luo et al. used NEMD constant-state to measure the graphs interfacial thermal resistance of graphene material or graphite-polymer systems. Graphene originates that longer wavelength photons in graphene play an important role in the flow of heat through the polymer interface. 91 The presence of lower-frequency methods of about 2-16 Hz owing to vibrations beyond the aircraft makes for a good mixture of graphene ranges and polymer spectra, thereby enabling interfacial heat transfer. 92 Using molecular dynamics (MD) simulations, Chen et al. reported that FLG thermal conductivity increases exponentially with the thickness of the sheet, exceeding 90% of the bulk graphite cap at six coatings and finally at 13.4 nm (40 layers). With the NEMD simulation, Li et al. analysed the crystallization consequence of the silicon carbide (SiC) surface and observed that the amorphous interface was less than the Rint sparkling interface. 93 This effect is due to the uneven external characteristic of the amorphous artifacts and the large photon channel unlocked by the soft photon thickness of the amorphous SiC states. 94 However, Zhang et al. reported that the thermal dissipation performance of a 2D junction transistor is strongly dependent on the heat conductivity of the surface and that it is problematic to decrease the temperature of the hotspot by changing the resistance of the 2D material substrate boundary. 95,96 In addition to the reasons described above, the thermal properties of graphene materials are often influenced by other methods, such as strain engineering and the manufacture of phonetic crystal structures. 97
Consequence of Doping on Graphene Thermal Performance
Doping is excellent method to expose a large and tidy doping tune. The doping elements boron (B) and nitrogen (N) atoms are common contenders for doping in graphene, the equivalent molecular size of carbon (C) and their pore receptors, and the position of the electron donor in computing B-and N-doping, correspondingly. 98 However, doping influences the heat efficiency of graphene. Shi et al. studied the interfacial thermal tolerance and thermal enhancement of nitrogen-doped zigzag graphene (NDZG) using NEMD. Intercellular thermal resistance in place of nitrogen doping has been shown to induce a sharp decrease in heat conduction of NDZG. 99 Thermal correction of trivalent single-nitrogen doped graphene (SNDG) with increasing temperature. Defects of varying relations on the thermal conductivity of graphene and the effects of isotopic doping were studied by Chen et al., in which spectral photon refers to the lessening period and the standardized accumulated heat conduction of the free path photon, suggesting that extended mean free path (MFP) phonon modes are firmly repressed in deficient and incapacitated graphene, leading to repressed volume dependency and weakened temperature dependency of heat conductivity relative to ancient graphene materials. 100,101 Graphene Works to Increase
Heat-Dissipating Performance
Graphene in plain thermal transference can be improved by increasing the biochemical bond between the graphene and the surface. 102 The thermal control of the micro-heaters is thus significantly enhanced by the incorporation of alternate heat-dispersed channels by aminosilane molecules that are coated with functionalized graphene oxide (FGO) attached to the graphene film assembly (GF). 103 The GF is bound to the surface of the FGO by aminosilane molecules such as (3-aminopropyl)triethoxysilane (APTES) which contains the three-O-group and has the NH 2 end due to the general chemistry of APTES, which readily binds to two dissimilar surfaces. 104 The Si-O termination of the APTES is attached to the GO surface and the cross-linked Si-O assembly serves as a solid supplementary coating between the surface and the GF with the azanide (NH 2 ) end of APTES binds to the carboxyl groups of the functionalized graphene film. 105 The microscopic cause of improvement of thermal conductivity (k) in graphene film was examined by mode-based analysis of the photon reduction time through the addition of the APTES molecule, and the reducing period of the acoustic flexural mode plane mode (ZA) is greatly increased at lower frequency curves; however, the longitudinal and transverse modes are marginally diminished. 106 ZA mode refers to the increase of k of the graphene layer film bound to the surface of ZA mode when it contributes more than 77% to 300 K. Molecular dynamic simulations and abiotic measurements demonstrate that there is an error in the function. 107 Figure 7 shows the graphene film applications as a heat spreader. The high thermal conductivity allows graphene film (GF) to disperse heat efficiently from heat sources. Most importantly, the comfortable, lightweight and compact features promote the application of GFs in comfortable light-emitting devices. Shielding composites have great application prospects in lightweight, electromagnetic composites, portable and wearable electronic devices due to their excellent EMI shielding performance and thermal stability, as well as enhanced thermal conductivities. 108,109 Cross-plane scattering of low-frequency phonons, which increases the convection of the graphene layer film bond plane by overcoming the longer flexural photon life. 110 These findings offer proof that the graphene layer film dropped on the FGO surface offers an extremely desirable medium for heat and energy control progress. 111
Thermal Modelling
Different techniques are used for thermal modelling, in which each process yields different results, often with common materials. It also explains the core and important role of graphene materials and other 2D materials for practical use in micro thickness measurements. 113 Here is a description of the state-of-the-art thermal measurement procedures used with graphene structures and other 2D graphene materials.
Thermal Bridge
The time period of heat conduction of nanotubes or nanowires is restrained in the arrangement of palette or packages using one heater and dual sensors method and the restrained thermal conductivity is not inherent owing to the conquered inter-tube sprinklings. 114 Multilayer electron-beam lithography was used to create micro-devices appropriate for heat calculation, deuce suspended microelectromechanical systems (MEMS) were developed and used to deactivate lowerdimensional patterns and detect temperature changes at microscale. 115 Done for the submission. This thermal bridge process is considered to be one of the most effective methods for calculating heat conduction and thermal power of carbon nanotubes, nanowires and 2D graphene materials. 116 Platinum or gold resistive coil is prefabricated on top of separate SiNx membrane and beam, which serve as both heater (R h ) and temperature sensor which is dependent on the track of the distribution of heat. The suspended singlelayer graphene heats the two layers and electrically separates them from the resistive coil. 117 The unit should be put in a room equipped with vacuum pump to provide pressure environment greater than 1 × 10 −5 pa to remove circumstantial gas convection and thermal radiation between the layers. AC heater with a DC voltage bias is applied up to current usually from 100 nA. Any frequency a few thousand hertz frequencies and additional AC current sensor of the identical amplitude or frequency. 118 The DC current (I) is used to add microwatt heat to the heater and to increase its temperature (T h ) and AC current is used to test the resistance of R h /R s , which is T h /T s . Suitable for T s later resistive coils can be used as temperature controllers. At a stable state, the heat conductivity of the film these and the suspended SiNx beams can be obtained from σ l = (Q h + Q l )/(ΔT h + ΔT s ) and σ s = ΔT s σ l / (ΔT h − ΔT s ), where Q h , Q l , ΔT h , and ΔT s and the heating tee heater connected to the above-mentioned heating capacity, the heating energy on the SiNx beam, Rh and the temperature rise. 119 The thermal conductivity of the film can be obtained by K = L × σ s /S = L/(R × S) where L, S and R are the sample volume, cross-section area and whole thermal resistance. The thermal bridge approach is widely used to calculate the thermal conductivity of suspended single-layer or multilayer 2D materials, including graphene, boron nitride, black phosphorus, transition-metal dichalcogenides, and so on, but there are numerous challenges. At higher temperature, the thermal stability is beneficial to the film surface and the film has excellent thermal stability which can help them perform better by preventing thermal decomposition. Thermally conductive composite films have excellent thermal properties and can be used in a high-temperature setting for an extended period of time. 120,121 For example, graphene and MoS 2 thermal conductivity works spontaneously at times, retaining internal thermal conductivity with hot debate and disagreement. 122
EB Self-heating Method
The electron beam (EB) self-heating system delivers straight dimension of heat interaction resistance and hence of internal thermal conductivity. This procedure is updated on the basis of the current bond technique and the model is combined within the scanning electron microscope (SEM) compartment which improves the quantity. 123 In contrast to the thermal bridge process, Rh heat foundation acts as a temperature instrument, the binary layers are used as a temperature device and the electron stream of light is used as a heat source. 124 Figure 8 shows the cogeneration technique of energy storage systems. Electrical energy is produced by combined heat and power (CHP) plants, wind farms and other plants and thermal control demand are maintained by CHP plants. For calculations, the electron beam is perused laterally towards the extent of the model, throughout which time the energy storage system of the electrons is consumed and hence the resident space is heated. 125 The resident heat produced by the engrossed electrons streams to the dual layers and increases their temperature range. The calculated current resistance is applied from one layer to the heating area and where reflection loss (RL) is equal to heat resistance of the auxiliary beam, X is the detachment from the layer to H.
Pulsed Photothermal Reflectance
In the experimental setup of pulsed photosensitive thermal reflectance (PPR), the thermal conductivity (bulk or film) of the material is described by a thermal reflection technique in the nanosecond regime. Kiding et al. reported the original method using PPR technologies to test the thermal conductivity of SiO 2 thin films. 127 Figure 9 shows the photothermal reflectance coating measurement. To increase the heat absorption, the gold layer evaporates above the surface. In this method, the sample surface is kept at the first and the surface temperature of the sample increases rapidly and then relaxes over time. This leads to a temperature excursion profile, which depends on the thermal properties of the underlying films and the thermal resistance between the layers. The technology is an optical and non-contact system containing a laser impel and a laser probe. 128 Prior to the calculation of thermal conductivity, the metallic layer, ideally gold, is dropped on the top of the film surface with the intention of capturing temperature captivation and external temperature through thermal reflective technology. 129
Resistance Dependence on Temperature
The deferred model is impassioned by applying an alternating current (AC) and a direct current (DC) over the film and the sample temperature is determined by the resistance of the sample temperature. In this circumstance, the constant of resistance of the film is also restrained. 131 The flick is suspended between two copper heat sinks and mounted in a void cavity, and the electrode is located in a four-point discrepancy conformation. The sample length with a width ratio of less than 0.1 can be sliced into bits. 132 The heat transfer with GF can also be regarded as a 1D heat transfer. In the AC system, a current frequency of RMS = 1 mA at 2000 Hz is added to the DC current model, which increases its temperature by heating. Then the tolerance of the sample to various DC values is sample (DC) = blank, obtained from rms where RMS is the primary sympathetic voltage drop and transversely the sample as restrained by the lock internal amplifier. 133 Table 2 presents the thermal measurement methods. Throughout the self-heating process of the sample, the modification in electric resistance is restrained and the overall heat increase is determined. 134 The heat conduction of the film is then achieved by means of the added Joule heating control and the resulting typical heat increase as the input of the average heat conduction case of a 3D graphene spray structure. 135,136 Joule Heat Nanomaterials such as graphene and CNT are temperaturedependent on the Joule heating effect. In this process, the suspended graphene film is active by DC application and the heat increase of the model is estimated by infrared radiation (IR). Heating setup and sketch of models for Joule throughout the dimension, the sample is deferred and connected to both electrodes. 137 The DC current is powered by the solid and the heat of the material varies evenly. The DC current is calibrated to the target material with a temperature increase of less than 10 °C in order to avoid radiation and convection losses. The volumetric heat output is uniform due to Joule heating, Q = VI/(Lwt), where W, T and L are the distance, depth and measurement of the model, and V, I are voltage and current, correspondingly. 138 Due to the disparity in the heat volume and the heat sink of the target material, the heat travels laterally towards the film and reaches a stable state within a few microseconds after heating. Both tests were done in an airtight laboratory atmosphere to prevent the influence of air flow. 139 The temperature of the heat sink during calculation is held at a steady temperature of T L/2 = T −L/2 = T o . The temperature profile of the goaled solid is proportionally disseminated with deference to the core of the target solid is disseminated correspondingly with respect to the centre of the model. 140 Figure 10 explains the Joule heating method around the CNT structures. Since the electrons mainly pass inside the polymer matrix via CNT, the Joule heating process takes place around CNTs initially, and it occurs gradually over the area of high temperature around the CNTs and in the whole composite.
The highest temperature was reached in the centre of the T M sample. In view of the temperature at the advantage (T 0 ) and midpoint (T M ) of the sample, the thermal conductivity of the target material is described by the subsequent equation: where K is the thermal conductivity of the target material in the level. Thermal conductivity can be determined by calculating the temperature at the control and middle of the specimen with an IR meter, the voltage around the sample and the contemporary and measurement of the sample. 142 For example, Wang et al. used the Joule heating technique to calculate the thermal conductivity of better heat-conducting graphene adhesives, and the thermal conductivity of conductive adhesives with 3 wt.% graphene was 8 W m −1 K −1 . This is roughly four times the position conductive epoxy resin without graphene, with improved graphene picture and the temperature profile and conductive adhesive pattern. 143
Laser Flash
Laser flash technology is a commonly used tool for calculating thermal deviations and thermal conductivity of constituents owing to certain essential characteristics such as great accuracy and duplicability, low dimensions and a well-defined test area. 144 Laser flash analysis defines thermal conductivity of aircraft by means of millimetre-thin aircraft 116 In its simplest adiabatic conformation, the thermal diffraction thickness of the film is related to the thickness of L and the maximum temperature increase of t50 is half the time. More complex models allow for correction due to small pulse duration and heat loss. 145 As long as the properties of all layers are understood in advance, it is also possible to calculate multilayer structures with contact resistance between three or two layers. Figure 11 describes the measurement of heat conduction using laser flash method. The laser-flash sample heats one of the surfaces with a small pulse, and the infrared detector measures the temperature rise over the other surface over time.
From the temperature lifetime curve, the thermal diffusivity is calculated. 146 Such an uncontacted laser flash system has the benefit of eradicating the influence of connection heat resistance and is closely related to the high heat properties of the material. 147,150 In accumulation, the period determination of the participation optical maser and indicator has a durable outcome on the precision of the test outcomes. 148,151
Future Applications of Graphene Materials
Recently, some of the applications of graphene materials have been established for the use of graphene-based materials and other associated 2D materials for thermal management in consumer products. Graphene structures have been utilized to coat light-emitting diode (LED) strands, and graphene is reported to help dissipate heat in LED bulbs, thereby prolonging the generation and strengthening the performance of LEDs. Graphene-based materials of various thicknesses have been applied as heat spreader, with thermal conductivity of 3200 W m −1 K −1 . In addition, precipitously oriented graphene-based TIM materials can achieve thermal conductivity up to 1000 W m −1 K −1 direction of Z.
Graphene-based materials have also been used as thermal paste, with thermal conductivity greater than 10 W m −1 K −1 reported. It seems that graphene can efficiently lower the temperature of the mounted film vision. In order to further reduce the temperature, the efficiency of graphene film is required.
Conclusion
High-quality graphene materials and 2D materials play a major role in industrial applications. They have thus garnered considerable interest and led to significant advancements in the manufacture of graphene-related material for numerous applications in the science and engineering fields over the past few decades. The exceptional properties of graphene materials, especially for thermal management and energy storage control systems, have paved the way for thermal management applications in the scientific industries. This review article summarizes various graphene structures and graphene-based materials from science and engineering with a specific focus on thermal control systems due to their excellent structural and management properties. We are hopeful and confident that this will inspire further systematic studies and at the same time advance more marketable applications in this field, including materials such as alloy fillers, as well as the application of graphene polymer as heat repellents. This active technology has paved the way for the implementation of 2D thermal control materials in industry. It is our heartfelt expectation that this will encourage further research experiments while at the same time creating more commercial applications in this field.
Outlook
A large number of research articles and reviews report on the development of energy storage materials, and still many improvements are required. The literature study showed that nanostructure materials in two dimensions were produced in enormous amounts and that very work on three-dimensional materials. Future work should therefore focus on the development of such materials and slight changes to the structure that may improve the electrochemical characteristics. The materials with an excellent surface area and conductivity increased the specific capacity, rate capacity of the material enormously, and it must therefore be improved. Recently, a great deal of attention was paid to the scientist and technologist for new materials such as the metal-organic/inorganic frameworks. However, very few articles on metal-organic framework are being published and need more attention in the future. In general, the development of these materials will certainly improve the storage and stability, which are strongly required. | 10,617.6 | 2021-07-12T00:00:00.000 | [
"Materials Science",
"Physics",
"Engineering"
] |
An Analysis of the Factors Influencing the Value Promotion of Chinese E-Commerce Enterprises Based on Amos Model
Based on the data of 148 e-commerce enterprises in China, Amos in the structural equation model is used as the analysis tool to build the verification model of e-commerce enterprise value promotion. 12 observation variables, such as the proportion of R & D expenditure to sales revenue and the proportion of R & D expenditure to sales revenue, are selected to explore innovation competitiveness, scale competitiveness, difference of competitive power has an impact on the promotion of enterprise value. This paper puts forward the suggestions of strengthening supply chain cost control, diversification strategy and cooperation and sharing, which is of practical significance to improve the core value of e-commerce enterprises.
I. INTRODUCTION
With smart phones as the interface, the world has entered the era of big data. Big data not only changes people's lifestyle, but also changes the development mode of enterprises. Under the severe impact of the electric business enterprises, enterprises relying solely on offline sales have been unable to maintain normal operation, and have turned to the development mode of OTO integration. Since 2017, China's electricity supplier industry has been developing steadily. Domestic policies have further fallen into the transformation effect of "Internet +" strategy in all walks of life, and supply side structural reform and consumption upgrading have played a supporting role in promoting the steady development of the electricity supplier mode. The e-commerce industry has entered the rational adjustment period from the extensive and rapid development stage. Enterprises pay more attention to fine operation and regional development, and gradually improve their service capacity in the whole industry chain. At the end of 2018, the Ministry of Industry and Information Technology issued a 5G system test frequency license, and the Political Bureau of the CPC Central Committee issued top-level plans and designs for the development and utilization of the new generation of artificial intelligence in China. E-commerce enterprises will play a more powerful role in the economy. The industry will play an increasingly important role in the capital market. Enterprises will accelerate the development of the Internet of Things, logistics, electronic payment and distribution of supply chain finance. In order to improve the overall profitability of e-commerce enterprises, it is necessary to Manuscript received April 17, 2020; revised July 4, 2020. Lihua Xia is with Taizhou University, China (e-mail: 261488796@qq.com). manage and control the various elements in order to create greater value. In the face of such a development trend, how e-commerce enterprises can develop and grow in the increasingly fierce competition, and how to achieve value enhancement in the industry has become a very urgent issue.
Shuan Liu (2010) pointed out that analytic hierarchy process (AHP), Balanced Scorecard (BSC) and data envelopment analysis (DEA) are often used to evaluate the performance of e-commerce enterprises. The difficulties encountered in the performance evaluation of e-commerce enterprises are mainly manifested in the data analysis method, and other losses will offset the profits obtained by e-commerce enterprises [1]. Caihua Liu et al. (2012) used the Balanced Scorecard to design a set of performance evaluation index system for e-commerce enterprises, including financial indicators, market share, customer retention rate and other non-financial indicators, and constructed an analytic hierarchy process model to evaluate the performance of e-commerce enterprises [2]. Linqi Wang (2014) believes that the core values of e-commerce enterprises include eight aspects: network operation mode, marketing solutions to meet customer needs, building their own brand, commodity types, website technology, logistics and distribution, striving to build an information platform, and improving service level [3]. The above-mentioned evaluation of e-commerce enterprise value is mostly based on future cash flow discount model or user contribution model. However, the factors such as R& D innovation, risk competition and profit model have not been paid attention to, which is more suitable for the development of mature e-commerce enterprises, but there is no stable revenue and profit in the initial and growth stages [4]. Cash flow enterprises are not adapted. According to the particularity of e-commerce enterprises and the influencing factors of enterprise value, a new value evaluation and analysis model is constructed.
A. The Basic Principle of Structural Equation Modeling
Structural equation model (SEM for short) originated from path analysis and factor analysis. It is a method of establishing and testing causality model. In 1970s, Keesing (1972) and Wiley (1973) Joreskog (1973) integrated factor analysis technology into path analysis, and put forward a general model that marked the emergence of structural equation model [5]. It can be used to analyze the effect of individual indicators on the whole and the relationship between individual indicators, instead of multiple regression and covariance analysis. The advantage of structural equation Li-Hua Xia analysis over traditional analysis is that it can deal with multiple dependent variables at the same time. It can test whether the relationship between variables has changed and whether the mean difference of each factor is significant.
B. Construction of Structural Equation Modeling
There are two basic models in the equation model: measurement model and structural model. The measurement model consists of manifest indicators and latent variables, and is a linear function of observation variables. Manifest indicators, also known as observation variables, refer to the variables that can be measured by a specific method to obtain specific values. Latent variables, as opposed to explicit variables, cannot be measured or observed to obtain specific values, which need to be reflected by measured indicators. Structural model is a combination of factor analysis and path analysis. It mainly studies the relationship between manifest indicators and latent variables, as well as the relationship between latent variable [6]. In the SEM analysis model, the regression relationship of the measurement model is confirmatory factor analysis, which explores the causal relationship between latent variable and observation variables. The regression relationship of the structural model is equivalent to the traditional path analysis, which directly explores the causal relationship between latent variable.
The regression equation of the measurement model is as follows:
Xn=λnζn+δn Yn=λnηn+εn
The regression equation can be expressed by matrix equation as follows: X=Λxζ+δ Y=Λyη+ε where ζ is not correlated with δ, η and ε, and vice versa. Λx and Λy as indicator variables (X、 Y) , the factors of load, the delta, δ and ε measurement error of observation variable, ζ is manifest indicators, η is inside Latent variables. In the SEM measurement model, it is assumed that there can be no covariation or causal path between latent variable and measurement errors.
At present, the software that can deal with SEM includes AMOS, LISREL, EQS, Mplus, R. Amos (Analysis of Moment Structures) moment structure analysis, combined with linear model and common analysis technology, has similar structure analysis connotation as covariance matrix, estimates the initial model diagram, verifies the adaptation effect, and outputs the optimal adaptation model after modifying according to the revised index [7].
A. Sample and Variable Selection
Referring to the industry classification guidelines, this paper selected the 2018 data of six typical types of Listed Companies in the e-commerce industry of Shanghai and Shenzhen Stock Exchanges as samples, including 16 e-commerce companies, 25 interconnected finance companies, 32 e-payment companies, 75 Internet of Things companies, and 148 companies which removed the cross-duplication of business and were processed by ST. The data in this paper are collected from the research databases of Cathay Guotaian and RESSET series database. In order to better evaluate the company's value, we preliminarily select indicators and conduct exploratory factor analysis on the data, as shown in Table I. Sig. .000 The degree of sphericity P value is significantly less than 0.05, and the KMO value is 0.782. The correlation coefficients between variables and unit arrays are significantly different, which is suitable for factor analysis. Choose orthogonal rotation, according to the result of cumulative explanatory rate of factor variance, extract the cumulative value of square sum loading 79.35%, extract most information of the three principal components which can cover the original variables, select their components as exogenous latent variables, named innovation competitiveness, scale competitiveness, risk competitiveness and operational competitiveness respectively. The relationship between the latent variables and their corresponding observed variables is shown in Table II. B. Construction of Initial Model AMOS 24.0 software was used to construct a model for validation analysis and to explore the relationship between latent variables and observed variables. According to the set latent variables and corresponding observation variables, the initial model path map of Internet enterprise value analysis structural equation is constructed, as shown in Fig. 1. Because of the different data dimensions, in SPSS 19.0, the operation command of "Analysis -Description Statistics -Description" is executed, and the standardized values are saved as variables to reduce errors. The most commonly used method of SEM estimation is maximum likelihood estimation, which is used to estimate various parameters in the model. Sample data conform to the assumption of multivariate normality. The operation calculation is checked, and the standardized output parameters of the model are between 0 and 1 except the reverse index asset-liability ratio, which is in line with the actual operation. But the ratio of absolute fitness index CMIN/DF=7.517 (normal range is less than 3), mean error square root RMSEA=0.18 (better fit than 0.08), benign fitness index GFI=0.719, NFI=0.684 (better fit than 0.9), P value is 0.023, less than 0.05, rejecting the assumption of initial model setting, need to be revised according to parameter hints [8]. Table III is the first validation result of the model. At the significant level of p=0.05, only the path coefficients between innovation risk competitiveness and liquidity ratio are not significant, while the path coefficients between other latent variables, observation variables and observation variables are significant. Note: *** means P < 0.001, the same as below Looking at the "Modification Indices" option in the output results, through the covariance correction index table (see Table IV), it can be concluded that e4 and e12 are error variables of the observation variables "operating income" and "total assets turnover ". In the initial model, it is assumed that there is no correlation between them. From the calculation formula, there is a covariant relationship between them, if there is a correlation between them. In this system, if the error term between the observed variables is relaxed from the strictly unrelated assumption, the chi-square value can be reduced by at least 76.458, and the parameter estimation will be changed by 0.432. The two variables are connected by double arrows. According to the M.I. index value provided by AMOS, each time a parameter is released, a new path is added to the model. After each modification, the revised index of error variables E1 and e10, E4 and e12, E5 and E12 is higher. Corresponding observation variables show that the total R&D expenditure and income growth rate, business income and total asset turnover rate, total assets and total capital are higher. There is an inevitable relationship between the production turnover rate and the symbiotic relationship. We can relax the error terms of these observation variables in turn, which will significantly improve the fitting effect of the model [9]. Therefore, these pairs of error variables are connected by double arrows in the structural equation path diagram. According to the output results, the overall fitting situation is sorted out, as shown in Table V.
C. Fitting and Modifying Model
Journal of Economics, Business and Management, Vol. 8, No. 3, August 2020 From the above table, it can be concluded that after model modification, the fitting index can reach or surpass the fitting standard, reflecting that the fitting between model and data is good. The modified model can fit the sample data and meet the absolute adaptability requirements. The constructed model is in good agreement with the actual situation, and the path analysis results are reliable. The revised model path is shown in Fig. 2. There are 28 variables in the structural equation model, including 12 observation variables, 16 latent variable, 12 internal variables, 16 external variables (including 4 latent variable plus 12 error variables), and the index variables of the measurement model are observation variables. The number of unique sample moment elements is the number of sample data points, the value is k (k + 1) / 2 = 78, where k is the number of observed variables of CFA model. In the model, the chi-square value is 33.547, and the significance probability value is p=0.250>0.05. The null hypothesis is accepted. The hypothesis that the variance covariance S matrix derived from the observation data is equal to the variance covariance matrix derived from the hypothesis model is supported, i.e. the hypothesis that the model map is suitable for the observation data. In the estimation of measurement model, the parameter of λ certain index variable is fixed to 1 between the latent variable and the path coefficient of its observation variable. As to which index variable is fixed, it has no relation. Because after testing with specific data, the value-type maps of standardized estimation of measurement model are the same and the overall fitness statistics are the same.
IV. RESEARCH CONCLUSION
Empirical research shows that there is a positive causal relationship between endogenous latent variables innovation competitiveness, scale competitiveness, risk competitiveness, operational competitiveness and the path coefficient of core value of e-commerce enterprises. Innovation and scale competitiveness are important factors affecting core competitiveness at the current stage. In contrast, risk competitiveness and operational competitiveness have a weak impact, and companies can give priority to them. Enhance the scale and innovative technology, so as to enhance its value and core competitiveness faster.
The overall effect of R&D expenditure and total assets on the standardization of core value is at the highest level. Among them, the explanations of total assets on scale competitiveness are stronger, and the explanations of total R&D expenditure on innovation ability are stronger (the value of λ is greater than 0.89). Consistent with the current situation that e-commerce enterprises attach importance to R&D expenditure to enhance core value, the total R& D expenditure of leading enterprises in the industry and the proportion of R& D expenditure to business income are increasing year by year; the asset-liability ratio has a strong explanation of risk ability, and the influence of liquidity ratio and net cash flow on risk ability is weak, mainly due to the assets of e-commerce industry. The characteristics are light assets type and high liquidity ratio. Because of the favor of investors, the cash flow of enterprises in the industry is generally higher than that of other industries, which belongs to the industry commonness and cannot constitute a specific factor to explain the value of enterprises. The explanatory power of earnings per share to operational competitiveness is the strongest, and the corresponding explanatory power is close to 0.9. The explanatory power of return on net assets to operational competitiveness is slightly weaker than that of earnings per share (value of λ is close to 0.7); The explanatory power of growth rate of operating income and turnover rate of total assets to operational competitiveness is weaker, because many e-commerce enterprises adopt the mode of "burning money" in the initial stage. The growth of business income, especially in the initial stage, has some difficulties, which cannot be well used to explain the improvement of the core competitiveness of enterprises; the turnover rate of total assets is reflected by the ratio between business income and average total assets, while the turnover rate of business income and total assets has well explained the value of enterprises in terms of scale competitiveness instead of total assets turnover rate. If the company data is incomplete or lagged, the observation variables with high explanatory degree can be selected to measure.
Different from traditional industries and general understanding, the impact of operating capacity on the core competitiveness of Listed Companies in China's e-commerce industry is not the most significant. There is no significant correlation between the size of Listed Companies in e-commerce industry and their innovation risk ability, innovation and operation ability, scale and operation ability. It shows that there is a lag between the expansion of the company's scale and the profit and capital turnover movement. At the same time, the rapid development of the e-commerce industry, the long profit cycle and the high cash flow characteristics lead to the relationship between profit and risk is not as close as in the traditional industry.
V. COUNTERMEASURE AND ENLIGHTENMENT
A. Enhancing Scale Competitiveness E-commerce industry follows the law of increasing scale value; network externalities will have a positive impact on the value of enterprises. The effect of increasing scale returns of e-commerce enterprises is more obvious than other industries, because the main products of the Internet industry generally have the characteristics of extensive cost-free replication. With the expansion of the scale, the average cost is decreasing, the revenue is increasing, and the price of enterprises is increasing. Value will be increased, and at the same time, more users will be encouraged to enter, forming a cycle of increasing law. With the continuous development of the Internet, it has brought infiltration among various industries, cross-links of industrial relations and improvement of user stickiness. More and more enterprises are aware of the great value of common user groups among different kinds of websites. This index will promote the integration of e-commerce enterprises and provide a new reference dimension for enterprises to analyze user behavior more accurately. Sharing logistics, data and purchasing channels, in order to reduce the procurement costs and logistics costs of both sides, improve their service quality and level, and then enhance their own development capacity, achieve mutual benefit and win-win. This is in line with the conclusion drawn from the empirical study. Scale competitiveness is an important factor to enhance the value of e-commerce industry [10].
B. Creating a Characteristic Profit Model
Although most e-commerce enterprises adopt the operation mode of "burning money" when they start up, they still need to consider the distinctive profit points when they start up companies, so as to gain profits as soon as possible and achieve higher value. Through the combination of long-term strategic planning and short-term business plan to adapt to the external environment, enterprises and users will win-win situation. At present, the development direction of many e-commerce platforms is basically determined, but in the way of profit, it has been difficult to balance the relationship between profit and service. With the increase of the scale of e-commerce enterprises, there is only a problem of transaction function in the consumption terminal of e-commerce websites. Faced with an increasingly wide range of groups, the value of users lies in creating profits for enterprises while consuming, while enterprises strive to make users pair with each other. Its products or services produce loyalty and dependence, so we need products with characteristics and service quality should be higher. Only by providing comprehensive and diversified services to enhance the consumer experience of users, can we achieve higher competitiveness. It can make full use of its greater media value to tap its profitability. Secondly, we should realize rational allocation of resources, share logistics system in physical stores, reduce procurement costs, multi-terminal consumption experience and value realization of multi-dimensional user flow, which will greatly reduce procurement costs and further improve the profitability of enterprises.
C. Continuously Improve the Level of Innovation Technology
The industry is facing the challenges of diverse, complex and rapid changes in technology and services. New applications, new channels and new business models brought by new technologies such as artificial intelligence, VR, AR and block chain bring new opportunities and challenges for the development of e-commerce industry. In order to adapt to these changes, enterprises should carry out in-depth research and experiments on new technologies and new businesses to continuously improve their technological innovation capabilities. Using big data to strengthen the cost control of supply chain, and using supply chain finance to solve the problem of liquidity turnover in e-commerce trade due to the existence of accounts receivable, inventory or advance payment. China's e-commerce has huge market potential, but there are still large security problems. The funds and information of online transactions can not be guaranteed. Enterprises should strengthen research and development investment in information security. Through external expansion and internal collaboration, we will further optimize and improve the big data industry ecosphere, promote the deep integration of big data, cloud services and artificial intelligence, and focus on exploring a reasonable and efficient mechanism for industrial co-development. | 4,762.8 | 2020-01-01T00:00:00.000 | [
"Business",
"Economics"
] |
Piezoelectric MEMS Acoustic Transducer with Electrically-Tunable Resonant Frequency
The paper presents a technique to obtain an electrically-tunable matching between the series and parallel resonant frequencies of a piezoelectric MEMS acoustic transducer to increase the effectiveness of acoustic emission/detection in voltage-mode driving and sensing. The piezoelectric MEMS transducer has been fabricated using the PiezoMUMPs technology, and it operates in a plate flexural mode exploiting a 6 mm × 6 mm doped silicon diaphragm with an aluminum nitride (AlN) piezoelectric layer deposited on top. The piezoelectric layer can be actuated by means of electrodes placed at the edges of the diaphragm above the AlN film. By applying an adjustable bias voltage Vb between two properly-connected electrodes and the doped silicon, the d31 mode in the AlN film has been exploited to electrically induce a planar static compressive or tensile stress in the diaphragm, depending on the sign of Vb, thus shifting its resonant frequency. The working principle has been first validated through an eigenfrequency analysis with an electrically induced prestress by means of 3D finite element modelling in COMSOL Multiphysics®. The first flexural mode of the unstressed diaphragm results at around 5.1 kHz. Then, the piezoelectric MEMS transducer has been experimentally tested in both receiver and transmitter modes. Experimental results have shown that the resonance can be electrically tuned in the range Vb = ±8 V with estimated tuning sensitivities of 8.7 ± 0.5 Hz/V and 7.8 ± 0.9 Hz/V in transmitter and receiver modes, respectively. A matching of the series and parallel resonant frequencies has been experimentally demonstrated in voltage-mode driving and sensing by applying Vb = 0 in transmission and Vb = −1.9 V in receiving, respectively, thereby obtaining the optimal acoustic emission and detection effectiveness at the same operating frequency.
Introduction
Acoustic transducers based on micro electro-mechanical systems (MEMS) represent a lively research field and, at the same time, provide a significant number of concrete solutions and commercial devices. Specifically, thanks to the advantages provided by MEMS technology such as compact sizes, low production costs and high compatibility with IC technology [1], acoustic MEMS transducers have been extensively employed in different applications. In biomedical fields they have been exploited to monitor heart and lungs sounds [2,3] and for cochlear implants [4]. In the livestock sector MEMS acoustic transducers have been used to estimate the state of health of animals [5] while in industrial fields they have been used for noise and vibration measurements [6], as resonant photoacoustic combustion gas monitors [7] or as hydrophones for pipeline leak detection [8]. In recent years MEMS acoustic sensors and actuators have been extensively produced for consumer applications as microphones for wearable devices [9,10] and voice controllable
Fabrication Technology and Device Design
The top view of the proposed piezoelectric MEMS device, taken from the graphic design system (GDS) file, is reported in Figure 1. The proposed 9 × 9 mm MEMS device exploits a 6 × 6 mm highly doped silicon diaphragm with an aluminum nitride (AlN) piezoelectric layer deposited on top that can be actuated by eight interdigital transducers (IDTs), each composed by two interlocking metal comb-shaped arrays of twenty equally spaced fingers. The IDTs are placed on the inner and outer edges of the diaphragm and disposed symmetrically with respect to its centre. The doped silicon layer can be electrically connected employing the four metal pads located at the die corners. The layout of the device, and specifically of the IDTs, has been designed to create a general-purpose piezoelectric MEMS platform exploitable in different applications. In particular, the IDTs have been exploited to generate Lamb waves in the diaphragm at frequencies in the megahertz range to drive mechanical vortexes in liquids for biological applications [36].
Micromachines 2022, 12, x FOR PEER REVIEW 3 of 18 disposed symmetrically with respect to its centre. The doped silicon layer can be electrically connected employing the four metal pads located at the die corners. The layout of the device, and specifically of the IDTs, has been designed to create a general-purpose piezoelectric MEMS platform exploitable in different applications. In particular, the IDTs have been exploited to generate Lamb waves in the diaphragm at frequencies in the megahertz range to drive mechanical vortexes in liquids for biological applications [36]. In this work, on the other hand, to induce a planar static compressive or tensile stress in the diaphragm and to excite/detect acoustic signals, the IDTs have been employed as top plates over the piezoelectric layer, while the highly doped silicon layer has been used as a common bottom plate, thus configuring the electrodes to produce in all respects parallel-plate transducers. Therefore, the IDTs layout, and specifically the spacing between two consecutively electrodes, does not affect the presently proposed application.
The piezoelectric MEMS has been manufactured by employing the piezoelectric multi-user MEMS processes (PiezoMUMPs) technology developed by the MEMSCAP foundry [37]. The manufacturing process steps are illustrated in Figure 2a-f. The process employs a 150 mm <100> oriented silicon-on-insulator (SOI) wafer where the silicon, the oxide and the silicon substrate have thicknesses of 10 ± 1 μm, 1 ± 0.05 μm and 400 ± 5 μm, and are shown in Figure 2 with red, black and blue colours, respectively. A bottom side oxide layer, shown in green colour, is also present on the starting substrate. The process begins with the doping step of the wafer reported in Figure 2a. This step involves the deposition of a phosphosilicate glass (PSG) layer, shown in purple colour, and its annealing at 1050 °C for 1 h in argon. The PSG layer is subsequently removed using wet chemical etching. The piezoelectric film lift-off occurs as the second step of the manufacturing process, reported in Figure 2b. The piezoelectric film consisting of 0.5 μm of aluminum nitride (AlN), shown in cyan colour, is deposited over the wafer by reactive sputtering. The third step involves the pad metal lift-off, reported in Figure 2c. A metal stack consisting of 20 nm of chrome (Cr) and 1000 nm of aluminum (Al), shown in grey colour, is deposited by beam evaporation. In this work, on the other hand, to induce a planar static compressive or tensile stress in the diaphragm and to excite/detect acoustic signals, the IDTs have been employed as top plates over the piezoelectric layer, while the highly doped silicon layer has been used as a common bottom plate, thus configuring the electrodes to produce in all respects parallel-plate transducers. Therefore, the IDTs layout, and specifically the spacing between two consecutively electrodes, does not affect the presently proposed application.
The piezoelectric MEMS has been manufactured by employing the piezoelectric multi-user MEMS processes (PiezoMUMPs) technology developed by the MEMSCAP foundry [37]. The manufacturing process steps are illustrated in Figure 2a-f. The process employs a 150 mm <100> oriented silicon-on-insulator (SOI) wafer where the silicon, the oxide and the silicon substrate have thicknesses of 10 ± 1 µm, 1 ± 0.05 µm and 400 ± 5 µm, and are shown in Figure 2 with red, black and blue colours, respectively. A bottom side oxide layer, shown in green colour, is also present on the starting substrate. The process begins with the doping step of the wafer reported in Figure 2a. This step involves the deposition of a phosphosilicate glass (PSG) layer, shown in purple colour, and its annealing at 1050 • C for 1 h in argon. The PSG layer is subsequently removed using wet chemical etching. The piezoelectric film lift-off occurs as the second step of the manufacturing process, reported in Figure 2b. The piezoelectric film consisting of 0.5 µm of aluminum nitride (AlN), shown in cyan colour, is deposited over the wafer by reactive sputtering. The third step involves the pad metal lift-off, reported in Figure 2c. A metal stack consisting of 20 nm of chrome (Cr) and 1000 nm of aluminum (Al), shown in grey colour, is deposited by beam evaporation. Micromachines 2022, 12, x FOR PEER REVIEW 4 of 18 A front side polyamide protection coat, shown in Figure 2d in orange colour, is then applied to the top surface of the wafer. The wafer is then reversed, and the substrate layer is lithographically patterned from the bottom side, as reported in Figure 2e. Reactive ion etching (RIE) is used to remove the bottom side oxide layer while a DeepRIE (DRIE) is subsequently used to etch the substrate layer up to the silicon layer. Finally, the front side protection material is stripped during the release step, as reported in Figure 2f. Top and bottom views of the fabricated piezoelectric MEMS device are reported in Figure 3a,b, respectively. The proposed device embeds electrical terminals for each comb-shaped arrays of fingers, while four electrically shorted metal pads are placed in each corner of the device to contact the highly doped silicon layer beneath the AlN piezoelectric layer, as reported in Figure 3c,d, respectively. Each comb-shaped array of the IDTs includes fingers with width of 28 μm and pitch of 112 μm, as reported in Figure 3c. Figure 3a,b, respectively. The proposed device embeds electrical terminals for each comb-shaped arrays of fingers, while four electrically shorted metal pads are placed in each corner of the device to contact the highly doped silicon layer beneath the AlN piezoelectric layer, as reported in Figure 3c,d, respectively. Each comb-shaped array of the IDTs includes fingers with width of 28 µm and pitch of 112 µm, as reported in Figure 3c. A front side polyamide protection coat, shown in Figure 2d in orange colour, is then applied to the top surface of the wafer. The wafer is then reversed, and the substrate layer is lithographically patterned from the bottom side, as reported in Figure 2e. Reactive ion etching (RIE) is used to remove the bottom side oxide layer while a DeepRIE (DRIE) is subsequently used to etch the substrate layer up to the silicon layer. Finally, the front side protection material is stripped during the release step, as reported in Figure 2f. Top and bottom views of the fabricated piezoelectric MEMS device are reported in Figure 3a,b, respectively. The proposed device embeds electrical terminals for each comb-shaped arrays of fingers, while four electrically shorted metal pads are placed in each corner of the device to contact the highly doped silicon layer beneath the AlN piezoelectric layer, as reported in Figure 3c,d, respectively. Each comb-shaped array of the IDTs includes fingers with width of 28 μm and pitch of 112 μm, as reported in Figure 3c. The possibility to obtain an adjustable matching between the series and parallel resonant frequencies of the first flexural mode of the piezoelectric MEMS transducer to increase the effectiveness of acoustic emission/detection has been investigated by electrically tuning the mechanical characteristics of the diaphragm. The displacement and the section view of a fully-clamped square plate vibrating at the first flexural mode is reported in Figure 4a,b, respectively.
Micromachines 2022, 12, x FOR PEER REVIEW 5 of 18 The possibility to obtain an adjustable matching between the series and parallel resonant frequencies of the first flexural mode of the piezoelectric MEMS transducer to increase the effectiveness of acoustic emission/detection has been investigated by electrically tuning the mechanical characteristics of the diaphragm. The displacement and the section view of a fully-clamped square plate vibrating at the first flexural mode is reported in Figure 4a,b, respectively. Two IDTs, namely IDT1 and IDT3, placed on the opposite inner edges of the diaphragm, shown in the simplified schematic of Figure 4c in blue colour, have been shorted and employed as a top plate over the piezoelectric layer.
The doped silicon layer, contactable by employing the metal pads shown in Figure 4c in dark red colour, has been grounded and employed as a bottom plate. The four IDTs on the outer edges of the diaphragm, i.e., IDT5-IDT8, shown in black colour, have not been employed and thus left unconnected. The remaining two IDTs, namely IDT2 and IDT4, on the inner edges of the diaphragm, shown in green colour, have been adopted for excitation or detection of acoustic signals.
By applying a DC bias voltage Vb between IDT1 shorted with IDT3 and the silicon pad, it is possible to exploit the d31 mode in the AlN film to induce a planar static compressive or tensile stress in the diaphragm, depending on the sign of the bias voltage Vb, as reported in Figure 5a,b, respectively. The application of an electrically controllable mechanical stress allows the mechanical resonant frequency fR0 of the diaphragm to be shifted thus leading to tunable resonant frequency. Studies have proven that a stress induced to a clamped square plate leads to variations of the frequencies of its vibrational modes, including the first flexural mode [38,39]. Considering the spacing of 28 μm between two consecutive fingers and the thickness of the AlN layer of 0.5 μm, the electric field E induced within the AlN layer below each finger can be assumed as in the configuration of parallel plates, where the top plate is the corresponding finger, and the bottom plate is the highly doped silicon layer. Given the piezoelectric polarization vector P oriented along the negative direction of the z-axis, by applying a positive bias voltage Vb > 0, an expansion along the z-axis and a contraction along the x-axis of the piezoelectric material is produced at each finger, as reported in the inset of Figure 5a. Two IDTs, namely IDT1 and IDT3, placed on the opposite inner edges of the diaphragm, shown in the simplified schematic of Figure 4c in blue colour, have been shorted and employed as a top plate over the piezoelectric layer.
The doped silicon layer, contactable by employing the metal pads shown in Figure 4c in dark red colour, has been grounded and employed as a bottom plate. The four IDTs on the outer edges of the diaphragm, i.e., IDT5-IDT8, shown in black colour, have not been employed and thus left unconnected. The remaining two IDTs, namely IDT2 and IDT4, on the inner edges of the diaphragm, shown in green colour, have been adopted for excitation or detection of acoustic signals.
By applying a DC bias voltage V b between IDT1 shorted with IDT3 and the silicon pad, it is possible to exploit the d31 mode in the AlN film to induce a planar static compressive or tensile stress in the diaphragm, depending on the sign of the bias voltage V b , as reported in Figure 5a,b, respectively. The application of an electrically controllable mechanical stress allows the mechanical resonant frequency f R0 of the diaphragm to be shifted thus leading to tunable resonant frequency. Studies have proven that a stress induced to a clamped square plate leads to variations of the frequencies of its vibrational modes, including the first flexural mode [38,39]. Considering the spacing of 28 µm between two consecutive fingers and the thickness of the AlN layer of 0.5 µm, the electric field E induced within the AlN layer below each finger can be assumed as in the configuration of parallel plates, where the top plate is the corresponding finger, and the bottom plate is the highly doped silicon layer. Given the piezoelectric polarization vector P oriented along the negative direction of the z-axis, by applying a positive bias voltage V b > 0, an expansion along the z-axis and a contraction along the x-axis of the piezoelectric material is produced at each finger, as reported in the inset of Figure 5a. Consequently, a planar static tensile stress is induced into the diaphragm as indicated by the arrows, thus increasing the mechanical resonant frequency fR, i.e., fR > fR0. On the contrary, by applying a negative bias voltage Vb < 0 a contraction along the z-axis and an expansion along the x-axis of the piezoelectric material is produced at each finger, as reported in the inset of Figure 5b. Consequently, a compressive stress will be induced into the diaphragm as indicated by the arrows, thus decreasing the mechanical resonant frequency fR, i.e., fR < fR0.
Finite Element Analysis of the Piezoelectric MEMS Device
The electro-mechanical behaviour of the piezoelectric MEMS device described in Section 2 has been investigated by means of 3D finite element modelling in COMSOL Multiphysics ® . Top and bottom views of the developed 3D model of the device are reported in Figure 6a,b, respectively.
Consequently, a planar static tensile stress is induced into the diaphragm as indicated by the arrows, thus increasing the mechanical resonant frequency f R , i.e., f R > f R0 . On the contrary, by applying a negative bias voltage V b < 0 a contraction along the z-axis and an expansion along the x-axis of the piezoelectric material is produced at each finger, as reported in the inset of Figure 5b. Consequently, a compressive stress will be induced into the diaphragm as indicated by the arrows, thus decreasing the mechanical resonant frequency f R , i.e., f R < f R0 .
Finite Element Analysis of the Piezoelectric MEMS Device
The electro-mechanical behaviour of the piezoelectric MEMS device described in Section 2 has been investigated by means of 3D finite element modelling in COMSOL Multiphysics ® . Top and bottom views of the developed 3D model of the device are reported in Figure 6a,b, respectively. Figure 6c reports an enlarged view of the structural layers that have been included in the 3D model. The nominal dimensions reported in Section 2 have been considered, i.e., neglecting tolerances in layer thicknesses produced by the manufacturing process. In the reported 3D model, the metal layer has been considered as made by Al, thus Cr has been neglected, since the Al thickness is 50 times higher than the Cr thickness. The four IDTs on the outer edges of the diaphragm have not been included in the model since, as described in Section 2, they have not been actuated and they do not affect the mechanical properties of the diaphragm to any significant extent. The SiO 2 has been used as the oxide layer material while Si <100> has been adopted for the substrate and the silicon layer.
The piezoelectric coefficients d 31 = −2.78 pC/N and d 33 = 6.5 pC/N have been specified for the AlN piezoelectric layer as reported in [22,35]. A rotation of 180 deg around the x-axis of the coordinate system has been adopted for the piezoelectric layer to correctly align the poling direction with the negative direction of the z-axis. The piezoelectric effect has been considered by including in the simulation the piezoelectric multiphysics which combines the solid mechanics with the electrostatics physics. Figure 6c reports an enlarged view of the structural layers that have been included in the 3D model. The nominal dimensions reported in Section 2 have been considered, i.e neglecting tolerances in layer thicknesses produced by the manufacturing process. In th reported 3D model, the metal layer has been considered as made by Al, thus Cr has bee neglected, since the Al thickness is 50 times higher than the Cr thickness. The four IDT on the outer edges of the diaphragm have not been included in the model since, a described in Section 2, they have not been actuated and they do not affect the mechanica properties of the diaphragm to any significant extent. The SiO2 has been used as the oxid layer material while Si <100> has been adopted for the substrate and the silicon layer.
The piezoelectric coefficients d31 = −2.78 pC/N and d33 = 6.5 pC/N have been specified for the AlN piezoelectric layer as reported in [22,35]. A rotation of 180 deg around the x axis of the coordinate system has been adopted for the piezoelectric layer to correctly alig the poling direction with the negative direction of the z-axis. The piezoelectric effect ha been considered by including in the simulation the piezoelectric multiphysics which combines the solid mechanics with the electrostatics physics.
Regarding the solid mechanic physics, a fixed boundary constraint has been applied to the bottom surface of the substrate while for the piezoelectric layer a strain-charg constitutive relation has been specified including the AlN material properties. The gravit constraint has been applied to the domains of the whole structure.
Regarding the electrostatics physics, a charge conservation boundary condition ha been applied to the AlN layer. Terminal constraints have been specified to the domain o each comb-shaped arrays of fingers. The metal pads placed in the device corners to contac the silicon layer beneath the piezoelectric layer have not been included in the 3D mode since a ground constraint has been applied to the top surface of the silicon. The mes domain has been carefully designed to obtain a convergent solution while reducing th computational workload. Top and bottom views of the mesh domain are shown in Figur 7a,b, respectively. Layers that compose the diaphragm have been studied with a fine mesh, while layers laid on the outer edges of the diaphragm with a coarser mesh, a reported in Figure 7c. Specifically, top surfaces of the metal and AlN layers that compos the diaphragm have been meshed with a mapped resolution distribution of 1 μm and with a free triangular minimum element size of 36 μm, respectively. Whereas, a free triangula Regarding the solid mechanic physics, a fixed boundary constraint has been applied to the bottom surface of the substrate while for the piezoelectric layer a strain-charge constitutive relation has been specified including the AlN material properties. The gravity constraint has been applied to the domains of the whole structure.
Regarding the electrostatics physics, a charge conservation boundary condition has been applied to the AlN layer. Terminal constraints have been specified to the domain of each comb-shaped arrays of fingers. The metal pads placed in the device corners to contact the silicon layer beneath the piezoelectric layer have not been included in the 3D model since a ground constraint has been applied to the top surface of the silicon. The mesh domain has been carefully designed to obtain a convergent solution while reducing the computational workload. Top and bottom views of the mesh domain are shown in Figure 7a,b, respectively. Layers that compose the diaphragm have been studied with a finer mesh, while layers laid on the outer edges of the diaphragm with a coarser mesh, as reported in Figure 7c. Specifically, top surfaces of the metal and AlN layers that compose the diaphragm have been meshed with a mapped resolution distribution of 1 µm and with a free triangular minimum element size of 36 µm, respectively. Whereas, a free triangular mesh with a minimum element size of 90 µm has been applied to layers laid on the outer edge of the diaphragm and swept down to the substrate layer.
A two-step study with parametric sweep has been employed to evaluate the effect of the electrical DC bias to the resonant frequency of the diaphragm. The terminal voltage V b of IDT1 shorted with IDT3 has been varied within the range of ±8 V with a step size of 2 V while leaving the terminals of IDT2 and IDT4 electrically floating.
As a first step, a stationary study has been employed to analyse the mechanical effect of the electric static load, i.e., an electrically induced prestress, on the diaphragm. A two-step study with parametric sweep has been employed to evaluate the effect of the electrical DC bias to the resonant frequency of the diaphragm. The terminal voltage Vb of IDT1 shorted with IDT3 has been varied within the range of ± 8 V with a step size of 2 V while leaving the terminals of IDT2 and IDT4 electrically floating.
As a first step, a stationary study has been employed to analyse the mechanical effect of the electric static load, i.e., an electrically induced prestress, on the diaphragm.
The stationary study results of the z-axis displacement for Vb = 8 V and Vb = −8 V have been reported with a 3D representation, not in true scale, in Figure 8a,b, respectively. It can be noticed that, as expected, the convexity of the diaphragm deflection is function of the sign of the applied bias voltage Vb due to the induced planar static compressive or tensile stress. The z-axis displacement wp of the point laid on the top of the AlN surface in the centre of the diaphragm as a function of the bias voltage Vb is plotted in Figure 9. A displacement of 0.48 μm and −0.51 μm has been obtained at Vb = 8 V and Vb = −8 V, respectively. The stationary study results of the z-axis displacement for V b = 8 V and V b = −8 V have been reported with a 3D representation, not in true scale, in Figure 8a,b, respectively. It can be noticed that, as expected, the convexity of the diaphragm deflection is function of the sign of the applied bias voltage V b due to the induced planar static compressive or tensile stress. The z-axis displacement w p of the point laid on the top of the AlN surface in the centre of the diaphragm as a function of the bias voltage V b is plotted in Figure 9. A displacement of 0.48 µm and −0.51 µm has been obtained at V b = 8 V and V b = −8 V, respectively. With Vb = 0 V, i.e., without electrically induced prestressed, wp is equal to −15 nm due to the gravity effect included in the simulation.
As a second step, an eigenfrequency study has been employed to compute the first flexural mode of the structure considering the influence of the electric static load previously evaluated by means of the stationary study. The simulation results of the With V b = 0 V, i.e., without electrically induced prestressed, w p is equal to −15 nm due to the gravity effect included in the simulation.
As a second step, an eigenfrequency study has been employed to compute the first flexural mode of the structure considering the influence of the electric static load previously evaluated by means of the stationary study. The simulation results of the prestressed eigenfrequency study related to the first eigenmode of the structure are reported in Figure 10. prestressed eigenfrequency study related to the first eigenmode of the structure are reported in Figure 10. Specifically, the mechanical resonant frequency fR of the piston-like first flexural vibrational mode of the diaphragm is plotted versus Vb. The estimated resonant frequency varies from 5.06 kHz for Vb = -8 V up to 5.19 kHz for Vb = 8 V. Therefore, by adjusting the voltage Vb it is possible to electrically tune the resonant frequency of the diaphragm. As expected, this could provide the system with the capability of reaching the coupling between the series and parallel resonant frequencies of a piezoelectric MEMS acoustic transceiver. The tuning sensitivity S = 8.6 Hz/V of the system defined as the linearized Specifically, the mechanical resonant frequency f R of the piston-like first flexural vibrational mode of the diaphragm is plotted versus V b . The estimated resonant frequency varies from 5.06 kHz for V b = −8 V up to 5.19 kHz for V b = 8 V. Therefore, by adjusting the voltage V b it is possible to electrically tune the resonant frequency of the diaphragm. As expected, this could provide the system with the capability of reaching the coupling between the series and parallel resonant frequencies of a piezoelectric MEMS acoustic transceiver. The tuning sensitivity S = 8.6 Hz/V of the system defined as the linearized ratio between the resonant frequency shift and the applied bias voltage has been estimated by taking the angular coefficient of the linear fitting of simulated data shown in Figure 10.
Experimental Results
The possibility to improve the receiving-transmitting effectiveness through an applied DC bias voltage V b was experimentally investigated by testing the piezoelectric MEMS device in both acoustic receiver and transmitter modes.
The block diagram of the piezoelectric MEMS device configured as acoustic receiver is reported in Figure 11a. In receiver mode the direct piezoelectric effect was exploited by measuring the voltage signal v out (t) at frequency near the mechanical resonant frequency. The MEMS device can be represented by the equivalent Butterworth-Van Dyke model (BVD) reported in Figure 11b, where the effective mass, mechanical damping, and elastic compliance are represented by the inductance L m , resistance R m , and capacitance C m , respectively. The force induced by the impinging acoustic signal is represented by the voltage v a (t) in the mechanical branch while the parallel capacitance C p represents the dielectric nature of the piezoelectric material. According to such an equivalent circuit, the piezoelectric acoustic device, under voltage readout, displays the highest receiving response at the parallel resonance f p [29], defined as: A sinusoidal excitation voltage vexc(t) with peak amplitude Aexc = 1 V and frequency fexc within the bandwidth 5.3-5.6 kHz, provided by the lock-in amplifier (HF2LI, Zurich Instruments: Zurich, Switzerland), was applied to a speaker (FRWS5, Visaton: Haan, Germany) with a flat response in the frequency region of interest placed at 6.5 cm above the diaphragm, as shown in Figure 11c.
The output voltage signal vout(t) was measured across the parallel connection of IDT2 and IDT4, while the bias voltage Vb was applied between IDT1 shorted with IDT3 and the A sinusoidal excitation voltage v exc (t) with peak amplitude A exc = 1 V and frequency f exc within the bandwidth 5.3-5.6 kHz, provided by the lock-in amplifier (HF2LI, Zurich Instruments: Zurich, Switzerland), was applied to a speaker (FRWS5, Visaton: Haan, Germany) with a flat response in the frequency region of interest placed at 6.5 cm above the diaphragm, as shown in Figure 11c.
The output voltage signal v out (t) was measured across the parallel connection of IDT2 and IDT4, while the bias voltage V b was applied between IDT1 shorted with IDT3 and the silicon pad using a power supply (Polytec: Grenoble, France). The acquired voltage v out (t) was synchronously demodulated with the excitation signal v exc (t) by the lock-in amplifier, thus providing the magnitude ratio |v out |/|v exc | of the resulting receiving transfer function which is plotted as a function of f exc for different values of V b in Figure 12. A sinusoidal excitation voltage vexc(t) with peak amplitude Aexc = 1 V and frequency fexc within the bandwidth 5.3-5.6 kHz, provided by the lock-in amplifier (HF2LI, Zurich Instruments: Zurich, Switzerland), was applied to a speaker (FRWS5, Visaton: Haan, Germany) with a flat response in the frequency region of interest placed at 6.5 cm above the diaphragm, as shown in Figure 11c.
The output voltage signal vout(t) was measured across the parallel connection of IDT2 and IDT4, while the bias voltage Vb was applied between IDT1 shorted with IDT3 and the silicon pad using a power supply (Polytec: Grenoble, France). The acquired voltage vout(t) was synchronously demodulated with the excitation signal vexc(t) by the lock-in amplifier, thus providing the magnitude ratio |vout|/|vexc| of the resulting receiving transfer function which is plotted as a function of fexc for different values of Vb in Figure 12. The results of Figure 12 show that by acting on V b it is also possible to electrically tune the resonant frequency of the piezoelectric MEMS device configured as an acoustic receiver.
The tuning sensitivity S was derived by the linear fitting of experimental data reported in Figure 13. The uncertainty for f P was estimated as σ = 5 Hz, and the uncertainty of S was obtained exploiting the error propagation approach [40]. The tuning sensitivity S results 7.8 ± 0.9 Hz/V for the receiver mode. Given the electrical constraints imposed for the FEM simulation reported in Section 3 the simulated mechanical resonant frequency f R is expected to approach the parallel resonant frequency f P defined in Equation (1). The obtained values of sensitivity show a good agreement between simulated and experimental results, demonstrating that a tunability of the parallel resonant frequency can be obtained in the explored range for V b . Discrepancies between the simulated and experimental results of f p are probably related to the tolerances introduced by the fabrication process of the device which were not taken into full account in the simulations.
The block diagram of the piezoelectric MEMS configured as acoustic transmitter is reported in Figure 14a. In transmitter mode the converse piezoelectric effect was exploited by applying the alternating excitation voltage v exc (t) at frequency near the mechanical resonant frequency. The MEMS device can be represented by the equivalent BVD model of Figure 14b. The velocity of the diaphragm causing the emitted acoustic signal is represented in electrical formalism by the current i a (t). According to such an equivalent circuit, the piezoelectric acoustic device, under voltage excitation, exhibits the highest transmitting output at the series resonance f s [29] defined as: the FEM simulation reported in Section 3 the simulated mechanical resonant frequency fR is expected to approach the parallel resonant frequency fP defined in Equation (1). The obtained values of sensitivity show a good agreement between simulated and experimental results, demonstrating that a tunability of the parallel resonant frequency can be obtained in the explored range for Vb. Discrepancies between the simulated and experimental results of fp are probably related to the tolerances introduced by the fabrication process of the device which were not taken into full account in the simulations. The block diagram of the piezoelectric MEMS configured as acoustic transmitter is reported in Figure 14a. In transmitter mode the converse piezoelectric effect was exploited by applying the alternating excitation voltage vexc(t) at frequency near the mechanical resonant frequency. The MEMS device can be represented by the equivalent BVD model of Figure 14b. The velocity of the diaphragm causing the emitted acoustic signal is represented in electrical formalism by the current ia(t). According to such an equivalent circuit, the piezoelectric acoustic device, under voltage excitation, exhibits the highest transmitting output at the series resonance fs [29] defined as: The excitation voltage vexc(t) was applied by means of the lock-in amplifier to the parallel connection of IDT2 and IDT4. The DC bias voltage Vb was applied between IDT1 shorted with IDT3 and the silicon pad and swept within the range of ± 8 V with a step size of 2 V. The generated acoustic signal was measured by a microphone (2670, Brüel & Kjaer: Naerum, Denmark) placed at 2 cm above the diaphragm, as shown in Figure 14c. The microphone output was fed to an amplifier (Nexus 2690, Brüel & Kjaer: Naerum, Denmark) set with a sensitivity of 1 V/Pa. The measured output signal vout(t) was fed to the lock-in amplifier input for synchronous demodulation with the excitation signal. The magnitude ratio |vout|/|vexc| of the resulting transmitting transfer function is reported as a function of fexc for different values of Vb in Figure 15. The excitation voltage v exc (t) was applied by means of the lock-in amplifier to the parallel connection of IDT2 and IDT4. The DC bias voltage V b was applied between IDT1 shorted with IDT3 and the silicon pad and swept within the range of ± 8 V with a step size of 2 V. The generated acoustic signal was measured by a microphone (2670, Brüel & Kjaer: Naerum, Denmark) placed at 2 cm above the diaphragm, as shown in Figure 14c. The microphone output was fed to an amplifier (Nexus 2690, Brüel & Kjaer: Naerum, Denmark) set with a sensitivity of 1 V/Pa. The measured output signal v out (t) was fed to the lock-in amplifier input for synchronous demodulation with the excitation signal. The magnitude ratio |v out |/|v exc | of the resulting transmitting transfer function is reported as a function of f exc for different values of V b in Figure 15. The results of Figure 15 show that by acting on the prestress caused by the bias voltage Vb it is possible to electrically tune the resonant frequency of the piezoelectric MEMS also in transmitter mode. Considering the maximum of the magnitude of the transmitting transfer function vout/vexc, the series resonant frequency fs was estimated.
The tuning sensitivity S of the system was derived by the linear fitting of experimental data reported in Figure 16. The uncertainty for fS, as for the receiver mode, was estimated as σ = 5 Hz. The tuning sensitivity S results 8.7 ± 0.5 Hz/V for the transmitter mode. The reported data demonstrates that a tunability of about 130 Hz can be obtained in the explored range for Vb. The obtained experimental values of S in the receiver and transmitter modes, taking into account their uncertainties, are compatible with each other in metrological sense [40] and closely approach the simulated value. The results of Figure 15 show that by acting on the prestress caused by the bias voltage V b it is possible to electrically tune the resonant frequency of the piezoelectric MEMS also in transmitter mode. Considering the maximum of the magnitude of the transmitting transfer function v out /v exc , the series resonant frequency f s was estimated.
The tuning sensitivity S of the system was derived by the linear fitting of experimental data reported in Figure 16. The uncertainty for f S , as for the receiver mode, was estimated as σ = 5 Hz. The tuning sensitivity S results 8.7 ± 0.5 Hz/V for the transmitter mode. The reported data demonstrates that a tunability of about 130 Hz can be obtained in the explored range for V b . The obtained experimental values of S in the receiver and transmitter modes, taking into account their uncertainties, are compatible with each other in metrological sense [40] and closely approach the simulated value.
The measured tuning sensitivities and frequency shifts obtained in both receiver and transmitter modes demonstrate that matching of the series resonant frequency with the parallel resonant frequency can be obtained by acting on the bias voltage in either one of the two working modes. A comparison between the receiver and the transmitter modes in terms of the normalized measured magnitude ratio as a function of the frequency f exc without and with the applied tuning by V b is reported in Figure 17a,b, respectively.
Specifically, a bias voltage V b = −1.9 V was applied to the device configured as receiver to match the resonant frequency of the device configured as transmitter. Therefore, by electrically tuning V b it is possible to finely control the resonant frequency of the device, thus obtaining the optimal acoustic emission and detection characteristics with the same operating frequency in both voltage-mode driving and sensing. The measured tuning sensitivities and frequency shifts obtained in both receiver and transmitter modes demonstrate that matching of the series resonant frequency with the parallel resonant frequency can be obtained by acting on the bias voltage in either one of the two working modes. A comparison between the receiver and the transmitter modes in terms of the normalized measured magnitude ratio as a function of the frequency fexc without and with the applied tuning by Vb is reported in Figure 17a,b, respectively. Specifically, a bias voltage Vb = −1.9 V was applied to the device configured as receiver to match the resonant frequency of the device configured as transmitter. Therefore, by The measured tuning sensitivities and frequency shifts obtained in both receiver and transmitter modes demonstrate that matching of the series resonant frequency with the parallel resonant frequency can be obtained by acting on the bias voltage in either one of the two working modes. A comparison between the receiver and the transmitter modes in terms of the normalized measured magnitude ratio as a function of the frequency fexc without and with the applied tuning by Vb is reported in Figure 17a,b, respectively. Specifically, a bias voltage Vb = −1.9 V was applied to the device configured as receiver to match the resonant frequency of the device configured as transmitter. Therefore, by
Conclusions
This work has presented a technique to electrically tune the resonant frequency of a piezoelectric MEMS acoustic transducer to obtain matching between the series and parallel resonant frequencies. The piezoelectric MEMS device has been fabricated with the PiezoMUMPs technology exploiting a doped silicon diaphragm with an AlN piezoelectric layer deposited on top. Electrodes disposed symmetrically with respect to the centre of the diaphragm allow for actuating and sensing. By applying a bias voltage V b between the bottom doped silicon layer and top electrodes on the AlN layer, an electrically-controllable stress can be induced into the diaphragm, thus leading to the tuning of the resonant frequency.
The working principle of the proposed technique has been studied by 3D finite element modelling in COMSOL Multiphysics ® and experimentally verified configuring the piezoelectric acoustic transducer in both receiver and transmitter modes.
Experimental results have shown a tuning sensitivity S = 7.8 ± 0.9 Hz/V in receiver mode, whereas a frequency shift of 130 Hz for V b = ±8 V and a tuning sensitivity S = 8.7 ± 0.5 Hz/V have been reached in transmitter mode. A comparison between the receiver and the transmitter modes has been performed by applying a bias voltage V b = −1.9 V to the device configured as receiver to match the resonant frequency of the device configured as transmitter, thus obtaining the optimal acoustic emission and detection characteristics with the same operating frequency in voltage-mode driving and sensing.
Taking advantage of the non-directional response in the low-frequency range, the proposed device can be employed in pulsed-echo mode as a proximity/presence, or gesture detector. Furthermore, the proposed technique can be transferred to a properly down-scaled structure to obtain a tunable piezoelectric micromachined ultrasound transducer (PMUT). | 9,711.8 | 2022-01-01T00:00:00.000 | [
"Engineering",
"Physics"
] |
The Hierarchy Structure in Directed and Undirected Signed Networks
The concept of social stratification and hierarchy among human dates is back to the origin of human race. Presently, the growing reputation of social networks has given us with an opportunity to analyze these well-studied phenomena over different networks at different scales. Generally, a social network could be defined as a collection of actors and their interactions. In this work, we concern ourselves with a particular type of social networks, known as trust networks. In this type of networks, there is an explicit show of trust (positive interaction) or distrust (negative interaction) among the actors. In a social network, actors tend to connect with each other on the basis of their perceived social hierarchy. The emergence of such a hierarchy within a social community shows the manner in which authority manifests in the community. In the case of signed networks, the concept of social hierarchy can be interpreted as the emergence of a tree-like structure comprising of actors in a top-down fashion in the order of their ranks, describing a specific parent-child relationship, viz. child trusts parent. However, owing to the presence of positive as well as negative interactions in signed networks, deriving such “trust hierarchies” is a non-trivial challenge. We argue that traditional notions (of unsigned networks) are insufficient to derive hierarchies that are latent within signed networks.
Introduction
Structural analysis of complex networks has been a dynamic and challenging area of interest among researchers for the past few decades [1].In a generic sense, a network is a collection of nodes associated to the other through links [2].Several graph theoretic approaches over such networks have revealed certain fundamental facts.Evidently, network analysis could provide us with better insights in understanding the hidden aspects of individuals or groups involved within a network, the pattern of relationships, how they evolve etc [3].Any network could be represented as a graph consisting of a collection of nodes (units) and edges (interactions) [4].In a network, the manner in which one node interacts with the other displays an important feature, the connectedness among nodes.The nature of connectedness underlying a network also determines its complex topology.In other words, network complexity is an intrinsic property of any physical, chemical, biological or social system characterized by various nodes and their interactions [5].Examples include organizational networks, neural networks, protein interaction networks, Internet, the World Wide Web and social networks to name but a few.
The past decade witnessed a tremendous rise in the popularity of online social networks such as Twitter, Digg, Youtube, Delicious, Livejournal, Facebook etc.Our study mainly focuses on the analyses of similar online social networks in order to understand the underlying mechanism of the connections involved as well as to verify the existence of certain social phenomena within the networks.Broadly speaking, a social network could be directed or undirected depending on the type of edges present in them.Directed social networks are distinguished from undirected ones by the presence of directed edges between actors [6].An example (Figure 1) for directed network could be followership in Twitter where an actor simply 'follows' another.Alternatively, undirected social networks comprise of undirected edges between actors.Facebook is an example for undirected networks with edges depicting only mutual friendships.
Another type of classification termed as the trust networks deals with nature of interactions (positive or negative) involved in social networks.In this type of classification, a social network could be categorized as either signed or unsigned.Unsigned networks are described by the presence of a single type of interaction, usually being positive in nature.That is, in unsigned networks all actors are same, either friends or strangers.Generally, social networks are largely found to be unsigned in nature [7].Followership in Twitter and friendship in Facebook are typical examples.But in the real world, the relationships need not always be positive in nature.Signed networks, capture this aspect of society allowing explicit show of trust or distrust among actors.They can designate others as friends or foes [8].In this scenario, an actor is said to trust the other if an actor approves of one's opinion among themselves.At the same time, an actor is said to distrust the other if an actor disapproves of one's opinion.E-opinions, Slashdot Zoo network are some of the examples of signed networks that indicate trust/friends or distrust/foes explicitly among themselves using an edge-weight of +1 and −1 respectively.Mathematically, a signed network can be defined as a directed graph, G = (V, E) where i) V is the set of actors in a network, E ⊆ V × V is the set of edges such that (u, v) indicates a link between u ∈ V and v ∈ V s: E → {+1, −1} assigns the edge weight [9].Consider the following illustration in Figure 2. If node A is connected to node B as a friend, there should be a directed edge from node A to node B with a trust score of +1.Meanwhile, if A is connected to B as a foe, there should be an edge directed from A to B with a score of −1.
Background and Prior Work
Various aspects of hierarchy have been studied in many literatures till date.The general idea behind the concept of hierarchy can be stated as the emergence of a tree-like structure in a top-down fashion in the order of their ranks further depicting a specific relationship.Earlier studies on dominance relationship in animal societies, Bonabeau et al. suggest a process of self-organization of nodes depending on their roles and importance [10].This lead to the identification of important or 'leader' nodes within a community.Such nodes occupy the higher positions in the hierarchy.Therefore, it can be argued that in a hierarchy the higher node indicates a greater influence than the lower ones.Using the directional correlation function analysis, M. Nagy et al. found that similar dominance hierarchies exist in the case of pigeon flocks [11].
In 1984, Huseyn et al. [12] suggested that hierarchy is found in numerous complex systems.Hierarchical organization is also studied in different real networks such as actor network, language network, the Internet and World Wide Web by Ravasz and Barabási in 2003 [13].They proposed that many real networks are scale free and transitive in nature which can be seen as a consequence of the hierarchy underlying the network.Small groups of nodes rearrange themselves to form a hierarchy of larger groups.In order to examine the presence of hierarchical structure in real networks, they argued that the scaling law for the clustering coefficient Ck, is sufficient to quantify the existence of hierarchy of nodes [14].Likewise, hierarchy is observed in certain types of collaboration networks too.
Rowe et al. [15], proposed a novel algorithm to find social hierarchy in e-mail networks by introducing a social score S.This score is computed for each user as a weighted combination of several other measures including the number of e-mails exchanged.Several studies came up with different hierarchy measures that lacked universal applicability on all network types.Instead of employing different measures, the need for a single efficient measure for quantifying hierarchy in complex networks was inevitable.
Liben-Nowell and E. Gilbert et al. [16] [17] Studied on social networks dealt with link-prediction and tie-strength prediction.They addressed the link-pre-diction issue and discuss certain achievements based on proximity measures of nodes in a network.Rather than considering the network evolution, a static snapshot of the network along with some specific node attributes are taken into study.Link-prediction can be applied to social network analysis to find out interesting or promising interactions within its members.In 2009, E. Gilbert et al. provided a predictive model for tie strength.The model effectively distinguishes between strong and weak ties with over 85% accuracy.The model predicts the tie strength by observing the manner in which a user chooses to communicate to another user in particular regardless of the number other choices offered.
Apart from these, attempts have been made lately to explore hierarchies as well.Helic D. and Strohmaier M. [18] looked into usefulness of tag hierarchies in improving navigability in social tagging sites like Delicious, CiteULike and Flickr.This paper aims to explore the usefulness of tag hierarchies as directories to facilitate navigation or browsing in social tagging systems.In order to construct such a tag hierarchy, the authors have put forward a new version of an existing centrality based algorithm with a branching factor b as an input parameter which describes the maximum number of categories and sub-categories.It employs tag co-occurrence as the similarity measure and tag generality as the centrality measure over the tag-tag networks.In a tag network, each tag is considered as a node and is linked to the other node according to a certain occurrence threshold.In the process of building up a hierarchy, the nodes are first ranked in a descending order based on the degree of centrality (generality threshold) to obtain a centrality list.As a result, the most general tags are placed at the top order.
The proposed algorithm has two phase procedure to ensure that as much as the given tags are being connected to the main tree without the tree being fairly deep.In the first phase, it populates a forest of multiple trees with the most general node as the root node, iterates through the centrality list, identifies the most similar tag to the current tag in the tree computing the co-occurrence threshold and then appends the tag as a child to its most similar tags.It attaches a maximum of subcategories to a given category.Later the produced trees are sorted in descending order of their size (no: of categories they possess) and the largest tree is considered as the main tree.In the second phase, the algorithm appends the other trees to the main tree by connecting the root node of a particular tree to the most similar node in the main tree.In case the most similar mode consists of only one free sub-category spot, then a misc category is introduced into the free spot and then the given tree is appended to that misc category.However, the nesting of misc category is also necessary and cannot be avoided completely due to the very structure of tag-tag networks.Normally, in a typical power-law network, the nodes with high degree centrality are connected to a small number of high and mid degree nodes (high centrality) as well as to a large number of low degree nodes (low centrality).Such high centrality tags occupy the top positions of the hierarchy.Therefore, in the hierarchy building process, the algorithm first appends the adjacent high degree and the mid degree nodes as sub-categories to a given node using up all free sub-category spots followed by the addition of the low degree nodes through the misc category.It is to be noted that nested misc categories do not affect the semantics of the network but rather keep the tags away from the most related ones into its misc categories.The results and simulation studies illustrates that the proposed algorithm outperforms existing ones in constructing a tag hierarchy useful for better navigation.
Maiya and Berger-Wolf [19] introduced a simple and flexible method based on maximum likelihood to infer social hierarchy from weighted social networks.They have used a simple greedy algorithm to infer maximum likelihood hierarchy from a given network.This approach was evaluated against both simulated as well as real-world datasets for accuracy.This method can also be used to infer the generative interaction models that could lead to a social network.The results show that hierarchies can be inferred from the associations among different entities in a network, provides the frequency and occurrence of theses associations.
Gupte et al. [20] investigated the emergence of hierarchy in directed social networks.They propose a measure of hierarchy and illustrate how hierarchy and degree of stratification emerge with the increase in network size.The paper presents a measure of hierarchy and a polynomial time algorithm to find the largest hierarchy in directed networks.This paper also shows that with the increase in network size, the size of hierarchy grows significantly but the rate of stratification tends to be slow.The studies are based on the assumption that there exists a global social rank for every person in a network and each of them is aware of their ranks as well as the ranks of people they link to.It is been observed that when people of higher ranks in a hierarchy links (or recommends) people of lower ranks there occurs a considerable amount of social agony depending on the difference between their ranks.
Global Reaching Centrality (GRC)
In 2012, the problem has been examined and a universal hierarchy measure has recently been put forward by Mones et al. [21].Known as the Global Reaching Centrality (GRC), this new measure captures the heterogeneous distribution of local reaching centralities in a network.Unlike other measures so far suggested, GRC claims to overcome many drawbacks and is widely applicable to all classes of complex networks.They propose a universal hierarchy measure based on Global Reaching Centrality and a visualization technique for any type of complex real-world network [22].In a network, complexity often arises due to the interactions between similar units or as a result of nature of interactions (edges) and units (nodes).Hierarchy is an imperative feature of any complex network.However, the emergence of hierarchy within a network depends largely on the extent to which a node influences the other as well as the system in whole.Therefore, a node with the strongest influence can be regarded 'central' to a network.In other words, the nodes with a stronger impact can be at a higher order (rank) in a hierarchy.In fact, determining such nodes becomes crucial in defining a measure of hierarchy.Apart from a tree-like network, a real-world network is much more complicated with the existence of relationships between nodes of the same level, cycles of connected nodes, clusters, edges moving upwards etc. Hence hierarchy detection in networks is very demanding.Hierarchical measures so far been suggested cannot be applied on different complex systems due to many shortcomings.In order to define a measure based on reaching centralities, the paper essentially focus on flow hierarchies in real and adjustable hierarchical (AH) networks.
The concept of Global Reaching Centrality (GRC) [21] measures the heterogeneous distribution of local reaching centralities in a graph.Local reaching centrality is largely based on a generalized case of m-reach centrality with m = N; where N is the no: of nodes in a given graph.In an unweighted directed graph, local reaching centrality is the ratio of number of nodes with finite positive directed distance from a particular node to the maximum number of reachable nodes from the same node.Therefore, the GRC of an unweighted directed graph can be defined as the difference between the highest and the average local reaching centralities within a network, given by [21]: V = set of nodes; ( ) For weighted undirected graphs, the generalization of GRC is quite straightforward based on local reaching centrality as defined for unweighted direct graphs.In the case of weighted directed graphs, the sum of lengths of all outgoing directed paths from node 'i' to node 'j' as well as the weight of edge along the path is taken into account.If there exist more than one directed shortest path from i to j, then the path with maximum weight (i.e.maximum connection strength) is considered.Similarly, for an undirected unweighted graph, GRC can be obtained by excluding computation of weights of shortest path between two nodes.Further, GRC is observed on an adjustable hierarchy (AH) model.In an AH model, all nodes in a directed tree is assigned to a level 'l' such that the level of the root node is equal to the total number of levels and those at the bottom level has l = 1.If a node has a level l, then the level of the child nodes would be l − 1. Thereafter an additional no: of random edges are included in the model in such a way that 1 − p proportion of edges is totally random.That is, two nodes chosen, say A and B, are connected if they were not already connected in the (AB) direction.The p proportion of the edges are connected as (AB) only if, to preserve hierarchy.Randomization of real networks is done by generating a random network with the same in and out degree with respect to the original model and followed by choosing two random edges AB and CD and then changing the endpoints to obtain AD and CB.
Analysis on a few classical networks such as Erdős-Rényi (ER) graphs [23], Scale-Free (SF) graphs and directed trees reveal that the GRC values are more acceptable than standard deviations of local reaching centralities to measure the hierarchical properties.The GRC for an adjustable hierarchical (AH) network is found to change continuously and monotonously in an interval of a highly random state to a fully hierarchical one.In the case of real networks, the edges are directed so that the origin of the edge has a greater impact on terminal.It has been observed that GRC depends largely on the average degree and network structure.A network with higher average degree has a smaller GRC indicating the existence of a lower hierarchy.However, the comparison of the actual GRC value with GRC of the randomized versions of the original networks exhibits slight variations.In order to analyze the correlation between hierarchy and controllability of a network, GRC is then compared with the number of driver nodes under switch board dynamics.Here, driver nodes are nodes that control the state of every edge.For a total control over an easily controllable network, the no: of driver nodes to be controlled are few.The results so obtained tend to exhibit a negative correlation between the two quantities, i.e.GRC and are inversely proportional to each other.This clearly suggests that a hierarchical network is better controllable.
The proposed hierarchical visualization technique for large graphs assigns each node into different levels on the basis of a local quantity.For an unweighted digraph this local quantity is equal to the local reaching centrality.
Therefore, an ER graph posses a two layered hierarchical structure and arborescence has many layers.The structure of an SF graph lies in between an ER graph and an arborescence with a few clearly separated layers.To avoid different hierarchical lay-outs for single graphs of same graph model, ensembles of ER, SF, directed AH and real networks are visualized.In short, the proposed hierarchy measure, GRC quantifies the heterogeneity of local reaching centrality in whole network by introducing bidirectional edges among equivalent nodes.It is free from the drawbacks of the hierarchy measures so far been suggested.Hence, it can be concluded that GRC is a more suitable measure for hierarchy in any network.
Classification of Networks
Networks could be of different types.Some of them include: 1) Physical networks comprising of physical entities and their interactions.
Examples could be road network, world-maritime network etc., where cities/ ports are nodes and their routes are links, 2) Biological networks like protein-interaction networks, gene-regulatory network where proteins/genes form the nodes and their interactions form links, 3) Social networks where people or other entities become the nodes depending on the social context and their interactions being links.teraction between nodes due to the threaded discussion.One major difference between the two networks is a link may lose its importance during the course of time in Infrastructure network i.e. a link might languish (or in other words stay static).For example, if one does not interact with a person on a regular basis then the link which connects both of them loses its importance with time.But in interaction network a link never loses its importance with time, as the nodes continue to interact with each other regularly.
A network can also be classified as signed and unsigned networks.In Unsigned networks, link between the nodes doesn't say about the nature of the link.
Online social networks like Facebook, twitter, friendster, and etc come under
Hierarchy in Signed Networks
Studies so far reveal only certain typical statistical properties shared by most of the complex networks.Some of distinctive properties include small-world phe-nomena [24], power-law degree distributions [25], clustering also called as network transitivity [26], community detection [22] etc.However, there still remain certain issues that are open.Hierarchy being one such issue has attracted many scientists.
Connectedness is a property exhibited by all networks and it determines the arrangement of nodes within a network.Such an arrangement gives rise to different classes of nodes based on certain factors that serve as a measure.In online social networks, actors tend to connect with each other within and across different classes on the basis of their perceived social hierarchy.The concept of social hierarchy can be stated as the emergence of a tree-like structure comprising of actors in a top-down fashion in the order of their ranks, describing a specific parent-child relationship.The total prestige owned by an actor could be considered as a measure of status.Therefore, a social hierarchy conveys a structure of authority and could be latent in every social network and needs to be extracted.
Different literatures present a variety of approaches and measures for mining hierarchy in complex networks.Attempts have also been made to mine hierarchy in social networks.These are further discussed in the related literature section.However, in signed networks the hierarchy is far less discernible.The presence of negative interactions in signed networks, pose an additional challenge in deriving trust hierarchies from signed networks.Hence, we argue that the traditional notions are insufficient to derive hierarchies underlying signed networks.
In order to extract hierarchies from signed networks, we have considered the Slashdot and Epinions networks [27].Slashdot is a technology related website, which has a feature named "Zoo" through which each user connects to other user as friend or foe based on the comments in a threaded discussion on an article.
In the dataset, a friend is represented by directed edge of weight +1 and foe by directed edge of weight −1.Epinions is a consumer review site where members of the network could decide whether to "trust" each other or not.
In this work, we attempt to mine hierarchies that remain latent in a signed network that represents the trust of nodes from the bottom to the root.It also based on a node's immediate neighborhood of trust relationships.Therefore, the trust hierarchy shows the nature of nodes trusting each other and at the same time preserves the locality of trust.These hierarchies are termed as locality-preserving trust hierarchies.
Being highly dynamic in nature, social networks have always reflected interesting patterns of connections among the nodes.These connections mostly lead to a parent-child relationship forming hierarchies among themselves.The hierarchical structure of a population in a social network often shapes the nature of the social interactions of individuals and, thus, provides insights into the underlying structure of the network.Understanding the mechanism by which hierarchies evolve is a fundamental question that still remains vague.Our approach could be relevant to a number of interesting current applications of social net-works including information dissemination, community structure detection and a framework for local self-governance among the population.The crux of our work lies in the fact that we seek to mine hierarchies based on the trust locality of a node in a signed network.That is, the hierarchies should be an abstract portrayal of local community structure.
Interpretations of Hierarchies in Signed Network
As discussed earlier, owing to the presence of positive (trust) as well as negative (distrust) interactions in signed networks (trust networks), the traditional notions of hierarchy were found to be inadequate to derive trust hierarchies.With the purpose of modeling both these interactions effectively into a hierarchy, we introduce two interpretations of trust or goodness into the trust networks.Trust is represented in terms of two different aspects namely, presence of trust and absence of distrust.In fact, these two interpretations could be considered as duals of trust signifying the degree of goodness of an actor.Presence of trust would imply how good an actor is where as an absence of distrust would imply how less bad the actor actually is.Consequently, the trust-based hierarchies thus obtained would consist of several actors arranged in the order of their degree of trust.This could be illustrated in Figure 3, as follows: Figure 3 and Figure 4, illustrates the trust-based hierarchies existing among actors.It is to be noted that in a trust network, high distrust and low trust need not necessarily be the same.In Figure 3, the trust earned is high at the root node indicating high goodness and decreases gradually as we move down the hierarchy.That is, the actors at the bottom of the hierarchy would have a comparatively lower trust than those at the top.On the same note, in Figure 4, the absence of distrust is found to decrease as one moves down the hierarchy beginning at the root.That is, at the root node the absence of distrust is much higher Alternatively, it could be viewed that 'goodness' decreases as we move down a trust hierarchy and 'badness' increases as we move down the distrust hierarchy.
This view puts forward a question of the manner in which an actor is considered to be genuinely good or bad.The philosophy behind this view could be explained in terms of the collective opinion of the population.Social networks comprise of autonomous agents capable of expressing opinions on their own.These opinions solely are based on their independent cognitive processes or inferences.In other words, the opinion of an actor is not hampered by any party or an interest group in particular.Therefore, a collective opinion regarding the trustworthiness of an actor cannot be ruled out as a co-incidence.With time, an architecture entirely based on trust emerges.This emergent trust-based architecture eventually becomes acceptable to the whole population.Thus an actor who has earned the trust (distrust) through the unanimous opinion of the majority is considered to be genuinely trustworthy (untrustworthy).By means of this emergent architecture it is possible to gain new insights into patterns underlying a network.An interesting example in this regard could be the collaborative editing of content in Wikipedia pages.Readers are allowed to edit information related to a particular topic and over time, an information architecture evolves eventually reaching consensus among the editors.
However, so as to convey both trust and distrust effectively in a single hierarchy, the trust as well as the distrust earned by an actor are taken together in terms of their aggregate deserve values.That is, deserve of node u is the aggregate of the trust and distrust it earns from its neighbors.The trust/distrust from node v to node u is dampened based on its bias towards trusting or distrusting the population at large.Therefore, the higher the bias, the lower is the effect of v's vote to u.Thus a consolidated hierarchy of actors is formed by way of a par-
Figure 1 .
Figure 1.Examples of directed and undirected network connectivity.
Figure 2 .
Figure 2. Examples for signed network connectivity.
local reaching centrality; max R C = highest local reaching centrality; N − 1 = maximum traversals possible.
this category.As opposed to unsigned network, in signed network a link carries +1 sign which represents a positive relationship or −1 sign which represents negative relationship among nodes.Both signs can be interpreted differently in different networks.For example in Eopinions network, +1 represents Trust while −1 represents Distrust, while in Slashdot Zoo network +1 represents friendship and −1 represents Foe ship between people.
Figure 3 .
Figure 3. Hierarchy based on the presence of trust.
Figure 4 .
Figure 4. Hierarchy based on Absence of distrust. | 6,340.6 | 2017-10-11T00:00:00.000 | [
"Computer Science"
] |
Comparison of Regression and Artificial Neural Network Models for Surface Roughness Prediction with the Cutting Parameters in CNC Turning
Surface roughness, an indicator of surface quality, is one of the most specified customer requirements in machining of parts. In this study, the experimental results corresponding to the effects of different insert nose radii of cutting tools (0.4, 0.8, 1.2 mm), various depth of cuts (0.75, 1.25, 1.75, 2.25, 2.75 mm), and different feedrates (100, 130, 160, 190, 220 mm/min) on the surface quality of the AISI 1030 steel workpieces have been investigated using multiple regression analysis and artificial neural networks (ANN). Regression analysis and neural network-based models used for the prediction of surface roughness were compared for various cutting conditions in turning. The data set obtained from the measurements of surface roughness was employed to and tests the neural network model. The trained neural network models were used in predicting surface roughness for cutting conditions. A comparison of neural network models with regression model was carried out. Coefficient of determination was 0.98 in multiple regression model. The scaled conjugate gradient (SCG) model with 9 neurons in hidden layer has produced absolute fraction of variance (R2) values of 0.999 for the training data, and 0.998 for the test data. Predictive neural network model showed better predictions than various regression models for surface roughness. However, both methods can be used for the prediction of surface roughness in turning.
INTRODUCTION
Metal cutting is one of the most significant manufacturing processes in material removal.Metal cutting can be defined as the removal of metal from a workpiece in the form of chips in order to obtain a finished product with desired size, shape, and surface roughness.There are different methods of metal cuttings and turning is one of the commonest among these methods.Turning is the process of machining external cylindrical and conical surfaces.It is usually performed on a lathe [1].
The quality of machined components is evaluated by how closely they adhere to set product specifications for length, width, diameter, surface finish, and reflective properties.Dimensional accuracy, tool wear, and quality of surface finish are three factors that manufacturers must be able to control at the machining operations [2].
In machining of parts, surface quality is one of the most specified customer requirements where major indication of surface quality on machined parts is surface roughness.Surface roughness is mainly a result of process parameters such as tool geometry (i.e., nose radius, edge geometry, rake angle, etc.) and cutting conditions (feed rate, cutting speed, depth of cut, etc.) [3].
Surface roughness is harder to attain and track than physical dimensions is, because relatively many factors affect surface roughness.Some of these factors can be controlled and some cannot.Controllable process parameters include feed, cutting speed, tool geometry, and tool setup.Other factors, such as tool, workpiece and machine vibration, tool wear and degradation, and workpiece and tool material variability cannot be controlled as easily [4].
A considerable number of studies has studied the effects of the speed, feed, depth of cut, nose radius, and other factors on the surface roughness.In recent studies, Lin et al. [5], Feng [6], Wang and Feng [7], Risbood et al. [8], Lou et al. [9], Choudhury and El-Baradie [10], Özel and Karpat [3] evaluated the effects of some factors on surface roughness and developed models.
The aim of this study was to set up a multiple regression models and a neural network model to predict the surface roughness of a machined workpiece, using turning operation.Other objectives of this study were the following: (1) to develop prediction models using machining parameters, such as feed rate, insert radius, and depth of cut, as predictors, (2) a prediction accuracy of above 90%, (3) to compare the different prediction methods for surface roughness to find the best model.
SURFACE ROUGHNESS
The surface parameter used to evaluate surface roughness in this study is the roughness average (R a ).This parameter is also known as the arithmetic mean roughness value, arithmetic average (AA), or centerline average (CLA).Within the presented research framework, the discussion of surface roughness is focused on the universally recognized.R a is recognized universally as the commonest international parameter of roughness.The average roughness is the area between the roughness profile and its center line, or the integral of the absolute value of the roughness profile height over the evaluation length (Figure 1) [11][12][13].Therefore, R a is specified by the following equation: when evaluated from digital data, the integral is normally approximated by a trapezoidal rule: where R a is the arithmetic average deviation from the mean line (μm), L is the sampling length, and Y is the ordinate of the profile curve.Graphically, the average roughness is the area (shown in Figure 1) between the roughness profile and its center line divided by the evaluation length (normally five sample lengths with each sample length equal to one cutoff).
Test specimens
Due to the experimental investigations AISI 1030, steel test samples of dimensions ø150×450 mm are prepared and used in tests.Chemical composition of test samples obtained by spectral analysis has been given in Table 1 and the mechanical properties of them are given in Table 2, respectively.
Cutting tools and lathe
In attempts to evaluate the effects of insert radius and cutting parameters on surface roughness values, as equivalent to ISO P20 grade for common carbon containing steel, it has been used cemented carbide cutters manufactured by Mitsubishi, coated with three layers of (TiN, Al2O3, TiC), the outermost CVD TiN.In tests, TNMG 160404-MA, TNMG 160408-MA, TNMG 160412-MA inserts, and MTJNR 2525 M16N tool holder were used.The type of the machine used for the turning test was a Johnford T35 Industrial type CNC lathe machine.The lathe equipped with continuously CNC lathe variable spindle speed from 50 to 3500 rpm, and a 10 KW motor drive was used for machining test.Orthogonal machining of AISI 1030 was used in turning.
Surface roughness measuring instrument
Surface roughness values of finish-turned workpieces were measured by MAHR-Perthometer M1 while measuring instrument and the measurements are repeated three times.To measure roughness of the surface formed while processing the workpiece, the cutoff length is taken as 0.8 mm and the sampling length as 5.6 mm.The temperature of environment was 20 ± 1 • C.
Design of experiment
As recommended in ISO 3685, the cutting speed of 300 m/min has been chosen according to the advice for cutting tools quality given by the manufacturing companies.The experiment includes three controllable process factors, whose levels are presented in Table 3.In this research, 75 sets of experiment are sorted using the standard ordering and carried out according to full factorial design.To obtain the surface roughness values were used the TiN-coated tools by CVD method in the machining of AISI 1030 steel.All of the turning tests were run under dry conditions.
Data processing and analysis were performed using Microsoft Windows versions of Microsoft Excell, Statistical Analysis System (SAS) software and SAS Institute JMP statistical software for the regression analysis.
REGRESSION-BASED MODELING
Regression analysis is a technique for modeling the relationship between two or more variables.Regression models quantitatively describe the variability among the observations by partitioning an observation into two parts [14,15].The first part of this decomposition is the predicted portion having the characteristic that can be ascribed to all the observations considered as a group in a parametric framework.The remaining portion, called the residual, is the difference between the observed and the predicted values and must be ascribed to unknown sources.The goal of the multiple regression analysis was to determine the dependency of surface roughness to selected machining parameters such as feed rate, depth of cut, and insert nose radius.In addition to the main effects of these variables, effects of the interactions of them were included in the analysis.
It can be written being linear and exponential empirical models for surface roughness as functions of feed rate ( f ), depth of cut (d), and insert nose radius (r), A logarithmic transformation converts the nonlinear form of (1) into the following linear mathematical form: The equation is rewritten as where y is the logarithmic value of the measured surface roughness, β 0 , β 1 , β 2 , β 3 are regression coefficients to be estimated, x 0 is the unit vector, x 1 , x 2 , x 3 are the logarithmic values of the feed rate, depth of cut, and insert nose radius, and ε is the random error.
The above equation can be written in scalar notation as you see, is the first-order model.The first-order model, with interaction term, and the second-order model, are utilized in this research.
The above equation can be written in matrix notation as Thus, the least-squares estimator of β is The fitted regression model is The difference between the experimentally measured and the fitted values of response is a residual In the regression analysis, the general null hypotheses, were described as the effects of depth of cut, feed rate, and insert nose radius on the surface roughness, do not significantly differ from zero; that is, The alternative hypothesis could also be expressed as follows: H 1 : at least, one of the β i is not equal to zero.
NEURAL NETWORK MODEL
A neuron is the basic element of neural networks, and its shape and size may vary depending on its duties.Analyzing a neuron in terms of its activities is important, since understanding the way it works also helps us to construct the ANNs.An ANN may be seen as a black box which contains hierarchical sets of neurons (e.g., processing elements) producing outputs for certain inputs.
Each processing element consists of data collection, processing the data and sending the results to the relevant consequent element.The whole process may be viewed in terms of the inputs, weights, the summation function, and the activation function (Figure 2) [16,17].
According to the figure, we have the following.
(1) The inputs are the activity of collecting data from the relevant sources.(2) The weights control the effects of the inputs on the neuron.In other words, an ANN saves its information over its links and each link has a weight.These weights are constantly varied while trying to optimize the relation in between the inputs and outputs.(3) Summation function is to calculate of the net input readings from the processing elements.( 4) Transfer (activation) function determines the output of the neuron by accepting the net input provided by the summation function.There are several transfer functions like summation function.Depending on the nature of the problem, the determination of transfer and summation function are made.A transfer function generally consists of algebraic equations of linear or nonlinear form [18].The use of a nonlinear transfer function makes a network capable of storing nonlinear relationships between the input and the output.A commonly used function is sigmoid function because it is self-limiting and has a simple derivative.An advantage of this function is that the output cannot grow infinitely large or small [19].(5) Outputs accept the results of the transfer function and present them either to the relevant processing element or to the outside of the network.
The functioning of ANNs depends on their physical structure.An ANN may be regarded as a directed graph contain-ing a summation function, a transfer function, its structure, and the learning rule used in it.The processing elements have links in between them forming a layer of networks.A neural network usually consists of an input layer, a number of hidden layers, and an output layer [17].
Determination of data and the network model
The training and test data have been prepared using experimental patterns.In this study, we have 75 patterns obtained from the experiments.Among them, five patterns have been randomly selected and used as the test data.Depth of cut, feed rate, insert radius have been used as input-layer, while the surface roughness was used as output-layer of the ANNs.
In the ANN model, logistic transfer function has been used and expressed as follows: where NET is the weighted sum of the input.Input and output values are normalized between 0 and 1.
The training of the network
Generally, there are 3 different learning strategies.Firstly, the trainer may tell the network what it should learn (supervised learning), secondly, the trainer may indicate whether or not the output is correct without telling what the network should learn (reinforcement learning), and finally, the network learns without any intervention of the trainer (unsupervised learning).The learning set consists of the inputs and the outputs used in training the network.The required outputs take place in this set in the case of supervised learning, while in other cases, they are not found in it [20].In our case, we have used supervised learning approach.
Since the number of neurons found in the input and output layers are known, the best performance of the network with the number of hidden layers is determined using trial error method.Using limited number of neurons with limited number of hidden layers causes lesser learning, while increasing these numbers too much, decreases the speed of learning, and in some cases prevents the learning entirely.Usually, an algorithm is used for the learning process, this algorithm determines the weights.There are various learning methods using these strategies [20].The back propagation learning algorithm has been used with SCG and LM versions at the training and testing stages of the networks [21].The computer program has been developed under MATLAB [22].In the first step of the training, a determination of the learning algorithms is made.The number of hidden layers and the number of neurons for each hidden layer are determined.Then, the number of iterations is entered by the user, and the training starts.The training continues either to the end of the iterations or reaching the target level of errors.Figure 3 illustrates the ANN predictions against the experimental results.
TESTING THE ACCURACY OF BOTH REGRESSION ANALYSIS AND ANN-BASED APPROACH
In order to understand whether a multiple regression analysis or an ANN is making good predictions, the test data which has never been presented to the network is used and the results are checked at this stage.The statistical methods of RMSE, R 2 , and MEP values have been used for making comparisons [23][24][25][26].The same data obtained from the regression analysis is used to determine the mentioned values.These values are determined by the following equations: where t is the target value, o the output, and p the number of samples.
RESULTS AND DISCUSSION
The 3 3 full factorial design was used to study the effect of the three process parameters: depth of cut, feed rate, and insert nose radius on surface roughness.Therefore, the experiment includes three controllable process parameters, whose levels are presented in Table 3.After 75 specimens were cut for experimental purpose, they were measured with a profilometer to obtain the surface roughness average value R a and were recorded.All original 75 samples are shown in Table 4.
The experimental data were applied with a statistical analysis system (SAS) software for multiple regression analysis and neural network analysis.
The results of analysis of variance (ANOVA) of the firstorder model also supported linear relationships in the model (Table 5).F value of regression was 100.88.This F value indicated a great significance (α < 0.0001) for model in rejecting the null hypothesis (H 0 ) that every coefficient of the predictor variables in the model was zero.Instead, the alternative hypothesis, at least one of these coefficients are not equal to zero, was accepted.Therefore, the linear relationship between the predicted variable (R a ) and predictor variables is significantly exist.
Correlation coefficient represents the relationship between the variables.Pearson correlation coefficients between depth of cut, feed rate, insert radius, and surface roughness are presented in Table 6.As seen in Table 6, feed rate and insert nose radius were found to have significant correlation coefficients but insert radius has negative effect.
According to calculated coefficients of main factors, the multiple regression first-order model of surface roughness was built as shown in (16), The determination coefficient (R 2 ) of this model was 0.810 which showed that 81% of observed variability in R a could be ascribed from linear relation.
Equation ( 16) can be transformed into the following form: R a = 52.022d 1.308 f 0.182 r −11.986 . ( The above equation shows that the surface roughness decreases with the increasing of insert radius, whilst it increases with the increasing of feed rate or depth of cut.The expected effects of regressors on the response were observed.Figure 4 shows the main effect on the surface roughness produced by variables d, f , and r.Note that it is preferable to maintain insert nose radius (r) on its highest level and feed rate ( f ) on its lowest level.When the depth of cut is increased, the surface roughness slowly increases, therefore, depth of cut does not have a significant impact on surface roughness as other two variables.
The scatter plot between the actual surface roughness and the predicted surface roughness of all 75 samples as shown in Figure 5 indicated that the relationship between actual sur- face roughness and predicted surface roughness was accepted as a linear.It is seen that most of the points lie close to the line for prediction.
The scatter plot of surface roughness residual versus predicted surface roughness was illustrated in Figure 6.In the model adequacy checking, the regression model was found correct and assumptions were satisfied form in Figure 6.The residual deviations from the mean line were among from −1.5 to 2.5.
The results of analysis of variance (ANOVA) of the model also supported linear relationships in model (Table 7).F value of regression was 179.19.This F value indicated a great significance (α < .0001)for model in rejecting the null hypothesis (H 0 ) that every coefficient of the predictor variables in the model was zero.Instead, the alternative hypothesis, at least one of these coefficients did not equal to zero, was accepted.Therefore, the linear relationship between the predicted variable (R a ) and predictor variables is significantly exist.
In the regression analysis based on the first-order model with interaction terms, pearson correlation coefficients between surface roughness and depth of cut, feed rate, insert nose radius, interaction terms are presented in Table 8.Feed rate ( f ), insert nose radius (r), depth of cut and feed rate interaction (df ), depth of cut and insert nose radius interaction (dr), feed rate and insert nose radius interaction (fr) were found to have significant correlation but insert radius, dr, and df have negative effect.
The developed full model includes df interaction terms that is not significant.Advanced modeling would, therefore, include model reduction and elimination of term that is not significant in the way that statistical hierarchy is not violated.The model reduction is either stepwise or it follows backward elimination.The analysis of variance proved that the feed rate, insert nose radius, depth of cut, and fr interaction most significantly affect the surface roughness.The surface roughness is additionally affected by the dr interaction.The surface roughness model has been developed in a form of reduced equation in term of factors.When only significant factors were considered in the multiple regression analysis, a statistical model was created by regression function in SAS from the tested data, Y = −7.793+ 0.912d + 0.081 f + 6.063r − 0.669dr − 0.061 f r, R 2 = 0.940.
(18) R 2 was 0.940, which showed that 94% of the observed variability in R a could be explained by the main effect and their interactions of independent variables.
Figure 7 shows the actual surface roughness versus predicted surface roughness.A line inclined at 45 • and passing through the origin is also drawn in the figure.For perfect prediction, all points should lie on this line.Here, it is seen that most of points are close to this line.Hence, this model provides reliable prediction.
Residual surface roughness versus predicted surface roughness is illustrated in Figure 8.In the model adequacy checking, the regression model was found correct and assumptions were satisfied from Figure 8.
Figure 9 shows the effects of interactions on the surface roughness parameter produced by variables d, f , and r.Note that there is a nondisregarded interaction between factors dr and fr.As It can be seen in Table 9, the smallest P-value correspond to the fr interaction, and consequently it is the most important.This fact can be observed by analyzing the graphic shown in Figure 9, where the straight lines shown in the dr interaction are more parallel.Figure 9 indicates that a larger insert nose radius working with a higher feed rate would result in a smoother surface.
The results of analysis of variance (ANOVA) of the second-order model supported linear relationships in model (Table 9).F value of regression was 317.29.This F value indicated a great significance (α < 0.0001) for model in rejecting the null hypothesis (H 0 ) that every coefficient of the predictor variables in the model was zero.Instead, the alternative hypothesis, at least one of these coefficients did not equal to zero, was accepted.Therefore, there is a significantly linear relationship between the predicted variable (R a ) and predictor variables.
In the regression analysis based on second-order model, pearson correlation coefficients are presented in Table 10.Feed rate, insert nose radius, df interaction, dr interaction, f r interaction, f 2 and r 2 correlation coefficients were found to have significance on the surface roughness but r, dr, f r, and r 2 have negative effect.
Developed full model includes some quadratic and interaction terms that are not significant.Advanced modeling would, therefore, include model reduction and elimination of terms that are not significant in the way that statistical hierarchy is not violated.The analysis of variance proved that the feed rate, insert node radius, depth of cut, and f r interaction most significantly affect the surface roughness.The surface roughness is additionally affected by the dr interaction.The surface roughness model has been developed in a form of reduced equation in term of factors.
Quadratic model was created by regression function in SAS from the data.The R 2 was 0.977 which showed that 97.7% of the observed variability in R a could be explained by the independent variables.
A second-order model was postulated to extend the variables range in obtaining the relationship between the response and the independent variables.The model is given by This model has a coefficient determination (R 2 ) of 0.977 which indicates a strong relationship between the factors and response.
Actual surface roughness versus predicted surface roughness and residual surface roughness versus predicted surface roughness are illustrated in Figures 10 and 11. Figure 10 shows the plot of actual versus predicted surface roughness.It is seen that most of the points lie very close to the line for strong prediction.For perfect prediction, all points should lie on this line.Hence, this model provides reliable prediction.
Residual surface roughness versus predicted surface roughness is illustrated in Figure 11.In the model adequacy checking, the regression model was found correct and assumptions were satisfied form Figure 11.
Figure 12 shows the interactions on the surface roughness parameter produced by variables d, f , and r.Note that there is a nondisregarded interaction between factors dr and f r.As It can be seen in Table 11, the smallest P-value corresponds to the f r interaction, and consequently it is the most important.This fact can be observed by analyzing the graphic shown in Figure 12, where the straight lines shown in the dr interaction are more parallel.
The ANOVA results also show that all regression models are valid at a high significance (α < 0.01).The secondorder multiple regression model was the better for prediction of surface roughness.They were calculated R 2 = 0.977 and RMSE = 0.375 of multiple second-order regression model.In this study, multiple regression analysis along with the neural network analysis have been applied to measured surface roughness data for making predictions.For accurate results, we have used a single hidden layer by altering the number of neurons used at the hidden layer (e.g., from 3 to 10) to get the best network in terms of the statistical errors that it provides.Table 11 illustrates the behaviors of networks with varying number of neurons.This table has been prepared by selected results obtained from the SCG and LM algorithms.As the table illustrates, the network based on the SCG algorithm with single hidden layer of 9 neurons has provided the best results (Figure 13).Then, the ANN model-as illustrated in Figure 13-is set up using 3 neurons in the input layer with a single hidden layer and finally one neuron is used at the output layer.The representation of knowledge is accomplished by the weights in between the layers.The values of these weights are given at Tables 12-13.
Finally, the surface roughness value can be calculated by Figure 3 corresponds to the R a values for the training data.
It is seen that most of the points lie very close to the original based approach has also been implemented.As Figures 15, 16, and 17 illustrate, for each insert radius value, the predictions of the ANN are very close to the experiment based results.These graphs show that the ANNs may be used as a good alternative in analyzing about the effects of cutting tool geometry and processing parameters on the surface roughness.As a result, the ANN model has been very successful at the training stage and the results for the test data have provided error levels well below critical acceptance level.
The comparison of accuracy values of multiple regression models and neural network model are presented in Table 15.As seen from the table-based on critical values-secondorder multiple regression model has been the best model in the regression analysis, but the neural network model has provided better results than the second-order multiple regression model.
CONCLUSIONS
In this study, the application of regression analysis and neural network analysis on the experimental surface roughness values are compared and discussed.The developed models which are limited with their boundary conditions are compared in terms of the prediction accuracy to the surface roughness.For a long time, modeling techniques have been developed for prediction of the surface roughness.From the results of this study, the following conclusions are drawn.First-order with interaction terms and second-order model predicting equations for surface roughness have been interaction and feed rate and insert nose radius interaction terms and the square terms of feed rate and insert nose radius are statistically significant.Moreover, it is seen that the depth of cut and feed rate interaction and square of depth of cut are insignificant.
The predicted values and measured values are fairly close, which indicates that the developed surface roughness prediction model can be used effectively to predict the surface roughness from the cutting process for the secondorder model.However, based on the statistical error analysis methods, using SCG technique for surface roughness, the R 2 value for the training data set was 0.9997, while for the testing data it became 0.9983; the RMSE values are 0.00058 and 0.0033, respectively; and the mean error values are %2.42 and %2.71, respectively.Therefore, the surface roughness values are accurately determined by the ANN, by using 3 input parameters (i.e., cut depth, feed rate, insert radius), the surface roughness of the steel parts may be predicted with less errors when compared to error of regression models.However, the degree of error can be ignored.Regression requires an explicit function to be defined before the least squares parameter estimates, while a neural network depends more on training data and the learning algorithm.Although predictive neural network model seemed to give better predictions than various regression models for surface roughness, both methods can be used for the same purpose, because the difference in R 2 is very small.The weighted sum of the input to the ith processing element
X j:
The output of the jth processing element W i j : The weights of the connections between ith and jth processing elements The weights of the biases between layers
Figure 2 :
Figure 2: The artificial representation of the biological neuron.
Figure 3 :
Figure 3: The ANN predictions against the experiment-based results.
Figure 4 :
Figure 4: The main effect plots of depth of cut, feed rate, and insert nose radius on the surface roughness.
Figure 5 :
Figure 5: The actual surface roughness versus predicted surface roughness of main effects model.
Figure 6 :
Figure 6: The residual surface roughness versus predicted surface roughness of main effects model.
Figure 7 :
Figure 7: The actual surface roughness versus predicted surface roughness of main effects and interaction terms model.
Figure 8 :Figure 9 :
Figure 8: The residual surface roughness versus predicted surface roughness of main effects and interaction terms model.
Figure 10 :
Figure 10: The actual surface roughness versus predicted surface roughness of second-order model.
Figure 11 :
Figure 11: The residual surface roughness versus predicted surface roughness of second-order model.
Figure 13 :
Figure 13: ANN architecture with 9 hidden neurons in a single hidden layer.
Figure 14 :Figure 15 :
Figure 14: The residual surface roughness versus predicted surface roughness of ANN model.
Table 1 :
The chemical composition of test specimens (weight %).
Table 2 :
Selected mechanical properties of test specimens.
Table 3 :
Design factors and their levels for AISI 1030.
Table 4 :
The average surface roughness values depending on depth of cut, feed rate, and insert nose radius.
a (μm) d f Insert radius Insert radius Insert radius r = 1.2 mm r = 0.8 mm r = 0.4 mm
Table 5 :
ANOVA for surface roughness first-order model in turning of AISI 1030 using cemented carbide tools.
Table 6 :
Pearson correlation coefficients to the surface roughness.
Table 7 :
Analysis of variance for surface roughness first order and interactions terms model in turning of AISI 1030 using cemented carbide tools.
Table 9 :
Analysis of variance for surface roughness quadratic model in turning of AISI 1030 using cemented carbide tools.
Table 10 :
Pearson correlation coefficients of second-order regression model.
Table 11 :
Statistical errors for the surface roughness using various algorithms.
Table 14 .
While Figure14presents the residual surface roughness versus predicted surface roughness of ANN model.The residual values of surface roughness by calculated ANN were less than by calculated regression analysis.For the analysis and simulation of the effects of different insert radii of cutting tools, different depths of cut and different feed rates, on the surface quality of the workpiecesdepending on various processing parameters-an ANN-
Table 12 :
The weights corresponding to the input layer and hidden layer.
Table 13 :
The weights corresponding to the hidden layer and output layer.
Table 14 :
Maximum deviation for surface roughness values.
Table 15 :
Comparison of accuracy values of models. | 7,114.8 | 2007-01-01T00:00:00.000 | [
"Materials Science"
] |
A highly accurate delta check method using deep learning for detection of sample mix-up in the clinical laboratory
Objectives: Delta check (DC) is widely used for detecting sample mix-up. Owing to the inadequate error detection and high false-positive rate, the implementation of DC in real-world settings is labor-intensive and rarely capable of absolute detection of sample mix-ups. The aim of the study was to develop a highly accurate DC method based on designed deep learning to detect sample mix-up. Methods: A total of 22 routine hematology test items were adopted for the study. The hematology test results, collected from two hospital laboratories, were independently divided into training, validation, and test sets. By selecting six mainstream algorithms, the Deep Belief Network (DBN) was able to learn error-free and artificially (intentionally) mixed sample results. The model ’ s analytical performance was evaluated using training and test sets. The model ’ s clinical validity was evaluated by comparing it with three well-recognized statistical methods. Results: When the accuracy of our model in the training set reached 0.931 at the 22nd epoch, the corresponding accuracy in the validation set was equal to 0.922. The loss values for the training and validation sets showed a similar (change) trend over time. The accuracy in the test set was 0.931 and the area under the receiver operating characteristic curve was 0.977. DBN demonstrated better performance than the three comparator statistical methods. The accuracy of DBN and revised weighted delta check (RwCDI) was 0.931 and 0.909, respectively. DBN performed significantly better than RCV and EDC. Of all test items, the absolute difference of DC yielded higher accuracy than the relative difference for all methods. Conclusions: The findings indicate that input of a group of hematology test items provides more comprehensive information for the accurate detection of sample mix-up by machine learning (ML) when compared with a single test item input method. The DC method based on DBN demonstrated highly effective sample mix-up identification performance in real-world clinical settings.
Introduction
Reducing patient harm through minimizing the risk of laboratory error is a major safety principle of laboratory practice.In the clinical laboratory testing process, preanalytical, analytical, and postanalytical phases are the three phases of laboratory practice and are referred to as the total testing process (TTP) [1][2][3].However, preanalytical errors account for approximately 60-70% of all errors found in TTP [4,5] with the primary source of error being related to the clinical sample.Common causes of errors include patient or sample misidentification, sample labeling errors, sample contamination, and measurement interferences in samples.
Delta check (DC), an error screening tool, calculates the difference between the current and the preceding results, and compares this difference against a predefined limit.If this difference is within a predefined DC limit, the result can be released to the clinical team.Otherwise, if the difference is greater than the predefined DC limit, this raises the possibility of an error in the pre-analytical stage.The concept of DC was introduced by Nosanchuk and Gottman in 1974 as a QC technique to identify misidentified samples [6].In 1975, Ladenson [7] described the first use of computers to automatically compare patient's current and previous results in real time.With the widespread use of auto-verification in various areas of laboratory medicine, DC is becoming a mandatory component of autoverification rules to identify results that require additional review before release to the medical record [8].
With more emphasis on proper sample labeling, the prevalence of mislabeled samples may be reduced in certain settings.While efforts to improve labeling practices may mitigate one source of sample mix-up, the ever-expanding scope of tests offered and the sharp increase in sample volumes processed in modern large clinical laboratories introduces high levels of complexity that counteract improvement efforts leaving a sample mix-up rate of 1.2%.Considering the potentially serious health risks posed by unidentified sample mix-up errors to the patient, DC may be as a useful tool to mitigate these risks through early identification of potential sample mix-up errors.Furthermore, DC is unaffected by the prevalence of mislabeled samples.
Issues such as low accuracy of error detection and significant variations in the implementation of DC by different laboratories are, in part, a consequence of the DC method itself and differences for DC limits.Related studies have indicated that the accuracy of DC methods available ranged from 15% to 76% [9].In addition, DC rules are typically defined for individual analytes of interest.However, in practice, multiple items are often tested and results reported as a group or panel.In such instances, multiple DC rules can be combined according to the common test panel, and the interpretation of DC limits for a grouped test panel should be different from a single analyte, since the number of hypothesis tests (i.e. the number of DC rules) applied is much higher and should be taken into account [8,10].
A more detailed and formal definition of machine learning (ML), first introduced by Arthur Samuel in 1959, was described as a computer program that by learning from experience (E) with respect to some class of tasks (T) and performance measure (P), if its performance at tasks in T, as measured by P, improved with experience E [11].In recent years, the widespread recognition of data-driven methods has made ML algorithms widely used in bioinformatics studies, and biomolecular correlation prediction [12].However, to our knowledge, there are no related studies demonstrating how to use deep ML technique to establish a DC method to date.
In this work, employing hematology test item results, we tried to establish a highly accurate DC method by using deep ML to detect sample mix-up in clinical laboratories.The performance of the deep ML approach was assessed by comparison with three well-statistical DC methods.
Data collection and exclusion criteria
In ML, data can be divided into a training set, a validation set, and a test set.The validation set can be understood as a part of the training set to monitor the process of model training.The three datasets are independently separated.In our study, 423,290 deidentified hematology test results measured on the XN-9000 (Sysmex, Kobe, Japan) from 01/2018 to 12/2018 were extracted from the Laboratory Information System (LIS) of the Beijing Chaoyang Hospital.The data from 01/2018 to 10/2018 was used as the training set and the data from 11/2018 to 12/2018 was used as the validation set.Twenty-two thousand four hundred sixty hematology test results from 01/2018 to 12/2018 measured on the BC-5390 (Mindray, Shenzhen, China) were extracted from the LIS of the Beijing Long-fu Hospital to be used as the test dataset.Data filtering rules applied to both the XN and the Mindray datasets.Filter rules included: 1) patients with only one result during the study period were excluded; 2) the first pair of results of each remaining patient was included; 3) Tukey's criteria [13], which defined outliers as values lying three interquartile ranges below the 25th percentile or above the 75th percentile, was applied to remove outlying data; 4) patients with two results after applying Tukey's criteria were included for further analysis; 5) in consideration of gender-dependent and age-dependent differences in distributions of test results, all test results were separated into male and female groups for all test items, and 6) the results of patients aged from 14 years old to 60 years old were included; 7) the time interval of DC was defined to one year [9].The information of deidentified results included: patient type, sex, age, sample number, sample type and all test item result respective values and units.The test results were randomly sorted by a shuffle function in Python 3.7.3 and then automatically matched the current data and preceding data from different patients to generate a mismatched data, simulating a switched sample scenario.The original paired test results were assumed to error-free.The absolute and relative differences were assessed by original matched and mismatched data.
ML method: data pre-processing
After filtering data by predefined exclusion rules, the data was assessed for consistency of analyte and unit parameters and possible missing values for each pair of data.Following assessment, the data was normalized with the Standard Scaler tool in soft package python 3.7.3.Then absolute and relative differences of data were calculated.Isolated forest algorithm was used for removing extreme values in delta data.
ML method: algorithm
The classification problem can be implemented by using classifiers with different algorithms.In our work, six mainstream classifiers were tested and evaluated by confusion matrix.They were Deep Belief Network (DBN), Random Forest (RF), Support Vector Machine (SVM), Logistic Regression (LR), K-Nearest Neighbor (KNN), Naive Bayesian Classifier (NBC).The introduction to the six algorithms is depicted in Supplementary Materials and Methods.
DBN belongs to a deeper neural network in the field of deep learning, which consists of Restricted Boltzmann Machine (RBM) and neural network (NN).DBN was selected for establishing our model.It was implemented by deep learning framework Keras in Python 3.7.3.The main tuning parameters included: 1) "learning_rate_rbm" for controlling the rate of learning; "batch_size_rbm" for selecting the number of sample each time; "n_epochs_rbm" for training iterative epochs; "activation_function_nn" for realizing the nonlinearity between the input and output of neuron.
ML method: implementation
Data pre-processing and model analysis were performed by "numpy" and "pandas" tools in Python 3.7.3 and by "sklearn" and "tensorflow"encoding frameworks in Python 3.7.3.All software packages were accessed from the sklearn library_2.4.0 in the public Python.Python is a computer language that can be used in scientific computing and data analysis, and is currently a mainstream programming tool of artificial intelligence.
Reference change value (RCV ) method
RCV limits of each test item dependent on biological variability (BV) [14] were estimated using the following formula: Coverage factor K was varied from 1.5 to 3.3 in steps of 0.1, coefficient of variation (CV a ) was analytical imprecision, and CV i was within-subject BV.CV a was calculated from the mean CV, which was considered a representative interval of long-term imprecision.There were two-type CV a (CV a, 1 , CV a, 2 ) calculated.CV a,1 used whole data.For the data of CV a,2 , we excluded pairs of test results if both test results constituting the pair were within the reference interval (CL).Extended CL here referred to twice the upper limit value of the CL.
Empirical delta check (EDC) method
EDC limits of each test item were calculated using the absolute or the absolute difference.For each patient, the relative difference for patient, △x r , was given by: where x 1 and x 2 corresponded to the early and later dates of the patient, respectively.The absolute difference for each patient, △x a , was given by: For relative difference, the DC limits were varied from 1% to 200% range in steps of 0.1%, whereas for absolute difference, the DC limits were varied from 1% to 200% of the average test result in the same step.
Revised weighted delta check (RwCDI) method
For all test items, a distribution of values for each test was transformed into approximately Gaussian form by using the Box-Cox formula [15].To make data comparable and unaffected by measurement units, all the transformed test results were standardized to a uniform scale on the basis of reference interval (RI) as described by Ichihara [16].As a next step, we used Formula (1) to get the absolute difference for each test item and calculated a new index termed weighted cumulative delta index (wCDI).We got three panels (including 5-item, 10-item and 22-item) to compute new parameter, and continued following the EDC method.The details of the procedure are described in Figure 1.
Evaluation metrics
The four parameters were defined below as [17]: 1) True Positive (TP): delta check limit was exceeded CL of mismatched queue; 2) False Positive (FP): delta check limit was exceeded CL of matched queue; 3) False Negative (FN): delta check limit was not exceeded CL of mismatched queue; 4) True Negative (TN): when the delta check limit was not exceeded CL of matched queue.
The parameters on confusion matrix were calculated, including true positive rate (TPR), true negative rate (TNR), false positive rate (FPR), false negative rate (FNR), accuracy rate (ACC).We evaluated our model using receiver operating characteristic (ROC) analysis, and the area under the curve (AUC) was calculated, which ranges between 0.0 and 1.0, with values of 0.5 for random classification and 1.0 for perfect classification.(peakedness of distribution) ranged from −0.83 (RBC) to 16.73 (MCHC).
Performance evaluation of six ML algorithms
We evaluated six types of classifiers: SVM, KNN, RF, LR, NBC and DBN model.The evaluation metrics of each model in absolute male data are shown in Table 1 and the detailed ROC curves are depicted in Supplementary Figure 1.We also evaluated the performance of six ML methods for different number of combinations of hematology test items (10-item and 22-item).The performance of all ML methods combined by 22-item was better, as shown in Table 1.As a result, we selected the 22-test item ML model for model training.The 22 hematology test items were input as a multi-label classification task in the ML method, as shown in Figure 1.
Performance of improved DBN model
The robustness and fault tolerance of RF, KNN and SVM to noise data were low, the learning ability of LR and NBC to multi-attribute nonlinear data was weak as well.Compared with the above 5 ML algorithms, NN performed stronger robustness and fault tolerance to noise data, stronger learning ability to complex nonlinear correlation, and higher classification accuracy.However, NN algorithm was also not omnipotent, with the shortcomings of slackness of learning rate or relatively inadequate accuracy.
We designed an improved DBN with restricted Boltzmann machine (RBM) as shown in Figure 3. DBN consisted of two parts: a feature learner with multi-layer RBMs and a classifier with a back propagation (BP).Model training initialized, RBM enabled to be self-encoded to strengthen data features, thus enlarging significant difference between positive data and negative data.Intra-and-inter RBM learning method not only dramatically improved learning rate, but also prevented exploding gradient and vanishing gradient problems, thus to assure capturing the higher accuracy than traditional NN as much as possible.
Comparison with three statistical DC methods
To evaluate the performance of the DBN model, it was compared with three statistical methods which had been proven to have high performance in their respective domains.
Absolute difference and relative difference of all test items were shown on male/female dataset.Nineteen thousand eight hundred seventy-six test results of Long-fu hospital were used to compare the DBN model parameters with three DC methods.Figure 4 demonstrated that seven parameters of four methods collected including TPR, TNR, FPR, FNR, PPV, NPV and ACC.Meanwhile, Figure 4 depicted the absolute difference results in male data among four methods.For the sake of space, the absolute and relative difference results in male and in female are shown in Supplementary Tables 1 and 2 and in Supplementary Tables 3 and 4. Experimental results illustrated DBN was better than the three statistical methods.Of all test items performed, absolute difference DC yielded higher accuracy than relative difference for all methods.The same simulation study was performed by artificially generating cases of female samples.
Discussion
Our model enabled the accurate detection of sample mix-up in real-world settings, illustrating powerful performance when compared to previous studies [10,16,18].The main reasons for these results were as follows: data preprocessing was adopted, mainly including data transformation and removal of extreme values for delta data.The difference was that DBN got rid of extreme values by isolated forest algorithm, while that RwCDI by simple truncation limits.Isolated forest algorithm was a relative robust method to remove extreme values.Its working principle was similar to the density map method.The number of extreme values were able to be adjusted according to the degree of density and balance of data.In this study, isolated forest algorithm in this step removed about 3% of the extreme data, while RwCDI excluded about 1% of the extreme data.For RCV and EDC, the original data only filtered by the first-step rules.The experimental results showed that the accuracy of DBN and RwCDI was 0.9310 and 0.9089 separately.DBN was better than RwCDI and was significantly superior to RCV and EDC.
DC limit setting was the key step to detect sample mix-up.Due to the different control limit settings in various laboratories, the maximum variation in the error detection rate of sample mix-up among laboratories reached up to 76% [9].In this study, two types of DC control limit setting methods were compared.The control limit for EDC was optimized by a dense grid search within a broad range of 0-200% in steps of 0.1.The control limit for RCV was calculated according to individual biological variation and optimized by adjusting k value or excluding pairs of test results within reference intervals or directly extending the original control limits.Our results illustrated that the accuracy for different test items for EDC after optimization ranged from 0.5825 to 0.7804, while for RCV from 0.5631 to 0.8145, which was similar to the results reported in the literature [18].The accuracy of EDC and RCV far lagged behind that of DBN.This might be related with method itself.The working principle of both methods was based on simple DC control limits to distinguish error samples from correct samples.Thus, they are difficult to capture nonlinear effects and interaction in real-world clinical scenarios.
Previous studies reported that the amount of test items affected the accuracy of error detection for sample mix-up [19].Most of DC methods only used a single test item as an input index.If a combination of test items was used as input indexes, ML features would be strengthened.Here k was introduced, which represented the number of test items (k=5-22).Our results proved that the accuracy of DBN adopting 22 test items (k=22) as input indexes reached up to 0.9310, which was higher than 10 test items (k=10).Teppei's study stated that AUC and sensitivity increased proportionately for test items k<10 but remained almost unchanged for k>10, and the cut off value decreased until k=10 and remained unchanged for k>10.This might be related with the way of weighting in the calculation.In Teppei' method [16], a weighting factor was conversed by standard deviation of a given test item.But correlations among test items involved in the calculation did not be taken into consideration.
For DBN model established in this study, the accuracy was regarded as the primary evaluation matric.The most basic component of DBN model was a neuron.Neurons receiving output signals from other neurons (x 1 …x n ) regarded as next input signals, these input signals transferred between neurons by connections with different weights (ω 1 …ω n ).A total input value received by neurons would be compared with a threshold, called θ.Then, the output of neurons was processed by an "activation function" ( y) (Figure 1).RCV and EDC were mainly optimized by adjusting DC limits at different strength.Our experimental data showed that EDC was better than RCV in DC limit optimization, but the input signal of the two methods was only a single dimension, that was x 1 .In the Teppei's method, the input signal was multi-dimensional, i.e. x 1 … x n .This was similar to DBN method.But Teppei's method was one-way correlation to input signals, the number of weight (ω) was the same as the input dimension, and the size of each weight was related to the dispersion of each input signal x, that was ω = 1 aSD 2 .In our DBN model, input signals were transferred in a multi-layer and crossstructured way, and the number of weight was tremendous and complicated.In general, parameters ω i and θ obtained by the way of on-going ML.In particular, perceptron (that was, it had only one layer of neurons) had limited learning ability and mainly solved the linear separable problem.For the nonlinear indivisible problem, we needed to consider the use of multi-layer functional neurons.The learning process was actually to adjust the "connection weight" between neurons and the threshold θ of each functional neuron according to the training set data.The results showed that the accuracy of the four methods was DBN>RwCDI≫EDC>RCV.
The generalization of the model was another important evaluation metrics for assuring a valuable clinical application.In this study, hematology test results were selected due to high testing frequency and high levels of standardization.Data from two laboratories in different hospitals were used to establish our training dataset, validation dataset, and test dataset.The test dataset came from one hospital, the training dataset and the validation dataset set data were from the other hospital.The training dataset and validation dataset were separated independently to avoid overestimation of the accuracy of unknown data by the established model.The experimental results did show that the accuracy training dataset from Chaoyang Hospital was approximately 93%, equal to the accuracy of test dataset from Long-fu Hospital.In addition, in the process of ML algorithm selection, it was found that both the RF algorithm and the DBN algorithm demonstrated acceptable performance characteristics.The DBN algorithm was slightly better than the RF algorithm on the current dataset.However, in clinical complex scenarios, when the data distribution difference became smaller, RF algorithm might be prone to worse, while DBN would represent stronger generalization ability.
In conclusion, our data demonstrate that utilizing the full panel of all available hematology test result items provides more significant information for sample mix-up detection by ML than what is offered by a single test item input.The DC method based on the DBN has demonstrated highly effective sample mix-up identification performance in real-world clinical settings.
Total 445,750 data was included from two hospitals, 123,365 pairs of data in matched queue and 123,365 sets of data in mismatched queue.We split the 423,290 data of Beijing Chao-yang Hospital dataset into a training set from 1/2018 to 10/2018 and a validation set from 11/2018 to 12/2018.We used the 22,460 data of Long-fu Hospital from 1/2018 to 12/2018 as test set.Prior to conducting further analysis, data distribution characteristics were examined; the distribution of MCH and MCHC had a skewness (Sk) close to zero (−0.15 and −0.02) resembling a normal distribution.The other test items examined had skewed distribution with |Sk|>0.3ranging from 0.31 (MCV) to 2.02 (NEUT).All items kurtosis
Figure 1 :
Figure 1: A comprehensive process and architecture of DC detection of sample mix-up.
Figure 2 :
Figure 2: DBN training process flowchart.(A-B) Represents the change of parameters with time for certain layer in the training dataset.(C-D) Represents the change of the accuracy and loss value with time in the training dataset and validation dataset from Beijing Chao-yang Hospital.In each diagram, red colored line represents the training dataset; green colored line the validation dataset.(E) Represents the results of ML algorithm selection.(F) Represents DBN ROC curve of the test dataset from Beijing Long-fu Hospital.
Figure 2C and D shows the change curve of the accuracy and loss value with time in the training set and the validation set at the current training model.The DBN model clearly achieved the highest accuracy on the test dataset, shown in Figure 2F.RF achieved closely competitive performance for current dataset.As shown in Figure 2E, the performance of DBN was obviously superior to those of the other five ML algorithms in DC method.
Figure 3 :
Figure 3: DBN parameter tuning chart.DBN consisted of two parts: a feature learner with multi-layer RBMs and a classifier with a back propagation (BP).Parameter tuning was realized at RBM and BP parts separately.
a) DC methods reported were prone to be affected by the data distribution patterns of test results, DC limits, and the amount of test items.b) Dramatically heterogeneous and extreme results exist in real-world clinical laboratory data and individual biological variations enlarge data fluctuation.c) Assuming analytical variation was ignored, matched data was mainly affected by within-individual biological variation, data distribution pattern and extreme values.Viceversa, mismatched data was mainly affected by between-individual biological variation, data distribution patterns, and extreme values.d) Simple statistical analysis was not use in uncovering cases of sample mix-up.For both DBN and the improved RwCDI, at the first, raw data was filtered by pre-defined rules, and further a series of subsequent
Figure 4 :
Figure 4: Comparison of DBN method with three optimized DC methods using absolute difference results of male samples.
Table :
Prediction scores of models created by different ML algorithms. | 5,570.6 | 2021-12-29T00:00:00.000 | [
"Computer Science"
] |
Multi-sensor particle filtering with multi-step randomly delayed measurements
This paper develops particle filtering for multi-sensor systems with randomly delayed measurements, where the general case that random delay can be multi-step rather than one-step or two-step is considered. Moreover, different sensors can have different delay steps and delay probabilities. Random delays are assumed to be mutually independent for different sensors and modelled by a separate sequence of random variables obeying discrete dis-tributions. Since random delay leads to the actual measurements being dependent rather than independent given states, and this dependence becomes more complicated with the increase of random delay step, a new formula of the local likelihood density is proposed and then a new weighting scheme is adopted in particle filtering to deal with these difficulties. The proposed method is applied to two examples to testify its effectiveness and superiority.
and diverge especially for high-dimensional state space. Moreover, due to the inherent Gaussian assumption, they will yield a poor filtering estimate in strongly non-linear or non-Gaussian cases.
Particle filtering (PF) is a promising alternative to Gaussian approximation filters since it is not subject to the Gaussian assumption and can provide a decent estimate with sufficient samples [13][14][15]. Thus it is attractive to utilise PF to deal with RMD. PFs are proposed to deal with one-step RMD and multistep RMD in [16,17] and [18], respectively. For the multi-sensor system with multi-step RMD, if randomly delays of all sensors occur in a synchronous manner, the PF proposed in [18] can be directly used by augmenting the measurement equation. However, different sensors generally have different delay steps and delay probabilities, that is to say, random delays are asynchronous for different sensors. To tackle this case, a PF is proposed in [19].
All aforementioned PFs assume that the current actual measurement is independent of the previous actual measurements conditioned on the state sequence, which is impractical in the presence of measurement delay and brings a theoretical problem to these PF algorithms (See Remark 1 in Section 4 for details). To solve this problem, an improved PF is developed for the onestep or two-step delay in [20] and [21], respectively, where a new local likelihood density is designed. We find that the calculation of the local likelihood density requires to take all possible combinations of a segment of actual measurements into consideration, and the number of all possible combinations is a power exponential function of delay step (See Remark 2 in Section 4 for details). When delay step is small, one can enumerate all possible combinations as has been adopted in [20] and [21]. But for multi-step RMD, which is more likely to occur in practical applications, combinations become extremely intricate, and enumeration method is obviously impractical. Hence, the motivation of this paper is to design a particle filtering which can deal with the measurement dependence for multi-step asynchronous RMD in multi-sensor systems.
The remainder of the paper is organised as follows. In Section 2, we present the multi-sensor measurement model with asynchronous random delays. In Section 3, we review the core of the generic PF and the standard multi-sensor PF. In Section 4, we study the dependence of multi-step randomly delayed measurements and then propose a new PF to deal with multi-step RMD. Numerical simulations are shown in Section 5. Section 6 concludes the paper.
PROBLEM FORMULATION
This paper considers the following non-linear dynamic system with multiple independent sensors: where x k is the latent state with known initial probability density p(x 0 ), z k,m the ideal (undelayed) measurement of the mth sensor at time step k ∈ ℕ, and M the number of sensors. The system noise {v k , k ≥ 0} and the measurement noise {e k,m , k ≥ 0} M m=1 are white noise processes with arbitrary probability densities p v k (⋅) and p e k,m (⋅), respectively. Now assume that the ideal measurements {z 0,m } M m=1 are always available for use but that for k ≥ 1, the ideal measurements {z k,m } M m=1 may be randomly unavailable. Then the multi-sensor measurement model with asynchronously multi-step random delays can be modelled as where y k,m is the actual measurement of the mth sensor at time step k, and k,m denotes the delay step of the mth sensor at time step k. Let d m be the user-defined maximum possible measurement delay for the mth sensor. Then k,m obeys a d m -valued discrete distribution where the random event k,m = s represents that an s-step measurement delay occurs for the mth sensor at time step k, and p s k,m is the s-step delay probability for the mth sensor at time step k. It is assumed that k 1 ,m is independent of k 2 ,m when k 1 ≠ k 2 , which means that delays occur at different time steps are mutually independent for the same sensor m. In addition, we assume that k 1 ,m 1 is independent of k 2 ,m 2 when m 1 ≠ m 2 . This indicates that delays between different sensors are mutually independent. We further assume the mutual independence of . For notational simplicity, we denote by z k = {z k,1 , … , z k,M } the ideal measurement set of M sensors, y k = {y k,1 , … , y k,M } the actual measurement set of M sensors, and a i: j the set {a i , a i+1 , … , a j } of any general sequence{a k }.
For the non-linear system (1) with randomly delayed measurements (2), our objective is to design particle filtering to approximate the filtering density p(x k |y 0:k ). To fulfil this goal, we quote two widely used properties in the Bayesian filtering framework as follows [22].
Generic particle filtering
In particle filtering context, the joint density p(x 0:k |y 0:k ) is usually approximated by a set of weighted random samples as where (⋅) is the Dirac delta function. The importance weight w (i ) k is computed recursively by where q(x k |x 0:k−1 , y 0:k ) is known as the importance density function.
Standard multi-sensor particle filtering
In the multi-sensor case, since sensors are assumed to be independent of each other, the joint likelihood density p(y k |x 0:k , y 0:k−1 ) can be computed as where p(y k,m |x 0:k , y 0:k−1,m ) is called the local likelihood density. In the undelayed case, the actual measurement sequence y 0:k is exactly the ideal measurement sequence z 0:k . Then by using Properties 1 and 2, the importance weight (5) is simplified into where the joint likelihood density (6) is simplified into Moreover, if we choose q(x k |x k−1 , y k ) = p(x k |x k−1 ) and conduct resampling at each time step, the standard multi-sensor particle filtering, also known as the central particle filter [23] can be obtained.
MULTI-SENSOR PARTICLE FILTERING WITH RMD
In the case of RMD, the actual measurements are disordered, repeated and missing in general. This fact leads to that Properties 1 and 2 cannot be directly applied to actual measurements, and resultantly the importance weight (5) cannot be simplified into (7). Hence, when the standard multi-sensor particle filtering is used by roughly ignoring the RMD, it will inevitably yield a poor state estimate.
To develop a new particle filtering which can capture the information of actual measurements correctly, the importance weight (5) is required to be reformulated. More specifically, we need to figure out expressions of p(x k |x 0:k−1 , y 0:k−1 ) and p(y k,m |x 0:k , y 0:k−1,m ). To this end, we shall give the following two lemmas.
Lemma 1.
Conditioned on x k−1 , x k is independent of x 0:k−2 and y 0:k−1 , that is, Proof. Since the actual measurement set {y 0 , y 1 , … , y k−1 } is a subset of the ideal measurement set {z 0 , z 1 , … , z k−1 }, (9) can then be obtained by using Property 1. □ For the simplified form (10), according to the assumption that elements of { s,m } k s=1 for the fixed m are mutually independent and also independent of the state sequence and measurement sequence, we have
Lemma 2. The local likelihood density is simplified into
where By dividing all possible combinations of j k−d m ,m , … , j k,m into the following d m + 2 cases, we can obtain explicit expressions of ( j k−d m ,m , j k−d m +1,m , … , j k,m ).
In this case, the number of combinations is (d m + 1) d m , and the probability of these combinations is p 0 k,m . Since y k,m takes the quantity z k,m , and none of y k−d m :k−1,m can take the quantity z k,m , by using Property 2 we have Case 2: j k,m = 1 and j k−1,m ≠ 0.
In this case, the number of combinations is d m (d m + 1) d m −1 , and the probability of these combinations is p 1 k,m (1 − p 0 k−1,m ). Since y k,m takes the quantity z k−1,m and none of y k−d m :k−1,m can take the quantity z k−1,m , y k,m does not depend on y k−d m :k−1,m conditioned on x k−d m :k,m . By using the Bayes formula and Property 2, we have Hence In this case, the number of combinations is d s−1 m (d m + 1) d m −s+1 , and the probability of these combinations is where we have where C j k,m denotes the indicator function of set C j k,m . In summary, we obtain the following theorem.
where a j k,m = In fact, the measurement y k,m can only occur a k-step delay at most when k < d m , that is, p j k,m = 0 for k < j ≤ d m . Thus Equation (22) in Theorem 1 can be simplified by discarding the combinations of j k−d m ,m , j k−d m +1,m , … , j k,m with zero probability.
In Theorem 1, the dependence of the actual measurements y k−d m :k,m is indeed considered. But the local likelihood density given by Theorem 1 has an unusual form because of the presence of Dirac delta functions and indicator functions, which brings a trouble for practical calculation. However, when y k and at least one of y k−d m :k−1,m take the same quantity, the second term in (22) is ∞, and resultantly the local likelihood density is ∞. This means that y k,m is uninformative about x k−d m :k and should be discarded. The above analysis can provide us with a solution for this trouble. Let k,m be the index satisfying whered m = min{k − 1, d m }. Then the local likelihood density can be assigned as Remark 3. k,m has the ability to distinguish the actual measurements which are repeated. Moreover, the probability that y k,m is repeated is ℙ( k,m = 1) = a k,m . Repeated measurements are regarded as valueless and will be discarded in (27).
Weighting: • Compute w (i ) k by (28), and normalisew In this paper, we use the transition density as the proposal density and conduct resampling at each time step. The recursive formula of importance weight (5) is thus simplified into (28) The resulting particle filtering for multiple sensors with RMD is presented in Table 1.
NUMERICAL SIMULATIONS
In this section, the efficiency of the proposed particle filtering is shown by two typical non-linear examples. One is the nonstationary growth model, and the other is the two-dimensional radar tracking model. The following filtering algorithms will be used.
Here, newMSPFRMD is our proposed method, while others are used for comparison. It is worth noting that
Example 1: Non-stationary growth model
The non-stationary growth model has been widely used as a benchmark problem to testify the performance of non-linear filtering. It is described as where the initial state follows x 0 ∼ (0, 25), and noise processes {v k , k ≥ 0} and {e k,m , k ≥ 0} 3 m=1 are mutually independent, white and Gaussian distributed according to v k ∼ (0, 10) and e k,m ∼ (0, 1) for m = 1, 2, 3, respectively.
Suppose that the actual measurement y k,1 suffers from a twostep random delay, and y k,2 and y k,3 suffer from a three-step random delay. The concrete delay probabilities are given in Table 2.
To make a fair comparison, we conduct K = 1000 independent Monte Carlo runs with T = 500 time steps. In each Monte Carlo run, N = 200 particles are employed for all particle filtering methods, and the initial state are randomly chosen from its initial probability distribution.
We firstly use the single sensor 2 to investigate the performance of all filtering methods. For clarity, we only present the RMSE curves of the first 100 and last 50 time steps for all filtering methods in Figures 1 and 2, respectively. Table 3 displays the TARMSEs and the total running time. Clearly, SMSPFideal performs best in terms of RMSE as it uses ideal measurements. By contrast, SMSPFignored has a very low estimation accuracy. The remaining three methods, CQKFRMD, MSPFRMD and newMSPFRMD, have mechanisms to handle the multi-step RMD. But they show different performances. Both MSPFRMD and newSMSPFRMD can deal with RMD well, but the latter performs slightly better and consumes less time than the former. This is because the dependence of measurements is considered, and repeated measurements are discarded. CQKFMRD gives the worst results due to its inherent Gaussian assumption, and it requires the most running time due to state augmentation.
Since the multi-sensor CQKFRMD has not been developed yet, we use four other methods to estimate the state in the multisensor case. The curves of RMSEs for the first 100 and last 50 time steps are plotted in Figures 3 and 4, respectively, and simulation results are summarised in Table 3. Similarly to the single sensor case, SMSPFideal gives the best result, followed by newMSPFRMD. By contrast, SMSPFignored has the worst performance. Even though MSPFRMD costs more running time, its estimation error is bigger than newMSPFRMD.
Example 2: Two-dimensional radar tracking
Assume that the target is moving in the x − y plane according to the following dynamics The target is observed by two radar stations, and the corresponding measurement equations are given by where z k,m is the ideal measurement at time step k for the mth radar, which contains the radial distance r k,m and the bearing k,m for m = 1, 2. The measurement noises e k,1 and e k,2 are mutually independent white noise processes and obey e k,1 ∼ (0 2 , R k,1 ) and e k,2 ∼ (0 2 , R k,2 ), respectively, where R k,1 = R k,2 = diag(100 m 2 , 10 −6 rad 2 ∕s 2 ). It is assumed that the measurements from these two radars suffer from asynchronously random delays, and the actual measurement y k = [y k,1 , y k,2 ] satisfy (2). The maximum possible measurement delays for two sensors are set to d 1 = d 2 = 3, and the corresponding delay probabilities are given in Table 4.
The following results are for K = 500 Monte Carlo simulations with T = 500 time steps and N = 2000 particles in each Table 5 lists TARMSEs and the total running time. It is clear that SMSPFideal has the smallest estimation error, while SMSPFignored has the largest estimation error. Our proposed method, newMSPFRMD, outperforms MSPFRMD in terms of RMSEs for all four state components, position and velocity. Moreover, it can save almost 15% of the total running time compared with MSPFRMD.
CONCLUSION
A new multi-sensor particle filtering is proposed to deal with the dependence of multi-step randomly delayed measurements. The simulation results validate the superiority of this method in terms of estimation accuracy and running time. It is very general and can be applied to various cases including: (a) The single sensor with multi-step random delay. (b) Multiple sensors with different delay steps and delay probabilities. (c) Delay probabilities can be time varying for the single sensor or multiple sensors. | 3,855.6 | 2020-12-22T00:00:00.000 | [
"Engineering",
"Computer Science"
] |
LEO Onboard Real-Time Orbit Determination Using GPS/BDS Data with an Optimal Stochastic Model
The advancements of Earth observations, remote sensing, communications and navigation augmentation based on low Earth orbit (LEO) platforms present strong requirements for accurate, real-time and autonomous navigation of LEO satellites. Precise onboard real-time orbit determination (RTOD) using the space-borne data of multiple global navigation satellite systems (multi-GNSS) becomes practicable along with the availability of multi-GNSS. We study the onboard RTOD algorithm and experiments by using America’s Global Positioning System (GPS) and China’s regional BeiDou Navigation Satellite System (BDS-2) space-borne data of the FengYun-3C satellite. A new pseudo-ambiguity parameter, which combines the constant phase ambiguity, the orbit and clock offset error of the GPS/BDS broadcast ephemeris in the line-of-sight (LOS), is defined and estimated in order to reduce the negative effect of the LOS error on onboard RTOD. The analyses on the variation of the LOS error in the GPS/BDS broadcast ephemeris indicate that the pseudo-ambiguity parameter could be modeled as a random walk, and the setting of the power spectral density in the random walk model decides whether the pseudo-ambiguity can be estimated reasonably and the LOS error could be reduced or not. For different types of GPS/BDS satellites, the LOS errors show different variation characteristics, so the power spectral density should be set separately and differently. A numerical search approach is presented in this paper to find the optimal setting of the power spectral density for each type of GPS/BDS satellite by a series of tests. Based on the optimal stochastic model, a 3-dimensional (3D) real-time orbit accuracy of 0.7–2.0 m for position and 0.7–1.7 mm/s for velocity could be achieved only with dual-frequency BDS measurements and the broadcast ephemeris, while a notably superior orbit accuracy of 0.3–0.5 m for position and 0.3–0.5 mm/s for velocity is achievable using dual-frequency GPS/BDS measurements, due to the absorption effect of the estimated pseudo-ambiguity on the LOS error of the GPS/BDS broadcast ephemeris. Compared to using GPS-alone data, the GPS/BDS fusion only marginally improves the onboard RTOD orbit accuracy by about 1–3 cm, but the inclusion of BDS satellites increases the number of the tracked GNSS satellites and thus speeds up the convergence of the filter. Furthermore, the GPS/BDS fusion could help suppress the local orbit errors, ensure the orbit accuracy and improve the reliability and availability of the onboard RTOD when fewer GPS satellites are tracked in some anomalous arcs.
Introduction
In recent years, along with the continuous development of the high-resolution Earth observations, remote sensing, communication and navigation based on low Earth orbit (LEO) satellites, real-time, precise and reliable orbits of LEO platforms are crucial for many space applications, such as altimetry, gravimetry, synthetic aperture radar (SAR) interferometry, atmospheric sounding or navigation signal augmentation [1][2][3]. Precise onboard real-time orbit determination (RTOD) using the space-borne data of a global navigation satellite system (GNSS) is recognized as the mainstream navigation technology for LEO satellites, due to its global coverage, abundant observations and low cost. The onboard RTOD, as the name implies, is the process of real-time orbit determination completed in the embedded system onboard satellites. It has no dependence on ground-based tracking assets, and the orbit results are required to be delivered within minutes, seconds or even a fraction of a second after observations are made. Although high-precision GNSS orbits and high-rate clocks have been widely used in post-processing precise orbit determination (POD), they are not always available in real time for LEO satellites. Therefore, in general, only the GNSS broadcast ephemeris is available for the onboard RTOD due to its real-time and autonomous requirements.
Over the past 30 years, the onboard RTOD using the space-borne data of America's Global Positioning System (GPS) has been in continuous developments and applied to many LEO satellites. The Goddard Space Flight Center (GSFC) began research on the onboard RTOD algorithm and developed the GPS Enhanced Orbit Determination Experiment (GEODE) space flight navigation software in the 1990s [4]. The 3-dimensional (3D) position and velocity accuracies of the onboard RTOD using dual-frequency GPS pseudo-range measurements were about 7.8 m and 5.9 mm/s, respectively, when applied to process the space-borne data of the TOPEX satellite collected in the absence of selective availability (SA) [5]. The position accuracy was improved to 1.0 m when using SA-free data after Goldstein's updates to the algorithm of GEODE [6]. NASA's Jet Propulsion Laboratory (JPL) also developed the Real-Time GIPSY (RTG) software for the onboard RTOD, and the orbit results with a position accuracy of 1.5 m were obtained when RTG was applied to process the dual-frequency pseudo-range measurements of the SAC-C satellite [7]. The German Aerospace Center (DLR) developed a high-precision onboard RTOD algorithm for BIRD, X-SAT, SunSat and PROBA-2 satellites [8][9][10][11]. Tests demonstrated that the onboard RTOD could provide real-time navigation accuracy at the 1.0 m level. Montenbruck et al. [12] were the first to present the onboard RTOD algorithm using GPS carrier-phase measurements. A series of experiments were made on various LEO satellites, such as CHAMP, GRACE, TerraSAR-X, ICESat, SAC-C and MetOp, and the tests demonstrated that a real-time position accuracy of about 0.5 m was feasible with dual-frequency carrier-phase measurements. A detailed study into the error sources for the onboard RTOD was also conducted in our past research [13][14][15][16], which revealed that the orbit and clock offset errors of the GPS broadcast ephemeris in the line-of-sight (LOS) are the main factors in restricting the accuracy improvement of onboard RTOD. Therefore, a new RTOD method was presented, where a large part of the LOS error could be absorbed into a new estimated parameter, named pseudo-ambiguity, which combines the phase ambiguity, the orbit and clock offset errors of the GPS broadcast ephemeris together [15]. The experiments of processing real data of China's HY2A and ZY3 satellites demonstrated that the position and velocity accuracies of 0.3-0.5 m and 0.3-0.5 mm/s, respectively, were achieved using dual-frequency GPS carrier phases [16].
For most LEO satellites, only space-borne GPS observations are available, so the research and experiments of the onboard RTOD in the past were usually based on GPS measurements. With the construction and development of other GNSSs such as China's BeiDou Navigation Satellite System (BDS) [17], Russia's Global Navigation Satellite System (GLONASS) [18] and Europe's Galileo Navigation Satellite System (Galileo) [19], the onboard RTOD based on multiple GNSS (multi-GNSS) integration has become possible. BDS, the global navigation system developed by China, provides an additional data source to the LEO orbit determination. A few Chinese LEO satellites, such as FengYun-3C [20], are equipped with the GPS/BDS receivers. The FengYun-3C satellite, a Chinese meteorological satellite launched in 2013, was equipped with a GNSS occultation sounder (GNOS) instrument which could provide dual-frequency pseudo-range and carrier-phase data of GPS and the earlier regional BDS (BDS-2). It should be noted that the global BDS (BDS-3) was not available at that time, and the GNOS was designed only to track BDS-2 satellites, as well as GPS satellites; nevertheless, post-POD processing of the FengYun-3C satellite using GPS and BDS-2 measurements has been carried Remote Sens. 2020, 12, 3458 3 of 20 out, and high-precision orbit results at centimeter level have been obtained [21,22]. Hence, this paper focuses on the onboard RTOD algorithm and experiments for processing the space-borne GPS and BDS-2 data of the FengYun-3C satellite. In order to achieve decimeter precision, the solution using both high-precision carrier-phase and pseudo-range data, the same as the algorithm presented occasionally in previous literature [12][13][14][15][16], is still adopted, where the pseudo-ambiguity parameter combining the phase ambiguity, the orbit and clock offset errors of the GNSS broadcast ephemeris in the line-of-sight (LOS) together is defined for each visible GNSS satellite, and estimated to absorb a large part of the LOS error [15,16]. Obviously, the stochastic modeling of pseudo-ambiguity decides whether it can be estimated reasonably and the LOS error could be absorbed effectively or not. Along with the use of multi-GNSS data, it is very important to obtain the optimal stochastic models of the pseudo-ambiguity for different types of GNSS satellites whose broadcast ephemeris error may show different variation characteristics. Therefore, this paper first analyzes the variation characteristics of the LOS error for different GNSS satellites in-depth, including the GPS satellites and the BDS satellites on the geosynchronous, inclined geosynchronous and medium Earth orbits (GEO/IGSO/MEO), and then derives out the reasonable stochastic model of the pseudo-ambiguity. A numerical search approach is presented to find the optimal stochastic model setting for each type of these GNSS satellites. Furthermore, the analysis on the model optimization will be performed in detail, and a series of discussion will be conducted which presents the effect of the pseudo-ambiguity with optimal modeling on the onboard RTOD. In addition, the analyses will be carried out about the effect of GPS/BDS fusion on the availability of GNSS satellites, orbit accuracy, filter convergence and reliability of the onboard RTOD. All the onboard RTOD experiments are made with the self-developed software SATPODS (Space-borne GNSS AuTonomous Precise Orbit Determination Software). The post-high-precision (2-4 cm) POD orbits from the PANDA (Positioning and Navigation Data Analyst) software [23] will be used as a reference for the accuracy assessment of the onboard RTOD results.
In the following section, the basic algorithm of onboard RTOD will be briefly introduced first, including the dynamical models, GNSS measurements and parameter estimation equation. Secondly, the orbit and clock offset error in the line-of-sight (LOS) caused by the GPS/BDS broadcast ephemeris will be analyzed in-depth, followed by the stochastic modeling of the pseudo-ambiguity parameter. Then, a numerical search approach will be presented and tested to find the optimal setting of the stochastic model of pseudo-ambiguity. Afterward, based on the onboard RTOD experiment results with the optimal stochastic model, the absorption effect of the estimated pseudo-ambiguity on the LOS error in the GPS/BDS broadcast ephemeris will be discussed and analyzed in-depth, as well as the impact of GPS/BDS fusion on the onboard RTOD. Finally, some summarized discussions and conclusions are made.
Basic Onboard RTOD Algorithm
The onboard RTOD algorithm integrates the dynamical models and GNSS observations to make an optimal estimation of the satellite orbit along with the process noise adjusting the quality of force models. Considering the limited computation capability of onboard satellite processers and the real-time and autonomous requirements, the algorithm deserves special care in some aspects, including the simplification of dynamical models, construction of observation solutions and stochastic modeling of estimated parameters.
Dynamical Models
The equations of the motion of an LEO satellite can be expressed as [24] ..
where r, r are the position, velocity and acceleration vectors of the satellite in a geocentric inertial coordinate frame, respectively, and p represents other parameters in the force models. The total Remote Sens. 2020, 12, 3458 4 of 20 forces of the satellite include the gravity of Earth, lunisolar gravitational perturbation, solid earth tide, atmospheric drag, solar radiation pressure and so on. However, not all perturbations should be considered in onboard RTOD processing, and some force models must be simplified or neglected because of the computation capacity limit of the space-borne processor. So, three empirical accelerations in radial, tangential and normal directions (e R , e T , e N ) modeled by a first-order Gauss-Markov process are estimated in the Kalman filter to compensate those unmodeled perturbation errors. The drag and radiation coefficients (C d , C r ) will also be estimated as scaling factors to account for the inaccuracy of atmosphere drag and solar radiation pressure models. Detailed settings about the dynamical models will be described in the onboard RTOD strategies.
In summary, the estimated dynamical parameters in onboard RTOD processing are usually p = (e R , e T , e N , C d , C r ), and the state vector of an LEO satellite can be expressed as X = r, . r, p . Following Equation (1), the first-order differential equation of the state vector X can be formed easily [24]: where F(·) represents the explicit function and u . X is considered as the white noise. According to Equation (2), the discrete state propagation equation can be derived out easily [24]: where X k denotes the state vector at epoch t k , W k is considered as the white noise and Φ(·) represents the implicit state transition function. The detailed computation of the state propagation equation is well documented in the literature [25,26].
GNSS Measurements
The dual-frequency GPS/BDS receiver onboard LEO satellites can provide the dual-frequency pseudo-range (C 1 , P 2 ) and carrier-phase measurements (L 1 , L 2 ), so the ionosphere-free linear combinations of two basic pseudo-range and carrier-phase measurements are usually considered to eliminate the ionospheric error [27]: where T denotes the transformation matrix from the inertial system, in which the LEO satellite orbit is determined, to the Earth-fixed system, δ R represents the clock offsets of the receiver, r * and δ * denote the orbit and clock offset of the GNSS satellite that are calculated by using its broadcast ephemeris and c is the speed of light in vacuum. The bias B represents the combined ambiguities of two carrier phases, and ε P C and ε L C are the pseudo-range and carrier-phase measurement noises, respectively. Obviously, |T · r − r * | is the inaccurate geometric range between the LEO and GNSS satellite that are computed by using the GNSS broadcast orbit r * , so dρ is set as the range error in the line-of-sight (LOS) caused by the orbit error of the GNSS broadcast ephemeris. Similarly, dδ is set as the clock offset error of the GNSS broadcast ephemeris, so δ * + dδ represents the precise or true clock offset. Therefore, the term dρ − cdδ can be treated as the total LOS error caused by the GNSS broadcast ephemeris. It is obvious that the LOS error is unknown if only the GNSS broadcast ephemeris is available, and it is also difficult to correct or eliminate this unknown error. Without appropriate processing, the LOS error of the GNSS broadcast ephemeris would result in less accurate orbit results.
In the embedded RTOD solution capable of processing carrier-phase and pseudo-range measurements, the phase ambiguity parameter B must be estimated because of the use of carrier phases. Furthermore, the unknown LOS error should be modeled and estimated in the Kalman filter to reduce its influence [14]. If the LOS error is estimated, too, according to Equation (4), the LOS error and ambiguity parameter are coupled and can be combined as one parameter. So, a new parameter that accounts for the LOS error dρ − cdδ and the ambiguity B is defined, which is called the "pseudo-ambiguity" [15]. The pseudo-ambiguity A and the observation equation can be improved as [15] where it should be noted that the pseudo-range measurement P C is not related to the pseudo-ambiguity A. All pseudo-ambiguities can be set as a vector a = (A 1 , A 2 , . . . , A n ), where n denotes the number of tracked GNSS satellites. Then, the state vector of an LEO satellite expands to X = r, . r, p, h, a , where h is the receiver clock parameters, including the clock offset δ R , the clock rate . δ R and the system bias µ between GPS and BDS if the GPS/BDS data are processed in fusion. The pseudo-range measurement P C cannot be used to update the state X in the Kalman filter directly. Therefore, the observation equation can be expressed as [15] where l(·) denotes the related function between the carrier-phase measurement L C and the state vector X. If the pseudo-ambiguity is estimated with a reasonable stochastic model, the effect of the LOS error can be reduced, and then the accuracy of the onboard RTOD can be improved significantly. Obviously, it is very important to model the variation characteristics of the pseudo-ambiguity correctly.
Parameter Estimation
In total, the state vector of the LEO satellite can be listed as X = r, . (6), the Kalman filter equation can be set up easily [24]:
r, p, h, a . Based on the state propagation Equation (3) and GNSS observation Equation
where Y k denotes the measurement vector, Φ(·) and H(·) represent the dynamical and observation equations, respectively, W k is the process noise, which is usually treated as the white noise, but for different to-be-estimated parameters, the variance is set differently, and V k is the error of measurements. It should be noted that, of all parameters, the first-order Gauss-Markov random process is used to describe the state propagation of three empirical accelerations, and the coefficients of atmosphere drag and solar radiation pressure are modeled by two random walk processes. Furthermore, each pseudo-ambiguity should be represented by a reasonable stochastic model, too. The dynamical and observation equations are highly non-linear. Thus, the extended Kalman filter is used to estimate the unknown parameters. The unified formula of the extended Kalman filter can be derived out [24]: where Y k is the measurement vector, Φ(·) and H(·) represent the dynamical and observation functions, respectively, and w k is the process noise, which is usually treated as the white noise but with different variances for different to-be-estimated parameters. The linearization is based on the nominal filter state X * k−1 at the epoch t k−1 , which usually adopts the estimated stateX k−1 to reduce the linearization error. X * k = Φ X * k−1 is the one-step predicted state from epoch t k−1 → t k . φ k,k−1 = ∂X k /∂X k−1 is the state transition matrix. Both X * k and φ k,k−1 are computed by numerical integration. x k = X k − X * k is the estimated correction of the filter state X k relative to the predicted state X * k . H X * k is the computed measurements and y k = Y k − H X * k is the prior residuals equal to the observed measurements minus computed measurements. h k is the design matrix, also known as the observation matrix and v k is the observation error. The detailed linearization formulas and the proceeding of the extended Kalman filter are well documented in the literature [25,26].
Pseudo-Ambiguity Optimal Stochastic Modeling
A reasonable stochastic model of the pseudo-ambiguity parameter is crucial to the absorption of the LOS error in the GNSS broadcast ephemeris [14,15]. In the following section, the variation characteristics of the LOS error will be first analyzed in-depth for different GPS/BDS satellites, and then the optimal stochastic modeling of the pseudo-ambiguity will be elaborated in detail.
Pseudo-Ambiguity Modeling
According to the definition in Equations (4) and (5), the pseudo-ambiguity consists of two parts: the true ambiguity B and the LOS error dρ − cdδ. The true ambiguity B is constant by its nature during uninterrupted signal tracking, so the stochastic model of the pseudo-ambiguity depends on the variation characteristic of the LOS error. From the definition, the LOS error is related to both the transmitter (GNSS) and receiver (LEO). Several Analysis Centers (ACs) of the International GNSS Service (IGS) have released the high-precision GNSS orbit and clock products with the accuracy at centimeter level, which is two orders of magnitude higher than that of the GNSS broadcast ephemeris at meter level [28]. Taking the FengYun-3C satellite as the example, precise GNSS ephemeris products released by the German Geosciences Research Center (GFZ) are used as references to compute and assess the LOS error caused by the GPS/BDS broadcast ephemeris. Due to the extremely high accuracy, the use of the precise GNSS ephemeris from other recognized IGS ACs will result in the same analysis and assessment results.
In consideration of different types of GPS/BDS satellites on different orbits, sub-graphs (a), (b), (c) and (d) of Figure 1 show the LOS errors of the four types of GPS/BDS satellites on day of year (DOY) 71, 2015. The thick lines represent the GPS or BDS satellites being tracked, and the intervals between consecutive lines indicate that the GPS/BDS satellite is not visible. Only for the tracking arc, the LOS error can be computed. The main property of the LOS error is that the LOS error in most tracking arcs is smooth and slowly changing regardless of the type of GPS/BDS satellite. In sub-graphs (a) and (b), the LOS error curves of two GPS satellites, G01 and G02, are displayed, and it is seen clearly that, whether being G01 or G02, the LOS error has the main property. Similarly, for C02 and C03 of BDS GEO, C10 and C09 of BDS IGSO and C14 and C12 of BDS MEO, the property holds. However, it cannot be ignored that there is a jump for some arcs. The second tracking arc of G01, the first one of C02, the third one of C10 and the seventh of C14 are zoomed in and showed in sub-graph (c), while the third one of G02, the second one of C03, the second one of C09 and the fourth of C12 are shown in sub-graph (d). As can be observed more clearly, for the C10, C12 and C14 satellites, the curves of the LOS errors are discontinuous and the jumps appear. This phenomenon was called the "ephemeris switch". The reason for the jump is that the GNSS broadcast ephemeris is released every one or two hours and the orbit error has a jumping change when a new ephemeris is used. Since two satellites in a type (e.g., GPS or BDS MEO) of GNSS satellite have nearly the same error property, it is therefore reasonable to only choose G01, C02, C10 and C14, respectively, to represent GPS, BDS GEO, BDS IGSO and BDS MEO in the following analysis. In fact, choosing other satellites would result in almost the same conclusion.
On the basis of the practical analysis in the above, the random walk process is used to represent the pseudo-ambiguity in a tracking arc. For the epoch with the ephemeris switch, the sudden jump of the LOS error can be modeled by enlarging the variance of the corresponding pseudo-ambiguity additionally. For one pseudo-ambiguity A, the random walk process can be expressed by a differential equation and the state propagation can be derived out easily [29]: Remote Sens. 2020, 12, 3458 7 of 20 where u is a white noise, σ 2 is the power spectral density of the process noise, ∆t is the epoch interval and A k represents the pseudo-ambiguity at the epoch t k . w k is considered as white noise and Q k = σ 2 ∆t is the variance of w k .
Stochastic Model Setting
Obviously, the power spectral density 2 is an essential parameter for the stochastic model of the pseudo-ambiguity. The setting of 2 decides whether the pseudo-ambiguity can be estimated reasonably and the LOS error could be reduced or not. If 2 is too small, the pseudo-ambiguity will degenerate to a constant phase ambiguity, which means the LOS error can be estimated and absorbed. Conversely, if 2 is too large, the constraint on the pseudo-ambiguity will be too loose, which means other errors may be absorbed into the estimated pseudo-ambiguity. All in all then, the inappropriate 2 is harmful to the absorption of the LOS error, and then leads to the decrease in the RTOD accuracy. In theory, the variance of k w depends on the variation of the LOS error. Within a tracking arc, the power spectral density 2 could be set based on the change rate of the true LOS error. An approximate relation between the change rate and the power spectral density 2 can be derived easily:
Stochastic Model Setting
Obviously, the power spectral density σ 2 is an essential parameter for the stochastic model of the pseudo-ambiguity. The setting of σ 2 decides whether the pseudo-ambiguity can be estimated reasonably and the LOS error could be reduced or not. If σ 2 is too small, the pseudo-ambiguity will degenerate to a constant phase ambiguity, which means the LOS error can be estimated and absorbed. Conversely, if σ 2 is too large, the constraint on the pseudo-ambiguity will be too loose, which means other errors may be absorbed into the estimated pseudo-ambiguity. All in all then, the inappropriate σ 2 is harmful to the absorption of the LOS error, and then leads to the decrease in the RTOD accuracy. In theory, the variance of w k depends on the variation of the LOS error. Within a tracking arc, the power Remote Sens. 2020, 12, 3458 8 of 20 spectral density σ 2 could be set based on the change rate υ of the true LOS error. An approximate relation between the change rate υ and the power spectral density σ 2 can be derived easily: The root mean square (RMS) statistics of the change rate υ for different GPS/BDS satellites on DOY 69-75, 2015, and the relevant power spectral density σ 2 with the epoch interval of 30s (∆t = 30s) are listed in Table 1. The statistics values of υ, σ 2 for different types of GNSS satellites are different, and for BDS satellites they are larger than that of GPS satellites. In addition, the change rates of BDS IGSO satellites are close to that of BDS MEO satellites. Obviously, in order to reduce the LOS error effectively, it ought to be set appropriately and differently for different types of GNSS satellites. However, it should be noted that the statistics value in Table 1 is not necessarily the optimal setting for the onboard RTOD. Therefore, a numerical search approach is adopted here to obtain the optimal setting of σ 2 . The procedure of this approach is actually a series of numerical tests. First, the change rate υ varies step by step around the statistics value. Thus, the relevant power spectral density σ 2 will also change gradually. With the power spectral density σ 2 of different values, the corresponding onboard RTOD tests are carried out, and then the σ 2 value resulting in the best orbit accuracy will be regarded as the optimal setting. Obviously, the optimal settings of three σ 2 parameters σ 2 G , σ 2 Cg , σ 2 Cim should be searched separately for three types of GNSS satellites, namely GPS, BDS GEO and BDS IGSO/MEO satellites. According to Table 1, the statistics value of the change rate υ of the LOS error for different GNSS satellites is about 1.0 × 10 −3 -2.0 × 10 −3 m/s, so the values of three parameters Cim are set to be ranging from 10 −9 -10 +1 m 2 /s along with the variation of change rate υ in the interval of 1.0 × 10 −5 -7.5 × 10 −1 m/s. In total, 20 sets of values are tested, which are listed in Table 2 and marked as S01-S20 for convenient expression in the following.
Orbit Results Analysis
A series of numerical tests will be carried out to search the optimal setting of the stochastic model of the pseudo-ambiguity parameter for each type of GPS/BDS satellite. Based on the onboard RTOD test results with the optimal stochastic model of pseudo-ambiguity, the absorption effect of the estimated pseudo-ambiguity on the LOS error in the GNSS broadcast ephemeris will be analyzed in-depth. At the same time, the analyses will be performed on the impact of GPS/BDS fusion on the availability of GNSS satellites, orbit accuracy, filter convergence and reliability of the onboard RTOD. In addition, it should be noted that the space-borne GPS/BDS data of the FengYun-3C satellite are not publicly available for users, and the data we can use are limited, so only three datasets that cover from DOY 69, 2015, through DOY 75, 2015, from DOY 33, 2018, through DOY 39, 2018, and from DOY 275, 2013, through DOY 300, 2013, will be processed by SATPODS to verify the effect of GPS/BDS fusion. It must be emphasized that the GNSS receiver onboard FengYun-3C can only track GPS satellites and the GEO, IGSO and MEO satellites of the earlier regional BDS-2, as per the receiver design back to 2013. In other words, the receiver on FengYun-3C is not able to track the operating BDS-3 satellites. At the same time, the main property of LOS errors caused by the GNSS broadcast ephemeris, shown in Figure 1, remains unchanged over the time. Therefore, the stochastic models of the pseudo-ambiguity are applicable to GNSS tracking data in any time, albeit the optimal settings are needed from time to time, which motivated the research presented in this paper. Hence, it can be inferred that the developed optimal setting method, which is experimented with these three datasets in 2013, 2015 and 2018, is generally applicable to any data at any other time, as long as the error property of the data is the same.
Numerical Search Tests
In the numerical search tests, the datasets of the FengYun-3C satellite that cover from DOY 69, 2015, through DOY 75, 2015, are processed by SATPODS. The software is capable of simulating the onboard operation scene. The strategies take the autonomy, timing and accuracy requirements of onboard RTOD into account. Only the GPS/BDS broadcast ephemeris is used, and the pseudo-ambiguities are estimated instead of the true constant ambiguities. Then, the dynamical models are simplified to the maximum extent to reduce the computational load. At the same time, the neglected perturbations, including the ocean and pole tide, earth radiation and relativistic effect, are small enough and have no notable effect on the real-time orbit accuracy. In addition, only simplified precession and nutation models and rapid predicted Earth orientation parameters (EOP) are applied for the transformation of coordinate systems. All the specific model settings are listed in Table 3. Furthermore, in order to assess the orbit accuracies from SATPODS, the precise orbit results generated by PANDA are used as references. The accuracy of the post-POD orbit results generated by PANDA are at the 2-4 cm level [23], so they can be treated as the reference to assess the real-time orbit accuracy. The optimal values of σ 2 G , σ 2 Cg , σ 2 Cim will be determined by the onboard RTOD tests with four types of solutions: (a) dual-frequency GPS ionosphere-free phase and pseudo-range measurements are used, and the solution is abbreviated as "GPS-alone"; (b) dual-frequency measurements of GPS and BDS IGSO/MEO satellites are used, and the solution is shorted as "GPS+BDSN", where BDSN means the BDS GEO satellites are not included; (c) the solution using dual-frequency measurements of GPS and BDS satellites is shorted as "GPS+BDS"; (d) the solution only using BDS dual-frequency measurements is abbreviated as "BDS-alone". The orbit accuracy of the onboard RTOD test will change when the values of σ 2 vary. The orbit accuracies (3D RMS) of the onboard RTOD trials for the GPS-alone and GPS+BDSN solutions in DOY 69-75, 2015, are shown in Figure 2, and the results of the GPS+BDS and BDS-alone solutions are illustrated in Figure 3. As can be observed clearly, the orbit accuracy of the GPS-alone solution is improved from 1.757 to 0.381 m monotonically when σ 2 G increases from S01 to S11, but the accuracy deteriorates from 0.381 to 2.256 m if σ 2 G increases from S11 to S20 continuously. So, for the GPS-alone solution, S11 is the optimal value of σ 2 G , which corresponds to the best orbit accuracy of 0.381 m for 3D position. For the GPS+BDSN solution, σ 2 G is fixed with S11 being the optimal value of the GPS-alone solution, but σ 2 Cim varies from S01 to S20. The optimal σ 2 Cim is again S11 with which the best orbit accuracy of 0.349m is achieved. For the GPS+BDS solution, σ 2 G also takes the optimal value S11, and both σ 2 Cg and σ 2 Cim range from S01 to S20, therefore 400 trials will be conducted. The orbit accuracies of all 400 trials are shown in 20 curves, where each curve corresponds to a fixed value of σ 2 Cg and 20 values of σ 2 Cim . When σ 2 Cg , σ 2 Cim take the values of (S13, S11), the best orbit accuracy of 0.348 m is obtained. Similarly, for the BDS-alone solution, both σ 2 Cg and σ 2 Cim change from S01 to S20, and the best accuracy of 0.899 m is achieved when σ 2 Cg , σ 2 Cim take the value of (S12, S09). Table 4 lists the optimal values of σ 2 for the four solutions and the corresponding best orbit accuracies. In the GPS+BDS solution, the optimal values of σ 2 for the BDS GEO and IGSO/MEO satellites are (S13, S11). The optimal setting for BDS IGSO/MEO satellites is the same as that of GPS satellites. The optimal value for BDS IGSO/MEO satellites in the GPS+BDSN solution is also the same as that in the GPS+BDS solution. This proves the consistency of pseudo-ambiguity stochastic models for the same types of GNSS satellites. However, if only BDS measurements are used, the optimal values of σ 2 Cg , σ 2 Cim are (S12, S09), not (S13, S11), and a merely inferior orbit accuracy of 1.164 m is obtained if (S13, S11) are adopted. The different optimal values of σ 2 for BDS satellites in the GPS+BDS and BDS-alone solutions demonstrate that the stochastic model setting of pseudo-ambiguity not only depends on its own characteristics but also relates to the measurements in the Kalman filter. In addition, the optimal setting of pseudo-ambiguity does not match the statistics value of υ, σ 2 based on the true LOS error in Table 1 very well. This also illustrates that the optimal setting of the stochastic model in the filter seems to be correlated with multiple factors, such as the characteristics of estimated parameters, the measurement errors, data distribution and so on.
Effect of Pseudo-Ambiguity
The numerical search tests indicate that the real-time orbit accuracy (3D RMS) of superior to 0.4 m for position could be achievable with the optimal stochastic model of pseudo-ambiguity in the GPS + BDS solution, which is much better than not only the accuracy of 1.0 m level in the solution of using single pseudo-range data [25] but also that of those GPS + BDS solutions with an inferior stochastic model setting. One reason for the accuracy improvement is that the measurement noise of carrier-phase data is very low. More importantly, the estimated pseudo-ambiguities absorb a large part of the LOS error caused by the GNSS broadcast ephemeris. Figure 4 shows the original true LOS error and the estimated LOS error fused in the pseudo-ambiguity of G01/C02/C10/C14 satellites on DOY 71, 2015. To separate the estimated LOS error from the pseudo-ambiguity, the true ambiguity in the pseudo-ambiguity is computed by the post-POD. As can be seen clearly, the two LOS error curves in sub-graph (a) present the same trend, indicating that the LOS error caused by the GNSS broadcast ephemeris is well absorbed through the estimate of the pseudo-ambiguity. The two LOS error curves in several special tracking arcs as shown in Figure 1 are also zoomed in and presented in sub-graph (b). Obviously, the LOS errors for both GPS and BDS satellites are well absorbed. Even though there is an ephemeris switch for the latter two arcs, the LOS errors are still well estimated due to enlarging the variance of the corresponding pseudo-ambiguity. All results demonstrate that the estimated pseudo-ambiguity can absorb the LOS error and improve the real-time orbit accuracy effectively. Figure 5 shows the overall statistics of the original LOS errors and the residuals after pseudo-ambiguity estimation for each type of GPS/BDS satellite at DOY 69-75, 2015. As can be observed, the original LOS error statistic (RMS) of GPS satellites is 0.869 m and most of the LOS errors are absorbed into the estimated pseudo-ambiguity parameters, so the residual LOS errors are reduced to 0.236 m. The original LOS errors of BDS satellites are up to 3-5 m, which is caused by the poor orbit and clock accuracy of the BDS broadcast ephemeris, and the different reference clock in the broadcast ephemeris and GFZ's precise clock products. Although the difference of the reference clocks is not excluded in the original LOS errors, it integrates into the estimated receiver clock and has no effects on the absorption of the pseudo-ambiguity. According to the statistics, the residual LOS errors of all BDS satellites are only 0.197 m, of which 0.172 m for GEO and 0.207 m for IGSO/MEO satellites. The reason for the slightly smaller residual LOS error of BDS satellites compared to that of GPS satellites is probably that the BDS measurements do not play a leading role for the GPS/BDS fusion solution. When GPS measurements are used, other estimated common parameters such as position and receiver clock are constrained tightly. Thus, with the constrained common parameters, the LOS error could be absorbed into the estimated pseudo-ambiguities of BDS satellites more effectively once BDS measurements are added. As a whole, the residual LOS errors of GPS/BDS satellites are reduced to the 0.2 m level due to the absorption effect of the pseudo-ambiguity, which is a contributing factor to the orbit results with the accuracy of 0.348 m for the GPS + BDS solution.
Impact of GPS/BDS Fusion
The most intuitive difference of different GPS/BDS fusion solutions is reflected in the number of tracked GNSS satellites, which is shown in Figure 6. For the GPS/BDS receiver onboard the FengYun-3C satellite, only the BDS-2 regional navigation system could be tracked. Thus, it can only observe four-six BDS satellites in the Asia/Pacific region and less than four satellites in other regions for the BDS-alone solution. GPS is a global navigation system with full operational capability, so more than four GPS satellites can be tracked by FengYun-3C in most places of the world. Compared to the GPS-alone or BDS-alone solution, the GPS + BDSN and GPS+BDS solutions increase the number of tracked GNSS satellites notably, especially in the Asia/Pacific region where GEO/IGSO satellites of BDS are tracked easily and two-six BDS satellites are added.
One should note that the increase in the number of tracked GNSS satellites does not mean the same increase for the real-time orbit accuracy. The daily position and velocity accuracies (3D RMS) with four GPS/BDS fusion solutions at DOY 69-75, 2015, DOY 33-39, 2018, and DOY 275-300, 2013, are shown in Figure 7, respectively. The label "All" is for the whole interval. According to sub-graphs (a) and (b), if only using BDS measurements, due to a few tracked BDS satellites and the obvious orbit and clock errors caused by the BDS broadcast ephemeris, the daily orbit accuracy is between 0.7 and 2.0 m for position and 0.7 and 1.7 mm/s for velocity. If only using GPS measurements, the daily orbit accuracy is improved dramatically by 0.3-0.5 m for position and 0.3-0.5 mm/s for velocity. Obviously, the orbit accuracy of using GPS-alone data is fairly superior to that of using BDS-alone data. However, the orbit accuracies of GPS + BDSN and GPS + BDS solutions are only slightly better than those of using GPS-alone data. For example, the overall 3D position accuracies of using GPS measurements at DOY 69-75, 2015, and DOY 33-38, 2018, are 0.381 and 0.392 m, while those of using GPS + BDS measurements are 0.348 and 0.363 m, which are only improved by 3.3 and 2.9 cm, respectively. The same conclusion can also be obtained from the orbit accuracy comparisons at DOY 275-300, 2013, as shown in sub-graph (c). The overall position accuracy in the whole 26-day interval for the GPS-alone and GPS + BDS solutions is 0.440 and 0.430 m, respectively, where the accuracy is only marginally better by 1.0 cm. The daily results demonstrate that the inclusion of BDS measurements could only slightly improve the accuracy by 1-3 cm, and if GPS measurements have been already used, the BDS measurements do not play a major role in the GPS/BDS fusion solutions. In addition, it should be noted that the difference between the GPS + BDSN and GPS + BDS solutions is also small. This indicates that there is no obvious orbit accuracy degradation or improvement, no matter whether BDS GEO satellites are involved or not.
The small increase seems to be of little significance or even meaningless from the view of accuracy, but the fusion of GPS/BDS data is able to speed up the convergence of the Kalman filter. Figure 8 shows the 3D position error curves of the four GPS/BDS fusion solutions at the convergent stage of the filter. If only using BDS measurements, the filter needs about 4 h for the position error to converge to less than 1.0 m, and it requires about 1 h to converge to 1.0 m when only using GPS measurements, while the convergent time decreases dramatically to 0.5 h for the GPS/BDS fusion solutions. More importantly, the main advantage of the GPS/BDS fusion is not the orbit accuracy improvement or shortened convergence time, but the improved reliability of the onboard RTOD. Figure 9 shows
Impact of GPS/BDS Fusion
The most intuitive difference of different GPS/BDS fusion solutions is reflected in the number of tracked GNSS satellites, which is shown in Figure 6. For the GPS/BDS receiver onboard the FengYun-3C satellite, only the BDS-2 regional navigation system could be tracked. Thus, it can only observe four-six BDS satellites in the Asia/Pacific region and less than four satellites in other regions for the BDS-alone solution. GPS is a global navigation system with full operational capability, so more than four GPS satellites can be tracked by FengYun-3C in most places of the world. Compared to the GPSalone or BDS-alone solution, the GPS + BDSN and GPS+BDS solutions increase the number of tracked GNSS satellites notably, especially in the Asia/Pacific region where GEO/IGSO satellites of BDS are tracked easily and two-six BDS satellites are added. Figure 7, respectively. The label "All" is for the whole interval. According to sub-graphs (a) and (b), if only using BDS measurements, due to a few tracked BDS satellites and the obvious orbit and clock errors caused by the BDS broadcast ephemeris, the daily orbit accuracy is between 0.7 and is also small. This indicates that there is no obvious orbit accuracy degradation or improvement, no matter whether BDS GEO satellites are involved or not. The small increase seems to be of little significance or even meaningless from the view of accuracy, but the fusion of GPS/BDS data is able to speed up the convergence of the Kalman filter. Figure 8 shows the 3D position error curves of the four GPS/BDS fusion solutions at the convergent stage of the filter. If only using BDS measurements, the filter needs about 4 h for the position error to converge to less than 1.0 m, and it requires about 1 h to converge to 1.0 m when only using GPS measurements, while the convergent time decreases dramatically to 0.5 h for the GPS/BDS fusion solutions. More importantly, the main advantage of the GPS/BDS fusion is not the orbit accuracy improvement or shortened convergence time, but the improved reliability of the onboard RTOD.
Application Prospect Discussion
A numerical search approach is presented in this paper to acquire the optimal setting of the power spectral density for the estimated pseudo-ambiguity parameters of different types of GNSS satellites. With the optimal setting, the real-time orbit accuracies (3D RMS) of 0.3-0.5 m for position and 0.3-0.5 mm/s for velocity are obtained for the FengYun-3C satellite when processing the space-borne GPS/BDS data. In practical applications, the similar numerical search tests could be carried out for a certain LEO satellite equipped with a GNSS receiver first, and then the searched optimal power spectral density of the pseudo-ambiguity can be adopted and uploaded for the onboard RTOD processing of this LEO satellite. Since the space-borne GNSS data condition and orbital characteristics for a specific LEO satellite and the LOS error variation in the broadcast ephemeris of a GNSS generally remain stable within a certain period, the re-prepared optimal value would be effective for a long time. Therefore, the numerical search approach is applicable to obtain the optimal stochastic model for onboard RTOD processing.
The GPS/BDS fusion could increase the number of tracked GNSS satellites effectively. Although the GPS/BDS fusion only improves the real-time orbit accuracy slightly by 1-3 cm, it could speed up the convergence of the onboard RTOD filter significantly. More importantly, it could make more GNSS satellites be observed to suppress the local variation of real-time orbit errors and improve the reliability and availability of the onboard RTOD, especially in some abnormal arcs where only a few GPS satellites are tracked. Due to the comprehensive advantage of multi-GNSS fusion in the availability of GNSS satellites, orbit accuracy, filter convergence and reliability of the onboard RTOD, more and more LEO satellites, even the large-scale LEO constellations, are expected to be equipped with the space-borne multi-GNSS receiver. In addition, although only the poor real-time orbit accuracy with 0.7-2.0 m for position and 0.7-1.7 mm/s for velocity could be achieved for the BDS-alone solution due to the few tracked regional BDS-2 satellites, it is expected to be improved along with the utilization of the BDS-3 global navigation system onboard LEO missions.
Summary and Conclusions
We present the onboard RTOD algorithm and experiments using space-borne GPS/BDS measurements of the FengYun-3C satellite. In the RTOD algorithm, a new pseudo-ambiguity parameter, which combines the constant phase ambiguity, the orbit and clock offset error of the GPS/BDS broadcast ephemeris in the line-of-sight (LOS), is defined and estimated in order to reduce the negative effect of the LOS error on onboard RTOD. The stochastic model setting of pseudo-ambiguity decides whether the pseudo-ambiguity can be estimated reasonably and the LOS error could be reduced or not. For different types of GPS/BDS satellites, the LOS errors show different variation properties, and accordingly, the stochastic model of pseudo-ambiguity is set separately and differently. A numerical search approach is presented to obtain the optimal setting of the power spectral density for each type of GNSS satellite by a series of tests. With the optimal setting, the best real-time orbit accuracies (3D RMS) of 0.3-0.5 m for position and 0.3-0.5 mm/s for velocity are achievable when using GPS/BDS carrier-phase and pseudo-range measurements, which are much better than those of the solutions with the inappropriate setting. The error analysis illustrates that these notable accuracy improvements result from the absorption effect of the pseudo-ambiguity, namely that the large part of the LOS error caused by the GNSS broadcast ephemeris is fused into the estimated pseudo-ambiguity.
The different combinations of multi-GNSS have a different influence on the onboard RTOD. If only using BDS measurements, the position and velocity accuracies are at the 0.7-2.0 m and 0.7-1.7 mm/s levels, respectively. The poor performance is caused by the few tracked satellites for BDS which is only a regional system and the obvious orbit and clock errors in the broadcast ephemeris. However, compared with the GPS-alone solution, GPS/BDS fusion only increases the real-time orbit accuracy slightly by 1-3 cm. Despite the marginally improved accuracy, GPS/BDS fusion could speed up the convergence of the Kalman filter, and the time for the position error to converge to 1.0 m is shortened to about 0.5 h for the GPS/BDS fusion compared to 4.0 h for using BDS-alone data and 1.0 h for GPS-alone data. Furthermore, the main advantage of the GPS/BDS fusion is not the orbit accuracy improvement or shortened convergence time, but rather the ability to make more satellites be observed to suppress the local variation of orbit errors and improve the reliability and availability of the onboard RTOD, which is particularly important when only a few GPS satellites are tracked in some abnormal arcs.
In summary, with the optimal stochastic model of the pseudo-ambiguity, the real-time orbit accuracy at 0.3-0.5 m is feasible for the onboard RTOD using space-borne GPS/BDS carrier-phase and pseudo-range measurements and the broadcast ephemeris. The onboard RTOD algorithm and software with refined multi-GNSS data processing have been developed in this paper. With the continuous development of Earth observations, remote sensing, communications and navigation augmentation based on LEO satellites, the onboard RTOD system is expected to fly on more LEO missions to provide a real-time service.
Author Contributions: In this study, X.G. designed the algorithm, performed experiments and wrote this paper. J.S. and F.W. designed the software, performed experiments, analyzed data and edited the manuscript. X.L. supervised its analysis and edited the manuscript. All authors have read and agreed to the published version of the manuscript.
Funding: This research was funded by the National Natural Science Foundation of China (Nos. 91638203). | 11,370.8 | 2020-10-21T00:00:00.000 | [
"Engineering",
"Environmental Science",
"Physics"
] |
Variational Disentanglement for Rare Event Modeling
Combining the increasing availability and abundance of healthcare data and the current advances in machine learning methods have created renewed opportunities to improve clinical decision support systems. However, in healthcare risk prediction applications, the proportion of cases with the condition (label) of interest is often very low relative to the available sample size. Though very prevalent in healthcare, such imbalanced classification settings are also common and challenging in many other scenarios. So motivated, we propose a variational disentanglement approach to semi-parametrically learn from rare events in heavily imbalanced classification problems. Specifically, we leverage the imposed extreme-distribution behavior on a latent space to extract information from low-prevalence events, and develop a robust prediction arm that joins the merits of the generalized additive model and isotonic neural nets. Results on synthetic studies and diverse real-world datasets, including mortality prediction on a COVID-19 cohort, demonstrate that the proposed approach outperforms existing alternatives.
INTRODUCTION
Early identification of in-hospital patients who are at imminent risk of life-threatening events, e.g., death, ventilation or intensive care unit (ICU) transfer, is a critical subject in clinical care (Bedoya et al. 2019).Especially during a pandemic like COVID-19, the needs for healthcare change dramatically.With the ability to accurately predict the risk, an automated triage system will be well-positioned to help clinicians better allocate resources and attention to those patients whose adverse outcomes can be averted if early intervention efforts were in place.
Despite the great promise it holds, with the richness of modern Electronic Health Record (EHR) repositories, the construction of such a system faces practical challenges.A major obstacle is the scarcity of patients experiencing adverse outcomes of interest.In the COVID-19 scenario, which we consider in our experiments, the mortality of patients tested positive at the Duke University Health System (DUHS) is slightly lower than 3%.Further, in another typical EHR dataset we consider, less than 5% of patients are reported to suffer adverse outcomes (ICU transfer or death).In these low-prevalence scenarios, commonly seen in clinical practice, standard classification models such as logistic regression suffer from majority domination, in which models tend to favor the prediction accuracy of majority groups.This is clearly undesirable for critical-care applications, given the high false negative rates (Type-II error), in which patients in urgent need of care could be falsely categorized.
Situations where the distribution of labels is highly skewed and the accuracy of the minority class bears particular significance (Dal Pozzolo et al. 2017;Lu, Guo, and Li 2020;Machado and Lopes 2020) have been associated with the name imbalanced dataset (He and Garcia 2009), whereas the methods dealing with such cases are coined extreme classification (Zong, Huang, and Chen 2013).Under such a setting, the lack of representation of minority cases severely undermines the ability of a standard learner to discriminate, relative to balanced datasets (Mitchell 1999).Consequently, these solutions do not generalize well on minority classes, where the primary interest is usually focused.
To address such a dilemma, several remedies have been proposed to account for the imbalance between class representations.One of the most popular strategies is the sampling-based adjustment, where during training, a model oversamples the minority classes (or undersamples the majority classes) to create balance artificially (Drummond, Holte et al. 2003).To overcome the biases and the lack of information that naive sampling adjustments might induce, variants have been proposed to maximally preserve the clustering structure of the original dataset (Mani and Zhang 2003;Yen and Lee 2009) and to promote diversity of oversampling schemes (Han, Wang, and Mao 2005).Alternatively, cost-sensitive weighting where minority losses are assigned larger weights provides another popular option via tuning the relative importance of minority classes (Elkan 2001;Munro et al. 1996;Zhou and Liu 2005).
While the above two strategies introduce heuristics to alleviate the issues caused by class imbalance, importance sampling (IS) offers a principled treatment that flexibly combines the merits of the two (Hahn and Jeruchim 1987;Heidelberger 1995).Each example is sampled with the probability of a pre-specified importance weight, and with the weight's inverse when accounting for the relative contribution in the overall loss.This helps to flexibly tune the representation of rare events during training, without biasing the data distribution (Heidelberger 1995;Shimodaira 2000;Gretton et al. 2009).It is important to note that, poor choice of importance weights may result in uncontrolled variance that destabilizes training (Robert and Casella 2013;Botev and Kroese 2008), calling for adaptive (Rubinstein and Kroese 2013) or variance reduction schemes (Rubinstein and Kroese 2016) to protect against such degeneracy.
Apart from the above strategies that fall within the standard empirical risk minimization framework, recent developments explicitly seek better generalization for the minority classes.One such example is the one-class classification that aims to capture one target class from a general population (Tax 2002).Meta-learning and few-shot learning strategies instead trying to transfer the knowledge learned from data-rich classes to facilitate the learning of data-scarce classes (Böhning, Mylona, and Kimber 2015;Finn, Abbeel, and Levine 2017).Additionally, non-cross-entropy based losses or penalties have been proved useful to imbalanced classification tasks (Weinberger and Saul 2009;Huang et al. 2016).For instance, the Focal loss (Lin et al. 2017) upweights the harder examples, and Cao et al. (2019) introduced a label-distribution-aware margin loss encouraging minority classes having larger margins.
In this work, we present a novel solution called variational inference for extremals (VIE), capitalizing on the learning of more generalizable representations for the minority classes.Our proposal is motivated by the observation that the statistical features of "rarity" have been largely overlooked in the current literature of rare-event modeling.And the uncertainties of rare-events are often not considered.Framed under the Variational Inference framework, we formulate our model with the assumption that the extreme presentation of (unobserved) latent variables can lead to the occurrence (or the inhibition) of rare events.This encourages the accurate characterization of the tail distribution of the data representation, which has been missed by prior work to the best of our knowledge.Building upon the state-of-the-art machine learning techniques, our solution features the following contributions: (i) the model accounts for representation uncertainty based on variational inference; (ii) the adoption of mixed Generalized Pareto priors to promote the learning of heavy-tailed feature representations; and (iii) integration of additive isotonic regression to disentangle representation and facilitate generalization.We demonstrate how our framework facilitates both model generalization and interpretation, with strong empirical performance reported across a wide-range of benchmarks.
BACKGROUND
To simplify our presentation, we focus on the problem of rare event classification for binary outcomes.The generalization to the multiple-class scenario is simple and presented in the Supplementary Material (SM)1 .Let D = {x i , y i } N i=1 be a dataset of interest, where x i and y i denote predictors and outcomes, respectively, and N is the sample size.Without loss of generality, we denote y = 1 as the minority event label (indicating the occurrence of an event of interest), and y = 0 as the majority label.
In the following, we will briefly review the three main techniques we used in this work, namely, variational inference (VI), extreme value theory (EVT), and additive isotonic regression.VI allows for approximate maximum likelihood inference while accounting for data uncertainty.EVT provides a principled and efficient way to model extreme, heavy-tailed representations.Additive isotonic regression further introduces monotonic constraints to disentangle the contribution of each latent dimension to the outcome.
Variational inference
Consider a latent variable model p θ (v, z) = p θ (v|z)p(z), where v ∈ R m is the observable data, z ∈ R p is the unobservable latent variable, and θ represents the parameters of the likelihood model, p θ (v|z).The marginal likelihood p θ (v) = p θ (v, z)dz requires integrating out the latent z, which typically, for complex distributions, does not enjoy a closed-form expression.This intractability prevents direct maximum likelihood estimation for θ in the latent variable setup.To overcome this difficulty, Variational Inference (VI) optimizes computationally tractable variational bounds to the marginal log-likelihood (Kingma and Welling 2014;Chen et al. 2018).Concretely, the most popular choice of VI optimizes the following Evidence Lower Bound (ELBO): where q φ (z|v) is an approximation to the true (unknown) posterior p θ (z|v), and the inequality is a direct result of Jensen's inequality.The variational gap between the ELBO and true marginal log-likelihood, i.e., log p θ (v) − ELBO(v; p θ (v, z), q φ (z|v)), is given by the Kullback-Leibler (KL) divergence between posteriors, i.e., KL(q φ (z|v)||p θ (z|v)) = E Z∼q φ (z|v) [log q φ (Z|v)] − E Z∼q φ (z|v) [log p θ (Z|v)], which implies that the ELBO tightens as q φ (z|v) approaching the true posterior p θ (z|v).
For estimation, we seek parameters θ and φ that maximize the ELBO in (1).Given a set of observations {v i } N i=1 sampled from data distribution v ∼ p d (v), maximizing the expected ELBO is also equivalent to minimizing the KL divergence KL(p d (v) p θ (v)) between the empirical and model distributions.When p θ (v|z) and q φ (z|v) are specified as neural networks, the resulting architecture is commonly known as the variational auto-encoder (VAE) (Kingma and Welling 2014), where q φ (z|v) and p θ (v|z) and are known as encoder and decoder, respectively.Note that q φ (z|v) is often used for subsequent inference tasks on new data.
Extreme Value Theory
Extreme Value Theory (EVT) provides a principled probabilistic framework for describing events with extremely low probabilities, which we seek to exploit for better rare event modeling.In particular, we focus on the exceedance models, where we aim to capture the asymptotic statistical behavior of values surpassing an extreme threshold (Davison and Smith 1990;Tao et al. 2017), which we briefly review below following the notation of Coles et al. (2001).Without loss of generality, we consider exceedance to the right, i.e., values greater than a threshold u.For a random variable X, the conditional cumulative distribution of exceedance level x beyond u is given by F , where x > 0 and F (x) denotes the cumulative density function for X.
A major result from EVT is that under some mild regularity conditions, e.g., continuity at the right end of F (x) and others, F u (x) will converge to the family of Generalized Pareto Distributions (GPD) regardless of F (x), as u approaches the right support boundary of F (x) (Balkema and De Haan 1974;Pickands III et al. 1975) (Falk, Hüsler, and Reiss 2010), where GPD ξ,σ,u (x) is of the form where σ is a positive scale parameter.When ξ < 0 the exceedance x has bounded support 0 ≤ x ≤ u − σ/ξ, otherwise when ξ ≥ 0, x is unbounded.A major implication of this asymptotic behavior is that, for modeling extreme values, one only needs to fit extreme samples to the loglikelihood function of the GPD.
Additive Isotonic Regression
Also known as monotonic regression, isotonic regression is a non-parametric regression model that constrains the relation between predictor and outcome to be monotonic, (e.g., non-decreasing f (a) ≤ f (b) for a ≤ b) (Barlow et al. 1972).Such monotonic constraint is a natural and flexible extension to the standard linear relation assumed by many statistical models.To accommodate multi-covariate predictors, additive isotonic regression combines isotonic models for each individual one-dimensional predictor (Bacchetti 1989).Standard implementations often involve specialized algorithms, such as local scoring algorithms (Hastie 2017) and the alternating conditional expectation (ACE) method of Breiman and Friedman (1985).All these approaches typically require costly iterative computations and are not scalable to large datasets.Here we consider recent advances in unconstrained monotonic neural networks, which allow for efficient and flexible end-to-end learning of monotonic relations with robust neural nets based on standard training schemes such as stochastic gradient descent (Sill 1998;Wehenkel and Louppe 2019).
VARIATIONAL INFERENCE OF EXTREMALS
The proposed model is based on the hypothesis that extreme events are driven by the extreme values of some latent factors.Specifically, we propose to recast the learning of low-prevalence events into the learning of extreme latent representations, thus amortizing the difficulties associated with directly modeling rare events as outcomes.To allow for more efficient learning from the rare events, we make some further assumptions to regularize the latent representation: (i) effect disentanglement: the contribution from each dimension of the latent representation to the event occurrence is additive; (ii) effect monotonicity: there is a monotonic relation between the outcome likelihood and the values of each dimension of the latent representation.The key to the proposed approach is using an additive isotonic neural network to model the one-dimensional disentangled monotonic relations from a latent representation, which is obtained via variational inference.Specifically, we impose an EVT prior to explicitly capture the information from the few minority group samples into the tail behavior of the extreme representation.Below we provide the rationale for our choices followed by a description of all model components.
Disentanglement & additive isotonic regression.Consistent with assumptions (i) and (ii), we posit a scenario in which the underlying representation of extreme events is more frequent at the far end of the representation spectrum, for which additive isotonic regression is ideal.The disentanglement consists of modeling each latent dimension individually, thus avoiding the curse of dimensionality when modeling combinatorial effects with few examples.Further, the monotonicity constraint imposed by the isotonic regression model restricts possible effect relations, thereby improving generalization error by learning with a smaller, yet still sufficiently expressive, class of models (Bacchetti 1989).
EVT & VI.Note that the spread of representation of extreme events is expected to be more uncertain relative to those of the normal, more abundant events, due to a few plausible causes: (i) extreme events represent the breakdown of system normality and are expected to behave in uncertain ways; (ii) there is only a small number of examples available for the extreme events, so the learned feature encoder will tend to be unreliable.As a result, it is safely expected that the encoded features associated with the extremes events will lie outside the effective support of the Gaussian distribution assumed by the standard VI model.In other words, the representation of the events can manifest as a heavy-tailed distribution.This will compromise the validity and generalizability of a prediction model if not dealt with appropriately.So motivated, we explicitly model the distribution of the extreme underlying representations via EVT.Using EVT, we decouple the learning of the tail end of the representation distribution.Since EVT-based estimation only requires very few parameters, it allows for accurate modeling with a small set of tail-end samples.Further, in combination with the variational inference framework, it accounts for representation uncertainty via the use of a stochastic encoder, which further strengthens model robustness.
Benefits of heavy-tailed modeling.A few other considerations further justify modeling with a heavy-tailed distribution for the extreme event representation.One obvious benefit is that it allows better model resolution along the representation axis, i.e., better risk stratification.For light-tail representations, extreme examples are clustered in a narrow region where the tail vanishes, thus a standard (lighttailed) learning model will report the average risk in that region.However, if the representations are more spread out, then there is a more gradual change in risk, which can be better captured, as shown in Figure 1.Another argument for favoring heavy-tailed representations is that heavy-tailed phenomena are very common in nature (Bryson 1974), and these tail samples are often encoded less robustly due to the lack of training examples.Allowing long-tail representations relieves the burden of an encoder.
Model structure.We consider latent variable model p θ (y, x, z) = p θ (y|z)p θ (x|z)p(z), where v = {x, y} are the observed variables.Under the VI framework, similar to (1) we write the ELBO(v; p θ (v, z), q φ (z|v)) as where p θ (y|z) is specified as an additive isotonic regression model, p(z) is modeled with EVT, and the approximate posterior, q φ (z|v), is specified as an inverse auto-regressive flow.Note that unlike in the standard ELBO in (3), we have dropped the term E Z∼q φ (z|v) [log p θ (x|z)] because we are not interested in modeling the covariates.Note this coincides with the variational information bottleneck (VIB) formulation (Alemi et al. 2016).Additionally, the posterior q φ (z|v) will not be conditioned on y, but only on x, because in practice, the labels y are not available at inference time.Specifically, we rewrite the objective in (3) as , where β is a hyperparameter controlling the relative contribution of the KL term to the objective.Below we provide details for each component of the proposed approach.
Decoder: Additive Monotonic Neural Network
First, let us consider the following monotone mapping z l h(s; θ)ds + γ, consisting on integrating a non-negative function h(s; θ) specified as a neural network with onedimensional input, s, and parameterized by θ.The choice of the lower end l is arbitrary, and γ is a bias term.For multi-dimensional latent representation z ∈ R p , we write the additive monotonic neural network (AMNN) as where α j serves as a weight which controlling the effect directions.In other words, when α j > 0, it can be interpreted as an event stimulator; otherwise it is an event blocker.To ensure h(s; θ) is non-negative, we apply exponential activation function to the network's output.The integration of z is conducted with numerical integration by the Riemann-Stieltjes method (Davis and Rabinowitz 2007).
Latent Prior: Gaussian GPD Mixture
To better capture the tail behavior of the latent representation, we assume random variable Z ∼ p(z) is a mixture of a standard Gaussian distribution truncated at u and a GPD for modeling the tail end thresholded at u, i.e., F (z) = Φ(z) when z ≤ u and when z > u, where Φ(z) denotes the CDF of a standard Gaussian distribution.Note that for z > u, F (z) can be expressed as a GPD with parameters ( ξ, σ, ũ) (McNeil 1997), where ξ = ξ and if ξ = 0, σ = σ(1 − Φ(u)) ξ and ũ = u − σ((1 − Φ(u)) −ξ − 1)/ξ.Otherwise, when ξ = 0, σ = σ and ũ = u+ σ log(1−Φ(u)).Consequently, the CDF for the mixed GPD is given by For simplicity, we denote the set of parameters in GPD as ψ={ξ GPD ,σ GPD } and the threshold u is a user-defined parameter.In the experiments we set u to Φ −1 (0.99).
Latent Posterior: Inverse Autoregressive Flow
Considering we have adopted a long-tailed GPD prior, we seek a posterior approximation q φ (z|x) that is: (i) a flexible parameterization to approximate arbitrary distributions; and (ii) with a tractable likelihood to be able to evaluate the KL(q φ (z|x) p(z)) exactly.We need (i) because the true posterior is likely to exhibit heavy-tailed behavior due to the extended coverage of the GPD prior, and (ii) is to ensure accurate and low-variance Monte Carlo estimation of the KL-divergence at the tail end of the prior.These requirements invalidate some popular choices, e.g., a standard Gaussian posterior is light-tailed, and the implicit neuralsampler-based posterior typical in the work of adversarial variational Bayes (Mescheder, Nowozin, and Geiger 2017), does not have a tractable likelihood.
One model family satisfying the above two requirements is known as the generative flows (Rezende and Mohamed 2015), where simple invertible transformations with tractable log Jacobian determinants are stacked together, transforming a simple base distribution into a complex one, while still having closed-form expressions for the likelihood.In this work, we consider the inverse autoregressive flow (IAF) model (Kingma et al. 2016).The flow chain is built as: where µ t ∈ R p and σ t ∈ R p are learnable parameters, denotes the element-wise product, z 0 is typically drawn from a p-dimensional Gaussian distribution, z 0 ∼ N (µ 0 , Diag(σ2 0 )) where µ 0 and σ 0 are obtained from an initial encoder defined by a neural network given input x with parameter φ.A sample from the posterior q φ (z|x) is given by z T , obtained by "flowing" z 0 through (7).Provided the Jacobians dµt dzt−1 and dσt dzt−1 are strictly upper triangular (Papamakarios, Pavlakou, and Murray 2017), we obtain the following closed-form expression for the log posterior where e j = (x j − µ 0,j )/σ 0,j for the jth dimension.
Posterior Match with Fenchel Mini-Max Learning
We consider an additional modification that explicitly encourages the match of the aggregated posterior q φ (z) = q φ (z|x)p d (x)dx to the prior p(z), which has been reported to be vastly successful at improving VAE learning (Mescheder, Nowozin, and Geiger 2017).In our case, q φ (z) does not have a closed-form expression for the likelihood ratio of the KL formulation, which motivates us to use a sample-based estimator.We consider the mini-max KL estimator based on the Fenchel duality (Tao et al. 2019;Dai et al. 2018).Concretely, recall the KL can be expressed in its Fenchel dual form 2 where ν(z) is commonly known as the critic function in the adversarial learning literature, and we maximize wrt ν(z) in the space of all functions F, modeled with a deep neural network.We use ( 4) and ( 9) to derive an augmented ELBO that further penalizes the discrepancy between the aggregated posterior and the prior, i.e., Ψ β (x, y; p θ (y|z), q φ (z|x)) − λKL(q φ (z) p(z)), where λ is a regularization hyperparameter (Chen, Feng, and Lu 2018).Solving for this objective results in the following mini-max game where β and λ are regularization hyperparameters.In a similar vein to β-VAE and adversarial variational Bayes (AVB), our objective leverages β, λ > 0 to balance the prediction accuracy and the complexity of the latent representation via KL regularization.Further, from Ψ β (x, y; p θ (y|z), q φ (z, x)) in (4), note that the decoder p θ (y|z) is obtained from the additive neural network in (5), p ψ (z) is the Gaussian GPD mixture with CDF in (6), q φ (z|x) is the autoregressive flow implied by ( 7) and ν(z; ω) is the critic function specified as a neural network and parameterized by ω.
where KL = l post − log p ψ (z T ) end To avoid collapsing to suboptimal local minima, we train the encoder arm more frequently to compensate for the detrimental posterior lagging phenomenon (He et al. 2019).The pseudo-code for the proposed VIE is summarized in Algorithm 1 and detailed architecture can be found in the SM.
RELATED WORK
Rare-event modeling with regression.Initiated by King and Zeng (2001), the discussion on how to handle the unique challenges presented by rare-event data for regression models has attracted extensive research attention.The statistical literature has mainly focused on bias correction for sampling (Fithian and Hastie 2014) and estimation (Firth 1993), driven by theoretical considerations in maximum likelihood estimation.However, their assumptions are often violated in the face of modern datasets (Sur and Candès 2019), characterized by high-dimensionality and complex interactions.Our proposal approaches a solution from a representation learning perspective (Bengio, Courville, and Vincent 2013), by explicitly exploiting the statistical regularities of extreme values to better capture extreme representations associated with rare events.
Re-sampling and loss correction.Applying statistical adjustments during model training is a straightforward solution to re-establish balance, but often associated with obvious caveats.For example, the popular down-sampling and up-sampling (He and Garcia 2009) discard useful information or introduce artificial bias, exacerbating the chances of capturing spurious features that may harm generalization (Drummond, Holte et al. 2003;Cao et al. 2019), and their performance gains may be limited (Byrd and Lipton Our contribution is orthogonal to these developments and promises additional gains when used in synergy. Transferring knowledge from the majority classes.Adapting the knowledge learned from data-rich classes to their under-represented counterparts has shown success in few-shot learning, especially in the visual recognition field (Wang, Ramanan, and Hebert 2017;Chen et al. 2020), and also in the clinical setting (Böhning, Mylona, and Kimber 2015).However, their success often critically depends on strong assumptions, the violation of which typically severely undermines performance (Wang et al. 2020).Related are the one-class classification (OCC) models (Tax 2002), assuming stable patterns for the majority over the minority classes.Our assumptions are weaker than those made in these model categories, and empirical results also suggest the proposed VIE works more favorably in practice (see experiments).
EXPERIMENTS
We carefully evaluate the proposed VIE on a diverse set of realistic synthetic data and real-world datasets with different degrees of imbalance.Our implementation is based on PyTorch, and code to replicate our experiments are available from https://github.com/ZidiXiu/VIE/.We provide additional experiments and analyses in the SM.
Baseline Models We consider the following set of competing baselines to compare the proposed solution: LASSO regression (Tibshirani 1996), MLP with re-sampling and re-weighting (MLP), Importance-Weighting model (IW) (Byrd and Lipton 2019), FOCAL loss (Lin et al. 2017), Label-Distribution-Aware Margin loss (LDAM) (Cao et al. 2019), and SVD based one-class classification model (Deep-SVDD) (Ruff et al. 2018).We tune the hyper-parameters of baseline models on the validation dataset, and pick best performing hyper-parameters to evaluate test set performance.For detailed settings please refer to the SM.
Evaluation Metrics To quantify model performance, we consider AUC and AUPRC.AUC is the area under the Receiver Operating Characteristic (ROC) curve, which provides a threshold-free evaluation metric for classification model performance.AUC summarizes the trade-off between True Positive Rate (TPR) and False Positive Rate (FPR).AUPRC summarizes the trade-off between TPR and True Predictive Rate.Specifically, it evaluates the area under Precision-Recall (PR) curve.We discuss other metrics in the SM.In simulation studies, we repeat simulation ten times to obtain empirical AUC and AUPRC confidence intervals.For real world datasets, we applied bootstrapping to estimate the confidence intervals.
Ablation study for VIE
VIE applies a few state-of-art techniques in variational inference in order to achieve optimal performance.In this section, we decouple their contributions via an ablation study, to justify the necessity of including those techniques in our final model.To this end, we synthesize a semi-synthetic dataset based on the Framingham study (Mitchell et al. 2010), a long-term cardiovascular survival cohort study.We use a realistic model to synthesize data from the realworld covariates under varying conditions, i.e., different event rates, sample size, non-linearity, etc.More specifically, we use the CoxPH-Weibull model (Bender, Augustin, and Blettner 2005) to simulate the survival times of patients T = { − log U λ exp(g(x)) } 1/ν , where g(x) is either a linear function or a randomly initialized neural net.Our goal is to predict whether the subject will decease within a pre-specified time frame, i.e., T < t 0 .Via adjusting the cut-off threshold t 0 , we can simulate different event rates.A detailed description of the simulation strategy is in the SM.
We experiment with different combinations of advanced VI techniques, as summarized in Table 1.Limited by space, we report results at 1% event rate with g(•) set to a randomly initialized neural network under various sample sizes.Additional results on linear models and other synthetic datasets are consistent and can be found in the SM.IAF and GPD only variants perform poorly, even compared to the vanilla VAE solution.This is possibly due to the fact that priors are mismatched.Explicitly matching to the prior via Fenchel mini-max learning technique improves performance.However, without using an encoder with a tractable likelihood, the model cannot directly leverage knowledge from the GPD prior likelihood.Stacked together (mixed GPD+IAF+Fenchel), our full proposal of VIE consistently outperforms its variants, approaching oracle performance in the large sample regime.
Real-World Datasets
To extensively evaluate real-world performance, we consider a wide range of real-world datasets, briefly summa- 3. Note that InP, SEER and SLEEP are all survival datasets, among which SEER and SLEEP include censored subjects.We follow the data pre-processing steps in (Xiu et al. 2020).To create outcome labels, we set a cut-off time to define an event of interest the same as in the ablation study, and exclude subjects censored before the cut-off time.The excluded samples only account for less than 0.2% of the whole population, and therefore it is expected to have a very limited impact on our results.Datasets have been randomly split into training, validation, and testing datasets with ratio 6:2:2.See the SM for details on data pre-processing.Table 2 compares VIE to its counterparts, where the numbers are averaged over the bootstrap samples.We see the proposed VIE yields the best performance in almost all cases, and the lead is more significant with low event rates.Note that the one-class classification based DeepSVDD performs poorly, which implies treating rare events as outliers are inappropriate in the scenarios considered here.Reweighting and resampling based methods (IW, Focal) are less stable compared to those simple baselines (LASSO, MLP).The theoretically optimal LDAM works well in general, second only to VIE in most settings.To further demonstrate the stability of our method, we visualize the bootstrapped evaluation scores for the COVID dataset in Figure 3, and defer the additional cross-validation results to the SM.We see that VIE leads consistently.
We also verify empirically that the estimated GPD shape
CONCLUSIONS
Motivated by the challenges of rare-event prediction in clinical settings, we presented Variational Inference with Extremals (VIE), a novel extreme representation learningbased variational solution to the problem.In this model we leveraged GPD to learn the extreme distributions with few samples and applied additive monotonic neural networks to disentangle the latent dimensions' effects on the outcome.VIE featured better generalization and interpretability, as evidenced by a strong performance on real and synthetic datasets.In future work, we will extend this framework to the context of causal inference to quantify treatment effects in the label imbalanced setting (Lu et al. 2020).
Supplementary Material to "Variational Disentanglement for Rare Event Modeling" Contents When the prevalence of an event is extremely low, but the event itself has substantial importance, the methods to identify such targets are called rare event modeling.Accurate and robust modeling of rare events is significant in many fields, such as identifying patients in high-risk and hopefully to prevent adverse outcomes from happening based on early intervention.
The scarcity of rare cases can cause extreme imbalanced among the dataset.Therefore, rare event modeling is challenging for most standard statistical approaches.As we discussed in the main text, careful statistical adjustments and new methodologies are required to approach such imbalance.Otherwise, the classifiers would be driven to the majority side and give misleading results.Also, the lack of representation in the minority class may cause unadjusted models to wrongly capturing spurious features that cannot generalize well to other observations.The apex of the risk curve or the mass of risk density usually overlays with the tail of the feature representation distribution, as illustrated in Figure S1, traditional statistical methods (such as Gaussian based approaches) often ill perform at the tail end, which can lead to lack-of-fit and poor generalization ability.We approach a solution to such challenges with a variational representation learning scheme that models disentangled extreme representations.Further, we design a robust, powerful prediction arm that combines the merits of a generalized additive model and isotonic neural net.
A. Derivation of Mixed GPD tail distribution
An important theory in Extreme value theory (EVT) shows that under some mild conditions, the conditional cumulative distribution of exceedance over a threshold u follows Generalized Pareto Distribution, GPD(u, ξ, σ) (McFadden 1978), which has the cumulative distribution function (CDF) as: where σ is a positive scale parameter.According the shape parameter ξ, x could have different support.When ξ < 0, the exceedance x has bounded support 0 ≤ x ≤ u − σ/ξ, otherwise x is bounded by 0 on the left.u is the location parameter.The corresponding PDF is: if ξ = 0 Thus the log-likelihood function is:
B. Implementation Details
Our main algorithm was written in PyTorch (version 1.3.1)(Paszke et al. 2017).The experiments were conducted on an Intel(R) Xeon(R) and Tesla P100-PCIE-16GB GPU (except for the COVID dataset).The COVID dataset were stored and analyzed on a protected virtual network space with Inter(R) Xeon(R) Gold 6152 CPU 2.10GHz 2 Core(s).
Figure 1 :
Figure 1: Left: Distribution of a two-dimensional latent space z where the long tail associates with higher risk.Right: Tail estimations with different schemes for the long-tailed data in one-dimensional space.EVT provides more accurate characterization comparing to other mechanisms.
Figure 2 :
Figure 2: First latent dimension from the InP dataset (1% event rate).Left: Learned prior and posterior distribution, and monotonic predicted risks (right axis).Right: The latent representation values distribution grouped by event type.
Figure S1 :
Figure S1: Feature representation mismatch at the tail parts.The heavy-tailed distribution can exploit extreme behavior in the latent space.
Table 1 :
Ablation study of VIE with different combinations of architectures on realistic synthetic datasets with 1% event rate.The oracle model has used the ground-truth model parameters to predict.
Table 2 :
Average AUC and AUPRC from real-world datasets.
Table 3 :
Summary statistics for real-world datasets. | 7,567 | 2020-09-17T00:00:00.000 | [
"Computer Science"
] |
Insights into antimicrobial resistance among long distance migratory East Canadian High Arctic light-bellied Brent geese (Branta bernicla hrota)
Background Antimicrobial resistance (AMR) is the most significant threat to global public health and ascertaining the role wild birds play in the epidemiology of resistance is critically important. This study investigated the prevalence of AMR Gram-negative bacteria among long-distance migratory East Canadian High Arctic (ECHA) light-bellied Brent geese found wintering on the east coast of Ireland. Findings In this study a number of bacterial species were isolated from cloacal swabs taken from ECHA light-bellied Brent geese. Nucleotide sequence analysis identified five species of Gram-negative bacteria; the dominant isolated species were Pantoea spp. (n = 5) followed by Buttiauxella agrestis (n = 2). Antimicrobial susceptibility disk diffusion results identified four of the Pantoea spp. strains, and one of the Buttiauxella agrestis strains resistant to amoxicillin-clavulanic acid. Conclusion To our knowledge this is the first record of AMR bacteria isolated from long distance migratory ECHA light-bellied Brent geese. This indicates that this species may act as reservoirs and potential disseminators of resistance genes into remote natural ecosystems across their migratory range. This population of geese frequently forage (and defecate) on public amenity areas during the winter months presenting a potential human health risk.
Findings
Antimicrobial resistance (AMR) is the greatest challenge facing global public health [1]. The current proliferation of multidrug-resistant pathogens, prevalence of resistant bacteria in the environment and dissemination of resistance genes into novel biogeographic regions is unprecedented [2].
Accurate and meaningful information relating to the dissemination of resistance genes in bacteria among wildlife is of importance in assessing the potential human health risks, and ecological impacts the ingress of these elements have on natural environments [3]. Wild birds are increasingly being studied as vectors for the transmission of resistant bacteria and the resistance genes they harbour [4,5]. The East Canadian High Arctic (ECHA) light-bellied Brent goose (Branta bernicla hrota) undertakes one of the longest migrations of any Palaearctic goose species, migrating annually from their breeding grounds in the high Canadian Arctic to Ireland in winter [6]. Their preferred food is the intertidal marine grass (Zostera spp), but this resource becomes exhausted in mid to late winter and the birds switch to feeding on terrestrial grasses [7]. In many parts of their range this brings them into close contact with humans as in urban areas these terrestrial grasses tend to be found in public parks and sports grounds. In this study we aimed to investigate the prevalence of clinically relevant antimicrobial resistant Gram-negative enteric bacteria carried among this population of Brent geese during their winter staging on the east coast of Ireland, and inform if these migratory wild birds are potential disseminators of resistance genes into remote natural ecosystems.
The animal trapping and handling undertaken as part of this research was carried out under permit from the British Trust for Ornithology (Permit A4656) and the Irish Medicines Board (Authorisation number AE19130/ 141). The project was reviewed and deemed fit by the University of Exeter Ethics Committee. A total of 66 Brent geese were caught using explosive cannon nets on North Bull Island, Dublin on the east coast of Ireland (53°22′14.92 N, 6°9′9.98 W) 31st March 2015. All birds had faecal cloacal swabs taken. Samples were placed in a cooler box and transported to the laboratory where they were stored at 4°C prior to analyses.
Biochemical testing was conducted on all cultured isolates to test for the presence of Escherichia coli using indole and citrate utilisation tests. A sub-set of 16 plates shown to support good colony growth, and screened for E. coli, was selected for further bacterial identification using the Gram stain method. Ten isolates positively identified as Gram-negative bacterial rods following microscopic examination under an oil immersion lens were selected for DNA isolation and PCR. Following user protocol, DNA was isolated from the selected bacterial isolates using the UltraClean® Microbial DNA Isolation Kit and subsequently stored at 4°C for downstream PCR analysis. The PCR reaction mixture was prepared for 12 samples, including one positive and one negative control, 22 μl of PCR reaction mixture was combined with 3 μl of extracted template DNA [equivalent to approx., 100 ng] for each sample. The forward and reverse bacterial primer pair sequences used in the PCR amplification process were Bakt_341F (5′-CCT ACG GGN GGC WGC AG-3′) and Bakt_805R (5′-GAC TAC HVG GGT ATC TAA TCC-3′) with an amplicon size of 464 base pair(s) (bp) [8]. The thermal cycle for PCR amplification consisted of an initial denaturation step at 95°C for 30 s, 35 cycles of denaturation at 95°C for 30 s, annealing at 54°C for 1 min, extension at 68°C for 2 min and a final elongation step at 68°C for 2 min. A preliminary characterisation of the quality of the amplified DNA was conducted, using agarose gel electrophoresis, to establish that the required sequence had been successfully amplified for all ten PCR products along with two controls. Following electrophoresis the DNA bands were excised from the gel under UV light and prepared for sequencing using the Wizard® SV Gel and PCR Clean-Up System. The purified DNA was stored at 4°C before same day collection and shipment for Sanger sequencing by Source BioScience Sequencing. Five microliter of primers and DNA were sent per reaction at concentrations of 3.2 pmol/μl and 1 ng/μl per 100 bp respectively. The nucleotide sequence queries for the ten bacterial samples were loaded into the National Centre for Biotechnology Information (NCBI) Basic Local Alignment Search Tool (BLAST®) [9]. A standard nucleotide BLAST was conducted using the Megablast programme BLASTN 2.2.31 [10], optimised to identify highly similar sequences. The highest Query cover and Ident percentage scores were used to determine best fit for sequence alignment. Disk diffusion antimicrobial susceptibility testing was conducted on all sequenced and identified isolates in accordance with the standardised methodology developed by the European Committee on Antimicrobial Susceptibility Testing (EUCAST), Version 5.0 [11,12]. Isolates were tested for susceptibility to Amoxicillin-clavulanic acid (20/10 μg), Cefepime (30 μg), Imipenem (10 μg), Ciprofloxacin (5 μg) and Trimethoprim-sulfamethoxazole (1.25/ 23.75 μg). Inhibition zone diameters were recorded and categorised according to the EUCAST Clinical Breakpoint table v 5.0 [12].
Gram-stain tests identified a selection of Gram-negative rods (n = 11), Gram-negative cocci (n = 1), Gram-positive cocci (n = 1) and samples containing a mix of both Gramnegative rods and cocci (n = 2). Sanger sequencing successfully yielded nucleotide sequences for all ten isolated DNA samples and BLAST analysis identified five species of Gram-negative bacteria. Antimicrobial susceptibility disk diffusion results identified four of the Pantoea spp. strains, and one of the Buttiauxella agrestis strains resistant to amoxicillin-clavulanic acid. All strains were susceptible to the remaining antimicrobial compounds (Table 1).
Wild birds, particularly migratory waterfowl, can travel immense distances, inhabit a wide variety of environments and may consequently have a significant epidemiological role in the dissemination of resistant bacteria and genes [13]. Non-migratory Canada Geese (Branta canadensis) have previously been identified as reservoirs of multi-resistant strains of E. coli and implicated in microbial water contamination; although non-migratory, these birds could serve to disperse bacteria between widely separated locations [14]. The finding that ECHA light-bellied Brent geese are reservoirs of resistant bacteria has direct implications related to the potential of this species to act as disseminators of resistance into remote habitats throughout their migratory range. The potential for migratory birds to carry AMR bacteria over remarkable distances to remote locations in the Arctic, a region formerly considered one of the last outposts of wilderness, has been demonstrated [15]. Sjölund et al. [15] have shown E. coli isolates expressing multi drug resistance to as many as eight antibiotics among Glaucous gulls in the Arctic. This study also found a juvenile Western sandpiper sampled on the Siberian tundra had resistance to cefadroxil, cefuroxime, and cefpodoxime, a resistance pattern commonly observed in clinical isolates, supporting the theory of introduction by migration and subsequent bacterial transfer between birds. The potential for wild birds to act as vectors for the transmission of clinically relevant resistance genes is substantiated by the discovery of gulls harbouring the same CTX-M types dominant among human isolates in the same area [4,16]. Hernandez et al. [17] found the carriage rate of ESBL-producing bacteria among Franklin's gulls in central parts of Chile to be twice as high as those found among the human population in the same area, but the gulls were also found to share sequence types from clinical samples in central Canada, a known nesting place for the birds. These findings contribute to the accumulating evidence supporting the dissemination of resistance by migratory birds, and the reciprocal transmission of resistance determinants between humans and wild birds [18].
It seems likely that the greatest exposure to environmental sources of antibiotics and resistant bacteria the light-bellied Brent geese encounter throughout their migration is during the winter in Ireland, where they experience the most intense contact with anthropogenicinfluenced habitats. Previous studies have isolated AMR E. coli from herring gulls (Larus argentatus) sampled in Howth harbour located within 10 km of North Bull Island [19][20][21]. ECHA light-bellied Brent geese migrate north from Ireland during the spring and stage in western Iceland [7], before continuing their journey to breed in the Canadian Arctic [22], satellite tracking has also identified a number of staging grounds in east Greenland [23]. AMR found among this population of Brent geese identifies them as potential disseminators of resistant bacteria, and the genetic resistance determinants they possess, into various ecosystems throughout their range.
Resistance found among the Pantoea species isolates in this research is notable as multiple species groups within Pantoea are viewed as opportunistic pathogens [24]. Perhaps, the greatest zoonotic potential the birds sampled in this research present may be through their use of amenity grasslands. Faecal shedding of resistant bacteria and the persistence of such organisms in the environment may pose a health threat to humans [25]. A study by Benson [26] identified 60 inland sites used by light-bellied Brent geese as winter feeding grounds in Dublin, these include playing pitches, public parks, golf clubs and municipal green spaces in densely populated areas. The large amount of faeces resulting from congregating flocks on amenity grassland could present a possible health risk. Future research, in conjunction with the findings here, could help elucidate the persistence of resistant bacteria carried by the birds throughout their range at different times of the year, accurately appraise their ability to act as vectors for the dissemination of resistance and define where resistant bacteria is acquired. The enrichment of environmental bacteria with genetic elements containing resistant genes, the dynamic nature of prokaryotic genomes, and ease with which resistance determinants can be shared among commensal and pathogenic bacteria, conspire to present a threat to both human and animal health. This research has identified ECHA light-bellied Brent geese as reservoirs of resistant bacteria and potential disseminators of resistance genes into remote natural habitats in Iceland, Greenland and the Canadian Arctic. The findings in this study add to the accumulating evidence that wild migratory birds are disseminators of resistant bacteria and can play an important role in the epidemiology of resistance. AMR is a global health concern and reaches far beyond clinical settings, understanding the role wildlife plays, particularly migratory birds, is critically important in designing practicable and effectual mitigation measures to address this problem for the future. | 2,641.8 | 2016-09-15T00:00:00.000 | [
"Biology"
] |
In vitro and in vivo activity of meglumine antimoniate produced at Farmanguinhos-Fiocruz , Brazil , against Leishmania ( Leishmania ) amazonensis , L ( L . ) chagasi and L ( Viannia ) braziliensis
The leishmanicidal activity of four batches of meglumine antimoniate, produced in Farmanguinhos-Fiocruz, Brazil (TAMs), was assessed and compared to Glucantime®-Aventis Pharma Ltda. Using the amastigote-like in vitro model, the active concentrations of Sbv varied from 10μg/ml to 300 μg/ml for L. (L.) chagasi and from 50μg/ml to 300μg/ml for L. (L.) amazonensis, with no statistically significant differences among the four batches of TAMs and Glucantime®. The inhibitory concentrations (IC50) determined by the amastigote-infected macrophage model for TAM01/03 and Glucantime® were, respectively: 26.3μg/ml and 127.6μg/ml for L. chagasi, 15.4μg /ml and 22.9μg/ml for L. amazonensis, and 12.1μg/ml and 24.2μg/ml for L. (V.) braziliensis. The activities of the four batches of TAMs were confirmed in an in vivo model by assessing, during eight weeks skin lesions caused by L. braziliensis in hamster that were treated with 20mg Sbv/Kg/day for 30 consecutive days. The meglumine antimoniate produced by Farmanguinhos was as effective as the reference drug, Glucantime®-Aventis, against three species of Leishmania that are of medical importance in Brazil.
The protozoan parasite Leishmania causes a variety of clinical diseases that afflict 12 million people worldwide. Organic salts of pentavalent antimony have been used for the treatment of all clinical forms of leishmaniasis for more than 60 years. Antimonials are thought to act by inhibiting the enzymes of glycolysis and other metabolic pathways (Berman 1988). Two formulations of pentavalent antimonials are currently used: sodium stibogluconate (Pentostam®) and meglumine antimoniate (Glucantime®). Both treatments are given intravenously or intramuscularly and demonstrate similar efficacy when used in equivalent doses.
The Brazilian Ministry of Health recommends a oncedaily injection of 20mg Sb v /Kg for 20 to 40 days for the treatment of visceral leishmaniasis (VL) and 15mg/Kg/ day for 20 days for the treatment of cutaneous leishmaniasis (MS 2006a, b). The cost per treatment, including delivery and clinical monitoring, reaches approximately US$ 200. This is a relatively high expenditure for public health as about 38,000 patients are treated yearly. The cost of Glucantime® is about US$80, depending on the clinical form and schedule. Cheaper generic formulations of sodium stibogluconate have recently become available. Studies conducted in the Sudan, Kenya, Ethiopia, Bolivia, and Colombia showed that generic sodium stibogluconate (Albert David Ltd, Calcutta, India) was Financial support: CNPq, Fapemig + Corresponding author<EMAIL_ADDRESS>Received: 18 February 2008 Accepted: 10 June 2008 equivalent to Pentostam® and Glucantime® for the treatment of leishmaniasis in terms of both safety and efficacy (Veeken et al. 2000, Moore et al. 2001, Ritmeijer et al. 2001, Soto et al. 2004, Bermúdez et al. 2006. The reported cost of the generic pentavalent antimony stibogluconate is approximately 7% that of Pentostam® (~ US$ 13/treatment; Ritmeijer et al. 2001), and 20% that of meglumine antimoniate in the form of Glucantime® (Soto et al. 2004).
In 2002, due to the menace of an unsupplied market of meglumine antimoniate, Farmanguinhos-Fiocruz (TAMs), a pharmaceutical company that belongs to the Brazilian Ministry of Health, developed an alternative process to produce this API (Active Pharmaceutical Ingredient) in its R&D laboratory. This study presents the evaluation of the activity of the TAMs against the Leishmania species of medical importance in Brazil, compared to the reference drug, Glucantime®-Aventis Pharma.
MATERIALS AND METHODS
The leishmanicidal activities of four batches of TAMs were assessed and compared to Glucantime®-Aventis using the amastigote-like in vitro model. Next, the inhibitory concentration (IC 50 ) of one of the batches of meglumine antimoniate (TAM 01/03-Farmanguinhos) was determined using the amastigote-infected macrophage model. Additionally, an in vivo study was conducted to evaluate the efficacy of the four batches of TAMs against L. (Viannia) braziliensis.
For the in vitro assays, the four batches of TAMs were suspended in deionized water at 80°C and sterilized by passage through 0.2 µm membrane filters. AmB was reconstituted by rapidly adding 10 ml of water into the lyophilized cake and shaking the vial until the colloidal suspension became clear, as per the manufacturer's instructions. All subsequent dilutions were prepared in fresh RPMI 1640 (Sigma-Aldrich, St. Louis, USA) culture medium on the day of the assay. Glucantime® and TAMs were stored at room temperature in the dark until use. AmB was maintained at 2-8°C and was used for a maximum of 30 days (manufacturer's instructions).
Amastigote-like model for L. chagasi and L. amazonensis -The leishmanicidal activities of the four batches of TAMs were tested by using the amastigote-like model. The amastigote-like forms were added to the Schneider's medium in concentrations of 5x10 7 /ml for L. chagasi, and 2x10 7 /ml for L. amazonensis and incubated with TAM 01/03, TAM 02/03, TAM 03/03, TAM 021/02 and Glucan-time® (concentrations of Sb v : 5, 10, 50, 150 and 300 µg/ml) and AmB (0.2 µg/ml). Samples were seeded into 96-well flat-bottom microtrays, in triplicate, and were incubated at 35°C or 32°C. After 72h, parasites were counted using a Neubauer™ chamber. Three independent experiments were performed in triplicate on different days. The number of parasites counted in wells without drug was set as 100% parasite survival (parasite controls).
Amastigote-macrophage assay for L. chagasi, L. amazonensis, and L. braziliensis -Balb/c mice were injected intraperitoneally with 1.5 ml of 3% tioglicolate medium (Biobrás, Brazil). After 96 h, the peritoneal macrophages were harvested by peritoneal lavage using cold RPMI-1640 medium. Cells were counted, centrifuged, and resuspended at a concentration of 4x10 5 /ml in RPMI-1640 medium without supplements. Sterile round glass coverslips (13 mm) were placed in each well of 24-well culture plates. Macrophages were pipetted in a volume of 500 µl/well and allowed to attach to the coverslips for 2 h at 37ºC in 5% CO 2 . After 2 h, the medium was removed from the wells and replaced with 500 µl of warm (37ºC) RPMI containing 10% FCS and penicillin (50 U/ml) and streptomycin (50 µg/ml). The following day, a suspension of 4x10 6 amastigote-like L. chagasi, L. amazonensis, or L. braziliensis was added to each well in 500 µl of RPMI (macrophage:parasite ratio of 1:10). The plates were incubated for 4 h at 37ºC in 5% CO 2 , and the medium was aspirated to remove free-floating parasites. Fresh RPMI (1 ml) with or without the drugs TAM 01/03, Glucantime®, or AmB (0.2µg/ml) at the appropriate concentration of Sb v (250µg to 7.8µg for L. chagasi and 40µg to 1.2µg for L. amazonensis and L. braziliensis) was added to wells in triplicate. The plates were incubated for 72 h at 37ºC in 5% CO 2 . The medium was aspirated and the coverslips were removed, air-dried, and glued to microscope slides. After staining with Giemsa, the cells were counted. The assays were considered valid if at least 80% of the macrophages in the control wells were infected. Three independent experiments in triplicate were performed for each concentration to determine the meglumine antimoniate efficacy. The results are presented as the ratio of infected (number of amastigotes) proportions between treated and non-treated macrophage cultures.
In vivo evaluation: golden hamster model for L. braziliensis -WHO-MHOM/BR/75/M2903 L. braziliensis was isolated from fragments of cutaneous lesions present in the paws of previously infected hamsters. The fragments were excised, macerated in saline solution, and quantified in a Neubauer™ chamber. Fifty-four male hamsters, weighing 100 to 110 g, were infected with L. braziliensis amastigotes (8x10 5 /100 µl) by subcutaneous injection in the hind footpad. Treatment by intramuscular injection of each formulation of TAMs (20 mg Sb v /kg/day) was initiated five days after infection and was given over the course of 30 consecutive days as follows: TAM 01/03 (group B), TAM 02/03 (group C), TAM 03/03 (group D), TAM 021/02 (group E), Glucantime® (drug-reference group F). Group A is the untreated control group, which received intramuscular injections of distilled sterile water. The lesions were examined weekly for eight weeks by measuring the size of the infected footpad with a vernier caliper. The animals were euthanized 30 days after the end of the treatment.
Data analysis -The data were processed using MINITAB V. 13.1 or Graph Prism 4 software. The active drug concentrations were compared by analysis of variance (ANOVA) or Kruskal-Wallis tests, depending on the distribution of the variables and paired comparisons done by the Dunn's or Tukey tests. IC 50 values were calculated by linear regression analysis (MINITAB V. 13.1) or linear interpolation (Microsoft Office Excel 2003) (Huber & Koella 1993). Linear regression was used when the distribution was normal (parametric method) and linear interpolation (non-parametric) was applied when the distribution was not normal.
Ethics -Animals were handled according to local and federal regulations, and the research protocols were approved by the Fiocruz Committee on Animal Research (protocol P-0321/06; licence L-0024/8).
Activity of the four batches of TAMs using the amastigote-like model -The leishmanicidal activities against
amastigote-like forms of L. chagasi and L. amazonensis of the four batches of TAMs, as well as meglumine antimoniate, Glucantime®-Aventis, AmB, and the control, are shown in Fig. 1. Activity was detected for concentrations that varied from 10 µg/ml to 300 µg/ml for L. chagasi (Fig. 1A) and from 50 µg/ml to 300 µg/ml for L. amazonensis (Fig. 1B). No statistically significant differences were observed among the four batches of TAMs at any concentration. The mean value of parasite inhibition observed with the control drug (0.2 µg/ml AmB) was 94.6% for L. chagasi and 80% for L. amazonensis.
Activity of TAMs using the amastigote-infected macrophage model -Peritoneal macrophages infected with amastigote-like forms of L. chagasi, L. amazonensis, and L. braziliensis were treated with different concentrations of Glucantime® and one batch of TAMs: TAM 01/03. The Table summarizes the comparative inhibitory concentrations.
Comparison of the IC 50 of Glucantime®-Aventis among Leishmania species showed statistical differences between L. chagasi and L. amazonensis (p < 0.01), and L. chagasi and L. braziliensis (p < 0.01). The values were not statistically different (p > 0.05) between L. amazonensis and L. braziliensis. In contrast, the IC 50 of TAM 01/03 were similar (p = 0.368) among the studied species. The mean value of parasite inhibition observed with the control drug (0.2 µg/ml AmB) was 99.7% for L. chagasi, 99.5% for L. amazonensis, and 98% for L. braziliensis.
In vivo activity of TAMs and Glucantime®-Aventis against L. braziliensis -Hamsters were infected with an amastigote suspension of L. braziliensis and treated with 20 mg Sb v /Kg/day of the four batches of TAMs or Glucantime® beginning at five days post-infection and continued for 30 consecutive days. The progress of the lesions was assessed weekly for eight weeks (Fig. 2). Nodes were observed in the footpads of the animals five days after infection (before treatment), and no statistically significant difference was observed among the node size of the groups (p = 0.854). After seven days of treatment, skin lesions remained similar among the groups (p = 0.056). Fourteen days after treatment initiation, Fig. 1: activity of meglumine antimoniate Glucantime®-Aventis and the four batches from Farmanguinhos (TAM 01/03, TAM 02/03, TAM 03/03 and TAM 021/02) against amastigote-like forms of L. chagasi (A) and L. amazonensis (B). Amphotericin B (AmB) was used as the reference control drug. Values represent the mean ± SD of three repeated experiments, each performed in triplicate. Significant differences for L. chagasi were observed at 10 µg/ml: control versus Gluc® and TAMs (p < 0.05); 50 µg/ml and 150 µg/ml: control versus Gluc®, Gluc® and TAMs (p < 0.01); 300 µg/ml: control x Gluc® (p < 0.01). For L. amazonensis, 50 µg/ml: control versus Gluc® and TAMs (p < 0.05); 150 µg/ml: control versus Gluc® and TAMs (p < 0.01); 300 µg/ml: control versus Gluc® (p < 0.01). however, the lesions became thicker in the group of untreated hamsters compared with the group that received TAM 02/03 (p < 0.01) or TAM 03/03 (p < 0.05). No statistical differences were observed among the four batches of TAMs or between each TAM and Glucantime®. This trend persisted for 29 days after the beginning of treatment, when the lesions were significantly larger in the untreated group compared to all treated-groups (TAM 01/03, p = 0.0000; TAM 02/03, p = 0.0002; TAM 03/03, p = 0.0000; TAM 021/02, p = 0.0000; Glucantime®, p = 0.0298). The lesions of all groups worsened after treatment end. Statistical differences were still significant between the untreated group and TAM 01/03 and TAM 03/03 at eight and 16 days after treatment ended (p < 0.05; p = 0.0036 and p = 0.0357; p = 0.0357, respectively). At 29 days after the end of treatment, lesions had increased in all groups, but the differences among groups were not significant (p = 0.157).
DISCUSSION
Despite their toxicity, pentavalent antimonials (Sb v ) in the form of sodium stibogluconate (Pentostam®) and meglumine antimoniate (Glucantime®) remain the drugs of choice for the treatment of leishmaniasis in many countries, mainly due to their effectiveness and relatively low cost compared with the therapeutic alternatives currently available.
In this work, the in vitro and in vivo activity of the TAMs was assessed and compared to the reference drug Glucantime®-Aventis. Due to the simplicity of the amastigote-like assay and the avoidance of animals use, this model was chosen for the initial testing and screening. As this model showed that the four batches have similar in vitro activity, only one of the batches (TAM 01/03) was tested in the macrophage assay to determine the IC 50. Moreover, for the three species tested, the IC 50 remained below the dose that was toxic for macrophages, which was determined to be higher than 300 µg/ml for each of the four batches (data not shown).
Alternative generic drugs for the treatment of neglected diseases are tactical for the disease control programs. The price and sustainability of drug production are key points that may hinder control strategies or, even worse, risk the lives of patients. Nevertheless, the quality of the drug production needs to be carefully assessed. It is important to note that the activity obtained with the four batches of the TAMs was reproducible, as interbatch variation is one important shortcoming associated with antimoniate production.
The threat of antimoniate to the lives of patients has been experienced in different countries. In 2000, the Brazilian Ministry of Health received notification from public health care centers of the high frequency of side effects reported by patients receiving meglumine antimoniate produced by Eurofarma (registered as similar to the reference product at the National Regulatory Agency -Agência Nacional de Vigilância Sanitária -ANVISA). The epidemiological and the clinical investigation led to physical chemistry analysis of the batches, which resulted in the detection of heavy metals (Silva Junior 2001). The higher frequency of skin reactions in patients was attributed to the presence of heavy metals (Romero et al. 2003).
In Nepal, fatal cardiotoxicity occurred among VL patients treated with a batch of generic sodium stibogluconate from GL Pharmaceuticals, Calcutta, India. These instances are in contrast with the low total death rate and the low death rate due to cardiotoxicity observed among patients treated with generic sodium stibogluconate from Albert David Ltd (Rijal et al. 2003).
Our in vitro and in vivo results showed that the TAMs was as active as the reference drug Glucantime®-Aventis against species of Leishmania of medical importance in Fig. 2: evaluation of the activity of meglumine antimoniate in male golden hamsters infected with L. braziliensis. Animals were treated vial intramuscular with 20mg Sb v /Kg/day of meglumine antimoniate (Glucantime®-Aventis and TAMs-Farmanguinhos) for 30 days. Vertical bars represent the average and standard error of lesion size (diameter) for each group. The arrows indicate the start and end of treatment. The ellipse indicates the maximum activity between the treatment-groups and the control. No statistical differences were observed between the four batches of meglumine antimoniate produced at Farmanguinhos and the reference drug Glucantime®. Brazil. Further steps are required to complete the generic drug development. The most challenging issue is to achieve a stable formulation as described by Cabral et al. (2008), which developed formulations of meglumine antimoniate with higher stability regarding the increase in Sb III percentage. Quality control of produced batches, chemical analysis, and bioequivalence studies would then suffice to recommend the adoption of the generic product as a low cost alternative to the branded reference drug. A clinical study addressing efficacy and safety would complement the current study. | 3,812.2 | 2008-06-01T00:00:00.000 | [
"Biology"
] |
Conceptual Bases of Full Realization of Women's Labour and Entrepreneurial Activity
Today, the problem of poverty has emerged as a global problem in all countries of the world. Due to the difference in economic development, poverty is different and has a relative meaning. Prevention and reduction of poverty and improving the quality of life of the people, increasing the interest and aspirations of women in women's entrepreneurship, creating favorable conditions for the development of their entrepreneurial activities are the main foundations of today's reforms. In a country where there is a gap between the incomes of the population, it is clear that there will be poor people. Therefore, poverty cannot be eradicated, but it can be reduced through the development of entrepreneurship. The poverty rate is inversely proportional to the economic level of the country, i.e., in developed economies, the poverty rate is low, and in weak economies it is high. The whole world has turned its attention to solving this problem. The accession of the Republic of Uzbekistan to global economic processes requires more active participation of women in the economic life of the country. Entrepreneurship is becoming an independent factor of women's sexual freedom in the economic sphere. At a time when society is renewing and entering the world economy, the development of women's entrepreneurship is encouraged. This is the main source of development of the real sector of the economy. Through the socio-economic development of women's entrepreneurship, it is possible to observe a certain positive effect on achieving sustainable economic development of the country, especially in the prevention of poverty. This article highlights the role of women's entrepreneurship in the country's economy and the problems in its development and their solutions, conclusions and recommendations.
Introduction
The study of the essence, importance and necessity of women's entrepreneurship in the socio-economic life of the country is a topical issue of scientific research. Scientific, theoretical and methodological study of the peculiarities of its formation, sustainability and development, assessment of the impact on the social status of the population, ensuring the effective organization of women's entrepreneurship in the regional economy are among the priorities of the national economy. Women's entrepreneurship plays an important role in the employment of the able-bodied population, the transition of the economy to an importsubstituting and export-oriented basis. The reason for this is explained by the seriousness of the demographic situation. The growing population is creating new problems related to employment. Women's entrepreneurship can play a positive role in solving problems. They are: -GDP per capita lags far behind the world average. This figure is $ 1,741 in our country; -The persistence of demographic pressure in our country has led to an increase in unemployment and poverty. There are now 4-5 million poor people; -As a result of maintaining a strong demographic situation in the labor market, there is a problem of creating enough jobs. The results led to an increase in poverty.
As a result of the economic policy pursued in the country for the development of women's entrepreneurship, a significant positive shift is observed. A number of measures have been taken to further develop women's entrepreneurship, make full use of available resources, production and labor potential, and thus create new jobs, saturate the domestic market with local goods.
Analysis of the relevant literature
The almost equal proportion of women (50.3%) and men (49.7%) in Uzbekistan indicates the potential for the development of women's entrepreneurship. According to the Chamber of Commerce and Industry of Uzbekistan, the total number of women entrepreneurs is 52,000, or 10% of the total number of business entities. More than 82% of women work in health and social services, and less than 10% in science, education, culture and the arts1. Entrepreneurial activities by women are longer lasting than those of the stronger sex. With true hopes and dedication, the woman is more determined than others to follow the plan to ensure the continuity of enterprise production. In practice, women show a tendency to compromise, a desire to take into account ethical principles and norms when entering into a partnership. They pay special attention to the social aspects of entrepreneurial activity. Women entrepreneurs show a desire to be on an equal footing with all levels of government. The main purpose of this is not primarily a sense of financial interest, but the fact that the state is focused on raising the legal status and providing adequate information, raising the status of women's entrepreneurship. Thus, the modern female entrepreneur can be described as follows: she has professional and great creative abilities, characterized by such qualities as perseverance, responsible approach, organization and careful handling of economically risky operations. These allow for a high level of administrative and financial control, selection of employees, proper provision of incentives, leadership of the team among related enterprises. In order to attract more women to entrepreneurial activities, it is very important to create conditions to free them from the usual household chores. Creating opportunities for child care, such as increasing the number of children admitted to preschools, allows women to set aside time to organize and run their own businesses.
Research methodology.
Methods of observation, comparison and systematic analysis were used in the study. The role of women's entrepreneurship in the economic development of the country is analyzed. Through the observation method, the level of poverty resulting from imbalances between the incomes of the population was observed in different countries and their results were compared. The role of women's entrepreneurship in preventing poverty was assessed.
Analysis and results
One of the main reasons for the lack of development of women's entrepreneurship is the high level of poverty in society. In order to further develop women's entrepreneurship in the country, in particular, to encourage women's entrepreneurship and innovative initiatives, the Presidential Decree No. PP-3856 of 14 July 2018 "On measures to improve and increase the efficiency of employment" including the creation of a number of favorable conditions for the involvement of women in entrepreneurial activities. Including: -Individual entrepreneurs who have organized hairdressing services, tailoring services, shoe repair services in rural areas with a population of less than 5,000 people, as well as public baths are exempt from all taxes until July 1, 2023. was; -Individual entrepreneurs are exempted from paying a fixed tax on each employee; -to hire up to 3 permanent employees, enter into employment contracts with family businesses without a legal entity, as well as other close relatives of working age, including spouses of able-bodied children and grandchildren, able-bodied the right to involve older brothers and sisters, their spouses as participants in a family business entity without forming a legal entity; -Individual entrepreneurs who received a microcredit for the first time at the expense of the Employment Promotion Fund are exempt from paying a fixed tax for 6 months from the date of its receipt; -Also, the fixed tax rate for individual entrepreneurs providing freight services in road transport of more than 3 tons was unified, almost halved and set at 2 times the minimum wage. They acknowledged that the status of women in society and providing them with decent work, as well as the creation of permanent jobs for women, as well as the extensive development of family business, home-based work, handicrafts, horticulture are still lagging behind. Critical conclusions also relate to the problems of organizing, managing, and providing financial resources for women's entrepreneurship. It was noted that one of the main tasks is to improve the welfare of women and their social protection. In any country, there are low-income segments of the population. Poverty reduction means the awakening of entrepreneurial spirit in the population, the full realization of the inner strength and potential of man, the implementation of economic and social policies to create new jobs. Therefore, it is necessary to develop a poverty reduction program with the World Bank, the EBRD, the United Nations Development Program and other international institutions, and conduct a thorough study on the basis of international standards to create a new methodology that covers the concept of poverty, its criteria and assessment methods. It was emphasized that In developed and developing market economies, there are different views on the concept of poverty by experts. In particular, economists from developed market economies A.Smith, J.M. Keynes, Y. Shumpeter, Paul E. Samoulson, N.H. Williams gave a number of definitions of the term poverty in their scientific research. According to I. Fisher, R. Dornbush, and R. Schmalenzi, poverty is morally and culturally constrained, although incomes are not physically viable enough to meet the needs of society within the current normative requirements of consumption. This is www.psychologyandeducation.net because they claim that their income will be much lower than the set level. According to the Russian economist A.A Razumov, poverty affects various spheres of social life, there is a high level of negative barriers to the socio-economic development of a particular individual, family and society as a whole. This will lead to a deterioration in the quality of life of the population, an increase in socio-economic and gender inequality, and will be an obstacle to successful social development. Poverty is not just food and clothing, lack of housing, lack of access to education and health services that people need. And he described it as a lack of income for the population to get everything they need for life, at least to a minimum. A distinctive feature of poverty among the Russian population is that more than 50 percent of the Russian poor are able to work, and 42 percent of them are employed. The vast majority of the unemployed are women. Given this critical economic situation, it is possible to employ a certain segment of the unemployed population through the development of women's entrepreneurship in the eradication of poverty and destitution in the country. This will contribute to the sustainable economic development of the country.
Conclusions and suggestions
For the formation and development of women's entrepreneurship, we must pay special attention to the following proposals: -Consistent implementation of feminism, that is, the creation of equal opportunities based not on gender, but on the qualities, knowledge, skills and abilities of women. This will give a new impetus to the economic development of the country; -Encouraging the employment of employed women through the unconditional implementation of preferential approaches of international financial corporations in supporting their entrepreneurship; -Conduct separate research and study the problems of the Ministry of Labor and Employment in the recruitment of unemployed women in labor exchanges, and on this basis to improve employment services; -In the formation of women's entrepreneurship, it is necessary to pay attention to the biological differences between men and women. But not to discriminate against women entrepreneurs, especially those with disabilities. They need to be provided with state social support, infrastructure (elevators, ramps, customized public transport) and amenities in the form of certain working conditions. Certain conditions are also necessary for the realization of the rights of women entrepreneurs. From the above, we can conclude that the role of women entrepreneurs in the socio-economic development of our country is invaluable. In the renewed Uzbekistan, women's entrepreneurship makes a worthy contribution to the innovative development of enterprises and organizations at the macro and micro levels. | 2,605.4 | 2021-02-04T00:00:00.000 | [
"Economics",
"Business"
] |
The role of apoptosis in early embryonic development of the adenohypophysis in rats
Background Apoptosis is involved in fundamental processes of life, like embryonic development, tissue homeostasis, or immune defense. Defects in apoptosis cause or contribute to developmental malformation, cancer, and degenerative disorders. Methods The developing adenohypophysis area of rat fetuses was studied at the embryonic stage 13.5 (gestational day) for apoptotic and proliferative cell activities using histological serial sections. Results A high cell proliferation rate was observed throughout the adenohypophysis. In contrast, apoptotic cells visualized by evidence of active caspase-3, were detected only in the basal epithelial cones as an introducing event for fusion and closure of the pharyngeal roof. Conclusion We can clearly show an increasing number of apoptotic events only at the basic fusion sides of the adenohypophysis as well as in the opening region of this organ. Apoptotic destruction of epithelial cells at the basal cones of the adenohypophysis begins even before differentiation of the adenohypophyseal cells and their contact with the neurohypophysis. In early stages of development, thus, apoptotic activity of the adenohypophysis is restricted to the basal areas mentioned. In our test animals, the adenohypophysis develops after closure of the anterior neuroporus.
Background
The adenohypophysis (Rathke pouch) is derived from the ectoderm and develops during the embryonic stage in the pharyngeal roof in front of the pharyngeal membrane before the anterior neuroporus closes. According to Starck (1975), the primordial Rathke pouch (saccus hypophysealis) is a transverse depression in the pharyngeal roof abutting the bottom of the diencephalon without interposed mesenchymal cells [1]. Later, the pouch loses connection with the pharyngeal roof, while a multitude of mesenchymal cells moves between the pharygeal roof and the bottom of the brain [2]. These mesenchymal cells later differentiate into the primordia of the cranial base. The cells of the adenohypophysis proliferate toward the bottom of the brain and further differentiate into hormonesecreting cells. An epithelial bridge may persist between the closed adenohypophysis and the pharyngeal roof (canalis craniophanryngeus) for a longer period. Occasionally, in 2% of the cases [1], this connection develops to the persisting form of a pharyngeal roof hypophysis [3].
Apoptosis is involved in fundamental processes of life, like embryonic development, tissue homeostasis, or immune defense. Defects in apoptosis cause or contribute to developmental malformation, cancer, and degenerative disorders. Apoptosis can be induced in response to many external stimuli (extrinsic pathway) including activation of death receptors such as tumor-necrosis factor (TNF)receptor 1 or Fas/CD95 by interaction with their cognate ligands [4,5]. Alternatively, various sensors of cellular stress receive signals, for example after DNA damage or growth factor deprivation leading to mitochondrial release of cytochrom c and other apoptogenic factors [6]. Both pathways converge on a cascade system of proteases, called caspases (cysteinyl aspartic proteinases). Activated caspases are the central initiators and executioners of the apoptotic program. Cellular apoptotic events as a consequence of programmed physiological cell death are histomorphologically identifiable by characteristic features such as cell shrinkage, membrane blebbing, and condensed and fragmented nuclear chromatin.
Methods
38 fetuses from pregnant LEW.1A-rats were collected by caesarean section on day 13.5 of gestation. The fetuses were fixed in 4% buffered formalin solution for 24 hours, embedded in paraffin, and serial frontal sections (5 μm) of the heads were stained with haematoxylin and eosin (HE). Immunohistochemistry on dewaxed and rehydrated sections was performed with the Vectastain Universal Quick kit (Vector Laboratories) according to manufacturer's protocol, and as discribed by Lotz et al. (2004) [7]. To determine apoptotic cell death caspase-3 activity was detected by an antibody that specifically binds to the cleaved and thereby activated form form of caspase-3 (Anti-ACTIVE Caspase-3 pAb; Promega, Cleaved Caspase-3 (Asp 175) Ab; Cell Signaling). Proliferating cells were detected using Anti-Rat Ki-67 Ab (MIB-5; DakoCytomation) as primary antibody after heat-induced epitope retrieval in 0.01 M Citrate buffer pH 6.0. Immune complexes were visualized with diaminobenzidine tetrahydrochloride precipitates, and the sections were subsequently counterstained with nuclear fast red (Vector Laboratories).
Results
At day 13.5 of embryonic development, the primordial adenohypophysis of the rat embryo presents as a cup-like indentation of the pharyngeal roof and, hence, originates from the ectoderm.
Regarding the development of the remaining cranial area, it may be mentioned that the development of the primary nasal ducts is largely completed and the lamina oronasalis at the end of the duct is not yet open. Thus, neither the nasal septum nor a primary palate has developed at this time. Moreover, the maxillary bulges have not yet fused with the nasal bulges, and the nasolacrimal duct between the maxillary bulge and the frontonasal bulge is partly open and not yet completely closed.
The developmental stage outlined here is in agreement with Keibel's (1937) findings [8], however, in his experiments involving Rattus norwegicus Erxleben this stage was reached on the 12 th embryonic day, while a connection between the adenohypophysis and the pharyngeal roof no longer existed on day 13.5.
Moreover, the trigeminal ganglion is clearly noticeable on both sides of the primordial adenohypophysis. Above the adenohypophysis, the diencephalon with the 3 rd ventricle is located (Fig. 1).
In Figure 2, the base of the adenohypophysis and the lumen of this primordial gland are shown in magnification. At the interface with the stomadeal ectoderm two opposite horizontal epithelial cones are noticed whose distance represents the pharyngeal opening of the Rathke pouch and amounts to 100 μm at maximum. A vertical plug running toward the stomadeum as described by Hinrichsen (1993) [9] in a SEM image was not detected in our specimens. Ki-67 marking (Fig. 3) reveals distinct proliferative processes both of the epithelial cells of the adenohypophysis and the surrounding mesenchymal cells. Particularly numerous proliferative cells were noticed in the multi-layered cell assembly and the lateral walls of the Rathke pouch, but not at its base, i.e., the cone-shaped interface with the oral cavity ectoderm. Thus, the marked proliferation of the mesenchymal cells at both sides of the adenohypophyseal base appears to exert lateral pressure on this primordial gland.
In contrast, the epithelial cells of the adenohypophyseal base and at the pharyngeal roof show a strong apoptotic activity around the bulge visualized by immunohistochemical evidence of active caspase-3 (Fig. 4). A similarly strong activity was not found in the remaining areas of the developing adenohypophysis. Figure 5 shows the cranial part of the adenohypophysis proliferating toward the diencephalon. Fewer mesenchymal cells are noticeable directly between the adenohypophysis and the diencephalon. Moreover, an increase of neuroectodermal cells is found at the base of the diencephalon.
Discussion
Histological serial sections of rat fetuses on day 13.5 of gestation were analyzed for proliferative and apoptotic cell activities during embryonic development of the adenohypophysis. Marked cell division was observed in the cranial area, whereas apoptotic processes were revealed primarily in the basal cells of the adenohypophysis. The process of pharyngeal roof closure has not yet been clearly described in literature thus far. Even though a temporary connection between the pinching off hypophysis and the pharyngeal roof has been reported, clear evidence of development-related cell processes during pharyngeal roof fusion is lacking.
Immunihistochemistry on paraffin section (4 μm) using anti-body against active caspase-3 visualizes apoptotic events in the basal epithelial cones of the Rathke pouch (arrows), S = stomatodeum, 20× The closure of the Rathke pouch mentioned by Hinrichsen (1993) [9] can thus be described in further detail. Epithelial closure appears to be initiated by induction of apoptotic counts in cells located at the pharyngeal (basal) side of the adenohypophysis. Apoptotic cell death is clearly noticed both histomorphologically and immunohistochemically by evidence of markedly increased amounts of active caspase-3. In contrast to observations of Han et al. (1998) [10] we failed to find apoptotic activities in all adenohypophyseal areas. The epithelial basal closure, therefore, compares to processes occurring during nasolacrimal duct development. The latter closes at the ectodermal surface involving apoptotic processes, while the lumen-directed double-walled epithelial sheet may persist for some time [11].
Regular apoptotic events in the adenohypophysis, e.g. depending upon hormonal influences, were described by several authors. However, all of these studies were confined to adult subjects whose adenohypophysis features fully differentiated cells [12][13][14][15]. The embryonic test individuals studied here have not yet arrived at such a stage of maturity.
Impressingly, the apoptotic processes begin long before cell fusion in the pharyngeal roof at a time when the opposite cells to be fused are not yet in contact, with a distance of about 100 μm.
The diencephalon overall features a narrow band of neuroectodermal cells. Distinct proliferation and reinforcement of the neuroectoderm is absent but opposite to the Rathke pouch. This site represents the early neurohypo-physeal primordium whose proliferating cells, in contrast to the marginal cells of the Rathke pouch, fail to display a high-prismatic shape (Fig. 2, 5). A direct connection between the neurohypophyseal and adenohypophyseal primordia without interposed mesenchymal cells after Starck (1975) [1] was not confirmed.
Conclusion
Programmed cell death (apoptosis) plays an important role in embryonic development and tissue homeostasis. In agreement with others our data suggest that a temporally and spatially regulated pattern of apoptosis is also essential for the development of the basis of adenohypophyseal structures. We can clearly show an increasing number of apoptotic events only at the basic fusion sides of the adenohypophysis as well as in the opening region of this organ. Apoptotic destruction of epithelial cells at the basal cones of the adenohypophysis begins even before differentiation of the adenohypophyseal cells and their contact with the neurohypophysis. In early stages of development, thus, apoptotic activity of the adenohypophysis is restricted to the basal areas mentioned. In our test animals, the adenohypophysis develops after closure of the anterior neuroporus. It should be stated that in the top of the developing adenohypophysis no apoptotic cells were detectible. And it is notable that this is a temporary result of the turning of the adenohypophysis out of the pharyngeal roof. | 2,293.2 | 2008-07-23T00:00:00.000 | [
"Medicine",
"Biology"
] |
Quantum internet using code division multiple access
A crucial open problem inS large-scale quantum networks is how to efficiently transmit quantum data among many pairs of users via a common data-transmission medium. We propose a solution by developing a quantum code division multiple access (q-CDMA) approach in which quantum information is chaotically encoded to spread its spectral content, and then decoded via chaos synchronization to separate different sender-receiver pairs. In comparison to other existing approaches, such as frequency division multiple access (FDMA), the proposed q-CDMA can greatly increase the information rates per channel used, especially for very noisy quantum channels.
Q
2][3] .Recent achievements in quantum information 4-10 have brought quantum networking much closer to realization.Quantum networks exhibit advantages when transmitting classical and quantum information with proper encoding into and decoding from quantum states [11][12][13][14][15][16][17] .However, the efficient transfer of quantum information among many nodes has remained as a problem yet to be solved [18][19][20][21][22][23][24] , which becomes more crucial for the limited-resource scenarios in large-scale networks.Multiple access, which allows simultaneous transmission of multiple quantum data streams in a shared channel, may provide a remedy to this problem.
Popular multiple-access methods in classical communication networks include time-division multiple-access (TDMA), frequency-division multiple-access (FDMA), and code-division multipleaccess (CDMA).See Fig. 1 for an illustration of different multiple-access methods.In TDMA, different users share the same frequency but transmit on different time slots, but timing synchronization and delays become serious problems in large-scale networks.In FDMA, different users share the same time slots but operate on different frequency bands.However, only a narrow band of the data transmission line has a low leakage rate and the bands assigned to different users should be sufficiently separated to suppress interference.Unlike TDMA and FDMA, CDMA utilizes the entire spectrum and time slots to encode the information for all users, while distinguishes different users with their own unique codes.Therefore, CDMA is adopted as the key technology of the currently-used third generation mobile communication systems, and can accommodate more bits per channel use 25 compared with TDMA and FDMA.It has achieved great success in commercial applications of classical communications.
Although FDMA has already been used in quantum key distribution networks [26][27][28][29][30] , to the best of our knowledge, CDMA has not yet been applied in quantum networks and internet 1 .A q-CDMA network would require that the states sent by each transmitting node of the quantum network are encoded into their coherent superposition before being sent to the common channel, and the quantum information for each of the intended receiving node is coherently and faithfully extracted by proper decoding at the end of the common channel.This, however, is not a trivial task but rather a difficult one.
In this paper, we propose a q-CDMA method via chaotic encoding and chaos synchronization among senders and receivers, which require a quantum channel to transmit quantum superposition states and N classical channels for chaos synchronization to decode the quantum signals at the receiver nodes.It can be seen that the proposed q-CDMA provides higher transmission rates for both classical and quantum information, especially in very noisy channels.
Results
To present the underlying principle of our method, we consider the simplest case, where two pairs of sender and receiver nodes communicate quantum information, encoded into quantized electromagnetic fields with the same frequencies, via a single quantum channel [see Fig. 2 The actions of the chaotic devices CPS i=1,2,3,4 induce the phase shifts exp [−iθ i (t)], where θ i (t) = t 0 δ i (τ ) dτ .Thus, to achieve faithful transmission between the senders and the receivers, the effects of δ 1 (t) and δ 2 (t) on the quantum signals should be minimized in the fields received by the nodes 3 and 4. Intuitively, this could be done by simply adjusting the system parameters such that δ 1 (t) = δ 3 (t) and δ 2 (t) = δ 4 (t).However, such an approach is impractical, because any small deviation in the system parameters can be greatly amplified by the chaotic motion, making it impossible to keep two chaotic circuits with the same exact parameters and initial conditions.
Instead, auxiliary classical channels between senders and the intended receivers can be used to synchronize the chaotic circuit as shown in Fig. 2(b).This classical chaotic synchronization helps to reduce the parameter differences between the chaotic phase shifters and to extract the quantum information faithfully.
Modelling of quantum CDMA network.Hereafter, for the sake of simplicity, we assume that CPS 1 (CPS 2 ) and CPS 3 (CPS 4 ) have been synchronized before the start of the transmission of quantum information, i.e., θ The whole information transmission process in this quantum network can be described by the input-output relationship where a † LA and a BS are the creation and annihilation operators of the auxiliary vacuum fields entering the linear amplifier LA and the second beamsplitter BS 2 .For the pseudo-noise chaotic phase-shift θ i (t), we should take an average over this broadband random signal, which leads to In Eq. (2), S δ i (ω) is the power spectrum density of the signal δ i (t), and ω li and ω ui are the lower and upper bounds of the frequency band of δ i (t), respectively.Equation (1) can then be reduced to For a chaotic signal with broadband frequency spectrum, the factor M i is extremely small, and can be neglected in Eq. (3).This leads to a 3 ≈ a 1 and a 4 ≈ a 2 , implying efficient and faithful transmission of quantum information from nodes 1 and 2 to nodes 3 and 4, respectively.
In our q-CDMA network, the information-bearing fields a 1 and a 2 , having the same fre-quency ω c , are modulated by two different pseudo-noise signals, which not only broaden them in the frequency domain but also change the shape of their wavepackets [see Fig. 2(b)].Thus, the energies of the fields a 1 and a 2 are distributed over a very broad frequency span, in which the contribution of ω c is extremely small and impossible to extract without coherent sharpening of the ω c components.This, on the other hand, is possible only via chaos synchronization which effectively eliminates the pseudo-noises in the fields and enables the recovery of a 1 (a 2 ) at the output a 3 (a 4 ) with almost no disturbance from a 2 (a 1 ).This is similar to the classical CDMA.Thus, we name our protocol as q-CDMA.
Quantum state transmission.
Let us further study the transmission of qubit states over the proposed q-CDMA network using a concrete model.The qubit states and , to be transmitted are encoded in the dark states of two Λ-type three-level atoms; i.e., atom 1 in cavity 1 and atom 2, in cavity 2, as shown in Fig. 3(a).The qubit states are transferred to the cavities by Raman transitions and are transmitted over the q-CDMA network.At the receiver nodes, the quantum states are transferred and stored in two Λ-type atoms; i.e., atom 3 in cavity 3, and atom 4 in cavity 4. We assume that the four coupled atom-cavity systems have the same parameters.Let |g i , |e i , and |r i be the three energy levels of atom i.As shown in Fig. 3(a), the |g i → |r i and |e i → |r i transitions are coupled with a classical control field and a quantized cavity field with coupling strengths Ω i (t) and g i (t).By adiabatically eliminating the highest energy level |r i , the Hamiltonian of the atom-cavity system can be expressed as where c i is the annihilation operator of the i-th cavity mode; g i (t) = gΩ i (t) /∆ is the coupling strength which can be tuned by the classical control field Ω i (t); and ∆ is the atom-cavity detuning.
The cavity fields c i are related to the travelling fields a i by where κ is the decay rate of the cavity field; and a 1,in , a 2,in (both in vacuum states) and a 3,out , a 4,out are the input and output fields of the whole system, respectively.
The chaotic phase shifters CPS i=1,2,3,4 are realized by coupling the optical fields to four driven Duffing oscillators, with damping rates γ, described by the Hamiltonian where x i and p i are the normalized position and momentum of the nonlinear Duffing oscillators, ω 0 /2π is the frequency of the fundamental mode, µ is a nonlinear constant, and is the driving force.The interaction between the field a i and the i-th Duffing oscillator is given by the Hamiltonian where g f−o is the coupling strength between the field and the oscillator.Under the semiclassical approximation for the degrees of freedom of the oscillator, the interaction Hamiltonian H i leads to a phase factor exp −i t 0 g f−o x i (τ ) dτ for the field a i .To simplify the discussion, we assume that all of the four Duffing oscillators have the same ω 0 , µ, f d , and ω d , but different initial states.Finally, the chaotic synchronization between CPS 1 (CPS 2 ) and CPS 3 (CPS 4 ) is achieved by coupling two Duffing oscillators by a harmonic potential The nonlinear coupling between the optical fields and the Duffing oscillators and the chaos synchronization to achieve the chaotic encoding and decoding could be realized using different physical platforms.For example, in optomechanical systems, the interaction Hamiltonian (7) can be realized by coupling the optical field via the radiation pressure to a moving mirror connected to a nonlinear spring (see Fig. 3(b)).Chaotic mechanical resonators can provide a frequencyspreading of several hundreds of MHz for a quantum signal, and this is broad enough, compared to the final recovered quantum signal, to realize the q-CDMA and noise suppression.Chaos synchronization between different nonlinear mechanical oscillators can be realized by coupling the two oscillators via a linear spring.This kind of synchronization of mechanical oscillators have been realized in experiments 32 , but it is not suitable or practical for long-distance quantum communication.Chaos synchronization with a mediating optical field, similar to that used to synchronize chaotic semiconductor lasers for high speed secure communication 33 , would be the method of choice for long-distance quantum communication.The main difficulty in this method, however, will be the coupling between the classical chaotic light and the information-bearing quantum light.
This, on the other hand, can be achieved via Kerr interactions.There is a recent report 34 that proposes to use Kerr nonlinearity in whispering gallery mode resonators to solve this problem.
Another approach for chaotic encoding and chaos synchronization between distant nodes of the network could be the use of electro-optic modulators (EOMs).See, e.g., Fig. 3(c).In this case, the input information-bearing quantum signal is modulated by the EOM driven by a chaotic electrical signal 35 .The EOM can prepare the needed broadband signal, and there have been various proven techniques of chaotic signal generation and synchronization in electrical circuits.Indeed, recently experimental demonstration of chaos synchronization in a four-node optoelectronic network was reported 36 .
To show the efficiency of state transmission in q-CDMA, let us calculate the fidelities states to be transmitted.By designing the control parameters g i (t), using the Raman transition technique 18 , we find for the particular chosen quantum states that the fidelities F 1 and F 2 can be approximated as When the Duffing oscillator enters the chaotic regime, we have M ≈ 0, leading to fidelities F 1 , F 2 ≈ 1, which means that the qubit states can be faithfully transmitted over the q-CDMA network.
Information transmission rates.
Next we consider the maximum transmission rates of classical and quantum information over the proposed q-CDMA network, and compare them, under certain energy constraints, with the achievable bounds of transmission rates in a q-FDMA network and in quantum networks without any multiple access method (i.e., single user-pair network).Here the classical information transmission rates are calculated in terms of the Holevo information 37,38 and the quantum information transmission rates are defined by the coherent information [39][40][41] .We assume that the frequencies allocated to different user pairs in the FDMA network are equally spaced such that the number of users is maximized and cross-talks between adjacent channels are suppressed.Moreover, we restrict our discussion to Gaussian channels and Bosonic channels, respectively for the transmissions of quantum and classical information.
We briefly summarize the main results here and in Fig. 5(a)-(c).(i) For lossless channels (i.e., η = 1 where η denotes the transmissivity of the central frequency of the information-bearing field), upper bounds of classical and the quantum information transmission rates for the proposed q-CDMA network are higher than those of the quantum FDMA and the single user-pair networks if the crosstalk in the q-CDMA is suppressed by setting M ≪ 1. (ii) With the increasing number N of user-pairs in the networks, q-CDMA increasingly performs better than the q-FDMA for classical and quantum information.(iii) Information transmission rates for the q-CDMA is more robust to noise.For fixed N, quantum information transmission rates of the q-FDMA and the single userpair networks degrades very fast to zero as the loss 1 − η increases from zero (ideal channel) to 1/2, whereas the q-CDMA network retains its non-zero rate even for very noisy channels.For the classical information transmission, the situation is similar except that the transmission rates of q-FDMA and the single user-pair network drops to zero when η = 0 which corresponds to a completely lossy channel.
The robustness of the proposed q-CDMA network for noisy channel can be explained as follows.The chaotic phase shifters in the q-CDMA network spread the information-bearing field across a broad spectral band.Thus, the energy distributed in a particular mode is almost negligible, and thus the photon loss is also almost negligible.Therefore, increasing η has very small effect on the transmission rates.In Fig. 5(b)-(c), we consider the noise to be broadband, and shows that the transmission rates of classical and quantum information over the q-CDMA network change only slightly.
Discussion
We have introduced a q-CDMA network based on chaotic synchronization where quantum information can be faithfully transmitted with fidelities as high as 0.99 between multiple pairs of nodes sharing a single quantum channel.The proposed quantum multiple-access network is robust against channel noises, and attains higher transmission rates for both classical and quantum information when compared to other approaches.A q-CDMA network based on our proposal requires the realization of two important issues.First, quantum interference of signals from different chaotic sources.This has recently been demonstrated by Nevet et.al. 42 .Second is the implementation of chaotic phase shifters and their synchronization.These could be implemented in various systems, including but not limited to optomechanical, optoelectrical 35 , and all-optical systems 33 .
In particular, whispering-gallery-mode (WGM) optical resonators are possible platforms as chaotic behavior in a WGM microtoroid resonator has been reported in Ref. 43 .Although synchronization of self-sustaining oscillations in directly coupled microring resonators have been demonstrated 44 , and mechanical mode synchronization in two distant resonators coupled via waveguides has been proposed 45 , demonstration of chaos synchronization in such optomechanical resonators are yet to be demonstrated.Although the tasks to be fulfilled are not trivial, we believe that we are not far away from such realizations due to the rapid pace of experimental and theoretical developments we have seen in the field in the past few years.We think that our proposal will pave the way for long distance q-CDMA networks, and will give new perspectives for the optimization of quantum networks.
Methods
Averaging over the chaotic phase shift.A chaotic signal δ i (t) can be expressed as a combination of many high-frequency components, i.e., where A iα , ω iα , φ iα are the amplitude, frequency, and phase of each component, respectively.Then the phase of the signal at any given time t can be written as Using the Fourier-Bessel series identity 31 : with J n (x) as the n-th Bessel function of the first kind, we can write If we take average over the "random" phase θ i (t), the components related to the frequencies ω iα should appear as fast-oscillating terms and thus can be averaged out.This treatment corresponds to averaging out the components that are far off-resonance with the information-bearing field, and keeping only the near-resonance components.Hence, only the lowest-frequency terms, with n α = 0, dominate the dynamical evolution.Thus, we have Since the chaotic signal δ i (t) is mainly distributed in the high-frequency regime, we have A iα ≪ ω iα .Using the expressions J 0 (x) ≈ 1 − x 2 /4, log (1 + x) ≈ x for x ≪ 1, it is easy to show that where Consequently, from Eqs. ( 9) and ( 10), we obtain the equation Input-output relationship of the quantum CDMA network.Here we calculate the input-output relationship of the quantum CDMA network shown in Fig. 6, we can express the input-output relationships of the chaotic phase shifters CPS i=1,2,3,4 as and those of the two beam splitters BS 1 and BS 2 and the linear quantum amplifier "LA", respectively, as and Then, using Eqs.(12-15), we obtain the total input-output relationship of the quantum network as where θ 1 and θ 2 are independent chaotic "noises" as we have not considered chaos synchronization yet.
1. Kimble, H. J. Upper bounds of classical (quantum) information transmission rates of different methods for noisy channel with 0 < η < 1.The correction factor in the q-CDMA network is M = 0.01.FDMA is constrained by the frequency bandwidth δω/ω = 0.2.All the methods are constrained with the total energy ǫ/ω = 1.C η c(q),CDMA(FDMA) denote the classical (c) and quantum (q) information transmission rates in q-CDMA and q-FDMA networks with transmissivity η.The rates for the single user-pair channel are C η c,sig and C η q,sig .
(a)].The schematic diagram of our strategy is shown in Fig. 2(b).The quantum information sent by the nodes 1 and 2 is first encoded by two chaotic phase shifters CPS 1 and CPS 2 , whose operation can be modelled by the effective Hamiltonian δ i (t) a † i a i , with δ i (t) being time dependent classical chaotic signals and i = 1, 2. This encoding spreads the spectral content of the quantum information across the entire spectrum.The two beams are then combined at the 50 : 50 beamsplitter BS 1 and transmitted via a common channel to the receivers.At the end of the channel, the quantum signal is first amplified by a phase-insensitive linear amplifier (LA), then divided into two branches by a second 50 : 50 beamsplitter BS 2 , and finally sent to nodes 3 and 4 through two chaotic phase shifters CPS 3 and CPS 4 , which are introduced to decode the information by applying the effective Hamiltonian −δ j (t) a † j a j , with j = 3, 4. Amplifier gain is set as G = 4 to compensate the losses induced by the beamsplitters.
Fig. 4
Fig.4(a), it is seen that there are three distinct regions representing how the chaotic motion affects
Figure 1 :
Figure 1: Illustration for different multiple-access methods.(a) TDMA: the users share
Figure 2 :
Figure 2: Diagrams of the quantum multiple access networks.(a) Quantum information
Figure 3 :
Figure 3: Quantum state transmission over q-CDMA network.(a) The broom-shaped or
Figure 4 :
Figure 4: Fidelities of quantum state transmission.(a) Fidelities F 1 and F 2 versus the
Figure 5 :
Figure 5: Quantum information transmission rates.(a) Fidelities F 1 and F 2 versus the
Figure 6 :
Figure 6: Input-output structure of quantum CDMA netwrok.The black dashed lines | 4,583.6 | 2012-04-08T00:00:00.000 | [
"Computer Science"
] |
Geometrical study on diseaserelated ncRNAs based on Z-curve method
The Z curve is a very useful method for visualizing and analyzing DNA sequences. It is a three-dimensional space curve that constitutes a unique representation of a given DNA sequence. It becomes more and more important to study non-coding regions in the recent years. Using Z curve method, 15 disease-related ncRNAs and some snoRNAs and miRNAs sequences are selected from the NONCODE database in this paper, which relate to Alzheimer Disease. The corresponding Z curves of the studied ncRNAs, sequences have been mapped and compared. The statistical features of the Z curves are obtained. These features indicate that the ncRNAs sequences playing same roles in the celluar process have almost the same Z-curves. And the base content in these sequences is almost same too.
INTRODUCTION
It is widely accepted that Non-coding sequences play important roles in the process of translation in organisms ranging from bacteria to mammals [1,2,3].At the present time the research on non-coding region and its function is still a hot field all over the world.Among the researches about non-coding sequences, the study on nonprotein-coding RNAs (ncRNAs) is becoming increasingly important and has been made great progress already.
Traditionally, most RNA molecules were regarded as carriers conveying information from the gene to the translation machinery [4].However, since the late 1990s, it has been widely acknowledged that other types of non-protein-coding RNA molecules are present in organisms ranging from bacteria to mammals, which affect a large variety of processes including plasmid replication, phage development, bacterial virulence, chromosome structure, DNA transcription, RNA processing and modification, development control and others [5].These observations suggest that the traditional view of the structure of the genetic reg ‰ ulatory systems in organisms is far from complete.And the considerable number of non-coding RNAs (ncRNAs) that has been detected in the past few years was largely unexpected [6].
As new members and classes of ncRNAs being progressively discovered, the understanding of the importance of ncRNAs in basic cellular processes is ever increasing.Although the functions of the many recently identified ncRNAs remain mostly unknown, increasing evidence stands in support of the notion that ncRNAs represent a diverse and important functional output of most genomes [7].
Furthermore, the understanding of the significance of ncRNAs as central components of various cellular processes has risen sharply over the recent years.However, there are so many unsolved problems in this field and many of these ncRNAs still have uncharacterized functions.
Some diseases, which have constituted a threat to human beings, are related to different ncRNAs.Such as Alzheimer disease, cancers, diabetes, heart diseases, etc. [8].Among these diseases, Alzheimer disease has become the fourth-biggest cause of the illness threaten the old men , s lives, next below the cancers，heart diseases and cerebrovascular diseases.Alzheimer disease is a progressive degenerative disorder of the brain characterized by a slow, progressive decline in cognitive function and behavior.As the disease advances, persons with Alzheimer disease have tough time with daily usage of things like using the phone, cooking, handling money, or driving the car.The disease is more common in elder population.It is estimated that Alzheimer disease affects 15 million people worldwide and approximately 4 million Americans [9].The neuropathologic hallmarks of the disorder are amyloid-rich senile plaques, neurofibrillary tangles, and neuronal degeneration.
It has reported that three genes with autosomal domi-
SciRes Copyright © 2009 HEALTH
nant mutations have been identified that may lead to Alzheimer symptoms in carriers before they reach age 60. [10].The clinical features of Alzheimer disease overlaps with common signs of aging, and other types of dementia, hence the diagnosis remains difficult.We make use the ZCURVE method, which is proposed by Professor Zhang Chun-ting, to analysis ncRNAs related to Alzheimer disease.ZCURVE is a geometrical approach to study DNA sequences.Based on the Z curve method, some global and local features of the sequence can be detected in a perceivable way [11].
In this work, we download 15 Specific ncRNAs (BC200 RNA) sequences from the NONCODE database, which relate to Alzheimer disease and come from different organisms.The corresponding Z curves of the selected sequences have been mapped and shown.By analyzing and comparing the Z curves, the common features of them are found and the features may be as a criterion to study same type of disease-related ncRNAs.
Material
The NONCODE database is an integrated knowledge database designed for the analysis of non-coding RNAs (ncRNAs).Since NONCODE was first released 3 years ago [15], the number of known ncRNAs has grown rapidly, and there is growing recognition that ncRNAs play important regulatory roles in most organisms.In the updated version of NONCODE (NONCODE v2.0), the number of collected ncRNAs has reached 206 226, including a wide range of microRNAs, Piwi-interacting RNAs and mRNA-like ncRNAs.The improvements brought to the database include not only new and updated ncRNA data sets, but also an incorporation of BLAST alignment search service and access through our custom UCSC Genome Browser [12].
All ncRNAs in NONCODE were filtered automatically from GenBank and the literature, and were then later manually curated.With the exception of rRNAs and tRNAs, all classes of reported ncRNAs are included.In addition to containing sequence data, NONCODE provides a user-friendly interface, a visualization platform and a convenient search option, allowing efficient recovery of sequences, regulatory elements in the flanking sequences, related publications and other information [13,14].
We pick up 15 ncRNA (BC200 RNA) and 20 snoRNA sequences from this database, which belong to specific ncRNAs and relate with Alzheimer disease.Adequately,we select miRNA of human, virus and sequences from miRNA database.All selected sequences can be directly downloaded from the webpage.
Method
The Z curve is a unique three-dimensional space curve representation for a given DNA sequence in the sense that each can be uniquely reconstructed given the other.Consider a DNA sequence read from the 5' to the 3'-end with N bases.Inspect the sequence one base at a time, beginning from the first base.Let the number of the inspecting steps is denoted by n, i.e., n =1, 2... N. In the nth step, count the cumulative numbers of the bases A, C, G and T, occurring in the subsequence from the first to the nth base in the DNA sequence inspected.Denoting the cumulative occurring numbers of the bases A, C, G and T in the above subsequence by A n , C n , G n and T n , respectively.The Z curve is a three-dimensional space curve and composed of a series of nodes P 0 , P 1 , P 2 , . . ., P N , whose coordinates , and (n = 0, 1, 2, . . ., N, where N is the length of the DNA sequence being studied) are uniquely determined by the Z-transform of DNA sequence.
. The three components of the Z curve, i.e., ， and , represent three independent distributions that completely describe the DNA sequence being studied.Furthermore, the three independent components , and have a clear biological meaning, respectively [11].It is noted that the Z curve defined above is generally not smooth at each node.Sometimes, a smooth procedure is needed.The B-spline functions are used to smooth the Z curve.For more detailed information about the Z curve defined, please refer to references [16,17,18].
In summary, the Z curve is the unique representation for a given DNA sequence in a three-dimensional space and each can be uniquely reconstructed from the other.It offers an intuitive and convenient approach to study DNA sequences geometrically.
Where, sequence 1, 14 and 15 belong to BC200 RNA and other sequences belong to BC200-alpha RNA.Their
SciRes Copyright © 2009
cellular roles are regulators, but their sequence length is different and coming from different organisms, respectively.and compare them, respectively (see Figures 1-6).From the obtained pictures, we can see obviously that all corresponding curves for BC200-alpha RNA are almost no disparity, not only having same shapes but also same tendency (see Figures 1-3).The same condition occurs in the BC200 RNA sequences (see Figure 4).
Using Z-plotter and Origin7.5 software, corresponding Z curves of the selected 15 sequences are mapped and part of typical curves are selected shown in Figures 1-6.In addition to mapping Z-curves, the base (A, C, G, T and GC) content of the studied sequences is respectively calculated based on the Z Curve Theory.The typical results are shown at Table 1.However, the corresponding Z curves of BC200 RNA and BC200-alpha RNA sequences have obvious disparities (see Figures 5,6).The fact shows the Z curves are different too, in spite of the studied sequences all related with one type disease but their functions are different.It means the shapes and tendency of Z curves is related with functions of ncRNA sequences.
We also select snoRNAs and microRNAs of human, Arabidopsis thaliana and virus in NONCODE and miRNA database, respectively.Then map the corresponding zcurves based on Z CURVE method and analyze them.
In addition, the n y n curves for the studied sequences show a global maximum at the position of about 120bp (BC200 RNA) or 190bp (BC200-alpha RNA).
Results are shown in Figure 7,8.
Discussion
We pick up part of typical Z curves of studied sequences And then, in the n z n curves of BC200 RNA and BC200-alpha RNA sequences, all <0 (see Figure 1) means strong H-bond bases (G/C) are in excess of weak H-bond bases (A/T).It indicates that this type of ncRNA is a stable structure and not mutated easily.At the same time, about
SciRes Copyright © 2009 HEALTH
Then we calculate the base content of A, C, G, T and GC in the studied sequences (see Table 1).For BC200 RNAs and BC200-alpha RNAs the results are 33%-35%, 28%-29%, 24%-25%, 13%-14%, and 52%-53%, respectively.This fact indicates that there is no obvious disparity on base content in the two types studied sequences.That is to say, the base content in the two types BC200 RNA's sequences is almost equal.
Adequately, we map and compare the Z-curves of snoRNAs, microRNAs, We can see the z-curves of one type ncRNA (miRNA) are very similar (see Figure 7).The same conditions occur in the sequences of snoRNA (see Figure 8).And the base content is almost equal in the same type ncRNA sequences.
CONCLUSIONS
Based on the above compare and analysis, a initial conclusion is drawn that all kinds of Z-curves (i.e.On top of this, the curves for the studied sequences show a global maximum at the position of about 120bp (BC200 RNA) or 190bp (BC200-alpha RNA).Furthermore, the almost same in each base content in the two types ncRNA sequences indicate that base content is related with their functions or playing roles.Furthermore, about all curves of BC200 RNA sequence, <0.Unfortunately, we don't know the biological signification about the above results.So many works will be done in our future research.
We do more tests for other type ncRNAs to test the conclusion.By mapping and comparing the Z-curves of snoRNAs, microRNAs, we can know that other type ncRNAs, also have the same statistical character as the BC200 RNA, both in sample of the z-curves and base content in the sequences.
ACKNOWLEDGMENT
This work was supported by Chinese National Key Fundamental Re-search Project (Grant No. 90403120) and Shandong Fundamental Research Project (Grant No. Y2005D12).We are grateful to Key Lab for Biophysics in Universities of Shandong for help with us.We also thank our colleagues for advice and for sharing protocols.
Figure 1 .Figure 2 .
Figure 1.The Z n -n curves for two BC200 RNA sequences 2 and 10 (coming from orangutan and crab-eating macaque, respectively).The curves are very similar in global and local.
Figure 3 .
Figure 3.The 3D curves for two BC200 RNA sequences 2 and 5 (coming from human and orangutan, respectively).The curves are same in global and local.
BFigure 4 .Figure 5 .SciResFigure 6 .
Figure 4.The Z / n -n curves for two sequences 1 and 14.The curves are very similar in global and local and all <0.n z
Figure 7 .
Figure 7.The Z curves of microRNAs of human, Arabidopsis thaliana and virus.The curves are very similar in global. | 2,786 | 2009-06-03T00:00:00.000 | [
"Biology",
"Computer Science"
] |
Flexural isostatic response of continental-scale deltas to climatically driven sea level changes
. The interplay between climate-forced sea level change, erosional and depositional processes
Introduction
Flexural isostatic processes in large deltaic depocenters generally reflect the accumulation of sediment loads on the crust over tens of millions of years and, within the depositional basin itself, can be treated as unidirectional subsidence at the first order (Fig. 1; e.g., Driscoll and Karner et al., 1994;Reynolds et al., 1991;Watts et al., 2009).Over timescales of 10 3 to 10 6 years; however, the nature of flexural isostatic response to crustal loading and unloading by ice during Quaternary climate-forced glacial-interglacial cycles is well documented: such glacial isostatic adjustments (GIAs) include bidirectional (subsidence and uplift) and cyclical deflections of the Earth's surface (e.g., Clark et al., 1978;Mitrovica and Peltier, 1991;Lambeck et al., 2014;Whitehouse, 2018).In addition to near-field flexural responses to the direct loading and unloading by ice, GIA includes far-field effects and is associated with hydro-isostatic responses of shorelines to climate-forced global changes in sea level (e.g., Chappell, 1974;Thom and Chappell, 1978;Lambeck and Nakada, 1990;Lambeck et al., 2003;Potter and Lambeck, 2003;Muhs et al., 2020, and many others).The cumulative deep-time (10 7 to 10 8 years or more) subsidence within the depositional basin represents the accommodation that results in preservation of the stratigraphic record and is generally well understood, but the stratigraphic imprint of bidirectional and cyclical flexural responses to changes in sediment and water loads over shorter timescales (< 10 6 to 10 7 years; Fig. 2), as driven by global climate and sea level change, remain less well known.Allen and Allen (2005), Jouet et al. (2008), andFrederick et al. (2019).Processes that can result in both uplift or subsidence, referred to as bidirectional, are shown in green.Sea level change produces hydro-isostatic adjustments that affect both the landward portion of the continental margin and the ocean basin.
A large body of data demonstrates close coupling over multiple timescales between climate and the globally coherent component of sea level change that reflects global changes in ice volumes (Fig. 3; Miller et al., 2020, and references therein).Interglacial periods are characterized by lower ice volumes, higher sea level, and more landward shoreline positions, and they generally result in sediment accumulation in more proximal parts of the shelf.In con-trast, glacial periods are characterized by higher ice volumes, lower sea level positions, and a corresponding basinward shift in shorelines, and they generally result in the preferential transfer of sediment to the shelf margin and deepwater parts of sedimentary basins (Johannessen and Steel, 2005;Carvajal et al., 2009;Blum et al., 2008;Sømme et al., 2009;Sweet and Blum, 2016;Mason et al., 2019;Sweet et al., 2020Sweet et al., , 2021)).Bidirectional high-frequency glacio-and Figure 2. Conceptual model illustrating the complex interaction between surface processes (erosion, red; deposition, black), sea level change, and flexural isostasy.Panel (a) shows the uplift generated by the erosion of the sediment load and the subsidence created by the deposition of the prograding sediment load.Panels (bc) show how the position of the sediment load changes with the position of the sea level (SL) and how hydrostatic uplift takes place as a result of sea level fall.These processes and interactions have been proposed to be key to explaining the concept of continental levering (e.g., Lambeck and Nakada, 1990;Clement et al., 2016;Whitehouse, 2018).HS stands for high stand, while LS stands for low stand.The dashed black line represents the elevation and or bathymetry of the initial surface.Modified from Watts (1989).
In this paper, we use numerical models to examine the impact of climate-forced global sea level change, erosional and depositional processes, and flexural isostatic response on the stratigraphic evolution of large deltaic depocenters over timescales of 10 6 to 10 7 years (Fig. 3).We use the basin and landscape dynamics code BADLANDS (Salles and Hardiman, 2016;Salles et al., 2018) to develop a continental-scale fluvial-deltaic to deep-water sediment dispersal system and perform a series of tests where we impose synthetic and climate-driven sea level curves of different frequencies to explore the relationships between patterns of sediment accumulation, flexural subsidence, and load partitioning.The coupling of surface processes with flexural isostasy allows us to investigate feedbacks between global sea level change, erosional and depositional processes, and flexural uplift and subsidence within a dynamic framework.Hence, we use an iterative approach because the flexural response can also generate relative changes in sea level, which in turn can modify the spatio-temporal characteristics of isostatic compensation.Our approach is more comprehensive than previous studies that have investigated the role of high-frequency flexural isostatic adjustments in a static manner (e.g., Blum et al., 2008;Jouet et al., 2008), and our experiments produce the clear stratigraphic impacts of high frequency, bidirectional flexural adjustments that are driven by climate-forced global sea level change on the morphology, architecture, and long-term stratigraphic evolution of large deltaic depocenters.
Components of vertical motion in passive margin deltaic depocenters
Accommodation from subsidence in large passive margin deltaic depocenters is generated by multiple processes that operate over different timescales, at different depths, and at different rates.(Fig. 1).At the largest spatial and temporal scales, tectonic subsidence is initiated during the initial rifting phase by stretching and thinning of the lithosphere (e.g., Mackenzie, 1978;Galloway, 1989), followed by thermal cooling and contraction of upwelled asthenosphere, which causes the margin to subside during the post-rift phase.
In some cases, the amount of post-rift subsidence may exceed the possible thermal contribution, in which case anomalous subsidence has been attributed to dynamic processes that reflect changes in mantle flow (e.g., Wang et al., 2017Wang et al., , 2020)).Vertical movements in passive margin deltaic depocenters over timescales of 10 6 years and longer are also known to inhttps://doi.org/10.5194/esurf-12-301-2024 Earth Surf.Dynam., 12, 301-320, 2024 clude flexural isostatic response to sediment loading, which produces bending of the lithosphere (Watts et al., 2009).Flexural responses of this kind are typically characterized by subsidence in the depositional basin and development of low-amplitude peripheral bulges hundreds of kilometers onshore and/or alongshore from the depocenter.Published examples include the Amazon (Driscoll and Karner, 1994;Watts et al., 2008), Niger (Doust andOmatbola, 1990), andMissis-sippi (e.g., Jurkowski et al., 1984;Feng et al., 1994;Frederick et al., 2019) systems.Over shorter Milankovitch orbital timescales of 10 4 to 10 6 years, GIA results from climatedriven growth of land ice and loading of continental land masses and the inverse, i.e., melting of land ice and unloading (glacial-isostatic adjustment or GIA) (e.g., Sella et al., 2007;Milne, 2015;Whitehouse, 2018).The growth and decay of land ice also results in global sea level fall and rise, with unloading and loading of the continental shelves.The impacts of GIA and associated hydro-isostasy extend through the marine to terrestrial parts of the system to different degrees but are bidirectional and cyclical over Milankovitch timescales.GIA has a large footprint that includes development of a foredeep proximal to the ice load and a low-amplitude peripheral forebulge located thousands of kilometers away, with each migrating as a flexural wave in response to growth and decay of ice masses.By contrast, hydro-isostatic impacts are generally limited to the marine environment and the lower reaches of large fluvial-deltaic systems but can result in cycles of uplift and subsidence that have been hypothesized to drive changes in the magnitude of fluvial incision and filling of coastal plain to cross-shelf valley systems as the river mouth migrates basinward and landward (Blum et al., 2013).
Growth faults (e.g., Gagliano, 2003;Armstrong et al., 2014;Shen et al., 2016;Pizzi et al., 2020), salt tectonics (Morley et al., 2011;King et al., 2012;Jackson and Hudec, 2017), and compaction of the sediment column (Athy, 1930;Meckel et al., 2007) also contribute to long-and short-term subsidence in deltaic depocenters.Subsidence patterns associated with these drivers can affect coastal and fluvial depositional processes; for example, the steering of fluvial channels and their associated patterns of deposition due to creation of local topographic highs and lows (hundreds to thousands of square kilometers, e.g., Gagliano, 2003;Armstrong et al., 2014;Morón et al., 2017).Moreover, for the modern Mississippi and other large deltas, delta plain subsidence over timescales of 10 4 years or less is dominated by dewatering and consolidation of recently deposited muddy and/or peatrich Holocene sediments within the uppermost tens of meters of strata (Törnqvist et al., 2008;Keogh et al., 2021;Steckler et al., 2022).
Subsidence processes in deltaic depocenters operate at different rates, and there is generally an inverse relationship between measured or estimated rates and the time interval over which rate measurements are integrated (e.g., Sadler, 1981Sadler, , 1999;;Frederick et al., 2020;Steckler et al., 2022).On one end of the spectrum, annual cycles of uplift and subsidence with amplitudes up to 60-75 mm have been recorded in the Amazon basin and the Ganges-Brahmaputra delta due to loading by, then drainage of, seasonal rains (Bevis et al., 2005;Steckler et al., 2010).Moreover, over timescales of decades to centuries, subsidence rates associated with consolidation and dewatering of Holocene deltaic sediments can be 5-15 mm yr −1 (e.g., Törnqvist et al., 2008;Jankowski et al., 2017;Keogh et al., 2021;Steckler et al., 2022).On the other end of the spectrum, time-averaged rates of vertical motion associated with deep-seated subsidence drivers (tectonic and/or dynamic subsidence, long-term loading and lithospheric flexure, and deep-seated long-term compaction) can be very low (Frederick et al., 2019): for example, over timescales of 10 6 years or more, flexural uplift of ∼ 0.1 mm yr −1 characterizes the landward margin of the Mississippi depocenter.Then, moving south from a hinge line on the delta plain, long-term subsidence rates increase from zero to ∼ 3 mm yr −1 at the shelf margin (e.g., Frederick et al., 2019;Fig. 11e).
Numerical methods
Measuring vertical motion from the longer-term stratigraphic record within a sedimentary basin tends to result in a sense of unidirectional subsidence with low rates that decrease as the interval over which they are measured increases.By contrast, studies of the more recent stratigraphic record of deltaic depocenters make it clear that isostatic processes can be driven by climate change and are both bidirectional and cyclical (e.g., Blum et al., 2008;Jouet et al., 2008;Bevis et al., 2005;Steckler et al., 2010).In this paper, we use the BAD-LANDS modeling software (Salles and Hardiman, 2016;Salles et al., 2018) to examine the effects of climate-forced sea level change, erosional and depositional processes, and bidirectional flexural isostasy on the morphology, architecture, and stratigraphic evolution of large deltaic depocenters in deep time.BADLANDS links landscape and basin dynamics through simulation of erosion, landscape evolution, and sedimentation where the rate of river incision is described using the stream power equation (Howard, 1980).The hillslope processes are simulated based on a linear sediment transport law, which sets the transport flux equal to a linear function of topographic slope: the simple diffusion law.The isostatic response to changes in water and sediment load is calculated using a two-way coupling between BADLANDS and gFlex (Wickert, 2016), which computes the flexural deflection of the Earth's surface by using an elastic plate flexure solution.For a more detailed description of BADLANDS' constitutive laws see Salles and Hardiman (2016) and Salles et al. (2018).
Our experimental design consists of a series of simulations where the initial configuration of the modeling domain resembles the topography of a natural source-to-sink system with 3400 m elevation in the headwaters, a length of 4500 km, a downstream-decreasing fluvial channel slope, and successive inflections in gradient associated with transitions from the coastal plain to continental shelf and from the shelf to slope (Figs.4a, S1, S2).To ensure that our simulated drainage basin produces a point-source for sediment input to the marine domain we imposed a longitudinal topographic low in the middle of the model.The resultant drainage basin, river channels, deltas, shallow marine shelves, and shelf margins are then self-generated as a result of landscape dynam- ics that are in turn controlled by hillslope and channel processes, sediment diffusion, uniform precipitation, sea level changes, and flexural isostasy.The three-dimensional spatial mesh size and temporal resolution were chosen to adequately reproduce the geomorphology and architectural evolution of deltaic systems at the first order (Fig. 4c) and to ensure that the frequency of sea level fluctuations is adequately captured.Details about the boundary conditions and input parameters used in the modeling simulations can be found in Table 1 and the data repository.
Our simulations produce catchment areas, river lengths, and volumes of deposited sediment that are consistent with the ranges observed in continental-scale source-to-sink systems such as the Mississippi and Amazon rivers (Fig. 4).We do not aim to reproduce the geological history of a particular continental-scale source-to-sink system, our goal is to better understand first-order relationships between sea level changes, flexural isostasy, and erosion and deposition.The general similarity between our simulated river length and catchment area and large-scale natural source-to-sink systems (Fig. 4b), supports choices made for our initial settings and allows us to draw general comparisons between simulated results and natural systems.
After the initial drainage basin and river system is formed, we perform a series of simulations that first hold sea level constant, then impose synthetic and climate-forced sea level curves.Synthetic curves were constructed to reflect frequencies of 500 kyr and 5 Myr, whereas climate-forced sea level curves were selected from intervals in geologic time with contrasting climatic regimes and therefore contrasting sea level amplitudes and directions: we extracted data from Miller et al. (2020) that correspond to the icefree Paleocene, ca.66-56 Ma (hereafter greenhouse) and the ice-dominated Oligocene ca.33.9-23.9Ma (icehouse) periods (Fig. 3b, c).These simulations are then compared to a suite of non-flexurally compensated models.The sea level curves we use are resampled to time steps of 50 kyr with outputs generated every 100 kyr.We let each simulation initialize and run for 2 Myr without any sea level fluctuations so that the delta can reach dynamic equilibrium without any disturbances in base level, and we then impose both the synthetic and climate-forced sea level changes.
For flexurally compensated models, we use an effective elastic thickness (T e ) of 50 km.This value is within the range observed in passive margins (Tesauro et al., 2012) and similar values have been used in previous flexural studies of continental-scale deltaic depocenters (e.g., Driscoll and Karner, 1994;Watts et al., 2009).Moreover, we test the sensitivity to T e by running a series of simulations where T e varies from 10 to 100 km: results show similar patterns of flexural response when using T e values < 60 km, whereas larger values result in a smaller flexural amplitude and larger flexural wavelength (Fig. S3), which is consistent with observations made for rigid lithosphere in cratonic regions (Tesauro et al., 2012).We use a spatially uniform T e value in all simulations and do not account for possible spatial variations that could generate small spatial differences in flexural deflections.However, we consider such differences to be negligible when compared to the flexural response caused by the large sediment loads typical of passive margins.To ensure the computed flexural response is in phase with our imposed sea level cycles, we use a temporal resolution of 100 kyr: here we have also conducted sensitivity tests with higher temporal resolution (up to 10 kyr), and the emergent behaviors are similar to those presented here.Our models do not include the effect of waves, tides, and shelf currents, although BADLANDS is capable of simulating wave-induced longshore drift (see Salles et al., 2018, which implements the equations from Longuet-Higgins 1970).We recognize the importance of waves, tides, and shelf currents in sediment transport, but those processes move sediment volumes generally in the vicinity of deltaic depocenters and therefore would not play a crucial role in sediment load redistribution.Our simulations hold precipitation constant over time and do not include tectonic or dynamic uplift in upstream regions or lithological changes in subsurface sedimentary layers.Such parameters clearly play a role in the sediment volumes and rates delivered to the marine environment, and therefore the accumulated sediment load, but are not considered here to avoid complicating the analysis of the effect of high-frequency, high-resolution isostatic adjustments on changes in crustal loading due to both the sediment and water load, the primary effect that we aim to unravel in this paper.Our simulations also do not account for sediment compaction or sediments of different grain sizes.
Model description
All of our simulations develop a drainage basin where rivers merge on the coastal plain and form a delta and act as a point source for sediment input to the marine domain.The simuhttps://doi.org/10.5194/esurf-12-301-2024 Earth Surf.Dynam., 12, 301-320, 2024 lated sediment flux is controlled by river incision and hillslope diffusion, whereas river incision is governed by precipitation, drainage area, the topographic gradient, and erodibility.Hillslope diffusion is dependent on diffusivity and the topographic gradient.Once the sediment load is deposited in the marine domain, a continental shelf emerges and shelfmargin clinothems (Fig. 4c) are generated.
Our results feature the response of the fluvial-deltaic system to flexural isostasy driven by climate-forced sea level change and illustrate significant impacts on the morphology, architecture, and stratigraphic evolution of continental-scale deltaic depocenters in deep time.Below, we highlight three distinct results: (a) the effects of flexural isostasy on the morphology and stratigraphy of large deltas, (b) the flexural isostatic response to sea level change and its effect on delta evolution, and (c) the effects of hydro-isostasy on delta evolution.
Effects of flexural isostasy on the evolution of deltaic depocenters
The comparison between flexurally compensated simulations and their non-flexurally compensated counterparts allows us to isolate the effects of flexural isostasy on delta evolution.
In non-flexurally compensated simulations, elevation is controlled by erosion whereas bathymetry is controlled by deposition and base-level changes caused by sea level fluctuations.By contrast, in flexurally compensated simulations vertical changes in erosion and deposition are amplified by sediment unloading and loading, which causes uplift and subsidence, respectively (Fig. 5d).In addition, the flexurally compensated simulations display peripheral bulges landward and alongshore of the deltaic depocenters, which are the result of the uplift and subsidence generated by flexural isostasy (Figs.4a, c, S5, S6).As expected, the spatial variation in flexural response is a consequence of the spatial and temporal variations of sediment and water loads, and the bulges and curvatures noted above are absent in simulations without flexural compensation (Figs.4c, S5, S6).Hence, our models capture features that have been used in natural systems to provide evidence of flexural isostasy.In all simulations, mean river mouth transit distances and lateral river mouth migration distances are significantly greater in non-flexurally compensated simulations when compared to their flexurally compensated counterparts (Figs. 5, 6).We attribute these patterns to reduced accommodation when flexural compensation is absent, which limits sediment accumulation close to the river mouth, forces the delta to prograde farther into the coastal ocean, and forces the river channel to migrate or avulse laterally, creating laterally extensive delta plains (Fig. 5a).Conversely, in simulations with flexural compensation, accommodation increases near the delta front as the load accumulates, which reduces cross-shelf transit distances and lateral migration of the river channel (Fig. 5c).To the extent that mean maximum rates of sediment accumulation at the river mouth are an indicator of accommodation that has been generated, they are significantly smaller in non-flexurally compensated simulations when compared to their flexurally compensated counterparts (Table S1).
4.3 Flexural isostatic response to sea level change and its effect on delta evolution and stratigraphy Our simulations show a close correspondence between crossshelf transit distances of river mouths and sea level position (Figs.6, 7), confirming the role that sea level plays as an important driver for sediment dispersal on passive margins (e.g., Carvajal et al., 2009;Sømme et al., 2009).Hence, we first describe the flexural response to sediment and water load at the river mouth.
Simulations where we impose synthetic sea level changes show that flexural responses are bidirectional at a rate of change and direction that amplifies the climate-forced sea level changes themselves.These results therefore demonstrate that significant bidirectional flexural compensation can be important at high frequencies, in this case with sea level cycles of 500 kyr and 5 Myr (Figs. 6f-h, 7b).In each case there is a background monotonic increase in the accumulated flexural subsidence as the sediment load increases through time, but flexure is uniformly unidirectional if, and only if, sea level is held constant (Figs. 6d,7b).The synchronicity between sea level change and flexural response is confirmed by spectral analyses that show peaks at frequencies similar to the imposed synthetic sea level fluctuations and flexural response at the river mouth (Fig. 7).Simulations that use climate-driven empirical sea level curves also show that flexural response is bidirectional, cyclical, and in-phase with the frequencies and amplitudes of sea level fluctuations as well (Fig. 6i-j), and thus reflects climate-driven changes in sediment loading and unloading, and climate-driven changes in sea level that produce change in thickness of the water column.
The synthetic stratigraphy produced by our simulations records the coupling of climate-driven sea level change, flexural compensation, and generation of accommodation, which in turn drives river mouth lateral migration and cross-shelf transit distances (Fig. 6p-y).First, simulations without flexural compensation show a larger proportion of hiatuses in the stratigraphic record when compared to flexurally compensated simulations (Fig. 6u-y), which we attribute to the absence of flexural subsidence Second, flexurally compensated simulations with no sea level change record a simple normal regression (Fig. 6u), whereas simulations with imposed synthetic and empirical sea level curves display significant hiatuses during sea level fall and basinward transit of the river mouth and shoreline (sensu the forced regression of Posamentier and Allen, 1999) but fewer hiatuses during sea level rise when the river mouth and shoreline migrate landwards.In a natural system, significant hiatuses would be produced (Salles et al., 2018).ND indicates no deposition.Model abbreviations are listed in Table 1. by the valley incision and cross-shelf extension of river systems that accompanies sea level fall, whereas few hiatuses would be produced during the period of valley aggradation and filling that accompanies shortening of the river system as the shoreline steps landward during sea level rise (Fig. 6u-y).Third, simulations with higher-amplitude icehouse sea level curves show a higher number of erosional hiatuses when compared to simulations with lower-amplitude greenhouse sea level curves.We attribute this observation to more frequent and extensive subaerial exposure of the shelf, with longer and more frequent cross-shelf river mouth transits in the icehouse case and less frequent and extensive exposure of the shelf with shorter and fewer cross-shelf transits of the river mouth in the greenhouse case.
Hydro-isostatic effects on delta evolution
In our simulations, net vertical changes are the result of the interplay between flexural isostasy (of the sediment and water loads) and erosion and deposition.By running a series of simulations where we impose different sea level scenarios, we were able to show that hydro-isostatic effects play an important role in the spatial and temporal evolution of large deltaic depocenters (Fig. 5).To further demonstrate the importance of hydro-isostasy we isolated the component of vertical changes that are solely associated with the water load by computing changes in the water column in the marine part of the modeling domain for each time step and then calculated hydro-isostatic responses in gFlex (Wickert, 2016): we chose this approach because BADLANDS and gFlex are fully coupled to calculate the flexural isostatic compensation of both the sediment and water load, and we therefore were able https://doi.org/10.5194/esurf-12-301-2024Earth Surf.Dynam., 12, 301-320, 2024 show the imposed sea level curve and the round symbol the point in time when the simulation output was taken.The variations in the direction of hydro-isostasy through time for GH and IH, which contrasts with the results from the simulations with no sea level change.Note that the length of the modeling domain is 4500 km; therefore, in some cases because of the scale it might look like the hydro-isostasy is not affecting the areas proximal to the shoreline.
to extract the water load component in the post-processing stage.As expected, our results show that the spatial variation of hydro-isostatic adjustments is a function of the magnitude and frequency of sea level change and related to the loading of the prograding sediment wedge .In simulations where sea level is held constant, changes in relative sea level and hydro-isostasy are associated with the progradation of the sediment wedge, which changes the distribution of water depths and in turn causes variations in the gradient of the shelf, shelf margin, and slope.By contrast, simulations with sea level change experience bidirectional hydroisostatic adjustments that vary in amplitude from the shoreline to the shelf margin (Fig. 8).We calculated the mean hydrostatic variation both along a cross-section (Fig. 9b) and at the river mouth (Fig. 9c) for each time step and found that simulations driven by a greenhouse sea level curve had a distribution skewed towards positive hydro-isostasy, which corresponds to uplift.These results are in agreement with the shelf-to-slope profile in Fig. 10 that shows that simulations with a greenhouse sea level curve have the highest elevations and bathymetries.As expected, simulations where the icehouse sea level curve was imposed display the largest range in hydro-isostatic response (Fig. 9), while the downdip shelf-to-slope profile has elevations and bathymetries that are 20 % lower than simulations with a greenhouse sea level curve.
Our simulations also incorporate the effects of erosion and deposition in a dynamic manner, and in doing so we are able to interrogate the cumulative effects of feedbacks between hydro-isostasy and erosional and depositional processes.First, our simulations with no sea level change produce large cross-shelf transits of the river mouth, as well as extensive progradation of the shelf margin and slope: without flexural compensation, the final elevation of the coastal plain is simply a result of the initial topography, erosion, and deposition and base-level changes (Figs. 9, 10).Second, in natural settings, shelf margin width and water depth correlate to the length scale and gradient of river systems and the magnitude of climate-driven sea level change (e.g., Blum and Womack, 2009;Blum et al., 2013): if sea level falls, the shelf is subaerially exposed, and river mouths transit the shelf to the shelf margin, which is located in shallow water depths, whereas if sea level rises, the shelf is submerged with the shelf margin residing in significant water depths.In our simulated icehouse scenario, with high-amplitude sea level changes, the shelf is inherently wide and the shelf gradient is steeper because of the long-distance back-and-forth cross-shelf river mouth transits: the depth of valley incision during sea level fall is amplified by large magnitude hydro-isostatic uplift, whereas corresponding large magnitude hydro-isostatic subsidence in response to sea level rise enhances accommodation and results in increased valley filling as the river mouth migrates landward.In contrast, in the greenhouse scenario and holding the river-system-scale constant, the shelf is inherently narrow due to low-amplitude sea level changes, river mouth transit distances are inherently short, and the shoreline remains proximal to the shelf margin (see Blum and Womack, 2009;Blum et al., 2013): the slope of the shelf is shallower than in the icehouse case because of relatively continuous and widespread delta topset aggradation.
Finally, our simulations demonstrate the potential role of high-frequency climate-forced sea level changes in generating high-frequency flexural responses because they drive changes in the location and volume of fluvial-deltaic sediment accumulation, which then drives changes in the spatial distribution of water depths.Our results show that accommodation is not first created and then filled later but instead coevolves in sync with the actual responses of fluviodeltaic systems to climate-driven sea level change.In this way, climate-forced sea level changes set up a feedback mechanism that results in self-sustaining destruction of accommodation and generation of hiatuses during sea level fall and self-sustaining creation of accommodation and sediment accumulation during sea level rise.
Implications for natural systems
The results of our simulations demonstrate that climateforced sea level changes result in self-sustaining creation and destruction of accommodation into which sediment is deposited and therefore plays a major role in delta morphology and stratigraphic architecture.This self-sustaining process might help explain the discrepancies between clinothem thicknesses, relative sea level, and global mean sea level estimates (Sømme et al., 2023).
We argue that the results of the non-flexurally compensated simulations can be extrapolated to ancient deltaic systems that formed over a rigid lithosphere.We interpret that such systems will have inherently long progradation distances, areally extensive delta plains, and relatively low accommodation because of the rigid lithosphere they overlie.For example, the Triassic sequences of the North West Shelf of Australia (Martin et al., 2018;Morón et al., 2019) and the Boreal Ocean (Klausen et al., 2019) represent deltaic successions that prograded > 500 km into the basin, creating widespread delta plains.Based on the fact that the reconstructed paleo-bathymetric relief of the Triassic Boreal Ocean is an order of magnitude smaller than other modern shelves, Klausen et al. (2019) conclude that a shallow marine basin is the most important prerequisite to create such a widespread delta plain.We equate those conditions to those observed in our non-flexurally compensated simulations.Another possible example of this phenomenon might be the Early Cretaceous McMurray Formation in the Alberta foreland, which represents a continental-scale river system (see Blum and Pecha, 2014;Wahbi et al., 2022), whose stratigraphic record is very thin and lacks an obvious thick deltaic depocenter within the area of preserved McMurray strata.
Our simulations show patterns and rates of bidirectional vertical motions that are similar to the chronostratigraphic https://doi.org/10.5194/esurf-12-301-2024 Earth Surf.Dynam., 12, 301-320, 2024 data analyzed by Frederick et al. (2019;see also Fillon, 2016) from the Mississippi deltaic depocenter (Fig. 11); however, vertical motions in our models are generated by flexural isostasy alone.We therefore argue that the correspondence between subsidence and uplift patterns in the Gulf of Mexico and in our models can be used to support the idea that the majority of the vertical motions in the Gulf of Mexico at timescales of 10 6 years or more are generated by flexural isostatic responses to sediment and water loads.
Rates of subsidence from our study can also be compared with rates of motion generated by other processes that include fault motion, compaction, and annual water loading.As noted above, in large deltaic depocenters, such as the Mississippi and Niger deltas, the locations of growth fault systems tend to migrate basinward as the shelf margin progrades (e.g., Galloway et al., 2008;Pizzi et al., 2020), and they illustrate this inverse relationship between rates and the time over which rates are integrated.Total throw on these faults can be measured in kilometers, but motion is inherently Where the flexural isostatic compensation is derived from both the sediment and water loads, the latter referred to as hydro-isostasy.Note how the shelf rollover depth adjusts to changes in sea level amplitude and frequency.In the models without flexural compensation the final elevation of the coastal plain is the result of the initial topography, in that case only erosion and base-level changes control the vertical adjustments, as illustrated by the lack of relief in the case with no sea level change.The grey-shaded background represents the longitudinal profile before the sea level condition is imposed (t = 2 Myr).Horizontal dashed lines show the minimum and maximum sea level for each of the scenarios, which correspond with the upper and lower boundary of the shelf's slope.episodic, and when faults are active they likely produce rates of vertical motion of tens of millimeters per day for short periods of time.However, long periods of quiescence in between episodes of active motion ensure that time-averaged rates of fault slip are very low over timescales of 10 4 years or more (e.g., Shen et al., 2017).
Our study focuses on the impact of the interplay between climate-forced sea level change, changes in sediment and water loads, and associated flexural adjustments on the evolution of continental-scale deltaic depocenters in deep time, but the results of our simulations also help explain the spatial variation of relative sea level changes in the Holocene as well (e.g., Lambeck et al., 2014;Khan et al., 2015;Whitehouse, 2018).Previous empirical studies that have calculated sea level fluctuations have noted that sea level estimates around deltas do not correspond with other adjacent sites (Lambeck et al., 2014).For example, many observations from large deltaic systems in Asia and comparisons of such data with observations from adjacent sites often indicate lower sea levels of the former, corresponding to differential subsidence rates on the order of 1 mm yr −1 for the past ∼ 8 ka.Interestingly, Lambeck et al. (2014) did not include data from the deltas of Chao Phraya of Thailand, the Mekong and Red rivers of Vietnam, and the Pearl and Yangtze rivers https://doi.org/10.5194/esurf-12-301-2024 Earth Surf.Dynam., 12, 301-320, 2024 of China, for the purpose of estimating the equivalent sea level function.We suggest the mismatch between subsidence rates in large deltaic depocenters and adjacent non-deltaic sites can be attributed to the interplay between spatially nonuniform progradation and aggradation of the deltaic sediment wedge and the spatially non-uniform flexural isostatic response to changes in sediment and water load sediment load as forced by climate-driven sea level change.
Conclusions
We use conceptual simulations to unravel the flexural isostatic response to the interplay between climate-forced sea level change, sediment erosion, and deposition in deep time on passive margin deltaic depocenters.Our results illustrate how cumulative high-frequency, bidirectional isostatic adjustments play an important role in driving along-strike and cross-shelf river mouth migration and sediment accumulation.Our results also show that accommodation coevolves in sync with flexural isostatic adjustments to climatedriven sea level change, and the associated responses of large fluvial-deltaic systems.We find that only simulations with no imposed sea level change produce a true monotonic increase in flexural subsidence and unidirectional deltaic progradation, regardless of timescale.We therefore view cyclical and bidirectional flexural responses to be inherent in fluvialdeltaic systems because they are coupled to the increasingly well-understood nature of how surface dynamics in fluvialdeltaic systems respond to climate-forced sea level changes to produce the stratigraphic record.
Figure 1 .
Figure 1.Conceptual diagram showing the components of vertical motions in passive margins with deltaic depocenters at different spatiotemporal scales.Rates of vertical motions are from Allen and Allen (2005), Jouet et al. (2008), and Frederick et al. (2019).Processes that can result in both uplift or subsidence, referred to as bidirectional, are shown in green.Sea level change produces hydro-isostatic adjustments that affect both the landward portion of the continental margin and the ocean basin.
Figure 3 .
Figure 3. Conceptual model for the evolution of (a) deltaic depocenters in response to sea level changes during (b) icehouse and (c) greenhouse periods.The frequent and large-amplitude variation in the sea level (purple line) during icehouse periods results in the exposure of the continental shelf and deep fluvial incision and sediment loads being deposited deep into the basin.In contrast, during greenhouse periods the smaller amplitude of the sea level variation (purple line) results in sediment being deposited closer to the continent.We test the flexural isostatic response of the aforementioned water and sediment partitioning.Figure modified from https://pubs.usgs.gov/(Schwab et al., 2009).(d) Evidence of the coupling between climate and global sea level change that reflects global changes in ice volumes.The coupling is particularly strong in ice-free (ca.66-56 Ma, greenhouse) and ice-dominated (ca.33.9-23.9Ma, icehouse) periods, which are the focus of this study.Data sources are as follows: sea level data are from Miller et al. (2020), temperature estimates are based on Mg/Ca values of Cramer et al. (2011), benthic foraminiferal δ 18 O data are from Miller et al. (2020), and the carbon isotope record δ 13 C data are from Zachos et al. (2001).
Figure 4 .
Figure 4. (a) Example of the outputs from numerical simulations that show elevation and bathymetry (top) and cumulative flexure (bottom).Discharge is shown in both maps to visualize the paths of the fluvial-deltaic system Model dimensions are 4500 km × 2000 km, with a vertical exaggeration of 100×.(b) Scatter plot of river length (top) and shelf width (bottom) versus catchment area from river systems.Data are from Somme et al. (2009), Nyberg et al. (2018), Blum et al. (2013, 2017), and simulations presented in this study.Pal represents Paleocene, Oli represents Oligocene, and PM represents Paleo-Mississippi.(c) Example of synthetic stratigraphy from a simulation without (left) and with flexural compensation (right).Vertical exaggeration is 150×.Cross-section A-A shows where the Wheeler diagrams presented in Fig. 5 were constructed.
Figure 5 .
Figure 5. (a) Output of numerical simulations with imposed synthetic sea level curves with different frequencies showing elevation, bathymetry, and discharge of the river mouth at 8 Myr.(b) River mouth (RM) location for each time step for all simulations, which illustrates the river mouth transit distance and lateral (L) migration.(c) Violin plots (modified kernel density plots) show that in all simulations mean river mouth transit distances (white dots) are significantly greater in non-flexurally compensated (NFC, lighter shades) simulations when compared to their flexurally compensated (FC, darker shades) counterparts.(d) Comparison of the down-dip shelf-to-slope profile for all the simulations.These results show that flexurally compensated models have significantly smaller progradation distances in the river mouth and the shelf break, smaller areal extent, and larger coastal elevations (due to flexural uplift) compared to their non-flexurally compensated counterparts.IH stands for icehouse, and GH stands for greenhouse.
Figure 6 .
Figure 6.Outputs of numerical simulations for the different scenarios where we impose synthetic (a-c) and empirical (d-e) sea level curves (SL).Outputs represent the temporal evolution of the flexural deflection (f-j) and travel distance of the river mouth (k-o).Wheeler diagrams show the changes in bathymetry through time for the flexurally compensated (p-t) and non-flexurally compensated (u-y) simulations.The location of the cross-section used to construct the Wheeler diagram is shown in Fig.4.Wheeler diagrams are constructed using the postprocessing module of BADLANDS(Salles et al., 2018).ND indicates no deposition.Model abbreviations are listed in Table1.
Figure 7 .
Figure 7. (a) Synthetic sea level curves with a frequency of 5 Myr (red), 500 kyr (grey) and no sea level change (SL = 0, black).Comparison of (b) flexural deflection and (c) transit distance extracted at the river mouth (RMT).(d) Power spectral density (PSD) for simulations with a frequency of 5 Myr (red) and 500 kyr (grey).The correspondence in peak frequencies between sea level, flexural deflection, and river mouth migration provides further evidence of the synchronicity between flexural deflection and sea level changes.
Figure 8 .
Figure 8. Map views showing the temporal evolution of the change in water column (left) and associated hydro-isostasy (right) at (a) 2.5 Myr and at (b) 10 Myr in the numerical simulations with different sea level scenarios: SL0 representing 0 constant, IH standing for icehouse, and GH standing for greenhouse.These result show the vertical changes that are solely associated with the water load.Insets in the left panelsshow the imposed sea level curve and the round symbol the point in time when the simulation output was taken.The variations in the direction of hydro-isostasy through time for GH and IH, which contrasts with the results from the simulations with no sea level change.Note that the length of the modeling domain is 4500 km; therefore, in some cases because of the scale it might look like the hydro-isostasy is not affecting the areas proximal to the shoreline.
Figure 9 .
Figure 9. (a) Temporal evolution of sea level, elevation, water column change, and hydro-isostasy along a down-dip section in the middle of the modeling domain for the simulations with different sea level scenarios: SL0 representing 0 constant (first row), GH standing for greenhouse (second row), and IH standing for icehouse (third row).The spatial variation of hydro-isostatic adjustments is a function of the magnitude and frequency of sea level change and related to the loading of the prograding sediment wedge.The positive values in the hydro-isostasy fluctuate because of bathymetric changes caused by the progradation of the sediment wedge and flexural subsidence.Dashed horizontal lines in the second column show the minimum and maximum sea level for each of the scenarios.Violin plots (modified kernel density plots) show the range of hydro-isostatic adjustments for the three sea level scenarios along a (b) cross-section (XS) and at (c) the river mouth (RM).Note how the largest range in the hydro-isostatic response occurs in the simulations where the icehouse sea level was imposed.
Figure 10 .
Figure 10.Comparison of the down-dip shelf-to-slope profile for scenarios with no sea level change (dark grey) greenhouse conditions (green) and icehouse conditions (blue) for (a) flexurally compensated and (b) non-flexurally compensated models.The net vertical change presented in this figure is the result of the interplay between flexural isostasy and erosion and deposition.Where the flexural isostatic compensation is derived from both the sediment and water loads, the latter referred to as hydro-isostasy.Note how the shelf rollover depth adjusts to changes in sea level amplitude and frequency.In the models without flexural compensation the final elevation of the coastal plain is the result of the initial topography, in that case only erosion and base-level changes control the vertical adjustments, as illustrated by the lack of relief in the case with no sea level change.The grey-shaded background represents the longitudinal profile before the sea level condition is imposed (t = 2 Myr).Horizontal dashed lines show the minimum and maximum sea level for each of the scenarios, which correspond with the upper and lower boundary of the shelf's slope.
Figure 11 .
Figure 11.Elevation bathymetry (top) and associated subsidence rates (bottom) from the Gulf of Mexico (a-e) and from simulations with no sea level change (SL = 0, b, f) greenhouse conditions (GH c, g) and icehouse conditions (IH d, h).(i) Vertical rates of motion measured over different timescales, illustrating a "Sadler" type rate decrease as a negative log-log relationship with timescales of measurement.
Table 1 .
Summary of the different input parameters used in the numerical simulations presented in this study.All the boundary conditions and parameters can be found in https://github.com/saraemp/egusphere-2023-53(last access: 23 January 2024).SL represents sea level, f represents frequency, A represents amplitude, FC represents flexurally compensated simulations, NFC represents non-flexurally compensated simulations, and T e represents elastic thickness. | 9,657.2 | 2024-02-01T00:00:00.000 | [
"Environmental Science",
"Geology"
] |
Synthesis and Application of Magnetic Nanoparticle Supported Ephedrine as a New Sorbent for Preconcentration of Trace Amounts of Pb and Cu in Water Samples
Um novo adsorbente para a determinação de Pb e Cu por espectrometria de absorção atômica de chama (FAAS) foi sintetizado e caracterizado por diferentes técnicas. O efeito de vários parâmetros como pH, tipo e volume do eluente, quantidade de adsorbente, volume da amostra e íons interferentes foram otimizados. Nas condições otimizadas, um gráfico de calibração linear foi obtido para a determinação de Pb(II) e Cu(II). As faixas lineares foram 15-500 μg L−1 e 18-500 μg L−1 para chumbo e cobre, respectivamente. Os limites de detecção e quantificação foram 4,3 e 14,5 μg L−1 para chumbo e 5,0 e 16,7 μg L−1 para cobre, respectivamente. O desvio padrão relativo para determinações de oito replicatas de 80 e 200 μg L−1 de Pb(II) foram 2,9 e 1,4% e para Cu(II) foram 3,5 e 1,9%, respectivamente. A aplicabilidade do método foi avaliada ao analisar traços de chumbo e cobre em diferentes amostras de água.
Introduction
In the past decade, the entry of pollutants of heavy metals has increased in the global ecosystem. 1The amount of entering of heavy metals into the environment is far beyond the amount that is removed by natural processes.Their accumulations in water, air and soil and their nonmetabolized is an important environmental problem.Moreover, they have deposited and accumulated in tissues such as fat, muscle, bones and joints.This causes several diseases and complications in the body.Most heavy metals in aquatic systems are Cu, Zn, Cd, Hg, Pb and Ni.It has long been found that, in the appropriate concentrations, many metals including Fe (hemoglobin), Cu (respiratory pigments), Zn (enzymes), Co (vitamin B 12 ), Cr (carbohydrate metabolism), Se (antioxidant role), Mo and Mn (enzyme) are essential to living organisms but may be toxic at high concentrations.Metals such as Hg, Pb, Sn, Ni and As are generally not required for metabolic activity and are toxic at quite low concentrations. 2herefore, the measurement of trace amounts of heavy metals in environmental samples is very important for analytical chemists.Several techniques for measuring these elements have been employed such as flame atomic absorption spectrometry (FAAS), 3,4 electrothermal atomic absorption spectrometry (ETAAS), 5 inductively coupled plasma mass spectrometry (ICP-MS), 6 inductively coupled plasma optical emission spectrometry (ICP-OES) [7][8][9][10] and spectrophotometry. 11Among these, FAAS, due to its simplicity and its lower price is more common than other instruments.But this technique has no sufficient sensitivity in direct determination of metals, therefore, separation and preconcentration methods including: coprecipitation, [12][13][14] liquid-liquid extraction, 15 cloud point extraction, 16 solid phase extraction 17,18 can solve this problem.Solid phase extraction (SPE) is widely used due to its simplicity, consumption of small volumes of organic solvent, high preconcentration factor, high recovery, rapid phase separation and the ability to combine with different modern detection techniques. 19SPE sorbent selection is critical to obtaining efficient SPE recovery.Until now, various adsorbents such as chelating resin, 20 activated carbon, 8 modified or bonded silica gel, 21 polyurethane foam, 22 naphthalene 23 and cellulose 24 have been used for metal ion sorption.Recently, nanomaterials have applied as one of the most promising adsorbents in SPE.A new method of solid phase extraction, based on the use of magnetic adsorbents has been developed. 25,26Magnetic nanoparticles offer many advantages due to the unique size, easy separation from the solution using a magnet and high surface area. 3,7he suitable surface coatings and effective protection strategies have developed to prevent the accumulation of nanoparticles, chemical analysis in specific environments as well as changes of the magnetic properties in complex environmental samples and biological systems. 10n this paper, a new adsorbent, magnetic nanoparticle Fe 3 O 4 -immobilized ephedrine (MNPs-ephedrine) for the determination of Pb and Cu by FAAS is synthesized and characterized with different techniques.In the best of our knowledge, this adsorbent has not been used to the separation and preconcentration of trace amount of metal ions.Applicability of the method was evaluated by analyzing trace amounts of lead and copper in different water samples.
Experimental
Instrumentation A Shimadzu model AA-670 atomic absorption spectrometer (Kyoto, Japan) equipped with lead and copper hollow cathode lamps and air-acetylene flame was used for determination of the metal ions.All pH settings were carried out by a Metrohm E-691 digital pH meter (Switzerland) with a combined glass electrode.The infrared spectra were recorded using an infrared spectrometer (Bruker-Vector 22, Germany) with KBr disks in the range of 4000-400 cm −1 .The scanning electron microscopy (SEM) image was obtained by VEGA TESCAN (Czech Republic).The XRD data were collected on an X'PertMPD Philips diffractometer (Netherland) with Cu Ka radiation source (λ = 1.54050Å) at 40 kV voltage and 40 mA current.The TGA was carried out on a Bähr STA 503 instrument (Germany) under air atmosphere, heating rate 10 °C min −1 .The magnetic measurements were carried out in an Alternating Gradient Force Magnetometer (AGFM, Meghnatis Daghigh Kavir Co., Made in Iran) at room temperature.A magnet (Nd-Fe-B, 1.2 T, 50 × 40 × 20 mm) was used for magnetic separation.
Reagents and solutions
All materials used in this work were of analytical grade from Merck Co. (Darmstadt, Germany).Stock standard solutions of Pb(II), Cu(II) (1000 mg L −1 each one) were purchased from Merck.Solutions with lower concentrations were prepared daily by suitable dilution of the stock solution with deionized water.(3-chloropropyl)-trimethoxysilane (CPTMS) and ephedrine hydrochloride were utilized for the synthesis of sorbent.The pH 4.5 of the sample solution was adjusted using acetate buffer.The HCl, H 2 SO 4 and HNO 3 solutions, using as eluents were prepared by dilution of the concentrated solutions with deionized water.
Preparation of large-scale the magnetic Fe 3 O 4 nanoparticles (MNPs) Masses of 4.865 g FeCl 3 •6H 2 O and 1.789 g FeCl 2 •4H 2 O were added to 100 mL deionized water and sonicated until the salts dissolved completely.Then, 10 mL of 25% aqueous ammonia was added quickly into the reaction mixture in one portion under N 2 atmosphere at room temperature followed by stirring about 30 min with mechanical stirrer.The black precipitate was washed five times with doubly distilled water. 27
Preparation of MNPs coated by (3-chloropropyl)trimethoxysilane (MNPs-CPTMS)
Mass of 1.500 g MNPs powder was dispersed in 250 mL ethanol/water solution with volume ratio, 1:1 by sonication for 30 min, and then 2.5 mL CPTMS (99%) was added to the mixture.After mechanical stirring under N 2 atmosphere at 33-38 °C for 8 h, the suspended substance was separated with centrifugation (for 30 min).The settled product was re-dispersed in ethanol by sonication.The final sample was separated by an external magnet and washed five times with ethanol.The product was stored in a refrigerator to use. 27
Preparation of MNPs-ephedrine ligand
To prepare the MNPs-ephedrine ligand, 1.00 g MNPs-CPTMS was dispersed in 6-8 mL dry toluene by ultrasonication for 10 min.Subsequently, 0.403 g ephedrine Vol. 25, No. 11, 2014 hydrochloride and 0.672 g sodium bicarbonate were added and the mixture was refluxed for 28 h.Then, the final product was separated by magnetic decantation and washed twice by dry CH 2 Cl 2 and ethanol respectively to remove the unattached substrates.The product was stored in a refrigerator until use.
Preparation of environmental waters
The water samples, such as tap water (Sanandaj, Iran), sea water from Caspian Sea (Rudsar, Iran), wastewater of oil refinery (Kermanshah, Iran) were stored in pre-cleaned polyethylene bottles for use.The pH water samples were adjusted to 1 with concentrated HNO 3 .Before the analysis, the sea water and wastewater of oil refinery samples were filtered through a filter paper to remove suspended particular solids.Then, all samples were placed in refrigerator at approximately 4 °C.
General procedure
A 100 mL aqueous solution containing 10.0 µg Pb(II), Cu(II) was prepared and pH was adjusted at 4.5 with 10.0 mL acetate buffer solution (0.01 mol L −1 ).The solution was added to 10.0 mg adsorbent in 100 mL beaker.The sample solution was sonicated for 10 min to simplify sorption of lead and copper ions.Then, a strong magnet was used and the magnetic adsorbent separated after a few minutes and the supernatants were decanted.For elution adsorbed analyte ions from nanoparticles, 2.0 mL H 2 SO 4 and HCl mixtures (0.5 mol L −1 each one) with 2:1 volumetric ratio were added and the solution was again sonicated for 6 min and exposed on the magnet to deposit the magnetic nanoparticles.Afterwards, the eluate containing metal ions was determined by FAAS, using the conditions recommended by the manufacturer with a flow rate of 2.0 mL min −1 .
Preparation and characterization of MNPs-ephedrine
The process of the preparation of MNPs-ephedrine is shown in Scheme 1.
The metal ions are easily adsorbed on MNPs-ephedrine because ephedrine possesses a hydroxyl group and a nitrogen atom, which can complex with Cu (II) and Pb(II). 28he MNPs-ephedrine has been characterized by SEM, X-ray diffraction (XRD), thermo gravimetric analysis (TGA), Fourier transform infrared spectroscopy (FTIR) and alternating gradient force magnetometer (AGFM).
The XRD pattern of MNPs-ephedrine is shown in Figure 1.According to the XRD analysis, the peaks with 2θ at 30.4°, 35.6°, 43.3°, 57.3° and 62.8° indicated the characteristic peaks of Fe 3 O 4 .Weak broad bands (2θ = 11.5-23°)appeared in XDR pattern which could be attributed to the amorphous silane shell formed around the magnetic cores.
The SEM image of MNPs-ephedrine is shown in Figure 2. It was confirmed that MNPs-ephedrine were made up of uniform nanometer-sized particles 20-32 nm.
Figure 3 shows FTIR spectra for MNPs, MNPs-CPTMS, and MNPs-ephedrine.The bands at low wavenumbers ≤ 700 cm −1 come from vibrations of Fe-O bonds of iron oxide, which for the bulk Fe 3 O 4 samples appear at 570 and 375 cm −1 but for Fe 3 O 4 nanoparticles appear at 624 and 572 cm −1 as a blue shift, due to the size reduction.The FTIR spectra of MNPs-CPTMS and MNPs-ephedrine show Fe-O vibrations in the same vicinity.The introduction of CPTMS to the surface of MNPs is confirmed by the bands at 1005 and 1128 cm −1 assigned to the Fe-O-Si and C-Cl stretching vibrations, respectively. 27Reaction of MNPs-CPTMS with ephedrine produces MNPs-ephedrine, in which the presence of ephedrine is demonstrated with stretching vibrations at 3341 and 3380 cm −1 , which incorporates the N-H and O-H bonds and vibrations in the range of 1428-1652 cm −1 are attributed to the phenyl ring.
One indication of bond formation between the nanoparticles and the ligand can be inferred from TGA.The TGA curve of the MNPs-ephedrine shows the mass loss of the organic functional group as it decomposes upon heating (Figure 4).The weight loss at temperatures below 200 °C is due to the removal of physically adsorbed solvent and surface hydroxyl groups.Organic spacers have been reported to desorb at temperatures above 260 °C. 27he curve shows a weight loss about 19% from 260 to 600 °C.The loading of the ligand in MNPs-ephedrine can Scheme 1.The synthesis of ephedrine-functionalized magnetic Fe 3 O 4 nanoparticles.be calculated from TGA, which confirmed loading approximately 0.35 mmol g −1 .
Superparamagnetic particles are beneficial for magnetic separation; the magnetic property of MNPs and MNPs-ephedrine were characterized by AGFM.The room temperature magnetization curves of MNPs and MNPs-ephedrineare are shown in Figures 5a and 5b.As expected, the bare MNPs, showed the higher magnetic value (saturation magnetization, Ms) of 74.3 emug −1 , the Ms value of MNPs-ephedrine is decreased due to the silica coating and the layer of the grafted ligand (30.4 emug −1 ).As a result, the MNPs-ephedrine has a typical superparamagnetic behaviour, and can be efficiently attracted with a small magnet.
Unfortunately, due to the magnetic properties of MNPs-ephedrine, it is actually impossible to further characterize this material by using solid-state nuclear magnetic resonance (NMR) spectroscopy.
The initial experiments indicated that the lead and copper ions are effectively adsorbed on MNPs-ephedrine.In order to use the present method for preconcentration and determination of trace amounts of metal ions, various parameters including, the pH of the sample solution, the type, concentration and volume of eluent, the amount of adsorbent, the sorption and desorption time, the volume of solution the effect of different coexisting ions were optimized.The optimization of each of these parameters for determination Pb(II) and Cu(II) is described in the next sections.
Effect of pH
Due to the surface charge of the adsorbent and the solution chemistry of the metal ions, the pH of the aqueous solution is an important controlling factor in the uptake process of metal ions on adsorbents.Thus, the effect of pH, on the preconcentration of solutions containing 100 µg L −1 Pb(II) and Cu(II) in the pH range of 3.0-10.0was evaluated.The pH of solutions was adjusted using either hydrochloric acid or sodium hydroxide solutions (0.1-1.0 mol L −1 ) and the result is shown in Figure 6.As can be seen, low extractions of Pb(II) and Cu(II) at pH < 4.0 may be due to the competition of proton with analyte ions for sorption on the MNPs-ephedrine surface and in the higher pHs (pH > 5.0), may be due to formation of metal hydroxide species such as Pb(OH) + , Pb(OH) 2 , Pb(OH) 3 , [29][30][31] that leads to the decrease in the efficiency of extraction.As can be seen from Figure 6, lead and copper ions were quantitatively recovered (≥ 95%) in the pH range of 4.0-5.0.So, the pH 4.5 was selected as the optimum value for subsequent experiments.This optimum pH was adjusted with 10.0 mL of acetic acid-acetate buffer solution.
Effect of type and volume of eluent
The desorption process of Pb(II) and Cu(II) from sorbent is influenced by the type and concentration of eluents.Various eluents such as EDTA, HNO 3 , HCl, H 2 SO 4 with different concentration and the mixture of HCl and H 2 SO 4 (0.5 mol L −1 ) were tested.As shown in Figure 7, a mixture of 2:1 v/v of H 2 SO 4 and HCl (0.5 mol L −1 each one) provided higher efficiency compared to other eluents.Thus, a mixture of 2:1 v/v of H 2 SO 4 and HCl was chosen for further studies.
The effect of the eluent volume on the extraction of metal ions was also studied.Different volumes of eluent in the range of 2.0 to 5.0 mL were examined.According to the results shown in Figure 8, 2.0 mL of the mixture of 2:1 v/v of H 2 SO 4 and HCl (0.5 mol L −1 each one) was sufficient for quantitative recovery of analyte ions.
Effect of amount of adsorbent
In the sorption step, the selection of a proper amount of adsorbent is very important.In order to investigate the effect of the quantity of adsorbent on preconcentration of lead and copper ions, various amounts of modified MNPs 5.0 15.0 mg were used.The results shown in Figure 9, indicate that the quantitative recovery (> 95%) for Pb(II) and Cu(II) was obtained when the amount of adsorbent was greater than 5.0 mg.Therefore, in the further experiments, 10.0 mg of adsorbent was applied because it showed higher values of recovery.
Effect of sorption and desorption time
The ultrasonic times of analytes sorption and desorption were evaluated.According to the experimental results, the quantitative recovery of lead and copper ions was obtained when ultrasonication time was greater than 10 min for sorption and greater than 6 min for desorption.Therefore, the optimum times of sorption and desorption were 10 min and 6 min, respectively.
Effect of sample volume
For the preconcentration of trace elements, it is important to have high preconcentration factors.In order to achieve, the preconcentration factors, maximum applicable sample volume must be examined.For this purpose, the sample volumes of 50, 75, 100, 150, 200, 300 and 500 mL containing 10.0 µg of Pb(II) and Cu(II) were studied according to the recommended procedure.The quantitative recoveries were achieved when the volume of solution was less than 200 mL.Hence, a sample volume of 200 mL was selected as the largest usable sample volume.Thus, in this work, by using 2.0 mL of elution solution, a preconcentration factor of 100 is obtained.
Effect of diverse ions
In order to investigate the selectivity of the method, the interference effect of different ions on the recovery of metal ions under the optimized conditions was evaluated.The concentration of diverse ions which resulted in an error ± 5% in determination 100 mL of 100 µg L −1 Pb(II) and Cu(II) was considered as the tolerance limit.The results summarized in Table 1 demonstrate that the method is relatively selective for determination Pb and Cu in real samples.
Sorption capacity
A batch method was used to calculate the sorption capacity.Langmuir isotherms were used to describe the sorption process at the solid-liquid interface which is represented by the following equation: where C (mg L −1 ) is the equilibrium concentration, q (mg g −1 ) is the of adsorbed per unit mass of adsorbent at equilibrium, q m (mg g −1 ) is the maximum amount of sorption in monolayered sorption systems and k (L mg −1 ) is the Langmuir constant, which can be considered as a measure of sorption energy.
A linear plot of C/q against C was applied to obtain the values of q m and k from the slope and intercept of the plot.
In order to calculate the sorption capacity, 3.0-10.0mg L −1 of metal ions were added to 10.0 mg of adsorbent.The results indicated that the sorption capacity of MNPs-ephedrine for Pb(II) and Cu(II) are 9.4 and 0.6 mg g −1 , respectively.
Analytical figures of merit
The analytical performance characteristics of the method are shown in Table 2.Under the optimized conditions, a linear calibration graph was obtained for determination Pb(II) and Cu(II).The linear ranges were found to be 15.0-500 µg L −1 and 18.0-500 µg L −1 for lead and copper, respectively.The regression equations for the lines were A = 0.0017 C − 0.0003 with r = 0.9992 for lead, A = 0.0016 C + 0.0325 with r = 0.9982 for copper, where A is absorbance and C is concentration of lead and copper ions in µg L −1 .The limit of detection and limit of quantification were defined as 3S b /m and 10S b /m where m is the slope of calibration graph and S b is the standard deviation of ten blank determinations were 4.3 and 14.5 µg L −1 for lead and, 5.0 and 16.7 µg L −1 for copper, respectively.The relative standard deviations for eight replicate determinations of 80 and 200 µg L −1 of Pb(II) were 2.9 and 1.4% and for 80 and 200 µg L −1 of Cu(II) were 3.5 and 1.9%, respectively.The reusability of the sorbent in several successive sorption and desorption processes was studied.The obtained results showed that the sorbent could be reused two times without any considerable loss in its sorption efficiency.
Application
The present method was applied to determination of Pb and Cu in water samples.The results are shown in Table 3.To evaluate the accuracy of the results, different The interference of Fe 3+ up to 10-fold was overcome by the addition of 1.0 mL of 1000 mg L −1 of F − solution.
Conclusions
In this magnetic Fe 3 O 4 modified with ephedrine has been applied for the first time for preconcentration of lead and copper in water samples.It was found that this procedure was relatively selective, simple, fast, low cost and eco-friendly and with a good preconcentration factor, wide linear dynamic range.A comparison with some of the previously reported works is also given in Table 4.
Table 1 .
Effect of diverse ions on the determination of 100 µg L −1 of Pb(II) and Cu(II)
Table 2 .
Analytical characteristics of proposed method at the optimum conditions
Table 3 .
Determination of Pb and Cu in water samples a Not detected
Table 4 .
Comparison of the proposed method with other to determination the metal ionsThis work amounts of investigated metal ions were spiked to the real sample by standard addition.The recovery values for added concentration of analyte ions were quantitative. | 4,652.2 | 2014-11-01T00:00:00.000 | [
"Chemistry"
] |
Reconstruction of the Lateral Collateral Ligament Using a Suture Tape Anchor for Iatrogenic Hallux Varus
Iatrogenic hallux varus is a difficult complication of hallux valgus surgery. Although tendon transfer combined with bony correction is performed for hallux varus, tendon transfer has several disadvantages, such as the complicated nature of the procedure and the donor site morbidity. We describe the case of a 70-year-old woman with iatrogenic hallux varus treated by lateral collateral ligament (LCL) reconstruction using a suture tape anchor with bony correction. Tarsometatarsal joint arthrodesis was performed to correct the narrow intermetatarsal angle (IMA), and the varus deformity of the great toe at the metatarsophalangeal joint was corrected by anatomical reconstruction of the LCL using the suture tape anchor. One year postoperatively, the Japanese Society for Surgery of the Foot Hallux Metatarsophalangeal-Interphalangeal Scale had improved from 37 to 90 points. Radiography confirmed that the hallux valgus angle had been corrected from -24° to 4° and the IMA from 0° to 8°. Reconstruction of the LCL using suture tape anchor is an easy procedure for iatrogenic hallux varus which can achieve good stabilization.
Introduction
Although the reported incidence of iatrogenic hallux varus is 2% to 15.4% after hallux valgus surgery, this complication remains challenging to treat because of several surgical factors including loss of osseous support, overcorrection of the proximal articular set angle or intermetatarsal angle (IMA), and muscular imbalance of the proximal phalangeal base [1][2][3][4][5]. Arthrodesis of the first metatarsophalangeal (MTP) joint for iatrogenic hallux varus has historically provided analgesia with no recurrence of severe instability [1]. However, motion at the first MTP joint is essential for activities requiring push-off power such as running and jumping. Therefore, other options have recently been introduced, including reverse osteotomies and tendon transfer procedures [1]. However, tendon transfer has several problems such as high demand techniques to adjust tendon and uncer-tain long-term clinical outcomes due to the loss of the muscle tension [1]. Osteotomies to correct IMA should be considered the healthy osteotomy site to avoid the nonunion in the case which the prior osteotomy was performed [1]. Therefore, an easier and more convenient technique to preserve the joint function of the first MTP is needed. Joint preservation surgery in the first MTP joint should be performed unless there is a severe osteoarthritis (OA) change in the joint, especially in iatrogenic case [6]. The two most important outcomes in joint preservation of the first MTP joint after hallux valgus surgery are a balanced MTP joint and a corrected IMA. The IMA is corrected by osteotomy but reobtaining balancing in the first MTP is challenging. Recently, suture tape devices have been developed and widely used to stabilize unstable joints [7][8][9]. We describe the case of a patient whose iatrogenic hallux varus was treated with tarsometatarsal (TMT) joint arthrodesis and lateral collateral ligament (LCL) reconstruction of the first MTP joint using a suture tape anchor device, with satisfactory clinical outcomes.
Case
A 70-year-old woman underwent a modified Mann's procedure for her right hallux valgus at a previous hospital three years ago. Postoperatively, the hallux varus deformity occurred with dorsal dislocation of the second phalanx at the MTP joint. She complained of severe pain at the metatarsal head of the plantar side of the second toe. Therefore, she was referred to our hospital for further treatment.
Her right foot exhibited a hallux varus deformity, and the second and third toes were dislocated dorsally (Figure 1(a)). Skin erosion at the dorsal aspect of the proximal interphalangeal joint of the second and third toes was observed. The range of motion (ROM) of the first MTP joint was 40°in dorsiflexion and 10°in plantarflexion. There was a tender callosity at the plantar aspect of the second metatarsal head. On plain radiographs, the hallux valgus angle was -24°, and the IMA was 0° (Figure 1(b)). The joint space at the first MTP joint was maintained.
Three-dimensional computed tomography (3DCT) revealed the hallux varus deformity and dorsal dislocation of the second toe (Figure 2(a)). Osteophyte formation was observed at the proximal end of the phalanx. On the coronal image, the surface area of the medial side of the articular sur-face at the first metatarsal head was decreased by the resection of the bony prominence ( Figure 2(b)). Because her symptoms such as pain and gait disturbance had worsened, surgery was performed.
The patient was placed in a supine position under saphenous and sciatic nerve block, and a pneumotourniquet was applied to a lower leg. To correct the hallux varus deformity after the proximal osteotomy, we planned three steps, namely, correction of the IMA by corrective arthrodesis of the TMT joint, a shortening osteotomy of the second metatarsal bone, and reconstruction of the LCL of the first MTP joint ( Figure 3).
First, the fixation plate and screws inserted at the proximal side of the metatarsal bone during the modified Mann procedure were removed using the previous skin incision. The skin incision was extended 1.5 cm proximally, and the first TMT joint was exposed to facilitate arthrodesis. The metatarsal base was resected in a wedge shape, with the medial side 4 mm from the TMT joint. The medial cuneiform bone was cut at a thickness of approximately 1.5 mm parallel to the line of the articular surface to prepare for the arthrodesis to correct the IMA and pronation deformity. The TMT joint was tentatively fixed with a 1.5 mm Kirschner wire, and the appropriate IMA aiming at 8°was confirmed under fluoroscopy. Then, the TMT joint was fixed using a cannulated cancellous screw (Asnis microscrew 3.0 mm, Stryker, Mahwah, NJ, USA) to apply a compression force, and additional fixation was achieved with a locking Second, a 3 cm incision was made on the medial side of the first MTP joint and a 3 cm incision at the first dorsal intermetatarsal area using the previous skin incision. To correct of the second toe at the MTP joint, a shortening osteotomy of the second metatarsal bone at the distal side was performed. The metatarsal bone was resected obliquely in a sagittal plane to a length of 3 mm and fixed with a cannulated cancellous screw (Asnis microscrew 2.0 mm, Stryker, Mahwah, NJ, USA). Then, reduction of the second toe was performed, and the MTP joint was temporarily fixed with a 1.2 mm Kirschner wire (K-wire).
Finally, the LCL reconstruction in the first MTP joint was performed using suture tape anchor (DX Suture Anchor with 1.3 mm Fiberwire Suture Tape, Arthrex, Inc., Naples, FL.). The adhesion of the abductor hallucis muscle was released through the lateral incision, and the joint capsule on the lateral side was dissected in the U shape including the loosened LCL. The footprint of the LCL at the proximal phalanx was exposed, and an anchor with the suture tape was inserted. Then, two bone tunnels were created at the 3 Case Reports in Orthopedics attachment of the proximal end of the LCL in the metatarsal head using a 1.2 mm K-wire, and the suture tape was pulled out from lateral to medial by suture relay. With the holding of the great toe at the appropriate alignment aiming at 5°of the hallux valgus angle, poly-L-lactic acid (PLLA) pins (SuperFixsorb, Depuy, Rayhnam, MA) were inserted into the bone tunnels to fix the suture tape while it was manually pulled to the maximum tension. Prior to the fixation of the suture tape, the first MTP joint motion was confirmed to determine the degree of the tension. Then, suture tapes were sutured to the medial capsule. Keeping the good alignment of the great toe, the first MTP joint was temporarily fixed with a 1.2 mm K-wire. The U-shape lateral capsule was sutured with 2-0 thread at the lateral side of the metatarsal head while maintaining tightness. Postoperatively, a short leg cast was applied for three weeks. Full-weight bearing was permitted from two weeks using a short leg cast with a rubber walking heel. Three weeks postoperatively, the cast was removed, and the patient was administered an insole and allowed to walk without a propulsive toe off. Eight weeks postoperatively, the patient was allowed to be active without any particular limitation. One year postoperatively, there were no residual complaints of pain. The hallux valgus deformity had improved to the extent that she could wear commercial shoes without difficulty (Figure 4(a)). The ROM of the first MTP joint was 65°in dorsiflexion and 15°in plantarflexion (Figures 4(b) and 4(c)). The Japanese Society for Surgery of the Foot (JSSF) Hallux Metatarsophalangeal-Interphalangeal Scale had improved from 37 to 90 points [10,11]. On the patient-reported SAFE-Q score, all subscale scores improved from pre-to postoperatively, as follows: pain and pain-related questions, from 18.9 to 71.7 points; physical functioning and daily living, from 34.1 to 59.1 points; social functioning, from 54.2 to 75 points; shoerelated questions, from 0 to 58.3 points; general health and well-being, from 25 to 60 points; and sports, from 0 to 1.1 points [12,13]. Radiographic findings showed that the hallux valgus angle had corrected from -24°to 4°, and the IMA had corrected from 0°to 8°( Figure 5). Progression of OA of the first MTP joint was not observed. Case Reports in Orthopedics
Discussion
Several studies related to the difficulty of the surgical treatment of the iatrogenic hallux varus have been published. However, we successfully treated it using a suture tape anchor with an easy technique. One year postoperatively, improvement of ROM of the first MTP joint was obtained with good alignment, suggesting that an LCL reconstruction using suture tape could achieve physiological and anatomical repair of the MTP joint. The tension of the LCL appears appropriate because progression of OA of the first MTP joint was not observed, and there was good ROM of the great toe. Various surgical treatments have been performed for iatrogenic hallux varus. In cases of OA with pain and stiffness in the MTP joint, arthrodesis is recognized as the most appropriate treatment [14]. Indeed, MTP arthrodesis is reliable because there is no risk of recurrence or improvement of the pain at the MTP joint. However, in cases without OA change and with maintenance of the ROM, joint-
Case Reports in Orthopedics
preserving surgery should be performed. The first-line procedure is medial capsule release in the retracted part of the first MTP joint including the abductor hallucis tendon, which provides the deforming force toward hallux varus [15]. Subsequently, fibrosis is released in the first intermetatarsal space to restore the intermetatarsal divergence and valgus phalanx positioning. These procedures are convenient, but insufficient to maintain reduction, and usually require additional techniques such as tendon transfer and bony correction [16]. Tendon transfer with the abductor hallucis or extensor hallucis longus tendon aims to compensate for the LCL and uses either a dynamic technique with the muscle body or a static technique without. Although this may achieve stabilization of the lateral side of the MTP, there are several disadvantages. In dynamic transfer, tension is difficult to adjust, and there is a risk of long-term tension loss, while adjusting the tendon in static transfer is simplified and stabilizes over time. However, incorrect positioning of the transfer in either method can make the joint nonfunctional, and the procedures are complicated. Above all, donor site morbidity, including functional loss and surgical invasion, should be considered. To avoid these complications, we performed the LCL reconstruction using suture tape.
Recently, joint stabilization techniques using artificial ligaments have been developed to achieve the reconstruction of the collateral ligament in various joint [7][8][9]. Cho et al. demonstrated the stabilization of the first MTP joint in a patient with chronic varus instability using suture tape [17]. In their procedure, 2.7 mm bone tunnels were created at the proximal phalanx and the metatarsal bone under fluoroscopy, whereafter suture tape was fixed using the 3.0 mm biotenodesis screws with congruent reduction. Other reports have shown the utility of a suture button device (mini Tigh-tRope R , Arthrex) for medial instability of the first MTP joint and traumatic hallux varus deformity [18][19][20]. Using these devices will enable less invasive joint stabilization surgery with a lower recurrence rate. We believe that adding these devices to the treatment of iatrogenic hallux varus deformity will improve the clinical outcome even if the pathology is complicated. In addition, the anchor system with suture tape allowed us to insert the suture tape precisely at the LCL attachment of the proximal phalanx, which is expected to facilitate more anatomical motion of the great toe.
In addition to the LCL reconstruction using suture tape, TMT joint fusion to correct the IMA was performed because a closed IMA is one of the factors for iatrogenic hallux varus [1]. Generally, IMA < 6°after correction of the hallux valgus should be revised to reopen the IMA. We aimed at an IMA of over 6°, but excessive correction of the IMA leads to the recurrence of hallux valgus. Therefore, we aimed at an IMA of 8°by adjusting the osteotomy at the TMT joint. This angle is a physiological IMA; however, it risks recurrence of the hallux varus due to the slight bony defect at the medial side of the metatarsal head. LCL reconstruction using suture tape provided a strong constraint in the varus direction, and we obtained good clinical outcomes without recurrence. We believe that ligament reconstruction using suture tape could be a useful procedure for chronic varus instability of the first MTP joint.
Conclusion
In conclusion, we described the successful treatment of an iatrogenic hallux varus using LCL reconstruction with suture tape anchor and TMT joint fusion. Reconstruction of the LCL with a suture tape anchor is an effective technique for iatrogenic hallux varus with little progression of arthritis, and this technique can be considered for the treatment of iatrogenic hallux varus.
Data Availability
Data supporting the results of the manuscript are included within the manuscript.
Conflicts of Interest
No authors have conflicts of interest to declare. | 3,227 | 2021-10-26T00:00:00.000 | [
"Medicine",
"Engineering"
] |
Annals of Clinical Microbiology and
Background: Antimicrobial therapy is considered an important component in the medical management of most patients with acute exacerbation of chronic bronchitis (AECB). The three predominant bacterial species isolated are nontypeable Haemophilus influenzae, Moraxella catarrhalis, and Streptococcus pneumoniae. Staphylococcus aureus is also frequently isolated while atypical bacteria are thought to cause up to 10% of exacerbations. Antibacterial resistance is increasing worldwide and little surveillance data exist concerning pathogens isolated from patients with AECB.
Introduction
The World Health Organization (WHO) estimates that chronic obstructive pulmonary disease (COPD) is the fourth leading cause of death worldwide. In the year 2000, it was estimated that 2.74 million people died from COPD worldwide [1]. COPD is defined by the presence of irreversible or partially irreversible airway obstruction in patients with chronic bronchitis or emphysema [2,3]. The disease is characterized by recurrent (1-4 per year) acute exacerbations of chronic bronchitis (AECB), defined by a subjective increase from baseline of one or more symptoms including shortness of breath, cough, sputum production, and sputum purulence [4]. The precipitating factors for AECB have been extensively researched and determined to be heterogeneous with complex aetiology [5][6][7][8][9][10].
Results from a number of placebo-controlled clinical investigations have demonstrated that antibacterial agents are of significant clinical benefit in the treatment of AECB, particularly for those patients with at least two of the three cardinal symptoms of AECB (worsening dyspnoea, increased sputum volume, and increased sputum purulence) and/or severe airway obstruction [11][12][13]. Other clinical trials measuring non-traditional endpoints have shown that antibiotic therapy reduces the time to symptom resolution and has long-term benefits including greater intervals between episodes of exacerbation [14,15]. Consequently, antibiotic therapy is considered an important component in the medical management of patients with AECB.
Amoxycillin, ampicillin, sulfamethoxazole-trimethoprim (trimethoprim-sulphamethoxazole), tetracyclines, and erythromycin are considered first-line antimicrobial therapy for AECB [17]. The clinical utility of these agents is, however, being hampered by the increasing global spread of pathogens with resistance to one or more of these agents. Up to 40% of H. influenzae isolates and more than 90% of M. catarrhalis isolates produce β-lactamase and this limits the value of penicillins and some other βlactams [18]. Furthermore, resistance to penicillin and macrolides has spread rapidly among isolates of S. pneumoniae [19]. Other agents used include extended spectrum cephalosporins, amoxycillin/clavulanate, azithromycin, clarithromycin, and levofloxacin.
Telithromycin is the first ketolide available for clinical use. Derivatives of erythromycin-A, the ketolides, like the macrolides, exert their antimicrobial action by binding to the bacterial ribosome. Although both macrolides and ketolides bind strongly to a region of domain V in the 23S rRNA of the ribosome, telithromycin has additional strong binding to a region in domain II to which the macrolides bind weakly [20]. Ketolides are also poor substrates for the efflux pump (mefA) responsible for macrolide resistance in S. pneumoniae [21]. Consequently, telithromycin has been found to have potent activity against macrolide resistant S. pneumoniae with methylase, efflux or ribosomal mutations as the mechanisms of resistance [22,23].
There is a need for alterative therapeutic options for the treatment of AECB and surveillance data are needed to help determine the suitability of new agents. The PRO-TEKT (Prospective Resistant Organism Tracking and Epidemiology for the Ketolide Telithromycin) study is an international, longitudinal, antibacterial resistance surveillance study, which was initiated in 1999 to monitor the spread of resistance among respiratory tract pathogens worldwide. Here we analyze the in vitro antimicrobial activity of bacterial isolates obtained from patients clinically diagnosed with AECB in 3 consecutive years of the PROTEKT study. Using these data, and previously published clinical data, the potential role of telithromycin in the treatment of AECB will be discussed.
Patients and bacterial isolates
Details of the study design, including the selection of patients and the methodology for the identification of isolates and their storage in the PROTEKT study has been described previously [24]. Isolates in this study were obtained from patients diagnosed with AECB from in 85 centres in 29 countries (Table 1). To be included in this analysis, an isolate was deemed pathogenic in AECB by clinical and laboratory findings. Isolates were only acceptable if the patient was ≥ 30 years old and the specimen was obtained from blood, bronchoalveolar lavage (BAL), or sputum. Isolates from patients diagnosed with AECB obtained from other sites (e.g., ear, throat, nasopharynx) and isolates obtained from patients <30 years of age were excluded from this analysis because AECB is more likely to be present in patients ≥ 30 years of age and the responsible bacterial pathogen is more likely to be correctly isolated from the blood, BAL, or sputum. In
Minimum inhibitory concentrations (MIC) of each antibacterial were determined using the National Committee for Clinical and Laboratory Standards (NCCLS) broth microdilution methodology and lyophilised microtitre plates (Sensititre, Trek Diagnostics) at a central laboratory (GR Micro Ltd., London, UK) [26]. NCCLS breakpoints [25,26] were used to interpret the MIC data and to determine susceptibility status. The NCCLS breakpoints for telithromycin for S. pneumoniae and for S. aureus are ≤ 1 mg/ l is susceptible, 2 mg/l is intermediate, and ≥ 4 mg/l is resistant, and for H. influenzae ≤ 4 mg/l is susceptible, 8 mg/l is intermediate, and ≥ 16 mg/l is resistant [27].
Results
A total of 3043 bacterial pathogens were isolated from patients in 29 countries around the world, with by far the largest number of specimens (1841, 60.5%) coming from Europe ( One hundred and three (9.6%) S. pneumoniae isolates (from 51 and 53 patients in the 30-64 and >64 year old age groups respectively)) were resistant to both penicillin (MIC ≥ 2 mg/L) and erythromycin (MIC ≥ 1 mg/L) and this was reflected in resistance to amoxycillin, cefuroxime, clarithromycin and azithromycin also (Table 4). These isolates were found in 35 centres in 16 countries. Sixty of these resistant isolates were also resistant to both trimethoprim-sulphamethoxazole and tetracycline. Both telithromycin and levofloxacin had good activity against these
Discussion
The primary cause of COPD is exposure to tobacco smoke, the major risk factor being cigarette smoking. The demography of the disease in this study and others reflects this, as the majority of patients in this analysis were male and half were elderly (>64 yrs of age) (2). S. pneumoniae is most frequently isolated in the least severe cases of AECB, whereas H. influenzae is more commonly isolated from moderate to severe cases, with P. aeruginosa occurring in severe hospitalised cases [28]. Telithromycin does not have good activity against Pseudomonas spp. (GR Micro Limited, data on file, internal report number 141-02-99) and hence may not be an appropriate empirical therapeutic option for AECB patients with severe underlying disease who are hospitalized for an acute exacerbation.
Whether the isolation of a pathogen during AECB represents an infection responsible for the exacerbation has been debated for many years [29][30][31]. Bacteria have been isolated almost as frequently from patients with stable COPD as those with an AECB, and clinical trials of antibiotic therapy in AECB show contradictory and sometimes unconvincing results [30]. The presence of bacteria in the lower airways is, however, regarded as abnormal since these airways are sterile in healthy adults, and it has been hypothesized that the presence of bacteria in stable COPD represents a low-grade smouldering infection. In addition, a recent study has shown that infection with different strains of pathogens that are new to the patient is associated with development of exacerbation [32,33].
Amoxycillin-clavulanate, azithromycin, and levofloxacin have been shown to be effective in the treatment of AECB, however, there is concern regarding their long-term usefulness, because of the development of resistance to these agents among the causative pathogens [34,35]. Telithromycin has a more focused spectrum of activity than the βlactams and the fluoroquinolones; it is specifically targeted against pathogens causing community-acquired respiratory disease, including those most commonly associated with AECB. In addition, it is active against penicillin-and macrolide-resistant strains of S. pneumoniae and hence offers a viable potential option for the empiric treatment of AECB in non-hospitalised patients [36].
The data in this study demonstrate that telithromycin has high in vitro activity against the commonest bacterial pathogens causing AECB. These data also show that telithromycin has the highest overall activity against bacterial isolates from patients with AECB, regardless of species. Almost 10% of S. pneumoniae isolated were resistant to penicillin, macrolides, and at least one of the other antibiotics tested, with only telithromycin and levofloxacin retaining high activity against these isolates (99.0% and 98.1%, respectively). The validity of this finding is strengthened as the isolates were obtained from a large number of patients over a wide geographical distribution.
Although atypical pathogens were not examined in the PROTEKT study, telithromycin has been shown to have superior activity in vitro against Chlamydophila pneumoniae to the other macrolides with the exception of clarithromycin and has similar activity to the fluoroquinolones [37]. In guinea pig models, telithromycin had better activity than erythromycin against Legionella pneumophila infections [38]. In vitro, the activity of telithromycin against L. pneumophila was similar to levofloxacin but better than erythromycin [38]. β-lactams and cephalosporins have no activity against Mycoplasma pneumoniae as this species lacks a typical bacterial cell wall, the site of activity for these drugs. Telithromycin has been found to have higher activity than doxycycline and levofloxacin against M. pneumoniae [39]. As the atypical pathogens can represent up to 10% of infections associated with AECB, the efficacy of telithromycin against these pathogens could be a consideration in the selection of empiric therapy for AECB.
Telithromycin has been shown to penetrate into respiratory tissues well [40]. The concentration of telithromycin in alveolar macrophages and epithelial lining fluid exceeds that of plasma markedly and remains at therapeutic levels for 24 hours after dosing. Bactericidal levels are also maintained in plasma. A good post-antibiotic effect has also been observed [41]. Telithromycin causes only moderate ecological disturbance to oral and intestinal flora comparable to that associated with clarithromycin and it does not significantly increase the development of resistance in the normal flora, although the MIC of oral streptococci can be slightly raised [42].
Telithromycin can be administered once a day for AECB. Clinical studies have demonstrated that 800 mg administered once daily for 5 days was as effective and well tolerated as a 10-day course of amoxycillin/clavulanate (500/ 125 mg 3 times daily for 10 days), cefuroxime axetil (500 mg twice daily for 10 days) or clarithromycin (500 mg twice daily for 10 days) [43]. Other clinical studies have also confirmed the safety and tolerability of telithromycin 800 mg administered for 5 -10 days [44]. Once a day dosing schedules and shorter courses may promote patient adherence to therapy, and this in turn could delay the development of resistance.
Although this study provides valuable information on the overall antimicrobial profile of bacteria causing AECB, care should be taken when interpreting data related to specific demographics. The prevalence of species could not be calculated in this study as a major limitation, inherent to most surveillance studies, is the requirement for collecting centres to fulfil a specified quota of isolates over a defined time period (1 year). If, for instance, a centre managed to fulfil the quota for S. pneumoniae isolates from patients with community-acquired pneumonia, it could then only send H. influenzae from patients with AECB to fulfil the quota for this organism. In addition, atypical pathogens were not sampled and they can represent up to 10% of the causative pathogens [28].
In summary, the data presented here demonstrate that telithromycin has good in vitro activity against H. influenzae, S. pneumoniae, and M. catarrhalis, respiratory pathogens commonly isolated in AECB. It is as active as or more active than agents that are currently used in this clinical setting. Additionally, although not shown here, telithromycin has better in vitro activity against atypical pathogens than other agents; an important advantage in this clinical setting as these pathogens may represent 10% of AECB associated infections.
The development of resistance will always be a threat to the usefulness of antibacterial compounds, however surveillance studies such as PROTEKT allow the rapid detection and characterization of resistance mechanisms and highlight the need for and examine the in vitro efficacy of newer antibacterial agents. Providing careful surveillance for the development of resistance is maintained, telithromycin currently offers a useful agent in the treatment of AECB. | 2,854.6 | 2005-01-01T00:00:00.000 | [
"Medicine",
"Biology"
] |
Effect on Compton Scattering Spectra by Hermite–Gaussian Light
: In this study, we measured the Compton scattering spectra of Al, Ag and Au metals changing the harmonic order of X-rays from an undulator. The width of the Compton scattered X-ray spectrum changed depending on the harmonic order of X-rays. This indicates that Compton scattering spectra shape reflects a momentum perpendicular to the traveling direction in Hermite– Gaussian (HG) light.
Introduction
Since the establishment of electromagnetism at the end of the 19th century, it has been known that light generally has angular momentum, and that the spin angular momentum (helicity) of light corresponds to the polarized state. Allen pointed out that laser light with a Laguerre-Gaussian (LG) mode, which possesses a helicoidal wave-front surface around the propagation axis, has an orbital angular momentum [1]. Furthermore, it has been discovered that a phenomenon in which the light with an LG mode gives a rotational motion to an object [2]. This LG light is associated with the orbital angular momentum of light. After this discovery, it has been considered that the angular momentum of light consists of the spin angular momentum and the orbital angular momentum.
The orbital angular momentum of light has already been widely studied in laser physics. The orbital angular momentum of the LG wave function represented in the cylindrical coordinate system is a quantum number of this light.
The Hermite-Gaussian (HG) light is associated with the Cartesian coordinate system, and is in a conjugate relationship with the LG light as a linear combination. Both HG light and LG light are characterized by having momentums perpendicular to their traveling direction [1].
The wave function of HG light in the Cartesian coordinate system shows that the "number of nodes" of light is a good quantum number, and the momentum perpendicular to the traveling direction of light is quantized. The LG light, which is often called optical vortex, has been studied in laser physics and in synchrotron radiation research, among others. Theoretical and experimental studies have reported that X-rays generated as synchrotron radiation would generate an optical vortex [3][4][5]. Sasaki et al. reported that the synchrotron radiation with a circularly polarized component from the undulator has an optical vortex represented by LG light, and the linearly polarized light from the undulator has a "node" represented by HG light [6].
Recently, Nairat et al. showed that the initial angular momentum carried by the incident photon beam transfers to the scattered beam in Compton scattering process from the energy-momentum conservation laws [7]. Furthermore, Maruyama et al. reported theoretical calculations of a Compton scattered X-ray spectrum with LG and HG light, and pointed out that the Compton scattered X-ray spectrum depends on the number of nodes of HG light, which reflects the momentum perpendicular to the traveling direction of the quantized light [8,9].
In this study, we measure Compton scattered X-ray spectra to detect the momentum perpendicular to the traveling direction, which reflect the number of nodes of HG light of an undulator.
Compton Scattering between X-rays and Electrons
If we consider Compton scattering between X-rays and electrons, we get the following equations from the momentum conservation and energy conservation.
where p denotes an electron momentum in a material. p' denotes a momentum of a recoiled electron.hk andhω denote an incident photon momentum and energy.hk andhω denote a scattered photon momentum and energy. From Equations (1) and (2), we get where K = k − k . If we consider momenta of electrons, p 1 = (p 1x , p 1y , p 1z ) and p 2 = (p 2x , p 2y , p 2z ), in a material, we get From Equation (5), we can understand that the full width of half maximum (FWHM) of the observed Compton scattered X-ray spectrum is therefore proportional to the FWHM of the electron momentum distribution of the material.
If we consider that the HG light has the momentum perpendicular to the traveling direction of the quantized light, ∆k,hk in Equation (1) can be replaced byh (k + ∆k). Therefore Equation (5) for the HG light can be expressed by For the scattering angle of 90 • with the scattering plane perpendicular to the linear polarization vector, as discussed in Materials and Methods later, ∆k is perpendicular to K. If we choose the z-axis along to K in momentum space, Equation (7) can be expressed by For an isotropic material such as a polycrystal metal, Equation (8) becomes Because we can regard as |K| ≈ √ 2|k| for the scattering angle of 90 • , we can estimate the momentum perpendicular to the traveling direction in the HG light from the analysis of a full width at half maximum (FWHM) of a Compton scattered X-ray spectrum.
Materials and Methods
In this paper, we detect the spectral change of Compton scattered X-rays due to the momentum perpendicular to the traveling direction in the HG light. Measurements were performed on SPring-8 BL37XU [10] with linearly polarized X-rays emitted from the undulator with a polarization vector in the horizontal plane. Figure 1 shows observed X-ray intensities, which are monochromatized to 114 keV by a Si 333-Si 511 double crystal monochromator, with changing undulator gaps as an example. By changing the undulator gap, a HG mode with a given harmonics number was chosen. In the present experiment, we used X-rays with 18, 19, 20, or 21st harmonics number (order of light) of the undulator. The emitted X-rays were monochromatized to 100 keV (energy width ∆E/E = 2 × 10 −5 ) by a Si 333-Si 511 double crystal monochromator, and irradiated on the sample. The center part of the X-ray was selected by a slit (0.5 mm × 0.5 mm) at the front end (30 m from the light source point). Figure 2 shows the experimental layout. Compton scattered X-rays were detected by a Ge solid state detector with a scattering angle of 90 • . The scattering plane was set perpendicular to the linear polarization vector in the present experiment (90 • , geometry in Figure 2). Polycrystalline metals of Al, Ag and Au were measured as samples, because their electric state is described as a nearly free electron model and regarded as the isotropic electron momentum distribution system. The Compton scattering X-ray spectra were measured for Ag and Al with an 18, 19, 20, or 21st harmonic order, and Au with an 18 or 21st harmonic order of undulator. The measurements were performed at room temperature. Figure 3a shows the Compton scattering spectra of Ag and Al samples for the incident X-ray with an 18, 19, 20, and 21st harmonic order of undulator. Although the spectral shapes depend on the sample clearly, dependences on the harmonic order of undulator are not clear, as shown in Figure 3a. Therefore, we performed analysis for the Compton scattered X-ray spectra by superimposing two Gaussian fittings. Figure 3b shows an example of fitting results for Al samples of the 21st harmonic order of the undulator. We call the broad FWHM, which corresponds to a momentum distribution of the core electrons, Width_1 (broad), and the narrower FWHM, which corresponds to a momentum distribution the valence electrons, Width_2 (narrow) [11]. Figure 4a shows the harmonic order dependence of Width_1 (broad) and Width_2 (narrow) for Al, Ag and Au. Here, we focus on Width_1 (broad) of Al and Width_2 (narrow) of Al, Ag and Au to avoid BG contributions such as lead fluorescence. The dependences on the harmonic order of undulator for Width_1 (broad) and Width_2 (narrow) are small. However, we seem to catch oscillations of Width_1 (broad) and Width_2 (narrow), as shown in Figure 4a. In order to highlight the oscillations of the half-width, the half-widths are normalized by an average value among the four half-widths of harmonic orders of the undulator, as the following.
Results and Discussions
where n = 1 or 2. Width_n_Ave denotes the average value of Width_n among the four half widths of harmonic orders of the undulator. Figure 4b shows the Normalized Width_n obtained from Equation (10) for Al, Ag and Au. The oscillation behaviors of Normalized Width_n are observed in Figure 4b. The oscillation behaviors show that the normalized Width_n increases on the odd harmonic orders of the undulator, and decreases on the even harmonic orders of the undulator. This behavior may come from the difference in the node of HG light. The difference of Normalized Width_n between the odd and even order is about 0.01 (1%) on average. Since a momentum perpendicular to the traveling direction in the HG light is reflected by an FWHM of a Compton scattered X-ray spectrum, as discussed in Equation (10), the dependence on the harmonic orders of the undulator for the normalized Width_n can be ascribed to difference of the momentum perpendicular to the traveling direction in the HG light. The order of the value of momentum perpendicular to the traveling direction is estimated to be 1% of the momentum of the incident X-rays.
Conclusions
We measured Compton scattering spectra of simple metals, changing the harmonic orders of the undulator. The FWHM of Compton scattered X-ray spectrum depends on the harmonic order of the undulator. The order of the value of momentum perpendicular to the traveling direction is estimated to be 1% of the momentum of the incident X-rays.
This indicates Compton scattering spectra analysis is a candidate to study HG light in high energy X-ray regions.
Data Availability Statement:
The data presented in this study are available on request from the corresponding author. | 2,216.2 | 2021-06-08T00:00:00.000 | [
"Physics"
] |
MLP and CARP are linked to chronic PKCα signalling in dilated cardiomyopathy
MLP (muscle LIM protein)-deficient mice count among the first mouse models for dilated cardiomyopathy (DCM), yet the exact role of MLP in cardiac signalling processes is still enigmatic. Elevated PKCα signalling activity is known to be an important contributor to heart failure. Here we show that MLP directly inhibits the activity of PKCα. In end-stage DCM, PKCα is concentrated at the intercalated disc of cardiomyocytes, where it is sequestered by the adaptor protein CARP in a multiprotein complex together with PLCβ1. In mice deficient for both MLP and CARP the chronic PKCα signalling chain at the intercalated disc is broken and they remain healthy. Our results suggest that the main role of MLP in heart lies in the direct inhibition of PKCα and that chronic uninhibited PKCα activity at the intercalated disc in the absence of functional MLP leads to heart failure.
M LP (Muscle LIM protein, encoded by the Csrp3 gene) was initially discovered as a protein up-regulated in skeletal muscle following denervation 1 . It was subsequently shown to be expressed only in the heart and in adult slow-twitch skeletal muscle, and suggested to play a role during muscle differentiation 1,2 . MLP consists of two LIM domains, structural domains composed of two zinc fingers, which are well known for their role in protein-protein interactions 3,4 . Among the many binding partners that were described for MLP are the cytoskeletal proteins actin, a-actinin, N-RAP, telethonin (T-cap) and spectrin as well as the skeletal muscle transcription factors MyoD, MRF4 and myogenin [5][6][7][8][9] . Based on these interactions and the presence of a nuclear localization signal it was proposed that MLP acts as a signalling protein between the myofilaments or the cytoplasm, and the nucleus in myocytes, which is responsive to pharmacological or mechanical stimuli 10 . Pathological mutations in MLP can lead to familial hypertrophic cardiomyopathy (HCM) 11 or dilated cardiomyopathy (DCM) 8 .
Mice deficient for MLP (MLP knockouts) count among the first published models for DCM in a genetically manipulated animal 12 . They show all the anatomical and physiological hallmarks of DCM, and present with up-regulated expression levels of classical biomarkers for hypertrophy such as ANF (atrial natriuretic factor), BNP (brain natriuretic peptide) and b-myosin heavy chain as well as stress markers such as CARP (Cardiacspecific ankyrin repeat protein, CARP1/Ankrd1) 12 . While MLP knockout mice have been used by many laboratories as a mouse model to investigate DCM, the exact role that MLP plays in myocytes remains unclear. It was proposed that MLP could act as a mechanosensor at the Z-disc transmitting stress signals to the nucleus 8,10,13 . While signalling roles of MLP in the heart are well characterized, its function as a mechanosensor is less clear. Considering its molecular structure and subcellular localization, it is not obvious how a 20 kDa protein that exclusively consists of LIM domains can sense changes in mechanical force. Given that LIM domains have well-known protein-protein interaction interfaces 3,4 , it seems more likely that MLP functions in signal transmission rather than as a direct mechanosensor. Additionally, the exclusive Z-disc localization has been challenged, since several groups have reported a more widespread distribution throughout several subcellular compartments in myocytes both for endogenous and transfected MLP, including the nucleus, plasma membrane, cytoplasm, cytoskeleton and myofibrillar localizations other than the Z-disc 6,10,11,13,14 .
Over the years numerous rescue models were published using MLP knockout mice including double knockout mice with the SERCA-2A (Sarcoplasmic reticulum Ca þ þ ATPase) regulator phospholamban 15 and overexpression of calcineurin 16 . Other MLP knockout rescue reports involve inhibition of adrenergic signalling 17,18 and interference with PKCa (Protein Kinase C) signalling 19,20 .
Increased PKCa expression and activity are well established in end stage heart failure models in rodents (for recent reviews see 21,22 ). The phosphorylation substrates for PKCa range from phospholamban to sarcomeric proteins such as troponin and titin, and phosphatases such as PP2A with ensuing effects on calcium handling and contractility 19,[23][24][25] .
Intrigued by recent propositions that MLP may interact directly with PKCa 26 , and reports that MLP expression is down-regulated in failing mouse and human hearts 7,27 , we speculated that MLP may directly affect PKCa activity.
Our assays reveal that the presence of MLP inhibits autophosphorylation of PKCa as well as phosphorylation of downstream targets such as phospholamban. We demonstrate that in failing hearts PKCa is concentrated at the intercalated discs (ID) (specialized cell-cell contacts in cardiomyocytes) in a complex with the adaptor protein CARP1 and PLCb1. In double knockout mice for CARP1 and MLP (CMP1), PKCa is no longer detected at the ID and the DCM phenotype does not develop. This is accompanied by normalized PKCa phosphorylation and expression levels. We propose a signalosome complex consisting of muscle ankyrin repeat proteins (MARPs) and PKCa that is regulated by MLP and whose persistent activation may play a role in chronic stress signalling in the failing heart.
Results
MLP directly inhibits PKCa activity. To investigate the possibility that MLP can directly modulate the activity of PKCa, we carried out in vitro phosphorylation assays using recombinantly expressed protein. Autoradiography showed that increasing amounts of MLP lead to decreased PKCa autophosphorylation and a decreased phosphorylation of the PKCa substrate phospholamban (Fig. 1a, for quantification see Fig. 1b). Furthermore, we found that MLP itself is a potential substrate of PKCa (Fig. 1a) and that gene mutations associated with human HCM result in reduced phosphorylation of MLP, while DCM-causing mutations show increased phosphorylation (Fig. 1c,d). Intriguingly, MLP phosphorylation appears to be significantly increased in heart samples of DCM patients (IDCM) as compared to non-failing (NF) controls (Fig. 1e,f). A similar shift in MLP phosphorylation was observed in Ga (q) overexpressing mice ( Supplementary Fig. 1a,b), a mouse model for cardiac contractile failure 28 . A potential role of wild-type (WT) MLP in modulating PKCa signalling and activity in long-term heart failure may explain why the lack of MLP leads to DCM, and why all strategies that interfered with PKCa signalling prevented the development of a DCM phenotype in the MLP-deficient background 19,20 .
PKCa concentrates at the ID in failing hearts. While an increase in PKCa expression and activity is well established in the failing rodent heart 21 , the situation in humans is less clear 25,29,30 . Analysis of PKCa/b expression in heart samples from patients with idiopathic dilated cardiomyopathy (IDCM) by immunoblotting revealed a marked up-regulation of PKC phosphorylation at Thr638/641 in most patient samples (Fig. 2a). Confocal microscopy on immunostained heart sections from DCM patients showed a remarkable concentration of PKCa at the ID (Fig. 2b, quantification in 2c), which was not as evident in heart samples from NF controls. This supports published data that reported an increased signal for PKCa at the ID in human heart failure 29 . How does PKCa specifically target the ID? CARP1 is a stress marker that is consistently up-regulated in heart failure 12,31 and a well-known scaffold protein (for reviews see refs 32,33). Immunoblot data (Fig. 2a) showed increased expression of CARP1, but not the closely related CARP2/Ankrd2 in failing human heart samples. At the cellular level CARP1 is known to localize to the sarcomere and the nucleus, with increased nuclear presence following mechanical strain 34 . We observed no differences in nuclear versus cytoplasmic localization of CARP1 in either normal or DCM hearts ( Supplementary Fig. 1c). However, in all functionally compromised hearts that we studied (MLP knockout and human DCM samples), CARP1 translocated from the sarcomere to the ID (Fig. 2d, for quantification see Supplementary Fig. 1d,e), whereas the localization of CARP2 was less affected (Fig. 2d, for quantification see Supplementary Fig. 1e). ID association for both proteins was recently also described by Jasnic-Savovic and colleagues 35 . Consistent with a previously reported interaction between CARP1 and phospholipase-C (PLC) 36 , we demonstrated in pull-down assays that the coiled coil region of PLCb1, the main PLC isoform in hearts 37 , binds to both CARP1 and CARP2 (Fig. 2e). Protein complementation assays revealed that all CARP family members can interact with the coiled coil region of PLCb1 and that this interaction happens in an antiparallel fashion (Fig. 2f). To investigate whether this complex also occurs in cardiac tissue in situ, we carried out immunoprecipitation experiments on cardiac extracts from MLP knockout hearts. CARP1 is exclusively expressed in cardiomyocytes and cannot be detected in fibroblasts ( Supplementary Fig. 1f), therefore confirming that we specifically assayed the cardiomyocyte signalosome. Using CARP1 antibodies, we identified the presence of a signalosome complex in MLP knockout hearts containing CARP1, CARP2, PKCa and PLCb1 (Fig. 2g). In contrast, these proteins were not pulled down from WT or CARP1 knockout hearts, adding support to our hypothesis that this complex only assembles in DCM/failing hearts. Using recombinant CARP1 or CARP2, we demonstrated that it is also possible to pull-down PLCb1 and PKCa from MLP knockout hearts ( Supplementary Fig. 1g). In left ventricle extracts from a subset of end-stage DCM patients the entire complex could be pulled down (Supplementary Fig. 1h). These results indicate that CARP1 and CARP2 can sequester PKCa into a complex with PLCb1. In the failing heart CARP1 is up-regulated and relocates from the sarcomere to the ID. This relocation and the ensuing retention of PKCa at the ID appear to be the crucial step in maladaptive signalling in failing hearts.
PKC activity affects CARP and MLP expression. CARP1 is reportedly a target for many protein kinases including PKCa 38 , which we recently demonstrated through in vitro kinase assays 39 . Since CARP1 up-regulation and subcellular relocation appear to be crucial in setting up the pathological signalosome complex together with PKCa and PLC1b at the ID, we considered the possibility of an initial crosstalk between PKCa and CARP. Addition of kinase inhibitors to cultures of neonatal mouse cardiomyocytes (NMC) revealed that exposure to PKC inhibitors (bisindolylmaleimide (BI) or calphostin C (Ca)) not only reduced PKCa phosphorylation levels at Thr638/641, but also downregulated CARP1 expression levels (Fig. 3a). No effect was seen with a Map kinase kinase inhibitor (U0126; U0), again supporting the idea that the effect on CARP1 expression levels may be due to PKCa activity (Fig. 3a). The reduction in CARP1 expression following PKC inhibition could also be observed in cardiomyocytes cultured over several days as well as in vivo, when MLP knockout mice were peritoneally injected with BI ( Fig. 3b,c, Supplementary Fig. 1i). The effect was not limited to CARP1, but also extended to CARP2 expression (Fig. 3b,c), which was significantly down-regulated in hearts of BI treated mice. More intriguingly, however, was the effect we observed on MLP phosphorylation and expression levels. PKC inhibition by BI decreased MLP expression levels, and more importantly MLP phosphorylation in cardiomyocytes (Fig. 3d). These data indicate that MLP is a direct phosphorylation target of PKC, and that this phosphorylation may affect MLP activity and protein levels ( Supplementary Fig. 1k). In immunofluorescence-stained heart sections, PKC inhibition shifted CARP1 but not CARP2 localization from a more pronounced ID association to an increased sarcomeric signal (Fig. 3e, for quantification see 3f). Label-free measurements of NMC contractility demonstrated that inhibition of PKCa by BI normalized the higher beating frequencies usually observed in neonatal MLP knockout cardiomyocytes compared to WT controls (Fig. 3g,h).
These results show that elevated PKCa phosphorylation increases CARP expression levels. This is accompanied by the subcellular relocalization of both PKCa and CARP1 to the ID. The assembly of a signalosome consisting of CARP, PKCa and PLCb1 at the ID leads to a stabilization of PKCa signalling activity. Inhibition of PKCa results in down-regulation of CARP expression, changes its subcellular targeting and probably leads to the disassembly of the signalosome. In addition, PKCa inhibition Dotted lines indicate that samples were run on the same gel, but non-consecutive. (b) Immunofluorescence staining of human NF and IDCM heart sections using antibodies against PKCa (right, green). Plakoglobin and a-actinin were used as counterstains (red and blue in overlay, respectively). Scale bar ¼ 10 mm. (c) Quantification of immunofluorescence intensity (F À F 0 ) over IDs from (b) shows that PKCa levels at the ID are increased in IDCM samples compared to NF controls. Sample sizes for analysed ID were n ¼ 24 from three NF heart samples and n ¼ 46 from five IDCM hearts. (d) Immunofluorescence staining of human NF and IDCM heart sections using antibodies against CARP1 (top, green in overlay) and CARP2 (bottom; green in overlay). The sarcomeric association (arrows) of CARP1 and CARP2 decreases in IDCM in favour of increased ID localization (arrowheads; for quantification see Supplementary Fig. 1d,e). Plakoglobin and a-actinin were used as counterstains (red and blue in overlay, respectively). Scale bar ¼ 10 mm. (e) Pull-down assay of GFP-PLCb1 (phospholipase-C) with GST-CARP1, GST-CARP2 or GST indicated the coiled-coil domain within PLCb1 as minimal binding site. Ponceau stain was used to visualize GST-CARP1 and GST-CARP2 protein levels, while equal amounts of GST (not shown) were used as a control. (f) Protein complementation assay using YFP (split-fluorescent protein assay) demonstrating antiparallel association (green in overlay) between coiledcoil domains of PLCb1 tagged on the N-terminus with a YFP N-terminal fragment (YN) and CARP1 tagged on the C-terminus with YFP C-terminal fragment (YC; upper right panel). No association was seen when CARP1 was tagged on the N-terminus with YFP C-terminal fragment (upper left panel). Split-fluorescent protein assay also demonstrated antiparallel association (green in overlay) between coiled-coil domains of PLCb1 tagged on the N-terminus with a YFP N-terminal fragment (YN) and either CARP2 (lower left panel) or CARP3 (lower right panel) tagged on the C-terminus with YFP C-terminal fragment (YC). Transfection efficiency was validated with a GFP antibody (red in the overlay), and DAPI (blue in overlay) was used as counterstain. Scale bar ¼ 20 mm. (g) Co-immunoprecipitation of PLCb1, PKCa, CARP2 and CARP1 from soluble heart extracts of adult WT, MLP knockout or CARP1 knockout mice by using either CARP1 antibodies or normal rabbit IgG as control (visible in Ponceau stain).
in WT cardiomyocytes leads to a decrease in MLP phosphorylation and expression. We propose that acute signalling of this complex may be beneficial for the adaptation of the heart to stress 40 , while its chronic or pathological activation is maladaptive.
Absent DCM phenotype in MLP CARP1/CARP2 double knockout mice. If CARP1 is indeed a crucial adaptor required to recruit PKCa stably to the ID, and to trigger pathological signalling in failing hearts, then its removal should prevent the development of heart failure in MLP knockout mice. Genetic ablation of any of the MARP family members in mice (CARP1/ Ankrd1/MARP-encoded by the Ankrd1 gene; CARP2/Ankrd2/ ARPP encoded by the Ankrd2 gene; CARP3/Ankrd23/DARPencoded by the Ankrd23 gene) had no effect on heart function at baseline, even in the case of a triple knockout 41 . To test our hypothesis, we crossed MLP knockout mice with the three MARP knockout lines and examined their cardiac phenotype. Immunoblot analysis of MLP and MARP expression revealed basal expression of CARP1, but no CARP2 in control hearts ( Fig. 4a and Supplementary Fig. 2a,b). Hearts of MLP knockouts displayed increased expression as well as altered posttranslational modification of CARP1 and induced CARP2 expression (Fig. 4a, Supplementary Fig. 2a-c). CARP3 was not expressed in control hearts, and was not induced in MLP knockout hearts ( Supplementary Fig. 2a). Histological and functional analyses revealed the previously reported phenotype of MLP knockout hearts (that is, dilation and heart failure 12 ); in contrast double knockout mice for MLP and CARP1 (CMP1), and MLP and CARP2 (CMP2) had normal healthy hearts. Thus, the DCM phenotype does not develop in the absence of CARP1 or CARP2 (Fig. 4b,c, Supplementary Fig. 2d,e, Supplementary Tables 1 and 2). The hearts of double knockout mice for MLP and CARP3 (CMP3) were dilated and exhibited heart failure similar to the single MLP knockout mice (Fig. 4b,c, Supplementary Fig. 2d,e, expression levels in NMC compared to untreated controls ( À ). Dotted line indicates that samples were run on the same gel, but non-consecutive. GAPDH was used as loading control. (b) Immunoblot analysis of total cardiac samples from untreated ( À ), vehicle treated (DMSO, V) or BI treated adult MLP knockout mice (24 h after vehicle or BI-injection), blotted for phospho-PKCa (Thr638/641), PKCa, CARP1 or CARP2. GAPDH was used as a loading control. Shown are representative blots and quantification of changes for PKCa phosphorylation as well as CARP1 and CARP2 protein levels (c). Sample sizes for untreated controls and BI-treated animals were 4 and 3, respectively. P values are indicated in the figure. (d) BI treatment of NRC for 24 h leads to decreased MLP phosphorylation and protein level (see also Supplementary Fig. 1j,k). SDS samples of protein extracts from vehicle treated (V) or BI-treated NRC were run on conventional SDS PAGE (middle) and on 12% polyacrylamide gels containing 50 mM Phostag reagent (top) and immunoblotted for MLP. Phosphorylated MLP proteins (P1) migrate slower due to their interaction with the Phostag reagent, compared to unphosphorylated protein (P0). Phostag profile plot analysis of band intensities (right panel) revealed presence of three distinct P1 MLP phosphorylation bands (P1-a, P1-b and P1-c), which are putatively caused by several distinct phosphorylation sites in MLP. One of these phosphorylation sites (P1-c) is changed significantly upon BI treatment (red curve) compared to vehicle control (blue curve). Shown are normalized average band intensities and s.e. GAPDH was used as loading control (bottom). (e) Representative immunofluorescence images of adult heart sections from BI and vehicle treated MLP knockout mice stained with antibodies against CARP1 (green), a-actinin (blue) and plakoglobin (red). Scale bar ¼ 10 mm. (f) Analysis of fluorescence intensity ratios (I ID / S ) of CARP1 (left panel) and CARP2 stainings (right panel) between the ID region and the sarcomere (S) of either BI-treated or vehicle-treated 4-month-old MLP mice (see Supplementary Fig. 1d for methodology). Sample sizes (n in base of the bar; from two hearts per sample) and P values are indicated in the figure; $ denotes low expression of CARP1/CARP2 in BI treated hearts (as shown in Fig. 3b). (g,h) Label free impedance analysis of spontaneously contracting WT and MLP knockout NMC using the RTCA cardio system. Representative images of real-time beating activity of WT and MLP mouse cardiomyocytes (g) and quantification of their beating frequency (h). Effect of BI treatment on MLP cardiomyocyte beating frequencies (h, right panel). Bar graphs depict mean values and s.e. The number of independently measured wells (n in base of bar) and P values are indicated.
Supplementary Table 3). CMP3 hearts had elevated expression levels of CARP1 and CARP2 as well as increased PKCa phosphorylation, suggesting activity of the same maladaptive signalling complex that we propose for MLP knockout mice ( Supplementary Fig. 2a,f). The absence of a failing heart phenotype in the CMP1 animals was accompanied by redistribution of CARP2 to the sarcomeres (Fig. 4d, for quantification see Supplementary Fig. 2g). Most importantly PKCa no longer concentrated at the ID, while the PLCb1 localization was indistinguishable between genotypes (Fig. 4e, Supplementary Fig. 2h,i). Analysis of PKCa phosphorylation levels revealed their normalization to control levels in the CMP1 heart compared to the MLP knockout heart (Fig. 4f). This suggests that it is the concentration of PKCa at the ID that triggers its pathological activity. Previously we characterized pathological changes at the cellular level in murine and human DCM samples such as elevated expression levels of N-RAP at the ID and re-expression of the foetal M-band marker EH-myomesin 7,42,43 . These changes were no longer seen in CMP1 or CMP2, but were still evident in CMP3 hearts ( Supplementary Fig. 3a,b). However, while the lack of CARP1 or CARP2 expression prevents the development of a DCM phenotype, biomarkers for hypertrophy such as ANF, BNP and skeletal actin continued to be elevated at the mRNA level in CMP1 and CMP2 mice ( Supplementary Fig. 4a-c). Also, double knockout hearts still showed some evidence of fibrosis ( Supplementary Fig. 4d-f), although the amount was markedly reduced compared to MLP knockout mice.
In conclusion, we propose that the existence of a multiprotein complex consisting of CARP, PKCa and PLCb1 at the ID of failing hearts is crucial for the maintenance of pathological signalling activity of PKCa. MLP can directly inhibit PKCa activity, and its absence is sufficient to lead to a DCM phenotype over time. Removal of CARP leads to the dissolution of the complex, where PKCa no longer concentrates at the ID, preventing the DCM phenotype in MLP knockout mice.
Discussion
We demonstrate here the existence of a multiprotein complex composed of PKCa, PLC1b, CARP1 and CARP2 at the ID specifically in failing hearts. The retention of PKCa to this signalosome at the ID may lead to an amplification of signalling. The persistent shift in subcellular localization makes the difference between acute activation of PKCa that is beneficial for the heart to cope with transient stress, and chronic activation, which leads to detrimental downstream effects on calcium handling and contractile parameters. We propose that the upregulation in expression and relocation of CARP1 or CARP2 from a sarcomeric localization to the ID is the crucial switch between acute and chronic PKCa activation (Fig. 5). We further show that MLP can directly inhibit PKCa activity in vitro, and demonstrate that removal of CARP by genetic means prevents the formation of this maladaptive signalling complex, thereby preventing the morphological, functional and molecular phenotype of DCM in MLP knockout mice.
These results explain at a molecular level, why previous strategies that prevented the development of a DCM phenotype in the MLP knockout mouse were successful 15,19,20 . Under normal circumstances MLP reduces excessive PKCa signalling in the heart. In MLP knockout mice this does no longer happen and DCM develops. Direct interference with the maladaptive PKCa signalling pathway by genetic or pharmacological strategies thus rescues the effect of the lack of MLP. The novel role of MLP as a PKCa inhibitor could also provide an explanation why overexpression of MLP does not have drastic effects on heart function under basal conditions 44 . Removal of phospholamban, which was found to rescue the MLP knockout phenotype 15 , leads to a reduction of free cytosolic calcium known to be required for PKCa activation.
We also noted that MLP itself is a substrate for PKCa activity. Indeed, DCM patient samples that show higher PKCa phosphorylation levels correlated with increased MLP phosphorylation levels compared to NF controls. Moreover we observed that DCM-causing MLP mutants were hyper-phosphorylated by PKCa in vitro, while MLP mutants that were associated with the development of HCM were hypo-phosphorylated. Inhibition of PKCa activity in vitro leads to decreased MLP phosphorylation and expression in cardiomyocytes. The exciting finding that the MLP phosphorylation state can be modulated by PKCa, correlates with different forms of cardiomyopathy and may affect its protein levels and/or activity again underlines the important role of this protein in cardiac stress signalling. At present it is unknown whether mutations in MLP or its various posttranslational modifications have an effect on oligomerization, which determines the subcellular localization and actin bundling activity of MLP 13,14 . Phostag blots of neonatal rat cardiomyocytes (NRC) indicate the presence of several distinct MLP phosphorylation sites, one of which was significantly changed upon PKC inhibition. It will be intriguing to investigate which posttranslational modifications of MLP indeed change its cellbiological properties. We propose that these posttranslational modifications of MLP are key to its biological function, and may ultimately determine where, when and how effectively it interferes with PKCa signalling. CARP1 has for a long time been associated with mechanosignalling 34 and was shown to be up-regulated in heart failure by several studies 12,31,45 . Recently its upregulated expression was shown to correlate directly with heart failure progression in IDCM 46 . So far most studies have proposed that nuclear signalling of CARP via ERK1/2 to molecules such as p53 and GATA4 is crucial for a hypertrophic response 47 . While there are conflicting results on the effect of CARP1 on ERK1/2 activation 40,47 , it appears to be increasingly recognized that stress sensing complexes associated with the titin I-band region, containing CARP1, or FHL1/2 and RAF/MEK/ERK may be partially responsible for mediating a hypertrophic response in hearts 48 . Our results now show that in heart failure CARP1 seems to be involved in a different signalling cascade at a distinct subcellular localization, namely boosting maladaptive PKCa signalling at the ID. Mutations in CARP1 were shown to cause both HCM and DCM 49,50 and to display both gain of function and dominant negative effects on the contractile behaviour of engineered heart tissues 51 . We propose that whether mutant CARP1 leads to HCM or DCM may be correlated to differential effects on protein-protein interaction and in particular distinct activation of the two signalling pathways via ERK1/2 from the sarcomere to the nucleus (HCM) or via PKCa at the ID (DCM).
Our results demonstrate that one of the main differences between a healthy and a failing heart is the presence of PKCa at the ID. The analysis of PKCa/b expression levels in human hearts was inconclusive and it seems to be the change in subcellular targeting rather than mere expression levels that makes the crucial difference between a healthy and a failing heart. Why would active PKCa/b at the ID be so detrimental for the heart? Cell membranes are known to be the major site of PKC function (for review see ref. 52) and the signalosome complex that we describe here appears to be similar to previously described scaffold interactions that were shown to be crucial for PKC regulation 52 . Increased translocation of PKCa from the cytosolic to the membranous fraction marked the transition from pressure overload induced hypertrophy to congestive heart failure in guinea pigs 53 . In addition, constitutively active PKCa at the ID may interfere with another level of regulation in the challenged heart. b-adrenergic receptors are known to become insensitive following PKCa activation 54 and adrenergic signalling is well known to play a role in heart disease 55 . Adrenergic receptors were previously shown to be concentrated at the ID in cardiomyocytes 56 . A failure to respond to fine-tuning by adrenergic signalling as shown by the altered contractile behaviour that we observed in cultured cardiomyocytes from MLP knockout mice, may over time lead to a failing heart phenotype.
Taken together, our results highlight novel roles for MLP and CARP in pathological signalling via their interactions with PKCa. This signalosome seems to be crucial for eliciting the maximal maladaptive response in the stressed heart, and CARP may be the proposed missing link 57 . Modulation of this complex may offer new therapeutic options to prevent heart failure. The reaction mix was supplied with a combination of 1 mg ml À 1 phosphatidylserine (1,2-diacyl-sn-glycero-3phospho-L-serine) and 2 mg ml À 1 diacylglycerol (DAG; 1,2-Dioleoyl-sn-glycerol) dissolved in resuspension buffer (10 mM HEPES pH 7.4, 0.3% Triton X-100). Following incubation at 30°C for 30 min, samples were mixed with SDS-sample buffer (BioRad) and separated using SDS-PAGE. For radioactively labelled proteins, gels were coomassie-stained or fixed, dried on a BioRad gel dryer and analysed using X-ray films and autoradiography 39 . Densitometric quantification of phosphorylation levels was done by scanning X-ray films followed by measurement of band intensity using ImageJ (NIH).
Methods
Human tissues. Human specimens were obtained from the Sydney Heart Bank at the University of Sydney and signed patient consent was obtained for all samples in this tissue bank. All failing samples were collected in the heart transplant theatres at St Vincent's Hospital from patients in end-stage heart failure and were frozen in liquid nitrogen within 15-20 min of the loss of coronary artery flow (cross-clamp). Human Research Ethics Committee approval was obtained by St Vincent's Hospital (#H03/118), and by the University of Sydney Human Research Ethics Committee (#12146, #159401). Human tissue was used in accordance with the ethical guidelines of King's College London (College Research Ethical Committee 04/05-74; REC reference 12/EM/0106), under current UK law. For protein, histological and immunofluorescence analysis, samples from the left ventricular free walls of patients in end-stage heart failure with familial (FDCM) or IDCM were used. NF donor hearts, which were cardiopleged, but not required for heart transplantation, were provided by the Australian Red Cross Blood Transfusion Service. Typically these hearts were not transplanted because of tissue incompatibility.
Generation of protein complementation constructs was done through in-frame ligation of the coiled-coil region of CARP1 (amino acids 1-122), the coiled-coil region of CARP2 (amino acids 1-146), the coiled-coil region of CARP3 (amino acids 1-109), and the coiled-coil region of human phospholipase C b1 (PLCb1; amino acids 980-1140) into the protein complementation vectors (split-YFP) YN-C1, YC-C1 or YC-N1 (ref. 39). This effectively generates either N-terminally tagged fusion proteins with the N-terminal (YN) or C-terminal (YC) half of YFP, or C-terminally tagged fusion proteins with the C-terminal half of YFP.
All constructs were verified for in-frame integration and correct sequence by sequencing.
Protein expression analysis. Total protein extracts of hearts for immunoblot or biochemical analysis were generated by homogenizing ventricular samples either directly into SDS-sample buffer (BioRad) or ice-cold IP-buffer (150 mM NaCl, Tris-HCl pH 8, 1 mM DTT, 1 Â Complete Protease Inhibitor EDTA-free (Roche), 1 Â PhosSTOP (Roche), 0.2% NP-40, 0.2% SDS) by using a polytron blade homogenizer (Pro Scientific Inc). Total proteins from NMC, neonatal cardiac fibroblasts or transfected COS-1 cell cultures were extracted by washing cells with PBS at room temperature and solubilization of cells directly into SDS-sample buffer or ice-cold IP-buffer using a cell scraper. Protein extracts were immediately transferred into Eppendorf tubes and stored on ice for immediate use or snap-frozen into liquid nitrogen and stored at À 20°C or À 80°C. Protein amounts were normalized by densitometry of Coomassie stained SDS PAGE.
Normalized total protein extracts were run on uniform 15% acrylamide, 10% acrylamide, or 4-20% acrylamide gradient SDS-PAGE gels (BioRad, Invitrogen), followed by immunoblotting on nitrocellulose membranes (BioRad) using wet blot technology. The nitrocellulose membranes were stained with Ponceau Red, blocked with 5% non fat dry milk (Sainsbury's) or 3% Bovine Serum Albumin (Sigma) in Low Salt Buffer (0.9% NaCl, 9 mM Tris pH 7.4, 0.1% Tween-20) and sequential incubation with the appropriate primary and secondary antibodies with intermittent washing in Low Salt Buffer was performed. Results from the chemiluminescence reaction were visualized on Fuji medical X-ray films. For 2D gel analysis, protein samples were diluted 1:1 into IEF sample buffer (Life Technologies), and first dimension was run on IEF pH3-10 gels (Life Technologies), followed by 12% acrylamide SDS-PAGE according to the manufacturer's instructions.
Phostag gels. For analysis of MLP phosphorylation levels, 20 mg of frozen cardiac tissue samples were homogenized into 200 ml freshly prepared ice-cold TLB lysis buffer (20 mM Tris HCL pH 7.6, 138 mM NaCl, 5% glycerol, 1% Triton, 5 mM DTT, 0.5 mM Sodium ortho-vanadate, Mini Protease inhibitor tablet (no EDTA; Roche), Phos-stop tablet (Roche)). Samples were briefly sonicated, incubated on ice for 20 min, centrifuged at 4°C at 14,000 r.p.m. for 15 min, and supernatants were used for further analysis via Phostag gel analysis 58 after determination of protein concentrations. Protein samples with normalized concentrations were supplemented with SDS sample buffer, boiled for 5 min, and either run on a 12% poly-acrylamide SDS-PAGE, or a 12% poly-acrylamide SDS PAGE supplemented with 50 mM Manganese Phostag (WAKO Chemicals). After end of electrophoresis, gels were washed once in immunoblot transfer buffer supplemented with 10 mM EDTA for 10 min, followed by a wash in immunoblot transfer buffer for another 10 min, and prepared for immunoblot transfer and subsequent detection and analysis following standard procedures.
Co-immunoprecipitation and GST pull-down assays. For biochemical protein-protein interaction assays, protein extracts in IP-buffer were sonicated for 1 min at 30% output on ice (Vibracell, Sonics & Materials Inc.), followed by centrifugation at 14,000 r.p.m. (4°C) for 10 min to separate insoluble proteins. Soluble proteins were either incubated with 1 mg of primary antibody or control serum for 2 h, or incubated with 2 mg of GST or GST-fusion protein at 4°C. Following immuno-complex formation in co-immunoprecipitation assays, protein extracts were incubated with protein G-linked magnetic beads (Dynabeads; Invitrogen) for an additional hour at 4°C with agitation. For GST pull-down assays, protein extracts were incubated with glutathione-sepharose 4B resin (Pharmacia) for an additional 2 h at 4°C with agitation. After incubation, beads were washed three times with ice-cold PBS, resuspended in sample buffer (BioRad), and analysed by SDS PAGE, followed by immunoblotting on nitrocellulose membranes 39 .
Animals. All procedures were reviewed and approved by the Animal Care and Use Committee at the University of California San Diego. MLP and MARP knockout mice as well as Ga (q) overexpressing mice were previously described 12,28,41 . All animals were in a mixed sv129/black swiss background. For physiological experiments only male mice were used, samples used for immunohistology and immunoblotting were from both genders. Unless stated otherwise, animals in the age range between 4 and 8 months were used for the experiments. Oligonucleotides used for genotyping via PCR analysis of tail DNA can be found in Supplementary Table 4.
The mice were fed ad libitum with a standard diet and maintained in a temperature and light-controlled room (22°C, 14 h light/10 h dark). Treatment of MLP knockout animals with PKC inhibitor was done similar to a previously described method 20 . In short, 4-months old male MLP mice were injected subcutaneous with a single dose of either PKC inhibitor BI I-HCl (BI; sc-24004, Santa Cruz Biotechnology, 1 mg g À 1 body weight) or DMSO (vehicle). Ventricular heart tissue of mice was dissected after 24 h following the injection and snap frozen in liquid nitrogen for further analysis, or processed for immunofluorescence analysis.
The protocol for animal handling and treatment procedures was in accordance with the guidelines of the Laboratory Animal Services at the University of California, San Diego and guidelines presented in the National Research Council's (NCR) 'Guide for Care and Use of Laboratory Animals' published by the Institute for Laboratory Animal Research of the National Academy of Science, Bethesda, MD, 2011.
For echocardiography analysis 2-, 4-, 6-and 12-month-old male control, MLP, MARP and MLP-MARP double knockout mice were analysed. For protein, mRNA, histological, immunofluorescence and morphological analysis, the left ventricular free wall of hearts from 3-9-month-old adult male and female mice were used. Investigation of gross cardiac morphology by histology used whole hearts that included ventricles and atria.
Transthoracic echocardiography. Adult male mice at 2, 4, 6 and 12 months of age were analysed as previously described 41 . Briefly, mice were anesthetized with 1% isoflurane and cardiac function was measured with a Philips Sonos 5500 machine (Philips Medical Systems, Andover, MA) equipped with a 15-MHz transducer. M-mode tracings of semi-conscious mice were recorded and analysed for left ventricular posterior wall and inter-ventricular septal thickness, as well as left ventricular chamber dimensions (LVID) at both end systole and end diastole. Heart contractility, shown as fractional shortening (%FS) was calculated as previously described 41 .
All fluorescently or enzymatically linked secondary antibodies were either from DAKO, Jackson ImmunoResearch or GE Healthcare Life Sciences and used at 1:100 (fluorescent) or 1:1000 (enzymatic) dilutions.
Quantitative real-time PCR analysis (qPCR). Ventricular samples were homogenized directly into Trizol reagent (Invitrogen) to extract total RNA according to the manufacturers instructions. First strand cDNA was generated from 2 mg of total mRNA using random hexamers and Superscript II reverse transcriptase. Oligonucleotides optimized for qPCR (Supplementary Table 5) of murine ANF, BNP, skeletal actin, CARP1 and GAPDH were used in reactions employing the PerfeCTa SYBR green real-time PCR mix (Quanta BioSciences) and a CFX96 thermocycler (BioRad). Samples were normalized to GAPDH. If not stated otherwise, three biological replicates were analysed per sample group.
Treatment of NMC and NRC using BI I-HCl (BI; sc-24004, Santa Cruz Biotechnology, 10 mM), calphostin C (Ca; sc-3545, Santa Cruz Biotechnology, 5 mM) or U0126 (U0; sc-222395, Santa Cruz Biotechnology, 1 mM) was done by supplementing the maintenance medium for 24 h with inhibitors or vehicle, followed by biochemical analysis of protein extracts.
For analysis of CARP1 and MLP expression in cardiomyocyte versus cardiac fibroblasts, cardiac fibroblasts were separated during the NMC isolation procedure, and cultured for an additional 2 days in DMEM (high glucose), supplemented with 10% foetal calf serum and 1% penicillin/streptomycin medium.
COS-1 cells were cultured in DMEM with 10% foetal calf serum and transfected mixing 1 mg of plasmid DNA and 3 ml of Lipofectamine-2000 (Life Technologies Carlsbad, CA) in 300 ml DMEM and adding the mix to the cells.
Immunofluorescence and histology. Immunostaining of frozen sections on glass-slides (Colorfrost Plus, Thermo Scientific) and cells in cell-culture dishes (Nunc) was done as described previously 7,39 . In brief, immunostaining of frozen heart sections was achieved by fixing 12 mm sections (Leica CM1850 cryostat) in ice-cold acetone for 5 min at À 20°C, followed by rehydration of the tissue section with PBS at room temperature for 5 min. Cultured NMC or COS-1 cells were fixed with 4% paraformaldehyde solution in PBS for 5 min at 37°C, followed by a wash with PBS at room temperature. After permeabilization using 0.2% triton-X100 in PBS for 5 min, sections or cells were incubated with primary antibodies in gold buffer (GB; 20 mM Tris-HCl, pH 7.5, 155 mM NaCl, 2 mM ethylene glycol tetraacetic acid, 2 mM MgCl 2 , 1% bovine serum albumin; for antibody dilutions see Antibodies paragraph) supplemented with 5% normal donkey serum over night at 4°C in a humid chamber. After washing of sections or cells for three times with PBS to remove unbound primary antibodies, sections or cells were incubated with secondary antibodies in GB supplemented with 5% normal donkey serum and 2 mg ml À 1 4 0 ,6-diamidino-2-phenylindole (10 mg ml À 1 DAPI, Sigma-Aldrich; for antibody dilutions see Antibodies paragraph) at room temperature in a humid chamber. Following washing of sections or cells with PBS (three times) for 5 min at room temperature, sections or cells were embedded in fluorescent mounting medium (DAKO) and mounted with coverslips.
For hematoxylin-eosin staining, 20 mm frozen sections were fixed for 5 min at room temperature using a 4% paraformaldehyde solution in PBS. Following fixation, sections were washed in water, incubated in Weigert's Hematoxylin solution (HT1079, Sigma-Aldrich) for 5 min, briefly rinsed with water and counterstained with Eosin-Y solution (Richard Allan Sci.) for 2 min. After dehydration of sections using increasing percentages of alcohol (50% for 2 min, 75% for 2 min, 95% for 2 min, twice 100% for 5 min), samples were washed twice in Histo-Clear (National Diagnostics) for 10 min and mounted using Permount (Fisher Sci.).
Confocal microscopy. Fluorescently stained samples were imaged using an Olympus Fluoview 1000 confocal microscope equipped with 40 Â oil immersion lens, or a LEICA TCS-SP5 confocal microscope equipped with a 63 Â glycerol immersion objective in sequential scanning mode and zoom rates between one and three. Histology samples were recorded by a SPOT camera and imaging software, using a Nikon epifluorescence microscope equipped with a 5x air objective in bright-field mode.
Image processing and statistical analysis. Images were processed using ImageJ (NIH) equipped with the LOCI bio-formats plugin and Photoshop (Adobe). Statistical analysis was done using Excel (Microsoft). Data presented are mean values±s.e. Significance was evaluated by the two-tailed student's t-test. Sample size (n-values) and P values are indicated in the figures and/or figure legends.
For analysis of PKC localization over ID (Fig. 2c, Supplementary Fig. 2h), profile plots were measured over aligned ID using the ImageJ plugin 'RGB profile plot'. Plot values (PKC fluorescence intensity and plakoglobin fluorescence intensity) were exported into Excel. The plakoglobin profile was used to establish the precise localization of the ID. The corresponding PKC intensity (F) at the ID was subtracted with the background PKC intensity (F0) to generate the background corrected profile PKCa plots (F À F0). Profile plots were centred at the ID, and the average profile plot and s.e. were calculated for each group, and used to generate the figure. Data presented in Fig. 2c and Supplementary Fig. 2h are mean values±s.e. Significance was evaluated by the two-tailed student's t-test over fluorescence intensity values (F À F0) at the ID. Sample size (n-values) and P values are indicated in the figures and/or figure legends.
For Phostag profile plot analysis (Fig. 3d), MLP bands were analysed using ImageJ and the 'Plot Profile' tool. Data were exported to Excel, aligned and normalized, and the resulting average plot profile for each group, including s.e. were used to generate the figure. Significance was evaluated by student's t-test analysis.
For analysis of CARP1, CARP2 and a-Actinin fluorescence intensity ratios (I ID / S ) between the ID and the sarcomere (S), identical-sized boxes encircling a portion of the ID, and an adjacent sarcomere were selected, and RGB values were measured in ImageJ using the 'RGB measure' plugin. After measurement of fluorescence intensities, individual and average intensity ratios as well as s.e. and significance was calculated in Excel. Data presented in Fig. 3f, Supplementary Figs 1e and 2g are mean values±s.e. Significance was evaluated by the two-tailed student's t-test. Sample size (n-values) and P values are indicated in the figures and/or figure legends.
Label-free measurement of cardiomyocyte contraction. Live label-free impedance measurements (cell index) of cardiomyocyte beating frequency (beats per minute) and contraction behaviour were done using the xCelligence RTCA cardio system (ACEA Biosciences) 60 . Basal contractile behaviour of cells was analysed two days after plating of NMCs into 96 well plates. The effect of increasing concentrations (0.5, 1, 5 and 10 mM) of the PKC inhibitor BI I-HCl (BI) in comparison to untreated cells was evaluated 21-22 h after addition of the small molecule inhibitor to the maintenance culture medium.
Data availability. All relevant data are available from the authors. | 9,751 | 2016-06-29T00:00:00.000 | [
"Biology"
] |
Froissart Bound on Total Cross-section without Unknown Constants
We determine the scale of the logarithm in the Froissart bound on total cross-sections using absolute bounds on the D-wave below threshold for $\pi\pi$ scattering. E.g. for $\pi^0 \pi^0$ scattering we show that for c.m. energy $\sqrt{s}\rightarrow \infty $, $\bar{\sigma}_{tot}(s,\infty)\equiv s\int_{s} ^{\infty} ds'\sigma_{tot}(s')/s'^2 \leq \pi (m_{\pi})^{-2} [\ln (s/s_0)+(1/2)\ln \ln (s/s_0) +1]^2$ where $m_\pi^2/s_0= 17\pi \sqrt{\pi/2} $ .
boundedness at fixed momentum transfer, were obtained by Epstein, Glaser and Martin [8] in the more general framework of the theory of local observables of Haag, Kastler and Ruelle [9].
Recently, Azimov has revisited the Froissart bound in a paper [10], Sec. 2 of which is similar to the 1962 and 1963 works of Martin [11].These papers were a precursor to Martin's later paper [2] which proved the bound rigorously from axiomatic field theory. Azimov has raised doubts about "application of the ideas and methods of axiomatic local field theory to hadron properties". His main point is that, "hadrons, consisting of quarks and gluons, cannot be pointlike", and might not be associated to local fields. However, Zimmermann [12] has shown that local fields can be associated to composite particles (for instance deuterons). We postulate that this construction applies to hadrons made of quarks. This is not obvious because, in spite of the practical successes of QCD, nobody knows how to incorporate particles without asymptotic fields in a field theory. Anyway this is a much weaker assumption than that of the validity of Mandelstam representation. In particular, we do not use the Froissart-Gribov representation of physical region partial waves for fixed s.
The Froissart-Martin bound has triggered much work on high energy theorems (see e.g. [13], [14]) and on models of high energy scattering [15]. Recently, Martin proved a bound on the total inelastic cross section at high energy [16] which is one-fourth of the bound σ max (s) on the total cross-section. Wu, Martin, Roy and Singh [17] obtained a bound on σ inel (s) in terms of σ tot (s) which vanishes both when the total cross-section vanishes and when it equals the unitarity upper bound.
In spite of all this progress, these bounds share severe shortcomings [17]. (i)They are deduced assuming that the absorptive part A(s, t), 0 ≤ t < t 0 is bounded by Const.s 2 / ln(s/s 0 ) for s → ∞. In fact, the Jin-Martin theorem on twice subtracted dispersion relations only guarantees that where s th is the s-channel threshold. As stressed by Yndurain and Common [18], this does not imply that A(s, t) ≤ Const.s 2 / ln(s/s 0 ) for all sequences of s → ∞.
(ii)The bounds are expressed in terms of σ max (s) which still contains the unknown scale s 0 of the logarithm, and the unknown positive parameter ǫ which can be chosen arbitrarily small but = 0. If ǫ is not fixed s 0 cannot be fixed since the advantage of a larger s 0 can be offset by a larger ǫ. We now remove both these shortcomings. We report definitive bounds on energy averages of the total crosssection in which the scale s 0 is determined in terms of C(t) which is a low energy (in fact below threshold) property in the t−channel. In some cases , e.g. for pion-pion scattering, for t → 4, C(t) is proportional to the Dwave scattering length [19] which is known phenomenologically; hence we obtain bounds on energy averages in terms of that scattering length. Even more exciting is the fact that for π 0 π 0 scattering we are able to obtain absolute bounds (in terms of pion-mass alone) on C(t) below threshold without assuming finiteness of the Dwave scattering length; this yields absolute bounds on the asymptotic energy averages of the total cross-section.
Normalizations. Let F (s, t) be an ab → ab scattering amplitude at c.m. energy √ s and momentum transfer squared t normalized for non-identical partcles a, b such that the differential cross-section is given by with t being given in terms of the c.m. momentum k and the scattering angle θ by the relation, Then, for fixed s larger than the physical s−channel threshold, F (s; cos θ) ≡ F (s, t) is analytic in the complex cos θ -plane inside the Lehmann-Martin ellipse with foci -1 and +1 and semi-major axis cos θ 0 = 1 + t 0 /(2k 2 ). Within the ellipse ,in particular, for |t| < t 0 , F (s, t) and the s-channel absorptive part F s (s, t) = A(s, t) have the convergent partial wave expansions, (2l + 1)P l (z)Ima l (s), (7) with the unitarity constraint, Correspondingly, the optical theorem gives, for a = b, For identical particles a = b e.g. for π 0 π 0 scattering, or for pion-pion scattering with Iso-spin I, we have the same formula for the differential cross-section, and the same form of the unitarity constraint, but the partial waves a l (s) → 2a I l (s) in the partial wave expansion,i.e.
With this normalization, F I (4, 0) = a I 0 , the S-wave scattering length for Iso-spin I. and for pion-pion scattering the identical particle factors lead to, In the following, we shall consider non-identical particles a = b for detailed derivations and quote the identical particle results when needed. Convexity Properties of Lower Bound on Absorptive Part in terms of Total Cross-Section. Martin has proved unitarity lower bounds on A(s, t) for 0 < t < t 0 in terms of σ tot (s) [2],and in terms of σ inel (s) [16]. He has also proved [20] that these bounds are convex functions of σ tot (s), and σ inel (s) respectively. We recall first the convexity properties which will be crucial for our proofs of lower bounds on C(t) in terms of energy averages of total cross-sections. We work at a fixed-s , and suppress the s-dependence of Ima l (s), and σ tot (s) for simplicity of writing. Using 0 ≤ Ima l ≤ 1, the lower bound on A(s, t) for given σ tot is obtained by choosing, where, the fraction η, 0 ≤ η < 1, and the integer L are determined from the given σ tot . Thus, where, Hence, A(z) is a monotonically increasing function of Σ tot with piecewise constant positive derivative. Denoting Int(x) = integer part of x, which increases with L since z > 1, and hence with Σ tot when it crosses square of an integer. This proves that the lower bound A(z) is a convex function of Σ tot , and that, and, for Σ tot > 1 Using integral representations for P µ (z) and for the modified Bessel function I 0 we obtain for µ ≥ 0, z > 1, This yields the strict inequality ( without any high energy approximation ), At high energy, this gives, which is a convex function of σ tot (s).
Upper bound on energy-averaged total crosssection . Defining, and we obtain, since the average of a convex function must be greater than the convex function of the average [21]. At high energies ifσ tot (s, ∞) goes to ∞, the asymptotic expansion of I 1 (ξ) yields, where, To extract a bound on the cross-section, we need the following lemma [20]. If ξ > 1, and then, Proof. It is enough to prove this for y = √ ξ exp ξ, since the right-hand side is an increasing function of ξ.Taking logarithms ,and using ξ = ln y − (1/2) ln ξ ≡ ξ 1 repeatedly, For fixed y the derivative of the right-hand side with respect to ξ 1 is (4ξ 2 1 ) −1 which is positive, and ξ 1 < ln y for ξ > 1. Hence the stated upper bound on ξ follows.
Instead of the s-dependent C s (t) we shall use the simple s-independent upper bound , which follows by using A(s, t) > A(s, 0) for 4 > t > 0 and improves the value C(t) if low energy total crosssections are known. The integral of the weight function multiplying σ tot can be done. Thus, where,σ With f (y) as defined above, the upper bound on the average total cross-section in terms of C x (t)is, We may also find bounds on the average of the total cross-section in the interval (s, 2s), The lower bound on A(s, t) and its convexity yield, Asymptotically we obtain a bound of the same form as before, but with the scale factor in the logarithm being s 0 /2, Note that σ tot (s) <σ tot (s, 2s) if the cross-section increases with s in the interval (s, 2s);the above bound on energy averages therefore immediately yields a bound on σ tot (s) in that case.
For identical particles there are only even partial waves in the partial wave expansions, but the lower bound on the absorptive part is again a convex function of the total cross-section; the identical particle factors multiplying the partial waves ensure that inspite of only even partial waves contributing, the largest partial wave L in the variational bound which is of O( sσ tot (s)) has only O(1) corrections with respect to the non-identical particles case.The quoted asymptotic bounds on the absorptive part in terms of σ tot and on the energy averaged total cross-section in terms of C x (t) therefore remain unchanged.
Phenomenological Bounds in terms of D-wave Scattering Length. Rigorous results from axiomatic field theory do not guarantee finiteness of the D-wave scattering lengths. However if we use phenomenological values for them we can choose ǫ = 0 and evaluate C(t = 4) . We shall use, F π + π 0 →π + π 0 = 1/2(F 1 + F 2 ), F π 0 π 0 →π 0 π 0 = 1 3 F 0 + 2 3 F 2 , the crossing relation, and the total crossing symmetry of the π 0 π 0 → π 0 π 0 amplitude. If we denote F (s, t) = G(t, s) = G(t; z t ) where F (s, t) denotes the π + π 0 → π + π 0 or the π 0 π 0 → π 0 π 0 amplitude , then the corresponding G(t, s) has only even partial waves, For 0 < t < 4, with the absorptive part F s (s, t) defined by Eq. (7), the fixed-t dispersion relations with two subtractions imply the Froissart-Gribov formula rigorously for l ≥ 2, (39) where Q l denotes the Legendre function of the second kind.The positivity of the absorptive part then implies the positivity of g l (t) for 0 < t < 4; further, If the t-channel D-wave scattering lengths exist,the definitions of C(t) and of the D-wave scattering lengths a I 2 for iso-spin I yield, and Here we have defined the l-wave scattering lengths a I l as the q → 0 limits of the phase shifts δ I l (q) divided by q 2l+1 where q is the c.m. momentum . Then an S-wave scattering length is indeed a length, with dimension m −1 π , and the D-wave scattering lengths have dimension m −5 π . Then, phenomenologically [19] we have, and Roy [14] has obtained from low energy data, for x = 50,σ π 0 π 0 tot (x) = 8.2 ± 4 mb;σ π + π 0 tot (x) = 17 ± 3.5 mb. (44) With ǫ = 0 ,t = 4 and the values of C x (t = 4) given in terms of the scattering lengths, and the low energy total cross-sections, we have, from Eqs. (31)- (33), with x = 50, where we have indicated the separate contributions of the D-wave scattering lengths and low energy total crosssections to C x (4) but have not indicated the (substantial) errors on them which imply corresponding errors on the scale factors. Our bounds on average total cross-sections for π + π 0 and π 0 π 0 scattering therefore do not contain any unknown constants but the scale factor s 0 has large phenomenological errors. We cure this problem in the next section at the cost of getting poorer bounds.
Absolute bounds on the D-wave below threshold for π 0 π 0 scattering. Although threshold behaviour cannot be proved from first principles, it was shown long ago [22] that |f l (s)| < C(4 − s) l−1 must hold for 0 < s < 4. We derive an absolute bound of this form and use it to derive a rigorous asymptotic bound on energy averaged total cross-section for π 0 π 0 scattering without unknown constants.As noted already, for 0 < s < 4 and l ≥ 2, the Froissart-Gribov formula implies that f l (s) > 0. Hence, for 0 < s < 4, 4 − s < t < 4 the convergent partial wave expansion, is in fact a sum of positive terms and yields an upper bound for the l ≥ 2 partial waves if we can obtain a bound on F (s, t) − F (s, 0) using analyticity. The twice subtracted fixed-t dispersion relations in s can be rewritten in terms of the convenient variable z ≡ ( and the positivity of the absorptive part then yields, If s 1 < s < 4 and z 1 ≡ (s 1 − 2 + t/2) 2 , then and hence for z 1 < z < z 0 , Inserting this into the dispersion relation we have,for 4 > t > 4 − s − s 1 , and t ≥ 0, We now use absolute bounds on pion-pion amplitudes first discovered by Martin [23], and improved successively by [4], [24] and [25] in the improved final form, The partial wave expansion of F (s, 2)−F (s, 0) now yields for 3 < s < 4, f l (s) ≤ 6.25 + 33.99 which implies in particular, With s replaced by t in this formula, the Froissart-Gribov formula now yields, Absolute bound on energy averaged total crosssection for π 0 π 0 scattering at high energy. Inserting the bound on C(t) into the average cross-section bound, the optimum value of t turns out to be t = 4 − (1/8 ln(s/s 0 )) −1 , and the optimum bound, Forσ tot (s, 2s) we obtain the same form of the bound ,but with half the value of s 0 . Outlook and Acknowledgements. Our basic bound on the absorptive part, Eq. (20), is valid at all energies and its energy integral may be used for comparisons with experimental cross-section data which have a large non-asymptotic contribution at current energies. We have highlighted the simpler asymptotic upper bounds on average total cross-sections .
We believe that our result is important as a matter of principle. However, we also believe that the magnitude of the coefficient in front of the Froissart bound is not satisfactory, especially if one decides to beiieve that the Froissart term is universal and compares with p-p and ppbar cross-sections at the ISR [26],at the SppbarS [27], at the Tevatron [28] and at the LHC [29]. All these indicate the existence of a Froissart like contribution with a much smaller coefficient, and a much larger scale and are well reproduced by, for instance, the BSW model [15] which incorporates automatically the Froissart behaviour. Returning to ππ scattering, can the situation be improved? Yes, because one has to enforce crossing symmetry and unitarity. Kupsch [30] has constructed a crossing symmetric model satisfying Eq.(8), but never tried to get numerical results. Also, we believe that unitarity in the elastic strips could be important. This led to the discovery by Gribov [31] that the behaviour sF(t) for the total amplitude is impossible. If you remove the elastic unitarity constraint [32] the Gribov theorem disappears. To attack the problem one could use a variational approach taking as an input the inelastic double spectral function in the Mandelstam representation. All we need is to find someone courageous not looking for a job.
Similar bounds on inelastic cross-sections without any unknown constants will be reported separately [33]. | 3,791.2 | 2013-06-21T00:00:00.000 | [
"Materials Science",
"Mathematics"
] |
The Inexorable Spread of a Newly Arisen Neo-Y Chromosome
A newly arisen Y-chromosome can become established in one part of a species range by genetic drift or through the effects of selection on sexually antagonistic alleles. However, it is difficult to explain why it should then spread throughout the species range after this initial episode. As it spreads into new populations, it will actually enter females. It would then be expected to perform poorly since it will have been shaped by the selective regime of the male-only environment from which it came. We address this problem using computer models of hybrid zone dynamics where a neo-XY chromosomal race meets the ancestral karyotype. Our models consider that the neo-Y was established by the fusion of an autosome with the ancestral X-chromosome (thereby creating the Y and the ‘fused X’). Our principal finding is that sexually antagonistic effects of the Y induce indirect selection in favour of the fused X-chromosomes, causing their spread. The Y-chromosome can then spread, protected behind the advancing shield of the fused X distribution. This mode of spread provides a robust explanation of how newly arisen Y-chromosomes can spread. A Y-chromosome would be expected to accumulate mutations that would cause it to be selected against when it is a rare newly arrived migrant. The Y can spread, nevertheless, because of the indirect selection induced by gene flow (which can only be observed in models comprising multiple populations). These results suggest a fundamental re-evaluation of sex-chromosome hybrid zones. The well-understood evolutionary events that initiate the Y-chromosome's degeneration will actually fuel its range expansion.
Introduction
Our understanding of sex chromosome evolution has increased immensely in the past decade. Theoretical expectations [1,2] have been experimentally verified in a wide variety of organisms, including fish, fruitflies, mammals and plants [3][4][5][6][7][8]. For example, it is widely documented that over a series of generations, a Ychromosome will eventually stop recombining with the X over most of its length. As a consequence there are increased rates of transposition, degeneration, heterochromatinization and loss of function of genes on the Y, amongst other changes [5,[8][9][10][11][12][13].
It appears then, that the inexorable fate of Y-chromosomes is degeneration and perhaps loss. It is even possible that all sexually dimorphic species lacking a Y have previously passed through a Ypossessing stage [1,14], as is the case for Caernohabditis elegans [15] (the logic would also apply to equivalent W chromosomes in species with heterogametic females). The persistence of Ychromosomes to the present day therefore suggests that they can repeatedly arise de novo. One straightforward way in which new Ys can be created is by the fusion between an autosome and Xchromosome followed by its fixation. This paper models the evolution of such neo-XY sex chromosome systems and, in particular, asks why they should become established throughout a species' range. The analysis suggests that the spread of neo-Ys is much more likely than suggested by current models, and that this new proposal could be tested by analysis of sex chromosome hybrid zones.
A concrete example can be a useful guide for explaining and constructing evolutionary models, so we make use of the wellstudied example of the neo-XY race of the grasshopper Podisma pedestris. Phylogenetic comparison [16] suggests that the ancestral P. pedestris karyotype had females with two X-chromosomes and males with one (but no Y): this is known as an XX/XO sexdetermining system [17]. The system changed following the centric fusion of the X with an autosome (A u ) to create a larger metacentric neo-X. The fused karyotype has become fixed in populations in the southern part of the species' distribution in the French Alps. In these fixed populations, the females contain two neo-X chromosomes, and hence no unfused A u . In males, however, the unfused A u chromosomes have continued to pair with the homologous section of the neo-X. These unfused A u are now restricted to males and are consequently designated neo-Y chromosomes. The karyotypes of the original unfused, and derived fused race are illustrated in Figure 1 (karyotypes A, B, F, G).
Sex chromosomes are often involved in fusions. Indeed, human sex chromosomes are believed to be the products of at least three chromosomal fusions [6,18,19], as is the Drosophila Y-chromosome [20]. The occurrence of a fusion is insufficient to explain the genesis of a neo-XY system however. Following the fusion event, the new karyotypes must also become fixed throughout the species range (or part of it). The establishment of neo-XY systems does appear to occur repeatedly in evolution. Good evidence comes from the Orthoptera, which have conveniently large chromosomes for surveys of karyotype. White calculates that there have probably been six independent fixations of the XY system from an ancestral XO condition in the Australian subfamily Morabinae alone (based on karyotypes from about 80 species). More generally, the fixation of the XY system has been reported in at least 21 genera of Acrididae [16](and references therein).
Attempts to explain this establishment fall broadly into two traditions. Firstly, cytogeneticists have noticed that chromosomal rearrangements often confer reduced fertility in heterozygotes. The cause may be a direct effect of meiotic aberrations [21,22]; in other cases selection against changed recombination patterns is suspected (see [23] for an example). These forms of selection actually act against the fusion when it is rare; but it could nevertheless become established in small isolated populations if genetic drift elevated its frequency until it became the commoner type and hence favoured by selection [24]. The difficulty with this explanation is to account for how the fusion would subsequently spread from a single isolated population to other populations. Lande [24] proposed that the fused race could colonize sites left vacant by local extinctions, whereas Hewitt [25] argues that spreading would be more effective if the initial population was Figure 1. The karyotypes occurring in the hybrid zone. The karyotypes are organised by sex and the generation they first appear (in a cross between neo-XY and XO populations). The lower pink chromosome is the ancestral X. The upper chromosome can be an unfused A u (yellow), or fused to the X (as part of the neo-X, also shown in yellow), or a Y-chromosome (green). Each karyotype has been labelled by a letter, for reference. The expected frequency of each karyotype at equilibrium is shown for the neutral case after a cross between equal numbers of new-XY and XO. Note the 1:4 ratio of the Y to unfused A u in the parental populations (made of A, B, F, G). doi:10.1371/journal.pgen.1000082.g001
Author Summary
Comparisons between related species have shown that, over evolutionary time scales, Y-chromosomes tend to degenerate and can be completely lost. How then can we explain the persistence of Y-chromosomes to the present? One possibility is that losses are counter-balanced by the origin of new Y chromosomes, which then spread throughout the species in which they have arisen. The first of these two processes, the generation of new Y chromsomes, is more readily understood: it can occur if an autosome (a non sex chromosome) fuses with an X chromosome. This form might become established in one locality. However, its subsequent geographic spread has been more challenging to explain. Problems arise if gene flow carries them to another part of the species range. Crosses can then occur which introduce the new Y chromosome into females, who are expected to suffer reduced fitness. The new sex chromosomes are therefore selected against when they are in the minority. We use simulations to show that they can nevertheless spread, if they meet the ancestral forms at a front so the chromosomes intermingle in a hybrid zone. Paradoxically, the degeneration of the Y will actually intensify selection, thereby speeding its spread. located on the expanding margin of the species range, as it spreads into new territory-most likely during an episode of rapid climatic change.
A second perspective comes from consideration of the alleles that were segregating on autosomes before the fusion occurred. Fusion with a sex chromosome might bring alleles into linkage with the newly created sex chromosome and confer a fitness benefit; the key alleles might be sexually antagonistic (benefiting one sex at the expense of the other) [26] or deleterious recessives [27] (especially in strongly inbreeding populations). In both cases the selection is expected to be much more effective in promoting Y-autosome fusions, and such events might repeatedly add new genetic material to existing Y-chromosomes. Linkage with sexually antagonistic alleles could also produce selection for the fixation of new X-autosome fusions, and hence the creation of neo-Ys.
There is some evidence that sexually antagonistic alleles may indeed have appreciable effects. Rice conducted an imaginative breeding design in which a haploid Drosophila genome was restricted to one sex for several generations and then returned to the other sex [28][29][30]. The results were striking. In less than 30 generations, sex specific fitness differences had become established in the sex-restricted genome. The rapidity of the response was interpreted as showing that sexually antagonistic alleles had been segregating in the founder population.
Even with strong selection on sexually antagonistic alleles, the advantage provided to the fused chromosome would be weak [26]. Nonetheless, this selective process, or the action of drift, might establish the neo-XY system locally in part of the species distribution. The spread throughout the whole species range is more difficult to explain. Any advantage to the fusion when rare is expected to be transient, because of the well-understood evolutionary events affecting new sex chromosomes. Alleles reducing female fitness, can accumulate readily on the Y [1,2], particularly if they also had beneficial effects in males.
Our analysis, has uncovered a paradoxical effect that nevertheless favours the geographic spread of the neo-XY system. If sexually antagonistic alleles have become established on the Y, the genetic interactions at the boundary between neo-XY and ancestral populations can favour the spread of the neo-X. Surprisingly, the results hold even if the net effect of selection against the neo-Y in females outweighs the benefits in males.
Materials and Methods
The pattern of chromosome segregation in crosses involving individuals with different sex chromosome combinations is illustrated in Figure 2. The letters correspond to the karyotypes shown in Figure 1, the area of each cell is proportional to the number of each karyotype in the offspring. This scheme was translated into a set of equations for the frequency of each karyotype as a function of the frequencies in the previous generation, assuming random mating and after weighting each karyotype by its fitness. A program to iterate the equations was written in the statistical language R [31], and is listed in the supporting information (Dataset S1). The initial analysis revisited, and then extended the results of [26]. Consider two sexually antagonistic alleles that might be segregating on the autosome A u during the period before the chromosomal fusion. The two alleles (a and b) have different fitness in the two sexes (specified by w Raa , w Rab & w Rbb for females, and w =aa , w =ab & w =bb for males). The a allele was assumed to be favored in males, and the b in females so w Rbb = w =aa = 1. The program calculated the outcome of selection for all possible combinations of w =bb and w Raa in the range 0-1 at intervals of 1/40 (with a specified dominance). If there is polymorphism at this locus before the fusion takes place, the fusion will (in some cases) generate linkage disequilibrium leading to selection for the fixation of the fusion. The fitness combinations leading to such polymorphism can be illustrated by initiating a simulation with only the ancestral XX:XO karyotypes, and including a chromosome carrying the a allele (equivalent to the green chromosome in Figure 1) at low frequency. Figure 3A illustrates a range of fitness combinations producing polymorphism (achieved from an initial a frequency of 0.1% with all genotypes in Hardy-Weinberg proportions). The combinations of fitness that can lead to selection for the fixation of the fusion have been explored in some detail [26]. In Figure 3, they correspond to the b allele becoming linked to the X by the fusion. This selection for the fusion could be demonstrated in our simulations by initiating the fusion at a low frequency (0.4%, corresponding to the frequency of the fusion, if migration had introduced the Y at 0.1%) and then iterating the equations for 1000 generations. The analysis was extended to investigate the effect of the slightly reduced fertility that is expected in females heterozygous for the fusion. This additional selection was set at s = 0.01, the value estimated for P. pedestris. The recombination rate between the b allele and the X centromere was set at zero to maximize the selection for the fusion [26]. As the Y-chromosome evolves, recombination is expected to be reduced over a greater proportion of the Y [1], hence the modelled effects are increasingly likely to occur. Indeed, in the case of P. pedestris, which is assumed to have a young neo-XY system, the recombination is already displaced way from the Y centromere [32], perhaps simply as effect of the fusion itself, and very strong linkage disequilibrium has been found even in the middle of the zone for an X-marker [33].
Simulation of a Contact Zone between XO and Neo-XY Populations
These initial calculations involved a single panmictic population. The outcome can be different when the population is subdivided. The next step was therefore to consider a situation in which the neo-XY system had become established in an isolated area, and come into contact with the ancestral (XX:XO) karyotype. Gene flow between the two chromosomal races would then produce a hybrid zone. A computer simulation of a linear array of 40 populations was used to model this situation. Initially the left hand 20 populations were fixed for the ancestral karyotype and the remainder for the neo-XY. There was gene flow of 8% between adjacent populations (total gene flow of 16%). Population size was uniform across the simulated populations. In other words there was no density trap to pin the zone down to a particular location (as described by [34]). The two ends of the array could either be set to receive gene flow from populations fixed for the ancestral karyotypes, or to only receive gene flow from their more central neighbour. Both options were used to check for any effect on the outcome of the simulations. For each generation, after gene flow, the expected frequencies of genotypes in the next generation were calculated as before.
In this extension of the model, the green chromosome in Figure 1 is considered to be a neo-Y-chromosome. The a allele would have been fixed on the neo-Y which could also have accumulated additional sexually antagonistic alleles tightly linked to the X centromere (due to the extension of the non-recombining region). We have argued that we would expect some alleles to be selected against in males and favoured in females, and for there to be selection against the chromosomal heterozygotes. However, it is helpful to understand the combination of these effects by first examining the behaviour of these three forms of selection individually. We therefore summarize the results by defining three selection regimes. The first two correspond to points on the X and Y axes of Figure 3: Firstly, male-beneficial variants could have become established on the Y (w =(AU?) ,1), were 'A u ?' represents genotypes containing A u -the autosomal homologue of the neo-Y; secondly, female-deleterious variants could occur on the Y (w R(Y?) ,1). The third simple case is selection against females heterozygous for the fused X (w R(FU) ,1). We then simulated all We consider an ancestral autosomal locus, which had two sexually antagonistic alleles: a was favoured in males and b in females. 3A. The outcome of selection on the b allele in the ancestral population as a function of the fitnesses of the two homozygotes. The central area of fitness combinations results in a stable polymorphism (delineated by the contours p a = 0.001 and p a = 0.999). In this example the alleles were additive. 3B. The evolutionary dynamics change if one of the b-bearing autosomes fuses to the X-chromosome. We illustrate this effect by plotting the frequency of the b-neo-X haplotype 1 000 generations after it has been introduced at low frequency (0.04). The neo-X spreads for some fitness combinations (the central area enclosed by 0.001 and 0.999 contours). The spread of the neo-X was opposed by weak selection against females heterozygous for the fusion (genotypes H and I from Figure 1 were assumed to suffer a 1% reduction in fertility). Note, that mild sexually antagonistic selection (i.e. the region near the point (1,1)) is insufficient to favour the fused X. doi:10.1371/journal.pgen.1000082.g003 possible combinations of the fitness regimes. Table 1 sets out the karyotypes with reduced fitness in each regime.
Having determined the basic patterns produced by the different fitness regimes, we assessed the spread of the neo-XY system throughout the possible parameter range shown in Figure 3. The simulations were run until fixation or until 10 000 generations. We explored the full range of values for s f and s m in the presence of selection against chromosomal heterozygotes, which was set to that estimated in P. pedestris of s h = 0.01 (fitnesses specified in the first and second rows of Table 1, combined multiplicatively). The dominance of the sexually antagonistic selection (d m and d f in Table 1) was varied: from d m = 1 or 0.5 for the male effect (not zero since recessive male-beneficial alleles would have had no advantage on the Y), and d f = 1, 0.5 or 0 for the female effect. In addition to these forms of selection, we also considered the possibility that there had also been evolution of coadaptation or dosage compensation between the sex chromosomes in the established neo-XY and XO populations, giving rise to the fitnesses in the last two rows of Table 1.
The model zone width was converted to values that could be observed in the field using the relationship s 2 = mD 2 , were s 2 is the variance in parent-offspring dispersal and is a measure of migration. For P. pedestris, it has been estimated by mark-releaserecapture experiments to be 400 m 2 per generation [35]. The value D represents the distance in the field equivalent to that between adjacent simulated populations. Since the simulated migration rate, m, was 0.16, the real hybrid zone width of 800 m [34] is equivalent to the distance between 16 simulated populations.
When all fitnesses were set to one, the simulated width of the zone increased with time-matching the neutral expectation w = 2.51s!t, where w is the width, s is the parent-offspring dispersal per generation and t is time in generations [36] (results not shown). Similarly, in the case w R(FU) ,1, the simulated width of the fusion cline fitted analytical expectations, as long as the femalespecific nature of selection was taken into account (see Results). The program to simulate the structured populations was written in Java and the full source is available from the authors upon request. Figure 3B shows the neo-XY chromosomes can invade an ancestral XO population when they are introduced at low frequency, if there is strong sexually antagonistic selection (the central area of the Figure). The spread of the fused X (the neo-X) was accompanied by fixation of the a allele. The model included weak selection against females heterozygous for the fusion, hence the fusion was selected against when rare. Consequently there were combinations of low to moderate selection (i.e. around (1,1)) for which the fusion did not spread. This failure to spread occurred irrespective of the dominance of the a allele in males, or the b allele in females (results not shown). Figure 4 summarizes the various outcomes of the simulated meeting between the two chromosomal races to form a hybrid zone. The relative frequency of the fusion has been plotted as against distance along the array of populations. This relative frequency was calculated as f f /(f f + f u ), where f i specifies the frequency of the chromosome of type iM{f, u, Y, Au}, representing fused X, unfused X, Y and autosome respectively. Similarly, the frequency of the Y was calculated as f Y /(f Y + f Au ). Note that the denominator increases with f u , since unfused individuals carry more of these chromosome (i.e. more Au and/or Ys, see Figure 1). Once the two races meet, gene flow produces a sigmoidal transition in the frequency of the fused X and the Y-chromosomes. For some parameter values the Y cline or the fusion cline spread as a wave of advance, indicated by arrows in Figure 2, in other cases the clines were stable or decayed (see Discussion).
Results
Existing analytical models do not describe much of this behaviour, but there are well known results for the case of heterozygote disadvantage w R(FU) ,1 under which the fusion cline would assume a fixed position and width. The relationship between the strength of selection (in the range 0.05-0.9) and width (the inverse of the maximum slope [37]) closely fitted the expected relationship w = !8s/!s [38] as long as the selection coefficient, s, was multiplied by 2/3 to compensate for selection acting on females only (r 2 .99.99%, regression coefficient = 1.016).
The examples in Figure 4, in which the neo-XY system spreads (upper two panels), involve strong sexually antagonistic selectionchosen to clearly illustrate the qualitative difference in outcome from single population models (in which it did not for these values). The neo-XY system spreads even when there is strong selection against the Y in females.
We investigated the rate of spread of the neo-XY system under less severe selection ( Table 2). Selection against the X chromosome heterozygotes of 1% (s h = 0.01) slowed down the rate of spread slightly (up to 50%), but did not prevent it even when the sexually antagonistic selection was weak (e.g. s m = s f = 0.005). Other forms of selection against introgression could both accelerate or retard the rate of spread. We investigated the fitnesses that might be generated by coadaptation and dosage compensation of the sex chromosomes (Table 1). Both forms of selection accelerated the spread when the selection for the Y in males was greater than the disadvantage in females (s m .s f ) and retarded or even slightly reversed the direction of spread under the converse (s m ,s f ) If more than one form of selection was acting, the values in the corresponding columns were multiplied. The forms of section shown in Figure 4 ( Table 2). The effect of dominance was minor over most fitness combinations and the outcomes were qualitatively unaffected. We show the effects of male dominance in Table 2, but omit the female for brevity.
Discussion
In the Introduction we outlined how the fixation of an Xautosome fusion could be explained by selection in favour of sexually antagonistic alleles linked to the fused centromere. Charlesworth and Charlesworth [26] have shown that there is net selection in favor of the fusion for fitness combinations that lead to polymorphism at the sexually antagonistic locus: which fall in the central shown in Figure 3A. However, Figure 3B suggests that this form of selection might be readily counteracted, even by a very minor (1%) reduction in the fertility of female fusion heterozygotes, w R(FU) . In particular, even relatively strong sexually antagonistic selection is overwhelmed: notice that when this additional selection is applied ( Figure 3B) the fusion does not spread for fitness values within 0.15 of the point (1,1) (i.e. selection coefficients of up to 15%) even if they fall within the polymorphic area in Figure 3A. Selection against female fusion heterozygotes is considered likely because of non-disjunction at meiosis [21], and is indeed suspected to occur in P. pedestris [37]. In these circumstances, it is easier to envisage the fusion becoming established by genetic drift, than deterministically under the action of selection.
Whatever the reason for the fusion initially becoming established in one locality, once it is fixed, the A u autosome will then be restricted to males, and would consequently have become a neo-Y. The subsequent evolution of the sex chromosomes would therefore take a course that would at first sight seem to make the spread of the neo-XY system even more unlikely. In particular the neo-Y is expected to accumulate further sexually antagonistic effects, which would in turn select for the loss of recombination, and its eventual degeneration [1]. The selection acting would therefore be stronger (further away from the point (1,1) in Figure 3B) and for the most part this would lead to even stronger selection to eliminate the neo-XY system: only if selection is strong and of similar order in males and females (the triangular area) would selection favour its spread.
The model illustrated in Figure 3B assumes a small starting number of neo-XY individuals. Remarkably, the outcome is completely overturned if the XO and neo-XY populations are assumed to meet in a hybrid zone. The spread of the neo-XY system would actually be driven by the selection regimes that lead to its elimination in Figure 3B.
Understanding the Simulations of a Hybrid Zone
It may be simplest to start interpreting the results using the biologically unrealistic case of selection only against females containing Y-chromosomes, w R(Y?) ,1. Since only females with an unfused X chromosome can contain a Y, this regime leads to selection against them, causing the fusion cline to advance (Figure 4, w R(Y?) ,1). However, in the absence of other selection, this effect is transient since the direct selection on the Y removes it from populations containing unfused chromosomes. In other words, the autosome (A u ) advances because it is favoured by selection. The Y persists only in the heartlands of the fused chromosome range, because there it experiences no disadvantage because it cannot enter females.
There is a comparable indirect effect on the fusion in the case of selection only in favour of Y-chromosomes in males, w =(Au?) ,1. In populations that are polymorphic (for Y/A u ), unfused males are more likely to contain at least one advantageous Y because they have double the number of these chromosomes (in fused males, the X replaces one of them). Hence w =(Au?) ,1 results in selection against the fusion (Figure 4, w =(Au?) ).
We can extend these explanations to the most interesting and biologically relevant result-the wave of advance for both the Y and the fusion clines under sexually antagonistic selection (w =(Au?) ,1 & w R(Y?) ,1). When the selection against the Y in females is stronger, the Y-chromosome tends to be removed from the fusion cline as under w R(Y?) ,1. However, as the fusion advances (for the same reason as under w R(Y?) ,1) the Y follows behind, up to the margins of the fusion cline, thereby indefinitely maintaining the selection for the advance of the fusion. Note that this neo-XY success The speed is expressed as the number of generations for the fusion cline to advance 1 Km. The location of the cline centre was output every 50 generations, and speed was estimated as the time taken to reach the end of the array (equivalent to 1 Km in P. pedestris). In those cases where the end was not reached within 10000 generations, the speed was estimated from the movement to that time. doi:10.1371/journal.pgen.1000082.t002 depends on the gene flow continually bringing the Y-chromosomes into the zone, which is why it did not occur in the single partially isolated population of Figure 3B. The neo-XY system also spread when there is a net advantage to the Y (under sexually antagonistic selection). However, in this case, the Y spreads as a traveling wave ahead of the fusion cline. The fused X will also spread because, as the Y becomes common, the selection against unfused females gets stronger whereas the benefit to unfused males is reduced (since both fused and unfused males tend to carry the favourable Ys once they become common).
The observed speed of spread of the fusion was relatively small compared to dispersal: the fastest being equivalent to 450 generations to move 1 km in P. pedestris, or 1/10th of the dispersal distance per generation. It would be difficult to observe by repeated sampling, but would cause consistent movement over evolutionary time. In addition, the sexually antagonistic effects could increase in magnitude with time, because newly evolved sexually antagonistic alleles arising in within the neo-XY range would be expected to spread until they met the hybrid zone.
We consider the forms of dominance in Table 2 to be most likely, although we have explored other combinations and found no qualitatively different outcomes. Dominant male beneficial effects would fix faster initially, making them more likely to evolve, however such alleles could spread in XO or neo-XY populations, so they need not be restricted to the neo-XY race; hence there could be a range of dominance for male beneficial effects. The first female deleterious effects on the Y would probably be due to loss of function mutations, and would therefore be less than fully dominant because the functional ancestral allele would still be present.
In addition to heterozygote disadvantage we also considered alternative forms of selection that might act against recombinant karyotypes in the hybrid zone. As long as these additional effects are of the same order as the sexually antagonistic selection, the results are essentially unchanged. It is, of course, possible to find some fitness combinations that slow down, or even slight reverse the spread, and it could be counter-acted by assymetrical gene flow (e.g. due to a density gradient [34]) or other selection (e.g. due to change in the environment). However, Rice's experiments [28][29][30] suggest that the alleles with large sexually antagonistic effects are segregating in natural populations, so we would expect their effects to predominate in the zone as soon as it was formed. If this interpretation is correct, more detailed analysis of hybrid zones should provide additional evidence of this sexually antagonistic selection (see below).
Strong Selection Can Produce Broad Clines
Interestingly, strong selection on the Y chromosome resulted in a broad fusion cline (Figure 4). The result emphasizes that the width of the cline in the character for which the hybrid zone was originally discovered, need not indicate the strength of selection. In fact the term ''hybrid zone'' can be misleading in these cases; it is preferable to refer to different clines. The broad fusion cline in the presence of strong selection is particularly relevant to the P. pedestris hybrid zone, where strong selection is detected in the F 1 in lab crosses [37,39] and in the field [40], yet the fusion cline is much wider than expected from the observed selection [41]. Previously, this discrepancy has been explained by a model in which selection is spread over many loci [41] only some of which need be linked to the fusion, but our results offer an alternative possibility: that the action of the selection is indirect and due to the well understood initial events in sex chromosome evolution.
Sex Chromosome Hybrid Zones as Natural Sex Chromosome Evolution Experiments
One implication of the results is that sex chromosome hybrid zones are a valuable, yet unexploited, source of information on early sex chromosome evolution. We suggest that it will be rewarding to obtain markers that distinguish the Y-chromosome from its homologous autosome (A u in our notation) and to survey their geographic distribution across known sex chromosome hybrid zones. Often the clines of different characters coincide (have the same centre) [42,43], however we would expect them to be displaced in the case of sexually antagonistic selection. Comparing the Y and the fusion cline as in Figure 4 using real hybrid zone data is a robust way to identify the selection regime operating.
The conventional explanation for a narrow sex chromosome hybrid zone is that there is selection against the chromosomal heterozygotes [34,37]. In that case there would be a narrow transition for the chromosomal fusion, but the distribution of the Y would be very similar to the neutral case (compare Figure 4, w R(FU) ,1 with the neutral case). However, if sexually antagonistic selection is operating, then these two clines will be displaced and the position on the Y cline relative to the fusion cline will indicate the relative strength of male beneficial and female deleterious effects on the Y. For example faster male evolution [44] would be supported if the Y cline were ahead of the fusion. This novel information on the forms of selection affecting young Y-chromosomes in natural populations has not previously been tapped.
A second indication of sexually antagonistic selection would be cline movement. In some cases it has proved possible to detect the actual movement of hybrid zones by repeated surveys or reviewing museum collections e.g. [45,46]. In other cases the movement would be too slow, or held back by barriers to gene flow or gradients in population density [47]. It should still prove possible to identify slow or historical movement by surveys of other loci throughout the nuclear and cytoplasmic genome (for a review see [48]).
The realisation that surveys of sex chromosome hybrid zones can answer questions relating to the early evolution of sex chromosomes is exciting because such hybrid zones are already known and waiting to be analyzed. Examples include Drosophila americana [49], the morabine grasshopper Vandiemenella (Warramaba) viatica [25] and the grasshopper Podisma pedestris [34]. A great advantage of hybrid zone studies is that they involve wild populations [25] and some may represent snapshots of the actual establishment of a neo-XY system in nature, thus allowing the testing of theoretical predictions in biologically realistic conditions.
Supporting Information
Dataset S1 Listing of the simulation which generated Figure 3. It is code which runs in R: the free software environment for statistical computing and graphics. Found at: doi:10.1371/journal.pgen.1000082.s001 (0.04 MB RTF) | 7,942.2 | 2008-05-01T00:00:00.000 | [
"Biology"
] |
Torus and Subharmonic Motions of a Forced Vibration System in 1:5 Weak Resonance
The Neimark-Sacker bifurcation of a forced vibration system is considered in this paper. The series solution to the motion equation is obtained, and the Poincaré map is established. The fixed point of the Poincaré map is guaranteed by the implicit function theorem. The map is transformed into its normal form at the fifth-order resonance case. For some parameter values, there exists the torus T1. Furthermore, the phenomenon of phase locking on the torus T1 is investigated and the parameter condition under which there exists subharmonic motion on the torus T1 is determined.
Introduction
In this article, we investigate the torus and subharmonic motions of the following system: _ y 1 = −y 2 , _ y 2 = y 1 + a 1 μy 2 + by 2 2 + sy 3 1 + cy 3 2 where μ, ε, and δ are small parameters; f is a 2π periodic function; and a 1 , s, b, and c are constants. ðq + δÞ is the frequency of the external force. If q is a positive integer with q ≥ 5, we say that the system (1) is in 1 : q weak resonance. For q being a positive integer satisfying q ≤ 4, the system (1) is referred to as 1 : q strong resonance. h:o:t: represents the "higher-order terms" than those that have been written out, and the same is true below. There are some mechanical models whose dynamical behaviors can be described by Equation (1), for example, the system shown in Figure 1, see Ref. [1] for more information. In this paper, we investigate abstractly system (1) and give a method for analyzing its Neimark-Sacker bifurcation. The value of relevant param-eters and constants depends on some specific mechanical models whose dynamics can be described by Equation (1). Therefore, we do not introduce the given parameters or constants here, and only choose several sets of values for numerical simulations in Section 4. For ε = 0, Equation (1) undergoes the Hopf bifurcation under certain conditions, and then, for ε ≠ 0 and jεj, sufficiently small torus or the qth order subharmonic motions can occur to Equation (1). The problem of 1 : q resonance of a closed orbit in R 3 (or in C × S 1 ) leads to the study of the Z q -equivariant planar vector (see Refs. [2,3]) whose versal unfolding has been studied for q ≠ 4 and been conjectured for q = 4 by Arnold [2]. Bifurcation sequence inventory at 1 : 4 resonance has been presented by Krauskopf [3]. Gambaudo [4] considered the general study of the periodic perturbation of a one family of autonomous differential equations in the plane satisfying conditions for a generic Hopf bifurcation. Iooss [5] investigated the subharmonic motion in the 1 : 3 and 1 : 4 resonance case. Wan [6] analyzed the Neimark-Sacker bifurcation in the 1 : 4 strong resonance case for the planar map. The width of the resonance tongue at a distance σ from the unit circle given by Arnold [2,7] is of the order of σ ðq−2Þ/2 . Shilnikov et al. [8] and Iooss [9] computed the Arnold tongue in weak resonance case for the planar map. The Neimark-Sacker bifurcation of an oscillator with dry friction was observed in Ref. [1]. Periodic-impact motions and bifurcations of vibroimpact systems near the 1 : 4 strong resonance point are considered in Ref. [10]. Results on other types of forced vibrations can be found in some literatures, see, for example, Refs. [11][12][13][14][15][16].
Judging from the above statement, we know that there is a lack of efficient criteria depending on the coefficients of the original differential equations, based on which we can talk about the asymptotic behaviors of trajectory. In this paper, we will restrict our attention to the case of 1 : 5 resonance, namely, q = 5 in Equation (1), and obtain criteria. For higher-order resonance, the procedure is essentially the same as this case but needs tedious computation. This paper is organized as follows. In Section 2, the Poincaré map is established according to power series solution to Equation (1). The map is further transformed into its normal form. In Section 3, the Neimark-Sacker bifurcation is investigated. When there is a circle bifurcating from the fixed point, the phenomenon of phase locking on the invariant circle is studied and the parameter region in which subharmonic motion can occur is determined. In Section 4, choosing a set of parameters, the theoretical results stated above are verified by numerical simulations.
The Numerical Simulations
To illustrate the results stated above, numerical simulations will be presented in this section. As mentioned in the introduction, f is a 2π periodic function. For simplicity, let f ðð5 + δÞtÞ = sin ðð5 + δÞtÞ in Equation (1), which is a simple form of f ðð5 + δÞtÞ. It follows from (6)- (14) that Furthermore, we take the set of parameters a 1 = 2, b = 1, s = 0, c = −4: We can calculate that Re ðα 0 λ 0 Þ = −1:8850 < 0 and the derivation of Re ðλ 1 Þ with respect to parameters μ and ε is nonzero, which means that the supercritical Neimark-Sacker bifurcation takes place for map (25) [17]. 4 Advances in Mathematical Physics
Advances in Mathematical Physics
Choosing ε = −0:3, μ = −0:2, and δ = 0, we get Re ðλ 1 Þ = −0:1919 and then assert that map (25) possesses a stable fixed point (see Figures 2(a) and 2(b)), namely, a stable periodic solution of Equation (1) (shown in Figures 2(c) and 2(d)). If the dynamic behaviors of the model in Figure 1 can be described by Equation (1) with the present parameters, the period of vibration of the mass body is the same as the external force.
Choosing ε = −0:3 and μ = 0:2, we have Re ðλ 1 Þ = 0:3108 and then assert that map (25) possesses a stable invariant circle [17], namely, a stable torus motion of Equation (1). Because the limitation of map (25) on the stable invariant circle is a circle diffeomorphism, the trajectory on the torus is quasiperiodic or subharmonic motion, which depends on expression (24). As will be investigated below.
For δ = 0, By straight computation, we obtain Figures 4(c) and 4(d)). If the dynamic behaviors of the model in Figure 1 can be described by Equation (1) with the present parameters, the vibration of the mass body is subharmonic, whose period is five times than that of the external force.
Because the analysis method of this paper is for system (1) with abstract coefficients, it can be applied in other mechanical models whose dynamics can be described by Equation (1), for example, the forced Van der Pol equations [6], the forced dry friction system [18], the vibration of railway bow net, and the forced vibration of cantilevered flowconveying pipe.
Conclusions
In this paper, we study the Neimark-Sacker bifurcation of a forced vibration system by theoretical analysis and numerical simulations in the 1 : 5 resonance case. The Poincaré map is established by the analytical method. By means of analyzing the map, it is shown that there exist quasiperiodic and subharmonic solutions on the torus. Numerical simulations agreed with the theoretical results. It is certain that the method applied in this paper can be applied to some other analogous systems.
Data Availability
The data used to support the findings of this study are available from the corresponding author upon request.
Conflicts of Interest
The author declares that they have no conflicts of interest. | 1,792 | 2020-09-26T00:00:00.000 | [
"Mathematics"
] |
A Cognitive Linguistic Study of Verb-copying Sentences in Mandarin Chinese
This study concerns the verb-copying structure “S+V+Object+V+Resultative” in Chinese from the perspective of cognitive grammar. It views the construction as composited from component structure “S+V+Object” and “V+Resultative” and reveals the compositional path and mental representation of this construction. The results indicate that component SVO elaborates the schematic trajector of V in Component VR. This conceptual correspondence lays the foundation for the integration of component SVO and VR, which constitutes the internal motivation of SVOVR construction. Component SVO serves as the cognitive reference point for the conceptualization. The elements in the dominion of action that SVO designates are extracted as conceptualization target. SVOVR construction is organized in the pattern of Baseline and Elaboration. In local terms, each stratum provides the potential for the next and the conceptualization result the linear structure is completed in a cumulative fashion. In global terms, component SVO and VR serve as dual baseline whose mutual elaboration yields a composite structure of greater complexity. SVOVR construction imposes some restrictions on the repeated verb. The verb needs to designate the actions that can be repeated and continued. Structure One cannot have any tense or the aspect markers. SVOVR construction does not stand in isolation from other units in linguistic system; it is categorized by the conventionalized SVO structure and predicate-complement VR structure in Chinese language system, which is the external motivation for this construction.
INTRODUCTION
Verb-copying construction (VCC) is typical in Mandarin Chinese. It refers to a grammatical process in which a verb is "copied" after its direct object when in the presence of certain adverbial elements. Thus, we have a construction of the following form:
(subject) verb direct object verb adverbial element
What is labeled direct object can be either an actual direct object or the object component of a verbobject compound. The term adverbial element is meant to be a cover term for four different types of adverbial expressions including quantity adverbial phrase, complex stative construction, locative Xu, Y. (2020). A Cognitive Linguistic Study of Verb-Copying Sentences in Mandrain Chinese. Advancees in Social Sciences Research Journal, 7 (4) 164-176. symbolic structures: semantic, phonological and symbolic. Semantic structures are conceptualizations exploited for linguistic purposes, notably as the meanings of expressions. The phonological structure includes not only sounds but also gestures and orthographic representations. A symbolic structure is bipolar: S is its semantic pole, and P is its phonological pole [22]. There is no strict division of lexicon, morphology and syntax. They form a continuum fully reducible to assemblies of symbolic structures. The central notion of CG is that grammatical structure is inherently symbolic. In CG, grammatical patterns are represented by means of schemas. A construction is defined as either an expression (of any size), or else a schema abstracted from expressions to capture their commonality (at any level of specificity) [22]. Expressions and the patterns they instantiate are all symbolic in nature, which only differ in degree of specificity.
In CG, describing the grammar of a language consists primarily of describing its constructions. There are four basic factors in the description of construction: correspondences, profiling, elaboration and constituency. Of the four descriptive factors to be considered, correspondences are perhaps the most fundamental. A construction is an assembly of symbolic structures (form-meaning pairings) linked by correspondence. They indicate how component and composite structures fit together in a coherent assembly. At the semantic pole, they specify the conceptual overlap between component structures, thus providing the basis for their integration [22]. The phonological pole is mostly ignored, the semantic structures under discussion represent just one pole of symbolic assemblies. It is typical in constructions for the composite semantic structure to profile the same entity as one of the component structures. The component structure that inherits profile to the composite structure is referred to as the profile determinant in CG. It is typical in a construction for one component structure to contain a schematic substructure which other component serves to elaborate, i.e. characterize in finer-grained detail. The schematic element elaborated by another component is called an elaboration site, or e-site for short.
In CG, a full, finite clause profiles a grounded instance of a process type. In the simplest case, this type is directly specified by a lexical verb. The grounded structure consists of more than just the verb. Minimally, it further includes the verb's "arguments": the nominals that specify its profiled participants. Since a process is conceptually dependent on its participants, a verb evokes them schematically as an inherent aspect of its meaning [22]. Thus, the verb is the head and the nominals are complements because they elaborate salient substructures of it. A verb's complements are not limited to the subject and object nominals that specify its focal participant. Often its meaning incorporates additional schematic entities that are sufficiently salient to function as elaboration sites. Among these are participants which happen not to be focused as trajector or landmark [22]. For example, in She sent him flowers, the agent and recipient are of focal prominence, making she the subject and him the object. The mover in this construction is clearly a central participant which is essential to the meaning of send. An elaborating nominal in this case is a complement. A verb's semantic structure can also incorporate a schematic relationship that functions as an e-site. The verb put profiles the action that the trajector moves the landmark and results the latter in a new location. Besides the subject and object, it therefore takes a locative expression as a complement that specifies where the object winds up as in She put the flowers in the vase. CG takes the position that grammar consists in a person's grasp of an inventory of established linguistic conventions which are referred to as "units". Importantly, the inventory is claimed to be structured in the sense that each unit participates in relations of various kinds to many other units. Baseline and elaboration is the latest theoretical development of CG. The notions baseline and elaboration pertain to asymmetries observable in any facet of language structure or its conceptual and phonological basis [23]. B/E organization is utterly ubiquitous in the broadest sense. In one way or another, the baseline (B) is already, in place, or under control. Its elaboration (E) can be characterized abstractly as a function mapping B onto the higher-level structure BE. E is a kind of cognitive processing which is dynamically described as an operation which consists of augmentation, adaptation, or additional processing activity. B has some kind of priority and is generally substantive than E as well. Naturally, an elaborated structure (BE) can function as baseline for further elaboration representing a higher level of B/E organization [23]. The levels are layered and dynamically evolved. Baseline, elaboration and strata are diagramed in Figure 1. The boxes indicate strata (S).
Figure 1. Baseline, elaboration, and strata
The key notion is that a structure does not just spring into existence ex nihilo, but always emerges from a substrate that supports it and creates the potential for its emergence. A given stratum, Si, comprises an array of resources available for structure building; they include both mental capacities and the structures already in place.
COMPOSITE STRUCTURE SVOVR
A defining property of human language is the formation of complex structures out of simpler ones. In CG, Composite structure is a structure that results when two or more structures in a given domain (phonological, semantic, or symbolic) combine in a valence relation. In this study, SVOVR construction is defined as a composite structure of the two components SVO and VR with the same verb. VO and VR are component structures which integrate in a combinatory relationship (particularly a grammatical valence relation). In this section, we will discuss how the component structures VO and VR are integrated and fit together in a coherent way.
SVO component
Component SVO is canonical linguistic pattern in Chinese language. Inside this component structure, the verb's schematic trajector and landmark are elaborated by its focal participants which serve as subject and object of this syntactic structure, in a finer-grained way. In Chinese, verbs do not always appear as a single morpheme; there are a very large number of compound verbs like "paobu" running, "zoulu" walking and "shuijiao" sleeping etc. So, this component structure can be abstracted as SV with only the schematic trajector to be elaborated by the verb's focal participants in some circumstances.
In the composite structure SVOVR, the first component SVO is reified as a nominal expression and designates a specific event; its temporal profile has faded into the background while the second verb in component VR profiles relationship; that is, SVO component is in subordinate position of the composite structure despite the same verb with component VR. In English grammar, verbs should be non-finite, in infinitive or participle form, if they do not serve as predicate verbs in a sentence. The infinitive and participle form have corresponding morphological markers like to-, -ing and -ed. Unlike English, Chinese is not rich in morphological change. In Chinese grammar, a syntactic structure is a phrasal expression if it is contained within a more complex structure; it is realized as a sentence when used alone [9]. Therefore, the profile of Chinese verbal expression, things or relationship, can only be judged from the specific linguistic expressions. In SVOVR construction, the VO component as a whole is nominalized and profile a thing. VO cannot be negated by negative adverbials and reject tense and aspect markers. From the standpoint of usage event, this component is rarely used alone. When the addressee hears bare SVO expressions without any additional elements, they will expect another new information. For example, this bare SVO structure can appear in parallel form like "Wo change, ni tiaowu" I sing, you dance. Shen [12] named this kind of sentence as bound sentence and proposed that there are not only free and bound morphemes but also free and bound syntactic structures in Chinese language. The opposition between free and bound is also the opposition of boundedness and unboundedness in human cognition. The unbounded syntactic structure designates an action but not an event for it has no clear internal starting point and endpoint. In contrast, the bounded syntactic structure designates an event which has explicit starting and end points. Thus, component SVO is unbounded and dependent from the perspective of usage.
VR component
Component VR designates a processual relationship and bequeaths its profile to the composite structure as a whole. The schematic substructure of the verb in this component is not specified by its nominal participants; its schematic trajector is elaborated by component SVO which is nominalized and profiles a thing; its schematic landmark is elaborated by complement R which designates the results of the action designated by SVO. Negative adverbials, tense and aspect markers can be used in this component, which is different with component SVO; it is bounded in cognition. Besides, the semantics of complement R orients to the elements in the frame of SVO. Take a look at the following linguistic instantiations of SVOVR construction:
Wo zuofan zuo huai le guo
My cooking broke the pan.
Wo zoulu zou de jiao qi pao le
Walking made my feet blister.
Wo kanshu kan lei le
Reading made me tired.
Wo jiao meishu jiao chu jingyan le
Teaching art produced me some expertise.
In (1), the "guo" pan in R complement orients to instrument in the action of cooking. In (2), the "jiao" feet are indispensable parts of body involved in the action of walking. In (3), "lei" tired is the description of the physical state of the agent in the action of reading. In (4), the semantics of "jingyan" expertise orients towards the action of teaching itself. The designation of complement R is the resultative state of the elements in the action of component SVO. The action can result in the appearance of a new state or a new thing. In example (1) and (3), the action "zuofan" cooking and Advances in Social Sciences Research Journal (ASSRJ) Vol.8, Issue 4, Apr-2020 "kanshu" reading cause a new condition of the instrument "guo" pan and agent "wo" I respectively; in example (2) and (4), the action "zoulu" walking and "jiao meishu" teaching art cause a new thing appear in the parts of body involved in the action and the action itself.
Compositional path of SVOVR construction
In CG, how an expression's composite meaning relates to those of its components (at successive levels of organization) is called its compositional path [22]. The conceptual correspondence motivates the integration of component SVO and VR. Integration depends on correspondences. For two semantic structures to combine syntagmatically, they must have some point of overlap; more precisely, a substructure of one is placed in correspondence with a substructure of the other, and these two substructures are construed as designating the same conceived entity. It is by virtue of having one or more such entities in common that two component structures can be integrated to form a coherent, more elaborate conceptualization [21].
At the lower level of organization, inside component SVO, the verb's schematic substructure is elaborated by its focal participants which serves as the subject and object of the verb. In SVOVR construction, SVO component as a whole profiles non-processual relationship and is reified as a nominal. The reified VO serves to elaborate the schematic trajector of the verb in VR. The verb in VR is conceptually dependent and profiles temporal relationship. Its schematic landmark is elaborated by R complement which designate the resultative state of the elements in the frame of SVO. The schematic trajector of this verb is elaborated by SVO component which designates the cause of the resultative state. Rather than seeing a composite structure as an edifice constructed out of smaller components, we can treat it as a coherent structure in its own right: component structures are not the building blocks out of which it is assembled, but function instead to motivate various aspects of it [21]. The composite structure inherits the profile of VR. Linguistic phenomena lend themselves more easily to a claim of partial rather than full compositionality [21]. The compositional path of SVOVR construction is demonstrated as Figure 2. Component SVO is reified and profiles complex non-processual relationship. Component VR designates a processual relationship. The schematic TR2 is elaborated by component SVO. The dotted line indicates correspondence and solid arrow leads from e-site to its elaboration. Component SVO designates a relationship which does develop through time but is non-processual by virtue of being viewed holistically, so that its temporal evolution is backgrounded. The bar along the time arrow in component VR indicates that its evolution through time is focused rather than backgrounded. At higher levels of organization, the composite structure inherits the profile of component VR and designates a processual relationship. Component SVO is viewed holistically in the composite structure and profiles a thing which is indicated by a surrounded ellipse. The component structures are not invoked for their own sake, but as "stepping-stones" for purpose of "reaching" the composite conception. The component structures constitute a symbolic assembly of complexity. Within this assembly, component structures serve as stepping-stones for arriving at composite structure, at two levels of organization. The ultimate target, shown at the top, comprises the composite form and meaning of the full expression. These stand in foreground. The path followed in reaching the final composite structure is a secondary but significant aspect of an expression's form and meaning.
The composite structure is not merely the sum of the component structures it is based on. It is an entity in its own right and has emergent properties not inherited or strictly predictable from the components and the correspondences between them. From Figure 3 alone, one could not predict that the composite expression Wo zuofan zuo huai le guo profiles the component VR zuohuai le guo rather than Wo zuofan. These are properties of the expression as a whole, emerging only at the composite expression. As a general matter, component structures should be thought of as resources drawn on-along with others-in arriving at the composite expression. While they motivate the composite structure to varying degrees, they should not be thought of as building blocks that need only be tacked together to form the composite whole. The relation between them is one of partial (rather than full compositionality) [22].
MENTAL REPRESENTATION OF SVOVR
CG emphasizes the psychological reality of linguistic representation. The representation of SVOVR construction fully corresponds to cognitive regularities of human beings. In this construction, component VR is the information focus and the profile determinant. The linguistic expressions are linear structures where the more specific first component provides potential for the second which is the conceptualization target and facilitates its emergence. This production process conforms to how we perform in the natural world. When we want to precisely describe the address of some places to a person, we may refer to another more well-known places or architectures to locate the target for the latter has more cognitive priority. Reference point is best described as the ability to invoke the conception of one entity for purpose of establishing mental contact with another, i.e., to single it out for individual conscious awareness [21]. The entity first invoked is called a reference point, and one accessed via a reference point is referred to as a target. A particular reference point affords potential access to many different targets. The salient reference point will pave the way for us to locate something. From the perspective of human cognitive processing, component VO lays the foundation for the whole conceptualization activity in VOVR construction.
In this construction, SVO has some kind of priority and provide substrate for the whole expression. This substrate allows the formation of more elaborate structures. In local terms, component SVO serves as the baseline which elaborates the trajector of the verb in component VR, forming a higher level B2. This higher level of baseline B2 is further elaborated by a Resultative complement, forming a higher level B3. Each component element is apprehended in relation to the one it directly follows. So moving from one stratum to the next, we encounter structures of greater complexity and conceptual sophistication [23]. SVOVR construction as a whole designates the resultative state of certain events. This cognitive processing, or operation, is completed in such order, which can be diagrammed as Figure 4. SVOVR is a linear structure where VO evokes the baseline, also the reference point, and then the target VR can be accessible. The whole structure is scanned sequentially. The first baseline VO fades into the background after being evoked and the target VR comes to the foreground in a cumulative fashion. The mental representation of the SVOVR structure also corresponds to the properties that we access the target via the reference point. VO works as the reference point via which the conceptualizer establish mental contact with the conceptualization target, that is, the component VR. The reference point needs to be salient enough and thus can provide the substrate for the whole expression. By directing attention to a salient reference point to a salient reference point (R), the conceptualizer can readily access anything in the reference point's dominion (D), one such element being the target (T). This natural and efficient strategy is a basic feature of cognitive processing, evident in numerous aspects of linguistic structure [22]. The essential semantic import resides in the very act of mental scanning: evoking first the reference point and then a target it renders accessible. It is thus inherently and quintessentially dynamic, for how it unfolds through processing time actually constitutes its value (Langacker 2008: 84). Let's take the above sentences as example.
Wo zuofan zuo huai le guo
My cooking broke the pan. 2. Wo zoulu zou de jiao qi pao le Walking made my feet blister.
Wo kanshu kan lei le
Reading made me tired. 4. Wo jiao meishu jiao chu jingyan le Teaching art produced me some expertise.
In Example (1), VO zuofan provides the substrate and works as the baseline upon which the conceptualizer can access the target, i.e. the instrument in the event that VO designates. In (2), VO zoulu is the baseline and the body part of the agent involved in the action is the conceptualization target. In (3), paobu is the baseline and the whole body of the agent is the target for conceptualization. Likewise, jiao meishu is the baseline and the event as a whole is the target in (4). In these sentences, each stratum provides potential for the next, forming expressions with growing complexity.
Moreover, from a global perspective component SVO and VR function as dual baseline whose mutual elaboration yields a composite structure of greater complexity. The direction of composition, proceeding from "lower" to "higher" levels, is nothing other than the priority inherent in B/E organization, where each stratum provides the basis for arriving at the next [23]. The two Advances in Social Sciences Research Journal (ASSRJ) Vol.8, Issue 4, Apr-2020 component structure involves in different ways of operation from the perspective of B/E organization. From "lower" to "higher" level, component SVO involves a process of modification whose profile changes from a processual relationship to a thing; component VR involves an operation of argumentation whose schematic trajector is specified by component SVO.
RESTRICTIONS OF THE VERB IN SVOVR CONSTRUCTION
SVOVR is a constructional schema abstracted from specific language use; it is a composite structure with SVO and VR as the component structure. In this construction, component VO provides the substrate and serves as the baseline based on which the conceptualization activity is completed in a cumulative fashion. Therefore, component VO needs to be substantive and have priority. There are some restrictions for verbs in SVOVR construction: 1. Component VO has more tendency to designate continuable and repeatable activities, which can be attributed to the fact that these activities can stimulate the following cognitive processing unfolded within the processing time. In other words, activities that are transient and unrepeatable like zhayan "blink" and tiaohe "drowning" cannot be trigger for the following conceptualization activity. Thus, verbal expressions that profile such activities have no occurrence in VOVR construction. Example (5) is ungrammatical because tiaohe "drowning" indicates the end of life according to our encyclopaedic knowledge. This event cannot be repeated and trigger the following conceptualization. On the contrary, tiaoshui "diving" is a kind of repeatable physical exercise. It can serve as the trigger for the following conceptualization activity.
Diving makes sb tired.
2. VO cannot be negated by the negative adverbials like bu and mei for the negation means the termination of the following cognitive processing. Component SVO cannot serve as baseline once negated. Take Example (1) as an example, SVOVR will be ungrammatical if VO is negated. However, the negative adverbials can exist in component VR which means that the resultative state does not happen in the event.
3. The aspect markers like zhe/le/guo in Mandarin Chinese have no existence in VO for the same reason as the previous one. VO has been reified as an object and serves as a reference point for the conceptualizer to access the target in its dominion. It rejects any aspect or tense markers of verbal expressions. Verb in VR structure profiles relationship and the VR structure designates the action results of SVO. In other words, despite the fact that there are two identical verbs in SVOVR construction, the two verbs have different meanings because of construal difference. The first verb profiles a kind of thing while the second one profiles a kind of relationship. As a reference point, VO structure will fade into the background and the Xu, Y. (2020 target VR structure will be the figure. The whole composite structure will inherit the profile of VR and VR structure is the profile determinant of the composite structure.
Consequently, the traditionally accepted fact that verb copying is used to resolve the dilemma of object and adverbial fighting for the same verb is problematic from the standpoint of CG which claims that meaning is conceptualization; an expression's meaning is not just the conceptual content it evokes-equally important is how that content is construed [22]. Therefore, the two verbs in SVOVR construction have the same form with nonetheless different meanings; they function differently in the composite structure. Component SVO is construed as a thing and serves as a reference point while Component VR is target for conceptualization and profile determinant of the composite structure. In this construction, the repeated verb does not have the same meaning although it evokes the same conceptual content for the same verb is construed differently in the construction as a whole. In this sense, verb-copying is not for resolving the dilemma of object and adverbial competing for verb but a step-by-step statement of certain event and its result. On the other hand, the theory of competition is groundless due to the fact that in Chinese the first component can be Subject + Verb + Object and Subject + intransitive compound Verb as well. In the second case, it is impossible for object and adverbial competing for the same verb for there is no object.
MOTIVATIONS OF SVOVR CONSTRUCTION
According to CG, the grammar of a language can be characterized as a structured inventory of conventional linguistic units. The inventory is said to be structured in the sense that the units do not constitute encapsulated chunks of information; on the contrary, each unit stands at the hub of a network of relations to other units [20]. The inventory is structured in a sense that each established conventions participates in relations of various kinds to others. Doing CG consists, very largely, in elucidating these relations. It is these relations which, cumulatively, motivate a linguistic structure, in that they create a "niche" for the structure within the larger language system [20]. In other words, construction is unlikely to be represented in human mind without internal and external sanctioning conditions; these conditions motivate constructional representation both internally and externally. Therefore, SVOVR construction is not isolated from other facts about a language; its representation has both internal and external motivations. The conceptual overlap between the component structures, that is, the conceptual basis for the integration, is the internal motivation of the construction, which has been elaborated in Section 3. The external motivation gets its source from some established conventions in human's language system. In SVOVR construction, the two component structures are categorized by the established SVO and predicate-complement structures VR in Chinese language system. The two structures are schemas abstracted from linguistic expressions of various kinds, which have therefore been conferred on unit status in language system. Thus, SVOVR construction is more highly motivated than others due to the fact that its position in the language is supported by two highly established units.
CONCLUSION
This study unearths the cognitive mechanism of verb-copying sentences in Mandarin Chinese from the standpoint of CG; it investigates how the syntactic construction SVOVR is formed by the integration of component structure SVO and component VR. In this construction, SVO changes its profile from a processual relationship to a thing; the SVO structure elaborates the schematic trajector of structure VR in a specific and detailed way, which facilitates the integration of the two components into a composite structure. Component SVO designates the cause and component VR Advances in Social Sciences Research Journal (ASSRJ) Vol.8, Issue 4, Apr-2020 the effects; the latter component serves as the profile determinant of the composite structure. SVOVR construction involves both a serial and hierarchical B/E organization. Locally, each stratum provides potential for the next and thus forms the linear structure of growing complexity. In global terms, the two component structures serve as dual baselines whose mutual elaboration forms the composite whole. Conceptual correspondence of the component structures constitutes the internal motivation of the SVOVR construction; SVOVR construction is categorized by the conventionalized SVO and predicate-complement VR in Chinese language system, which forms the external motivation of SVOVR construction. This study has a different take on the traditional view that verbcopying serves to resolve the dilemma of object and adverbial competing for the same verb; it is an ordinal statement of an action and its result. This study contributes to the growing body of research on Chinese syntax and elucidates the relations each linguistic unit participate in with others. | 6,594 | 2020-04-24T00:00:00.000 | [
"Linguistics"
] |
Porphyromonas gingivalis Fimbriae Induce Osteoclastogenesis via Toll-like Receptors in RAW264 Cells
The effect of Mfa1 fimbriae of Porphyromonas gingivalis on the progression of bone resorption remains unclear, especially compared with another fimbriae, FimA. We investigated the effect of Mfa1 on osteoclastogenesis together with FimA. We also investigated the role of Toll-like receptors (TLRs) in Mfa1 recognition during osteoclast differentiation. Receptor activator of nuclear factor κβ ligand (RANKL)-prestimulated RAW264 cells were used to examine the effects of purified Mfa1 fimbriae. The number of osteoclasts was examined by tartrate-resistant acid phosphate (TRAP) staining, osteoclast activation was investigated by bone resorption assays, and gene expression of differentiation markers was examined by quantitative real-time PCR. Transfection of Tlr2 and Tlr4 siRNAs into RAW264 cells was also employed and their role in Mfa1 recognition was investigated. Mfa1 effectively induced the formation of TRAP-positive multinucleated cells and activated osteoclasts. Mfa1 also increased gene expression of Acp5, Mmp9, and Ctsk in RANKL-prestimulated RAW264 cells compared with the control. The osteoclastogenesis induced by Mfa1 was significantly decreased in cells transfected with Tlr2 or Tlr4 siRNAs compared with control siRNA. Our results revealed the role of Mfa1 fimbriae in osteoclastogenesis that may contribute to the partial elucidation of the mechanisms of periodontal disease progression and the development of new therapeutic strategies.
Introduction
Periodontitis is a chronic inflammatory disease that is mainly caused by three species of bacteria called the Red complex (Porphyromonas gingivalis, Tannerella forsythia, and Treponema denticola) [1]. P. gingivalis is a gram-negative, biased anaerobic rod considered to be a keystone bacterium in the etiology of periodontal disease caused by multiple bacteria. In general, a significant infectious capacity is required for periodontopathic bacteria to destroy periodontal tissues, including attachment to periodontal tissue, invasion and disturbance of the host immune response, direct destruction, and escape from host immunity. The major virulence factors of P. gingivalis include fimbriae, lipopolysaccharide (LPS), and gingipain [2].
P. gingivalis has two types of fimbriae, long-type FimA fimbriae and short-type Mfa1 fimbriae, which are proteinaceous filamentous appendages [3]. They protrude from the bacterial cell surface and are thought to play important roles in attachment to the host tissue, biofilm formation, and coaggregation with streptococci and dendritic cells [4]. The fimA gene encodes the fimbrial major protein FimA, and the genotypes of FimA are I-V and Ib [5]. Genotype II is 2 of 12 often detected in patients with severe periodontitis, whereas genotype I is typically detected in patients with mild periodontitis [5][6][7]. Mfa1 fimbriae consist of five proteins (Mfa1-5) with Mfa1 being the major subunit, which polymerizes on the fimbria axis [8].
FimA stimulates macrophages, gingival epithelial cells, and gingival fibroblasts to produce proinflammatory cytokines, such as interleukin (IL)-1, tumor necrosis factor (TNF)α, and IL-6, which promote osteoclast differentiation and alveolar bone resorption [9][10][11]. P. gingivalis ATCC 33277 (wildtype) induces alveolar bone resorption in rats, and the fimAdeficient strain causes more bone resorption than the mfa1-deficient strain [12]. These findings suggest that Mfa1 has a more substantial effect on alveolar bone resorption than FimA. Additionally, fimA and mfa1 double-deficient strains completely lose their ability to adhere to host cells. These results also suggest that both FimA and Mfa1 are important for P. gingivalis virulence.
Toll-like receptors (TLRs) on the cell surface recognize pathogen-associated molecular patterns. The host innate immune response to pathogens is mediated primarily by TLR signaling [13]. TLR2 and TLR4 are the most widely investigated extracellular innate immune receptors that recognize a variety of pathogen-associated molecular patterns and are closely related to the pathogenesis of periodontal disease [14]. In particular, TLR2 is important for P. gingivalis to produce proinflammatory cytokines [15,16]. It has been suggested that TLR2 and TLR4 may be involved in the recognition of P. gingivalis fimbriae [17,18], but a consensus has not been reached so far.
In the present study, we investigated the effects of P. gingivalis fimbriae, with particular attention to Mfa1 fimbriae, on the differentiation and activation of osteoclasts, which cause bone destruction of periodontal tissue compared with FimA fimbriae. Additionally, we examined the effects of TLR2 and TLR4 knockdown on osteoclast differentiation after fimbria stimulation.
Mfa1 and FimA Fimbriae Promote RANKL-Mediated Osteoclast Differentiation
RANKL-treated RAW264 cells and RANKL and M-CSF-treated mouse osteoclast precursor cells were stimulated with various concentrations of Mfa1 or FimA fimbriae to determine whether they affect RANKL-dependent osteoclastogenesis. Mfa1 and FimA stimulation of both cell types induced the formation of TRAP-positive multinucleated cells compared with controls in a dose-dependent manner ( Figure 1A,B). Mfa1 had a significantly higher ability to induce osteoclast differentiation compared with 1 and 10 µg/mL FimA ( Figure 1A,B).
Mfa1 and FimA Fimbriae Promote RANKL-Induced Osteoclastic Bone Resorption
To evaluate the effect of fimbriae on osteoclast activation, RAW264 cells were plated on bone slices and hydroxyapatite (HA) mineral surfaces and cultured for 120 h in the presence of Mfa1 or FimA fimbriae after RANKL prestimulation. The bone slice surface was partially resorbed by osteoclasts derived from RAW264 cells treated with RANKL and Mfa1 or FimA ( Figure 2A). Interestingly, both Mfa1 and FimA significantly promoted bone pit formation compared with the control ( Figure 2B). The osteoassay surface was also partially resorbed by osteoclasts derived from RAW264 cells treated with RANKL and Mfa1 or FimA ( Figure 2C). Mfa1 significantly promoted HA pit formation compared with the control ( Figure 2D). However, no significant pit formation areas were observed in the FimA group compared with the control ( Figure 2D). Mouse osteoclast precursor cells were also plated on bone slices and cultured for 120 h in the presence of Mfa1 or FimA fimbriae after RANKL and M-CSF prestimulation. The bone slice surface was partially resorbed by osteoclasts derived from osteoclast precursor cells treated with RANKL and Mfa1 or FimA ( Figure 2E). Interestingly, both Mfa1 and FimA significantly promoted bone pit formation compared with the control ( Figure 2F).
In the present study, we investigated the effects of P. gingivalis fimbriae, with particular attention to Mfa1 fimbriae, on the differentiation and activation of osteoclasts, which cause bone destruction of periodontal tissue compared with FimA fimbriae. Additionally, we examined the effects of TLR2 and TLR4 knockdown on osteoclast differentiation after fimbria stimulation.
Mfa1 and FimA Fimbriae Promote RANKL-Mediated Osteoclast Differentiation
RANKL-treated RAW264 cells and RANKL and M-CSF-treated mouse osteoclast precursor cells were stimulated with various concentrations of Mfa1 or FimA fimbriae to determine whether they affect RANKL-dependent osteoclastogenesis. Mfa1 and FimA stimulation of both cell types induced the formation of TRAP-positive multinucleated cells compared with controls in a dose-dependent manner ( Figure 1A,B). Mfa1 had a significantly higher ability to induce osteoclast differentiation compared with 1 and 10 μg/mL FimA ( Figure 1A,B). Mouse bone marrow-derived osteoclast precursor cells were prestimulated with 50 ng/mL RANKL and 25 ng/mL M-CSF for 24 h and then stimulated with 10 ng/mL, 100 ng/mL, 1 µg/mL, and 10 µg/mL Mfa1 and FimA fimbriae or 50 ng/mL RANKL every 48 h. After 96 h, the number of TRAP-positive multinucleated cells was counted. Differences between groups were analyzed by ANOVA and Tukey's test. Data are expressed as the mean ± SD (n = 3). * p < 0.05, ** p < 0.01, **** p < 0.0001.
Mfa1 and FimA Fimbriae Increase Gene Expression of TLR2 and TLR4 in RANKL-Induced Osteoclasts
To examine whether fimbriae affect the expression of TLRs, RANKL-prestimulated RAW264 cells were stimulated with fimbriae or RANKL and then mRNA expression of TLR2 (Tlr2) and TLR4 (Tlr4) was investigated. As a result, Mfa1 and FimA significantly increased the expression of Tlr2 and Tlr4 in RANKL-induced osteoclasts ( Figure 4).
Transfection of Tlr2 and Tlr4 siRNAs into RAW264 Cells
To determine whether TLRs on osteoclasts recognized Mfa1 and FimA, we examined the effects of silencing TLR2 and TLR4 expression. Tlr2 siRNA and Tlr4 siRNA-transfected RAW264 cells showed clear knockdown of Tlr2 and Tlr4 mRNAs, respectively, compared with control siRNA-transfected RAW264 cells (see Supplementary Figure S1A). Flow cytometric analysis also showed that surface expression of TLR2 and TLR4 was suppressed in siRNAtransfected RAW264 cells compared with control cells (see Supplementary Figure S1B).
Mfa1 Fimbriae Induce Osteoclast Differentiation Primarily through Recognition by TLR2
RANKL-prestimulated RAW264 cells with suppressed expression of TLR2 and TLR4 were stimulated by fimbriae and then osteoclast differentiation was examined by TRAP staining. Notably, among Tlr2 siRNA-transfected cells stimulated by Mfa1, the number of TRAP-positive cells was markedly decreased compared with control siRNA-transfected cells ( Figure 5). Additionally, suppression of Tlr4 partially but significantly weakened the effect of Mfa1 on osteoclast differentiation ( Figure 5). In the case of FimA stimulation, suppression of Tlr2 and Tlr4 resulted in significantly weakened osteoclast differentiation ( Figure 5). ever, no significant pit formation areas were observed in the FimA group compared with the control ( Figure 2D). Mouse osteoclast precursor cells were also plated on bone slices and cultured for 120 h in the presence of Mfa1 or FimA fimbriae after RANKL and M-CSF prestimulation. The bone slice surface was partially resorbed by osteoclasts derived from osteoclast precursor cells treated with RANKL and Mfa1 or FimA ( Figure 2E). Interestingly, both Mfa1 and FimA significantly promoted bone pit formation compared with the control ( Figure 2F). . Differences between groups were analyzed by ANOVA and Tukey's test. Data are expressed as the mean ± SD (n = 3). * p < 0.05, *** p < 0.001, **** p < 0.0001. (E) Mouse bone marrow-derived osteoclast precursor cells were prestimulated with 50 ng/mL RANKL and 25 ng/mL M-CSF for 24 h and then stimulated with 1 µg/mL Mfa1 and FimA fimbriae or 50 ng/mL RANKL every 48 h. After 120 h, images of the bone slice were obtained. Representative images are shown for each group. White and black bars indicate 50 µm widths. (F) Average area of pits. Differences between groups were analyzed by ANOVA and Tukey's test. Data are expressed as the mean ± SD (n = 3). * p < 0.05, *** p < 0.001, **** p < 0.0001.
Mfa1 Fimbriae Induce Osteoclast Differentiation Markers Primarily through Recognition by TLRs
RANKL-prestimulated RAW264 cells with suppressed expression of TLR2 and TLR4 were stimulated by fimbriae and then osteoclast differentiation marker expression was examined by qPCR. Expression of osteoclast differentiation markers Acp5, Mmp9, and Ctsk after Mfa1 stimulation was significantly decreased in Tlr2 siRNA and Tlr4 siRNAtransfected cells compared with control siRNA-transfected cells ( Figure 6).
Mfa1 and FimA Fimbriae Increase Gene Expression of TLR2 and TLR4 in RANKL-Induced Osteoclasts
To examine whether fimbriae affect the expression of TLRs, RANKL-prestimulated RAW264 cells were stimulated with fimbriae or RANKL and then mRNA expression of
Transfection of Tlr2 and Tlr4 siRNAs into RAW264 Cells
To determine whether TLRs on osteoclasts recognized Mfa1 and FimA, we exami the effects of silencing TLR2 and TLR4 expression. Tlr2 siRNA and Tlr4 siRNA-transfec RAW264 cells showed clear knockdown of Tlr2 and Tlr4 mRNAs, respectively, compa with control siRNA-transfected RAW264 cells (see Supplementary Figure S1A). Flow staining. Notably, among Tlr2 siRNA-transfected cells stimulated by Mfa1, the number of TRAP-positive cells was markedly decreased compared with control siRNA-transfected cells ( Figure 5). Additionally, suppression of Tlr4 partially but significantly weakened the effect of Mfa1 on osteoclast differentiation ( Figure 5). In the case of FimA stimulation, suppression of Tlr2 and Tlr4 resulted in significantly weakened osteoclast differentiation ( Figure 5). multinucleated cells was counted after 96 h. Differences between groups were analyzed by ANOVA and Tukey's test. Data are expressed as the mean ± SD (n = 4). *** p < 0.001, **** p < 0.0001.
Mfa1 Fimbriae Induce Osteoclast Differentiation Markers Primarily through Recognition by TLRs
RANKL-prestimulated RAW264 cells with suppressed expression of TLR2 and TLR4 were stimulated by fimbriae and then osteoclast differentiation marker expression was examined by qPCR. Expression of osteoclast differentiation markers Acp5, Mmp9, and Ctsk after Mfa1 stimulation was significantly decreased in Tlr2 siRNA and Tlr4 siRNA-transfected cells compared with control siRNA-transfected cells ( Figure 6). Values are expressed as fold changes. Differences between groups were analyzed by ANOVA and Tukey's test. Data represent the mean ± SD (n = 3). ** p < 0.01, *** p < 0.001, **** p < 0.0001.
Discussion
In the present study, we demonstrated that P. gingivalis Mfa1 and FimA fimbriae promoted osteoclast differentiation and activation. Notably, the effects of Mfa1 on osteoclast differentiation appeared to be stronger than those of FimA. Hiramine et al. reported that the 67-kDa fimbriae (corresponding to Mfa1 fimbriae) of P. gingivalis induce osteoclast activation [19]. In their study, they used mouse primary bone marrow cells and stromal cell lines to confirm resorption pit formation in dentin slices under treatment with macrophage colony-stimulating factor (M-CSF), RANKL, dexamethasone, and 1α,25(OH) 2 D 3 . Additionally, bone marrow cells and stromal cells were cocultured and many osteoclastogenesisinducing factors were involved in their study. The findings suggested that Mfa1 fimbriae may promote osteoclastogenesis by stimulating receptors on stromal cells and increasing RANKL production. In the current study, we consider this indirect osteoclastogenesis to be a major difference because we examined direct effects on osteoclast progenitor cells. FimA fimbriae also stimulate bone resorption activity of calvarial bone cells on a bovine bone slice [20]. However, no direct effects of FimA on osteoclast differentiation or activation have been reported to our knowledge. Therefore, this is the first report on the direct effects of FimA and Mfa1 fimbriae on osteoclastogenesis. We observed no increase in Nfatc1 expression induced by fimbriae, which may be due to the stimulation time (48 h). In fact, we observed an increase in Nfatc1 expression at 8 and 24 h of stimulation with fimbriae (Supplementary Figure S2). In particular, the absorptive capacity of FimA-stimulated cells for minerals was not increased in a statistically significant manner unlike on bone slices. These data suggest that the ability of FimA to generate hydrogen ions via the ATP6i complex from osteoclasts is not as high as that of Mfa1. The differences in the inducibility of osteoclast differentiation and activation by various P. gingivalis fimbriae need to be investigated further. Recently, FimA and Mfa1 fimbriae of P. gingivalis were classified as novel V-type fimbriae [21], and their detailed structure, including trace components such as Mfa2-5, has been identified [4]. In studies investigating the relationship between bone resorption and P. gingivalis fimbriae, such fimbrial structures have not been considered. Future studies should focus on the detailed fimbrial structure to further explore the direct mechanism of osteoclastogenesis.
The proinflammatory mediator IL-1 has been suggested to directly promote osteoclast activation [22]. TNF-α has also been suggested to directly promote osteoclast differentiation [23]. LPS stimulates the survival and fusion of osteoclasts independently of RANKL, IL-1, and TNF-α [24]. LPS also induces various cytokines and mediators, such as IL-1, TNF-α, and prostaglandin E2, which play crucial roles in osteoclast differentiation and activation [25]. LPS appears to affect osteoclastogenesis in a complicated manner. Simultaneous stimulation by RANKL and LPS or staphylococcal lipoteichoic acid inhibits osteoclast differentiation induced by RANKL [26,27], whereas stimulation of osteoclast progenitors by LPS appears to promote osteoclastogenesis as well as osteoclast survival and activation [26,28,29]. Our preliminary results indicated that RANKL-unstimulated RAW264 cells, with a macrophage-like state, produced various proinflammatory cytokines upon the addition of Mfa1, and stimulation by Mfa1 alone did not differentiate macrophages into osteoclasts. Indeed, Hamada et al. reported that Mfa1 increases the expression of IL-1, IL-6, and TNF-α in mouse peritoneal macrophages [30]. Therefore, the effects of Mfa1 on osteoclast differentiation and activation in the present study were considered to occur in RANKL-induced osteoclast precursor cells. RANKL-primed macrophages promote osteoclastogenesis in a TNF-α-independent manner by P. gingivalis [31]. It has been reported that 41-kDa fimbriae of P. gulae, a periodontopathogenic bacterium in dogs, promote osteoclast differentiation upon costimulation with RANKL and 1α,25(OH) 2 D 3 [32]. In actual bone resorption, many bone resorption-promoting factors would be produced in the surroundings, and the action of Mfa1 would be further enhanced.
TLRs recognize many bacterial components and play a decisive role in innate immunity [13]. Although E. coli LPS is recognized by TLR4 [13], P. gingivalis LPS is reported to be recognized by both TLR2 and TLR4 [33], suggesting that the biological response to the bacterial components of P. gingivalis is more complex. Recombinant FimA enhances inflammatory mediator production in human peripheral blood monocytic cells via TLR4 [34]. Moreover, FimA-like lipoproteins or lipopeptides related to FimA have been suggested to induce, at least in part, TLR2-mediated signaling and subsequent TNF-α production in macrophages [35]. In terms of Mfa1, the recognition receptor remains controversial. We recently reported that TLR4 is mainly important for the recognition of Mfa1 by gingival fibroblasts [17]. Takahashi et al. reported that TLR2 may be important for Mfa1 recognition by bronchial epithelial cells [18]. Our results suggested that TLR2, along with TLR4, may be important for the recognition of Mfa1 and FimA by osteoclast progenitor cells. Indeed, 67-kDa fimbriae of P. gingivalis induce osteoclast activation and its effect is attenuated by TLR2-neutralizing antibodies [19]. Future analysis using knockout mice or the CRISPR-Cas9 system should clarify the recognition receptors of Mfa1 and FimA.
This study has several limitations. First, because this was a cell-based in vitro study, future in vivo studies using infection model animals with an emphasis on analysis of alveolar bone resorption using a fimbria mutant strain of P. gingivalis should be conducted to clarify the role of Mfa1 in periodontitis. Second, osteoclasts are also induced to differentiate and activate by proinflammatory mediators derived from osteoblasts, macrophages, gingival fibroblasts, and epithelial cells. Therefore, it is necessary to study the effect of multiple cell types on the intercellular network of Mfa1. Third, this study did not examine the intracellular signaling of TLRs, the putative receptors of Mfa1. Such studies may determine which intracellular signaling pathways of TLRs regulate osteoclast differentiation and activation induced by Mfa1.
Cell Culture
The mouse macrophage cell line RAW264 was purchased from the RIKEN Cell Bank (Ibaraki, Japan). Mouse bone marrow-derived osteoclast precursor cells were purchased from COSMO BIO (Tokyo, Japan). Both cell types were cultured in α-minimum essential medium (MEM) (Thermo Fisher Scientific, Wilmington, DE, USA) containing 10% fetal bovine serum (MP Biomedicals, Santa Ana, CA, USA), 100 U/mL penicillin, and 100 µg/mL streptomycin in a 5% CO 2 incubator at 37 • C.
Purification of Fimbriae
Purification of Mfa1 fimbriae from JI-1 (fimA deleted) was performed using a standard protocol [36]. Briefly, P. gingivalis cells disrupted in a French pressure cell (OHTAKEWORKS, Osaka, Japan) were ultracentrifuged, and then the supernatant was precipitated with ammonium sulfate (50% saturation). The Mfa1 fimbrial fraction was separated by ion exchange chromatography (DEAE Sepharose Fast Flow chromatography, GE Healthcare Bio-Sciences AB, Uppsala, Sweden). The purity and identity of Mfa1 fimbriae were verified by sodium dodecyl sulfate-polyacrylamide gel electrophoresis and transmission electron microscopy. FimA fimbriae from SMF-1 (mfa1 deleted) were purified in accordance with the protocol by Yoshimura et al. [40].
Tartrate-Resistant Acid Phosphatase (TRAP) Staining
RAW264 cells were prestimulated with 50 ng/mL receptor activator of nuclear factor κβ ligand (RANKL) (PeproTech, Rocky Hill, NJ, USA) for 24 h. Mouse bone marrowderived osteoclast precursor cells were prestimulated with 50 ng/mL RANKL and 25 ng/mL macrophage colony-stimulating factor (M-CSF) (Sigma-Aldrich) for 24 h. Then, the cells were stimulated with 10 ng/mL, 100 ng/mL, 1 µg/mL, and 10 µg/mL Mfa1 and FimA fimbriae or 50 ng/mL RANKL every 48 h. After 96 h, the cells were washed with phosphate-buffered saline and fixed in 3.3% formaldehyde, a citric acid solution, and acetone for 5 min. The staining solution was prepared by mixing a Fastgarnet GBc BASE solution, sodium nitrite solution, Naphthol AS-BIPb, acetic acid solution, tartaric acid solution (all purchased from Sigma-Aldrich), and distilled water. After fixation, cells were washed with distilled water and stained in the solution for 30 min. After the cells were washed with distilled water and dried, TRAP-positive multinucleated cells containing three or more nuclei were counted under an optical microscope (BZ-X700, KEYENCE, Osaka, Japan).
Bone Resorption Assay
After washing a bone slice (BioVendor R&D, Brno, Czech Republic) and bone resorption activity evaluation plate (Osteo Assay Stripwell plate, Corning Lifesciences, Corning, NY, USA) with α-MEM, seeded RAW264 cells were prestimulated with 50 ng/mL RANKL for 24 h and mouse bone marrow-derived osteoclast precursor cells were prestimulated with 50 ng/mL RANKL and 25 ng/mL M-CSF for 24 h. Then, every 48 h, the cells were stimulated with 1 µg/mL Mfa1 and FimA fimbriae or 50 ng/mL RANKL. After 120 h, the cells were lysed in 5% sodium hypochlorite for 5 min. After washing with distilled water and drying, the absorption pit area was measured under the optical microscope. Three areas where pits had formed were randomly selected and the total area was measured using a BZ-X Analyzer (KEYENCE).
Real-Time qPCR
RAW264 cells were prestimulated with 50 ng/mL RANKL for 24 h. Then, the cells were stimulated with 1 µg/mL Mfa1 and FimA fimbriae or 50 ng/mL RANKL for 48 h. Total RNA was then extracted using NucleoSpin RNA (Macherey-Nagel Inc, Bethlehem, PA, USA) in accordance with the manufacturer's protocol. The sample concentration was measured by a Thermo NANO DROP LITE (Thermo Fisher Scientific) and underwent cDNA conversion using a Biosystems GeneAmp PCR System (Thermo Fisher Scientific). Conditions were 37 • C for 15 min, 50 • C for 5 min, 98 • C for 5 min, and hold at 4 • C. Then, to quantify mRNA expression, real-time qPCR was performed using Taqman gene expression assays (Thermo Fisher Scientific) for mouse Acp5 (Trap) (Mm00437135-m1), Mmp9 (Mm00442991-m1), Ctsk (Mm00484039-m1), and Nfatc1 (Mm00479445-m1) with TaqMan Universal PCR Master Mix (Thermo Fisher Scientific). mRNA levels were normalized to eukaryotic 18S rRNA (Hs99999901_s1). qPCR was performed using a StepOnePlus™ Real-Time System (Thermo Fisher Scientific). The thermocycling conditions were 40 cycles of 10 min at 95 • C, followed by 40 cycles of 15 sec at 95 • C and 1 min at 60 • C. Relative changes in gene expression were calculated using the 2 −∆∆Ct method. 18S rRNA (Hs99999901-s1) was used as an internal control.
siRNA Transfection
RAW264 cells at 60-80% confluence were transfected with siRNAs targeting Tlr2 and Tlr4 (Silencer Select Pre-designed siRNAs, Ambion, Austin, TX, USA) or non-targeting control siRNA using Lipofectamine RNAiMAX (Thermo Fisher Scientific) in Opti-MEM (Thermo Fisher Scientific). After 24 h, cells were collected, prestimulated with RANKL for 24 h, and then divided into two experimental systems. First, the cells were stimulated with 1 µg/mL Mfa1 and FimA fimbriae or RANKL for 48 h and then collected to analyze gene expression by qPCR. Second, cells were stimulated with 1 µg/mL Mfa1 and FimA fimbriae or 50 ng/mL RANKL every 48 h. After 96 h, the cells were stained with TRAP.
Statistical Analysis
Statistical analysis was performed using PASW Statistics version 18.0 (SPSS Japan, Tokyo, Japan). Results were compared by analysis of variance (ANOVA) and Tukey's multiple comparisons test. Comparisons between two independent groups were made using Student's t-test. Data are expressed as the mean ± standard deviation (SD). Significant differences were accepted at less than 0.05.
Conclusions
Our findings demonstrate that Mfa1 fimbriae directly promote osteoclast differentiation and activation. Recognition of Mfa1 by TLRs on osteoclasts is important to facilitate osteoclastogenesis. Further studies focusing on the intracellular signaling of TLRs are necessary to reveal the underlying mechanism and develop therapeutic strategies for periodontal disease. | 5,404 | 2022-12-01T00:00:00.000 | [
"Biology"
] |
A Bayesian Inference Framework for Gamma-Ray Burst Afterglow Properties
In the field of multi-messenger astronomy, Bayesian inference is commonly adopted to compare the compatibility of models given the observed data. However, to describe a physical system like neutron star mergers and their associated gamma-ray burst (GRB) events, usually more than ten physical parameters are incorporated in the model. With such a complex model, likelihood evaluation for each Monte Carlo sampling point becomes a massive task and requires a significant amount of computational power. In this work, we perform quick parameter estimation on simulated GRB X-ray light curves using an interpolated physical GRB model. This is achieved by generating a grid of GRB afterglow light curves across the parameter space and replacing the likelihood with a simple interpolation function in the high-dimensional grid that stores all light curves. This framework, compared to the original method, leads to a $\sim$90$\times$ speedup per likelihood estimation. It will allow us to explore different jet models and enable fast model comparison in the future.
Introduction
The gravitational wave emitted from a merging binary neutron star, GW170817 [1], followed swiftly by a Gamma-ray burst (GRB) of short duration known as GRB170817A [2,3] commenced the era of multi-messenger astrophysics (MMA). This is the first direct evidence showing neutron star mergers as the progenitor of short GRBs [4]. The union between electromagnetic (EM) and gravitational wave (GW) observations allow for astrophysical objects to be investigated with alternative probes, enabling the source properties to be studied under a new light.
The coalescence of compact objects with at least one neutron star is of particular interest. By the end of the third LIGO observing run (O3), one more binary neutron star (BNS) event, GW190425 [5], and two NS-BH mergers are reported [6]. However, GW170817 remains the only event with a confirmed GW-EM detection and has allowed a major leap in terms of its scientific implications.
A broadband search of the EM counterpart of GW170817 shows both thermal kilonova emission (AT2017gfo) generated from the radioactive decay of newly synthesized elements in the ejecta and a non-thermal synchrotron component from the relativistic jet [7]. The latter is the GRB afterglow that typically shines in X-ray, optical, and radio wavelengths and can last for months to years following the GRB. For the case of GRB170817A, the first confirmed X-ray counterpart was observed by the Chandra X-ray Observatory and announced at around 9 days post-merger [8]. Late-time monitoring of this event unveils the X-ray afterglow emission reaching its peak luminosity at around 155 days [9,10] and continues to shine at more than 1000 days after the GRB [11][12][13].
Modeling the afterglow light curves would shed lights on the physical environment of the system, as the afterglow peak time is dependent on the observers' viewing angle and the angular profile of the jet. Studies of the temporal behavior of GRB170817A afterglow suggest a structured jet seen off-axis, see, e.g., in [13][14][15][16][17][18]. Understanding of the jet structure is essential to explain the observations of off-axis BNS events such as GRB170817A and uncovering the underlying physical mechanisms behind their production. With better modelling of the jet structure, we are able to more confidently infer the rate of GRB production [19], and the inference of the viewing angle of individual MMA events improves, allowing for tighter constraints on inference of the Hubble constant [20,21].
In this paper, we describe a semi-analytical GRB afterglow model that takes into account the effects of lateral spreading in Section 2. Based on this model, we develop a Bayesian framework that would allow quick parameter estimation and explain it in Section 3. We present the results with this methodology and discuss its future application in Section 4.
GRB Afterglow
A broadband afterglow lasting hours-to-days typically follows a GRB, where this GRB afterglow is produced in the shock system that forms as the ultra-relativistic jet that emitted the GRB decelerates. Using energy conservation for a spherical blastwave, the dynamical behaviour of the decelerating system can be modeled, and the change in Lorentz factor, dΓ, with the change in swept-up mass found, see, e.g., in [22]. The instantaneous Γ and sweptup mass of the blastwave can be used to find the synchrotron emission responsible for the broadband GRB afterglow. By following the method in [23], a single jetted outflow with a half opening angle, θ j , is split into multiple emission components and the dynamics and synchrotron flux, relative to the observers line-of-sight from each segment, can be found. Summing across the multiple components at a given observer time, t, is equivalent to integrating across the equal arrival time surface for the jet, and by allowing each component to have a unique energy and Γ, then the afterglow from complex jet structures can be found, see, e.g., in [24].
This method naturally accounts for the edge-effect in a GRB afterglow, where the jet edge becomes visible for an on-axis observer when Γ(t) < 1/θ j , and the resulting jet-break in the lightcurve. We further add the effects of lateral spreading in the jet by considering the maximum sound speed of each component and the change in radius due to this lateral spreading (see in [23,25]). With this method, the afterglow from spreading and non-spreading jets, with a defined structure profile can be found for observers at any viewing angle i.e., within the jet opening angle for cosmological GRBs, see, e.g., in [26], and at higher viewing angles for gravitational wave counterparts to neutron star mergers, see e.g., in [15,16].
The semi-analytic approach used in generating the lightcurves divides the jet emission area over its radial and angular components into a grid of a given resolution. The dynamics and flux are solved for each resolution element within the jet individually before summing to give the resultant lightcurve. The duration of each afterglow iteration depends on the resolution, where for off-axis events the resolution can be set to low values >50, while for events with low viewing angle, wide opening angles or high Γ, the required resolution becomes much higher, and typically >10,000 to maintain the same level of accuracy. Fitting algorithms can require 100s of thousands of individual model lightcurves for a reliable fit. In our test model with four free parameters, it requires at least ≥1 ×10 4 iterations for a single analysis. A model that takes several seconds, or even minutes, per iteration is therefore not ideal.
Jet Structuring
For a power-law jet, the energy and Lorentz factor distribution of the ejecta with respect to jet central axis is scaled as In this scenario, the jet is parameterized by two characteristic angles, θ c and θ j . Starting from the center of the jet, the energy is uniformly distributed within the core, θ c . Outside of θ c , the energy distribution follows a power-law decay with a sharp decline at the edge, θ j . This structure was proposed to account for the observed power-law decay in GRB afterglow light curves. Here, we assume the power-law index k = 2 as used in Lamb and Kobayashi [24].
Parameter Estimation
The model described in Section 2 is used to simulate X-ray afterglow light curves from 0.15 to 1000 days after the GRB event. Based on the X-ray fluxes, we generate random noise assuming a minimum signal-to-nose ratio (SNR) of 3. The injection data are the sum of both the signal and the noise components. We perform the parameter estimation using Bayesian inference, from which the posterior distribution of jet parameters is given by, according to Bayes theorem, where L(d|θ) is the likelihood function of the data given a set of model parameters θ, and π(θ) is the prior distribution for θ and Z is the evidence, or the observed data. A nested sampling algorithm is used to calculate the evidence and the posterior densities. We carry out the analysis with Bilby, a Python-based Bayesian inference library [27], and the Dynesty sampler [28]. However, one of the issues with stochastic sampling over a complicated physical system is that the likelihood evaluation in the process are computationally expensive. In order to increase the efficiency in investigating different jet models, we employ an interpolated GRB model and substitute the original likelihood function with it. This is done by first simulating each model parameter across a certain range in the parameter space and storing the resulting light curves in a multidimensional grid. The detailed configuration of the grid is specified below: where θ j is the jet half-opening angle, θ c is the core width of a structured jet, θ obs is the angle between the observer and the jet central axis, and 0 is the isotropic equivalent energy of the jet. For each dimension, the parameter range is linearly cut into subsets. We increase the density of splits for parameters that have higher impacts to the resulting light curves . The total number of entries in our 4D Cartesian grid is then 0 × θ j × θ c × θ obs = 40 × 50 × 40 × 50 = 4,000,000. Figure 1 shows how afterglow light curves distribute in the three-dimensional space for an observer at different viewing angles. When the inclination angle increases, the afterglow flux peaks at later times. Jet parameters in this dataset are θ j = 0.2 rad (∼11.5 • ), θ c = 0.15 rad (∼8.6 • ) and 0 = 2 × 10 52 ergs −1 . To put the emphasis on the jet structure and save computational power, we keep the rest of the afterglow parameters constant throughout the simulation. The fixed parameters include the initial bulk Lorentz factor Γ = 100, the ambient particle number density n = 0.001 cm −3 , the shock accelerated election index p = 2.15, the microphysical parameters B = 0.01 and e = 0.1 [24], and the luminosity distance d = 40 Mpc [1] for a GW170817-like source . The resolution in the jet model is set to 50 for all light curves.
We adopt uniform priors with upper and lower bounds the same as the simulation range for every parameter of interest. At each Monte Carlo sampling point, instead of calculating the likelihood from the original model, it performs a linear interpolation between the adjacent parameter values in the grid. A comparison between the interpolated and theoretical light curves with same parameter values is shown in Figure 2. The largest difference appears at observing angles near the jet edge, i.e., slightly off-axis case with~10% deviation. As the geometry between the jet and observers affects the light curves the most, we set the numbers of splits for θ j and θ obs larger than other parameters. Increasing the density of splits in the grid would raise the accuracy but the simulation time also dramatically grows. For postpeak observing epochs and the rest of off-axis scenarios, the interpolated light curves well resemble the theoretical values with a <10% deviation. As our input data has a SNR value of 3, the error caused by the lack of entries in the interpolation process is not the dominant source of uncertainty.
Results and Discussion
The input data, produced with the method described in Section 3, are the X-ray light curve generated from a GRB jet with a power-law profile. The width of the jet is set to θ j = 0.3 rad with the core angle being θ c = 0.15 rad. Considering an off-axis observer, we set the viewing angle θ obs = 0.36 rad (21 • ) and the isotropic energy is 0 = 2 × 10 52 erg s −1 . The remaining parameters have the same values as those used to simulate the 4D grid. This light curve is composed of the 0.4 keV fluxes at observing times from 0.16 days to 1000 days post-event and visualized as blue dots in Figure 3. After incorporating the interpolated model in the sampling process, the time required for calculating the likelihood per iteration reduces from 7.78 seconds to ∼0.09 seconds with only 1 CPU. The overall run time with this method decreases by ∼90 times. Figure 4 shows the joint posterior of the four parameters mentioned above. The injection values are well recovered within the 68% confidence interval. This serves as a proof-of-principle that this method is not only efficient, but also capable of capturing the jet features if the afterglow emission is intrinsically produced by such a system.
In this work, we perform an analysis on simulated light curves with a uniform signalto-noise ratio throughout all observing epochs. In reality, for the same patch of sky and the same observational instrument, the background fluence would remain at a similar scale. The temporal evolution of a GRB afterglow, which arises until reaching the peak luminosity at t peak and then declines with time, yields a varying observational significance. The X-ray emission from the position of GW170817, except for the first observation at 2.3 days, reports a significance of ≥3σ at all epochs. We set all SNR equaling to the detection limit for a conservative estimate. Recent studies show that the X-ray counterpart of GW170817 is still bright after 1200 days [11,12]. The 3σ excess compared to the prediction from an off-axis structured jet poses a challenge to previous agreements from multi-wavelength studies. It also makes modeling the afterglow of GRB170817 worthwhile even after four years as new evidence is still emerging. On the other hand, we use relatively wide uniform priors for jet structure parameters in the analysis. One of the strengths of multi-messenger astronomy is that we can combine the information obtained from various approaches and place tighter constraints on the source properties. As gravitational wave observation provides an independent measure of viewing angle and luminosity distance, the posteriors from GW parameter estimation can be passed to the afterglow analysis as prior function and thus increase the accuracy in search of the GRB afterglow parameters.
In this paper, we present the validity of a Bayesian framework combined with an interpolated GRB model. It effectively lessens the computational cost while being capable of constraining jet structure parameters in low SNR circumstances. In the upcoming era of a global gravitational wave detector network (LIGO-VIRGO-KAGRA) and new EM observatories like the Gravitational Wave High-energy Electromagnetic Counterpart All-sky Monitor (GECAM), it is foreseeable that we will find more GW-EM counterparts in the near future. A fast, functional parameter estimation will be beneficial for modeling the GRB afterglow light curves. With this methodology, more jet models, e.g., Gaussian or Top-hat jets, will be investigated and its application on the real data of GW170817, the only event so far with GW-EM detection, will be included in our future works.
Conflicts of Interest:
The authors declare no conflict of interest. | 3,437 | 2021-09-17T00:00:00.000 | [
"Physics",
"Computer Science"
] |
Benchmarking and Reconciliation With Time-Varying Cross-Coefficients
In this paper, the authors propose a method to obtain explicit solutions for simultaneous benchmarking and reconciliation problems for a system of variables when the cross-restrictions use time-varying coefficients. The method is based on a hierarchical Bayesian model with a normal-gamma specification for the prior distributions. The proposed solution provides explicit (not sequential) feasible estimations, including measurements for its statistical accuracy. One interesting feature of the proposed procedure is that it allows users to include one or several performance indicators and to estimate disaggregated values for incomplete years. The method is applied to obtain Quarterly Regional Accounts for the Spanish economy.
and Causey and Trager's (1981; available as an Appendix in Bozik & Otto [1988]) growth rates preservation principle to solve several reconciliation problems, including temporal or contemporaneous aggregation constraints, one or two-way time series systems or marginal benchmarking problems.
Second, the book by Dagum and Cholette (2006) analyses and solves benchmarking, calendarisation or reconciliation problems by using regression-based models. Said authors also derive solutions based on autoregressive integrated moving average (ARIMA) and structural time series models. Their approach includes distribution or interpolation problems. As in the Di Fonzo and Marini (2005) papers, the authors deal with one and two-way problems.
Present-day interest in studying the main topics concerning benchmarking and reconciliation for time series is evidenced by two recent publications: Firstly, the new version of the Quarterly National Accounts (QNA) Manual (International Monetary Fund [IMF], 2018) which dedicates chapter 6 to the systematic review of benchmarking methods that are relevant for compiling Quarterly National Accounts, although these methods may indeed prove helpful in addressing other problems, and secondly a special issue of Statistica Neerlandica (Chen, Di Fonzo, & Mushkudiani, 2018) devoted to the state-of-theart of these topics. Several papers address specific aspects and tools (for example, Chen, Di Fonzo, Howells, & Marini 2018;Guerrero & Corona, 2018;or Bisio & Moauro, 2018, among others), whilst others offer a compilation of methods or procedures. For example, Quilis' (2018) paper analyses different benchmarking procedures "in terms of practical feasibility, ease of use, and availability of dedicated software" (p. 448).
In this paper, the authors propose a method applicable to problems simultaneously involving both reconciliation and temporal benchmarking. The technique is herein applied to a Laspeyres-type volume index by employing a method derived from a proposal by Rojo and Sanz (2017), modified for use when cross-sectional restrictions employ weighted sums with time-varying weights. The solution is one-step, thereby simultaneously optimising benchmarking and reconciliation. The relevance of the problem is evident, for example, in the QNA Manual (IMF, 2018, pp. 194-195) which explores the possible inconsistences among aggregate QNA and their disaggregated estimates, with these inconsistencies deriving from the non-additivity of the Annual-Overlapping method. Said manual suggests reducing this inconvenience by presenting only percent measures of components' contribution to the aggregated variable. The transversal non-additivity for the popular Annual-Overlapping method is a key challenge in cross-sectoral reconciliation for quarterly estimates of volume index.
The proposed method allows several indicators to be used, and does not require them to be approximations for the value to be estimated. It should also be pointed out that the stochastic nature of the model proposed by the authors enables the dispersion of the solution obtained to be estimated, thereby providing Bayesian confidence intervals, see Zellner (1971, pp. 27-28), for said solution, and also obtains the expression of the linear model which relates the indicators to the high-frequency series to be estimated.
Di Fonzo and Marini (2015) or Dagum and Cholette (2006) propose alternative methods, albeit focusing on the case in which (only) one indicator approximates the high-frequency series to be estimated. Cuevas et al. (2011Cuevas et al. ( , 2015 designed a method focused on benchmarking and reconciliation for National and Regional Accounts. These authors resolve the two aspects separately and, therefore, do not ensure that the final solution respects the set of restrictions. Other methodological differences concern the use of indicators. Although the Cuevas et al. (2015) method does allow for the use of multiple indicators, it initially combines them by means of dynamic factor analysis, such that in methodological terms it is a single indicator procedure like the previous ones.
In the following section, a detailed description is given for the assumptions and development of the proposed methodology, obtaining the explicit expression of the solution. The third section applies the proposed solution to obtain Quarterly Regional Accounts (QRA) for Spanish regions, with both the Annual National Accounts (ANA) and QNA for the whole of Spain being known. The method may obviously be applied to any country's QNA and ANA. This example uses the same data as in the Cuevas et al. (2011) proposal, albeit over a longer time period. The final section summarises the most relevant findings. The work includes Supplementary Materials that contains colour tables and illustrations that are of collateral importance.
Bayesian Benchmarking and Reconciliation in the Context of Time-Varying Aggregation
Let a T i , T = 1,...,N , i = 1,...,R be a non-stochastic annual variable, observed for N consecutive years and for R disaggregated entities (typically, the disaggregation is linked to a sectoral or geographic classification). We also assume the annual variables resulting from the disaggregation over the classification, a T i , T = 1,...,N , to be known. Even though in the simplest applications the aggregation is achieved through either the sum or the mean, we consider a more general scheme such that the aggregate series is a 'linear combination' of the disaggregated ones, with the coefficients of the combination being time-varying, however. Specifically, we suppose The high or sub-annual frequency (usually monthly or quarterly) 1 aggregated series is also assumed to be known, q t, T , T = 1,...,N , t = 1,...,m where t, T refers to period t of year T , and where the number of periods for a year will be denoted by m. There may be an additional 'incomplete' year, in other words, with r sub-annual periods to be estimated, q t, N + 1 , t = 1,...,r, for r < m.
The aim is to estimate the sub-annual series for each disaggregated area, q t, T i , i = 1,...,R, t = 1,...,m, T = 1,...,N including, if any, the high-frequency values relative to the incomplete year, q t, N + 1 i , t = 1,...,r, with r < m. We assume that the cross-aggregation links at the sub-annual level are the same as are used for the annual series; that is to say considering, if appropriate, the same relation for the incomplete year. Finally, we should establish the temporal aggregation scheme for both levels of cross-aggregation. The arithmetic mean has been taken, although other classical schemes (sum, first or last values) are developed in an analogous fashion. Specifically, we assume last year is complete, and q i * = (q 1,1 i ,...,q r, N + 1 i )′ for incomplete last year schemes. We also denote by q* the contemporaneously aggregated chain series q* = (q 1,1 ,...,q m, N )′ or q* = (q 1,1 ,...,q r, N + 1 )′ for both schemes.
In addition, we denote by a* = (a 1 ,...,a N )′ the column vector of annual chained series for the aggregated area, and by a i * the column vector of annual chained series for the i-th individual area, i.e., a i * = (a 1 i ,...,a N i )′, i = 1,...,R. The cross-disaggregated series may then be stacked into the column vectors x = (q 1 *′,...,q R *′)′ and y = (a 1 *′,...,a R *′)′ Before continuing, one clarification concerning the dimensions of the matrices involved should be made; we denote by d x = N mR the column dimension of x, d y = N R the column dimension of y, and by d q = N m the number of high-frequency periods. These dimension values are, respectively, d x = (N m + r)R, d y = N R and d q = N m + r for incomplete year schemes. Obviously, d x = Rd q . By using this notation, we can compactly write sub-annual data as q i * = (q 1 i ,...,q d q i )′, i = 1,...,R, q* = (q 1 ,...,q d q )′ In sum, we have the annual and sub-annual series for the aggregated area, respectively, a* and q*. We also have the annual series a i * for the different disaggregated areas, i = 1,...,R. The main aim is to estimate the sub-annual series for those disaggregated units, q i *, i = 1,...,R, combined with the appropriate measures for the accuracy of the estimation. As usual, Bayesian strategy initially states the prior distributions for the parameters and variables involved. These distributions include a certain number of non-random hyperparameters, whose values shall be stated by using allocation procedures. In addition, we then establish a behavioural linear model explaining the sub-annual variables as a function of one or more relevant indicators or approximated series. This linear model allows the likelihood function to be established and, consequently, the posterior distribution for the quarterly series and other relevant parameters to be derived.
Joint Prior Distribution
Following Sanz (2005, 2017), the authors propose a hierarchical Bayes normal-gamma approach. Specifically, we assume a normal conditional prior density for x, given the precision, τ, precision matrix (the inverse of the variance-covariance matrix), and D = I R ⊗ D * , with I R being the identity matrix of order R and D * the (d q − 1) × d q firstdifference matrix 2 (with rank d q − 1 ). A greater degree of smoothness can be achieved by substituting D for D 2 = I R ⊗ D 2 * , with D 2 * being the (d q − 2) × d q matrix of rank d q − 2 that provides second order differences. Although only first order differences are used in the formal derivation of the estimates, the results for second order differences are obtained by simply replacing D′D with D 2 ′D 2 in the following formulae. We assume a gamma 3 prior distribution for τ, π τ y ∝ τ α − 1 e −τβ , τ > 0 (4) In sum, we obtain a normal-gamma prior joint distribution for x and τ, where x obeys restrictions (Equation 2) and (Equation 3). The prior distributions include a large number of parameters and, generally, a limited amount of data. These parameters are α and β (from prior distribution for τ ), matrices δ and P δ (involved in the prior for δ), matrices μ and P from the distribution for x and, finally, matrix P ε from the likelihood (see below).
Non-informative priors are accurate for problems in which we could take large-sized samples, allowing for a dominance of likelihood over priors. By contrast, for many Macroeconomic and Psychological studies, or for numerous analyses related with Natural Sciences, we have only a small amount of statistical information and, consequently, prior distributions dominate likelihood. The authors have chosen the option of using 'informative' priors, thus allocating reasonable values for the hyperparameters by using the information included in the marginal prior for the relevant parameters and variables. Readers can find the method in Rojo and Sanz (2005).
2) Series differentiation is often used to eliminate the tendency of non-stationary temporal series. In this work, on the other hand, it is used to penalize volatile behaviour in the series to be estimated.
Likelihood Function and Posterior Joint Distribution
In order to establish the likelihood function, we assume a linear model relating the disaggregated sub-annual series with k i indicators (maybe, thus, in different amounts for each disaggregated entity) The model has a similar expression if the last year is incomplete. We now write a matricial version for the proposed likelihood, which will provide more compact reasoning. We denote δ = (δ 1 ′,...,δ R ′ )′ with δ i = (δ 1,i ,...,δ k i , i )′, i = 1,...,R the parameters of the likelihood model for each disaggregated area, and we include the values of the indicators in the matrix Z = diag Z i i = 1 R , block diagonal matrix of size Denoting the perturbations for the models as ε = (ε 1 ′,...,ε R ′ )′, with ε i = (ε 1 i ,...,ε d q i )′ the linear model could be written as The likelihood is completely established by defining the prior for ε given τ, (ε τ) as N 0, τ −1 P ε −1 , assuming that Cov(ε i , ε j ) = 0, i ≠ j and V Cov(ε i ) = P ε i , and being the hyperparameter P ε a block diagonal matrix, P ε = diag P ε Hence, the likelihood function is given by being y the annual data vector for the disaggregated areas previously defined. We assume a normal prior distribution for δ, given τ, with k = i = 1 R k i , and P δ being the precision matrix, i.e., the conditional prior for δ is a N k δ, τ −1 P δ −1 .
Using Equation 5, Equation 6 and Equation 7
, and assuming prior independence between δ and the remaining parameters, except τ, the posterior distribution can be expressed as under restrictions (Equation 2) and (Equation 3).
A well-known result (Zellner, 1971, p. 30) states that for a quadratic loss function with C being a positive definite matrix, the minimum of the quadratic risk function (the average of the quadratic loss function with the posterior distribution) is achieved by taking for τ, x, δ the mean of that posterior joint distribution. The solution, therefore, involves obtaining said posterior means. Note, however, that the restricted distribution for x is a degenerate one. Obtaining the posterior average for x may be achieved by including the temporal and contemporaneous restrictions in the posterior distribution. This substitution removes several components of x. We then obtain the posterior mean for the 'active' parameters and variables and, taking into account the linearity of the restrictions, we derive the outcome for the remaining parameters.
Specifically, our aim is to write x = Wx r + y v , where W and y v are non-random matrix and vector, respectively, and x r will group the components that are not subject to restrictions for the sub-annual series. The temporal constraints (Equation 1) lead to one sub-annual period being excluded for each year (specifically, we have excluded the last sub-annual period for each year). Furthermore, the transversal aggregation constraint (Equation 2) implies the linear dependence of one disaggregated sub-annual series. Thus, the R-th series has also been excluded. Denoting by q i, r (r for 'restricted') the column vector whose components are the 'independent' values for the chained series corresponding to the i-th disaggregated area (q i, r = (q 1,1 i ,...,q m )′, depending on the presence of an incomplete last year), these independent values can be grouped into the column vector x r = (q 1,r ′,...,q R − 1,r ′)′. Note that the dimension of x r is equal to d x r = d x − d q − N (R − 1).
Note S1 in Supplementary Materials provides the linear relation linking x and x r x = W · x r + y v Now, integrating density (Equation 8), we obtain the posterior marginal distributions for x and for δ and, replacing restriction Equation 9 provides us with the posterior restricted distributions for both x r and δ, being M = P + D′D + P ε − P ε ZP δε −1 Z′P ε a square matrix, with P δε = P δ + Z′P ε Z, b r = b + M −1 u − y v ′M M −1 − W W′MW −1 W′ M M −1 u − y v , for b = s − u′M −1 u and s = 2β + μ′Pμ + δ′P δ δ − δ′P δ P δε −1 P δ δ scalars, and u = P ε ZP δε −1 P δ δ + Pμ a vector, being Posterior density (Equation 10) is a Student multivariate (henceforth, MS-t) 4 with υ x r = d q (1 + R) + 2α + N (R − 1) degrees of freedom. The scale matrix is υ x r b r W′MW and the position vector a r . Thus, the posterior variance-covariance matrix for x r could be The posterior first and second order moments for x are then E(x y, Z) = Wa r + y v and Posterior density for δ is obtained in a similar manner. We obtain π(δ) ∝ b δ + (δ − a δ )′M δ (δ − a δ ) − k + (2d x − d xr + 2α) 2 with M δ = P δ + Z′P ε (P ε −1 − Q r )P ε Z a square matrix with dimension k for Q r = W(W′P r W) −1 W′, a singular square matrix with dimension d x and rank d x r and being P r = P + D′D + P ε , and with b δ = 2β + y v ′D′Dy v + δ′P δ δ + μ − y v ′P μ − y v + y v ′P ε y v − Pμ − P r y v ′Q r Pμ − P r y v and being a δ the column vector a δ = M δ −1 P δ δ + Z′P ε y v + Q r (Pμ − P r y v ) .
The posterior distribution (Equation 13) for δ is, thus, an MS-t with 2d x − d x r + 2α degrees of freedom, position a δ and scale matrix (2d x − d x r + 2α)
An Example From the Quarterly Regional Accounts
We now present an example, which illustrates the above method. The resulting estimated variables are the quarterly regional chained volume series for Spanish regional Gross Domestic Product (GDP) 5 . Temporal restrictions impose consistency among the regional estimations (quarterly regional chained volume series) and annual chained volume series provided by the annual regional accounts (ARA) from official Spanish regional statistics.
We also force the transversal consistency of these quarterly regional series with the Spanish national chained series provided by the QNA, a consistency which states that the QNA chained series is a weighted sum of the regional chained volume indices.
Many national statistics institutes have used annual chain-linking series for ANA and corresponding quarterly series for QNA. Specifically, the US Bureau of Economic Analysis (BEA) has used quarterly chain-linking volume series since 1996. In the European Union (EU), a European task force was set up in 2007, co-chaired by Eurostat and the European Central Bank. The growing popularity of chain-linking series for both QNA and ANA has led to the need for efficient tools in the reconciliation and benchmarking of quarterly chain-linking series. Cuevas et al. (2011Cuevas et al. ( , 2015, for example, proposed a two-step-method for use with the annual overlap derivation of chain-linked QNA, a very popular technique frequently used in the EU (see Eiglsperger, 2008 for a detailed listing of its dissemination).
Chen, Di Fonzo, Howells, and Marini (2018) developed approaches for reconciling annual (preliminary) estimates of US national accounts aggregates subject to quinquennial benchmarks available from detailed input-output tables. Furthermore, in an update of the 2001 version of the IMF QNA manual (Shrestha [2013], a document prepared by M. Marini and Th. Alexander), the authors suggest that "the compilation of consistent quarterly estimates satisfying both low frequency benchmarks and accounting identities at the quarterly level has become more and more challenging for compilers" (Shrestha, 2013, p. 7). Although the UN's System of National Accounts (SNA) recommends use of the Fisher-type index, the authors of the manual recognise that the Laspeyres volume indices are an acceptable alternative in national accounts.
More specifically, the Spanish National Statistics Institute (INE) provides the annual GDP (chain-linked volumes, reference year 2010 6 ) for the 17 autonomous regions and for the two autonomous cities (hereinafter, 19 regions) for the nineteen-year period 2000-2018. The INE also estimates total quarterly GDP (from QNA), both at market prices (in euros) and by the volume-chained series (the annual overlap method is used here). In both cases, the raw series as well as the seasonally and working-day adjusted series (SA) are presented. At the time of writing this paper, the quarterly series was composed of 76 quarters 7 , from the first quarter of 2000 to the fourth quarter of 2018.
As mentioned before, the authors' aim is to estimate the 19-quarterly regional chained series, all being consistent with the annual regional chained series and with the total quarterly one. We only estimate the SA regional series, with the estimation for raw series following a similar development. We apply the procedure obtained in the second section for the period from quarter 2000:1 to 2018:4. Cuevas et al. (2011) solve this problem by using a multivariate two-step extension of the Denton (1971) method. They use a procedure proposed by Di Fonzo and Marini (2005). The former authors use total employment (regional social security contributors [SSC]) as high-frequency indicators, among others, due to its close linkage to real output. Although the widespread debate concerning the extent of cyclical synchrony between employment and output is well known, we do not lead or lag the SSC series in order to replicate their data selection. As already pointed out, Cuevas et al. (2011) use a two-stage procedure that is not simultaneous, unlike the one proposed in this work.
Taking into account that the estimated quarterly regional series will be seasonally adjusted, the SSC series have first been seasonally adjusted using the X12 method, as implemented in Eviews (Version 6) software. The procedure proposed by the authors was implemented in MATLAB (Version R2012b).
Before the method can be applied, an extra adjustment is needed. This is due to the lack of consistency among the whole set of annual regional chained-series and the total Spanish one, resulting from the existence of 'extra-regio' territories (extra-regional economic activities such as embassies, military or scientific bases or resource deposits in international waters, among others).
To sum up, we have regional series concerning the annual GDP for the 19 regions for the period 2000-2018, both at current prices and in terms of volume. We have also estimated Spanish quarterly GDP (without extra-regio), both raw series and SA series, 6) The INE also provides total GDP at market prices in euros. 7) INE, quarterly Spanish national accounts, series from quarter 1/1995 up to last published (4/2018). Data were extracted on May 2019, from http://www.ine.es/dynt3/inebase/es/index.html?padre=1691&dh=1.. also at current prices and in chained-linking terms. Finally, we know the raw SSC series for each region, and have previously derived the corresponding SA series. As pointed out earlier, our objective is to estimate the Quarterly Regional chained volume series for Spanish regional GDP.
We then apply the method obtained in the second section, whose notations we now describe.
The INE provides regional annual GDP at market prices, in euros, v T i , T = 1,...,19, i = 1,...,19, for the 17 autonomous regions and the two autonomous cities (R = 19 disaggregated areas) and for N = 19 years (from 2000 to 2018). National annual GDP at market prices v T , T = 1,...,19 is obtained by aggregation. It also provides the annual volume chain series, a T , a T i , T = 1,...,19, i = 1,...,19 at the national and regional level, respectively. Furthermore, the QNA provide the national chained volume series, q t, T , T = 1,...,19, t = 1,...,4, seasonally and working-day adjusted. Finally, the Labour, Migration & Social Security Ministry provides the indicator used, Z i = z t, T i T = 1,...,19,t = 1,...,4 , i = 1,...,19, the social security contributors, SSC, seasonally corrected by the authors. The 'annual overlap' method employed by the INE establishes the links between high frequency (quarterly) and low-frequency (annual) volume indices by using arithmetic means, both for national and for regional volume index. We thus impose temporal benchmarking with c T i = a T − 1 ω T − 1 i / a T − 1 i , T = 1,...,19,i = 1,...,19. We are, therefore, in the context foreseen in section 2, such that the method proposed allows us to estimate the chained volume series q t, T i , T = 1,...,19, t = 1,...,4, i = 1,...,19, grouped in vector x, and to obtain its variance-covariance matrix. Some representative tables and charts for the regional chain-linking SA series are shown in the Supplementary Materials. The authors have also obtained the results for the regional raw series, although these are not displayed in the work for reasons of length. Readers may observe the similarities and differences between our tables and charts and those obtained by Cuevas et al. (2011) for another time interval. These are shown in said reference. Table S1 in the Supplementary Materials presents the Pearson correlations between SSC and estimated series, and also between the growth rates (annual grow rates, t (1,1) = (y t − y t − 1 )/y t − 1 and quarterly growth rates, t (4,1) = (y t − y t − 4 )/y t − 4 ) for both series. Broadly speaking, the highest values are obtained for annual growth rates, except for Ceuta and Melilla 8 (the two autonomous cities). One surprising result is the low correlation in levels for Galicia, the Canary Islands, and the Comunidad Valenciana.
Tables S2 to S4 in the Supplementary Materials provide an overview of the results obtained for the Spanish regions, showing in particular regional growth behaviour during the recent Great Recession.
For its part, Figures S1 and S2 offer a graphic overview of said regional growth path. Table S5 in the Supplementary Materials presents the comparison between the cyclical signal of each region and the turning points for the aggregate Spanish reference by using ratios, as defined in Abad and Quilis (2004). Only the turning points are taken into account, and no specific attention is paid to the numerical values for the two series.
The first and second column show the so-called 'conformity ratio', comparing the paired 9 turning points of each regional cyclical signal with the individual turning points. R x thus compares such paired turning points as a percentage of the turning points for regional cycles, and R y compares them as a percentage of national ones. The conformity ratio varies between 0% and 100%, showing the extent to which the paired turning points reflect the overall cyclical signal of the region.
Readers may note that, with some exceptions, the agreement between regional and national cycles shown by indices Rx and Ry is relevant, with the exceptions corresponding to the Spanish regions of Ceuta and Castilla La Mancha. 8) Cuevas et al. (2011) suggest that their small size and particular economic structure might cause these effects. 9) For each region, we have paired its turning points with the closest Spanish ones.
The third column shows the global median delay (GMD) between regional and national cycles. Thus, the series are classified as coincident, lagged or leading with respect to the national cycle, respectively, for small (we take between -1 and +1), positive or negative GMD values. It should be noted that the Balearic Islands lead the national economy by two and a half quarters, and that Cantabria lags the national economy by two quarters.
The last column shows an index of cyclical coincidence 10 between regional and national series. Values near 1 suggest that regional and national series are highly procyclical, being highly countercyclical if the index approaches the value -1. For values near zero, we conclude that the regional series is non-classifiable compared to the national one. It should be noted that all of the values are positive, but that only Andalusia, Cataluña, the Comunidad Valenciana, Madrid and the Basque Country display high values.
We also performed the estimation with incomplete years. Now T = 18 (between 2000 and 2017). The regions involved do not change, and we take r = 3; in other words, we estimate the chained volume series for the three first quarters of 2018. The estimates are shown only for one region (Andalucía) in Figure S3 in the Supplementary Materials. Table S6 shows the Bayesian confidence intervals at 95% for this estimation (see Supplementary Materials).
Conclusions
In this article, the authors propose a method to obtain explicit solutions for simultaneous benchmarking and reconciliation problems for a system of time series when the cross-restrictions use time-varying coefficients. The method provides explicit solutions to the estimation problem and deals with concurrently solving temporal restrictions (benchmarking annual and sub-annual frequency series) and contemporaneous ones (reconciliation among disaggregated and aggregated sub-annual frequency series).
The Bayesian model involved belongs to the frequently used normal-gamma family and minimizes a risk function derived from a quadratic loss function. In addition, the design of the method allows users to include one or several performance indicators through the likelihood model, and to estimate quarterly values for incomplete years.
The stochastic nature of the proposed model allows Bayesian confidence intervals to be obtained for each of the values of the high-frequency series estimated. These intervals are particularly interesting when estimating incomplete years.
Comparisons with alternative methods are not, broadly speaking, feasible since the methods have a different statistical base (mathematical methods compared to statistical 10) Defining V x (V y ) as 1 for the growth phases and -1 for the declining phases, the coincidence index is then defined as COI N C x, y = 1/N t = 1 N V x, t · V y, t . methods) and are not nested models (none of them is a generalisation of the other). Nevertheless, certain differences may be pointed out.
Compared to the proposal put forward by Cuevas et al. (2011Cuevas et al. ( , p. 6, 2015) it is worth highlighting that our procedure allows for the use of several indicators when estimating the sub-annual disaggregate series. Said indicators need not even be the same for all of them. In contrast, the proposal of Cuevas et al., previously synthesises the indicators in a single indicator through dynamic factor analysis. In addition, these authors' procedure initially resolves temporal benchmarking and, subsequently, reconciliation of the disaggregated areas with the total aggregate. It is therefore (in any case) a sequentially optimal procedure, whereas the one proposed in this paper is globally optimal.
As regards the procedure of Di Marini (2011, 2015), we highlight that the design of their model implies using a single indicator. In fact, said authors follow the original idea of Denton (1971), such that the problem is one of adjustment through a single indicator, which is an approximation of the target series (for example, in Di Fonzo & Marini, 2011), they state that: "This procedure performs the constrained optimization of an objective function according to which the proportionate difference between the benchmarked and the original series ... must be as constant as possible through time" (emphasis ours; p. 148). In the words of IMF (2018), "the objective is to combine the quarterly movements of the indicator with the annual levels of the ANA variables" (p. 87). Strictly speaking, therefore, it involves not having an indicator but rather an approximation to the sub-annual series to be estimated.
Since it is an approximation, Di Fonzo and Marini (2011) suggest that one of the advantages of their method involves the similarity of the series estimated, whether either one or more, to the approximation (indicator) used. This apparent advantage proves to be a drawback when the indicator used as an approximation is too volatile, as tends to happen for small disaggregated areas. As seen in the second section, the proposed inclusion of an additional factor in the prior distribution seeks to correct, at least partially, that volatility, should it occur. The authors have simulated examples at different levels of volatility for the indicators, and have shown the ability of the Bayesian proposal to obtain smooth estimates. The comparison is carried out only for temporal benchmarking procedure and not for the reconciliation, given that Di Fonzo and Marini did not provide solutions for the reconciliation when time-varying links are present.
An example related to Spanish quarterly accounts, combining national and regional volume series, is presented, and evidences the method's feasibility and appropriateness. In addition, individual regional behaviour during the recent twin economic crisis is analysed using the estimated quarterly volume series.
As previously pointed out (and illustrated in Figure S3 and Table S6 in the Supplementary Materials), the proposed method allows, as in those of Di Fonzo and Marini (2011) and Dagum and Cholette (2006), the target series to be estimated when the final year is incomplete. The additional advantage of the proposal put forward is that it enables Bayesian confidence bands to be built based on the posterior distribution of the estimated series.
Funding: The authors have no funding to report.
Competing Interests:
The authors have declared that no competing interests exist. | 7,803.6 | 2020-01-01T00:00:00.000 | [
"Economics",
"Mathematics"
] |
The Reform of Blended Courses for Software Project Management
Blended teaching is a teaching method integrating classroom teaching and online learning. Based on OBE (Outcome-based education), this paper proposes some reform ideas for blended teaching by software project management curriculum. Project driven, flipped classroom, vivid animations, MOOC video and cases study etc. are adopted to improve the self-study ability and team work of students. Although it takes more time to study, it gain high praise and enhance the teaching quality.
Introduction
Online teaching has gained more and more attention because of the COVID-19. With the deep integration of Internet and teaching, the design and application of blended teaching has become one of the key contents on the reform of teaching methods in colleges and universities. [1] Software project management is a comprehensive curriculum which stress on both theory and practice. Through this course study, students should grasp the basic theory, method, processes, tools and apply them in software project management practice. [2] As we know, students should study the theories, methods and some popular tools in advance, so they can better apply the learned knowledge on work and make a good basis for an excellent software project manager later.
OBE (Outcome-based education) is a kind of educational philosophy which is "student -centered, achievement oriented and continuous improvement". Focus on "what learners can actually do with what they know and have learned". [3] Based on OBE continuous improvement, we update teaching methods through blended teaching, cultivate students' self-study ability and improve students' interest in learning, in order to realize the teaching goal of " knowledge, thinking and apply".
Based on the previous achievements on software project management curriculum, this paper introduces some reform methods for blended teaching. Teaching means has changed from "teacher-centered" to "student-centered". The teaching process mainly adopts "project driven -> flipped classroom -> method inspiration -> animation demonstration -> cases study ->quiz" to improve students' participation, promote their independent thinking and self-study ability. The purpose is make students better understand the learned knowledge and use them to solve the actual project management problems.
The Relationship for Graduation Requirements and Curriculum Objectives
Software project management is an important course that can transform students' early knowledge into ability and improve students' career development. Software project management consists of a number of activities, which includes planning of project, deciding scope of software product, estimation of cost in various terms, scheduling of tasks and events, and resource management. The basic idea of OBE is "student-centered", focus on students' learning achievements, emphasize what abilities students have by learning, what graduation requirements they should meet. There are 12 graduation requirements and students should achieve the goal after 5 years graduate.
[3] We should make clear the curriculum objectives and which of the 12 graduation requirements the course supports. There are three curriculum objectives for software project management, shown as follows: Curriculum objective 1. Students are required to master the characteristics of software project management, the basic concepts, principles and methods of 10 knowledge areas and 5 process groups of project management, and can manage software projects with engineering ideas, and enhance the ability of analysing and solving complex software engineering problems.
Curriculum objective 2. Students should grasp the basic technology and common tools in each stage of software project management, to ensure the successful implementation of the project. Should master how to write important technical documents. And understand the role and relationship of individuals, team members and leaders in complex software engineering practice under the multi-disciplinary background.
Curriculum Objective 3. Students should know how to conduct economic feasibility analysis and decision-making on software engineering projects, analyze and evaluate objectively the impact of software engineering practice and complex software engineering problem solutions on society, health, safety, law and culture.
The supporting relationship between graduation requirements and curriculum objectives is shown as table 1. [4] 3 Reform of Blended Teaching Method
Classroom Teaching
To learn software project management curriculum better is to know some management rules, study some good cases, consult a series of standard templates and master some popular tools for project management. [5] Software project management curriculum on our school was approved as the key curriculum of Shanghai on 2017, and the excellent curriculum of our university on 2019. A variety of teaching methods and means are used in the teaching. We have acculmulated rich and diverse curriculum resources. MOOC video have been recorded by ourseleves. After the use in the spring semester, the students reflected that the video content is comprehensive and explain clearly, which has made a very good foundation for the online and offline blended teaching. With the continuous improvement, this curriculum is deeply loved by students. The details of the classroom teaching reform are mainly shown as follows: First, we do many investigations and academic exchanges with enterprises and other colleges and universities. We must know newest technologies, methods and the needs of enterprises, and then integrate them into teaching. Meanwhile, the syllabus, teaching content, teaching methods and teaching means of the course are updated to ensure the timeliness to a certain extent.
Second, we have collected some classic cases and established a teaching case library. There are four complete cases which covering the whole processes of software project management, including object-oriented method and traditional structured method. Cases study teaching improves the teaching effect.
Third, professional MOOCS videos have been recorded and put on the course website for students to learn. Because the COVID-19 epidemic, online teaching was implemented in our school, the video was used and the students' evaluation was very good.
Fourth, we have written the experimental instruction. Software project management course is very practical, in order to master the actual process of software project management and grasp the theoretical knowledge, it also needs a lot of practical experience. In practice teaching, we stress on the teaching of integrating theory with practice. Simulating the real software develop environment, conduct the whole process of project management, more cases analysis and organize more discussions. Project practice can improve students' analytical ability and practical ability.
Fifth, the vivid course animation simulating the actual scene of software project management. These animations let students understand the actual operation of software project management better. Make more boring theoretical knowledge easier to remember and understand. Enhance students' interest in learning and achieved excellent results.
Sixth, the complete set of exercises has been compiled, including after class exercises and answers, PMP examination exercises and various exercises; the types of questions include single choice questions, fill in the blanks, short answer questions and case analysis questions. Homework and practice are very important to consolidate the learned knowledge. So, appropriate homework and practice work should be assigned to students. And they are important part of assessment.
Blended Teaching Reform
Teaching methods and means will be "student-centered", using "project driven and case study", online and offline blended teaching method. Online course resources are very rich. We have built a course website with complete learning resources based on ChaoXing platform, including: MOOC video by ourselves, course plan, syllabus, schedule, case library, exercise library, experiment guide, scene animation, catchy formula and document template and online communication etc. Through the website, students can obtain learning materials all day, communicate with each other, and submit assignments, which greatly facilitates students' learning. It is convenient to cultivating the students' autonomous learning ability.
Based on the concept of flipped classroom, the blended teaching process is shown as figure 1. [6] Fig. 1. The Blended Teaching Process 1) Before class, students should familiar with the blended teaching process, and know clearly about the learning objectives, learning requirements and the contents to be prepared before class.
2) Online learning, watching recorded MOOC videos, teaching materials and online learning materials. The video content is rich, the explanation is detailed, and some animation and exercises are interspersed, which improves the interest and practicability of the course video, and facilitates students' preview before class and review after class.
3) Teachers should track students' learning situation, collect online learning problems, and adjust the class process accordingly.
4) The key and difficult points of knowledge are mainly explained in class, and the learned knowledge is tested through questions and classroom exercises. Through "case analysis -> group discussion -> stage report ->mutual comments -> induction and summary", consolidate the knowledge points learned and improve the ability of knowledge transformation. Teaching activities should be fully designed in class to arouse students' learning enthusiasm and change the traditional "teacher-centered" to "student-centered" learning.
Assessment Reform
Process assessment is a very important part for this course reform. Online and offline learning process and results need to be evaluated. In order to ensure the effect of pre-class learning, Teachers should track and evaluate students' learning situation, collect online learning problems and ask questions in class. The details are as follows: 1) Online learning assessment 40%, include two parts: online learning participation 20% and examination score 20%. The main assessment of online course participation, including statistics of the number of video viewers, the proportion of average video viewing time, the number and quality of students' questions. Examination score is mainly about the completion of online quiz and online course assignments.
2) Classroom learning and final assessment: 60%. It mainly includes students' usual classroom performance, group discussion and final project work, etc. Assessment is very important to this project-driven teaching mode. The final examination is changed from the traditional paper examination to the project work. Divide 3-5 students into a team, select a project, and use the method of software project management to analyse, plan and manage the project. At each phase, we choose 1-2 teams to present their work of on class, other students listen, ask questions and discuss for their presentation. At the end of class, the teacher review and summary for the presentation. After the presentation, students can revise their project work according to the comments and suggestion. Each team have 2 opportunities to modify and rework. Such a repeated process is very helpful for students to master the theoretical knowledge of software project management. At last, each team should submit all their work including the project plan, source code, documents and prepare for a presentation. The final score for project work is also given in proportion.
3) Students are instructed to attend professional qualification certificate like "ruankao" etc. The professional certificate can strengthen student's professional direction, consolidate and check the learning knowledge. Professional certificate also can do help for students to find jobs. If students get some professional qualification certificate about project management, he can apply for exemption.
With the blended teaching and project practice, students could learn this curriculum better. Most students giving positive feedback. Due to space limitation, we give some feedback of students, shown as Figure 2.
Summary
Nowadays, the ability of project management gets more and more attention. As one of the core curriculum for software engineering and related majors, this paper proposes some reform ideas for blended teaching on software project management curriculum. The syllabus, teaching content, teaching mode and teaching methods based on OBE have been enriched and completed. The online and offline blended teaching mode and teaching methods suitable for cultivating application-oriented talents are explored. Although blended teaching increases the study time and pressure, it improves the quality of teaching and most students like the new teaching mode. We hope these reform ideas will do some help to your teaching improvement. | 2,648.8 | 2021-01-01T00:00:00.000 | [
"Computer Science",
"Education"
] |
Contribution of the Women’s Cooperative Societies for Living Standards of Rural Family Life
Poverty has become a global women’s issue, as majority of the world’s poor are women due to various reasons. In Sri Lanka, poverty number refers to the percentage of individuals whose household per capita consumption below the official poverty line. Sri Lanka has among the lowest extreme poverty rates among countries in the region. The past studies exhibits the contribution of Micro Finance Institutions, Samurdhi Program and other government programs as means of poverty alleviation but, there are no research on the Women’s Cooperative Societies and their role in rural community . In the present study, the researcher is focusing on to capture the contribution of Women’s Cooperative Society on uplifting living standards of rural families of Sri Lanka. The purpose of the study is to identify the impact of Women’s Co-operative Society on standard of living of rural community of Sri Lanka. Hence, it is identified four independent variables namely, loan size available to members, size of savings kept at the society, number of dependents of the family and years of membership with the society to measure the impact of Women’s Cooperative Society and one dependent variable namely, average monthly expenditure of the family to measure the living standards of rural family life. The results of correlation analysis revealed that the correlation between all four variables and living standards of rural families were significant and the results of multiple regression analysis revealed that the loan size available to members and number dependents are only significant in determining the relationship between living standards of rural families and the women’s cooperative society. These findings can be helpful for policy makers in the field of cooperative societies in designing their services further in the future.
Introduction
Co-operatives, which are consisted solely with women, can be recognized as women cooperatives. It gives more authority and power to women to take part in more social and economic activities rather than fully depending upon their male counterpart for their daily needs. In Sri Lanka, the Women's Co-operative Society is recognized as Women's Bank (Gamage & Keppetiyagama, 2004). This is built, owned, and operated by and for poor women of Sri Lanka. It was incorporated under the Cooperative Societies law in 1991 as a District Society and upgraded to national level in 1998 as the Sri Lanka Women's Development Services Cooperative Society Ltd (Gamage & Keppetiyagama, 2004). It is engaged in a mission to provide the resources, ideas, and support for its own members to solve their own problems using the cooperative principles of self-help and mutual aid.
It is important to understand that; the Women's Co-operative Society is not coming under the concept of Micro Finance. According to the performance report 2012 of Department of Cooperative Development Sri Lanka, Women's bank is recognized as an island wide primary Thrift and Credit Cooperative society. However, there are similarities and dissimilarities exist between micro finance institutions and the Women's Co-operative Society. Nonetheless, microfinance is also recognized as one of the new development strategies for poverty alleviation through social and economic development of the poor with special emphasis on empowering women (Singh, 2009).
According to the World Council of Credit Unions it can be clearly identified the differences between Micro Finance Institutions' (MFI) and co-operative societies. Although, the micro finance institutions indirect objective is to obtain social welfare, they are profit-oriented institutions and these institutions are funded by external loans, in contrast co-operative societies are not profit oriented and they are member owned financial cooperatives funded by voluntary member deposits. Earnings of the MFI's are used to build reserves or are divided among investors. Earnings of the co-operative societies are applied to lower interest on loans, higher interest on savings or new product and service development. MFI's are run by an appointed board of directors or salaried staff and where co-operative society members elect a volunteer board of directors and management from their membership and they are eligible for an allowance for the service rendered by them. The past literature has much focused on the role of MFI's as a mean of poverty reduction (Mawa,2008;Dahir A.M.2015;Adejoke,A.G 2010) and to the best knowledge of authors, there are no research on the Women's Co-operative Societies and their role as a contributor to the rural community's standard of living. Therefore, this research was undertaken to explore the role of women's cooperative societies in uplifting the standard of living of low income earning people of Sri Lanka.
Statement of the Problem
One of the major problems that the country facing today is poverty, according to the Department of Census and Statistics (Household Income and Expenditure Survey 2015) Sri Lanka's Poverty Headcount Ratio (PHR) in 2012/13, urban PHR in 2012/13, rural PHR in 2012/13, estate PHR in 2012/13 are 6.7 %, 2.1%, 7.6%, 10.9%, respectively. If we take into consideration the above PHR in 2009/2010 the national level PHR was 8.9% while the corresponding values for urban, rural, estate was 5.3%, 9.4%, 11.4% respectively. The results of the two surveys showed that there is a statistically reduction in poverty in Sri Lanka in year 2012/2013 over 2009/2010. And also, it is evident that the rural sector has got more poverty than the urban sector. According to the United Nation Development Program (2012), 1.4 to 1.8 billion in the world lives under the poverty line 1 of which 70% are women. Further, the studies conducted in a number of countries confirmed that the majority under the poverty line is women (Bernard, Lock Teng, & Khin, 2017). Women are the key in the society and their role is extremely important within the world. They give a big contribution not solely within the family, additionally within the society. The perception of women as good with money, including being better at paying back loans, has led them to be targeted in microfinance Programs (Bradshaw, 2013).
In developing countries poor people organize in many types of grassroots organizations whose purpose is to improve their wellbeing and addressed shared problems through collective action (ILO, 1966;Mayoux, 1993). Cooperatives have been viewed as one of the principle institutional hardware for enabling the financially feeble individuals from the general public. In the case of women co-operatives tend to serve dual functions, the first economic and second social. Co-operatives support women's productivity and income generating projects (Mayoux, 1993;Laming, 1983). Researchers have found that women who participated in co-operatives tend to be heads of households and frequently are the sole providers in their families (Molyneux, 1985) The general purpose of the study is to identify the impact of Women's Cooperative Society on standard of living of rural community of Sri Lanka. It is expected to analyze the ultimate impact of these co-operatives on the rural women families. The specific objectives of the study are two fold, firstly, to identify the factors that are more attractive to the rural women engage in Women's Co-operative Society and secondly, to identify the role of Women's Co-operative Society in economically empowering rural women in Sri Lanka.
Methodology Variables
With the aim to determine the impact of the women's cooperative society towards the living standards of rural family life the following framework is developed.
Living Standards of Rural
Family Life
Loan Size
Size of Savings
Years of Membership
The researcher has identified four independent variables to measure the impact of women's cooperative society namely loan size available to members, size of savings kept at the society, number of dependents of the family and years of membership with the society. The dependent variable, living standards of rural family life is measured by the average monthly expenditure of the family. Women, who are members of this Women's Co-operative Society, are the target population for the present study. For the purpose of data collection, the cluster sampling method was used and the authors were able to gather data from 100 participants to the survey and that was more than ten percent (10%) from the target population of Colombo and Gampaha Districts.
In this study primary data collected through a self-developed questionnaire which was filled by the researcher by interviewing each member personally and finally 100 questionnaires were able to fill. Secondary data were collected from the sources such as journals, research publications, annual reports, statistical reports, books, and Internet. The dimensions and indicators of the variables are illustrated in the table 1.
Model Specification and Analysis of Data
In the present study the researcher had identified four independent variables which showcase the impact of women's cooperative society such as, loan size, size of savings, number of dependents and years of membership (Sebhatu , 2015). The impact on living standards of rural family life is measured by the average monthly expenditure of a family (Slesnick, 1993). Data analysis was carried out by using Statistical Package for Social Science (SPSS) version 20.0 software. Descriptive statistics analysis and multiple regression model was used to analyze the contribution of women's cooperative society for the rural family life hence the following regression model was developed for the study. The correlation coefficient of the present study interpreted as, if correlation coefficient ranges from 0.00-0.19 the relationship is very weak, from 0.20-0.39 is weak, from 0.40-0.59 is moderate, from 0.60-079 is strong and 0.80-1 the relationship is very strong (Ratner, 2009
Results and Discussion
The summary of descriptive analysis carried out by the researcher is depicted in table 2. According to the sample the highest participation in the cooperative societies were in between the women's age of 40 to 50 (74%). And the lowest participation is from the age group of 30 to 40 (7%). The marital status of the sample shows that 81% of the members are married and 6% of the sample is widowed. The education qualification level of the respondents shows that 77% of the members are educated up to GCE advance level and 23% of the members are educated up to GCE ordinary level. According to table 3 illustrates the highest mean value comes from the years of membership with the cooperative society and lowest mean value is with the loan size. Hence it can be concluded that the most influential variable among the four independent variables which affects the living standards of rural family life is the years of membership with the society.
Y= a + β1X1 + β 2X2 + β 3X3 + β 4X4 According to the results of correlation coefficients, there is a positive strong correlation between average monthly expenditure and years of membership and the relationship is statistically significant. Correlation between average monthly expenditure and number of dependents is moderate and significant. There is a strong positive significant correlation exists between loan size and average monthly expenditure. And the correlation between average monthly expenditure and amount of savings is strong and significant Table 4 depicted above shows the summary of the overall model fit. R square is a statistical measure which shows how close the data are to the fitted line, sometimes called as coefficient of determination. In the present study 68% of R square value conveys that the proportion of variation in the dependent variable explained by the regression model is significant and the model does fit the data. As showed in the table 5, the total sum of squares represents how much explained by the regression and how much explained by the residual out of total model. As total sum of squares value is 52.640 and regression sum of squares value is 35.776 it symbolizes that 35.776 of variation of dependent variable of living standards of rural family life can be explained by regression and 16.864 of variation of dependent variable is explained by residual which means that there are few more independent variables amount to residual value which are not identified by the researcher in analyzing the contribution of women's cooperative society for living standards of rural family life. P value of 0.000 symbolizes that the overall model is significant hence it is less than 0.05 (P value < 0.05).
According to the table 6 illustrated below the multiple linear regression model was tested by using SPSS version 20.0. As per the results state the coefficient of constant (β0) is 0.816 and the P value is 0.001which denotes that is significant at the 0.05 significance level. Further it shows the impact of other variables which are not identified by the researcher in determining the living standards of rural family life through women's cooperative society. The coefficient of loan size 0.329 states that when loan size increases by 1 percent the living standards of rural family will rise by 0.329 percent. The coefficient of size of savings 0.170 states that when savings amount increase by 1 percent the living standards of rural family will increase by 0.170 percent. The coefficient of number of dependents 0.213 states that when number of dependents of a family increase by 1 person the living standards of rural family life will increase by 0.213 percent. The year of membership coefficient is 0.061. It states that when years of membership increase by 1 year the living standards of rural family life will increase by 0.061 percent.
Table 6. Multiple regression analysis
As mentioned in the table 6, loan size and number of dependents in family are significant at 0.05 significance level as their P values are less than 0.05. Hence it rejects the null hypotheses (H0=There is no relationship between X1, X3 with Y) and accepts the alternative hypotheses (H1). Findings from earlier studies indicated that the ability of the women to access more financial opportunities or loans under low interest rates and relaxed repayment opportunities increases their purchasing power and thereby increases their living standards. (Hashemi, 1996).Further as the number of dependents of a family increases women are more encouraged to participate in the cooperative societies (Okaya, 2013) and it does increase their living standards. Size of savings and years of membership are not significant at 0.05 significance level. Hence it accepts the null hypotheses (H0) and rejects alternative hypotheses (H1). Conferring to the results of multiple regression analysis illustrated in table 6, the impact of the variables of size of savings and the years of membership towards determining the living standards of rural family life have found insignificant. This is due to the reason: the availability of loan size variation between the respondents were minor and the majority of respondents were more than 5 years of membership and the variation of the respondents were little. Hence these variables were insignificant in determining the contribution of women's cooperative society towards the living standards of rural family life. However, these limitations can be resolved increasing the sample size and collecting data in a wider scope.
Conclusions
Women are the key in the society and their role is extremely important within the world. They give a big contribution not solely within the family, additionally within the society. Female economic empowerment is usually about the increased access of women to financial resources, income generating assets or activities, saving, increased financial decision-making power and more economic independence (Mayoux & Linda, 2006). According to the United Nation Development Program (2012), 1.4 to 1.8 billion in the world lives under the poverty line, one of which 70% are women. Hence it is clear that women are the most lacking context in the society who needs more attention. In the present study the researcher tried to investigate the contribution given by the local women's cooperative society of Sri Lanka for the rural women to uplift their living standards. Accordingly the researcher had identified four variables, such as; loan size available to a member, size of savings kept at the cooperative society, number of dependents in a family and years of membership with the cooperative society as independent variables in order to measure the contribution given by the women's cooperative society (Sebhatu, 2015). In order to measure the living standards of rural family life the researcher has identified the average monthly expenditure of the family (Slesnick, 1993).
According to the findings the researcher had found that the correlation between all four variables and the living standards of rural family life are positive and significant. Conferring to the multiple regression results it is found that though the all four variables have significant correlation with the living standards of rural family life, the size of savings and years of membership with the cooperative society found to be insignificant in determining contribution given by the women's cooperative society to uplift the standard of living of rural family life. This is due to the reason: of the loan size variation between the respondents were minor and the majority of respondents were more than 5 years of membership and the variation of the respondent's membership were little.
It is recommended to increase the awareness regarding the women's cooperative society among the rural and poor women all over the country because the society itself contribute to uplift the rural family life immensely. It is recommended to increase the availability of loan size to each member in the ways of relaxing the rules and regulations and with longer payback periods as it can be identified as the most influential factor through the multiple linear regression model in order to uplift the living standards of rural family life. | 4,020.8 | 2019-01-01T00:00:00.000 | [
"Sociology",
"Economics"
] |
Non-separable states in a bipartite elastic system
We consider two one-dimensional harmonic chains coupled along their length via linear springs. Casting the elastic wave equation for this system in a Dirac-like form reveals a directional representation. The elastic band structure, in a spectral representation, is constituted of two branches corresponding to symmetric and antisymmetric modes. In the directional representation, the antisymmetric states of the elastic waves possess a plane wave orbital part and a 4x1 spinor part. Two of the components of the spinor part of the wave function relate to the amplitude of the forward component of waves propagating in both chains. The other two components relate to the amplitude of the backward component of waves. The 4x1 spinorial state of the two coupled chains is supported by the tensor product Hilbert space of two identical subsystems composed of a non-interacting chain with linear springs coupled to a rigid substrate. The 4x1 spinor of the coupled system is shown to be in general not separable into the tenso...
I. INTRODUCTION
The phenomenon of entanglement of quantum systems is receiving an increasing amount of attention in the light of its importance in the development of quantum information science and computing. 1Entangled quantum states are those states, defined in a vector space obtained by a tensor product of subspaces, which cannot be written as a tensor product of states in the subspaces.One can distinguish between two forms of entanglement, namely entanglement between spatially separated systems or particles and entanglement between different degrees of freedom of the same system or particle. 2,3The former type of entanglement is nonlocal and exclusively quantum in nature, while the latter is local and may occur in classical systems.For example, bipartite classical entanglement has been achieved between the orbital angular momentum and the polarization states of electromagnetic waves. 4,5Tripartite classical entanglement has also been reported by controlling the path, polarization, and transverse mode degrees of freedom of a classical laser beam. 6From the point of view of nomenclature, since classical entanglement does not account for non-locality, it is more appropriate to use the term "non-separability" and describe correlated states of a single systems as "non-separable states". 7hile non-separability in the states of electromagnetic waves has been extensively studied, the objective of the present paper is to address the much less studied notions of separability and nonseparability of multipartite classical mechanical systems supporting elastic waves.More specifically, we consider separability relative to the choice of the partitioning of the multipartite system.Indeed, it is known that given a multipartite physical system, whether quantum or classical, the way to subdivide it into subsystems is not unique. 8Therefore, the separability and non-separability of multipartite quantum systems possess some ambiguity relative to their decomposition into subsystems.For instance, the states of a quantum system may not appear entangled relative to some decomposition but appear entangled relative to another partitioning.The question arises as to how to make the choice of a decomposition into subsystems of a classical elastic systems.The criterion for that choice may be the ability to perform observations and measurements of some degrees of freedom of the subsystems. 9n particular, we consider a bipartite classical mechanical system composed of two coupled onedimensional elastic chains whose elastic wave equations can be factored into a Dirac-like equation.The antisymmetric elastic waves exhibit quantum-like behavior.Indeed, the elastic wave functions, solutions of the Dirac-like equation, possess a plane wave orbital part and an amplitude that takes on the form of spinors.The 4x1 spinor amplitude expressed in a directional representation leads to constraints on the degrees of freedom associated with the direction of propagation along the chains.The components of the spinor part of the wave function are also measurable.1][12] While the states of the bipartite mechanical system are separable in a spectral representation, the possibility of measuring the spinor part of its wave function dictates another partitioning into two identical subsystems, each composed of a single elastic chain coupled to a rigid substrate.The states of elastic waves in these subsystems can also be described via a Dirac-like equation and possess 2x1 spinor amplitudes.These amplitudes are also again measurable through measurements of transmission coefficients.The 4x1 spinor states of the elastic bipartite system, defined in the tensor product Hilbert space of the two subsystems, is shown not to be expressible as tensor products of subsystems 2x1 spinor states except in some limiting cases.The non-separability of the states in terms of direction of propagation of the elastic bipartite system relative to single chain subsystems is analogous to the phenomenon of local correlation.These non-separable classical states and their analogy with local quantum states may prove to be useful in quantum information processing.
A. Coupled two-chain system
We consider two one-dimensional harmonic chains with linear springs coupling pairs of masses along the chains (see figure 1a).
In the long wavelength limit, the system is treated as continuous and the equations of motion are: u and v are the displacements in the top chain "a" and the bottom chain "b" respectively.The constants β and γ depend on the stiffness of the spring and the masses.Inserting plane wave solutions: FIG. 1. Schematic illustration of (a) the two-chain elastic system.The top and bottom chains are identified as "a" and "b", respectively; and (b) system composed of a single chain coupled to a rigid substrate via side springs.
u = Ae iωt e ikx and v = Be iωt e ikx yields the eigen value problem: There are two sets of eigen values: and These eigen values correspond to two nondegenerate dispersion curves in the one-dimensional band structure of the coupled two-chains system.The first eigen value (Eq.( 4)) passes through the origin and is the same as that of the uncoupled chains.Inserting its expression into equation ( 3) leads to the condition on the amplitudes: A = B.This is the well know symmetric mode where the displacements u(x, t) and v(x, t) of the masses in the two chains are in phase.There is no energy stored in the coupling springs.Using the second eigen value which takes on a finite value at the origin, leads to the condition on the amplitudes: A = B.This is the antisymmetric mode.The displacements of the masses in the two chains are out of phase (phase difference of π).There is energy exchanged and stored in the coupling springs.To analyze this system further, let us consider the problem of the coupled chains using Diraclike forms of the equations of motion.In the long wavelength limit, the Dirac-like equation for this coupled system is written in the form: γ. σ x and σ y are the 2x2 Pauli matrices given by σ x = 0 1 1 0 and σ y = 0 −i i 0 .In the language of quantum field theory the non-self dual Ψ and Ψ are 4x1 vectors, corresponding to "particles" and "antiparticles."In matrix form and for the minus part of the ± ("particles"), equation (6a) becomes: , we can rewrite equation ( 7) in a spectral representation: where
B. Single chain coupled to a substrate
We also consider the case of a single chain coupled to a rigid substrate through harmonic springs (Fig. 1b) to illustrate our reference for the decomposition of the coupled two-chain system.The long-wavelength equation of motion for the single chain system is: Equation ( 9) can be also factored in the form of Dirac equations and its adjoint: where σ x and σ y are again the 2x2 Pauli matrices and I is the 2x2 identity matrix.We have investigated that system in a previous publication. 11Here, we have written the solutions to Dirac equation in the form: where ξ k are two by one spinors.It is important to point out that the interpretation of the Dirac spinor is in terms of forward and backward propagating waves.Two limits of the spinor components of the single chain system will be useful in section III.B which are listed in Table II.
We have shown in reference 10-12, that the spinor part of the wave function when projected on the orthonormal basis 1 0 and 0 1 represents a superposition of states in the possible directions of propagation of the wave.In such a directional representation, these states are quasi-standing waves.
If one solves the classical Eq. ( 9) using plane wave solutions, one does find that the directions of propagation are correlated as required for quasi-standing waves.The amplitude of the forward propagating and backward propagating waves that make up the quasi-standing wave are not independent of each other.Introducing the Dirac formalism enables us to project the quasi-standing wave solutions on forward and backward propagating states.We also note that in conventional quantum systems, a measurement on a superposition of states would collapse the wave function into a pure state.A number of measurements is then required to obtain probabilistic information on the characteristics of the superposition of states.We have shown 11 that one can use measurement of the transmission coefficient to determine the spinor part of the wave function in the directional representation and through a single measurement determine the system's superposition of states.Indeed, by defining the number operator: N = ∫ dxψ † σ x ψ where ψ is promoted to an operator and ψ † is its Hermitian conjugate.These operators when expanded in plane waves can be expressed in terms of creation and annihilation operators, namely a † k , a † * k , a k , and We can rewrite the number operator for wave number k, as: The operator 1 2ω S − S + σ x corresponds to the occupancy of the opposite direction of propagation.Its Eigen values are given by: n The transmission coefficient is a measure of these eigen values through the relation 2 which represents a standing wave with zero transmission (n + k -n − k )=0.For k=∞, we find n + k = 1, n − k = 0 which represents a traveling wave wave with complete transmission (n + k -n − k )=1.
A. The two-chain system is separable in the spectral representation
Let us now consider a bipartite composite system constituted of two two-chain systems in two different non-degenerate states, ϕ 1 = ϕ(k 1 , t) and ϕ 2 = ϕ(k 2 , t).We rewrite equation ( 8) in the form of the equation for a two non-interacting two-chain system: The solution φ = ϕ 1 ⊗ ϕ 2 is now a 16x1 vector representing the tensor product of the states of the two two-chain systems.This can be generalized to a multipartite composite system of N non-interacting two-chain systems in the various states: ϕ 1 (k 1 , t), ϕ 2 (k 2 , t),. .., ϕ N (k N , t).Equation ( 11) is generalized to: The states of the N two-chain systems span the N tensor product space H ⊗ H ⊗ . . .⊗ H where H is the Hilbert space of the states of a single two-chain system in the spectral representation.It is straightforward to show that these states take the form of tensor products of single two-chain system states: Φ = ϕ 1 ⊗ ϕ 2 ⊗ ϕ 3 ⊗ . . .⊗ ϕ N .The multipartite system of N non-interacting two-chain systems is separable in the spectral representation.This argument could also be applied to multiple single chain systems to show that a composite system of N non-interacting single chain systems is also separable in the spectral representation.
B. The two-chain system is not always separable in the directional representation
We solve equation ( 7) by choosing plane wave solutions,Ψ = ×e iωt e ikx form the orthonormal basis for the solutions of the coupled system.Equation (7) transforms into the set of four linear equations: By inspection, we find that a 1 = a 3 is a solution.It leads to the eigen values: ω = ± βk which yield in turn a 2 = a 4 .The actual values of a 1 and a 2 are arbitrary, that is there are no correlations between the forward and backward directions of propagation of plane waves along the two coupled chains.This solution corresponds to the symmetric mode and will not be addressed further as solution does not exist for the single chain system coupled to a rigid substrate.Another solution of equation ( 13) can be found when a 1 = a 3 and a 2 = a 4 .In that case, the eigen values are ω 2 = β 2 k 2 + (2δ) 2 .Inserting 2δ = + ω 2 − β 2 k 2 (the + sign is chosen because δ physically represents a stiffness), into equations (13) yields: A possible solution of the set of equations ( 15) is: The negative signs in Eq. ( 16) reflect the antisymmetry of the displacement.
Note also that we could normalize equation ( 16) by a 0 √ + to obtain: In the directional representation using Dirac formalism, we can interpret â1 and â3 as relating to the amplitude of the forward component of waves propagating in the top and bottom chains in the coupled system.â2 and â4 relate to the amplitude of the backward component of waves propagating in the top and bottom chains in the coupled system.The displacement of the two coupled harmonic chains are constrained and the direction of propagation of waves in the two-chain system are not independent of each other.For instance, at k=0, the antisymmetric mode is represented by a standing wave with the amplitude of a forward propagating wave and a backward propagating wave being identical.As k → ∞, ω → + βk, the terms √ − in equation ( 16) go to zero and only one direction of propagation is supported by the medium (first and third terms in Eq. ( 16)).For any other value of the wave number k, the elastic modes supported by the coupled chains are quasi-standing waves which enforce a strict relation between the amplitudes of superposed forward propagating wave and a backward propagating wave.We recall that the constraint on the amplitude of superposed waves forming the antisymmetric modes does not exist for symmetric modes.If one solves the classical Eqs. ( 1) and (2) using plane wave solutions as is done to obtain Eqs. ( 4) and (5), one does not find that the solutions are quasi-standing waves with correlation of the amplitudes along the forward and backward directions.Introducing the Dirac formalism (Eq.( 7)) enables us to reveal and to project the quasi-standing wave solutions supported by the two coupled chains on forward and backward propagating states.
These amplitudes can be measured directly by measuring the transmission coefficients of the top chain and bottom chain.The measured transmission coefficient of the top chain relates to two of the components of the 4x1 spinor (Eq.16).Similarly, the transmission coefficient of the bottom chain relates to the other two components.The superposition of elastic states given by equation ( 16) is therefore measurable without wave function collapse in contrast to what would be the case for the superposition of states of a true quantum system.
In equation ( 16) a 0 is an arbitrary constant.Because of the relationship between the components of the solution (16), it cannot be expressed, in general, as a tensor product of the solutions of two single chain systems.Table I can be summarized by writing the states of an individual single chain system as with s 1 and s 2 are taking on the values +1 or -1.Here, we express the tensor product of two single chain systems a and b: By inspection, one sees that equation ( 16) cannot always be written in the form of equation ( 17) but only in a few specific cases.e +ikx e +iω k t e −ikx e +iω k t e +ikx e −iω k t e −ikx e −iω k t For instance, when k = 0, then ω = 2δ and 16) becomes: When δ → 0 then ω → βk and √ + → 2 βk and √ − → 0. Equation ( 16) reduces to The 2x1 vectors 1 −1 and 1 0 are visualized as spinor components of the single chain system.
In summary, the state of the two-chain system is not separable in the tensor product Hilbert space of states of two single chain systems except for the special cases: k = 0 and δ → 0 (or equivalently k → ∞).The choice of this decomposition is driven by the measurability of both the states of single chain systems and two-chain system via measurement of transmission coefficients along the relevant chains.
IV. CONCLUSIONS
We consider a bipartite classical mechanical system composed of two coupled one-dimensional elastic chains whose elastic wave equations can be factored into a Dirac-like equation.Using the Dirac formalism reveals the true nature of elastic waves in this system, namely that they are quasi-standing waves.In the Dirac formalism, the elastic wave functions have spinor components which characterize the strict relation between forward and backward propagating components of quasi-standing waves.This system allows us to address the separability and non-separability of the states of the bipartite elastic system with respect to the choice of elastic subsystems into which it is partitioned.We show that the states of the coupled two-chain system, like any elastic system, can be readily decomposed in the spectral representation.The states of the coupled system span the tensor product space of the energy spectrum associated with wave vector k.The states of the two-chain system are therefore separable into states associated with a spectral representation.In contrast, it is also possible to express the states of the coupled system in the tensor product space of only two subsystems, namely two noninteracting systems composed of a single chain coupled to a rigid substrate.While some of the states of the bipartite system can be written as tensor products of states of the subsystems, there are many states that display correlations that do not allow such decomposition.These states exist in the tensor product Hilbert space of two single chain systems but are not expressible by a simple tensor product.Then, where does the non-separability come from when we can decompose the coupled chains system into two single chain systems.The spectral decomposition describes the orbital part of the wave function.We find new correlations in the directional representation that may lead to nonseparability in the spinor part of the amplitude, i.e., correlations between the directions of propagation.The degrees of freedom associated with direction of propagation lead to non-separability.The choice of the decomposition into the tensor product Hilbert space of two single chain systems is motivated by the possibility of measuring directly these degrees of freedom through measurement of transmission coefficients.
To illustrate the applicability of non-separability of the elastic waves, we may consider a two-qubit algorithm, 13 in the spirit of the Deutsch-Jozsa algorithm 14 that exploits non-separability.The intended outcome of this algorithm is to distinguish between the even or odd nature of all possible binary functions for two bit inputs. 13To identify whether a function is even or odd, one just needs to identify if the final state of the two qubit system is separable or not.The challenge for quantum systems lies now in the measurability of the final states.There is no unambiguous single measurement of entangled states of quantum systems.To distinguish between separable and non-separable superpositions of states, one needs to make multiple measurements and obtain a statistical representation of the superpositions.This drawback could be overcome by using the spinor states of the coupled two-chain system.When the orbital state of the elastic plane wave is defined by k = 0 and ω = 2δ, the spinor state of the system is separable.When the state of the elastic plane wave is defined by a general wave number and frequency, the spinor state of the coupled chains takes a non-separable form.The determination of separable or non-separable elastic states can be done unambiguously via measurement of transmission.In the first case, there is no transmission along the chains and in the second case, transmission can be detected.
The concept of non-separable elastic waves can be extended to three coupled chains, whereby the spinor state of a three mutually coupled chain system is defined in the 2 3 -dimensional tensor product Hilbert space of the three single chain systems.Most states of the three-chain system are not separable and measurements of these states can still be achieved via measurement of transmission coefficient along the three chains.Finally, extension to N mutually coupled chains would allow us to conceive the possibility of measureable non-separable spinor states in a 2 N -dimensional space.Such states let us envisage the possibility of storing, processing, and efficiently measuring information in elastic systems with exponential complexity.
ξ k where we use the direction switching operators S + = part of the wave function with creation and annihilation operators from the spinorial part defines the operator 1 2ω S + S − σ x .This operator represents the occupancy of one of the directions of propagation along the chain of mass and springs.Its Eigen values are given by:
TABLE I .
Spinor components of ψ k , solutions of equations (10a).e+ikx e +iω k t e −ikx e +iω k t e +ikx e −iω k t e −ikx e −iω k t
TABLE II .
k → 0 and k → ∞ limits of the spinor components of ψ k . | 4,958.8 | 2017-04-26T00:00:00.000 | [
"Physics"
] |
Mach-Zehnder Interferometer Biochemical Sensor Based on Silicon-on-Insulator Rib Waveguide with Large Cross Section
A high-sensitivity Mach-Zehnder interferometer (MZI) biochemical sensing platform based on Silicon-in-insulator (SOI) rib waveguide with large cross section is proposed in this paper. Based on the analyses of the evanescent field intensity, the mode polarization and cross section dimensions of the SOI rib waveguide are optimized through finite difference method (FDM) simulation. To realize high-resolution MZI read-out configuration based on the SOI rib waveguide, medium-filled trenches are employed and their performances are simulated through two-dimensional finite-difference-time domain (2D-FDTD) method. With the fundamental EH-polarized mode of the SOI rib waveguide with a total rib height of 10 μm, an outside rib height of 5 μm and a rib width of 2.5 μm at the operating wavelength of 1550 nm, when the length of the sensitive window in the MZI configuration is 10 mm, a homogeneous sensitivity of 7296.6%/refractive index unit (RIU) is obtained. Supposing the resolutions of the photoelectric detectors connected to the output ports are 0.2%, the MZI sensor can achieve a detection limit of 2.74 × 10−6 RIU. Due to high coupling efficiency of SOI rib waveguide with large cross section with standard single-mode glass optical fiber, the proposed MZI sensing platform can be conveniently integrated with optical fiber communication systems and (opto-) electronic systems, and therefore has the potential to realize remote sensing, in situ real-time detecting, and possible applications in the internet of things.
Principle of Operation
A typical MZI sensor based on optical waveguide consists of a laser source, an optical detection unit and an MZI structure with an evanescent field sensing window, as shown in Figure 1. The MZI structure consists of an input waveguide and an output waveguide, a beam splitter (the left Y-junction), and a beam combiner (the right Y-junction), as well as two straight waveguides between the two Y-junctions as the sensing arm (with a sensitive window) and the reference arm respectively. In the operation, the monochromatic and polarized light from the laser source is coupled into the input waveguide and split equally at the beam splitter. Then the two guiding modes propagate a certain distance along the sensing arm and the reference arm respectively, and recombine at the beam combiner. Based on evanescent field sensing, a phase difference Δφ between the sensing arm and the reference arm occurs when the effective refractive index of the guiding mode in the sensing arm is changed by the biochemical reaction in the sensitive window, resulting in an intensity modulation caused by the interference of the two arms at the waveguide output. Measuring the interference intensity at the output waveguide, the biochemical reaction in the sensitive window is able to be detected.
where, λ represents the operating wavelength, Neff represents the effective refractive index of propagating mode of waveguide, L is the length of the sensitive window on the sensing arm. Assuming the detection target is the refractive index change of bulk solution, the sensitivity of MZI-based sensor can be expressed as: where, Δn represents the refractive index change of the sensitive region, and ΔP is the normalized output power change of the MZI sensor responding to a given Δn. As the partial sensitivity of the interference measurement and evanescent field sensing are defined as ΔP/Δϕ and ΔNeff/Δn respectively, the Equation (2) can be rewritten as: It can be found that the sensitivity of the MZI-based sensor is determined by the length of the sensitive window (L), and the partial sensitivity of the interference measurement and evanescent field sensing.
Intensity of Evanescent Field
The fundamental principle of optical waveguide MZI sensor is based on the evanescent field sensing. According to the Goos-Hanchen effects, the evanescent field is a fraction of optical field that extends to the cladding layer and the substrate layer of the waveguide. In general, there are two types of evanescent field sensing, homogeneous sensing, and surface sensing [14]. The homogeneous sensing and surface sensing for SOI waveguide with large cross section are shown in Figure 2a,b respectively. For the homogeneous sensing, the effective index variation of the propagating mode is produced by the change of the refractive index in the sensitive region. While for the surface sensing, the effective index variation is produced by the change of the thickness of the ultra-thin sensitive layer which is immobilized on the waveguide surface. The thickness of the surface sensitive layer is denoted as t (shown in Figure 2b), which is changed at a small range, about several nanometers. Homogeneous sensing is generally used to detect the concentration change of gas or liquid in the entire sensitive region, and the surface sensing is often applied to detect protein, DNA, virus, and bacteria with the help of immobilized receptor molecules [15]. The schematic diagram of sensitive region and sensitive layer at silicon-on-insulator (SOI) rib waveguide with large cross section. (a) Homogeneous sensing, the light olive region is the sensitive region; (b) Surface sensing, the blue region represents the sensitive layer, and its thickness is denoted by t.
Obviously, the greater the intensity of the evanescent field is, the more sensitive the sensor both in the case of homogeneous sensing and surface sensing will be. The intensity of the evanescent field can be represented by the confinement factor Γs, which is the ratio of the electric field intensity in the sensitive region or the sensitive layer to the entire mode distribution of the guiding mode [15], defined as Equation (4). (4) Generally, in order to achieve low-loss propagation and avoid negative influences due to multimode transmission or cross-polarization interference, the optical waveguides employed in the senor must possess single mode and single polarization. In this article, the SOI rib waveguides with large cross section will be confined by single mode conditions. Once the mode distributions are solved by mode solver programs, such as finite element method (FEM) and finite difference method (FDM), the intensity of the evanescent field can be calculated through area integration. Therefore, taking the maximization of the intensity of the evanescent field as the target, the optimization of waveguide section dimensions can be realized.
Optimization of Waveguide Section Dimensions
As an example, suppose the sensor based on SOI rib waveguide with large cross section is used to detect the concentration of glucose solution and the refractive index nc of the analyte solution is in the vicinity of 1.33. In order to guarantee high coupling efficiency with the standard single-mode fiber, the total rib height (H) of SOI rib waveguide is set to be 10 μm, and the operating wavelength is selected at 1550 nm so that the refractive index of silicon and SiO2 are 3.476 and 1.444, respectively. Considering too small a rib width (w) will reduce the restriction of the rib structure on the guiding mode and too big a rib width (w) will be difficult for system integration (such as the waveguide bends and branches analyzed in next section), the rib width (w) of SOI rib waveguide is set to be in the range of 2.5 μm to 10 μm.
The modes of rib waveguide can be denoted as HEnm or EHnm [16], where n = 0, 1, 2…, and m = 0, 1, 2…. HE mode and EH mode are commonly known as quasi-transverse-electric mode and quasi-transverse-magnetic mode respectively. Employing a strict single-mode condition [17], all the SOI rib waveguides with large cross section only support the fundamental guiding modes for each polarization, i.e. HE00 and EH00. Using the full-vector finite difference method (FDM) [18] to solve the fundamental guiding modes with different dimensions (including the outside rib height h and the rib width w), the intensity of the evanescent field for homogeneous sensing can be calculated, as shown in Figure 3. It can be seen from Figure 3, for an SOI rib waveguide with the total rib height of 10 μm, the evanescent field intensity of homogeneous sensing exhibits a maximum of 7.436 × 10 −3 % for HE polarization, corresponding to h = 5 μm and w = 4.5 μm, and 9.407 × 10 −3 % for EH polarization, corresponding to h = 5 μm and w = 2.5 μm.
Using the same method, dependence of the evanescent field intensity for surface sensing on rib width (w) with different outside rib height (h) also can be calculated. Assuming the refractive index of the sensitive layer nm = 1.45 and the thickness of the sensitive layer t = 10 nm, the evanescent field intensity for surface sensing of the SOI rib waveguide with the total rib height of 10 μm possesses a maximum of 5.089 × 10 −3 % for HE polarization and 7.924 × 10 −3 % for EH polarization, both corresponding to h = 5 μm and w = 2.5 μm. Both for homogeneous sensing and surface sensing, it is evident that EH polarization of the SOI rib waveguide with large cross section is significantly more sensitive than HE polarization due to the larger evanescent field intensity.
According to the variational theorem for dielectric waveguides, an analytical method can be used to estimate the partial sensitivity of evanescent field sensing based on the calculated evanescent field intensity [17,18]. Therefore, the partial sensitivity of homogeneous sensing (ΔNeff/Δn) possesses a maximum of 2.86 × 10 −3 for HE polarization (corresponding to h = 5 μm, w = 4.5 μm) and 3.6 × 10 −3 for EH polarization (corresponding to h = 5 μm, w = 2.5 μm), and the partial sensitivity of surface sensing (ΔNeff/Δt) exhibits a maximum of 4.89 × 10 −6 nm −1 for HE polarization and 7.61 × 10 −6 nm −1 for EH polarization (both corresponding to h = 5μm, w = 2.5μm). Therefore, the result of the optimization is that the waveguide mode is the fundamental EH mode, and the cross section dimension of the SOI rib waveguide is H = 10 μm, h = 5 μm, w = 2.5 μm. The simulation of the optimized mode is shown in Figure 4.
Conventional Implementations
In the waveguide MZI sensor, the evanescent field sensing is read out by the MZI configuration. The most critical components of the MZI configuration are the beam splitter and the beam combiner, which are waveguide branches and are often identical. For the conventional MZI configurations, there are three structures to realize waveguide branches, as listed in Table 1. The data in this table show the technical parameters and the minimum required length of a single branch to fulfill the separation distance of d = 50 μm between the two straight SOI rib waveguides with large cross section. Thus for the SOI rib waveguide with a large cross section, the conventional implementations of waveguide branches lead to an overlong structure which is difficult to be realized.
Multimode interference
Self-imaging effect [19] When d = 50 μm, The minimum length of a single branch: L 0 > 12 mm.
Trench-Based Bend and Branch
Studies have shown that the bends and branches of waveguides can be realized by using the medium trench, slot, or photonic crystal [20][21][22][23][24]. In this paper, media-filled trenches are used to achieve waveguide bends and branches for an SOI rib waveguide with a large cross section, as shown in Figures 5 and 6, respectively. The gray area in these figures represents the filling medium, which can be air, SU8 or other refractive index matching fluid. Figure 6. T-shaped branch geometry of SOI rib waveguide with large cross section.
Due to the large dimension of the SOI rib waveguide, the computational memory requirement and time-consumption of a three-dimensional finite difference-time domain (3D-FDTD) simulation are huge. The usual way to overcome this issue is to simplify the 3D structure to 2D by using effective refractive index method (EIM), and then simulate the interested structure using two-dimensional finite difference-time domain (2D-FDTD) [25]. In this paper, a 2D-FDTD method with a perfectly matched layer (PML) boundary condition [26] is employed to numerically simulate the above mentioned waveguide bends and branches.
The medium trench shown in Figure 5 corresponds to a corner mirror, which can change the propagation direction of the waveguide according to total internal reflection. In general, the length and width of the medium trench (L and W as shown in Figure 5) should be large enough to reflect instead of to transmit more mode energy. The parameter D is defined to account for the Goos-Hanchen shift compensation, and it is positive when the trench interface moves away from the bending center (corresponding to the case shown in Figure 5, D > 0). When the trench interface precisely passes through the bending center, D = 0. Setting the width and length of the air trench at 10 μm and 30 μm respectively, i.e. L = 30 μm, W = 10 μm, the electric intensity map of a 90° air-trench bend is shown in Figure 7.
The influence of the parameter D on the bend efficiency of the air-trench bend exhibits as a curve with oscillatory variation in a very small range, as shown in Figure 8. Due to the strong ability of light constraint, there is a high bend efficiency for SOI rib waveguide with large cross section, even when D = ±500 nm. The reason for this oscillatory variation is that there is an angle between the air-trench and the meshing direction in 2D-FDTD simulation to avoid oblique incident light, and the interface of the air-trench and the waveguide is jagged as shown in Figure 9. It can be verified that this effect will be attenuated if finer mesh sizes are adopted. Similarly, the electric intensity map of the asymmetric T-shaped air-trench branch with the trench width and length of 97 nm and 30 μm respectively is shown in Figure 10. Based on 2D-FDTD simulation with the mesh grid of 5 nm, the reflection efficiency is 0.497357 and the transmission efficiency is 0.502518. Figure 10. The electric intensity map in a plane 4 μm above the SiO2 layer for an asymmetric T-shaped air-trench branch at a wavelength of 1550 nm. The SOI rib waveguide with a large cross section possesses total rib height of 10 μm, outside rib height of 5 μm, and rib width of 2.5 μm. The guiding mode is the fundamental EH mode. The width and length of the air trench is 97 nm and 30 μm respectively.
It is found from the simulations that the splitting efficiencies, including the reflection efficiency, the transmission efficiency and the total transmission efficiency, are functions of trench width (w), as shown in Figure 11. Both for air-filled trench and SU8-filled trench, the transmission decreases and the reflection increases as the trench width increases. With the trench width of 96 nm for air-trench branch and 112 nm for SU8-trench branch, a splitting ratio of 50%:50% is achieved. Figure 11. Splitting efficiency as a function of trench width. A 50%:50% splitting ratio is achieved with 96 nm air-trench width or 112 nm SU8-trench width.
Supposing the trench branch is filled with index matching fluid and has a trench width of 120 nm, the influence of the refractive index of the index matching fluid is shown in Figure 12. As expected, the higher refractive the index of the filled material, the smaller the reflection efficiency and the larger the transmission efficiency. Thus, it can be concluded that the higher refractive index of the filled material results in the larger trench width for a 50%:50% splitting ratio. In addition, using index matching fluid as the filled material for the trench branch can prevent pollution of dust or impurities in the air. Overall, the trench-based bends and branches can be used for SOI rib waveguide with large cross section, and their performances are obviously superior to the traditional bending and branching structures, such as the higher transmission efficiency and the shorter length of a single branch. Note that the simulation errors mainly come from two aspects, the coupling between modes and the out-of-plane scattering loss, in the process of simplifying the 3D model to 2D [27]. Because the SOI rib waveguides discussed in this paper are confined by the single-mode condition and have strong capability of light constraint, the influences caused by these two factors are very small, and thus the error of the above results obtained from 2D-FDTD simulations is insignificant.
Proposed MZI Structure
A new MZI structure based on an SOI rib waveguide with a large cross section that consists of two trench-based waveguide bends, two trench-based waveguide branches, and two straight waveguides is proposed as shown in Figure 13. These two parallel straight waveguides can act as the reference arm and the sensing arm of the MZI sensor, and their spacing is denoted by d. The parameter S represents the horizontal distance of the two waveguide branches. Compared with the traditional MZI configurations, the proposed configuration possesses two out ports, which can be used solely, or synchronously as mutual reference. Figure 13. Schematic of the MZI using media trenches based on SOI rib waveguide with large cross section.
Using a 2D-FDTD simulation, the optical power map of such a trench-based MZI structure with d = 50 μm, S = 80 μm is shown in Figure 14. The guiding mode is the fundamental EH mode of the SOI rib waveguide with H = 10 μm, h = 5 μm, w =2.5 μm at wavelength of 1550 nm. It shows that only one output port has power output due to the interference effect of the two identical guiding modes in the sensing arm and the reference arm. The transmission of the electric field components in the Z direction Ez (out of plane) can be used to clearly explain this interference case, as shown in Figure 15. Figure 16a,b respectively show the normalized power of the SOI rib waveguide mode that propagates along the sensing arm and the reference arm. This normalized power is the ratio of the power on the transmission cross section to the input power of the light source. It can be seen that there are unstable regions at the waveguide bends and branches, which are caused by the interference of the waveguide modes. In particular, the normalized power of the interference enhancement is more than 1. It is found that the length of the unstable transmission segments at the waveguide bends or branches are less than 20 μm, which is a great advantage compared to the conventional implementations of MZI. In the ideal situation (the splitting ratio of every branch in the MZI is 50%:50%), the normalized output power of each output port is a function of the phase difference (ΔΦ) between the sensing arm and the reference arm, as shown in Figure 17. A complementary relationship between the two output ports can be found. The normalized output powers of the output port "output 1" can be expressed as Equation (5). There, ΔΦ0 is the initial phase difference caused by machining error or other unbalanced factor.
[ ] 0 1 = 1 cos( ) 2 P + ΔΦ+ΔΦ (5) Figure 17. The normalized output power in each output port as a function of the phase difference between the sensing arm and the reference arm.
MZI Sensing Platform
According to the above analyses and results, the trench-based MZI structure exhibits better performance than traditional configurations. The schematic of the MZI sensing platform based on SOI rib waveguide with large cross section is shown in Figure 18. There is a sensitive window at the sensing arm where selective biochemical sensitive material is used for a specific application. Due to the high coupling efficiency of the SOI rib waveguide with large cross section and the standard single-mode glass fiber, the input and output ports of the MZI sensing platform can be conveniently connected to a laser source and light power detection unit, and remote measurement based on fiber-optic communication can be achieved. With the help of a simple tapered mode converter, the butt coupling of input and output waveguides with standard single mode fiber can be easily realized. A set of simulations show that more than 80% of the coupling efficiency is very easy to achieve. In addition, the output signal can also be detected by integrated photoelectric detectors, as shown in Figure 18. Figure 18. Schematic of the MZI sensing platform based on SOI rib waveguide with large cross section.
As an example, an MZI sensor with the fundamental EH-polarized mode of SOI rib waveguide with H = 10 μm, h = 5 μm, w = 2.5 μm at wavelength of 1550 nm, when L = 10 mm and d = 50 μm, a homogeneous sensitivity of 7296.6%/refractive index unit (RIU) can be obtained according to Equations (3) and (5). Supposing the resolutions of the photoelectric detectors connected to the output ports are 0.2% of the optical power for the intensity measurement, the MZI sensor can achieve a detection limit of 2.74 × 10 −6 RIU. This detection limit of refractive index is smaller than that obtainable by Si3N4 rib waveguides (7 × 10 −6 RIU) [28] and close to that reported in the previous works [29,30]. More importantly, employment of the SOI rib waveguides with large cross sections make the device match well with the communication glass fibers, and the size of the MZI's branches are very small, so that the entire sensing platform can be very compact.
Discussion
Employing with the media-filled trenches, the MZI configuration based on the SOI rib waveguide was realized in micrometer scale, thus a micron-sized and compact biochemical sensing platform based on an SOI rib waveguide with a large cross section was obtained, which is the main advantage of this work. Moreover, high coupling efficiency with standard single-mode glass optical fiber is the most important advantage of SOI rib waveguide with large cross section, which enables the waveguide sensors to integrate with optical fiber communication systems and (opto-) electronic systems, and therefore to realize remote sensing, in situ real-time detecting and possible application in the internet of things.
According to the above analysis, the MZI sensor can perform bulk sensing with high sensitivity, which can be used to as chemical sensor. Due to the simple MZI configuration and strong adaptability of evanescent field sensing, the MZI sensing platform can also be used to detect biological reactions as long as suitable receptor molecules are immobilized at the waveguide surface in the sensitive window.
To achieve high sensitivity, the trenches of the waveguide branches in the MZI configuration needs to be very narrow, which at the present stage may involve expensive microfabrication processes such as electron beam lithography (EBL) and deep reactive ion etching (DRIE). Therefore a compromise should be made between the sensitivity and the cost of the process. Alternately, employment of filled medium with higher refractive index can lower the difficulty of the process without decreasing the sensitivity. From the analysis of this paper, it can be speculated that the surface roughness of processing and the change of the external environmental temperature in real measurement have effects on the splitting efficiencies of the MZI's branches, but the further effects on the performance of the sensing system need to be further researched in the experiments, which is the next step of this work. In addition, due to the strong anti-interference capability of MZI configuration, these effects might be very small. Fortunately, due to the mass production feature of the SOI process, a very low cost for each sensor chip can be achieved as long as a large market is found.
Conclusions
An MZI biochemical sensing platform based on an SOI rib waveguide with a large cross section is proposed in this paper. The optimization of the cross section dimensions of the SOI rib waveguide is performed through FDM simulations with the target of maximizing the evanescent field intensity. Medium filled trenches are employed to realize the MZI configuration based on the SOI rib waveguide. The performances of the MZI sensing platform are simulated by using 2D-FDTD method.
The optimization of the SOI rib waveguide is that the guiding mode is the fundamental EH mode, and the cross section dimension is that the total rib height H = 10 μm, the outside rib height h = 5 μm and the rib width w = 2.5 μm. When the length of the sensitive window L = 10 mm and the spacing distance between the sensing arm and the reference arm d = 50 μm, the MZI sensor based on SOI rib waveguide with large cross section at an operating wavelength of 1550 nm can achieve a homogeneous sensitivity of 7296.6%/RIU. Supposing the resolutions of the photoelectric detectors connected to the output ports are 0.2%, the MZI sensor can achieve a detection limit of 2.74 × 10 −6 RIU. | 5,754 | 2015-08-28T00:00:00.000 | [
"Physics"
] |
Machine learning based luminance analysis of a µLED array
In the past years, the development of µLED arrays gained momentum since they combine the advantages of LEDs, such as high brightness and longevity, with the high resolution of a micro-scaled structure. For their development, spatially resolved measurement of luminance and color of single µLEDs and the entire light-emitting surface are usually analyzed as they quantify the visual perception. However, studying individual µLEDs is time-consuming to measure and evaluate, while examining the entire light-emitting area suffers from interference from non-functioning µLEDs. This paper presents a method to perform both analyzes with a single measurement employing unsupervised machine learning. The results suggest that a precise reconstruction of the µLEDs and a more accurate characterization µLED arrays is achieved.
Introduction
Over the last decades, light-emitting diodes (LEDs) were established as a key light source in the consumer market, such as for general lighting or automotive. The accelerators for the far-reaching influence of LEDs are their durability, even under harsh conditions such as extreme temperatures, and their high-efficiency and luminance. However, for applications requiring high pixelation like displays, the high-performance LEDs are mostly too big because research has focused on the optimization of 1 mm 2 large high-performance LEDs 1, 2 .
Shrinking those LEDs into the micro-scale -edge length below 100 µm-allows to overcome this issue and should theoretically increase their efficiency. Already in 2000, Jin et al. published the first realizations of micro light-emitting diodes µLED 3 , and since then, numerous publications regarding this topic were submitted.
One advantage of µLEDs is the possibility of combining them to an array-structure, which can then be linked with an underlying electrical control unit. This two-dimensional stringing-together of µLEDs results in a µLED array. Such structures should have a higher illuminance and homogeneity than a single µLED, as well as provide a high brightness, contrast, resolution, and durability 4,5 . In combination, they may become superior against already established pixelated light sources such as organic LEDs (OLEDs) or liquid crystal (LC) based ones 2,6 . Furthermore, LED arrays gained much momentum in the industry as well 6 . However, there is no published data that the established manufacturing techniques, mass transfer-based or the monolithic-based fabrication, achieved a yield of close to 100% 7 . Consequently, most of the produced µLED arrays will contain non-emitting and therefore defect µLEDs.
Classical approaches to characterize the luminance of the light-emitting surface (LES), such as averaging over the entire area, do not distinguish between functional and defect, resulting in an underestimation of the actual µLED array behavior. If a significant fraction of nonfunctional µLEDs occurs, this becomes a problem in the development process. For instance, the "noise" created by the defect µLEDs could prevent an evaluation of a design change. Each µLED (pixel) should be classified, and defect ones may not be considered for the final analysis to overcome this blurring effect.
In this paper, a machine learning-based algorithm is applied to overcome this issue. Therefore, individual µLEDs are located in a luminance image of a µLED array and then classified as functional or defect with an unsupervised learner (KMeans). As a result, the behavior of the µLED array is studied without the blurring effect of the non-emitting µLEDs. The implemented analysis is evaluated by three performance measures including the reconstructed µLED size, the confusion matrix of the underlying classifier, and the noise reduction performance. In order to achieve this, a luminance camera with a resolution of 2448 × 2050 pixels and a white light-emitting µLED array with a total of 60 × 60 µLEDs are used. Each µLED of the light source has a size of 40 µm × 40 µm and can be addressed separately. A lens with a magnification factor of two is mounted on the camera. Due to the luminance camera pixel's size of 3.45 µm, a single µLED is represented by approximately 23 camera pixels.
Results
The developed analysis is tested with four luminance images, which are shown in Figure 1. The first image represents the Reference state, where all µLEDs of the array are turned on. The non-uniform luminance distribution originates from the underlying electronics and will not be discussed further. On the other images (Pixel map one to three), some µLEDs are turned-off on purpose to simulate a nonfunctional state. These pseudo-defect µLEDs were selected for each Pixel map randomly and represent a yield of 80%, hence 20% of the µLEDs are turned-off. Furthermore, the four luminance images are the direct outputs of the luminance camera, which explains the black bars framing the array structure. In addition, the sample was rotated to increase the difficulty and to stress the robustness of the method. The Pixel maps shown in Figure 1 are the inputs of the underlying analysis pipeline, which is illustrated in Figure 2. The first step of the pipeline performs a projective transformation to compensate for a tilted and rotated sample. Then, the grid reconstruction scopes to reassemble the array structure, so that individual µLEDs can be localized. This result is exploited by the Pixel analysis that studies the luminance and CIE color coordinates of each reconstructed pixel. The acquired data set is then guided into the machine learning part of the pipeline, which utilizes a principal component analysis (PCA) and an unsupervised learner (KMeans) to classify functional and nonfunctional µLEDs. Finally, the Array level analysis evaluates the entire light-emitting surface (LES), while considering the information about the functionality of the individual µLEDs. Note, each step of the pipeline will be discussed in more detail in the subsequent section. In total, three performance measures can be extracted from the analysis pipeline (see Fig. 2): Firstly, the reconstructed pixel size, secondly the confusion matrix of the pixel classification, and finally the mean luminance of the LES. Since the luminance image provides spatial information of the µLED array, the pixel grid framing the individual µLEDs can be extracted. Figure 3 shows the result of the underlying grid reconstruction algorithm for Pixel map 1. In addition, Figure 3 confirms that the applied projective transformation can correct the initial rotation of the sample. The method is even capable of reconstructing the grid at locations where several pixels are "defect". The quality of the extracted grid can be measured by its pixel size since the pixel size of the grid ideally corresponds to the number of camera pixels representing a µLED. Equation (1) shows the mean grid's pixel size d for all three Pixel maps (compare Figure 1): The standard deviation of the mean value is intentionally neglected because the uncertainty is less than 0.5 px. As a result, it is concluded that the pixel grid reconstruction is capable of consistently reconstructing the proposed pixel size. Note that pixels at the edges of the LES are currently not considered to diminish the influence of boundary effects.
Functional pixels Pixel grid
Cluster of defect pixels Figure 1).
Next, the reconstructed grid organizes the input luminance image (compare Figure 1) into individual pixel areas containing information about a certain µLED. Hence, from a single luminance image, information about hundreds of µLEDs can be extracted and used for statistical analysis. The maximum, mean, minimum, and standard deviation of the luminance as well as the mean CIE x and CIE y color coordinates are calculated for each pixel including the "defect" ones. Consequently, this procedure leads to a statistical data set that not only represents a base for a profound analysis of the µLEDs behavior but also serves here as training data for the unsupervised learner. In particular, the classification of defect pixels is of interest as elaborated in the introduction. The achieved performance of the used k-Means-algorithm is illustrated in the confusion matrices shown in Figure 4. The confusion matrices reveal for each Pixel map (compare Figure 1) the corresponding prediction accuracy for the classification functional / defect. Note that the general procedure of the classification step is explained in detail in the subsequent section. According to the matrices, the classifier predicts the pixel's status with a percentage of ≈ 99.5% correct. In the other case, the lower left matrix elements indicate that functional pixels are falsely classified as defective in less than 0.5%. Moreover, the obtained true positive and negative rates of about 80% and 20%, respectively, are consistent with the originally selected yield of 80%. In general, from the confusion matrix further evaluation quantities can be extracted including, but not limited to, the precision score P, recall score R, and the F1-score. The precision score measures the ability of a classifier to prevent the positive labeling of something that should be negative. In the case of this classifier, it quantifies the risk of predicting a defect pixel as functional. The recall score represents the capability of a classifier to identify all positive elements, here the ability to notice functional pixels. The F1-score considers both precision and recall score to give a balanced quantity 8 . Note, for all scores 0 is the worst and 1 is the best outcome. The scores can be calculated as follows: T P =True positiv, FP = False positiv, FN = False negative The calculation of those measures confirms the performance of the k-Means-based classifer: Finally, the results can be exploited to extract information about the LES even with nonfunctional pixels. As elaborated in the introduction, averaging over the LES, which includes nonfunctional pixels, lead to an underestimation of the actual performance of the LES. In order to quantify the improvement of the presented analysis, the mean luminance of the Reference state (left in Figure 1 The index raw indicates the mean luminance, including the defect pixels, whereas the index dn marks the mean luminance only for functional µLEDs. Consequently, the proposed analysis pipeline is capable of reconstructing a more representative value for the mean luminance compared to the classical approach.
Conclusion
The presented analysis technique is capable of reconstructing the pixel grid with high precision, even when clusters of µLEDs are not working. Further, the machine learning-based classification of the current pixel status also shows a high overlap with the simulated yield. In comparison, to a classical classification approach using, for instance, an arbitrary luminance threshold, no intense parameter tuning is required. Finally, exploiting the knowledge about the µLED status enhances the analysis of the entire LES and allows to infer the actual behavior of the LES more preciously as a classical approach. However, since the current pipeline uses an unsupervised learner (KMeans), it could behave differently on different µLED arrays. Swapping to a supervised learner such as a RANDOMFOREST could reinforce the robustness of the analysis, however, requires the presence of a labeled data set, which is time-intense to accumulate. Moreover, being able to classify thousands of µLEDs within a single measurement also offers the possibility to study the statistical distribution of each quantity for both functional and nonfunctional µLEDs.
Luminance measurement
The used setup, shown in Figure 5, employs the luminance camera LMK5-5 of the manufacturer TechnoTeam. The camera offers a resolution of 2448 × 2050 pixels and is equipped with a filter wheel to consider the V (λ ) curve and color-weight functions for the measurements of luminance and color, respectively. Besides, the luminance camera uses a lens with a magnification factor of two. On the sample side, a white light-emitting µLED-array with a total of 60 × 60 µLEDs, where each
Sample
Heat sink has size of 40 µm × 40 µm, is used. Due to the luminance camera pixel's size of 3.45 µm, a single µLED is represented by approximately 23 × 23 camera pixels. Accordingly, the setup is theoretically capable of resolving smaller µLEDs. However, the significance of the pixel-level analysis diminishes with fewer camera pixels per µLED. For instance, local anomalies on a µLED would be averaged out with fewer camera pixels, yielding a less accurate statistical representation of the µLED array's behavior. Furthermore, the grid reconstruction reaches a resolution limit for fewer camera pixels. Tests with a narrowing lens (size was halved) indicate that the lower threshold for the algorithm lays at a single µLED representation of at least with 10 × 10 camera pixels, beyond this point the accuracy of the grid can diminish.
Image correction
The projection transformation is a technique to project an input image into an equivalent image, keeping its properties using a linear transformation. A deeper insight in the mathematical description of the method is given in reference 9 . Figure 6 shows how this transformation removes projective distortions (tilting, rotation) from an image. Note that tilting can not only occur from a miss-aligned camera but also from the µLED array itself. The perspective transformation is performed using CV2.GETPERSPECTIVETRANSFORM and CV2.WARPPERSPECTIVE, which are implemented in the PYTHON package CV2 (based on OPENCV) 10,11 . The edges of the LES are located in the luminance image by thresholding the input image with CV2.THRESHOLD and then routing the output into the contour locator CV2.FINDCONTOURS. Figure 6. Illustration on how a projection transformation can remove tilting and rotation 9 .
Grid reconstruction
After ensuring an adequate alignment, the pixel grid is reconstructed by projecting the luminance image on the x-and y-axis, which can be formally written as: As indicated by Figure 7 the values for Σ x L(x i ) and Σ y L(y j ) are significantly smaller at the edge of two µLEDs, leading to minima. Through the minima locations in the projections, it is possible to detect the pixel edges and, therefore, localize the pixel grid. Moreover, since the minima are independent of the µLEDs itself, the presented method can reconstruct the pixel grid even at positions where multiple pixels are defective. Figure 3 shows the reconstructed grid for the first Pixel map. Remarkable is that the pipeline reconstructs the pixel position even for a cluster of nonfunctional pixels correctly.
Pixel classification
With the pixel location, a pixel-level analysis can be performed, whereby the exact type of analysis can be adapted for different use cases. In the case of the subsequent pixel classification, the mean, maximum, minimum, and standard deviation of the pixel's luminance are extracted. Additionally, the mean color coordinates CIE x, and CIE y are also calculated. In total, six parameters describe a single µLED (see Figure 8), creating a six-dimensional parameter space. This parameter space is exploited for machine learning. Although this would be ideal for training a supervised learner such as a RANDOM FOREST, the lack of labeled data motivated the use of an unsupervised learner. From the variety of promising methods, such as Gaussian Mixture Models, a simple k-Means-algorithm was chosen because it offers for this classification task a precision of over 99%. In addition, its simplicity allows people with less background in machine learning to quickly implement the presented analysis pipeline. Figure 8. Visualization of the feature set for each reconstructed pixel leading to a six-dimensional parameter space.
After extracting the information shown in Figure 8 for each pixel, the data is standardized with SKLEARN.PREPROCESSING.SCALE. Subsequently, a principal component analysis (PCA) aims to increase the information density of the feature set by reducing the number of the feature space to two. A PCA tries to find new axes in the input parameter space, which maximize the variance of the data. These axes correspond to the eigenvectors with the largest eigenvalues extracted from the samples input covariance matrix. After determination of the eigenvectors, the input data is projected onto these new axes 12 . The reference 13 provides a mathematical description. On the software side, the implementation SKLEARN.DECOMPOSITION.PCA of the PYTHON package SCIKIT-LEARN is applied 14 .
The outcome of the PCA is then guided into the actual classifier, which is a k-Means-algorithm. This algorithm tries to classify the input data into k clusters by minimizing the squared distance of all data point to the cluster centers. In the context of this paper, k equals two. A more detailed description can be found in reference 13 . For the analysis, the implementation SKLEARN.CLUSTER.KMEANS of SCIKIT-LEARN with a random seed of eight and n init = 100 (number of repetitions) is utilized. | 3,750.6 | 2021-01-01T00:00:00.000 | [
"Physics"
] |
A BLACK BOX ANALYSIS OF WEBRTC MOUTH-TO-EAR DELAYS A BLACK BOX ANALYSIS OF WEBRTC MOUTH-TO-EAR DELAYS
of Service (QoS). The internet was not designed with real-time capabilities in mind, therefore QoS support is required for some applications. This may be achieved in several ways including application and network configuration. Due to the time sensitivity of the voice packets, QoS is of exceptional importance when talking about VoIP applications, including WebRTC. Three main QoS indicators of the performance of VoIP applications are as follows: network jitter, packet loss rate and mouth-to-ear delay [4]. At the sender side, delays include encoding and packetization. These will depend on factors such as codec and frame size used. Other delays on sender side include operating system, sound card and NIC serialization delay. While being transmitted over the network, packets are subject to congestion delays at all intermediate routers as well as baseline propagation and serialization delay. At the receiving point packets experience similar delays to those at the sending
Introduction
VoIP uses IP based networks (either public or private) in order to transmit digitized voice between two or more endpoints in real-time.The development of VoIP has eased the cost constraints of long-distance communications as well as completely revolutionized the way we think about real-time communications (RTC).Modern day RTC provide a rich variety of functions including, but not limited to multi-point voice calls, video conversations or data and screen sharing.
In May 2011 Google released a product called WebRTC under an open-source license.From that point, some of the world's leading standardization bodies have been combining forces to provide guidelines and standards for native VoIP support within browsers.The World Wide Web Consortium (W3C) works on WebRTC APIs, which define JavaScript APIs and HTML elements necessary to develop WebRTC communications.The Internet Engineering Task Force and their RTCweb (Realtime communications in WEB-browsers) team are concerned with underlying standards.The WebRTC project by Google, Mozilla and Opera is an open framework that allows browsers to be RTC ready.With only just a few lines of JavaScript code, developers can set-up a Video or Voice over IP (VoIP) sessions.The aim of this research is to quantify M2E delays introduced by WebRTC applications under varying application configuration settings.Firstly, baseline M2E delays are measured using standard WebRTC settings (i.e.Opus Codec with 20 ms packet size and 48000 sampling rate).Once the baseline delays are established, the impact of various Opus settings -namely packet size and encoding rate as well as other WebRTC provided codecs is examined.
The rest of the paper is organized as follows: In Section 2 the importance of mouth to ear delay (M2E) as a QoS metric is discussed.Section 3 briefly reviews WebRTC.Section 4 and 5 describe the experimental test-bed and experimental results.Finally, Section 6 concludes the paper and suggests some future work.
Importance of M2E delay as QoS metric
One of the most important aspects of Real-Time communications is Quality of Service (QoS).The internet was not designed with real-time capabilities in mind, therefore QoS support is required for some applications.This may be achieved in several ways including application and network configuration.Due to the time sensitivity of the voice packets, QoS is of exceptional importance when talking about VoIP applications, including WebRTC.Three main QoS indicators of the performance of VoIP applications are as follows: network jitter, packet loss rate and mouth-to-ear delay [4].At the sender side, delays include encoding and packetization.These will depend on factors such as codec and frame size used.Other delays on sender side include operating system, sound card and NIC serialization delay.While being transmitted over the network, packets are subject to congestion delays at all intermediate routers as well as baseline propagation and serialization delay.At the receiving point packets experience similar delays to those at the sending
A BLACK BOX ANALYSIS OF WEBRTC MOUTH-TO-EAR DELAYS A BLACK BOX ANALYSIS OF WEBRTC MOUTH-TO-EAR DELAYS
Oliwia Komperda -Hugh Melvin -Peter Pocta * Due to the recent shift toward cloud based computing, some of the world's leading standardization bodies have combined forces to provide guidelines and standards for native implementation of RealTime Communication (RTC) in the browsers.The World Wide Web Consortium (W3C) works on WebRTC APIs [1], while the Internet Engineering Task Force (IETF) is concerned with underlying standards [2].Their efforts led to the development of WebRTC project by Google, Mozilla and Opera, which is an open framework that allows browsers to be RTC ready [3].In this paper, we examine WebRTC from a Quality of Service (QoS) perspective, focusing on Mouth-to-Ear (M2E) delays under various application configurations.
Test Design
M2E delay is one of the crucial factors of QoS.Identifying the extent of delays introduced by WebRTC application could help pinpoint flaws in the current WebRTC implementation and identify areas of the project that require improvements.The questions addressed by this paper are as follows: • How do various codecs impact on WebRTC M2E delay?• For Opus, how does packetization size impact on WebRTC M2E delay?• For Opus, how does encoding rate impact on WebRTC M2E delay?
A local server was used to enable a peer-to-peer WebRTC connection.Ping was used to estimate network RTT delay both in advance and during tests as our primary interest was in application performance.The SDP was used to modify the various application settings.An oscilloscope attached to both input and output acoustic interfaces ensured the most accurate results possible.The test bed is depicted in Fig. 1. point, as well as extra delays caused by jitter buffer [5, 6 and 7].The ITU Telecommunication Standardization Sector (ITU-T) provides some guidelines in the area of telecommunications.More specifically, ITU-T Recommendation G.114 [8] for mouthto-ear delay outlines that delays of 0-150 ms are acceptable to most users, delays of 150-400 ms are acceptable, but will impact on the quality of call, while delays of over 400 ms are generally unacceptable.Those values are only applicable if relevant echo management techniques are employed.
Delays can have a drastic impact on the interactive nature and thus the QoS for VoIP.More generally, there is a growing awareness of the importance of deterministic timing for many diverse application areas, including real-time communication (RTC).To achieve such temporal determinism, research is necessary in many areas in order to create so-called Time-Aware Applications, Computers and Communication Systems (TAACCS).TAACCS is a recently established interest group that have identified areas requiring research as follows: clock and oscillator designs, time and frequency transfer methods, the use of timing in networking and communications systems, hardware and software architecture, design environments, and the design of applications.Current systems use timing mostly as performance metric.In order to provide higher QoS in the area of telecommunications, better time and timing awareness between hardware, software and the network is thus necessary.Further details on the TAACCS project can be found at [9].
WebRTC
WebRTC is an innovative approach to real-time communications.Its target is to implement real-time communication functionalities into all browsers making them accessible to developers through HTML and JavaScript.Some of the major international standardization communities are currently working on standards and guidelines for implementation of WebRTC into browsers [10].Those standards are already being implemented by some of the browser vendors.WebRTC introduces a concept of peer-to-peer streaming into web browsers.In this new model two browsers are able to communicate directly between each other once the Session Description Protocol (SDP) offer has been negotiated.The Signaling Server is used to provide a signaling channel between the ends of the peer-to-peer connection.Communication is done using on-the-wire standard protocols, which normally use User Datagram Protocol (UDP) for transport.Real-Time Transport Protocol (RTP) and Real-Time Transport Control Protocol (RTCP) are used to provide more reliability when transporting time sensitive data over UDP [11].
packet size from 20 ms to 40 ms results in average M2E delay increase of 24 ms.This seems acceptable as an additional 20 ms of delay are added by increasing length of voice sample from 20 to 40 ms, the remaining 4 ms are possibly added due to longer processing requirements caused by encoding and transmitting of bigger packet.However -a further increase of packet size to 60 ms results in an additional 53 ms of delay -from 184 to 237 ms.Interestingly, reducing packet size from 20 ms to 10 ms did not result in M2E delay decrease.This could be caused by protocol headers overhead and the way in which WebRTC library handles 10 ms packet sizes for Opus codec.As evident from Fig. 2, a greater variation between single M2E delays for 10ms packets has been observed in comparison to that achieved for 20 ms packets.
Again, in the context of G.114, it is interesting to note the non-linear impact and the extent to which application settings can greatly impact on M2E delay.In the absence of forensic under-thehood analysis of the WebRTC codebase, the precise reasons for this non-linearity are not known.In section E below, we identify the jitter buffer behaviour as being the main contributor to this increase but also rule out sender side jitter in sending interval as a primary cause.Whilst such analysis is ongoing, we can speculate that issues such as drivers can be a contributory factor.Previous research by one of the authors has shown very unusual M2E behaviour casued by a mismatch between driver and application settings [5].As outlined above, this reinforces and highlights the more general concerns raised by the TAACCS interest group and the consequent need for better Time Awareness.
C. Sample Rate
Besides deploying different packet sizes, Opus can also operate with a wide range of sampling rates.The sampling rate can be defined as a number of samples taken for each second of the signal.It is important to remember that Opus is a variable bit-rate codec.Bit-rate is a number of bits sent over the network in each second of the connection.Generally as sampling rate increases, greater bitrate is required to transmit data.That means that depending on network conditions and current performance, Opus can adjust encoding rate in order to maximize a quality of call [12].The encoding rate will never exceed the maximum value as passed in SDP offer, however it may be set up to any value below it during the duration of call.Tests were performed to measure M2E delays under five encoding rates: 8000 Hz, 12000 Hz, 16000 Hz, 24000 Hz and 48000 Hz.The ping values again indicated negligible network delay with values between 0 and 3 ms.
As shown in Fig. 3, encoding rates did not have a noticeable impact on M2E delays.As expected, the bitrates present during the calls increase slightly as encoding rates rise.Opus supports bitrates of up to 510 kbps, however encoding bitrates of up to 48
A. Baseline
Firstly, M2E baseline delays for connections involving the default Opus codec were captured.It should be mentioned that ping results for this configuration were negligible with less than 2 ms round trip delays.Table 1 summarizes delays, outlining average and standard deviation.The experiment was performed using Voice only and then with Voice/Video call configuration.The results illustrate that in absence of significant network delays, the WebRTC default application produced M2E delays that approach the G.114 limit levels.
B. Packetization
The influence of packetization size on M2E delay for Opus codec was then examined.Figure 2 shows average delays for 10, 20, 40 and 60 ms packets respectively, as well as the full range for each packet size.For each call, ten delay samples were taken.The Ping facility, as above, was used to measure network related delays.For all of the packet size tests, ping was negligible with values never exceeding 1ms.
Fig. 2 Packet size delays
Interestingly, a non-linear increase in M2E delays was observed as packet sizes increase, as shown in Fig. 2. Increasing values of M2E delay are quite similar for both Opus and iSAC, the delay range for iSAC is also greater.Opus codec achieved slightly higher average values to those reported for G. 711, however minimum and maximum values of delays for both codecs suggest that delays imposed by both of them are of the same range.Moreover it is important to note that Opus as a wideband codec is much superior to G.711 (a narrowband codec) in terms of quality of experience.Moreover G.711 performs almost identical for both versions, namely A-law and μ-law.
E. Jitter Buffer
The non-linear increase in M2E delay for different packet sizes outlined above encouraged the authors to investigate the jitter buffer behaviour during those calls.It is worth noting here that the jitter buffer values can be observed using chrome:// webrtc-internals/.Our observations showed that when packet size is changed from the default value of 20 ms, the jitter buffer starts introducing additional delays, see Figs. 5 -8 for more detail.Both Opus and G.711 are mandatory to implement for any WebRTC solution.This is to make sure that all the endpoints of the connection support at least one common codec.The browser generates an SDP offer contains a list of all the codecs supported on a given machine.In this way, the end points negotiate at the beginning of the communication, by comparing their SDP offers, and agreeing on a codec for voice communication.Tests were performed between two endpoints to measure M2E delays for various codecs supported by WebRTC library.Ten calls were performed for each codec and ten samples were taken during each call.Before each sample a Ping signal was sent between two endpoints to measure network delay.As before, Ping values for those tests never exceed 1 ms.In this test, the following codecs were tested: iSAC operating at 16 kHz, iSAC operating at 32 kHz, Opus operating at 48 kHz, G. 711 A-law and G.711 μ-law operating at 8 kHz.The packet size for each codec is following: 30 ms (iSAC (16000)), 30 ms (iSAC (32000)), 20 ms (Opus), 20 ms (G.711 (PCMU)) and 20 ms (G.711 (PCMA)) respectively.
As shown above in Fig. 4, the delays imposed by various codecs are relatively very similar, with iSAC codec shows slightly higher delays than Opus and G.711 for both sampling rates.That could be explained by the fact that iSAC operates on speech frames involving 30 ms of speech samples as oppose to speech frames of length of 20 ms used for Opus and G.711.Although the average
Conclusions and further work
Overall, some very interesting results in terms of M2E delays of WebRTC have been presented in this paper.The results firstly showed that baseline delays for WebRTC in presence of minimal network delay and jitter (treated as negligible) resulted in delays very close to or greater than those defined in ITU-T Rec.G.114.This is of significant concern for QoS/QoE as network delay can easily introduce 10-100 ms baseline increase of delay with jitter adding further due to jitter buffer behaviour.Regarding other tests, the sampling rate and in turn bitrate for Opus codec has no significant impact on M2E delays that occur during WebRTC calls.However, the experiments with packet sizes showed its significant impact on M2E delay.The packet size of 20 ms has been identified as the preferable choice, as it resulted in the lowest M2E delays.For a packet size increase to 60 ms, the corresponding M2E delay increase is of particular note.Moreover the operation of jitter buffer has been identified as a source of additional delay.While the jitter buffer for 20 ms packets tends to stay within 30-50 ms range, once the packet size increases to 60 ms jitter buffer values increase to between 60 and 130 ms.The source of such behaviour was not evident from an analysis of sender side packets, and further research is needed.
The study also showed that codec choice had little impact on M2E delay.Delays introduced by iSAC codec for both sampling rates were slightly higher than those achieved for Opus as well as G.711 though packetisation size is the obvious reason for this.
In conclusion, the tests emphasize the need for RTC application developers to ensure Time Awareness in the design and implementation of RTC applications.More broadly, these test results highlight the essential message of the TAACCS project whereby the full chain of hardware and software need to be better integrated and Time Aware so as to provide better guarantees and temporal determinism.We initially speculated that one possible reason for such jitter buffer behaviour could be irregular transmission of packets at the sending side caused by sender side non-determinism.The authors investigated the sender-side time intervals for each packet size using the Wireshark network analyzer.We concluded that with exception of a few out of place packets, packets are being transmitted relatively regularly from the sending side, regardless of packetization.There is some jitter in the data -however as the issue occurs for all the packet sizes, it is clear that sending time intervals are not the source, and thus the cause of jitter buffer behaviour lies elsewhere.We plan to investigate this issue as a further step in this research. | 3,862.6 | 2016-02-29T00:00:00.000 | [
"Computer Science"
] |
Comment on tc-2021-219
L.5 : the term “relatively” is quite imprecise for an abstract, I suggest to remove it. L.7 : “Permafrost and soil layers” is inappropriate, since permafrost are considered as soil as well. Maybe replace it by “active and frozen permafrost layers” ? L.8 : “shear and bulk moduli” Introduction : L16 to 19 : I would add some references about permafrost thermal definition and permafrost basics. L16 : I would replace “upper” by “shallower” L17 : The expression “freeze-thawing cycles” is more common, maybe replace by it. L27 : I would add at least one reference for ice wedge definition. L28 to 37 : For these important applications that you mention, more reference and details are expected. For example, does the terms “thaw-stable” and “thaw-unstable” well documented ? L29 : I would remove “amount of” L38 to L50 : for all the geophysical methods on permafrost, maybe more references are expected. L51 to L60 : a reference for the MASW is expected. For passive methods using ambient seismic noise on permafrost sites, you can add recent references in mountain permafrost (Guillemot 2019, Lindner 2021, Albaric 2021), to develop the state of the art about these methods. L61 : In this paragraph, I would add some sentences to define shortly but precisely all the four terms that you mention in your approach : “hybrid”, “inverse”, multi-phase” and “poromechanical”. L64-L65 : I would remove these sentence about potential applications, since you already mention them above. Maybe you can even suggest these application in the discussion and/or conclusion parts. L70 : remove the article “the” in “for the assessment”
The article invites the use of this method to characterize a permafrost medium, as it appears to be more efficient and requires fewer a priori assumptions about the investigated medium.
The authors mention various applications to the detection and characterization of permafrost, ranging from civil engineering and infrastructure monitoring to the assessment of the potential vulnerability of certain areas to permafrost degradation and associated feedbacks.
The article is well structured and adequately written. A significant contribution is that authors used seismic data collected at a site in Svalbard, and applied their processing to this experiment, to show a real application of their method.
In my opinion the paper deserves publication after minor revisions. I suggest several edits.
First, the contribution of this study to the current knowledge of seismic waves propagating in permafrost is not very comprehensible to the reader. The lack of references about the poroelastic model and the lack of physical interpretation of the two Rayleigh waves should be corrected.
Also, the authors should include a fuller explanation of their field experiment in Svalbard (with a figure), to clarify what data they have collected and what their real contribution (instrumentation, data processing, ...) to this site.
More generally, there is a lack of references addressing issues which the authors mention. For an example, the applications (early warning systems and permafrost carbon feedback vulnerability) are frequently mentioned, but have to be more documented.
Finally, uncertainties of this new method must be addressed more quantitatively, in order to better assess its benefits and drawbacks over other methods.
Specific comments
Applications : early warning systems and permafrost carbon feedback vulnerability -> I suggest to add more details about what could be applied, and more referenced. Otherwise, these applications would be mention with caution only in the discussion part .
Discussion : In Figure 7c is shown the results of the inversion of shear modulus over the offset distance. The reader can observe a huge value of shear modulus in the permafrost layer located at a offset distance from 500m to 600m. Why this order of magnitude much higher than other parts of the whole profile ? To my mind, this results must be addressed in the discussion as well.
L237 : according to this sentence, the ground temperature is deduced from soil temperature among others. How did you get this soil temperature data (modeled, measured on the field ?) ?
Uncertainties : RMS values have to be systematically computed, in order to quantitatively assess the accuracy of all steps of your inversion algorithm. For example, in Figure B3 : why such a misfit between R1 experimental and numerical dispersion curves, comparative to other locations ? I suggest to add a discussion of this issue.
Technical corrections
Abstract : L.5 : the term "relatively" is quite imprecise for an abstract, I suggest to remove it. L.7 : "Permafrost and soil layers" is inappropriate, since permafrost are considered as soil as well. Maybe replace it by "active and frozen permafrost layers" ? L.8 : "shear and bulk moduli" Introduction : L16 to 19 : I would add some references about permafrost thermal definition and permafrost basics. L16 : I would replace "upper" by "shallower" L17 : The expression "freeze-thawing cycles" is more common, maybe replace by it.
L27 : I would add at least one reference for ice wedge definition.
L28 to 37 : For these important applications that you mention, more reference and details are expected. For example, does the terms "thaw-stable" and "thaw-unstable" well documented ? L29 : I would remove "amount of" L38 to L50 : for all the geophysical methods on permafrost, maybe more references are expected.
L51 to L60 : a reference for the MASW is expected. For passive methods using ambient seismic noise on permafrost sites, you can add recent references in mountain permafrost (Guillemot 2019, Lindner 2021, Albaric 2021), to develop the state of the art about these methods.
L61 : In this paragraph, I would add some sentences to define shortly but precisely all the four terms that you mention in your approach : "hybrid", "inverse", multi-phase" and "poromechanical".
L64-L65 : I would remove these sentence about potential applications, since you already mention them above. Maybe you can even suggest these application in the discussion and/or conclusion parts. L70 : remove the article "the" in "for the assessment" Methods: L74 : change "the overview" to "an overview" L75 : "surface wave measurements" Maybe you must develop the technique used in details, or precise if these seismic tests are active or passive. L100-102 : Are this statement and this equation for extracting Rayleigh wave dispersion relation ? If yes, please precise explicitly. L103 : I would replace "a constant frequency" by "one given frequency" L199 : I would add "respectively" in this sentence L122 : please, precise what are the two tuning parameters. Are they chosen among the optimization variables mentioned above ? L138 : the term "Here" is not clear, you must precise if you mind "in our model" or more focused on one layer of your model. L147 to L159 : This paragraph would be improved by adding some references or figures that illustrates your statements. Actually, it is not very clear for the readers whether the elements are your contribution, or from the current state of the art. For example : the existence of two Rayleigh waves, the respective dependency of R1 and R2 waves to parameters (mechanical and physical). If references exist about these questions, you must add them here. Overall, some physical interpretations will be appreciated : for example, is the higher R1 velocity than R2 velocity easily interpretable in a physical point of view ? Is the difference of sensitivity to physical and mechanical properties between R1 and R2 surprising or expectable ? Why ? L174 : if you can, precise the type of geophones (type, natural frequency) L181 : why "almost completely frozen" ? Precise why you choose the value 85% for the degree of saturation of unfrozen water. L195 : I would add a reference for illustrating this statement L210 : the term "sufficiently close" must be completed by a quantitative assessment (RMS ?).
L227 : "We also predicted" L239 : I suggest to replace "is highly related" by "could highly related" Discussion and conclusions: L249 : "makes the analysis more efficient" : you must tell more about this statement : What do you compare this method to ? And, have you done a quantitative assessment to support this discussion? L255 to 257 : this sentence must be documented by at least one reference.
L276 : for the case of a potential early warning system, how do you plan to deal with the seasonal variations (ex: freeze-thawing cycles of the active layer) that you would measure over one year? Do you have any idea how to model and remove such environmental influences that are not related to damage? And how to fix critical values ? If you have any ideas on this issues, you would be welcome to mention them, in order to strengthen your discussion on this potential application.
Figures: Figure 1: I would precise in the legend that variable (n, Sr, H, K, G) are defined for each layer (layer 1, layer 2, layer 3). (d), scaling has to be modified. And also for the sake of simplicity, the predicted average soil temperature distribution may be removed from this figure, since this variable do not seem to be useful for the study.
Appendices:
Appendices A, C and D: there is a lack of references in these parts. I suggest to add at least Carcione & Seriani (2001) and Leclaire (1994).
Appendix B : in all figures you show both saturation degree of unfrozen water and saturation degree of ice, but only one seems to be useful, since the two variables are directly linked together. Furthermore, what about the results of the layer thickness from this surface wave inversion ? It could be appropriate to show them as well. Again, for R1 and R2 experimental and numerical dispersion curves, it should be good to precise misfits through RMS values.
L297 : I suggest to add "respectively" L382 : "Convention" instead of "convection" L396 : "The values of each component" instead of "The value of each components" Appendix D : L432 : I suggest to remove "the matrix formed by" for consistency Powered by TCPDF (www.tcpdf.org) | 2,333.8 | 2021-09-06T00:00:00.000 | [
"Geology"
] |
Population genomics of an exceptional hybridogenetic system of Pelophylax water frogs
Background Hybridogenesis can represent the first stage towards hybrid speciation where the hybrid taxon eventually weans off its parental species. In hybridogenetic water frogs, the hybrid Pelophylax kl. esculentus (genomes RL) usually eliminates one genome from its germline and relies on its parental species P. lessonae (genomes LL) or P. ridibundus (genomes RR) to perpetuate in so-called L-E and R-E systems. But not exclusively: some all-hybrid populations (E-E system) bypass the need for their parental species and fulfill their sexual cycle via triploid hybrid frogs. Genetic surveys are essential to understand the great diversity of these hybridogenetic dynamics and their evolution. Here we conducted such study using RAD-sequencing on Pelophylax from southern Switzerland (Ticino), a geographically-isolated region featuring different assemblages of parental P. lessonae and hybrid P. kl. esculentus. Results We found two types of hybridogenetic systems in Ticino: an L-E system in northern populations and a presumably all-hybrid E-E system in the closely-related southern populations, where P. lessonae was not detected. In the latter, we did not find evidence for triploid individuals from the population genomic data, but identified a few P. ridibundus (RR) as offspring from interhybrid crosses (LR × LR). Conclusions Assuming P. lessonae is truly absent from southern Ticino, the putative maintenance of all-hybrid populations without triploid individuals would require an unusual lability of genome elimination, namely that P. kl. esculentus from both sexes are capable of producing gametes with either L or R genomes. This could be achieved by the co-existence of L- and R- eliminating lineages or by “hybrid amphigamy”, i. e. males and females producing sperm and eggs among which both genomes are represented. These hypotheses imply that polyploidy is not the exclusive evolutionary pathway for hybrids to become reproductively independent, and challenge the classical view that hybridogenetic taxa are necessarily sexual parasites. Electronic supplementary material The online version of this article (10.1186/s12862-019-1482-4) contains supplementary material, which is available to authorized users.
Background
Hybridization can promote adaptive divergence and speciation [1][2][3][4], but interspecific hybrids must first overcome the meiotic disorders associated with gametogenesis of diverged, non-coadapted genomes. The best-studied evolutionary strategies to bypass this barrier include clonal reproduction by parthenogenesis, gynogenesis/hybridogenesis [5] or allopolyploidy, by the production of unreduced diploid gametes [6,7]. If they are accompanied by reproductive isolation with the parental taxa, these mechanisms can represent the first stages towards hybrid speciation [8,9]. Nevertheless, hybrid taxa may still arise without clonality and polyploidization [1], and their contribution to biodiversity is presumed to be marginal [10]. Characterizing the processes responsible for the maintenance of hybrid taxa is thus a fundamental step towards understanding how they can lead to speciation.
Hybridogenesis and polyploidization are well-known attributes of Pelophylax water frogs [11]. The edible frog P. kl. esculentus is the hybrid between the pool frog P. lessonae (genomes LL) and the marsh frog P. ridibundus (genomes RR). It is often found in diploid form (LR) coexisting with one or both of the other parental species. In the lessonae-esculentus system (L-E), common in Western Europe, LR hybrids exclude their L genome and produce clonal R gametes. Inter-hybrid crosses yield unviable RR offspring because the R hemiclone irreversibly accumulated deleterious mutations through Müller's ratchet [12][13][14]. Pelophylax. kl. esculentus must then backcross with LL P. lessonae to perpetuate, and is therefore considered a sexual parasite. In Eastern Europe, the system is essentially the reverse (ridibundus-esculentus, R-E): LR P. kl. esculentus hybrids predominantly produce L gametes and rely on RR P. ridibundus to reproduce.
Interestingly, P. kl. esculentus can also be widely found by itself in so-called all-hybrid populations (E-E system), where the life cycle is fulfilled by the production of polyploids. Female hybrids can produce bivalent LR eggs that develop into triploid LLR and RRL frogs upon fertilization by haploid L or R hemiclonal sperm, respectively [15][16][17]. These triploids may also produce unreduced gametes (e. g. LL sperm from LLR males, [17]). As the diploid genomes (LL or RR) are able to recombine in triploids, the all-hybrid population as a whole becomes a sexually functional unit [18]. Yet it is worth noting the strong mutation load of such system, which induces RR and LL zygotes that do not reach sexual maturity [15]. Hence the respective proportions of each gamete (LR, LL, R, L) produced by each sex of each hybrid type (LR, LLR, LRR) is key to the persistence of all-hybrid P. kl. esculentus populations at an evolutionary stable but sensitive equilibrium [19].
Comparative population genetics of P. kl. esculentus can shed light on the origin, composition and evolutionary dynamics of all-hybrid populations [20,21]. Because of the high diversity of breeding systems, clonal genomes, sex-determination and genetic variation in this frog complex [21,22], comparative analyses of closely-related groups of populations are of prime interest, since their biogeographic history should not be a confounding factor. In addition, many European populations have been largely compromised by multiple invasions of Pelophylax alien species, resulting in genetic pollution and/or disruption of their hybridogenetic complexes [23][24][25][26].
The present study focusses on Pelophylax populations from southern Switzerland, namely the canton of Ticino. This area is mostly inhabited by P. lessonae and P. kl. esculentus in its northern parts (L-E system) but P. lessonae frogs are missing from most of the southern parts, which may consist only of E-E systems [12,27]. This set of populations could therefore provide a standardized framework in which to examine the composition and dynamics of L-E derived all-hybrid Pelophylax populations -P. ridibundus is naturally absent from the Apennine Peninsula, [28]. The Ticino area also has the advantage of being free of alien Pelophylax taxa [24].
South-alpine populations are also attractive from a phylogenetic perspective. Our recent phylogeography of water frogs from the Apennine Peninsula revealed a cryptic nuclear lineage basal to the two known pool frog taxa (P. lessonae and P. bergeri), and restricted to the Alpine catchments valleys of the Po plain (named "Pelophylax. n. t. 2", [29]). Mostly based on the intronic sequence marker Serum Albumin intron 1 (SAI-1), but lacking mitochondrial divergence, the origin of this lineage is pending additional analyses.
In order to characterize the genetic nature and hybridogenetic mechanisms of the subalpine Pelophylax populations, we conducted a population genomic and morphometric survey of nine sites inhabited by P. lessonae and P. kl. esculentus in southern Switzerland. The objectives were (1) to assess the genetic composition of putative L-E and E-E populations, (2) to understand whether E-E populations are maintained through triploid individuals or other mechanisms, and (3) to infer the nature and origin of the P. n. t. 2 lineage previously proposed [29].
Population genomics of Pelophylax in Ticino
In northern Ticino, we sampled both P. lessonae and P. kl. esculentus, occurring in syntopy at most sites ( Fig. 1, Table 1). In southern Ticino, we only found P. kl. esculentus from the three extant water frog populations known, expect a few hundred meters from site STA, where we captured 5 females of the P. ridibundus morphotype ( Fig. 1, Additional file 2: Table S1).
Genetic analyses corroborated our field observations. Based on 2521 SNPs, Bayesian clustering with STRUCT URE (k = 2) recovered the two main gene pools corresponding to P. lessonae (northern Ticino, Joux Valley, northern Italy) and P. ridibundus (STA). Hybrids P. kl. esculentus were accordingly assigned with half probabilities (Fig. 1). Analyses with increasing k separated P. lessonae from the north-alpine Joux Valley (k = 3) and distinguished the L genomes of southern Ticino hybrids (k = 4). The most likely k was k = 2 according to the Δk statistic (Δk = 3689.9) and k = 3 according to the L(k) statistic (L(k) = − 108,408.6). In south Ticino, one P. kl. esculentus (AGR08) featured the northern Ticino L genome, while all others had the southern Ticino L, or a mix of both (Fig. 1). The first axis of the PCA depicted a similar signal, highlighting the genetic structure within P. lessonae and between P. kl. esculentus from north and south Ticino (Fig. 2). Genetic diversity was accordingly higher for the hybrid P. kl. esculentus (H o = 0.18-0.21) compared to the parental P. lessonae (H o = 0.06-0.07) ( Table 1, Additional file 1: Figure S1). The five P. ridibundus specimens featured very low heterozygosity (H o = 0.02) ( Table 1, Additional file 1: Figure S1).
All frogs from Ticino possessed P. lessonae mtDNA (Fig. 3). A single cyt-b haplotype (LES25s) was sequenced in all populations but a few additional ones were private from northern (LES22s, LES24s and LES28s) and southern Ticino (LES16s, LES30s). The P. ridibundus females from STA featured two different P. lessonae haplotypes (LES25s and LES30s).
Phylogenomics of Pelophylax in Ticino
Phylogenetic reconstruction of RAD sequences (13.1 kb) recovered the six species included in the analysis (Fig. 4). Parental Pelophylax frogs from northern Ticino all belong to a fully supported monophyletic P. lessonae clade, sister of P. bergeri, with little intraspecific structure. The five frogs from southern Ticino identified as P. ridibundus were accordingly grouped with our P. ridibundus reference samples.
Identification of triploids
We could not find evidence of triploid hybrid frogs in Ticino. No tri-allelic genotypes were found at the three diagnostic L/R microsatellite loci: Res16, Rica1b5 and Rica2a34, see Methods). For locus Res16, allele 127 was fixed in the R genome, while five different alleles segregate on the L genome, including allele 127 and a null allele (Additional file 2: Table S1). For locus Rica2a34, allele 106 and a null allele could be isolated from the R genome, while 16 other variants segregate on the L genome (Additional file 2: Table S1). Because of this strong variation, both in terms of polymorphism and amplification success, it was not appropriate to quantify the peak height ratio (PHR) between the two alleles of hybrid frogs for Res16 and Rica2a34. For locus Rica1b5, allele 137 was R-specific in all populations, while alleles 122, 123, 127 and 145 were L-specific (Additional file 2: Table S1). Fortunately, the majority of P. kl. Fig. 5), which fall within the range obtained by Christiansen [30] for diploid LR frogs of an analogous genotype at this marker (120/136; PHR = − 0.23 -0.00). In contrast, Christiansen [30] reported PHR averaging − 0.39 (− 0.54 -− 0.29) for LLR triploids and 0.17 (0.09-0.25) for LRR triploids at this genotype (illustrated in Fig. 5 for comparison).
RAD markers also supported diploidy for all hybrid frogs. Among the 2521 SNP genotypes, we identified 376 RAD tags with fixed differences between P. lessonae and P. ridibundus. For these L/R diagnostic loci, relative allele coverage (highest allele coverage/lowest allele coverage) tended towards 1 for all loci (mean = 1.1×, range: 1.01× − 1.3×). This average value (1.1×) was also obtained separately for northern (putative L-E system) and southern Ticino (putative E-E system) (Fig. 5). No frogs showed relative coverage ratios anywhere near 2×, as expected for LLR or LRR triploids.
Morphometric analyses
A MANOVA analysis on Ticino frogs combining five morphometric variables (see Methods) suggested a significant effect of taxa (F = 54.1, P < 0.001) and population (F = 2.6, P < 0.001), but not of sex (F = 1.9, P = 0.10). PCAs on different sets of individuals (both sexes pooled) clearly differentiated the five P. ridibundus of site STA (Fig. 6, top), as well as between P. lessonae and P. kl. esculentus (Fig. 6, middle). The difference between northern and southern P. kl. esculentus was not obvious (Fig. 6, bottom), but remained significant even when including sex and population in the MANOVA (F = 4.7, P = 0.002).
Two contrasting hybridogenetic systems in Ticino
Several hybridogenetic systems are known from the Pelophylax model, where the hybrid P. kl. esculentus coexists with either P. lessonae (L-E system), P. ridibundus (R-E system), sometimes with both (L-R-E system) and sometimes with neither (E-E system), in which case the sexual cycle relies on triploid individuals (see Background). In Ticino, we characterized two putatively different hybridogenetic systems in otherwise closelyrelated populations.
In northern Ticino, the hybrid P. kl. esculentus was found together with P. lessonae at sites BIA, CAM, GUD, PIA and LOS. At GOL (a river bank) only hybrids were captured but these frogs most likely breed elsewhere, perhaps at the nearby site LOS, where P. lessonae occurs in large numbers. Hence, all these populations fit the expectations of an L-E system, where P. kl. esculentus exclusively produces R gametes and backcrosses P. lessonae to perpetuate (Fig. 7a).
In southern Ticino however, we did not find P. lessonae at any site, despite equivalent search efforts. All populations were exclusively composed of P. kl. esculentus of both sexes, except for five females at site STA that we unexpectedly identified as P. ridibundus (Table 1). Several clues indicate that these P. ridibundus were not parental frogs, but rather the offspring of P. kl. esculentus × P. kl. Haplotype numbers correspond to the~900 bp sequences published by Dufresnes et al. [24] but are labelled "s" for "short" since here only~500 bp was sequenced esculentus hybrid crosses. First, P. ridibundus is not naturally present in Ticino and northern Italy; the closest naturally-connected populations are in Croatia [28]. Second, all five possessed local P. lessonae mitotypes, while parental P. ridibundus normally conserve their maternal lineages throughout hybridogenesis, because of mating preferences: in mixed populations, the large P. ridibundus females are preferentially chosen by the males of the other smaller species, rarely the other way around (e.g. [26]). Third, the nuclear diversity of these females was extremely low (H o = 0.02), as expected for hemiclonal RR individuals. Fourth, these frogs were much smaller (SVL = 45-65 mm) than regular P. ridibundus (up to 170 mm, [28]). While they could be subadults, all our other observations at the same time of the year in Ticino involved sexually mature frogs, and their small size could rather reflect the mutation load and low fitness expressed by diploid clonal RR genotypes. Although rarely fit, non-hybrid frogs are supposed to arise every year in all-hybrid populations [15,31]. Hence, southern Ticino could be inhabited by all-hybrid E-E-systems that sometimes produce non-hybrid individuals. Surprisingly however, we did not find evidence for mixed ploidy. First, L/R-diagnostic SNP alleles received Fig. 4 Bayesian phylogenetic reconstruction of 13.1 kb of genome-wide nuclear data (RAD tags) for non-hybrid water frogs from Ticino (in bold) and reference samples. All south-alpine frogs belong to the P. lessonae clade, expect for the five P. ridibundus females collected at STA, which branch to the corresponding P. ridibundus clade. The intron-based phylogeny of Dubey & Dufresnes [29] is provided for comparison (~1.6 kb from two markers, including 1.4 kb from SAI-1); "P. n. t. 1": hemiclone sequenced in the Italian hybridogens P. kl. hispanicus; "P. n. t. 2": putative endemic lineage sequenced in south-alpine P. lessonae and P. kl. esculentus. Our genomic data clearly rule out the existence of the latter. Country codes used for the reference samples as follows. AL: Albania; FR: France; GR: Greece; IT: Italy; PL: Poland; SRB: Serbia; TR: Turkey even sequence coverage in all hybrids, as expected for LR diploids, where the L and R alleles should be present in equal quantities. While no confirmed triploid frogs could be included here as controls, the same approach efficiently disentangles diploids from asymmetric polyploids in other hybridogenetic systems (G Lavanchy, pers. com.). Second, no individual was tri-allelic at our diagnostic microsatellites. For the latter, the strong diversity of L alleles should have allowed to detect LLR individuals if these were present. Third, the allelic profiles of our hybrid frogs were far from the range of allele quantity difference reported for LRR frogs at the microsatellite locus RICA1b5 [30]. Hence, the data at hand suggests that the putatively all-hybrid P. kl. esculentus populations of southern Ticino are maintained without triploids, unlike in other parts of Europe. This assumption necessitates confirmation by direct evidence from experimental crosses to trace allele inheritance, and from cytogenetics.
Alternatively, P. lessonae could be cryptically present in southern Ticino, in which case populations would be composed of unnoticed L-E systems. In a bioacoustic survey over 2001-2002, Mattei-Roesli & Maddalena [27] mostly reported P. kl. esculentus in this area, but suspected the presence of P. lessonae at a few sites. A few years before, Vorburger [12] identified two P. lessonae among tens of P. kl. esculentus captured nearby STA (locality Seseglio, now extinct). These observations could represent a recent shift in the composition of these populations (as seen elsewhere [21]), but they could also be the scarce LL offspring from hybrid crosses (rather than parental P. lessonae). Occasional dispersal from northern to southern Ticino is also possible, as illustrated by one "northern" frog caught at site AGR (Fig. 2). Therefore, formally rejecting the hypothetical presence of breeding P. lessonae in southern Ticino will require additional monitoring efforts throughout an entire breeding season.
On the causes and consequences of a putative diploid allhybrid system
How could a diploid all-hybrid E-E hybridogenetic system perpetuate? To our knowledge, such situation has never been reported. Importantly, because sex is supposedly determined by an XY system [32], and because primary hybridization events preferentially involve P. lessonae males (L x L y ) with P. ridibundus females (R x R x ), the L hemiclone of P. kl. esculentus hybrids can carry Ticino (low panel). All fall within the range of LR diploid frogs bearing the analogous genotype 120/136 analyzed by Christiansen [30]. Ratio obtained for triploids LLR and LRR of the same genotype are provided for comparison. Note that given the high diversity of L-specific alleles (see Results), we would have also detected LLR triploids as tri-allelic individuals. Right: average differences in coverage between the L/R diagnostic alleles of each P. kl. esculentus frog sampled in northern and southern Ticino, based on the RAD data either an X or a Y, while the R is strictly X-linked, i. e. hybrid males are L y R x and hybrids females are L x R x [33]. Therefore, both sexes must provide L and R gametes so sons and daughters can be generated (Fig. 7b). As a consequence, R x R x females (like the ones we found in STA, see also [31]) and L x L y males should also be produced. However, it is worth noting that Vorburger [12] failed to obtain any viable metamorphs from a few inter-hybrid crosses from the extinct Seseglio site. This suggests an important hybridogenetic load and if they exist, Fig. 6 PCA on morphometric data combining all three species found in Ticino (top), P. lessonae with P. kl. esculentus (middle), and P. kl. esculentus only (bottom). Color and symbols discriminate geographic origin (dark blue: northern Ticino; light blue: southern Ticino) and species (squares: P. ridibundus; circles: P. lessonae; triangles: P. kl. esculentus) Fig. 7 Outcomes of crosses for the putative hybridogenetic systems proposed to inhabit Ticino. Northern populations correspond to the classic L-E system (a). In southern Ticino, the hypothetical maintenance of LR hybrid frogs without triploids and without P. lessonae requires that frogs from both sexes alternatively eliminate the L and the R genomes (b). RR and LL genotypes are supposedly unfit (hybridogenetic load) but can arise in populations, as we found in STA that hybrid frogs producing both L and R gametes might be rather infrequent. In counterpart, the occasional LL and RR individuals would provide opportunities for recombination to purge deleterious mutations from hemiclones.
The pre-requisites of this putative diploid E-E system are difficult to reconcile given our current knowledge of Pelophylax gametogenesis. In R-E populations, P. kl. esculentus males can produce sperm of either hemiclones [34], or sometimes both simultaneously by hybrid amphispermy [35,36]. We are not aware of reciprocal dynamics in female hybridogens, which either transmit their R hemiclone (in L-E systems), or both the R and L within diploid eggs (in regular E-E-systems with mixed ploidy) [34]. Alternated L or R genome exclusion between hybrid females, or between the germ cells of the same femalewhat could be referred to as "hybrid amphigamy"is yet to be documented. Nevertheless, gametogenesis and notably oogenesis might be more labile than previously assumed in diploid P. kl. esculentus [34], including the mechanism and timing of genome exclusion [37].
How could such a system arise? Lability in genome elimination could stem from a mixed origin of these frogs, involving secondary contact between L-and Reliminating lineages. The amphispermic hybrid reported by Ragghianti et al. [36] was a cross between P. lessonae from a L-E system and P. ridibundus from a R-E system. Water frogs most likely colonized the south-alpine region from a Central European refugia, where a great diversity of hybridogenetic systems (including L-E and R-E populations) and clonal lineages co-exist [21].
The effective maintenance of a diploid P. kl. esculentus system, although pending further investigations to confirm the total absence of P. lessonae and of triploids, contributes to the ongoing debate of the evolutionary fate of hybridogenetic hybrids. First, since both genomes would be transmitted, this system challenges the usual view that hybridogenetic hybrids are sexual parasites [38]. Second, it suggests that polyploidization is not required to become reproductively independent, as a preliminary stage of hybrid speciation. Triploidy is often seen as a springboard towards tetraploidy, from which hybrid species are easier to evolve [39]. The diploid hybridogens from Ticino would emphasize an alternative pathway, although whether they are truly self-sustainable (if they reproduce by "hybrid amphigamy") or rely on interdependent L-and R-eliminating lineages, remains an open question. In a later step, these populations could eventually evolve reproductive isolation from P. lessonae and P. ridibundus by allopatric divergence, leading to homoploid hybrid speciation. Such outcome yet appears unlikely given the frequent reshuffling of amphibian distributions throughout the Quaternary. For the time being, water frogs from Ticino represent some of the last genuine Pelophylax assemblages in Western Europe and offer a promising framework to study these fascinating aspects of hybridogenesis.
No evidence for a south-alpine endemic water frog lineage
In contrast to our previous investigations based on the SAI-1 intronic marker [29], we did not recover a southalpine lineage endemic to Ticino and northern Italy using genome-wide RAD data. Instead, all pool frogs belonged to a monophyletic, well-supported P. lessonae clade (Fig. 4) and all possessed P. lessonae mitotypes (Fig. 3). SAI-1 thus features strong ancestral polymorphism and does not always seem representative of the evolutionary history of species, perhaps because it contains a retro-transposon [40]. Hence, we recommend that phylogenetic and phylogeographic inferences relying on this marker to be treated with caution (e.g. [29,41,42]).
In particular, we previously hypothesized that the hemiclone of Italian hybrid frogs P. kl. hispanicus is related to an undescribed extinct lineage of Anatolian origin, based on SAI-1 variation ("P. n. t. 1", clearly differing from the north-Italian hemiclones; [29]). Alternatively, this "lineage" could thus simply represent intraspecific SAI-1 alleles of P. ridibundus, or of a related Middle Eastern taxon. Yet, despite several molecular surveys focusing on this region [41,42], such alleles have still never been reported from extant populations. Similarly, the Cyprus endemic P. cypriensis, which was described from mtDNA and nuclear SAI-1 divergence [42], deserves a re-evaluation. Pelophylax phylogeography and systematics are in clear need for more comprehensive molecular analyses, as offered by genome-wide loci.
Conclusions
Through a comprehensive genomic survey, we rejected the hypothesis of an endemic south-alpine lineage of Pelophylax water frogs, and emphasized two types of hybridogenetic systems from southern Switzerland: a rather classic P. lessonae -P. kl. esculentus (L-E) system in northern populations and a putatively all-hybrid P. kl. esculentus (E-E) system in the south, where frogs unexpectedly showed no sign of triploidy. Nevertheless, we cannot formally exclude the cryptic presence of P. lessonae in the latter, and call for future monitoring efforts in southern Ticino and nearby Italy. If confirmed, these allhybrid diploids could persist by labile genome elimination, i. e. frogs produce eggs and sperm carrying L or R genomes indiscriminately, which would challenge the classic views that hybridogenetic hybrids are sexual parasites and that they require transient polyploid steps to reach sexual independence.
Sampling
In July 2017 and 2018, nine localities were surveyed by day and night under good meteorological conditions for Pelophylax activity (sunny days, temperatures between 20°C and 30°C), covering the entire distribution of these frogs in the canton of Ticino (southern Switzerland) ( Table 1). A total of 115 individuals were captured and identified based on the shape of their metatarsal tubercle [28]. Buccal cells were sampled using non-invasive cotton swabs and adults (n = 111) were measured for the following variables, relevant for comparative morphometry in Pelophylax: snout-vent length (SVL), tibia length (LTi), total hind leg length (LTo), length (LMT) and height (HMT) of the metatarsal tubercle. Animals were sampled and measured directly in the field, and then immediately released at their place of capture.
DNA was extracted using the Qiagen BioSprint Robotic workstation. In complement, we included 18 DNA samples collected in the study area during Spring 2014, as well as 25 samples from other Pelophylax populations/species available from our previous studies [24,29], used as references to identify the taxa inhabiting Ticino. The latter consisted of six P. lessonae from the Joux Valley (northwestern Switzerland), four P. lessonae/P. kl. esculentus from northern Italy, nine P. bergeri from Italy and Corsica, two P. ridibundus from eastern Europe, two P. kurtmuelleri from Albania, one P. cretensis from Crete, and one P. c.f. bedriagae from Turkey. Full details are provided in Additional file 2: Table S1.
RAD-sequencing
We prepared a double digest RAD (ddRAD) multiplexed library following the protocol by Brelsford et al. [43], which performs nicely for population genomics in anuran amphibians (e. g. [44]), including Pelophylax [45]. The library contained the 133 frogs from Ticino plus the 25 reference samples, and was sequenced on two Illumina lanes (single read 125). Raw sequences were qualitychecked (FastQC v0.10.1) and processed with Stacks v1.48 [46] to demultiplex, stack and catalog homologous loci in all samples using the default -m -n, and -M values. We then called SNPs to conduct population genomic analyses on Ticino and the closely-related frogs from Joux and northern Italy. We flagged 17 Ticino samples featuring high rates of missing data with a custom python script (available at: https://github.com/DanJeffries/RADweek/ blob/master/code/Summary_plotter.py), and subsequently outputted a genotype matrix for 126 samples, considering SNPs present in 80% of individuals of each population (2521 SNPs). We also produced a sequence alignment (13,098 bp from 111 RAD tags) for 45 non-hybrid individuals from different Pelophylax taxa, to be used in the phylogeny (list in Additional file 2: Table S1).
Genetic detection of triploid hybrids
We followed two separate approaches to identify the ploidy of hybrids P. kl. esculentus. First, we genotyped microsatellite loci known to co-amplify and have specific L and R alleles: RICA1b5, Res16 and RICA2a34 (reviewed in [47]). Triploids may be tri-allelic, or, in the absence of polymorphism on the duplicated genome, one allele should be amplified in double quantity compared to the other. Because the amplification performance of microsatellites is also affected by allele size and potential nucleotide mismatch on the priming sequence, the height of the absorbance peaks must be interpreted with caution and independently for different genotypes. Christiansen [30] calibrated such approach for several northeastern European genotypes and showed that their peak height ratio (PHR) were not overlapping between LLR, LR and LRR hybrids, and could thus be used as an identification tool.
We amplified the three loci in 113 water frogs from Ticino by 10 μL multiplexed PCRs containing 3 μL of MPMM, 2.2 μL of milli-Q water, 3 μL of template DNA, as well as primers (10 μM) for RICA1b5 (0.1 μL each), Res16 (0.3 μL each) and RICA2a34 (0.5 μL) each. PCRs were conducted as follow: 95°C for 15′, 35 cycles of 94°C for 30″, 53°C for 45″ and 72°C for 1′, followed by 30′ at 60°C. Amplicons were diluted 4× and run on an ABI Prism 3100 genetic analyzer. Importantly, PCR conditions and dilution were optimized to ensure the readability of absorbance peaks. Peaks were scored and their height measured with GeneMapper 4.0 (Applied Biosystems). When comparable (see Results), we calculated the PHR as log(H R /H L ), where H R is the height of the R-specific allele, and H L is the height of the L-specific allele, following Christiansen [30].
Our second approach aimed at comparing the coverage (sequence depth) between RAD tags with fixed L-R differences. In LR diploids, the two alleles should have approximately been sequenced at the same depth, and so their relative coverage should on average tend towards one. In LLR and LRR triploids however, one allele should have been sequenced twice compared to the other and so their relative coverage should on average tend towards two. To compute L-R coverage differences, we first flagged SNPs that were fixed between the R and the L genome, i. e. with allele frequency differences of 1.0 between the frogs identified as P. lessonae (LL) and P. ridibundus (RR) in our study area (see Results). For each of these SNPs, we then calculated the ratio of the highest allele coverage by the lowest, for every P. kl. esculentus hybrid frog identified in the study area.
Population genetic analyses
We explored the genetic structure of the water frog populations from Ticino, Joux and northern Italy based on our matrix of 2521 SNPs (n = 126 individuals). First, we used the Bayesian clustering algorithm of STRUCTURE [48] with the admixture model and performed three replicate runs from k = 1 to 6, each with 100,000 iterations after a burnin period of 10,000. Because the divergence between the L and R genomes (~16Mya, G. Mazepa unpublished data) pre-date any intraspecific differentiation, we expected two major gene pools and thus k = 2 as the most informative solution, which we verified with STRU CTURE Harvester [49]. Second, we conducted a Principal Component Analyses (PCA) on individual genotypes with the R packages adegenet and ade4. We also computed average heterozygosity for each population with n ≥ 5, separately for P. lessonae, P. kl. esculentus and P. ridibundus.
In addition, we visualize mitochondrial sequence variation of our~500 bp cyt-b fragment in Ticino by an haplotype network (TCS, [50]).
Phylogenetic analyses
We conducted a Bayesian phylogenetic reconstruction of RAD sequences of non-hybrid frogs from Ticino, complemented by reference individuals from six different species (n = 45, 13.1 kb; Additional file 2: Table S1). This analysis was performed in BEAST (BEAST 2.4.8, [51]). We used a lognormal relaxed molecular clock calibrated to the divergence of the Cretan endemic P. cretensis at the end of the Messinian Salinity Crisis at 5.33 ± 1.0 Mya [52], and the midpoint root between pool frogs (here represented by P. lessonae and P. bergeri) and marsh frogs (here represented by P. ridibundus, P. cf. bedriagae, P. cretensis and P. kurtmuelleri) at 16 ± 3.0 Mya My (G. Mazepa, unpublished data), using normally distributed priors and a birth-death tree model. We applied a GTR + G substitution model (BEAST package bModelTest; [53]) and ran the chain for 100 million iterations, sampling one tree every 50,000. We verified stationarity and effective sample sizes of parameters with Tracer 1.5, and built maximum-clade credibility trees with the BEAST module TreeAnnotator, discarding the first 20% of sampled trees as burnin.
Morphometric analyses
The morphology of Ticino frogs was assessed using the field-measured variables LTi, LTo, LMT and HMT corrected by individual size (SVL), as well as the ratio LMT/HMT, which reflects the shape of the metatarsal tubercle (an important anatomical feature to compare Pelophylax taxa and their hybrids, [28]). Combining these five variables, the general morphology was first compared between species, sex and population by a MANOVA. Because there were no significant differences between sexes (see Results), we then conducted several PCA analyses (ade4 package in R) with males and females pooled together. | 7,404.6 | 2019-08-05T00:00:00.000 | [
"Biology"
] |
Addendum: Observation-based solar and wind power capacity factors and power densities (2018 Environ. Res. Lett. 13 104008)
‘Observation-based solar and wind power capacity factors and power densities’ (Miller and Keith 2018 Environ. Res. Lett. 13 104008) contained a methodological error in how we estimated wind plant area, leading to an underestimate of wind power densities. The method and revised results were published as a Corrigendum (Miller and Keith 2019 Environ. Res. Lett. 14 079501). Given the importance of these estimates to energy policy, here in this Addendum, we expand on these corrected results, while also describing the public release of data to allow verification by third-parties. Specifically, here we: (1) illustrate our method by showing in greater detail how it works for the 2 wind power plants from figure 1 of the original study, (2) identify potential selection biases in the sampling of wind power plants used in our study, (3) provide a comparative overview of the various prior published estimates in graphical form, and, (4) conclude with a description of the data we are releasing publicly.
shows the Bull Creek and Fenton wind power plants. Bull Creek (Plant_Code=56956) is comprised of 180 Mitsubishi MWT62/1.0 turbines, with a per-turbine rated capacity of 1.0 MW i and 69 m hubheight. Based on our Methods (main text and supplemental information of Miller andKeith 2018a, 2019), we calculated a Voronoi polygon around each wind turbine location. Each Voronoi polyon delineates the area closer to an individual turbine than to any other wind turbine in the entire USGS data set (Hoen et al 2018). As shown in figure 1(C), some Voronoi polyons are unrealistically large if considered as estimates of their wind turbine footprint. Indeed, a wind plant consisting of a single isolated turbine would have a huge Voronoi polyon area. This is the reason our method uses the area of the median Voronoi polygon as the basis for estimating the area of the wind plant which is computed by multiplying the median area (in this case 0.22 km 2 ) by the number of turbines (180) to yield an estimate for the total area (39.2 km 2 ).
Our method provides an estimate of the wind plant area, but it does not provide a unique outline of that area.
As an aid to visualizing and understanding our results, we can compute an outline that contains the same total area as our estimate while having a constant minimum offset between each turbine and the perimeter. This is computed by constructing disks of equal radii around each turbine, dissolving their overlapping area to prevent double counting, and adjusting the radii until the total area inside one or more disks is equal to the area we computed. The resulting perimeter for Bull Creek is shown in figure 1(A). This is a useful illustration of a wind-plant perimeter, particularly as compared to methods that use a fixed radius from each turbine.
The same approach was used for Fenton (Plant_Code=56617). Fenton is comprised of 137 GE1.5-77 turbines, with a per-turbine rated capacity of 1.5 MWi and an 80 m hub-height. Just like the method for Bull Creek, Fenton's median Voronoi polygon area of 0.53 km 2 was multiplied by the turbine count, yielding a total area of 73.2 km 2 ( figure 1(D)). This equivalent area is shown around the Fenton wind turbines in figure 1(B).
We investigated the impact of selection bias in our sample of wind plants. In the 2016 data, for example, Original content from this work may be used under the terms of the Creative Commons Attribution 3.0 licence.
Any further distribution of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI.
total capacity of wind plants that passed our qualityassurance filters was only 51% of the total capacity reported by EIA at the end of 2016. How representative is our sub-sample of the full data set? We first examine spatial sampling. Figure 2 shows the map of the all wind power plants from the EIA Power Plants (US Energy Information Administration EIA 2018a) and our subset (see public data release below: MillerKeith2018Data1.csv). Some locations, like Massachusetts appear to be missed. Upon further analysis, many of these 'plants' are actually 1 or 2 wind turbines, which because of their rated capacity >1 MW are included in the Power Plants dataset, but were excluded from our analysis because the Voronoi polygons containing these wind power plants were large, so the filter for capacity densities 0.1 MW i km −2 excluded these wind plants. Figure 3 compares the capacity cumulative distribution of wind plant size in the full data and our subset. Our sample is biased towards larger wind power plants. This highlights the fact that our method is not applicable to very small wind plants.
Assuming, as seems likely, that continued expansion of wind power will be associated with larger wind plants, then our under-sampling of small wind plants does not impact the applicability of our estimates. Moreover, the uncertainty or arbitrariness of the median-Voronoi method decrease with increasing scale. For a single turbine, we know of no unambiguous way to define the area over which it is extracting power. Calculating power density for small wind plants is inherently arbitrary, as the benefits of being isolated on a ridgeline or in a coastal region are clear, but the ability to later deploy additional wind turbines downwind is often limited. But the methodological errors in estimating the power density of medium sized wind power plants (30-200 km 2 ) using the median-Voronoi method are small as these plants typically contain 10-100 s of wind turbines all co-located into one region, often in rows (e.g. figure 1).
Large wind power plants (>300 km 2 ) should also be easily defined in the future, but as today's wind plants are just now growing to these dimensions via adjacent deployment in windy locations, we suspect that our results for these large wind power plants are also not perfect. In our analysis, the power densities associated with these large wind power plants are quite low (figure 7(B) of Corrigendum), leading us to believe that these are regions with very low capacity densities rather than low power densities resulting from wind plants operating upwind. As these regions are builtout with turbines, we would expect the area estimate for large wind plants to improve as well.
Finally, figure 4 compares results from this study with earlier estimates of wind's power density. The power densities from this study are consistent with physically-based models, and inconsistent with wind resource estimates that ignore interactions between wind turbines and the atmosphere.
To facilitate independent validation of these results, we have placed 2 datasets on OpenEI, a public USWTDB_ID-unique ID (text name) which also appears in MillerKeith2018_Data1.csv for matched wind power plants | 1,587.6 | 2019-07-01T00:00:00.000 | [
"Physics"
] |
Digital transformation of business processes in aic: starting conditions and priorities
. The article examines the starting conditions for digital transformation in agriculture in Russia, identifies the main stages in the history of the development of agricultural production. The author studied the regulatory documents on digital transformation, the components of the Central Information and Analytical System of State Information Support for Agriculture, analyzed the current state of the use of complex digital agricultural solutions by agricultural producers. It also assesses the achieved target i ndicators of the departmental project «Digital Agriculture», identifies trends in their change over the period. Drivers for the development of digital transformation of the agro-industrial complex in Russia have been formed and presented.
Introduction
Currently, digitalization is a priority area for the development of agriculture in Russia, since, starting in 2019, the Ministry of Agriculture of the Russian Federation has begun to implement the sectoral program «Digital Agriculture» [1].
Digital transformation is the process of integrating digital technologies into all aspects of business activities, requiring fundamental changes in technology, culture, operations and the principles of creating new products and services. Thus, digital transformation is not only investment in new technologies (artificial intelligence, blockchain, data analysis and the Internet of Things), but also a deep transformation of products and services, organization structure, development strategy, customer relations and corporate culture.
The main trajectories of digital transformation in agriculture are as follows: − at the national level: functioning of digital platforms of the Ministry of Agriculture of Russia, predictive analytics based on big data, with tools of a distributed ledger, artificial intelligence; − at the regional level: smart sectoral planning, smart contracts; − at the agribusiness level: mass implementation of complex digital agro-solutions, mass acquisition of digital competencies by specialists of agricultural enterprises.
Methods
During the study, the following regulatory documents were studied: -Decree of the President of the Russian Federation No. 204 of May 7, 2018 «On national goals and strategic objectives of the development of the Russian Federation for the period up to 2024» [2]; -The program «Digital Economy of the Russian Federation» (Order of the Government of the Russian Federation dated July 28, 2017 No. 1632-R) [3].
The main conclusions of the study were based on a statistical analysis of official data from the Ministry of Agriculture of the Russian Federation, as well as data from a survey of agricultural organizations in the Saratov region on the use of existing and implementation of new complex digital agricultural solutions.
Results
The process of digitalization of agriculture at the level of collection and accumulation of information can be represented in the form of a diagram ( Figure 1). The analysis showed that the use of certain elements of digital agriculture is taking place at a rapid pace, especially in those areas where scientifically grounded systems for supporting agrobioprocesses already existed and operated ( farming systems, systems for accounting and assessing land use, systems for assessing technosphere factors, accounting and management systems. The low percentage of coverage of regional systems of the agro-industrial complex on deep processing of products (58%) and the use of programs for processing Big Data (57%) is associated with high costs, in the first case, the purchase and acquisition of equipment, in the second case, with a shortage of specialists in Big Data.
Despite such starting positions in the field of digitalization of agriculture, tasks were set to achieve the following targets within the framework of the Digital Agriculture project ( ,5) 7 (5) 10 (7) 15 (10) 25 (20) The share of specialists of agricultural enterprises who have undergone retraining and have competencies in the field of the digital economy to work with digital products and technologies,% The analysis of target indicators indicates that by 2023 all regions of Russia will introduce digital sectoral planning of agricultural production based on the digital platform «Digital Agriculture», the process of forming resources in Big Data will also be completed, while the growth of labor productivity will be 190-200%.
The issue of training and retraining specialists of agricultural enterprises in the field of the digital economy to work with digital products and technologies will remain problematic.
To assess the results of the implementation of the Digital Agriculture program, the community of leaders of the agricultural industry «Smart Farming Club», within the framework of the plenary session «Practical tools for sustainable development of digitalization in the agro-industrial complex» of the exhibition «Golden Autumn», announced the start of the project «Rating of digital maturity of the agro-industrial complex» ... The project is being implemented with the participation of an independent consultant KPMG and expert support from Sberbank of Russia.
According to the methodology, at the first stage of the rating, it is planned to fill out questionnaires by its participants on the official website of the project. Agricultural enterprises will have to provide not only information about the digital solutions they use, but also about the strategic vision for the development of digitalization in their companies, as well as the level of qualifications and knowledge of personnel in the field of high technologies and other data. The constituent entities of the Russian Federation will present a report on their use of digital technologies, plans and prospects for its further development, measures to create favorable conditions for digital transformation, as well as the competences of specialized departments in the direction of digitalization, work on training young personnel for the agro-industrial complex.
VVRD 2021
The organizers of the project expect that more than 15 thousand domestic agricultural enterprises and all constituent entities of the Russian Federation will take part in the formation of the rating [4].
As a significant result of the implementation of the «Digital Agriculture» program in 2021, one can call the commissioning of the GIS «Single Window» into industrial operation, where the verified data of the regional authorities of the agro-industrial complex is collected. Based on the data received from the information systems of the Ministry of Agriculture of Russia, other federal executive bodies and control and supervisory bodies that have integration interaction with the GIS «Single Window» and create tools and models for forecasting industry indicators. [4] At present, in monetary terms, the IT market in agriculture is estimated by experts at more than 360 billion rubles. According to forecasts of the Ministry of Agriculture, it should grow 3-5 times in the next 10-15 years. According to the Ministry of Agriculture of the Russian Federation, the comprehensive digitalization of agricultural production will allow farmers to reduce costs by 23%, reduce crop losses, which, if the means of production are ineffectively used, can be up to 40% [5].
Discussion
As a result of ongoing research in the field of digitalization in agriculture, as drivers for the development of digital transformation of the agro-industrial complex in Russia, it is already possible to name a high degree of cooperation between digital service providers for this complex and the almost absence of competition, the presence of large agricultural holdings; government measures to develop the digital economy.
We agree with the opinion of experts who believe that the obstacles to digital transformation of the agricultural sector are: lack of specialized IT specialists; insufficient financial opportunities for large-scale modernization; digital inequality between cities and villages; foreign origin of most information resources [6,7].
However, given the favorable starting conditions and the rapid pace of development of digitalization in Russia, the main "points of application" of digital technologies in the agrifood industry can be identified: increasing crop yields and animal productivity; lower production costs; increasing labor productivity; timely response to climate change; reduction of transaction costs in sales by building a transparent supply chain of products from the field to the consumer; minimization of management risks; reducing the shortage of qualified specialists; informatization of rural producers; simplification of access to borrowed funds; gaining access to digital distribution channels for agricultural producers.
Conclusion
Thus, at the time of the transformation of business processes in the agro-industrial complex, there was a technological basis of Russian origin, that is, the agricultural digitalization project practically did not depend on the foreign IT sphere. In addition, the domestic agricultural sector at the end of 2019-2021, demonstrates positive results despite the impact of sanctions and the pandemic. We believe that in this situation, the transition to smart agriculture technologies allows Russia to become one of the guarantors of global food security. | 1,946 | 2022-01-01T00:00:00.000 | [
"Agricultural and Food Sciences",
"Computer Science",
"Business"
] |
Accounting Treatment of Research and Development Expenditure: A Critical Literature Review
: Intangible investments have been found to contribute largely to value enhancement in firms and economies that spend colossus resources on them. Despite being an important component of valuation, such investments are largely ignored or given subjective treatment by the existing accounting standards and consequently, not included on firm valuation. The American standard (FASB-S2) establishes standards of financial accounting and reporting for research and development (R&D) costs. This Statement requires that R&D costs be charged to expense when incurred. It also requires a company to disclose in its financial statements the amount of R&D that it charges to expense. On the other hand, the accounting for R&D under IFRS standards requires judgment of the expectation of future economic benefit that will flow to the entity due to R&D. If it can be “ascertained”, then these costs should be treated as an asset rather than an expense since they meet the definition of an asset as prescribed by the IASB Framework for the Preparation and Presentation of Financial Statements. This paper, therefore, seeks to critically analyze the literary works of various researchers on the treatment and hence the impact of accounting for research and development expenditure on: firstly, the value relevance of financial information to investors; secondly, allocation of equity and debt resources to the firm; thirdly, growth of intangible assets; and lastly, firm value in capital markets. Previously studies conducted under here have cut across the accounting treatment of R&D expenditure, and generally, internally generated intangibles using the International Financial Reporting Standards (IFRSs) and the U.S.’ Generally Accepted Accounting Principles (GAAPs). Majority of the studies analyzed agree that sufficient disclosure of R&D investment as well as other internally generated intangibles can supplement and improve the financial information provided by the firm. This in turn will improve the outlook of the financial statements which can improve their use and reliability to investors as well as give reliable inputs to financial analysts, thus improving the applied valuation models in computing dependable valuation figures for the firm. This, by and large, should avoid the negative consequences that may result from inadequate accounting treatment of R&D expenditures.
Background
Intangible assets (capital) such as R&D have been found to play an important role in the performance of firms [45]. Firms involved in high R&D expenditures are difficult to value because their future profits depend on the achievement of yet to be implemented and uncertain experimental models. However, there is insufficient reporting of R&D expenditure resulting in inadequate information for valuation purposes [18]. Aboody & Lev argue that there is a time lag in reporting profitability of R&D and thus suggest that it should be conservatively sustained at a certain level in accordance with prudence principle of accounting [1].
Expenses incurred by companies generally have three major; the first category is the operating expenditure which deals with expenses for the firm in the current period; second category consists of capital expenditure, which deals with expenses of the firm spanning multiple periods; and the third category consists of financial expenditure which are associated with debt capital of the firm. Critically analyzing these expenses, we note that financial expenses are paid for debt capital, while capital expenses are paid for long-term assets. Therefore, these two categories of expenditures should not form part of operating expenses and hence should not be A Critical Literature Review included when computing a firm's operating income. However, capital expenditure may be depreciated or amortized over the period that the firm benefits from them [22]. Particular concern, therefore, is the inclusion of R&D as an operating expense that can distort the estimation of a firm's operating income, and by extension, the net income.
The accounting standard requires R&D expenditure to be totally expensed when incurred (GAAP), whereas, IAS 38 under the IASB framework, requires the research component expensed while the development component capitalized after satisfying the inclusion criteria [46]. This is because the accounting standards consider the ultimate R&D product to be uncertain and hence not easy to quantify. Therefore, R&D as an internally generated asset largely misses out in the financial statements because of the insufficient description of an asset, which requires the identification, measurement, and control of these internally generated assets. This, therefore, leads to understating the firm's value [22].
Hall found the market value of listed firms and R&D investments to have a positive relationship [37]. The study also found that stock prices positively follow the announcements of new R&D investments. Related studies also found that markets react positively to R&D investments announcements [18,54]. On the contrary, however, another study found that the market value of firms that heavily invest in R&D is volatile over time [20]. This variability calls for the investigation of potential factors affecting the valuation of these firms' R&D investment and consequently the criteria used to evaluate such investments in the capital markets.
According to IAS 38, research is the original planned search of knowledge done in order to acquire new scientific and technological thinking, while development is the use of research findings to plan for new or significantly improved products and services ahead of commercial implementation thereof. The System of National Accounts, 2008 (2008 SNA), defines research and development as a detailed statistical outline for macroeconomic policy development and research. The System of National Accounting (2008) recognizes R&D as investment or an asset in the economy. The term 'Research and Development' is commonly used to describe the firm's activities undertaken to make new or to improve existing products and services. Among the activities included here are researches done by universities and laboratories and also product testing and refining before commercial or internal use (International Encyclopedia of Social Sciences, second edition, 2006). The Organization for Economic Co-operation and Development came up with most agreeable definition of R&D which states that research and development (R&D) include creative and systematic work done to increase the available knowledge and to come up with new usage of the existing knowledge [58]. Hence R&D is carried out with an aim of generating new knowledge for economic benefit, solving challenges in the society or just creating knowledge for its own sake [58].
Research and development (R&D) investments are among the intangible resources that constitute an increasingly important part of performance in companies in modern economies. Most firms have realized that intangible assets win them sustained competitive advantage than the tangible assets. The development of accounting standards to measure intangible assets value has taken long and is still controversial, making the disclosure of these unseen resources difficult [6]. Although firms are aware of the fact that all intangible assets are valuable and critical because they create value and decide future growth potential, it is evident that financial statements are not good at reflecting these assets [53]. Unlike the tangible assets, measuring intangible assets is complex because they are neither bought nor sold in an open market. Traditional accounting and management systems were designed during an era when tangible assets dominated the economies. However, this has changed significantly towards intangible asset investments, but the accounting systems have largely remained the same [53].
Firms investing heavily in R&D need to recognize it as an investment an asset in the balance sheet because it will ultimately generate cash flows to the firm. This recognition is in line with asset definition under the accounting standards. Unfortunately, investment in R&D is either expensed in total or given a subjective treatment as prescribed in the U.S. GAAP and IASB's IAS 38 respectively. Consequently, different analysts come up with different values for the same firms or comparative firms depending on how these firms account for R&D expenditures or how the analysts adjust their valuation inputs and figures to suit their beliefs.
This paper aims at providing critical and analytical review of the previous studies touching on the accounting treatment on R&D investments in relation to firm valuation. It is majorly supposed to contribute to: firstly suggesting to accounting standard setters on the accounting treatment and reporting of intangible investment so as to be more responsive to firm valuation; secondly, the study findings will help firms to formulate more realistic policies with regard to reporting R&D expenditures as it directly affects firm performance and resource allocation; and thirdly, the study findings will help in theory building. The ongoing debate on this subject is aimed at developing a universally acceptable theory through empirical studies. The gaps found will lead to further studies which will eventually lead to convergence of reporting standards on R&D expenditure.
This study paper is arranged into four sections. The first section looks at the study background; the second section deals with theoretical literature review of the accounting treatment of intangible assets, and focus more closely at theories of research and development (R&D) investments; section three looks at the empirical review of intangibles and R&D investments; and finally, section four discusses the study findings and summarizes the review.
Theoretical Literature Review
The theories discussed here looks at the most related conjectures put forward by authors to explain the accounting treatment of R&D investment. Financial statements provide important information relied upon by, among others, investors to make investment decisions, and financial analysts, to provide figures to value companies. Thus, there is need for accurate preparation of financial statements, carefully taking into account both tangible and intangible assets of the company. This section, therefore, looks at some of the theoretical literature involved.
The market-based view (MBV) explains the external market factors and industry characteristics that determine the performance of a firm [30]. The research work looked at the Structure-Conduct-Performance (SCP) and focused on the five-force model. The theory analyses the unique sets of activities performed by a firm that set it apart from its rivals. The strategic advantage of a firm is determined by how it does similar activities differently from other rival firms. The performance of the firm to is determined by its structure and competitive strength in the relevant industry.
Entry barriers, distinct products, competitor volume and demand level are the factors affecting firm's behavior [2]. Other researchers who advanced the SCP framework found that organizations can have competitive advantage over others by engaging in activities that responds to the structure of their industry [14]. Firms achieve this by basing their performance against the Porter's five force model. These are; entry barriers, substitute threats, supplier bargaining power, buyer bargaining power and competitor rivalry [62]. Also according to study by Bessieux-Ollier et al., three most common basis of market power are: monopoly, entry barrier, and bargaining power [13]. Firms that have monopoly exhibit strong market position which makes them perform better [24]. High barriers to entry reduce competition leading to better performance.
However, Porter's five-force factors have limitation because it assumes both a perfect and static market structures, which are not easy to achieve. The five-force model is also complex in terms of industry interrelationships [29]. A contrary opinion argues that important factors that determine firm profitability are unique to the firm, and not the industry [28]. Competitive advantage that depends on firm capabilities and resources outweigh those based on products and market positioning [24]. They consider heterogeneous resources as main contributors to firm's competitiveness. These studies shifted the focus and suggested that from 1980s onwards, studies in strategic management would move from the structure of the industry, that is, market based structure (MBV) to the firm's internal structure involving resources and capabilities, also known as resource based structure (RBV) [24].
The Resource Based View (RBV) focuses on the internal resources of a firm as well as their strategic strength to achieve sustainable competitive advantages [35]. These will enable the firm to compete in the external business environment by offering better products and services and ensuring improved quality of the supply and value chain. According to this theory, competitive advantage will only be achieved if the firm practices resource heterogeneity and immobility.
The theory focuses on "difficult to imitate" characteristics of a firm in order to come up with high quality products and services as compared to other competitors [25]. The RBV scrutinises why firms thrive or flop in the market [24]. Resource Based Theory views an organization as comprising, human resource, physical resource and organizational resource [4]. Organizations consider resources of higher value, scarce, not easy to imitate and not easily substituted as those contributing to sustained higher performance and competitive advantage of a firm [8].
A highly performing and competitive resource fulfils the criteria of VRIN; that is, 1) Valuable (V): which means that it should provide strategic value to the firm by helping the firm to explore opportunities and reduce threats in the market place; 2) Rare (R): firm's resources should be unique and rare and difficult to find among competitors in order to be competitive; 3) Imperfect Imitability (I): it should not be easy to copy resources used by an organization to make it competitively feasible; 4) Non-Substitutability (N): this means that it must be difficult to substitute resources controlled by a firm to deny competitors the ability to achieve equally better performance.
According to Barney, a resource is valuable if it leads the firm to make comparatively higher sales, incur lower costs, and thus achieve higher margins [8]. Additionally, the author underscored the need for resources to enable a firm to come up with strategies that make it efficient and effective in its performance. Therefore, RBV gives managers of firms an understanding of the importance of competencies as the most valuable asset of a firm in achieving superior performance [38].
Despite the argument by most RBV researcher that knowledge is a generic resource of a firm, some researchers take knowledge to be a valuable resource because of its unique feature [5,65]. Others argue that, in this information age, knowledge; intellectual capital, competencies and such like attributes are important when it comes to firm's superior performance [61]. Another study pointed out that the more the knowledge asset is used by a firm, the higher the value it creates as compared to physical assets of a firm [70]. Amrit et al., further argue that technology, capital of the firm, market share and product sources are easily imitated compared to knowledge [5]. The study further suggests that knowledge theories views knowledge in five levels ordered as data, followed by information, knowledge, expertise and capabilities. On his part, Zack classifies knowledge into core, advanced, and innovative knowledge respectively [70]. According to him, core knowledge is the basic and required by firms for short-run market survival; advanced knowledge allows a firm to make peer comparison for short run competition; while innovative knowledge is required by firms for ranking against its competitors in the market. Therefore an innovative firm has the capabilities of introducing innovative products and services and thus positions itself as a market leader [70].
The basic role of financial reports is to reflect a firm's value for decision making. The following are the perspectives of value relevance according to sampled authors: 1) value relevance examines the association between accounting figures and firm's equity value [52]; 2) another study analyses value relevance in the following two ways: firstly, that stock prices are influenced by figures in the financial statements since they are used to capture a firm's intrinsic value; secondly, to be value relevant, the financial information should have the variables used as input in the valuation models or at least assists in predicting those variables [31]; 3) value relevance is the ability of financial statement information to record and report relevant information for firm value [12]. Therefore, standard setting bodies such as the IASB and the FASB should come up with high quality standards geared towards enhancing the quality of financial reports issued to stakeholders.
Accounting information is value relevant if it can predict a firm' market share price and returns [40,48,56]. Easton and Beaver did an empirical study on the association between security prices and fundamental accounting variables and found that such fundamental factors are associated with the market value of securities [11,27]. Other studies looked at the adoption of International Financial Reporting Standards (IFRS) and its effect on a firm's earnings per share (EPS) and found that earnings announcements become value relevant if they influence investors to develop an appetite for shares [16,19,66]. This culminates in enhanced share price and share return.
From the FASB framework, R&D costs are expensed when incurred, whereas its benefits are recorded later [63,71]. This accounting treatment leads to distortion of the matching principle in accounting, and thus affects financial information value, specifically earnings and cash flows [46].
The analysis of a sizeable sample of US companies to disapprove the argument that exclusion of intangible asset investments negatively affects financial information value [23]. This study was supported by another study that used a sample of US companies taken between 1952 and 1994 [31]. However, from these studies, there was mixed reaction about the value of hi-tech versus low-tech firm. A similar study focusing on the periods between 1975 and 1999 also found mixed reaction about hi-tech versus low-tech firms [21].
Accounting Standards, IASB (2004) defines an intangible asset as "an identifiable nonmonetary asset without physical substance" [41]. Financial Accounting Standards Board (FASB, 2001) also defines an intangible asset as a "noncurrent, nonfinancial claim to future benefits that lacks a physical or financial term." The theoretical framework of accounting defines an asset as "a resource controlled by an entity as a result of past events from which future economic benefits are expected to flow into the entity." The term "intangible" covers intangible assets, intangible investments, as well as intangible capital (or intellectual capital, knowledge capital or goodwill). Hence intangibles are listed into three main categories; in the first category, property rights and markets exist and include copyrights, trademarks and patents; in the second category, property rights and markets are weak and these include R&D in progress, reputation, management structure and business secrets; and finally, intangible category with neither proper legal right nor markets such as human resource, structural assets, relational asset etc. [14]. Hence the last category has more accounting issues in comparison to the second, and much more compared to the first category. However, classifications given by accounting standard setting bodies divides intangible assets into two; internally generated, which are difficult to measure and externally acquired intangibles whose markets exist and hence can be quantified in terms of price.
Intangibles investment cut across manufacturing and service firms in the U.S. and such companies spend trillion of dollars annually on intangibles [56]. Studies also show that the share price of intangibles-intensive firms moves together with large premium as compared with book value. This shows that investors value investment in intangibles [46].
A study on 'Market Valuation of Companies investing heavily in R&D,' compares those supporting with those opposing the capitalization of R&D costs [57]. Those supporting expensing of R&D expenditure argue that there is uncertainty in reliably estimating such costs [42]. They further argue that allowing managers to decide on the amount amortized could reduce the quality of reported earnings [28]. However, those supporting the idea of capitalizing R&D costs argue that there is enough evidence attached to R&D investments in term of future income [44]. Another study shows positive relationship between accounting figures and stock prices when R&D is capitalized as a balance sheet item balance sheet and amortized in future income statements [17]. Also, there is evidence of higher earnings quality in terms of price/earnings association for firms capitalizing rather than those immediately expensing R&D investment [50].
To include intangible assets in the balance sheet, it must meet the set recognition criteria in the standards. Because of restrictive conditions for inclusion set by the standards, internally generated intangibles end up left out of the financial statements [13]. The accounting framework emphasizes "control" as an imperative inclusion criterion when defining an asset. The framework states that for a firm to have control over an asset, it has to exclusively benefit from it while disallowing other firms from such benefits. However, on the flip side, it is not easy to control human resource even after benefiting from the firm in terms of training and experience [45]. A study on "partial excludability", justifies the argument that such intangible resource should not be capitalized due to the uncertainty of employee-employer contract as employees can easily switch employment from one firm to the next [45].
The accounting framework is also concerned about the reliability of measurement of the asset's value. However, because of the difficulty involved in measuring R&D investment, which is an internally generated asset, the FASB framework recommends the immediate expensing of the cost, while the IASB framework recommends expensing the research costs, while capitalizing the development costs after satisfying the set recognition criteria of technical feasibility (IAS 38/SFAS 5). However, the FASB gives an exception to computer software firms by allowing the development cost of computer software to be capitalized after satisfying the conditions set for technological feasibility (SFAS 86).
Standard setting bodies have made efforts to development of sound standards in order to improve the quality of financial reporting. Thus, various accounting bodies and institutes have made contributions to these efforts because different stakeholders rely on financial reports for decision making. These bodies have made effort to come up with guidelines and models to improve the quality of financial reporting (AICPA, 1994; CICA, 1995; ICAEW, 2000; FASB, 2001). They have formulated various disclosure requirement for disclosing nonfinancial information and other specific disclosures on intangible assets [10,58]. These efforts are discussed next.
The US, through the Institute of Certified Public Accountants (ICPA) constituted the Jenkins Committee of 1991. This committee noted that financial statement users required both historical and futuristic information on firm performance. They recommended the replacement of traditional financial reporting with one known as the Business Model which reports both financial and wealthcreating information of an enterprise with strategic plans. Additionally, AICPA together with the FASB steering committee submitted a report in 2001 on "Improving Business Reporting: Insights into Enhancing Voluntary Disclosure." The objective was to formulate a reporting framework that encourages coherent and voluntary disclosures of key success factors of a firm.
Another body, the Canadian Institute of Chartered Accountants (CICA) initiated "Performance Measures in the New Economy," (1994). They identified nonfinancial reporting as key to strategic planning and hence shareholder value. On the other hand, The Institute of Chartered Accountants in England and Wales (ICAEW), through Charles Leadbeater, studied "New Measures for the New Economy" and recommended an incremental approach for intangibles. That is, traditional financial reporting is maintained as key requirement for corporate reporting and supplemented by nonfinancial reporting aimed at helping investors interested in intangible valuation.
Subsequently, the Securities and Exchange Commission (SEC) in the US and the Canadian Securities Administrators (CSA) put forth a requirement for publicly listed companies to issue on the face of their annual report known as Management Discussion and Analysis (MD&A). This is aimed at giving detailed narration and analysis on nonfinancial information. The European Commission also adopted these recommendations in their 'Business Review' (2005), and required directors to include such reports in order to enhance their reporting.
A reporting model known as Intangible Capital Statement was developed by the Danish Agency for Trade and Industry (DATI). This model combines narrations, graphics and figures for disclosing specific information on intangibles (DATI, 1998). The objectives of the model are two-fold; one is internal; for management decision making and the other is external; for reporting annual accounts. This model allows firms to give an account of their effort in developing knowledge resource.
Another specific model known as MERITUM project was developed by the European Union, in 2002. It set out the guidelines for management and reporting intangibles to improve intangible asset management and to encourage directors to voluntarily disclosure such assets. In Germany, the Federal Ministry of Economics and Labor developed an "Intellectual Capital Statement" reporting model in 2004. In 2005, the Japanese through Ministry of Economy, Trade and Industry (METI), developed a disclosure guideline for intangible assets for management. All these are geared towards improving the reporting of intangible investment of firms.
Review of Empirical Studies
Empirical studies highlighted here are organized as studies that have had impact on the knowledge area of accounting treatment of R&D expenditure on firm value. A majority of these studies have looked at the association between general and specific disclosure on intangibles. These studies have further paid attention on the effect of such disclosures on the following: 1) value relevance of financial information; 2) resource allocation of the firm; 3) growth of intangibles in the firm; and 4) market value of the firm. This section discusses the empirical evidence provided by different authors on the mentioned subheadings.
Voluntary disclosure serves as a way of compensating for the loss of financial statement value relevance. Managers disclose more information by way of press releases, give evidence of managers voluntarily disclosing such information through annual reports and conference calls [51]. A study of US cellular firms shows that nonfinancial information has more value than traditional financial information [3]. The findings of this study is supported by another study that used a sample of electronic companies in Taiwan to show that companies largely use nonfinancial information to supplement financial information for valuation purposes [47]. Additionally, another study supported the value relevance of financial information through their study of IT firms in Taiwan and found that combining both financial and nonfinancial information greatly adds value the explanatory power (given by R 2 ) of equity valuation model [69].
However, there is a contrary opinion that tangible and intangible assets are inseparable when it comes to measuring productivity of a firm [9]. The authors in the cited study supported this observation by looking at the macroeconomic variables of education reforms, legal reforms, networking, among others as being stimuli to encourage knowledge transformation for improved firm productivity. According to their study, it is not economical to separate intangible from tangible assets and hence managers can only disclose nonfinancial information.
A study conducted in Australian Health-Care Industry with regards to R&D's value relevance compared periods before A Critical Literature Review and after the introduction of IAS 38 and found R&D investment to be value-relevant [49,55]. The finding is attributed to capitalizing the development component of R&D expenditure as opposed to totally expensing it as before. This is in accordance to the requirement of the accounting standard IAS 38.
In summary, therefore, majority of the studies point to voluntary disclosure of R&D and other intangibles as possible solution to the value relevance inadequacy of financial information, and that incorporating such information into models of equity valuation compensates for the 'missing input' required by analysts in valuing firms.
Managers tend to voluntarily disclose more information about on intangible investments of their firms to avoid undesirable outcomes caused by nondisclosures of such information. A study on 312 firms listed in the US stock exchange found that most managers (44% as opposed to 17.4%) were in agreement that voluntary disclosure improved liquidity of their stock [34]. Another study empirically validated this finding by using AIMR rankings to conclude that disclosure improves liquidity of stock while at the same time reducing variations in analysts' forecasts [39].
On the cost of capital, Graham et al., found that 39.3% as opposed to 22% of managers contend that voluntary disclosure reduces the cost of capital [34]. This statement validates another study which used self-constructed disclosure index of US companies and found that firms that disclosed nonfinancial information in their annual reports experienced lower cost of capital [15]. A study of the EU companies' annual reports on voluntary disclosure of intangible assets found strong empirical evidence that disclosure of futuristic information is tied to lower cost of capital [43]. A related study of EU companies found that enhanced intangibles disclosure has a negative relationship with information asymmetry as well as cost of both equity and debt capital [59].
Dinh et al., posed a study question: "can capitalization of R&D improve investment efficiency?" They based their study on US GAAP by comparing investment in high-tech firms capitalizing R&D with those that don't. They found that software firms which are allowed to capitalize their R&D costs have improved resource allocation compared to those that don't capitalize [25].
Therefore, the analysis of these studies concludes that additional public disclosure on R&D as well as other intangibles reduces information asymmetry and hence the adverse selection in the capital market, resulting into greater liquidity and lower cost of capital. Overall, this may enhance market efficiency improve resource allocation efficiency in the capital market.
Empirical evidence show that enhanced disclosure of research and development investments and other intangible assets generated by firms can result in growth of intangible investments, consequently creating value to the firm [29]. A study on Canadian companies' disclosure practice and the level of growth in research and development in firms [31] found their relationship to be positive. Similarly, Gelb showed that companies would rather put more weight on voluntary disclosure than traditional accounting reporting so as to enhance R&D growth [32].
A study of the annual reports of EU companies found a positive relationship between intangible assets investment and intangibles assets disclosure [67]. Similarly, Zeghal et al., looked at a sample of Canadian companies and found a significant positive movement between R&D disclosures and R&D investment levels [72]. Additionally, other studies looked at the magnitude of intangible investments done by US companies in the high-tech sectors. They found that the higher the level of such investment, the more the frequency of disclosing such information [2,64].
From these studies, it can be concluded that management voluntary disclosure of intangible assets can be viewed as an efficient communication channel to stakeholders who require comprehensive information on R&D and other internally generated intangibles for their decision making. As observed in the analyzed studies, firms can use statutory reports, press statements, conferences, internet among other means to communicate nonfinancial information.
A number of studies also show that R&D investments and other internally generated intangibles disclosure influence the firm's market value. In another study, the average stock price in the US pharmaceutical industry rose to 1.13% up from 0.51% in the absence of such information [46]. Further, in addition to both qualitative (nonfinancial) and quantitative information, when the board announced the approval of drugs under research, the average reaction rose to 2.01%.
A survey conducted on finance professionals in Hong Kong found 88% agreeing that enhanced intangibles disclosure will positively influence a company's stock price [60]. Validating this finding is a study that looked at how disclosure of intangible assets is related to a firm's market capitalization [2]. This study found a significant positive relationship between voluntary disclosure and a firm's market capitalization. A study of internationally quoted telecommunication companies that largely use websites besides the annual reports to disclose intangibles found a significant positive movement between intangibles disclosure and market capitalization [33].
A theoretical review of mean-variance analysis on investment decisions of firms at different stages of business development that focused on Australian accounting information found a positive association between intangibles investment and return on equity [26]. This study finding validates the theory put forward by another study which suggests that for internally generated intangible to be considered an asset, its investment should be clearly irreversible [73].
Therefore, it can be concluded, from the analysis of different authors under this subheading that, when information about internally generated assets is publicly disclosed, such information can supplement the traditional financial reporting. Consequently, firms can be rewarded by the capital markets for disclosure of such nonfinancial information.
Study Gaps
The following table summarizes the gaps identified in the analyzed studies which require further researches in order to improve the theory in this study area. Firms with minimal accounting earnings are likely to disclose their pro forma earnings. Also in the absence of strategic information, firms disclose their pro forma earnings.
They could not clearly determine whether investors overreacted to pro forma earnings information at the expense of market efficiency or mispricing.
Basu & Waymire (2008)
Has the Importance of Intangibles Really Grown? And If So, Why?
They sampled international firms for their study.
They found intangibles to be synergistic, cumulative, and mostly inseparable from tangible assets and so it's wasteful to try to estimate such figures.
There is need to examine a broad set of benchmarks to ascertain whether accounting for intangibles is important. They found voluntary disclosure levels of Intellectual Capital (IC) to be low. They also found disclosure levels to be positively related to firm size. The information was largely qualitative rather than quantitative.
External validity may be questionable as a result of the relatively small sample size. Also managers were not observed in the survey, so management intent is simply inferred.
Summary, Conclusion and Recommendations
This study has critically analyzed the studies by various researchers in the area surrounding the consequences of accounting treatment on R&D investment in firms with significant R&D expenditures culminating to an internally generated intangible. There is significant statistical evidence that intangible assets are becoming the key creator of value for a sizeable number of firms in the modern economy. However, this kind of investment has been given inadequate accounting treatment as have been discussed in the earlier sections. This is because, within the accounting framework, there is a challenge in the identification, measurement and control of such intangible assets.
In summary, the lack of adequate accounting treatment on R&D investments has raised concerns with many researchers investigating the impact of current accounting treatment on R&D investment-intensive companies under the headings; the value relevance of financial reports; allocation of resources in the capital market; growth of intangibles investments; and the firm's market value.
Studies conducted to probe how accounting treatment of nonfinancial information affect the value of financial reporting has resulted into varied findings, with different authors giving opposing findings on the same. The differing conclusions arrived at by the authors are as a result of failing to capture information related to intangible assets in financial statements of firms, and hence missing it out as an input in the valuation models of companies. On the other hand, incorporating voluntarily disclosed intangible assets in the company's financial statements has been viewed as taking care of the loss of value on financial information. Profitability is also found to increase with increase in R&D investment as evidenced in studies by studies [7,68]. Studies on mean-variance analysis of firms by authors found a positive relationship between investment in intangible assets and equity returns [26].
Studies done on the misallocation of resources in the capital market has identified lack of adequate accounting treatment of R&D investments as cause of misallocation of resources for R&D-intensive firms. However, latest studies have indicated that enhanced disclosure of the existence of intangible investments contributes to improved efficiency in market operations at large and reduce the cost of capital while giving way to better resource allocation in the capital market.
Studies on growth of R&D, an internally generated intangible for companies, has shown that accounting treatment of R&D investment did not substantially hinder its growth. Supporting this finding is the fact that companies communicate information of their investments in intangibles annually by way of financial reports, conference calls, websites, and press releases among other channels of disclosure. Another study validated this finding using R&D data of sampled UK and other International companies within the period of 2001 to 2007. Some studies show that increased level of investment in R&D can lead to enhanced disclosure of internally generated intangibles [72]. A Critical Literature Review Studies on inadequate accountability of R&D investment as compared to the market value show that there is systematic misevaluation of companies that invest heavily in R&D. Some studies indicate undervaluation while others indicate overvaluation of such companies. However, these studies show that the misevaluation of companies can be mitigated by providing adequate disclosure on R&D and other internally generated intangibles. Capitalizing R&D costs has been found to mitigate the under-investment in high-tech firms as evidenced in computer software firms especially in the U.S., which allows such companies to capitalize these cost, as opposed to their counterparts in other industries [25]. However, there is evidence that firm managers may take advantage of capitalizing such costs to manage earnings [25,36].
From the discussions summarized above, it is evident that disclosure of intangibles can greatly improve the financial information provided to users of financial reports. This will as well reward firms in the capital markets. Therefore, the management of firms should provide adequate information about their intangibles investments, and specifically, the internally generated intangibles such as R&D more prominently. The accounting standards setters should pay attention to accounting for intangibles and come up with universal guidelines of reporting such internally generated intangibles.
Most researchers agree that R&D investment significantly influence the value of a firm and therefore needs to be capitalized in the financial statements in order to reflect this value. This would enhance efficiency and improve accuracy in valuing and analyzing firms. Analysts will therefore eliminate the problem of undervaluing these firms, and this, will in turn lead to efficiency in allocating resources to these firms by investors in the capital markets.
The critical analysis of literature and empirical studies still leaves room for discourse and study. This is because there is no conclusive theory about the universal accounting standards on R&D investment applied across the financial statements of firms, leading to non-uniformity of financial reporting and hence misevaluation of firms. Therefore, more studies should be carried out to find the best measurement and reporting models for R&D costs, and by extension, intangible assets that will eliminate subjective assignment of costs. Also studies should be carried out on the best methods of classifying companies with various levels or intensity of R&D investments and other intangible assets so as to find the best measurement and reporting models for each category. | 8,827 | 2021-04-07T00:00:00.000 | [
"Business",
"Economics"
] |
Nailfold capillaroscopy in Egyptian systemic lupus erythematosus (SLE) patients: correlation with demographic features and serum levels of IL 17A and IFNs I
In SLE patients, cytokines are linked to endothelial cell damage. Nailfold capillaroscopy (NFC) is a simple method for evaluating micro-vascular abnormalities in different connective tissue diseases (CTDs). The study aimed to detect the levels of interleukin 17A (IL 17A), type I interferons (IFNs I) in the serum, and NFC changes in Egyptian SLE patients compared to a control group and to correlate NFC findings with patients’ demographic features and serum levels of IL 17A and IFNs I. Serum levels of IL 17A, IFN α, and IFN β were significantly higher in SLE patients than in control group (P < 0.0001). About thirty nine patients (73.6%) of the 53 SLE patients showed abnormal NFC changes. Egyptian SLE patients had a high prevalence of the NFC non-specific pattern, with 32 (60.4%) patients showing non-specific changes and 7 (13.2%) patients showing scleroderma pattern, including 3 (5.6%) patients with active scleroderma pattern and 4 (7.55%) patients with late scleroderma pattern. Furthermore, Raynaud’s phenomenon (RP) was observed in 8 (15.1%) SLE patients, with 3 (5.6%) having normal NFC pattern and 5 (9.4%) having scleroderma pattern. All controls (n = 20) showed normal hairpin shape capillaries. Except for SLEDAI (P = 0.03) and the presence of RP (P < 0.0001), there were no significant differences in demographic and laboratory parameters between the three NFC patterns (normal, non-specific, and scleroderma); additionally, NFC score correlated significantly with SLEDAI (P = 0.021). As a result of the high disease activity, Egyptian SLE patients had elevated serum levels of IL 17A and IFNs I. The most common NFC pattern in Egyptian SLE patients was a non-specific pattern. NFC abnormalities in Egyptian SLE patients were correlated with disease activity but not with patients’ ages, disease duration, or serum levels of IL 17A and IFNs I. SLE patients with scleroderma NFC pattern and RP should be closely followed for the possibility of appearance of anti-U1 RNP antibodies and MCTDS.
Background
Systemic lupus erythematosus (SLE) is an inflammatory autoimmune disease with a wide range of clinical symptoms and a chronic course with flares and remissions.The pathophysiology of SLE is significantly influenced by the activation of endothelial cells, as well as immunological abnormalities [1].Endothelial cells produce a variety of substances that regulate vascular tone, the immune system, and the coagulation system; they are also a goal for cytokines produced during the inflammatory process [2].Vascular injury has been considered as the origin of tissue damage in SLE pathogenesis [1].
IL 17A and IFNs I, including IFN α and IFN β, are cytokines that damage endothelial cells [3,4].Many studies in SLE patients [4][5][6][7] found high serum levels of these cytokines.Recent studies suggests that IL 17A is associated with SLE pathogenesis and also has a crucial role in initiating angiogenesis by stimulating vascular endothelial growth factor (VEGF), an essential factor in the process of new vessels formation [8,9].IFN-α and β have anti-proliferative and anti-angiogenic activity [10], and their high levels reduce levels of endothelial progenitor cells (EPCs) [3], which are important in endothelial repair and contribute to endothelial dysfunction.
Nailfold capillaroscopy (NFC) is a highly effective, lowcost, and simple imaging method used in the morphological analysis of capillaries in the nailfold area [11], with the added benefit of detecting micro-vascular changes early in some inflammatory CTDs.While NFC has been widely utilized for diagnosing systemic sclerosis and scleroderma-like diseases [12], there is a growing body of evidence suggesting its potential utility in detecting early microangiopathic changes in SLE [13].Moreover, recent studies have indicated a correlation between microcapillary pathological alterations and the clinical progression of SLE [13].
Few studies have been conducted on the correlation between serum cytokine levels and microvascular abnormalities observed by NFC in SLE patients.Therefore, the study aimed to detect the levels of IL 17A, IFNs I in the serum, and NFC changes in Egyptian SLE patients compared to a control group and to correlate NFC findings with patients' demographic features, IL 17A and IFNs I levels.
Patients and methods
Fifty-three SLE patients (46 women and 7 men, mean age 32.5 ± 10.6 years) were recruited for the study, according to the updated 1982 American College of Rheumatology (ACR) criteria for SLE [14], from the out-patient clinics of immunology and rheumatology department from January 2021 to August 2022.The local ethics committee approved the study (D-1-2022) and all participants received informed consent before the beginning of the study.Patients with any of the following diseases were excluded from the study: diabetes mellitus, uncontrolled hypertension, overlap syndrome, malignancy, and chronic infections.On the same day of collecting blood samples, the physicians applied the physical examination for the patients and the rheumatologist performed NFC measurements for all subjects.The patients were also evaluated for the existence of Raynaud's phenomenon (RP).The systemic lupus erythematosus diseases activity index (SLEDAI) was used for the assessment of disease activity [15] with a maximum score of 105 points and a score more than or equal to 12 points was considered an active case.The control sera were obtained from 20 healthy volunteers matched for age and sex.
The blood samples were obtained from all subjects and the serum was isolated and stored as frozen at − 80 °C for later ELISA experiments.The laboratory investigations were performed for all subjects.Serum complement levels (C3 and C4) were determined by using a quantitative competitive fluorescent probe technique, AssayLite ™ Multiplex EFCIA Kit.Serum level of anti-phospholipid antibody (APL) (lupus anticoagulant) was measured by ELISA assay (MYBIOSOURCE kit).Serum levels of anticardiolipin antibody (ACL) were measured by using ELISA assay (DIAPHARMA kit); in addition, anti-double strand deoxyribonucleic acid (anti-dsDNA) and antinuclear antibody (ANA) concentrations were determined by using ELISA assay (SIGNOSIS kit).Serum levels of IL-17A, IFN-α and β were measured by using ELISA assay (ELABSCIENCE kit and RAYBiotech kit, respectively).
NFC measurements
NFC was performed for all subjects by a rheumatologist with a background in NFC using an XW 880 USB digital microscope with LED light, magnification power × 400, an 8 inch LCD monitor, and a 0.38 megapixel camera (Hefei Golden Brains optical instrument Co., China).Before the examination, each of the included subjects sat for 20 min in a room with a temperature of 20-24 °C.Both the middle and ring fingers from the right and left hands were examined to detect NFC abnormalities in Egyptian SLE patients.A drop of immersion oil was added to the nailfold bed to improve visualization.The following parameters were recorded: capillary density (number of capillaries counted in 1 mm area), capillary length (normal or elongated ≥ 300 μm), apical loop diameter including dilated capillaries (20-50 μm) and giant capillaries (> 50 μm), capillary morphology (shape of subject capillaries), and presence or absence of hemorrhages [16,17].According to Smith et al. [18], a qualitative analysis was adopted to evaluate capillaroscopic changes.The qualitative analysis includes three patterns: normal (pattern of typical hairpin-like capillaries with a regular distribution), non-specific (pattern with density, dimension, and morphological abnormalities but not meeting the definition of a 'scleroderma pattern'), and scleroderma (pattern with giant capillaries, hemorrhages, and avascularity).Scleroderma pattern was divided into three subgroups (early, active, late) according to the degree of NFC abnormalities, where "early scleroderma pattern" was described as a pattern with normal density and capillary morphology, but has giant capillary(> 50 μm) with or without occurrence of hemorrhages, "active scleroderma pattern" was explained as a pattern with lowered density (4-6 capillary/mm), giant capillary (> 50 μm), abnormal morphology, and with presence or absence of hemorrhages, "late scleroderma pattern" was defined as a pattern with very low density (≤ 3 capillary/mm), abnormal morphology, and with absence of hemorrhages and giant capillaries.According to Sulli et al. [19], a semi quantitative analysis was also adopted which rates capillaroscopic changes from 0 to 3 (0 = no changes, 1 = ≤ 33% of capillary alterations/reduction, 2 = 33-66% of capillary alterations/reduction, 3 = ≥ 66% of capillary alterations/ reduction, per linear mm).
Statistical analysis
The data were analyzed by t-test for unpaired samples and one way ANOVA to evaluate the significance of difference between means.The probability difference in frequency distributions was measured by using chi-square test or Fisher's exact test.The correlation between data was performed by using Spearman's rank coefficient.P value was considered significant when it was less than 0.05.
NFC changes were observed in 39 of 53 (73.6%)SLE patients, with 32 (60.4%) showing non-specific changes (Fig. 1) and 7 (13.2%)showing scleroderma pattern, including active scleroderma pattern in 3 (5.6%)(Fig. 2a), and late scleroderma pattern in 4 (7.55%) (Fig. 2b).Controls were compared to SLE patients in terms of age, sex, serum cytokine levels, and NFC changes as shown in Table 2. Serum levels of IL 17A, IFN α, and β were significantly higher (P < 0.0001) in SLE patients than in the control group.According to Table 2, most of SLE patients had elongated, crossed, tortuous, and dilated capillaries.In the control group, NFC changes were normal, with hairpin shape capillaries.When the previous two groups were compared in terms of NFC changes, there was a significant difference in capillary density (P = 0.02), capillary length (P = 0.0073), capillary elongation (P < 0.0001), capillary dilation (P = 0.015), and capillary tortuosity (P < 0.0001).
As shown in Table 3, NFC score has significant positive correlation with SLEDAI (P = 0.021).SLEDAI, on the other hand, showed significant positive correlations with disease duration, IL 17A, IFN α, and IFN β.
Table 4 shows the demographic and laboratory features of SLE patients based on NFC patterns (normal, nonspecific, and scleroderma).RP was seen in three (21.43%) of SLE patients with a normal pattern and five (71.4%)
Discussion
Many studies have recently focused on microvascular endothelial damage and its role in the pathogenesis of systemic organ involvement in rheumatic diseases like [21].IL 17 family consists of six members (IL 17A to IL 17F) and the previous studies has found that 17A and IL 17F, in particular, can initiate tissue injury through the secretion of chemokines that cause monocytes recruitment, maturation, and proliferation.[22].IL 17A is a cytokine with inflammatory properties that participates in the host defenses against bacterial and fungal infections as well as autoimmunity and tumors [23].Studies reported that IL 17A is associated with SLE pathogenesis because it can activate B cells and cause local inflammation and tissue injury, which are associated with different events in the pathophysiology of SLE [24].It can also magnify the immune response by increasing the secretion of autoantibodies by B lymphocyte activation, making it an appealing therapeutic target [25].The present study demonstrated that individuals diagnosed with SLE and exhibiting active disease displayed elevated levels of IL-17A in comparison to those with mild or moderate disease and control subjects.Furthermore, a notable significant positive correlation was observed between IL 17A and SLEDAI.This correlation suggests that raised IL-17A levels contribute to increased disease activity or vice versa.These findings were consistent with previous studies' findings [26][27][28][29].
Type I IFNs, which are primarily secreted by plasmacytoid dendritic cells (pDCs), participate in the pathogenesis of SLE [30].It was found that most of SLE symptoms are related to elevated levels of IFNs I.There are two types of cells that control vascular repair: bone marrowderived EPCs and myelomonocytic circulating angiogenic cells (CACs) [31,32].Type I IFNs have a possibility of disrupt EPC/CAC function by decreasing pro-angiogenic factors such as VEGF and increasing IL-18 [32].An in vitro study that found that blocking IFN-α in SLE patients' peripheral blood mononuclear cells (PBMCs) restored the normal angiogenic phenotype [31] provides evidence that type I IFNs promote abnormal vascular repair.The present study found elevated serum levels of type I IFNs in SLE patients when compared to the control group, with a significant positive correlation with SLE-DAI, implying that as serum levels of IFNs I increase, disease activity increases, which is in agreement with the previous studies [33][34][35].
NFC has been used as a non-invasive method for detecting microvascular involvement in rheumatic diseases [11].Capillary abnormalities have been reported in a variable prevalence in SLE patients [36][37][38].Furthermore, the capillaroscopic changes in the nailfold appear to be related to the disease activity score [36].Both the middle and ring fingers from the right and left hands were examined in the current study, as the previous multicenter studies showed that these fingers have a high sensitivity to detect capillary abnormalities [39,40].According to the findings of this study, 39 out of 53 (73.6%)SLE patients have NFC abnormalities, with nonspecific pattern being the most prevalent pattern in SLE patients, followed by normal and scleroderma patterns; additionally, patients with scleroderma pattern showed only active and late pattern.The previous findings were consistent with other studies [41,42] that found that the majority of SLE patients have a non-specific NFC pattern with rare occurrence of scleroderma pattern.
The tortuosity of capillaries exhibited a statistically significant increase in individuals diagnosed with SLE compared to the control group in this study.Tortuosity has been described as a normal variation in some studies [43,44], but it has also been defined as an SLE pattern in others [11,38].Moreover, a notable significant difference in capillary density was obtained between the control group and the patient group.Specifically, a majority of individuals diagnosed with systemic lupus exhibited capillary density of below 7 capillary/mm.The prevailing NFC alterations observed in the Egyptian individuals with SLE include dilatation, elongation, and capillary crossing.These findings provide evidence to demonstrate that most SLE patients have non-specific abnormalities.
In addition, there was a correlation observed between changes in NFC and SLEDAI scores, whereas discernible variations in scleroderma patterns were observed between patients with inactive SLE and those with active SLE.Patients diagnosed with inactive SLE exhibited normal NFC, whereas active SLE patients had an abnormal pattern.These findings are consistent with many other studies that have found a correlation between capillary abnormalities and disease activity in SLE [45,46].Furthermore, the occurrence of hemorrhages was higher in active SLE cases than in inactive cases, which is consistent with the previous studies that demonstrated a higher frequency of hemorrhages in SLE patients with higher disease activity [47].In contrast to the findings of Nakajima et al. [48], who found that NFC patterns correlated with subjects' age, the current study found no significant correlation between NFC score and the ages of SLE patients.
The study found that SLE patients with different NFC patterns, including normal, nonspecific, and scleroderma, had no significant difference in IL 17A, IFN α, and β levels.Furthermore, there was no significant correlation between NFC score and previous serum cytokine levels, indicating that there was no clear relationship between NFC abnormalities and immunologic markers.When the previous three NFC patterns were compared based on the presence of RP, a significant difference was found, with scleroderma pattern patients having a higher occurrence of this phenomenon than the other two patterns, which was consistent with the findings of Furtado et al. who discovered a significant correlation between the presence of RP and the scleroderma NFC pattern.Because no clinical signs of scleroderma were observed in the SLE patients with RP, these patients need close follow up for the possibility of the development of anti-U1 ribonucleoprotein antibodies (anti-U1 RNP) and mixed connective tissue diseases (MCTDs), as the previous studies [50,51] discovered that the development of anti-U1 RNP antibodies is related to the presence of RP, which may lead to MCTDs in these patients.Moreover, no significant difference in antiphospholipid antibodies was found in the current study between different NFC patterns, which agrees with Raeeskarami et al. [52] who demonstrated that there was no significant correlation between antiphospholipid antibodies and capillary alterations and contrasts with other studies that used capillaroscopic changes as a diagnostic test for the antiphospholipid syndrome in rheumatic disease [42,53,54].Our findings suggest that IL 17A and IFNs I have essential roles in the pathogenesis of SLE, but they may not have a specific effect on SLE microvascular abnormalities; additionally, NFC changes in Egyptian SLE patients are affected by disease activity rather than age or disease duration.NFC abnormalities in SLE can act as a mirror for microvascular involvement and disease activity.
Conclusion
The present study found that patients with SLE had elevated levels of IL 17A and IFNs I in their serum due to increased disease activity.Furthermore, the study's findings revealed that non-specific NFC patterns were more common in SLE patients.Notably, there was a significant positive correlation between NFC changes and disease activity but no correlation between NFC changes and studied cytokines.SLE patients with scleroderma NFC pattern and RP should be closely followed up for the possibility of appearance of anti-U1 RNP antibodies and MCTDS.
Fig. 1 Fig. 2
Fig. 1 Examples of non-specific pattern in SLE patients.A Abnormal capillary density.B Abnormal capillary dimension.C Abnormal capillary density, dimension, and morphology
Table 1
Detailed demographic, clinical, laboratory features and drug history of 53 SLE patients SLEDAI systemic lupus erythematosus disease activity index, RP + positive for Raynaud's phenomenon, ANA anti-nuclear antibody, Anti-dsDNA anti-double strand DNA, APL + positive for anti-phospholipid antibody, ACL + positive for anticardiolipin antibody, ESR erythrocyte sedimentation rate.SD standard deviation.Data were presented by mean ± standard deviation and number/ percentage (n/%)
Table 2
Differences in the age, sex, serum cytokines level, and NFC changes between SLE patients and control group IL-17A interleukin 17A, IFN-α interferon alpha, IFN-β interferon beta, NS non-significant.Data were presented by mean ± standard deviation and number/percentage (n/%).P was considered significant when it was ≤ 0.05 and non-significant when more than 0.05
Table 3
NFC score and SLEDAI correlations with age, disease duration, and serum cytokines levels in SLE patients NFC nailfold capillaroscopy, SLEDAI systemic lupus erythematosus disease activity index, IL-17A interleukin 17A, IFN-α interferon alpha, IFN-β interferon beta, NS non-significant.P was considered significant when it was ≤ 0.05 and non-significant when more than 0.05
Table 4
Demographic and laboratory differences in SLE patients according to NFC patterns Th 17 cells, CD 8 + cells, γδ T cells, and natural killer T cells SLEDAI systemic lupus erythematosus disease activity index, RP + positive for Reynaud's phenomenon, IL-17A interleukin 17A, IFN-α interferon alpha, IFN-β interferon beta, APL + positive for anti-phospholipid antibody, NS non-significant.Data were presented by mean ± standard deviation and number/percentage (n/%).P was considered significant when it was ≤ 0.05 and non-significant when more than 0.05ParametersNormal pattern (n = 14) Non-specific pattern (n = 32) Scleroderma pattern (n = 7) P value | 4,229.4 | 2023-09-19T00:00:00.000 | [
"Medicine",
"Biology"
] |
Progress of MRI Radiomics in Hepatocellular Carcinoma
Background Hepatocellular carcinoma (HCC) is the sixth most common cancer in the world and the third leading cause of cancer-related death. Although the diagnostic scheme of HCC is currently undergoing refinement, the prognosis of HCC is still not satisfactory. In addition to certain factors, such as tumor size and number and vascular invasion displayed on traditional imaging, some histopathological features and gene expression parameters are also important for the prognosis of HCC patients. However, most parameters are based on postoperative pathological examinations, which cannot help with preoperative decision-making. As a new field, radiomics extracts high-throughput imaging data from different types of images to build models and predict clinical outcomes noninvasively before surgery, rendering it a powerful aid for making personalized treatment decisions preoperatively. Objective This study reviewed the workflow of radiomics and the research progress on magnetic resonance imaging (MRI) radiomics in the diagnosis and treatment of HCC. Methods A literature review was conducted by searching PubMed for search of relevant peer-reviewed articles published from May 2017 to June 2021.The search keywords included HCC, MRI, radiomics, deep learning, artificial intelligence, machine learning, neural network, texture analysis, diagnosis, histopathology, microvascular invasion, surgical resection, radiofrequency, recurrence, relapse, transarterial chemoembolization, targeted therapy, immunotherapy, therapeutic response, and prognosis. Results Radiomics features on MRI can be used as biomarkers to determine the differential diagnosis, histological grade, microvascular invasion status, gene expression status, local and systemic therapeutic responses, and prognosis of HCC patients. Conclusion Radiomics is a promising new imaging method. MRI radiomics has high application value in the diagnosis and treatment of HCC.
INTRODUCTION
Hepatocellular carcinoma (HCC) is the sixth most common cancer and the third leading cause of cancer-related death worldwide (1). Although the diagnostic criteria of HCC continue to improve, its prognosis remains unsatisfactory (2). In addition to certain factors, such as tumor size and number and vascular invasion displayed on traditional imaging, some histopathological features and gene expression parameters are also important in the prognoses of patients with HCC. However, many current staging systems for HCC have not taken into consideration the above-mentioned histopathological features or genetic traits beyond the size and number and vascular invasion of the tumor (3,4). Most parameters are based on postoperative pathological examinations, which cannot help with preoperative decision-making. To better stratify HCC patients before surgery, make more accurate treatment decisions, and improve the prognoses of patients, there is an urgent need for a noninvasive method that can accurately predict the histopathological features and gene expression parameters before surgery. The rapid development of artificial intelligence has played an important role in personalized precision medicine (5). Radiomics, a new technology, can transform the potential histopathological and physiological information in images into high-dimensional quantitative image features that can be mined (6,7).The study of radiomics will contribute to the early diagnosis and treatment of HCC and ultimately improve survival (8,9). In recent years, many studies have confirmed the application values of magnetic resonance imaging (MRI) radiomics in the diagnosis and differentiation (10,11), histological grading (12,13), microvascular invasion (MVI) assessment (14,15), radiogenomics (16,17),prediction of relapse and prognosis after surgical resection (18)(19)(20), response to transarterial chemoembolization(TACE) (21,22) and systemic treatment efficacy of HCC (23).
To better understand the research hotspots and trends of MRI radiomics in HCC, we used PubMed to identify important recent publications on MRI radiomics in HCC, selected research articles and reviews and used bibliometric method to visually analyze the countries, institution, authors, and keywords of MRI radiomics in HCC. Meanwhile, this study reviews the radiomics workflow from image acquisition and reconstruction, segmentation, feature extraction, feature selection and modeling to model validation, and the research progress of MRI radiomics in HCC.
BIBLIOMETRICS OF MRI RADIOMICS IN HCC
only recognize files named "download *.txt", the files were renamed accordingly. The bibliometric software CiteSpace5.7.R2 (64 bits) was utilized for this study to visually analyze the countries, institution, authors, and keywords draw relevant charts. The articles originated from a total of 12 countries, and the top five countries were China (24), the USA (13), South Korea (5), Germany (3), and France (2). A total of 30 institutions published manuscripts independently or cooperatively. The top five institutions were the Chinese Academy of Sciences (12), Fudan University (10), GE Healthcare (8), Sun Yat-Sen University (7), and Sichuan University (5). Bin Song and Xin Li were the most prolific authors. Meng-Su Zeng, Jie Tian, and Dong-Sheng Gu were also active in this field. "Hepatocellular carcinoma" was the most important term, followed by "radiomics", "recurrence", and "microvascular invasion". According to the link strength of keyword cooccurrence, the network was divided into eight clusters, and the largest cluster was "tumor differentiation (#0)" (Figure 2).
RADIOMICS WORKFLOW
Radiomics extracts high-throughput features from images and transforms imaging data into high-resolution mining data spaces through machine learning (25). Quantitative radiological data can therefore be extracted and applied to clinical decisionmaking (25). The workflow of radiomics usually includes five steps (6), which are described below.
IMAGE ACQUISITION AND RECONSTRUCTION
Imaging techniques that can be used for radiomics include MRI, computed tomography (CT), positron-emission tomography, and ultrasound. Among them, MRI has the advantage of depicting more soft-tissue features. Radiomics is an imaging analysis method; thus, it is vital to standardize high-quality images (26)(27)(28). This makes it necessary to preprocess the imaging data; otherwise, a widely promoted standard scanning protocol is needed to reduce the variability in radiomic features and improve the performance of radiomic models (25,29).
IMAGE SEGMENTATION
Manual, automatic, and semiautomatic segmentation are often used to segment the volume or region of interest in a target tissue (30). Manual segmentation is most reliable, but it involves intraobserver and interobserver variability. Its labor and time cost are high. The segmentation of an image often requires multiple clinicians or the same clinician at multiple times. The intraobserver and interobserver variability can be improved by screening the intraobserver and interobserver consistency. The purpose of automatic segmentation is to mark the regions of interest automatically by a computer. Semiautomatic segmentation involves manual corrections. Automatic segmentation algorithms include image segmentation based on thresholds, image segmentation based on region growing, and image segmentation based on edge detection. Some classical algorithms perform well at delineating liver lesions (31,32).
IMAGE FEATURE EXTRACTION
Image features include semantic features and nonsemanticfeatures (33). Semantic features include qualitative (shapes, boundaries, etc.) and quantitative features, and their analysis depends on the radiologist's knowledge. Nonsemantic features are quantitative descriptors extracted from tissues of interest, including shape and statistical features (34). The shape features of objects in images include topological features, distances, perimeters, areas, geometric features, and descriptions of shape and orientation. Statistical features can be further divided into first-order, secondorder, and high-order features. First-order features are usually called density features, which involve gray-level histogram information simply describing the global distribution of gray levels in an image. Such features cannot describe the local distribution of gray levels in an image or the spatial position of each gray level (35). Second-order features are often called texture features. These reflect the relationships between adjacent voxels. High-order features are usually called filtering features and are generated by wavelet and Laplacian Gaussian filtering, for example, in addition to first-order and secondorder features.
FEATURE SELECTION AND MODELING
Many features can be extracted from a high-throughput image, but using all the features to analyze an image will lead to overfitting. The best features can be selected by dimensionality reduction to improve the efficiency of the model. The methods of feature selection can be divided into three categories: filter, wrapper, and embedded (36). The goal of radiomics is to establish a prediction model for clinical outcomes from selected features. The modeling methods include logistic regression, k-nearest neighbor, decision trees, ensemble learning, and support vector machines. It is recommended to test the effectiveness of several forecasting models to select the model with the best performance (37).
MODEL VALIDATION
The prediction model can be validated by internal crossvalidation, such that the model can be further optimized and the prediction performance can be maximized. Validation of the model should be carried out in a separate cohort (37). For differentiation analysis, the receiver operating characteristic (ROC) curve is the most commonly used method to evaluate the performance of the model. The area under the ROC curve (AUC) or the sensitivity and specificity of the model can be used to evaluate whether the model can predict clinical outcomes. For survival analysis, the concordance index (C-index) and the timerelated ROC curve are usually used for validation (38).
DIAGNOSIS AND DIFFERENTIATION
At present, the diagnosis of HCC is mainly based on imaging methods such as MRI, CT, and ultrasound. Because HCC has a typical enhancement mode, contrast-enhanced CT and dynamic contrast-enhanced MRI play important roles in the diagnosis of HCC (39)(40)(41)(42). The European Association for the Study of the Liver standard (40) and the Liver Imaging Reporting and Data System (43) are widely recognized. However, the evaluation of imaging features may be subjective because radiologists have different experiences and different familiarities with the system (44,45). Radiomics has important application value in the diagnosis of solid tumors because it uses advanced image processing technology to extract high-throughput data and quantitative analysis of tumor behavior and heterogeneity (6,(46)(47)(48)(49)(50)(51).
Radiomics signatures based on conventional precontrast T1-weighted imaging, postcontrast T1-weighted imaging, T2-weighted imaging, diffusion-weighted imaging (DWI), and intravoxel incoherent motion (IVIM), whether alone or in combination with clinical data, are all valuable for HCC differentiation (52)(53)(54)(55)(56)(57)(58)(59), and their differentiation efficiency is almost equal to that of experienced radiologists (10-year experience) (52). HCC, intrahepatic cholangiocarcinoma (ICC), and HCC-ICC have common risk factors (60,61), and their typical qualitative MRI features may overlap (24,(62)(63)(64). Therefore, the conventional MRI diagnosis of HCC is still uncertain. According to Liu et al. (54), the imaging features extracted from MR images have great potential to differentiate combined hepatocellular cholangiocarcinoma from cholangiocarcinoma and HCC, showing a maximum AUC of 0.77. Recently, Zhu et al. (56) studied the application value of histogram features on IVIM-DWI in the differential diagnosis of HCC. They found that the histogram parameters of IVIM-DWI could distinguish hepatic hemangiomas, hepatic cysts, and HCC and that the volume of the pseudodiffusion coefficient and perfusion fraction had better diagnostic value than other histogram parameters (56).
In recent years, DL technology has been developed and has achieved excellent performance in the classification of hepatic lesions (65-71). Hamm CA et al. (65) developed a proof-ofconcept convolutional neural network (CNN)-based DL system and classified 494 hepatic lesions from six categories on MRI. The system demonstrated 92% accuracy, 92% sensitivity and 98% specificity, and their results showed a 90% sensitivity for classifying HCC compared to 60%/70% for radiologists.
HISTOLOGICAL GRADING
The histological grading of HCC is key to determining the best treatment scheme and prognosis of a patient. High-grade HCC patients have a higher intrahepatic relapse rate than lowgrade HCC patients (72,73), and most high-grade HCC patients need larger safe resection margins and more frequent postoperative follow-up visits (74,75). The radiomic features of precontrast T1-weighted imaging, postcontrast T1-weighted imaging, and T2-weighted imaging, whether alone or in combination with clinical data (76), are all valuable for identifying poorly differentiated HCC (13,(76)(77)(78)(79)(80). In addition, recent studies have shown the application value of functional MRI radiomics based on IVIM-DWI in predicting the pathological grade of HCC (12,81,82
MVI
MVI is diagnosed depending on postoperative tissue specimens, but detection by conventional imaging is difficult. The presence of MVI indicates that the tumor has strong biological invasiveness, which can increase the relapse rate of HCC more than fourfold (83,84). Accurate preoperative prediction of MVI of HCC can help doctors adjust treatment strategies in a timely manner (such as expanding the resection range), optimize treatment plans, reduce the risks of postoperative relapse, and improve the prognosis (84,85). Enhanced MRI is helpful to predict MVI in HCC (83,(86)(87)(88)(89)(90)(91). MRI-based radiomics (15,(92)(93)(94)(95)(96)(97)(98)(99)(100)(101)(102) and DL systems (103-106) have shown good performance in predicting MVI in HCC. The increase in clinicopathological risk factors and qualitative imaging features can improve the prediction efficiency of the model (14,98,107,108). Li et al. (101) found that tumor volume-based IVIM histogram analysis can be used to predict MVI and that the fifth percentile of the true diffusion coefficient is most beneficial to predict MVI of HCC. Zhang et al. (107) extracted imaging features based on preoperative multimodal MR images and constructed an MVI prediction model (combined model) by combining the clinical features and qualitative imaging features of patients with HCC. The AUC in the validation cohort of their combined model was 0.858, which was higher than the AUC (0.820) in the validation cohort of the model constructed from individual radiomic features, indicating that the prediction efficiency of the combined model was higher. Song D et al. (104) predicted MVI using radiomics and DL in 601 patients with HCC based on preoperative MRI. Their results showed that the radiomics model achieved an AUC of 0.731, the DL model based only on MRI images achieved an AUC of 0.915, and a DL model combined with clinical parameters achieved an AUC of 0.931. These studies indicated that the model combining radiomics, DL, and clinical parameters showed the best predictive performance.
RADIOGENOMICS
The biological behavior of a tumor is closely related to its gene expression profile. Biopsy is a widely used method to evaluate gene expression before surgery, but biopsy is an invasive examination that may cause bleeding and other complications. Therefore, patients are often unwilling to undergo this examination. In recent years, radiogenomics has gradually become more widely applied in HCC research. The purpose of radiogenomics is to determine the relationship between semantic and quantitative image data and genomic and molecular measurements, thus constructing correlation diagrams related to results or other clinical measurements (33,109,110). Segal et al. (111) evaluated the correlation between radiogenomic features and the liver cancer gene phenotype and reported that 78% of liver cancer gene expression profiles could be reconstructed by this combination of features. To date, radiogenomics studies have described the semantic features obtained from MRI (112)(113)(114)(115). MRI radiogenomics has the value of predicting gene features with prognostic and therapeutic significance (16,17,(116)(117)(118)(119)(120)(121)(122)(123). Taouli et al. (113) found that there was a strong connection between imaging features, such as the "infiltrative pattern", "mosaic appearance", and "presence of macrovascular invasion", and an aggressive genomic signature determined previously. Shi et al. (82) found that histogram indices extracted from IVIM parameter maps could predict Ki-67 expression. Jun et al. (124) used an immunohistochemical method to detect the expression of programmed cell death-1 (PD-1) and programmed cell death ligand-1 (PD-L1) in 98 ICC patients and extracted radiological features from the arterial phase and portal venous phase of preoperative MR images. The results indicated that the AUCs of the models for predicting PD-1 and PD-L1 expression were 0.897 and 0.897, respectively. The prognoses of PD-1-positive and PD-L1-positive patients were worse than those of PD-1-negative and PD-L1-negative patients, and their 5-year survival rates were 12.5%, 48.3%, 21.9%, and 39.4%, respectively (P < 0.05). The results indicated that MRI radiomics could be used as a noninvasive biomarker to evaluate the expression of PD-1 and PD-L1 and the prognosis of ICC patients ( Table 1).
PREDICTION OF RELAPSE AND PROGNOSIS AFTER SURGICAL RESECTION
Surgical resection is still the main treatment for patients with early HCC (125). However, tumor relapse is still the main cause Radiomics features extracted from MR images correlate with quantitative expression of the immune markers CD3, CD68 and CD31and expression of the immunotherapy targets PD-L1 at the protein level, as well as PD1 and CTLA4 at the mRNA level. Wang W et al. (17) 2020 China 227 The radiomics-based model performs better than the clinico-radiological model for predicting biliary-specific marker CK19 status of HCC. Gu D et al. (117) 2020 China 293 The MRI-based radiomics signature is significantly related to GPC3positivity (a prognosis factor, was associated with metastasis and recurrence after resection) in patients with HCC. Ye Z et al. (118) 2019 China 89 Texture analysis on preoperative enhanced MRI can be used to predict the status of the cell proliferation marker Ki-67 after curative resection in patients with HCC. Fan Y et al. (16) 2021 China 133 Texture analysis based on enhanced MRI can help identify VETC-positive HCC (histological vascular pattern, micrometastases, early recurrence and poor prognosis). Li Y et al. (119) 2019 China 83 Texture analysis of multiphase MRI images is helpful for predicting expression of the cell proliferation marker Ki-67 in HCC. Wang HQ et al. (120) 2019 China 86 Texture analysis based on MRI can help identifyCK19-positive HCC(tends to be related to a worse prognosis). Fan Y et al. (121) 2021 China 151 A combined model including artery phase radiomics score and serum AFP levels based on enhanced MRI can preoperatively predict expression of the cell proliferation marker Ki-67 in HCC. Huang X et al. (122) 2019 China 100 MRI radiomics features can be used to preoperatively differentiate dual-phenotype HCC from CK7-and CK19 (markers of cholangiocellular carcinoma) -negative HCC. Chen S et al. (123) 2019 China 207 Radiomics obtained from enhanced MRI can help predict the immunoscore (density of CD3+ and CD8+ T cells) in HCC. of postoperative death, and the 5-year relapse rate after surgery is close to 70% (126). Improving the ability to preoperatively identify these high-risk patients will guide surgical management, postoperative monitoring, and treatment intervention (127,128). The radiomic model based on preoperative MRI can be used as a new tool to predict early relapse (18,19,(129)(130)(131)(132)(133)(134), relapsefree survival (135) and overall survival (OS) (136,137) in patients with HCC after surgery. Hui et al. (130) used preoperative MRI to extract 290 texture parameters to predict the relapse of HCC patients within 730 days after surgical resection. The results showed that the prediction accuracy of texture features based on dynamic contrast-enhanced MRI in the equilibrium phase was 84%. Combining clinical, laboratory, and radiomic data can improve the performance of quantitative models (20,129,135,136,138). According to Kim et al. (135), the combined clinical and radiomic model had the same performance as the clinicopathological model in predicting early relapse. Zhang et al. evaluated the effectiveness of contrast-enhanced MRI radiomic features in predicting the OS of HCC patients after resection. Their results showed that preoperative clinical features and semantic imaging features were significantly correlated with survival rate; the Barcelona Clinic Liver Cancer stage, uneven tumor margin, and combined rad-score were independently correlated with OS; and the combined model incorporating radiological and radiomic features had a better prediction performance than the clinicradiological model (136).
PREDICTION OF RESPONSE TO TACE
TACE is recognized as an effective treatment for advanced HCC (125), but its long-term efficacy needs to be further improved (139)(140)(141). MRI radiomics can be used to predict the response to TACE treatment and provide a reference for the formulation of individualized treatment plans (21,22,(142)(143)(144)(145)(146)(147)(148). Sun et al. (142) predicted the risk of early postoperative progression based on multiparameter MRI data before TACE. The results showed that the AUC of the model based on DWI features was 0.786 and 0.729 when b=0 and b=500, respectively, followed by the AUC of T2-weighted imaging features (0.729) and the apparent diffusion coefficient (0.714). Compared with any single MRI signal, the MP-MRI signal had a higher AUC, at 0.800. Song et al. (143) revealed that their combined model incorporating radiomic features and clinical radiation risk factors had the best predictive value (C = 0.802).
PREDICTION OF THE SYSTEMIC TREATMENT EFFICACY
The treatment of HCC has been a challenge. Systemic therapies for HCC are current research hotspots. Targeted therapy with sorafenib (149) and lenvatinib (150) and immunotherapy with immune checkpoint inhibitors, especially antibodies against PD-1/PD-L1 pathway members (nivolumab and pembrolizumab), have achieved excellent clinical results (151)(152)(153)(154)(155)(156)(157)(158). These results strongly indicated that immune checkpoint inhibitor-based strategies will soon be primary method in the treatment of advanced HCC, and immunotherapy will introduce a new era of HCC therapy. Traditional contrast-enhanced CT and MRI, including functional imaging, are the most commonly used biomarkers for evaluating the therapeutic response in clinical practice (159)(160)(161)(162)(163)(164)(165)(166)(167)(168)(169)(170)(171)(172). Research based on contrastenhanced CT and MR images has shown the value of radiomics and DL in predicting systemic treatment efficacy for advanced HCC (23,(173)(174)(175). Muléet al. (174) analyzed the CT texture features of 92 patients before receiving sorafenib and found that the entropy of portal phase-derived entropy at fine texture scales was an independent predictor of OS, which was confirmed in their validation cohort. Yuan et al. (173) established a radiomics nomogram and measured its ability to evaluate the therapeutic efficacy of anti-PD-1antibodies in the treatment of HCC by combining pretreatment contrast-enhanced CT images and clinical risk factors. The results indicated that the AUCs of the radiomics nomogram were 0.894 and 0.883 in the training and validation cohorts, respectively. In recent years, MRI radiomics has gradually become more widely applied to systemic treatment evaluation of brain tumors (176,177). There are still no reports on using MRI radiomics to evaluate the systemic treatments of patients with HCC. We believe that as research progresses, MRI radiomics will play an important role in the evaluation of systemic treatments for HCC in the near future.
CONCLUSION
As a new technology, radiomics can improve the diagnosis and differentiation of HCC, as well as predictions of the stage, histological grade, MVI, gene expression, treatment response, and prognosis of HCC. This is because it allows us to analyze the relationship between high-dimensional quantitative imaging features and clinical and genetic data. Moreover, it is a powerful tool for making personalized treatment decisions before surgery. With the rapid development of targeted therapy and immunotherapy for HCC, radiomics is expected to become a reliable radiological marker for predicting the therapeutic targets and therapeutic responses of HCC patients.
There are still some challenges and limitations in the clinical application of radiomics. First, a key challenge is to ensure that the academic community can obtain high-quality radiological and clinical resources that involve the establishment and promotion of imaging and clinical data acquisition protocols. Second, the analytical methods of radiomics need to be standardized. Third, many radiomics studies are retrospective, whereas a prospective research design is ideal. As technology advances and research progresses, MRI radiomics will play a more important and even irreplaceable role in the diagnosis and treatment of HCC.
AUTHOR CONTRIBUTIONS
X-QG and LY wrote the paper. Y-YT, Y-KW, NinL, XY, RW, JZ, and GY contributed to the literature search and manuscript preparation. NiaL and X-HH revised the paper. X-QW performed the data analysis and created graphs. X-MZ and J-DL designed the research. All authors contributed to the article and approved the submitted version. | 5,220.6 | 2021-09-20T00:00:00.000 | [
"Medicine",
"Engineering"
] |
Gene expression rearrangements denoting changes in the biological state
In many situations, the gene expression signature is a unique marker of the biological state. We study the modification of the gene expression distribution function when the biological state of a system experiences a change. This change may be the result of a selective pressure, as in the Long Term Evolution Experiment with E. Coli populations, or the progression to Alzheimer disease in aged brains, or the progression from a normal tissue to the cancer state. The first two cases seem to belong to a class of transitions, where the initial and final states are relatively close to each other, and the distribution function for the differential expressions is short ranged, with a tail of only a few dozens of strongly varying genes. In the latter case, cancer, the initial and final states are far apart and separated by a low-fitness barrier. The distribution function shows a very heavy tail, with thousands of silenced and over-expressed genes. We characterize the biological states by means of their principal component representations, and the expression distribution functions by their maximal and minimal differential expression values and the exponents of the Pareto laws describing the tails.
Scientific Reports
| (2021) 11:8470 | https://doi.org/10.1038/s41598-021-87764-0 www.nature.com/scientificreports/ distribution functions. That is, in the over-expression branch we count the number of genes with differential expressions greater than or equal to a given d. In the under-expression branch, we count the number of genes with differential expressions lower than or equal to a given d. A d ≈ 1 means that the expression level of a gene has not changed, whereas d ≫ 1 or d ≪ 1 correspond to over-expressed or under-expressed (silenced) genes, respectively. In the studied samples, we found two kinds of GE rearrangements after a change in the biological state. In the first case, most genes take values near the reference ones, and only a small fraction of genes take significant differential expression values. The distribution function is rapidly decaying as d departs from 1. Because of the Pareto character, the decay law is 1/d υ , with a relatively large value for the exponent. This situation corresponds to relatively close initial and final states, and a "continuous" transition.
The second general case, on the other hand, is characterized by radical expression rearrangements and heavy tails in the distribution functions (small exponents), involving thousands of differentially expressed genes. It corresponds to initial and final states far apart in GE space, and a "discontinuous" transition.
In the next section, we use an analogy with physics in order to build up an intuition with regard to these two kinds of transitions.
Continuous and discontinuous transitions in Physics
In Fig. 1a, we draw a nearly harmonic potential well (dashed line). Under the action of a small amplitude noise, the motion of a particle in the well is characterized by a mean value for its position �x� = 0 , corresponding to the potential minimum. This abstract picture may represent a biological system. The x-axis is a coordinate in GE space, and the y-axis is the fitness with a minus sign, such that the minimum of the potential is the state with maximal fitness. Now, a small amplitude electric field is applied in the x direction. The resulting effective potential is drawn with a continuous line. A non zero minimum emerges. As time evolves, the result of the noisy motion is a mean position displacement from �x� = 0 to the new potential minimum. In the biological analogy, the electric field may be interpreted as a change in the external conditions, exerting new selection pressures. In the LTEE, for example, a fixed daily quantity of nutrients induce adaptation to this new conditions and a rise of fitness. The random noisy motion can be viewed as the result of mutations or epigenetic changes.
Figure 1. (a) and (b)
Illustration of a continuous transition. The addition of a small electric field to a harmonic potential well causes a modification of the minimum from �x 1 � = 0 (blue circle) to a nonzero value (red circle). (c) and (d) Illustration of a discontinuous transition in a double well with distant minima. Random fluctuations may drive a particle, initially in the left well, towards the right deepest well. The barrier separating the two minima should be surpassed. www.nature.com/scientificreports/ If the electric field is relatively small, the initial and final potential minima are relatively close to each other and the cloud described by the particle motion realizes a continuous transition between the minima (Fig. 1b).
The second situation is depicted in Fig. 1c. A double well with two distant minima is represented. The right minimum is deeper (higher fitness). This situation seems to describe cancer.
The initial (normal) state is prepared in the left well. It means that the particle starts realizing random motions from �x� = 0 . If the motions are of small amplitudes, the particle will remain in the left well for a long time because of the barrier preventing the transitions to the right well. Once a jump over the barrier takes place, the transition to a non zero mean value of x occurs. It is seen as a discontinuous transition, or a jump in the mean position of the cloud described by the random particle motions (Fig. 1d).
Gene expression rearrangements in the LTEE
The LTEE 10 is a formidable controlled evolution experiment with 12 E. Coli populations, followed for more than 60000 bacterial generations. We have studied some of the results coming from it 18,19 with the purpose of creating a model of mutations 20 . In the present section, we use the reported GE data 21 , involving measurements in 4290 genes, in order to analyze the transition from the initial (ancestral) state to a final state at generation 20000. Data is provided for 8 harvested clones, coming from two of the twelve evolving populations in the experiment, called Ara+1 and Ara-1. 8 samples from the ancestral populations are also measured. The Ara+ and Ara-tags denote two particular mutations that were isolated from the main strain and from which all 12 populations (6 of each) were replicated, nevertheless this characteristic is not relevant for our purposes, since in effect they are simply populations that evolve independently.
The conditions stressing the bacterial populations, i.e. the scarcity of nutrients, act since the very beginning of the experiment. The transition to the new state seems to be continuous, as suggested by the observed quasicontinuous variation of fitness as a function of time 22 . We shall verify how this transition is reflected in the principal component (PC) representation and in the rearrangement of the GE distribution function.
We show the results of the PCA in Fig. 2a. A brief description of the procedures is given in the Methods section. We define new variables, y = log 2 (d) , from which the covariance matrix is constructed. Diagonalization of the matrix leads to new coordinate axes. www.nature.com/scientificreports/ The first principal component (PC1) axis, responsible for 43 % of the total data variance, seems to distinguish between the ancestral and evolved states. The coordinate x 1 is the projection along PC1, that is x 1 = y · u 1 , where u 1 is the normalized vector along the PC1 axis.
The mean value of the x 1 coordinate changes from �x 1 � = 0 to �x 1 � = 21.44 . The mean radii of the ancestral and evolved clouds of samples, measured from the standard deviations along the PC1 axis, are 3.08 and 12.77, respectively.
Let us stress that the evolved state at generation 20000 may be seen as an intermediate stage in the transit between minima in Fig. 1. Indeed, the fitness keeps increasing at least until generation 50000 22 .
The Fig. 2b,c show the GE distribution functions. They are integrated distribution functions, that is count the number of genes with differential expression greater (lower) than a given value. Notice that the slope of the over-expression log-log curve for 1 < d < 2 (the Pareto exponent) is around -10, whereas the slope in the underexpression curve for 1 > d > 1/3 is around 4. At these points, there are changes in the exponents to values -2 and +2, respectively (the dotted lines).
There are only 4 genes in the extreme region d > 2 (in the Ara+1 culture), and around 20 genes in the opposite region d < 1/3 . The total number of differentially expressed genes should be contrasted with the around 30 beneficial mutations detected at generation 20000 18,19 . Up to this point, gain of fitness is achieved mainly by turning off non active metabolic processes, i.e. by silencing the responsible genes 21 .
Summarizing the section, we may say that in the experimentally observed continuous transition in the LTEE, the initial (ancestral) and the final (evolved) states are relatively close in GE space, and the GE distribution functions of both states are also close, with only around 25 genes exhibiting significant values for the differential expression, that is a fraction of around 1/200 of the total number of genes. The latter criteria will be employed to assess the continuous character of the transition in the example studied in the next section.
Changes in brain white matter and Alzheimer disease
The second studied example is the GE data obtained post-mortem from a cohort of patients with Alzheimer disease (AD) and nondemented controls (ND), whose ages are above 77 years. The data comes from the Aging, Dementia and TBI study by the Allen Institute 12,23 .
In the Allen study, samples are collected from four brain regions known to show neurodegeneration and be related to pathologies as a result of AD and Lewis body disease (as described in 23 ): temporal and parietal neocortex (TCx and PCx), hippocampus (HIP) and white matter of the forebrain (FWM).
A general PCA picture of AD and ND samples can be found in supplementary Fig. S1. It is apparent that in the neocortex and the hippocampus, the clouds of ND and AD samples practically overlap. Samples from the white matter, on the other hand, are distributed over a wider sector in GE space, and it seems to be a clear distinction between the AD and ND zones.
Thus, below we focus on FWM. There are 47 ND and 28 AD samples, coming from different patients. The number of involved genes in the study is 50281. Notice that in the RNA-seq technology 12, 13 , not only proteincoding genes are detected, but also pseudogenes, long noncoding sequences with so far unknown functions, etc. The number of genes depends on the knowledge on genes at the moment the technology is created. Figure 3a shows the results of the PCA of the FWM data. The PC1 axis, which accounts for 24.7 % of the total data variance, discriminates between the ND and AD states. The transition between both states is accompanied by a change from �x 1 � = 0 to �x 1 � = 40.97 . However, the radii of the ND and AD clouds of samples are larger than the intercenter distance, that is 80.69 and 72.64, respectively. These results suggest a continuous transition in a very broad well.
It is well known the role of age in AD, specially in the elderly 24 . Then, we may use age as a time variable to follow the transition. In spite of the relatively small number of samples, a linear regression analysis of the mean position x 1 as a function of age in ND samples, Fig. 4a, shows that �x 1 � ≈ −287.12 + 3.24 age , P-value = 0.07 . In the AD samples, however, no correlation between x 1 and age is observed. Thus, the position of the AD zone is roughly fixed, and the cloud of ND samples shows a drift towards the AD minimum as age increases.
A better illustration of this fact comes from supplementary Fig. S2 Figure 4b shows the increase with age for both ND and AD samples of the NIA Reagan index for the neuropathological diagnosis of AD 25 . This may be simply interpreted as an increase of the fraction of brain microstates trapped in the AD zone.
Let us recall the physical analogy, mentioned above. The random motion of samples in GE space can not be ascribed to mutations because it is well known that the replacement rate of neurons is very low 26 . These random displacements or variations in GE space are instead related to accumulation of damage in the DNA of brain cells 27 or to accumulation of methylation events 28,29 . Both processes are related to aging and in general lead to a decrease of tissue fitness. The roughly independence of age position of the AD zone means that this is a definite region in GE space with higher fitness, a local maximum, which holds the disease state.
The following picture of late AD progression emerges. As age increases, the fitness of brain microregions decrease and a zone of GE space representing a local maximum (the AD zone) becomes reachable. The neocortex and other brain regions are attracted earlier to this zone. The white matter, responsible for the connections and probably defining the global AD brain state, shows higher resilience. Below, we shall come back to this picture. Figure 3b,c illustrate the rearrangement in the expression levels. The distribution function exhibits a fast decay when the differential expression departs from 1. The exponents of the Pareto laws are -8 and 9, respectively. There are around 100 genes with d > 2 , and only around 10 genes with d < 1/2 . The fraction of differentially expressed www.nature.com/scientificreports/ genes is ∼ 1/500 . The relatively small number of genes exhibiting high values of the differential expressions was stressed in the Allen Institute report 23 . We interpret it as a continuous transition between two close states: the normal aged state and the AD state. We notice that this "closeness" is only at the molecular level (not at the functional one), and that the main distinction occurs precisely in white matter, in charge of communication between brain sections. Summarizing the section, we may say that the data on GE in the white matter of aged brains seems to support a picture of a continuous transition from the ND to the AD state motivated by a modification of the potential (the fitness distribution) at ages below 77.
The transition from a normal tissue to a tumor
In this section, we consider a set of human tissues. In a lifetime span, the stem cells of some of them realize around 10,000 divisions 30,31 . If the tissue is in a tumor phase, an increase of the division rate is expected 32 . Thus, with respect to the number of cell divisions (generations), the data for tumor cells are comparable to that of the LTEE with bacteria.
We analyze GE data from the TCGA 13 for the 15 tumor localizations described in Table 1. Expression levels for 60483 genes are measured. Recall the comment above on the number of genes in the RNA-seq technology. Normal and tumor samples from different patients are recorded. Thus, we should make use of the ergodic hypothesis for the analysis of the data. We stress that a set of results coming from the PCA of this data is presented in Ref. 33 . Below, we focus on the rearrangements of GE levels.
Let us consider the Kidney Clear Cell Carcinoma (KIRC) in more details. The PCA is presented in Fig. 5a. The PC1 axis, responsible for 60 % of the data variance, discriminates between normal and tumor samples. The mean value of the x 1 coordinate varies from �x 1 � = 0 to �x 1 � = 171.80 in the transition from the normal to the tumor state. The radii of these regions are 28.70 and 36.00, respectively. Thus, the data suggests that there exist two distinct minima, occupying distant regions in GE space.
Notice that the number of samples in the intermediate region is scarce. This fact could be related to the common late detection of tumors 34 . Our interpretation is different. In KIRC, there are 72 normal and 739 tumor samples, large enough numbers. According to the ergodic hypothesis, the higher density of observed samples correspond to the potential wells (higher fitness regions). The deepest well seems to be the tumor state. The intermediate region 30 < x 1 < 130 , supports a low-fitness barrier which prevents the transition from the normal to the tumor state. In particular, 30 < x 1 < 80 defines a coexistence region, where both normal and tumor In our previous paper 35 , we have quantitatively estimated the number of available microstates in each region for a set of tumors by means of an entropy-like magnitude. This number is much greater for tumors than for the normal state. Thus, the barrier in the intermediate region is needed. Otherwise, the normal microstates could be continuously driven to the tumor region.
The progression of a normal sample to a tumor state could proceed as follows. The sample starts at a point near x 1 = 0 and realizes random motions due to somatic mutations, epigenetic changes or external carcinogenic factors. However, the barrier prevent the sample from leaving the normal region. Only when a jump over the barrier occurs the sample starts moving towards the tumor region.
The idea that the x 1 coordinate indicates progression towards the tumor region is supported by a set of facts. In paper 33 , we show in KIRC that the intermediate region is populated mainly by stage I tumors. In Ref. 36 we show in PRAD (prostate cancer) that x 1 shows strong correlation to clinical indicators of progression, in particular tumor cellularity, that is the fraction of cancer cells in the sample.
In Fig. 5b,c, we show the distribution function for the differential expressions in the over-and under-expression regions. The average tumor curves exhibit exponents near − 1.4 and 0.7, respectively, and there are thousands of differentially expressed genes. These results favor the picture of a discontinuous normal to tumor tissue transition.
Two additional curves were added to these figures. They reflect the average distributions of normal and tumor samples in the intermediate coexistence region, and show how the rearrangement of expression levels occurs in the progression to tumors. The greatest differences between normal and tumor distributions become apparent in the under-expression region. Roughly speaking, these are genes related to homeostasis, which are silenced in the tumor state. This fact was already noticed in paper 33 . Genes may be ranked according to their contribution to the unitary vector along PC1, the axis labeling progression to cancer. In lungs, for example, the most relevant silenced gene is Surfactant Protein C, in kidney it is Uromodulin, etc. All these genes play an important role in their respective tissue homeostasis.
The results for the other tumor localizations, studied in the present paper, are summarized in Table 1. The mean value of the x 1 coordinate in the tumor state (for the normal state we set �x 1 � = 0 ), the radii of the normal www.nature.com/scientificreports/ and tumor zones, the Pareto exponents, and the maximal and minimal reached differential expression values are given for each tissue.
We have grouped in a final supplementary Fig. S3 the distribution functions for all of the studied tumor localizations, which shows a kind of universal behavior in cancer.
Summarizing the section, we may say that the transition from a normal tissue to a tumor seems to be a discontinuous one. The differential distribution functions show very heavy tails with thousands of differentially expressed genes, around 1/10 of the total number of genes.
Concluding remarks
We use an analogy with the motion of a particle realizing random displacements in an external potential in order to analyze the GE rearrangements in a biological system, which experiences a transition from an initial to a final state. The random motion of the particle is associated to variations in the expressions of a group of genes as a result of mutations and epigenetic events, or even damages in the DNA. The external potential is the fitness landscape.
In the LTEE, the experiment conditions induce displacement towards a new minimum, away from the initial one corresponding to the wild or ancestral genotype.
In the study concerning late onset of AD, we observe an AD zone with a definite position in GE space, and a drift of the ND clouds of samples towards the AD zone as age increases.
Both are examples of continuous transitions, motivated by a modification of the fitness landscape. This modification is well understood in the LTEE. In the AD study, on the other hand, we think that the accumulation of damages and methylation events as a result of aging is not only the reason for the random motion in GE space, but leads also to a significant reduction of fitness in the microstates. Recalling the fitness landscape in the next example, tumors, we may say that aging makes the brain microstates to move away from the normal, homeostatic zone to the low-fitness region. It seems that the AD zone is located somewhere in this region and is a kind of local maximum for the fitness, to which the ND samples are attracted.
The idea of aging as a cause for reaching the low-fitness barrier is also consistent with the increase of cancer risk with age.
The conceptualized abrupt character of the transition in cancer shows similarities with the two-stages theory (initialization-progression) 37,38 . The initialization phase is identified with the initial jump moving the microstate out of the homeostatic region. Further elaborations of this theory, i.e. Vogelstein progression in colon cancer and beyond 39,40 , indicate that there could be a sequence of steps. This is not surprising because there is a long way from the normal to the tumor regions, as shown in our calculations of distances.
We make notice that in paper 41 we demonstrate for 8 tissues and no free parameters that the observed risks of cancer are consistent with a model of large jumps in GE space.
Continuous and discontinuous transitions are reflected in different ways in the GE distribution functions. The former corresponds to slight, whereas the latter corresponds to radical rearrangements.
We quantitatively describe the geometry of minima in GE space, and the tails of the GE distribution functions.
Methods
The GE data corresponding to the studied examples is analyzed by means of the PCA technique. The details of the PC analysis may be found in paper 33 . We briefly sketch them in the present section. The dimension of matrices in the Principal Component Analysis is equal to the number of genes in the data. The geometric mean is used in order to compute the average expression of the genes, where the data is slightly distorted to avoid zeroes. To this end, we added a constant to the expression (0.0001 in the LTEE data, 0.1 in the other two examples). By applying this procedure the differential expression of not statistically significant genes is regularized to one.
We define the reference expression for each gene, e ref , by taking the mean geometric average over normal or initial state samples. Then the normalized or differential expression is defined as: d = e/e ref . The fold variation is defined in terms of the logarithm y = log 2 (d) . Besides reducing the variance, the logarithm allows treating over-and sub-expression in a symmetrical way 33 .
Deviations and variances are measured with respect to the average over normal samples: y = 0 . Then, the covariance matrix is written: where the sum runs over the samples, s, and N samples is the total number of samples (initial or normal plus final or disease). y i (s) is the fold variation of gene i in sample s.
By diagonalizing σ ij we get the axes of maximal variance: the Principal Components (PCs). They are sorted in descending order of their contribution to the variance. PC1 accounts for a high percent of the variance, as notice in Ref. 33 for the case of cancer. Therefore, we restrict our analysis for all cases to PC2 vs. PC1. maps.
To process the data and perform the diagonalization of σ we employ a Python routine that was ran in a node of a local cluster with 2 processors, 12 cores and 64 GB of RAM memory. More details can be found is section "Availability of data and materials".
Data availibility
The information about the data we used, the procedures and results are integrated in a public repository that is part of the project "Processing and Analyzing Mutations and Gene Expression Data in Different Systems": https:// github. com/ Dario ALeon Valido/ evolp. www.nature.com/scientificreports/ The data we use for bacteria 10 and Alzheimer 12 are replicated in paths ../evolp/bases_external/LTEE/Gene_ Expression/ and ../evolp/bases_external/Aging_Brain/ respectively. While in the case of cancer, in the path ../ evolp/bases_external/TCGA/ we include the data for KIRC and provide instructions for downloading the data corresponding to any of the others cases from The Cancer Genome Atlas website 13 .
To process each data set we include specific scripts for bacteria, Alzheimer and cancer in ../evolp/PCA_ecoli/, ../evolp/PCA_Alzheimer and ../evolp/PCA_cancer/ respectively. There is also an additional script located in the last of the previous directories where we collect the routines we implemented for the Principal Component Analysis method. | 5,973.8 | 2017-06-29T00:00:00.000 | [
"Biology"
] |
A decoupled, modular and scriptable architecture for tools to curate data platforms
Abstract Motivation Curation is essential for any data platform to maintain the quality of the data it provides. Today, more effective curation tools are often vital to keep up with the rapid growth of existing, maintenance-requiring databases and the amount of newly published information that needs to be surveyed. However, curation interfaces are often complex and challenging to be further developed. Therefore, opportunities for experimentation with curation workflows may be lost due to a lack of development resources or a reluctance to change sensitive production systems. Results We propose a decoupled, modular and scriptable architecture to build new curation tools on top of existing platforms. Our architecture treats the existing platform as a black box. It, therefore, only relies on its public application programming interfaces and web application instead of requiring any changes to the existing infrastructure. As a case study, we have implemented this architecture in cmd-iaso, a curation tool for the identifiers.org registry. With cmd-iaso, we also show that the proposed design’s flexibility can be utilized to streamline and enhance the curator’s workflow with the platform’s existing web interface. Availabilityand implementation The cmd-iaso curation tool is implemented in Python 3.7+ and supports Linux, macOS and Windows. Its source code and documentation are freely available from https://github.com/identifiers-org/cmd-iaso. It is also published as a Docker container at https://hub.docker.com/r/identifiersorg/cmd-iaso. Supplementary information Supplementary data are available at Bioinformatics online.
Introduction
Improving the curation process on an existing data platform is often difficult. Curation workflows might be tightly coupled to the infrastructure, which increases the cost of any change. The platform might also no longer be technically maintained, its implementation might be outdated, or its original programmers might have left. In all of these cases, a curation tool that treats the underlying platform as a black box and only interacts with its existing application programming interfaces (APIs) would allow continuous expansions of the curation process without touching the underlying data platform.
We have used the identifiers.orgregistry data platform as a case study for this application note. Identifiers.org provides stable, globally unique identifiers for hundreds of data collections, mainly in the Life Sciences domain (Juty et al., 2013). To ensure these identifiers can be translated to a working URL (Wimalaratne et al., 2018), its registry stores manually curated, high-quality metadata for all collections, which must be kept accurate and up to date. Previously, identifiers.org used the HTTP response code of regular ping requests to determine whether dataset providers were still working (Juty et al., 2013). In case of failure, a curator would then still have to investigate the type of error manually. For instance, the curator would have to come up with and test out multiple different stable identifiers at several points in time to distinguish between a planned outage or an outdated URL. This existing process was not further automated because of the cost to change the infrastructure.
We have developed the application cmd-iaso as a case study of our decoupled curation tool architecture, which does not require any changes to the existing architecture, allowing for quicker prototyping of curation workflows. We designed cmd-iaso to help with the current curation workflows in identifiers.org and any future ones. The tool is run in two stages. First, cmd-iaso runs expensive and long-running data gathering and analysis tasks in the background without supervision from a human curator. For instance, cmd-iaso can observe the data providers identifiers.org lists over multiple days. Second, cmd-iaso allows the curator to perform interactive analysis of the collected data, and guides them through an augmented user-interface of identifiers.org to highlight any problems that have been identified. Figure 1 shows an overview of how we have implemented the proposed architecture. cmd-iaso is fully decoupled from identifiers.org, treats the platform as a black box and only communicates with it through its public APIs. In particular, technical knowledge of how the underlying data platform works is not required for implementing new curation workflows. As cmd-iaso is also designed as a modular tool, it can easily be extended with new analysis and curation workflows in an agile way. For instance, new data sources can easily be integrated into a workflow without changing any part of the underlying identifiers.org platform. To further reduce the technical barrier of prototyping extensions to the tool, cmd-iaso is written in the scripting language Python. Please refer to Supplementary Materials SIII and SIV for details on the implementation of cmd-iaso's modular plugin system and an example analysis using the tool, respectively.
Implementation
The most significant feature of cmd-iaso is its interactive curation workflow, during which the curator is guided through the identified issues. The tool can be run in a text-only terminal-based mode. However, cmd-iaso can best assist curation in its browserbased mode, in which it augments the existing web application of the platform. cmd-iaso uses pyppeteer, a Python port of the browser automation library Puppeteer. Puppeteer can launch or connect to a session of the Chrome browser and take complete control over it. For instance, the library can inject new information and control existing elements on any websites.
If cmd-iaso is run in its browser-based curation mode, it injects a control interface into identifiers.org's website so that the curator can quickly jump between the problematic entries. It also automatically navigates to the corresponding page in the registry and augments it with an information overlay containing information about the issue, relevant hyperlinks and any proposed corrections. Please see Supplementary Materials SI and SII for a visualization of a typical curation session using cmd-iaso and more implementation details. It is worth emphasizing that the entire augmentation only occurs locally in the curator's browser. This augmentation is the perfect example of how our proposed decoupled tool architecture can extend and improve the curator's existing interaction with their platform.
Discussion
We have proposed a decoupled, modular and scriptable architecture for a curation tool, opening up the possibility for agile development and a diverse, easily maintainable set of plugins. As a case study, we have implemented this architecture in cmd-iaso, demonstrating the benefit of the proposed decoupled architecture. In particular, cmdiaso's flexible analysis plugin system is promising. The proposed modular architecture can even be viewed as a general and highly customizable curation toolbox, as it would simplify the integration and curation of different data platforms with various analysis methods.
So far, we have only tested the architecture in our case study of cmd-iaso. However, the proposed approach can be generalized to any data platform. Furthermore, the flexible architecture is very suitable to close collaboration between curators and developers. Specifically, we also envision a modern interpretation of the curators' role, in which they have increasing ownership of and responsibility for the tools that support their curation workflows. The proposed architecture allows curators to be better equipped for the rapidly changing needs and magnitude of data in Life Sciences today (Tang et al., 2019). | 1,607.2 | 2020-09-29T00:00:00.000 | [
"Computer Science"
] |
A Cognitive Anycast Routing Method for Delay-Tolerant Networks
: A cognitive networking approach to the anycast routing problem for delay-tolerant networking (DTN) is proposed. The method is suitable for the space–ground and other domains where communications are recurrently challenged by diverse link impairments, including long propagation delays, communication asymmetry, and lengthy disruptions. The proposed method delivers data bundles achieving low delays by avoiding, whenever possible, link congestion and long wait times for contacts to become active, and without the need of duplicating data bundles. Network gateways use a spiking neural network (SNN) to decide the optimal outbound link for each bundle. The SNN is regularly updated to reflect the expected cost of the routing decisions, which helps to fine-tune future decisions. The method is decentralized and selects both the anycast group member to be used as the sink and the path to reach that node. A series of experiments were carried out on a network testbed to evaluate the method. The results demonstrate its performance advantage over unicast routing, as anycast routing is not yet supported by the current DTN standard (Contact Graph Routing). The proposed approach yields improved performance for space applications that require as-fast-as-possible data returns.
Introduction
With anycast routing, data can be delivered to (or from) any one of equally suitable sinks (or sources) by establishing a one to any-of-many relation. When correctly used, it can help to simplify network services while improving the data delivery performance. Practical uses of anycast routing today can be found mainly for the terrestrial domain. A common application involves information retrieval in content-delivery networks (CDN), where user requests can be served from the most appropriated data center hosting a geographically replicated service [1][2][3][4]. This alternative to unicast routing helps to reduce service latency that can originate from normal or intended network congestion, e.g., as produced by a denial-of-service attack [5]. In addition, it helps improve service resiliency to partial network outages [6]. Looking beyond 5G, anycast routing can serve as an efficient data delivery mechanism for integrated local, cellular, and satellite networks to handle massive and high-frequency data collection and dissemination tasks.
Despite not being previously explored in detail, anycast routing may help to mitigate some of the challenges involved in space networks [7,8]. For instance, consider a constellation of LEO satellites [9] providing intermittent communication services for a large number of sensor nodes of a smart environment (e.g., smart ocean [10]). The data from the nodes could be forwarded to different sinks through diverse paths, which may be automatically selected by the anycast service based on the location, network congestion levels, and satellite link availability. Similarly, consider a high throughput satellite (HTS) system with site diversity [11,12], where the ground gateway sites are interconnected via high-speed links. The use of the millimeter wave spectrum (mmWave) and free-space optical (FSO) in the feeder links makes the communication channels highly susceptible to atmospheric phenomena [13]. In this case, anycast routing could help to automate the traffic distribution among multiple feeder links and mitigate the impact of adverse weather affecting one or more gateway sites. As a third example, consider a space-ground sensor network with limited contact opportunities. In this scenario, it is permissible to expect that the data collected from the space sensors could be delivered to any ground station for processing or their central collection using ground channels. These are a few examples that show the potential benefits of the integration of anycast routing and delay-tolerant networking (DTN) [14].
This work explores how to achieve (near) optimal anycast routing on a space-ground DTN. The optimality of the routing is defined with respect to a metric of interest, commonly the response time of the data bundles. The main features of the network involve regular but mostly predictable contact opportunities, time-varying channel conditions that may not be observable network-wide by all nodes, and asymmetric and large propagation delays. The Consultative Committee for Space Data Systems (CCSDS) standard on space DTNs is Schedule-Aware Bundle Routing (CCSDS, 734.3-B-1), which uses Contact Graph Routing (CGR) to determine unicast paths [15]. However, the standard does not define anycast routing capabilities. Additionally, a subset of the network nodes may be geographically dispersed and subject to connectivity losses for extended periods, and therefore network partitions are possible. This feature conveys the general lack of knowledge about networkwide conditions, such as the actual buffer occupancies and the channel bit error rates, which creates additional challenges to the identification of the optimal paths from a bundle delivery time perspective.
The proposed method builds on top of the Cognitive Space Gateway (CSG) [16] approach to unicast routing that has been proposed for space DTNs. A spiking neural network (SNN) and learning method are used to formulate the Cognitive Network Controller (CNC) [17,18], which is used as the learning element of an autonomic loop. The CNC decides the optimal anycast routing option for data bundles and the expected utility (or cost) of such routing decision helps to rebalance the weights of the SNN and to induce a possible switch to a different routing alternative in the near future. The main advantages of the proposed anycast method is that routing is both decentralized and dynamic. The CNC design is also suitable for neuromorphic hardware, and therefore involves ultra-low power requirements, e.g., when using Intel's Loihi chip [19]. Because the autonomic loop continuously maps the routing decisions and the expected costs, the CSG is able to quickly adapt the next routing decisions to better exploit the parallel resources that may be available to reach any member of the designated anycast group.
The remainder of the paper is organized as follows. The next section discusses related works. Section 3 offers details of the CSG and the proposed anycast routing method. Details of the experimental evaluation environment are given in Section 4 and the performance measurements of the proposed anycast routing method are reported in Section 5. Finally, Section 6 provides concluding remarks.
Related Works
Prior works have investigated anycast for regular networks, i.e., holding the original Internet-design assumptions. Anycasting is supported by both versions 4 and 6 of the Internet Protocol (IP) as documented in RFCs 1546 and 4291, and the Border Gateway Protocol (BGP-4, RFC 4271). It plays a vital role for DNS and CDN and services, but different issues have been noted, including the detection latency of IP anycast prefixes [20], AS-level anycast path inflation [21], and client-server mapping limitations that arise in CDN routing [2]. Several enhancements have been discussed focusing on either performance, see, e.g., [22,23], or security, e.g., to mitigate distributed denial-of-service (DDoS) attacks [5]. It has been shown that anycast routing can achieve near-optimal response times and that it can be tailored to suit the QoS requirements of DiffServ networks [24], but these techniques often require knowledge about the global network state (i.e., router loads) [25,26]. Anycast services can also be used at other layers of the protocol stack. Representative examples include methods to achieve low-latency access for edge computing services [27], to overcome the high packet loss rates of Internet-of-Things (IoT) environments [6], to optimize the selection of replicated web servers [28], to improve the reservation system for electric vehicles and reduce the wait time [29], and to reduce the energy consumption of data centers [30] and wireless sensor networks [31,32].
However, DTNs and space networks are characterized by the lack of Internet-design assumptions, and therefore the works discussed above may not be directly applicable. One of the issues is the lack of a permanent end-to-end connectivity among all nodes [33]. Currently, the literature on anycast DTN routing is sparse. A group of works has addressed the probabilistic DTN case with random contacts. Along that line of research, the anycast social distance metric (ASDM) was introduced [34]. The ASDM helps to direct routing messages in the direction of the location where most group members are expected to be present. In the time constrained anycast (TCA) method [35], the one-hop and two-hop delivery probability to the anycast group are determined from the distribution of intercontact times (ICT). Routing decisions are then made based on an exponential distribution of ICTs. An extension to CGR has also been proposed [36], which considers the history of the observed contacts.
Theoretical studies of the problem include the application of epidemic routing to DTN anycasting, which has been studied under a Markovian assumption [37]. The semantics related to the address identifier to be used in DTNs have been examined at least for the case without recurrent network partitioning [38]. Larger anycast groups increase the chances of delivering the information quickly to any of the group members. This case was studied in simulation for a sparse mobile network communicating ad hoc and using the receiverbased forwarding (RBF) approach [39]. A genetic algorithm (GA) has been applied to DTN anycast routing for deterministic networks [40], showing reduced average delays compared to the shortest path but mainly with a small number of concurrent sessions, possibly because of the slow convergence time of GAs.
Anycast Routing
The network is identified by a topology that is dynamically changing due to node mobility, scheduling decisions, and signal propagation issues. The changes mainly follow well-defined patterns, e.g., resulting from orbital mechanics, and therefore, the future contact times between nodes can be predicted. The service unit is called bundle and the use of multiple contacts and nodes may be needed to deliver each bundle to the sink. Any node can operate as a gateway routing bundles for other nodes. It is assumed that the one-hop transmissions are provided by a reliable convergence layer protocol and the use of the Licklider Transmission Protocol (LTP) is assumed. The routing process aims to deliver each bundle both efficiently and with a minimum delay to any member node of the anycast group. The former aim prevents the use of bundle duplication, and therefore any form of epidemic routing, whereas the latter requires the optimization of both the member selection and the path to reach that node. Furthermore, the routing decisions must be done autonomously without any central coordination. The gateways are not aware of the global network state. The main concepts behind the cognitive space gateway approach, which was previously introduced for unicast routing, are concisely discussed followed by the proposed anycast extension.
CSG Routing Method
Routing decisions are determined independently by each gateway with the help of an SNN architecture (i.e., the cognitive network controller-CNC [18]). After choosing the optimal outbound link for each data bundle, the SNN is modified so that the CNC can adapt to the expected outcome of such a decision (i.e., the routing cost) and mitigate the performance impact of the latest bundle transmission to the next bundles. This process is achieved by modifying some of the synapse strengths of the SNN. To illustrate the process, a brief overview of the SNN model is first provided.
Spiking Neuron Model
The CNC uses the leaky-integrate-and-fire (LIF) spiking neuron model, although other spiking neuron models are also suitable [41]. The LIF model describes the electrical properties of biological neurons as given by a resistor-capacitor (RC) circuit that is driven by a current I(t). The membrane potential is a function of time u(t) that can be determined with the expression: τ d dt u(t) = −u(t) + RI(t), where τ = RC is the selected time constant of the circuit and I(t) the total stimulus.
As in the biological case, neurons can communicate with each other via nerve impulses or spikes. One spike is emitted by a neuron when its membrane's potential reaches a given threshold θ. A neuron produces spikes at times t ( f ) : u(t ( f ) ) = θ. After emitting one spike, the membrane potential drops significantly, i.e., the neuron requires to rest for some time. The potential drops to the level u r < θ for a given time period ∆ (i.e., the refractory period). During that time, the neuron ignores the effect of new spike arrivals or any external stimulus. The SNN connections determine which neurons will be affected by the emission of spikes and by how much as given by the values of the synaptic strengths (or connection weights). From neuron j (the pre synapse) to neuron k (the post synapse) the synapse strength is denoted by w jk , with w jk ≤ 0. Spiking neurons are either of excitatory or inhibitory nature. In the former case, the spikes emitted by such neurons tend to increase the membrane's potential of the post synapse neuron. In the latter case, the effect is the opposite.
The stimulus I(t) consists of the aggregated effect of the potential gradients produced by all of the neuron's dendrites, which include the effect of any external stimulus i e (t) and the spikes arriving from neuron j. The spikes arriving at the post-synaptic neuron are modeled by the impulse train i , which occur at the firing times f , delayed by d jk . Because more than one synapse may connect any two neurons, the notation emphasizes that the effect of the k-th synapse originated at neuron j. Therefore, the resulting stimulus can be expressed as
Cognitive Network Controller
The CNC [18] defines an SNN architecture of 2(K − 1) neurons, where K is the number of outbound link decisions (also known as actions) that are available for the bundle. Each action is represented by one excitatory neuron: N 1 , . . . , N K connected as a mesh with synapses w N i ,N j in both directions for any two neurons i, j = 1, . . . K, i = j. The goal of the other neurons is to provide membrane potential regulation as the excitatory signals of the core neurons can lead to potential saturation. This is achieved by making the neurons M 1 , . . . , M K−2 inhibitory with directed synapses toward the core neurons: To make a routing decision, the SNN is started by applying identical external stimulus I re f to all neurons. The SNN activity is then observed to determine the time of emission of the neurons' second spike. The first spike is used for bootstrapping the SNN as that spike carries the effect of the current synaptic connection values to the post-synapse neurons. Once the neuron that emits the earliest second spike is detected, the routing decision is given by the link being represented by that neuron, i.e., the CNC decides to use the link a = i * , provided that N i * = argmin N i t (2) is the firing time of the f -th spike of neuron x. If multiple neurons are detected to emit their second spike at the same time, the link decision is carried out randomly among those options. Furthermore, to prevent local minima, the routing decisions are modulated by a random walk with a (small) probability of P w .
CNC Learning
After a routing decision a has been made, the CNC computes the estimated cost (or negative reward) of that choice. In this paper, the interest is in optimizing response times, and thus the routing cost C is computed in terms of the expected delivery time of the bundle (at the destination) via the selected outbound link. The average observed cost can be iteratively calculated using the expression: G ← αC + (1 − α)G, where 0 ≤ α ≤ 1 is a hyperparameter. The gradient δ = G − C is used to update the SNN weights using the scaling factor η > 0 (the learning rate): As routing decisions are made and the SNN's weights are changed, the range of weight values is observed. When the range exceeds a predetermined threshold, a weight regulation process is applied to shift and scale the weights to a selected range. This is done to prevent synapse connections from becoming too large or too small, which may impact the emission of spikes. The learning algorithm has the linear complexity O(n), where n is the number of decisions.
Anycast CSG Routing
The proposed anycast routing method defines both the semantics of the anycast group and the mechanisms needed to optimally forward the bundles to any one member of the group. More than one anycast group may concurrently work. Consider one such group A = {k}, where k is a CSG node address.
Anycast group semantics
To fulfill the former requirement, a CSG gateway needs to determine whether it is part of A, which can be accomplished by defining the destination address d either as a compounded identifier or as a reserved identifier for the anycast group. The task is facilitated by the use of string identifiers by the CSG to name network nodes. In the first case, d = A, which offers the advantage that the source can dynamically modify the group composition without needing to communicate the details to other nodes, but requires extending the header size of the bundles with an overhead that is proportional to |A|. In the second case, d = a, where a is a string that identifies the group A. The size of a can be much shorter than |A| but the method requires informing A to each k ∈ A, and so any group membership modifications are more involved than with the first method. In either case, the anycast group members need not confirm their membership to the source node. For greater flexibility, the CSG anycast method supports both methods so that the decision of which method will be used would depend on the features and the logistic constraints of specific missions.
Optimal Anycast Bundle Forwarding
Each unicast or anycast destination is assigned a separated CNC at any CSG gateway i. The reward-shaping method used by the CSG [16] to estimate the cost C for each routing decision toward destination d (i.e., the negative reward) requires three terms: where T i,j , is the average bundle communication time over the i, j link, T j,d is the communication time from gateway j to destination d, and D i,d is the expected dormant time, i.e., the total disruption time of all links along the path. The first two terms include the radiation and buffering times but not the contribution of link disruptions. It is convenient to define T i,d = T i,j + T j,d . The bundle transmission time to the neighbor j can be measured after the event either by the convergence layer or through the single-hop bundle acknowledgement that is sent by the neighbor. In either case, gateway i keeps the average observation using: The single-hop bundle acknowledgements are also used by neighbor j to share its own estimation T j,d with the preceding node. The process repeats along the path with T d,d = 0. Gateway i stores the value T j,d for the next calculation.
As with unicast destinations, the gateway determines the dormant time introduced by link disruptions by analyzing the contact plan. The contact plan is given as a list of tuples (t s , t e , n s , n e , f ), where t s and t e are the start and end times of the contact between nodes n s and n e with expected channel features f . The latter term is ignored in this discussion for simplicity. The contact plan is updated by replacing n e ← a. The feasible paths are then calculated with respect to a (i.e., considering a as a supernode) using Algorithm 3 and after defining the graph from the contact plan with the procedures indicated by Algorithms 1 and 2. Algorithms 2 and 3 are identical to the one used for unicast addresses except for the last step of Algorithm 3. The time complexity of the algorithms is quadratic in the size of the contact plan. for all (n i , n j , t s , t e ) ∈ P do 6: if t e < t then 7: continue skip this entry 8: end if 9: if t s < t then 10: end if 12: if n j ∈ A then 13: n j ← a 14: end if 15: P ← P + (n i , n j , t s , t e ) append entry 16: end for 17: return P 18: end procedure The last step of Algorithm 3 filters the routes so that the list to be returned only includes as the next hop the optimal neighbor j * (as selected by the SNN) and the learning process is done with reference to that node. Cost C is given by the lowest value τ that is returned by Algorithm 3. The cost value is then used to adjust the weights of the SNN as defined by the learning method. Additionally, a simple extension needs to be added to the normal processing of incoming bundles. When a gateway k receives a new bundle with destination address d, it needs to check for anycast group membership when d is an anycast identifier (a = k) (or k ∈ A with the other semantic). If so, it passes the bundle to the upper layer. for j ← 1, n, i = j do 8: (n i , n j , t s , t e ) ← P[j] 9: if n j == n i then 10: if t s < t e then 11: (n i , n j , t s , t e ) ← P[h] 10: for all k ∈ G[h] do 11: (n i , n j , t s , t e ) ← P[k] 12: if k ∈ C or n j ∈ N then 13: continue avoid loop 14: end if 15: if k == 1 then destination contact 16: R ← (C + k, N, τ) 17: else 18: if τ > t e then 19: continue contact ended 20: else if eta < t s then 21: τ ← t s wait for contact 22: else if τ ≤ t e then 23: W ← W + (C + k, N + n j , τ) 24: end if 25: end if 26: end for 27: end while 28: return R sorted by cost (τ) and filtered so that for all r ∈ R the next hop is j * . 29: end procedure
Evaluation Testbed
The proposed anycast routing method was evaluated experimentally for a representative set of scenarios of space, air, ground, and sea. To this end, a network testbed of 10 Linux servers was deployed (PowerEdge Dell R220, Intel Xeon CPU E3-1270 v3 running at 3.50 GHz with 16 GB RAM). A machine virtualization hypervisor (VirtualBox 6.1) was used to create the virtual machines (VM) containing the anycast gateway software. The VMs were configured with 2 CPUs and 12 MB RAM without any CPU execution cap. Moreover, to prevent any performance degradation from multitenancy or the shared use of machine resources, a single gateway was created per host with direct access (i.e., using the bridged adapter) to each of the network ports. Each server was equipped with six network interface cards (NIC) in total, which were provided by a PCI express four-port NIC in addition to the two NICs embedded on the mainboard. The network topology is depicted in Figure 1.
The edges indicate the one-way propagation delays used for scenario 2, but the topology was the same for all of the tests. The testbed was build using point-to-point connections with a nominal transmission rate of 1 Gbps (i.e., UTP cables connecting NICs). The unicast routing performance of the implementation has been experimentally compared to that of ION-DTN [42,43], which suggests the correctness of the system implementation.
A tool was developed to achieve dynamic and repeatable control of the topology links' state during each experiment run, which allows testing different network conditions. The Network Environment Emulator (NEE) allows programmatic control of the link delays, rates, and packet drops through a distributed, master-slave architecture that works on top of Linux's traffic control, where the master node calculates the intended state of the network over time and communicates the changes when they are due to the relevant clients (agents) that run on the gateway nodes. The main task of the agents is to implement the desired changes. For example, the deactivation of the link (i.e., link disruption) is achieved by applying a filter (netem qdisc) to the desired interface that drops 100% of the packets received. A contact is emulated by removing such filters. Likewise, changes in the bit error rate (BER) and transmission rates of the links can be programmatically controlled by the NEE. In addition to implementing link impairments, the NEE also works as an information source for the contact plan, which is calculated for a given future window (60 s in the tests) and communicated to the gateways. The CSG software [16], which is in Python, was extended to support anycast routing. The gateway was configured with the Licklider transmission protocol (LTP) as the convergence layer (CL) protocol. LTP is a delay-tolerant protocol that can reliably deliver data blocks over an unreliable link with disruptions and long propagation delay. The main features of LTP were implemented: delay tolerance, block segmentation, reliable block delivery via multi-round transmissions, and the use of parallel transmission sessions (configured as 32 sessions). The latter feature allows the protocol to achieve high efficiency over large delay-bandwidth links. The same CL was used for all of the tested scenarios to obtain comparable results. For performance reference, the CGR algorithm was also implemented. CGR is the standard routing method used by Schedule-Aware Bundle Routing, and so it provides an adequate reference of the performance level of the current state-of-practice of DTN. The parameters used by the CNC are shown in Table 1. Each experiment consisted of observing the bundle delivery performance of a flow addressed to the anycast set {g27, g30}, and sent from node g21. The flow included 100 kB bundles, which were sent at a designated rate. The transmission was stopped after 1000 bundles. Since a reliable convergence layer was used, the final experiment statistics were collected after all of the bundles arrived at the anycast destination. Note that the intention is not to find the steady-state results, but to characterize the average performance of a selected number of bundle transmissions that represent as a whole, for example, the transmission of a large scientific dataset split into 1000 bundles. The experiments were repeated multiple times and where possible; the 95% confidence interval is shown.
Performance Measurements
Mobile networks for space, air, ground, and sea applications involve the use of channels of different characteristics that are affected by particular environmental conditions. To cover a wide range of cases while still allowing reasonable comparisons among the different cases, the evaluation study focused on a common network topology (depicted in Figure 1). Different cases were then evaluated by applying different conditions to the channel features and operating conditions.
Scenario 1: Homogeneous Propagation Delays and Random Packet Drops
The case of links sharing identical features was evaluated first. In this scenario, the propagation delay values that are depicted by the graph's edges in Figure 1 were replaced with a constant value of 100 ms. These delays were emulated by introducing controlled buffering to the outbound links of the nodes through the Traffic Control tool (NetEm). Three of the links were configured to randomly drop packets with a small probability: 0.01 (g22-g28), 0.02 (g23-g26), and 0.03 (g25-g27). These links emulate the use of wireless channels in the topology (e.g., via medium-orbit satellites). Figure 2a depicts the average response time achieved by the proposed anycast routing method. For comparison purposes, the results obtained with CGR (unicast as the protocol does not support anycast) to either of the anycast group members are also shown. Addressing the whole traffic with CGR to g27 rather than g30 produced better results. The performance achieved by the CSG method was observed to be better than either choice sent with CGR. CSG dynamically divided the flow delivering a number of the bundles to g27 and the rest to g30 as the SNN adapted to the expected routing cost. This effect can be observed in Figure 2b, which depicts the average path of the flows and sub-flows. The average observed split between the two sub-flows is depicted in Figure 2d. It is interesting to note that under low traffic levels, CSG attempted to balance in about the same proportion the workload of both anycast group members. However, as the traffic level was increased, there was a clearer preference for the g27 node because of the larger end-to-end bandwidth to that node. The response time deviation of the resulting sub-flow from the aggregate average was observed to be very small in this experiment. The average response time of the sub-flows is depicted by the discontinuous lines in the figure. The addition of both sub-flow throughputs yielded the final CSG throughput. Lower bundle response times imply a more balanced use of the parallel paths that are available for forwarding bundles in the DTN. Both routing methods use this metric as the minimization goal. By reducing the response times, it is also possible to improve the end-to-end throughput as indicated by the results in Figure 2c compared to what can be achieved with unicast CGR to a selected member.
Scenario 2: Earth-Moon Scenario
Mobile networks of nodes in the space, air, ground, and sea domains may include links of large one-way light times. To evaluate one such case, a possible earth-lunar network was considered by modifying the propagation delays of the testbed. In this case, data need to be downloaded from a node in the lunar section to any of the two designated ground stations. Three links were configured as long-haul connections: g23-g26, g25-g27, and g22-g28, with a 1.3 s propagation delay for each communication direction. No random packet losses were introduced in this test. The propagation delays within each of the sections were selected to be in the range of 7-150 ms to represent different ground, satellite-satellite, and satellite-ground links, with satellites at MEO and LEO orbits.
The link propagation delays that were in general longer that in the previous scenario directly impacted the response times of the flow. This effect is depicted in Figure 3a, where the average response times of either of the unicast routing transmissions were up to 10 times larger than the one obtained with the anycast CGR once congestion started forming. As with the first experiment, the lower response times were achieved by exploiting longer but less congested paths, as shown in Figure 3b. The g21-g30 sub-flow that was dynamically determined by CSG was transmitted over a longer path on average than the other sub-flow. As with the first scenario, a larger throughput was observed for medium to high traffic levels, as shown in Figure 3c. The average observed traffic split between the two group members as dynamically determined by the CSG anycast routing is depicted in Figure 3d with the g27 target achieving a slightly higher preference over the g30 target.
Scenario 3: Regular Link Disruptions
Link disruptions can occur in mobile networks for different reasons in addition to regular link and node failures. It is well known that random graphs show a zero-one phase transition for network connectivity; thus, if the node density drops below the critical threshold, the chances of a network partition greatly increase [44]. Environmental factors, such as obstacles, can also cause link disruptions. Satellite links in the LEO, MEO, or HEO can experience loss of visibility for several minutes or hours and even longer periods beyond the cislunar range.
Repetitive disruption patterns were dynamically applied to two of the long-haul links of the network as follows. Link g23-g26 was kept active for 11 s and then disabled for 5 s. The active-inactive time lengths for Link g25-g27 were chosen to be 7 s and 13 s, respectively. Given that the experiments were run in real-time, the contact and disruption times were chosen to be small to achieve manageable testing times. The disruption patterns were controlled at runtime by the NEE, which, as explained earlier, runs as a distributed process parallel to the CSG.
The link disruptions reduce the end-to-end carrying capacity of the network, which the CSG attempts to dynamically discover and exploit. For the given topology and disruption pattern, this task is achieved by using longer paths, as shown in Figure 4b. The penalty of this capacity reduction can be observed through the longer average response times of the flow bundles, as depicted in Figure 4a. Unlike the previous case, the disruptions accentuated the performance differences between the two anycast group members, with unicast routing to g30 offering greater bandwidth than to g27. It is interesting to observe that the CGR actually preferred g27 in larger proportion to g30 as the delivery target as it lay along the faster path, as shown in Figure 4d. These observations also apply to the flow throughput that is depicted in Figure 4c.
Anycast Group Size
Although it would depend on the layout of the network topology and the group member that are chosen, the general expectation is that larger anycast groups would involve shorter delivery times. To verify this observation, a set of experiments was designed with different group sizes. Starting from the two-member case of the prior sections, new group members were added in the following order: g29, g26, g28, g22, g24, g23, and g25, yielding group sizes from 2 to 9. To define this sequence, the farthest node from the source was selected to join the anycast group for the next experiment. Figure 5 depicts the average response time as a function of the anycast group size. Four cases are shown, which correspond to traffic levels of 10, 20, 50, and 100 bundles. The expected response times become larger with higher traffic levels, but smaller with larger groups. A significant change can be observed in the response time after increasing the group size to 6. One of the reasons for this outcome is that all of the first five gateways in the group belong the ground section of the network and so all feasible paths must cross one of the long-haul links. The first gateway located at the space section side, where the flow source is also located, is g22, which was added to make a group of size 6.
Another reason for the large change in the response time after adding node g22 is because of the network topology, whose structure allows a significant reduction in the path length once the group size becomes at least 6 (see Figure 6). Additionally, note that the differences on the average path length tend to diminish as the anycast group becomes larger due to the reduced path diversity. The observations regarding the average flow throughput follow a similar logic with one difference (see Figure 7). For light traffic levels (i.e., 10 or 20 bundles) the throughput remained almost constant regardless of the group size. This happened because no saturation was reached with those traffic levels and thus the observed throughput was basically identical to the flow rate. With large traffic levels, congestion is created in parts of the network and the throughput is less than the flow rate. For those cases, the addition of space-section gateways to the group (size 6 onwards) allowed boosting the flow throughput significantly, thus removing all congestion at least for the case of λ = 50.
Conclusions
By taking advantage of the autonomous route decision-making capability of the cognitive space gateway method, an anycast routing mechanism for delay-tolerant networks was developed. It allows network gateways to independently determine the optimal outbound link to be used that leads to the lowest delivery time to one of the members of the anycast group. Measurements obtained with a system prototype have shown significant performance improvement over unicasting given that anycasting is not currently supported by the standard space DTN approach (i.e., the CGR algorithm). Future work is expected to further evaluate the approach using representative traffic of actual applications.
Conflicts of Interest:
The author declare no conflict of interest.
Abbreviations
The | 8,437.8 | 2021-07-30T00:00:00.000 | [
"Computer Science",
"Engineering"
] |
Time-bandwidth compression of microwave signals
We report and demonstrate a reconfigurable photonic anamorphic stretch transform to realize time-bandwidth product (TBP) compression for microwave signals. A timespectrum convolution system is employed to provide an ultra-high nonlinear dispersion up to several nanoseconds per gigahertz, which is required for processing nanosecond-long microwave signals. The group delay of the system can be engineered easily by programming a WaveShaper. Based on the proposed scheme, the TBP of a double pulse microwave signal is compressed by 1.9 times. Our proposal can provide a more efficient way to sample, digitize and store high-speed microwave signals, opening up entirely new perspectives for generation of many critical microwave signal processing modules. © 2018 Optical Society of America under the terms of the OSA Open Access Publishing Agreement OCIS codes: (070.1170) Analog optical signal processing; (120.0120) Instrumentation, measurement, and metrology; (260.2030) Dispersion; (350.4010) Microwaves. References and links 1. A. S. Bhushan, F. Coppinger, and B. Jalali, “Time-stretche danalogue-to-digital conversion,” Electron. Lett. 34(9), 839–841 (1998). 2. F. Coppinger, A. S. Bhushan, and B. Jalali, “Time magnification of electrical signals using chirped optical pulses,” Electron. Lett. 34(4), 399–400 (1998). 3. A. M. Fard, S. Gupta, and B. Jalali, “Photonic time-stretch digitizer and its extension to real-time spectroscopy and imaging,” Laser Photonics Rev. 7(2), 207–263 (2013). 4. D. R. Solli, C. Ropers, P. Koonath, and B. Jalali, “Optical rogue waves,” Nature 450(7172), 1054–1057 (2007). 5. F. Qian, Q. Song, E. Tien, S. K. Kalyoncu, and O. Boyraz, “Real-time optical imaging and tracking of micronsized particles,” Opt. Commun. 282(24), 4672–4675 (2009). 6. C. Zhang, Y. Qiu, R. Zhu, K. K. Y. Wong, and K. K. Tsia, “Serial time-encoded amplified microscopy (STEAM) based on a stabilized picosecond supercontinuum source,” Opt. Express 19(17), 15810–15816 (2011). 7. F. Xing, H. Chen, M. Chen, S. Yang, and S. Xie, “Simple approach for fast real-time line scan microscopic imaging,” Appl. Opt. 52(28), 7049–7053 (2013). 8. K. Goda, K. K. Tsia, and B. Jalali, “Amplified dispersive fourier-transform imaging for ultrafast displacement sensing and barcode reading,” Appl. Phys. Lett. 93(13), 131109 (2008). 9. Y. Deng, M. Li, N. Huang, J. Azaña, and N. Zhu, “Serial time-encoded amplified microscopy for ultrafast imaging based on multi-wavelength laser,” Chin. Sci. Bull. 59(22), 2693–2701 (2014). 10. T. T. W. Wong, A. K. S. Lau, K. K. Y. Ho, M. Y. H. Tang, J. D. F. Robles, X. Wei, A. C. S. Chan, A. H. L. Tang, E. Y. Lam, K. K. Y. Wong, G. C. F. Chan, H. C. Shum, and K. K. Tsia, “Asymmetric-detection timestretch optical microscopy (ATOM) for ultrafast high-contrast cellular imaging in flow,” Sci. Rep. 4(1), 3656 (2015). 11. G. C. Valley, “Photonic analog-to-digital converters,” Opt. Express 15(5), 1955–1982 (2007). 12. J. Stigwall and S. Galt, “Signal reconstruction by phase retrieval and optical backpropagation in phase-diverse photonic time-stretch systems,” J. Lightwave Technol. 25(10), 3017–3027 (2007). 13. W. Ng, T. D. Rockwood, G. A. Sefler, and G. C. Valley, “Demonstration of a large stretch-ratio (m=41) photonic analog-to-digital converter with 8 enob for an input signal bandwidth of 10 ghz,” IEEE Photon. Technol. Lett. 24(14), 1185–1187 (2012). Vol. 26, No. 2 | 22 Jan 2018 | OPTICS EXPRESS 990 #309067 https://doi.org/10.1364/OE.26.000990 Journal © 2018 Received 12 Oct 2017; revised 17 Dec 2017; accepted 21 Dec 2017; published 10 Jan 2018 14. Y. Han, O. Boyraz, and B. Jalali, “Ultrawide-band photonic time-stretch a/d converter employing phase diversity,” IEEE Trans. Microw.Theory 53(4), 1404–1408 (2005). 15. J. Fuster, D. Novak, A. Nirmalathas, and J. Marti, “Singlesideband modulation in photonic time-stretch analogue-todigital conversion,” Electron. Lett. 37(1), 67–68 (2001). 16. J. Chou, O. Boyraz, D. Solli, and B. Jalali, “Femtosecond real-time single-shot digitizer,” Appl. Phys. Lett. 91(16), 161105 (2007). 17. K. Goda, K. K. Tsia, and B. Jalali, “Serial time-encoded amplified imaging for real-time observation of fast dynamic phenomena,” Nature 458(7242), 1145–1149 (2009). 18. K. K. Tsia, K. Goda, D. Capewell, and B. Jalali, “Performance of serial time-encoded amplified microscope,” Opt. Express 18(10), 10016–10028 (2010). 19. M. H. Asghari and B. Jalali, “Experimental demonstration of optical real-time data compressiona),” Appl. Phys. Lett. 104(11), 111101 (2014). 20. M. H. Asghari and B. Jalali, “Anamorphic transformation and its application to time-bandwidth compression,” Appl. Opt. 52(27), 6735–6743 (2013). 21. B. Jalali, J. Chan, and M. H. Asghari, “Time-bandwidth engineering,” Optica 1(1), 23–31 (2014). 22. B. Jalali and A. Mahjoubfar, “Tailoring wideband signals with a photonic hardware accelerator,” Proc. IEEE 103(7), 1071–1086 (2015). 23. Y. Park and J. Azaña, “Ultrahigh dispersion of broadband microwave signals by incoherent photonic processing,” Opt. Express 18(14), 14752–14761 (2010). 24. A. Mahjoubfar, C. L. Chen, and B. Jalali, “Design of warped stretch transform,” Sci. Rep. 5(1), 17148 (2015). 25. B. Li and J. Azaña, “Incoherent-light temporal stretching of high-speed intensity waveforms,” Opt. Lett. 39(14),
Introduction
Data capacity has been growing explosively with the rapid development of mobile internet and sensor technology along with wireless sensing networks. As a result, the analysis, measurement, storage and transmission of such huge amounts of data have been recognized as an important and urgent issues for the development of modern information techniques. However, there are two limiting issues to acquire, compress or digitize increasingly dynamic and fast information flow: 1) the sampling rate of conventional information processing techniques is supposed to be at least two times larger than the signal bandwidth, which is referred to as Nyquist's Theorem; 2) for dynamic signals, the signal with a bandwidth lower than Nyquist frequency will be oversampled, leading to a waste of system cost.
The bandwidth and sampling rate of analog-to-digital converters is the main bottleneck in real-time acquisition and processing of fast waveforms. The photonic time-stretch technology provides a unique and practical solution to this problem and has been widely applied in ultrafast optical spectrum analysis [1-4], imaging [5-10] and broadband analog-to-digital conversion [1-3,11 -18]. Photonic time-stretch transform is a technique that temporally stretches the fast signal based on linear photonic dispersion delay lines so that the bandwidth of the signal will be compressed by a factor M [15][16][17][18][19]. However, the record time of the stretched signal increases with the same factor M (see Fig. 1). In other words, the timebandwidth product (TBP) of signal remains unchanged. Therefore, dynamic signals with bandwidth lower than the Nyquist frequency are still oversampled. To solve this problem, anamorphic stretch transform (AST), a type of warped stretch transform, capable of TBP compression has been proposed (see Fig. 1). This technique replaces linear dispersive devices with an engineered nonlinear dispersive device [19][20][21]. The problem of insufficient sampling rate is solved, and simultaneously the time-bandwidth product of the stretched signal is compressed, leading to a shorter record length for the same bandwidth. Arbitrary stretch profiles and time bandwidth product can be synthesized using nonlinear dispersion primitives and stretch basis functions [21]. Reconstruction of the warped waveform requires information about the phase which can be obtained via coherent detection or phase retrieval techniques [22].
In many applications such as laser imaging, radar imaging and high-speed photography, the signal bandwidths are typically from several GHz to several tens of gigahertz. Whereas a time-stretch transform based on optical dispersive devices usually offer dispersion values much less than 1ns/GHz. These optical dispersive lines are unable to meet the requirement of stretching nanosecond-long microwave signals. Additionally, a nonlinear optical dispersive device [19] (e.g. chirped fiber Bragg grating) is usually designed to produce a specific group delay profile with no or limited reconfigurability. A more flexible approach would be to design a programmable dispersive line which can provide tunable group delay profiles.
In this article, we propose and realize a microwave AST system with ultra-high nonlinear dispersion and reconfigurable group delay profiles. The proposed scheme is implemented based on a time-spectrum convolution (TSC) system [23], providing a microwave dispersion up to several nanoseconds per gigahertz. The group delay profiles could be engineered easily by programming the WaveShaper. Based on the proposed system, TBP of a nanosecond-long microwave signal is compressed successfully with a compression factor up to 1.9.
Principle
TBP compression for different signals requires different types of group delay [24]. For an input signal having fast variations in the central region of spectrum (as shown in Fig. 2(a) and 2(b)), a sub-linear group delay is desired. The simulation results of linear group delay (green lines) are shown in Fig. 2(c)(e)(g), while the results of sub-linear group delay (blue lines) are illustrated in Fig. 2(d)(f)(h) for comparison. The output signal after linear group delay reflects the spectrum of input signal (see Fig. 2(b) and 2(e)). If a sub-linear group delay is employed, the temporal width of output signal will be compressed (see Fig. 2(e) and 2(f)). Figure 2(g) and 2(h) illustrate the corresponding short-time Fourier transforms, in which the dash boxes mark the temporal width and bandwidth with power density 20 dB less than the peak value. Compared with conventional time stretch, the temporal duration after sub-linear group delay is compressed, while the output bandwidth remains almost unchanged. That means the recording length of the output signal is decreased without increasing the sampling rate. This makes a more efficient use of the available samples. In terms of signals having slow variations in the central region and fast variations in the wings of the spectrum, a super-linear group delay with lower dispersion at the center of the bandwidth is needed (see Fig. 3(d)). In this case, the output temporal duration remains almost constant, while the bandwidth is decreased, smaller than the linear dispersion (see Fig. 3(g) and 3(h)). This means the output signal of the AST could be sampled with lower sampling rate, without increasing the recording time. Compared to the conventional time stretch with linear dispersion, the TBP of AST profile is compressed, leading to a more efficient and easier sampling. Since most signals in practice have spectra like Fig. 2(b), the spectrum of input RF signal in our experiment is feature-dense in the central region and sparse in the wings. Thus a sub-linear group delay should be provided by the dispersive line. The spectrum of input RF signal. (c) If a linear group delay is used to stretch the input signal, (e) the output signal is a linear scaled version of the input spectrum. (d) If a sub-linear group delay is used to stretch the input signal, the high-frequency parts of the input signal will be stretched less than baseband parts, resulting in a nonlinear frequency-to-time mapping (f). The short-time Fourier transform shows the temporal duration and bandwidth of signals after the linear dispersion (g) and nonlinear dispersion (h). The temporal duration and bandwidth are marked with green dash box (for linear dispersion) and blue dash box (for AST), with power density 20 dB less than the peak value.
Conventional photonic dispersive lines based on optical fiber or fiber Bragg grating (see Fig. 4(a)) usually offer limited amount of dispersion, (e.g. the dispersion value provided by a 100-km long single mode fiber (SMF) is about 1.36 × 10 −2 ns/GHz, which is much smaller than the dispersion required for processing nanosecond-long microwave signals. Recently, dispersive lines with ultra-high dispersion amount for microwave signals (see Fig. 4(b)) have been proposed based on a TSC system [23]. It is implemented by temporal modulating an incoherent light source with chirped spectrum. After a suitable length of dispersive medium, the output intensity is proportional to the convolution between the light source spectrum and the temporal intensity of input signals. A microwave dispersion value more than 1ns/GHz can be easily achieved using this method. delay is used to stretch the input signal, the low-frequency parts of the input signal will be stretched less than peripheral parts, resulting in a nonlinear frequency-to-time mapping (f). The short-time Fourier transform shows the temporal duration and bandwidth of signals after the linear dispersion (g) and nonlinear dispersion (h). The temporal duration and bandwidth are marked with green dash box (for linear dispersion) and blue dash box (for AST), with power density 20 dB less than the peak value.
Based on the method proposed in [23], the output temporal intensity profile is given by As such, the envelope of the output intensity is equivalent to the dispersed input RF signal. The effective dispersion is given by By properly choosing a small B 0 , D mw can be much larger than D 0 . Therefore, an ultra-high linear dispersion for microwave signal can is achieved. As shown in Eq.
(2), the frequency-totime mapping of ( ) S t is given by which is a linear function. To realize a sub-linear group delay, we have to find a function to describe the nonlinear frequency-to-time mapping. Tangent function provides a simple but feasible approach for this problem. It has been widely used in the AST of optical signals [20]. So ( ) t ω can be written as where C 1 and C 2 are arbitrary real numbers. And as such, ( ) S t is given by where It{} denotes temporal integration operation. If we use the same optical dispersive line where The spectrum of the incoherent light ( ) S ω is given by Therefore, the group delays supported by Eqs.
(2) and (6) are the same in the law frequency area. While in the high frequency part, the group delay supported by Eq. (2) is larger than that supported by Eq. (6). In particular, the value of C 2 is related to the TBP compression factor, which will be discussed later.
Experiment
The experimental setup of our TBP compression system is shown in Fig. 5. The broadband incoherent light is generated by a superluminescent diode (Covega SLD-6716-11748.5.A02) followed by a semiconductor optical amplifier (SOA, Covega BOA-3876) [25]. Then the light is specially filtered by a WaveShaper (Finisar 4000S) with an engineered profile. A RF signal generated from an arbitrary waveform generation (AWG, Tektronix 7122B) is modulated on the filtered incoherent light using an intensity modulator. The optical dispersive medium employed in this system is a section of dispersion compensating fiber (DFC) with a dispersion value of −1314 ps/nm. After amplified by an EDFA, the output optical signal is detected by a 45-GHz photodiode (PD) and monitored in an oscilloscope (Tektronix CSA 8200). By programming the WaveShaper, we can realize linear dispersion or nonlinear dispersion for microwave signals. The measured and simulated spectra of the shaped incoherent light source are shown in Fig. 6. Figure 6(a) represents the shaped spectrum for linear dispersion, where the chirp rate B 0 is chosen to be -6 2 2.4 10 / ns rad × . According to Eq. (3), the effective dispersion for microwave signals D mw is calculated to be 2 1.17 / ns rad , which enables our scheme to work on GHz-bandwidth microwave signals. Figure 6(b) and 6(c) denote the spectra required for AST with 7 2 5 10 / C r a d s = × and 7 10 10 / rad s × , respectively. And another parameter C 1 is determined based on Eq. (8). The corresponding group delay profiles are shown in Fig. 7. As can be seen, the group delay profiles of AST have lower dispersion in the wings of the bandwidth, leading to the sides of the spectrum stretched less than the central part. This is desirable as the spectrum of our input signal do not have fast oscillations in the wings, so there is no need to stretch them as much as the central region. In this way, the TBP of microwave signals can be compressed. Meanwhile, group delay profiles of the AST system can be easily engineered by adjusting the WaveShaper, showing a great reconfigurability. However, a programmable optical filter doesn't mean that the proposed scheme can work for arbitrary input signals. As explained in Principle section, input signals need to have the potential to be compressed. For sub-linear group delay used in our system, input signals having slow variations in the wing of spectrum are desired. Moreover, the signal will have greater potential to be compressed if the spectrum has a larger proportion of slow variation region. Fig. 6. The measured (red) and simulated (blue)spectra of the incoherent light after a WaveShaper: (a) Linear dispersion, (b) and (c) AST with C 2 = 5 × 10 7 rad/s, and C 2 = 10 × 10 7 rad/s respectively. The input RF signal and its spectrum are shown in Fig. 8(a) and Fig. 8(b). The input RF signal is a double waveform with different pulse widths. As be seen, the input spectrum has relatively slow oscillations at the sides of bandwidth. In the linear case, i.e. the input optical power spectrum is encoded with a linearly chirped envelope profile by the WaveShaper, and the envelope of the output signal is a linear scaled version of the input RF spectrum, as shown in Fig. 9(a). The blue line shows the measured output waveform, while the green line represents the numerically calculated output envelope of the measured input signal in Fig. 8(a). As shown in Fig. 9(a), the time duration of simulation output signal is 50 ns, while the duration of measured waveform is less than 40 ns. The missing information is caused by the bandwidth limitation of our microwave dispersion system. However, in the central region, the measured signal envelope mainly agrees with the simulation one. Thus we have reason to believe that the measured waveform would have the same time duration with the simulation signal if a larger operation bandwidth is provided. Fig. 8. The waveform (a) and spectrum (b) of input RF signal. Figure 9(b) and 9(c) show the output waveforms of AST with different value of C 2 . Again, there is a good agreement between the simulation (green) and the envelope of measured waveform (blue). However, we can still find some discrepancy, which mainly results from the ASE noise and the spectral unflatness of the broadband source and amplifier. Compared with the linear case, the wings of the output waveform of AST are stretched less than the central region, which leads to a shorter time duration (36ns and 24ns respectively). In order to compare the TBPs of the linear dispersion and AST, the short-time Fourier transform (STFT) of linear dispersion (Fig. 9(e)) and AST ( Fig. 9(f) and 9(g)) are calculated. The region of the STFT with power density 20 dB less than the maximum power density is marked by a blue dash line. The green dash box shows the time duration and the bandwidth of output envelopes. As observed, the output bandwidth of AST1 (3.6 GHz) remains almost unchanged, compared to the output bandwidth of linear dispersion, while the temporal duration is squeezed (from 50 ns of linear profile to 36 ns). When a larger nonlinear factor C 2 is employed, the time duration of output envelope will be compressed further, reaching 24 ns. However, the output bandwidth is expanded slightly to 4 GHz, compared to the linear dispersion. This is because the wing part of the output signal is stretched much less than the central part under a larger C 2 . In this case, the side part of output waveform has more fast variations than the central part, creating some new high-frequency components. The TBP of output envelopes in the three cases can be easily calculated as 180, 130 and 96 respectively. As can be seen, the TBP of output envelopes in AST 1 and AST 2 are compressed compared with the linear dispersion, with compression factors of 1.4 and 1.9 respectively.
By employing nonlinear dispersion, the recording time can be reduced under the same sampling rate, offering a more efficient way to sample, digitize and store high-speed microwave signals. However, when a larger nonlinear factor is employed, the output envelope bandwidth may increase, requiring a higher sampling rate. To avoid the increasing of sampling rate, the nonlinear factor of AST need to be controlled within a reasonable range. For a given input signal, the nonlinear group delay should be designed to compress the output temporal width as much as possible, without increasing the output bandwidth. 9. (a)-(c) Measured output signal (blue line) in case of linear system and AST system with C 2 = 5 × 10 7 rad/s, and C 2 = 10 × 10 7 rad/s. Green lines are simulated output envelope based on the measured input signal. (e-g) show the STFT of the output envelopes. The region of the STFT with power density 20 dB less than the peak power density is contoured by a blue dash line. The bandwidth and time duration are marked by a green dash box. The time duration of the output envelope is defined as the temporal range up to the fifth notch.
Conclusion
We proposed and demonstrated a reconfigurable AST system for GHz-bandwidth microwave signals based on an ultra-high nonlinear dispersion line. The ultrahigh dispersion is induced by using a spectrally shaped incoherent light source followed by a section optical dispersion medium. By using AST technique, the recording time of output envelope can be reduced without increasing the sampling rate. But when the nonlinear factor of group delay increases further, the output envelope bandwidth may be larger than that of linear dispersion. Thus the nonlinear factor of AST need to be controlled within a reasonable range. The TBP of the microwave signal is compressed, with a compression factor up to 1.9 compared with the TBP of linear dispersion. The proposed scheme can provide a more efficient way to sample, digitize and store high-speed microwave signals. By programming the WaveShaper, group delay profiles of the system can be tailored flexibly, opening the path for realization various critical instruments for measurement, generation and processing of high-speed microwave signals in a very simple and practical fashion. | 5,041.4 | 2018-01-22T00:00:00.000 | [
"Physics"
] |
Low Ethanol Concentrations Promote Endothelial Progenitor Cell Capacity and Reparative Function
Background Endothelial progenitor cells (EPCs) are recruited to injured endothelium and contribute to its regeneration. There is evidence that moderate ethanol consumption prevents the development and progression of atherosclerosis in a variety of in vitro and in vivo models and increases the mobilization of progenitor cells. Furthermore, there are studies that identified ethanol at low concentration as a therapeutic tool to mobilize progenitor cells in peripheral blood. At the same time, the cell number of EPCs represents a close link to cardiovascular system constitution and function and contributes to cardiovascular risk. The aim of this study was to evaluate the effect of low dose ethanol on typical features of endothelial colony-forming cells (ECFCs), a proliferative subtype of EPCs. Methods and Results We tested whether ethanol impacts the functional abilities of ECFC (e.g., migration, tube formation, and proliferation) using in vitro assays, the intercommunication of ECFC by exploring cell surface molecules by flow cytometry, and the expression of (anti-)angiogenic molecules by ELISA. Low concentrations of ethanol concentration promoted migration, proliferation, and tubule formation of ECFC. The expression of the cell surface marker VE-cadherin, a protein which plays an important role in cell-cell interaction, was enhanced by ethanol, while (anti-)angiogenic molecule expression was not impacted. Conclusion Ethanol at moderate concentrations increases the angiogenic abilities of endothelial progenitor cells thus possibly contributing to vasoprotection.
Introduction
Cardiovascular disease (CVD) is a major cause of death worldwide. Most CVDs are based on atherosclerosis, a degenerative process of the arterial vascular endothelium induced by oxidative stress and chronic inflammatory status. The classic risk factors include smoking, diabetes mellitus, arterial hypertension, changes in total cholesterol, obesity, and excessive consumption of ethanol. Several studies have shown that strong ethanol consumption in adults correlates with the occurrence of CVD [1][2][3][4]. On the contrary, moderate ethanol consumption is associated with decreased morbidity and mortality from ischemic heart disease [5][6][7]. The protective effect of ethanol on the cardiovascular system has been attributed to the modulation of blood lipoproteins and to reduced platelet activation and thus diminished formation of thrombi. Furthermore, other studies suggest that ethanol has a direct protective effect on the myocardium [5,8,9].
Tissue regeneration is the focus of therapy in the treatment of CVDs. Proper vascular formation is essential in this context. Endothelial progenitor cells (EPCs) and especially their proliferative subtype endothelial colony-forming cells (ECFCs) are of pivotal importance for endothelial homeostasis and vascular remodeling. In addition, they represent a promising cell source for the revascularization of damaged tissue. The concept that alterations in EPC biology impact endothelial cell function is supported by recent studies showing that decreased numbers of circulating EPCs correlate with impaired endothelial function [10,11]. Both a reduced number and impaired function of circulating EPCs have been observed in patients with CVD [12]. The number of circulating EPCs predicts the occurrence of cardiovascular events and can help identify patients at increased risk [13]. ECFC can be isolated from cord blood or peripheral adult blood and migrate to sites of vessel formation, possessing the ability to differentiate into mature endothelial cells, to participate in vessel repair, and to form de novo endothelium [14].
Angiogenic proteins and vascular cell adhesion proteins are important determinants of vascular health. Vascular endothelial growth factor (VEGF) has been demonstrated to promote atherosclerotic plaque progression [15]. Moreover, VEGF and placental growth factor (PIGF) stimulate endothelial cell proliferation and migration and mediate vascular growth and angiogenesis. VEGF and its soluble receptor soluble fms-like tyrosine kinase-1 (sFlt-1) are implicated in vascular damage and repair in CVD [16]. Vascular cell adhesion protein 1 (VCAM), platelet and endothelial cell adhesion molecule 1 (PECAM1), and vascular endothelial cadherin (VE-cadherin) play an important role in leukocyte extravasation, inflammatory processes, and vascular permeability and are mainly involved in the homing of cells to sites of endothelial repair and angiogenesis [17].
While in cohort studies moderate ethanol exposure is associated with reduced cardiovascular morbidity [18], little is known about the underlying pathomechanisms, specifically with regard to EPC biology. Therefore, in in vitro models, we investigated whether ethanol contributes to a poor EPC response or rather has a protective cardiovascular effect.
Materials and Methods
2.1. ECFC Isolation and Culture. ECFCs from cord blood were isolated as previously described [19]. Briefly, umbilical cord venous blood was collected immediately after delivery into sterile EDTA-coated tubes. Blood samples were centrifuged within 3 h of collection at 2,000 g for 5 min. Mononuclear cells (MCs) were isolated by density gradient centrifugation. The plasma was removed for collection and replaced with the same volume of plasma replacement buffer consisting of phosphate-buffered saline solution (PBS) supplemented with 0.025 M EDTA (Sigma-Aldrich, Steinheim, Germany or St. Louis, MO) and 1% (v/v) penicillin/streptomycin (Sigma-Aldrich). The sample volume was doubled by adding isolation buffer (PBS, 2% (v/v), fetal bovine serum (FBS, Biochrom KG, Berlin, Germany or Life Technologies, Carlsbad, CA), and 1% penicillin/streptomycin), and the sample was gently mixed. Samples were layered on Ficoll Plus (GE Healthcare, Buckinghamshire, England or Piscataway, NJ) and spun at 400 g for 40 min in a swinging bucket centrifuge with a brake in the off position. The MC fraction was collected and washed two times with an isolation buffer. Mononuclear cells were cultured in endothelial growth medium 2 (EGM-2, Lonza, Basel, Switzerland or Walkersville, MD), supplemented with supplier-recommended concentrations of human recombinant epidermal growth factor, VEGF, ascorbic acid, hydrocortisone, heparin, and recombinant insulin-like growth factor, 10% FBS and 1% penicillin/streptomycin. The MCs were plated at a density of 5 × 10 7 cells/well on collagen-coated 6-well plates (BD Biosciences, Heidelberg, Germany, or Billerica, MA) and incubated at 37°C, 5% CO 2 . The medium was changed daily for 10 days and then every second day. The first appearance of ECFC colonies was noted as well-circumscribed monolayers of >50 rapidly proliferating, cobblestone-appearing cells. Colonies were identified by visual inspection using an inverted microscope (Olympus, Tokyo, Japan; Zeiss, Thornwood, NY). Well-defined colonies were released from the plates using cloning rings and trypsin-EDTA and collected. The cells from each separate colony were placed into a well of a collagen-coated 6-well plate and after becoming 80-90% confluent, subsequently passaged into collagencoated T25 culture flasks. After reaching 80-90% confluence, the cells in the T25 flasks were passaged into gelatin-coated T75 flasks. At 80-90% confluence, these cells were harvested, phenotyped, and frozen in a freezing medium containing 92% FCS and 8% DMSO (Sigma-Aldrich, Steinheim, Germany).
Flow cytometric analyses to confirm the ECFC phenotype were performed using surface markers CD31, CD34, CD133, VEGFR-2, and CD45 as well as appropriate isotype controls.
All experiments were run with ECFC in passage 3 to 5 at 80-90% confluence. The concentrations of ethanol used in this study (0.5% (17 mM) and 1% (34 mM)) are doses that do not cause intoxication in vivo and correspond to 2 to 4 standard drinks.
2.2. In Vitro Angiogenesis Assay. We used an in vitro angiogenesis assay (endothelial tubule formation in Matrigel) to test the capacity of ECFC to form capillary tubule-like networks. In a 96-well plate, 17,000 cells per well were seeded in 150 μl treatment medium with 30 μl growth factor reduced in Matrigel (BD Biosciences, Bedford, MA). ECFCs were either treated with 0.5% and 1% ethanol for the duration of the assay or pretreated for 24 h with 0.5% and 1% ethanol. A corresponding control w/o ethanol was run in tandem. After 16 h of incubation at 37°C and 5% CO 2 supply, each well was photographed with a LEICA DMI 6000 B microscope. Total tubule length in each visual field was measured using the ImageJ software 1.52q (National Institutes of Health). All experiments were performed in triplicate wells from which values were averaged (n = number of experiments).
Cell Migration Assay.
To analyze ECFC migratory ability, 50,000 ECFCs were seeded in each well of a 12-well plate with a growth medium containing 2.44% supplements, 5% FCS, and 1.2% penicillin/streptomycin. After reaching confluence, the ECFC monolayer was scratched using a sterile P200 pipette tip to produce a lane free of cells as described before [20]. ECFCs treated with 0.5% and 1% ethanol or after 24 h of preincubation with 0.5% or 1% ethanol and a corresponding control w/o ethanol were run in tandem. Light microscopic images were obtained immediately after the scratch (T0) and at the end of the experiment after 18 h (T18). Migration into the scratch wound was analyzed using the ImageJ software and calculated as the percentage of wound closure (percentage of original area at T0 that became occupied by cells by migration into the wound area at T18). All experiments were done in quadruplicate wells from which values were averaged.
2.4. Cell Impedance Assay. Cell impedance was determined by real-time impedance analysis using the xCelligence Realtime analyzer (RTCA, Roche, Mannheim, Germany). The Cell Index (CI), which reflects cell adherence, is converted from impedance measurement by the xCelligence software (Version 1.2.1) and was continuously monitored every 20 min for at least 72 h and directly after specific treatments of the cells. For experiments done, 10,000 cells were seeded in triplicates onto gold-coated E-Plate VIEW 96-well plates (Roche, Mannheim, Germany) and proliferation was calculated through measuring increasing impedance. ECFCs treated with 0.5% and 1% ethanol or after 24 h of preincubation with 0.5% or 1% ethanol and a corresponding control w/o ethanol were run in tandem.
Cell Proliferation Assay.
To determine the proliferative capacity of ECFCs after treatment with ethanol, 50,000 cells were seeded per well of 6-well culture plates in EGM supplemented with 5% (v/v) FBS and 1% penicillin/streptomycin. After 24 h, 48 h, and 72 h of treatment, the cell number was counted in a Neubauer chamber with 1 : 2 trypan-blue dilution. Population doubling time was calculated as the following: log 2/ðlog ðNt/NoÞ/tÞ, t = time period ðhÞ, Nt = number of cells at time t, and No = initial cell number. ECFCs after treatment with 0.5% EtOH 24 h and 1.0% EtOH and a corresponding untreated control were run in tandem.
2.6. Flow Cytometry. Flow cytometry analysis was performed to detect adhesion molecule expression on the ECFC surface. ECFCs treated with 1% ethanol and a corresponding control w/o ethanol were run in tandem. Cells were harvested by incubating with Accutase (Capricorn, Ebsdorfergrund, Germany) for 10-15 min at room temperature. After washing with flow cytometry buffer (PBS, 2% BSA (Merck, Darmstadt, Germany)), 1 × 10 5 cells were blocked with normal IgG (Grifols, Paris, France) (5 mg/ml) for 1 min, followed by incubation with the appropriate antibodies (VCAM1 APC (BioLegend, San Diego, CA), PECAM1 FITC (BioLegend, San Diego, CA), and VE-cadherin PE (BioLegend, San Diego, CA)) and isotype controls at 4°C for 30 minutes. For each experiment, a positive control for apoptotic cells was included to exclude dead cells from analysis. Apoptosis was induced by UV-irradiation with a transilluminator (Biostep, Jahnsdorf, Germany) for 30 min, and cells were further incubated for 2 h at room temperature. Cells were harvested, washed, and resuspended in Annexin V binding buffer, stained with Annexin V FITC (BioLegend, San Diego, CA) and incubated for 20 min. After washing and prior to measurement, propidium iodide (10 μg/ml) (Sigma-Aldrich, Darmstadt, Germany) was added. Flow cytometry measurements were performed on a BD FACSCalibur Flow cytometer (Becton Dickinson, Heidelberg, Germany), and results were analyzed using the FlowJo X Software V10 (Tree Star, Ashland, OR).
2.8. Immunoblot for Quantification of VEGF, sFlt-1, and VE-Cadherin. For analysis of proteins VEGF, sFlt-1, and VE-cadherin, ECFCs were grown to 50% confluence in 100 mm dishes (Sarstedt, Nümbrecht, Germany) in an endothelial growth medium with 2.5% FCS with the respective additives. ECFCs were seeded into 100 mm dishes and treated with either only EGM (C = control) or EGM plus 1% ethanol. Cells were treated with 6 ml medium and cultivated for 24 h at 37°C, followed by harvesting of the cells, cell lysis, and protein quantification.
Discussion
In this study, we demonstrate a promoting effect of moderate ethanol concentrations on angiogenic capacities of ECFC.
In this context, we tested the ability of migration, proliferation, and tubule formation, which were enhanced after incubation with moderate concentrations of ethanol. These functional Figure 2: Effect of ethanol on ECFC migration. ECFCs were cultured in endothelial basal medium (EBM) +5% FBS and in the absence or presence of 0.5% or 1% ethanol directly prior assay start or 24 h before. The migration of ECFC into the scratch wound was assessed after incubation for 8 h. Results of at least 6 independent experiments represent mean ± SD percent wound filling. * P < 0:05 vs. untreated control.
Cardiovascular Therapeutics
properties represent cell characteristics important for angiogenesis and vasculogenesis, and they are markers of vascular health. Ethanol-treated cells showed a higher expression level of the cell surface marker VE-cadherin, a protein which plays a significant role in cell-cell interaction. Interestingly, the positive effect of ethanol seems not to be triggered by VEGF, PlGF, sEng, or sFlt-1. Additionally, expression levels of their mRNAs, protein, or soluble protein expression were not different after ethanol treatment compared to untreated control.
Our findings confirm data of previous studies in which a stimulating effect of ethanol on endothelial function is demonstrated [26,27]. To our knowledge, however, this is the first study to demonstrate enhanced functional properties of ECFC treated with ethanol. ECFCs are an endothelial cell type with a strong intrinsic clonal proliferation potential and the ability to contribute to de novo blood vessel formation in vitro and in vivo [21,22]. The ECFC recruited into the damaged tissue was derived either from the circulation or from the local vascular wall [23]. The endothelial integrity of the vascular wall is restored by migration and proliferation of ECFC [24,25]. Taken together, based on the current accumulated evidence, ECFCs are the most rational and promising cell sources that are able to incorporate into or form vessels directly in areas of tissue regeneration.
There are some epidemiological studies that demonstrate the benefits of moderate ethanol consumption in ischemic heart disease [5][6][7]. Interestingly, the results of these works suggest that in addition to lowering the rate of atherosclerosis through regular exposure to ethanol over many years, at least part of the ethanol-induced protection of the vascular endothelium can be induced within a few minutes by a short exposure. This protection can be provided by moderate ethanol doses, which correspond to about one or two alcoholic beverages. At this juncture, the concentrations of ethanol used in this study, i.e., 0.5% (17 mM) and 1% (34 mM), are levels that do not cause intoxication in vivo and correspond to 2 to 3 standard drinks per day [28]. Chen et al. showed that the treatment of cardiac muscle cells with 10-50 mM ethanol protects against ischemia by activating protein kinase C [26]. The ability of ethanol to activate protein kinase C has also been observed in several other cell systems [29,30] and has been reported to mediate the protection of the heart from ischemia. In vivo studies demonstrated cardiovascular protection after prolonged ethanol administration of 25-50 mM also in a guinea pig model.
Restoring blood flow after tissue injury or occlusion requires angiogenic germination of endothelial cells from 6 Cardiovascular Therapeutics nearby intact blood vessels, as well as vasculogenesis through circulating ECFC to allow invasion of new blood vessels to restore tissue perfusion. A short-term treatment with ethanol to increase the capacity of ECFC could be a possible therapeutic approach to enable the blood flow to be restored more quickly.
The exact reasons why ethanol can protect the cardiac tissue and endothelium have not yet been fully clarified. It is certain that ethanol can change the biophysical and biochemical properties of cell membranes. Ethanol interacts directly with membrane proteins to modulate their activity. In different cell models, exposure of ethanol impacted both the signal transduction mediated by protein kinase C [31,32] and the cAMP-dependent protein kinase [30,33].
In our study, we show an increased expression of the membrane surface molecule VE-cadherin, which is significant for cell-cell interaction. Processes, which are mediated by adhesion molecules, are activation of cell mobilization, migration, and proliferation via loosening of endothelial cell adhesion complexes. These mechanisms are impaired in CVD [17,34]. Widner et al. showed that the expression of VE-cadherin is reduced in plaque microvessels leading to vascular damage [35]. The observed increase of VEcadherin in our study might contribute to an increased repair capacity of the endothelium.
With regard to the cardioprotective effect, half a million adults and 15000 children undergo open-heart surgery annually in the United States each year in which the heart is exposed to controlled periods of ischemia. Despite advances in heart protection, myocardial dysfunction remains a major cause of morbidity and mortality in the immediate postoperative period. The exact time of the expected ischemia is known in this context. Therefore, the targeted use of a cardioprotective agent can prevent the consequences of ischemia in the endothelium and the remaining heart tissue. One could speculate that moderate doses of ethanol could be cardioprotective and this approach could be evaluated further.
ECFCs seem to be a promising cell source for revasculogenesis in damaged tissue in CVDs. They have excellent vascularization potential in vitro and in vivo, and in this study, we show that their potential is further increased by adding small amounts of ethanol. ECFC can be isolated in sufficient 7 Cardiovascular Therapeutics numbers from umbilical cord blood for possible use in newborns but if stored long-term could be used as an autologous source of reparative cells for the treatment of CVDs in adulthood. In older children and adults, primarily circulating peripheral blood ECFCs are less clonogenic, proliferative, and angiogenic than umbilical cord ECFC. Finally, it is known that certain disease states such as diabetes can reduce the frequency and function of the isolated ECFC to such an extent that these cells are not of sufficient quality for use as revascularization therapy [36,37]. Here, short-term therapy with ethanol could possibly increase the capacity and proliferation power of the ECFC in order to achieve a higher quantity and quality of these cells.
In conclusion, ethanol at moderate concentrations enhances the angiogenic abilities of ECFC. Ethanol-exposed cells showed a higher expression level of VE-cadherin, suggesting an increased endothelial repair capacity. Our findings support data from epidemiological studies demonstrating the benefits of short exposure of ethanol to prevent cardiovascular damage in adults [5][6][7]. However, the data provided here are limited due to the use of primary cells under in vitro conditions and show the short-term effect of ethanol on endothelial functionality. Further studies on the impact of long-term moderate ethanol exposure on ECFC biology are still needed.
Data Availability
The experimental data and row data used to support the findings of this study are available from the corresponding author upon request.
Conflicts of Interest
No conflicts of interest, financial, or otherwise are declared by the authors. | 4,379.2 | 2020-09-22T00:00:00.000 | [
"Biology",
"Medicine"
] |
Analysis and Design of a 2-40.5 GHz Low Noise Amplifier With Multiple Bandwidth Expansion Techniques
This paper analyzes the main factors limiting the bandwidth expansion of low-noise amplifiers (LNA) and designs a broadband LNA with a bandwidth of 2-40.5 GHz. The LNA is designed using multiple bandwidth expansion methods, including cascode, resistance feedback, and cascode Darlington amplifier. The amplitude-frequency characteristics and bandwidth expansion principle of the three structures are studied theoretically based on the small-signal equivalent circuit model. Thanks to these techniques, a three-stage LNA is designed in a 0.15- $\mu \text{m}$ GaAs pseudomorphic high-electron-mobility (pHEMT) process. The measured results show that the designed LNA achieves an average gain of 21.6 dB in 2-40.5 GHz while maintaining a noise figure (NF) below 3.6 dB. The measured output 1-dB gain compression point (OP1 dB) is from 4.5 to 12.8 dBm and the input/output return loss are better than 5 dB. The chip area is only 1.57 mm2, including input and output test pads.
end of ultra-wideband wireless communication systems, it is important to promote the design of high-performance wideband low-noise amplifiers (LNA).
Academically, many high-performance broadband lownoise amplifiers are reported. In [5], a cascode distributed amplifier (DA) is designed to achieve a 0.1-45 GHz low noise amplifier. However, the loss of the artificial transmission line is inevitable, making it difficult for DA to achieve a high gain. And DA is not suitable in many cases because of low unity-gain bandwidth, high noise figure (NF), and large area. In [6], a transformer feedback technique is proposed to compensate for transistor gain roll-off with frequency, and a broadband LNA is finally realized for 18-43 GHz applications. A frequency-dependent feedback structure is employed to achieve a 0.1-23 GHz LNA in [7]. Nevertheless, the thermal noise of the resistors introduced in this configuration deteriorates the NF. An alternative seven-octave bandwidth LNA using combination techniques of feedback and inductive peaking techniques is VOLUME 11, 2023 This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/ proposed in [8]. However, its bandwidth is only extended to 20 GHz.
In this paper, we analyze the effect of the parasitic capacitance of transistors on bandwidth detailly and propose a three-stage broadband LNA. In the designed LNA, different amplification structures are used for the three-stage amplifier, the amplitude-frequency characteristics of the three structures are analyzed separately, and the rationality of the theoretical derivation is verified by experimental simulation. Finally, the designed LNA achieves a bandwidth range of 2-40. 5 GHz.
The remainder of this paper is organized as follows: Section II investigates the effect of the three parasitic capacitances of the transistor on the bandwidth of the transistor; Section III analyzes the amplitude-frequency properties and bandwidth expansion principles of the three proposed amplification structures. Section IV presents the simulation and measurement results; Section V summarizes the conclusions drawn from this study.
II. ANALYSIS OF THE FACTORS AFFECTING AMPLIFIER BANDWIDTH EXPANSION
The gain roll-off characteristic [9] of transistors is an important aspect that influences the bandwidth of the amplifier. It means that the maximum available gain (MAG) of the transistor does not remain the same as the frequency increases, but will show a gradual decrease in an approximately inverse relationship, as shown in Fig. 1. The gain roll-off of the transistor causes an uneven gain in broadband, which is unacceptable in the design of a broadband LNA. To further explore the factors affecting the bandwidth expansion of the amplifier, this paper studies the effects of gate-drain parasitic capacitance C gd , source-drain parasitic capacitance C ds and gate-source parasitic capacitance C gs on the basis of the small-signal equivalent circuit model of the common-source amplifier. The equivalent small-signal circuit model of the common-source amplifier is shown in Fig. 2. The gate-drain parasitic capacitance C gd is the main contributor causing the gain roll-off [10]. The expression for the characteristic impedance of the parasitic capacitance C gd is: where ω is the angular frequency. As can be seen from (1), Z gd will gradually decrease as the frequency increases. Therefore, the presence of C gd will provide a path from the input to the output for high-frequency signals, leading to the signals leaking directly to the output without amplification. In addition, the C gd also produces a feedback loop from the output to the input, which severely affects the reverse isolation [11], and the isolation continues to deteriorate with increasing frequency. The impact of isolation on circuit design is primarily in impedance matching, where poor reverse isolation can lead to interactions when performing input impedance matching and output impedance matching [12]. To further investigate the effect of C gd on the gain of the amplifier, we decompose C gd into the input and output terminations with Miller's theorem [13], as shown in Fig. 3. The presence of m 1 C gd will affect the drain current, which can be expressed as: where i s is the input current and m 1 is the Miller multiplication factor. As can be seen from (2), the drain current is smaller at higher frequencies, and can seriously affect the gain of the amplifier. In addition, the presence of m 2 C gd will affect the gain of the amplifier, and the gain can be expressed as: where R ds is the drain-source parasitic resistance, m 2 is the Miller multiplication factor and R L is the load impedance. It can be seen from (3) that the gain of the common-source amplifier will continue to decrease as the frequency increases due to the presence of C gd and C ds . Based on the above analysis, there are two ways to solved the gain roll-off problem: One is to improve the gain flatness in the broadband range by losing a part of the gain in the low frequency band, the other one is to improve the gain flatness in the broadband range by increasing the high-frequency gain.
III. ANALYSIS AND DESIGN OF THE PROPOSED LNA
The simplified schematic of the proposed wideband LNA is shown in Fig. 4. The first stage is a common-source and common-gate amplifier with a resistive feedback network (CSGF). The second stage uses a common-source amplifier with a resistive feedback network (CSF). And the third stage introduces a common-source and common-gate Darlington amplifier (CSGD). The analysis and derivation procedure are shown in sections A, B, and C.
A. ANALYSIS OF THE CSGF
For CSGF, we can intuitively see that it is constructed by connecting a common-gate amplifier to a conventional common-source amplifier. This can significantly improve the isolation between the input port and the output port [11], [14], and then the gain roll-off characteristic of the amplifier is improved effectively. In addition, the low-frequency power gain can be tuned by the resistance feedback [15].
To facilitate the analysis of the CSGF, we consider it as a common-source amplifier and a common-gate amplifier. As shown in stage I in Fig. 4, the two structures of the amplifier are combined through point A, and their smallsignal equivalent circuits are illustrated in Fig. 5.
The small-signal equivalent circuit of the common-source amplifier is illustrated in Fig. 5(b), where R f 1 in the feedback network is decomposed into input and output terminations with Miller's theorem. Assuming that the input and output ports are well-matched. The power gain can be expressed as: where g m1 , C ds1, and R ds1 are the transconductance, drainsource parasitic capacitance, and drain-source parasitic resistance. m 1 is the Miller multiplication factor. Once the transistor size and bias voltage are determined, the parasitic capacitance and parasitic resistance of the circuit will be determined, and the frequency response of the circuit will mainly depend on the resistance R f 1 in the feedback network. In Equation (4), the power gain will decrease with the decrease of R f 1 . In order to verify (4), we simulate the MAG changes of the first-stage amplifier under different R f 1 values, as shown in Fig. 5. In the CSGF, the transistor sizes of M1 and M2 are 2 × 100 µm and 2 × 90 µm, respectively, where the value of C g1 is kept constant at 0.12 pF. The studied band is set in the range of 0-50 GHz, which fully covers the operating band of the designed LNA. The simulation results are shown in Fig. 6. It is shown that the resistance value of R f 1 has a significant impact on the MAG below 25 GHz. The MAG at 10 GHz decreases from 12.3 dB to 8.1 dB as the R f 1 changes from 250 to 100 , and then the gain flatness over a wide band range can be tuned by varying the value of R f 1 . In addition, for cascaded multistage amplifiers, the noise figure (NF) is formulated [16] as: where F n and G n are the NF and the power gain of the n th stage amplifier, respectively. From (5), it can be seen that the NF of the multi-stage amplifier depends mainly on the first stage. In this design, the thermal noise introduced by the feedback network in the first-stage amplifier causes the deterioration of the noise factor, which needs to be taken care of in the design. Therefore, we simulated the effect of the resistor values in the feedback network on the NF min of the first-stage amplifier, and the simulation results are shown in Fig. 7. The simulation results show that the smaller the resistance value of R f 1 , the larger the NF min . In summary, it is almost impossible to obtain the minimum NF and the best gain flatness at the same time, so the value of R f needs to be chosen reasonably to achieve the optimal design. The CSGF is used as the input stage of the designed three-stage amplifier, and it is necessary to achieve a good input impedance matching in the broadband range. Therefore, we analyzed the role of the source feedback inductor for input impedance matching [17], [18], and the input impedance of the common-source amplifier can be expressed as: From (6), it can be concluded that the source inductance can offset the imaginary part of the impedance introduced by C gs to a certain extent, making it much easier to achieve a good impedance matching in the broadband range. The effect of different source inductance values on the input impedance matching is simulated and the results are shown in Fig. 8. With the increasing value of source inductance, the input impedance will gradually approach to the impedance matching point, which is conducive to the input impedance matching in the broadband range, and the presence of Ls does not cause a deterioration of the noise factor. [8], [19]. The small-signal equivalent model of the common-gate amplifier is shown in Fig. 5(a), and the voltages [20] at nodes B and C can be expressed as: 13504 VOLUME 11, 2023 As indicated by (9), V gs2 can be changed by adjusting the value of the external capacitor C g1 , and thus the gain of the common-gate amplifier can be adjusted. Specifically, the relationship between V gs2 and C g1 is positively correlated. Fig. 9 presents the simulated MAG at different C g1 for the CSGF. The simulated results show that the gain improves significantly after 2.5 GHz for C g1 = 0.12 pF relative to C g1 = 0.04 pF. The gain roll-off characteristics of the amplifier can be improved by increasing the capacitance of C g1 . The theoretical analysis and simulation verification in Section III-A indicate that the CSGF can improve the gain roll-off and input impedance matching of the amplifier and realize bandwidth extension by optimizing C g1 , R f 1, and L s1 .
B. ANALYSIS OF THE CSF
The frequency response of the CSF is mainly determined by the feedback network in the circuit, where the effect of the resistance R f 2 on the amplitude and frequency characteristics of the amplifier has been described detailly in Section III-A, so we will not repeat it in this subsection. We focus on the effect of the inductor L FB2 in the feedback network on the frequency response [7] of the circuit in Section III-B and further indicated the advantages of CSF in the broadband LNA design. The equivalent small-signal circuit model of CSF is shown in Fig. 10. We decompose the feedback network and the gate-drain parasitic capacitance into the input and output terminals with Miller's theorem to simplify the analysis. Then the gain of CSF can be expressed as: In the feedback network, the effect of C f 2 on the frequency response is negligible. Here, the main role of C f 2 is to isolate the DC bias of the gate and drain. Therefore, C f 2 in (11) is neglected. The gain can be further formulated as: As demonstrated by (11), Z FB increases with L FB2 , and the increase of Z FB will successively enlarge A v as indicated by (12). Intuitively, the effect of L FB2 on the gain is easily understood. The presence of the inductor inhibits the feedback of the high-frequency signals, and the curtailment effect becomes more significant as the inductor value increases. To verify the above analysis, the MAG of the CSF is simulated for different L FB2 , and the results are shown in Fig. 11, where M3 is 2 × 100 µm, R f 2 = 150 , and C f 2 = 1.5 pF. As demonstrated by Fig. 11, when L FB2 increases from 200 pH to 800 pH, the maximum value of MAG will change from 10.2 dB to 11.3 dB. In addition, the frequency point corresponding to the maximum value of MAG gradually shifts to lower frequencies. In Fig. 11, this frequency point gradually decreases from 29.9 GHz to 24.9 GHz. Based on this feature, it is very convenient to compensate the gain at certain frequencies and thus adjust the gain flatness of the multi-stage amplifier.
C. ANALYSIS OF THE CSGD
To simplify the analysis of the CSGD, we use the same study method as in Section III-A. The CSGD is decomposed into two components consisting of a Darlington amplifier and a common-gate amplifier, and it's small-signal equivalent circuit is illustrated in Fig. 12. The characteristics of the common-gate amplifier have been analyzed in detail in Section III-A, so in this subsection we will focus on the characteristics of the Darlington amplifier. For a single transistor, the expression for the characteristic frequency is F T ≈ g m 2πC gs (13) where g m1 and C gs are the transconductance and gatesource parasitic capacitance. C gs1 and C gs2 in the Darlington amplifier are connected in series, making the gate-source parasitic capacitance of the Darlington amplifier approximately equal to half of the original one. According to (13), the characteristic frequency of the Darlington amplifier is also approximately equal to twice that of a single transistor without considering the g m variation [21]. (In fact, the g m is influenced by the frequency and the operating state of the transistor.) In addition, the bandwidth expansion principle of the Darlington amplifier can be understood by analyzing the equivalent transconductance of the amplifier. Assuming that the drain-source parasitic resistance of the amplifier is neglected, the equivalent transconductance of the Darlington amplifier can be expressed as [22]: where g m1 and g m2 are the transconductance of M4 and M6. C gs2 is the gate-source capacitance of the transistor. As the g m1 , g m2 and C gs2 are determined by the transistor's size and bias voltage, and then the enhancement of the equivalent transconductance G m can be achieved by adjusting the size and bias voltage conditions of the M4 and M6 transistors.
Then, the gain expression of the Darlington amplifier can be further expressed as: According to (15), the gain of the Darlington amplifier is closely related to the parasitic parameters of M6 [23], [24]. Therefore, we simulated the effect of different sizes of M6 on MAG, and the simulation results are shown in Fig. 13. Based on the simulation results we can conclude that the gain roll-off in the frequency range of 10-50 GHz can be significantly improved when M6 = 2 × 40 µm compared to M6 = 0 × 0 µm (It is a common-source amplifier in this state). When the size of M6 is increased to 2 × 50 µm, it will further improve the gain roll-off in 10-35GHz, but it will cause more drastic gain roll-off in the frequency band range greater than 35 GHz. Therefore, it is necessary to choose the size of M6 to achieve a compromise between gain and gain flatness when designing the amplifier.
D. CIRCUIT DESIGN
To validate the proposed techniques, a three-stage LNA is designed and implemented by 0.15-µm GaAs pHEMT process. Transistors M1-M3 have equal size of 2 × 100 µm, and transistors M4, M5 and M6 have sizes of 2 × 75 µm, 2 × 100 µm and 2 × 40 µm, respectively. The drain to source (V D ) and gate to source (V G ) voltages of M1-M6 are set to 3 V and −0.5 V with a total current of 46 mA. The V G of all transistors are supplied through by a large resistance of 5980 . In addition, large inductors are connected to the drain nodes of the transistors to prevent RF signals leakage to the drain supply DC terminals [11]. Here, L D1 = 1.2 nH, L D2 = 0.8 nH and L D3 = 1.3 nH. Moreover, for a multistage LNA, checking the stability of the input and output ports alone cannot strictly guarantee the stable operation of the amplifier. Therefore, it is necessary to ensure that each stage in the amplifier meets the stability factor µ > 1 [25]. So, we simulated the stability factor µ for each single stage, and the simulation results show that µ is greater than 1 in the frequency band of 0.01-100 GHz.
IV. MEASUREMENT RESULTS
The proposed broadband LNA is designed and fabricated by 0.15-µm pHEMT process. The thickness of the substrate is 100 µm. Fig. 14 shows a micrograph of the LNA with compact 2.1×0.76 mm 2 chip size, including the input/output testing pads and the DC power pads. The measurement setup includes a probe station (Cascade Summit 11000M), a vector network analyzer (Agilent N5244A) and a noise analyzer (Keysight N8976B). op1db versus frequency is 4.5-12.8 dbm. For wideband lna design, it is difficult to achieve a better than 20db return loss in the operating band. the output return loss in this design is better than 5db, which results in large output power loss and large op1db deviation. Table 1 summarizes and compares the designed LNA with the reported broadband LNAs. It should be mentioned that the designed broadband LNA not only provides a wide bandwidth of 38.5 GHz, but also achieves a high gain and a low NF. The definition of figure of merit (FOM) can be written as [26]: He is currently the Head of the Microwave Microsystems Research and Development Department, and also a Researcher and a Doctoral Tutor with the Aerospace Information Research Institute (AIR), Chinese Academy of Sciences. His current research interests include millimeter-wave/terahertz integrated circuit, phased array antenna microsystems, communication, and radar. He was awarded the First Place Winner of the Chinese nonferrous metals industry science and technology, in 2016. VOLUME 11, 2023 | 4,550 | 2023-01-01T00:00:00.000 | [
"Computer Science"
] |
Numerical Solutions for a Model of Tissue Invasion and Migration of Tumour Cells
The goal of this paper is to construct a new algorithm for the numerical simulations of the evolution of tumour invasion and metastasis. By means of mathematical model equations and their numerical solutions we investigate how cancer cells can produce and secrete matrix degradative enzymes, degrade extracellular matrix, and invade due to diffusion and haptotactic migration. For the numerical simulations of the interactions between the tumour cells and the surrounding tissue, we apply numerical approximations, which are spectrally accurate and based on small amounts of grid-points. Our numerical experiments illustrate the metastatic ability of tumour cells.
Introduction
The analysis of data obtained from the World Health Organization (WHO) [1] and the UN [2] databases shows that, at present, cancer is and probably will remain to be among the leading causes of death worldwide [3][4][5] being surpassed only by cardiovascular diseases. According to the data provided by the WHO, cancer disease is the cause of the death of roughly six million people yearly [1]. This explains the major significance of the fight against the malignant conditions, which includes prevention [6], cure [7,8], and cancer research.
Tumour development is a very complex multistep process involving many intracellular and extracellular phenomena which are strongly nonlinear and time varying [4,[9][10][11]. Genomic changes as well as microenvironmental factors such as the extracellular matrix (ECM), various growth factors, and substrate concentrations have been shown to play a major role in the process of carcinogenesis [12].
Generally, tumours can be classified as benign and malignant. The growth of benign tumours is self-limiting and their cells tend to stay in the same place. Malignant tumours may grow without limitations and their constituent cells are prone to migrate or metastasize to other parts of the organism [13][14][15]. The ability of malignant cancer to invade the local tissue and to spread throughout the organism is their most insidious and dangerous property. Metastasis is the predominant cause of most cancer deaths [14,16,17].
The process of metastasis includes angiogenesis and invasion. Tumour angiogenesis (rapid growth of blood vessels near the tumour cells) is induced by a secretion of various growth factors such as vascular endothelial growth factor (VEGF). These vessels facilitate the influx of oxygen and other nutrients needed for the development of the cancer [18]. The process of angiogenesis is followed by invasion and penetration of cancer cells into surrounding tissues and possibly by dissemination of cancer cells through blood vessels. Thus, tumour cells can be carried to a distant site of the body. There they can implant and initiate the development of a secondary tumour [14,16,19]. An important role in the process of cancer invasion is performed by matrix degradative enzymes (MDEs) such as metalloproteases (MMPs). They are produced by tumour cells and digest the ECM, which enables the migration of cancer cells through the tissue [13,14,17].
In the last half century, many mathematical models describing the process of tumourigenesis have been the subject of active research. Mathematical and computational methods have contributed to clarifying the factors that are sufficient to explain experimental and clinical data, to defining these factors in precise terms and to suggesting experiments for calculation of these factors [20]. In addition, analyses and simulations of mathematical models have been used for the reduction of the amounts of costly experiments needed for the development of therapies [21,22]. It is strongly believed that mathematical and computational methods will play a significant role in cancer research in the future. They may improve the understanding of some complicated features and details of tumour evolution as well as be effectively used in clinical laboratories, by means of appropriate model-based decision support systems [4]. We refer the readers to special issues [23][24][25][26][27] for more complete bibliography regarding the applications of mathematical and computational methods to cancer research.
Gatenby and Gawlinski present one of the first models of tumour invasion in the papers [28,29]. Gatenby [28] considers the competition between healthy host cells and modified (tumour) cells and proposes and analyses several models formulated in terms of ordinary differential equations. Gatenby and Gawlinski [29] present a reaction-diffusion model for the investigation of the role of the alteration of the microenvironmental acidity induced by cancer cells for their invasion into the organism. Subsequently, the series of papers, among others [30][31][32][33][34][35][36][37][38][39][40][41][42][43], have appeared offering models and detailed analysis of diverse features of cancer invasion. In this paper, we study the continuum models of avascular tumour growth investigated by Chaplain et al. (cf., e.g., [31,[34][35][36][37]). The first model of this series is proposed in Anderson et al. [31]. The authors consider three major variables involved in the process of cancer invasion, namely, cancer cells, ECM, and MDEs. In order to study in detail mainly the influence of the surrounding tissue on the process of migration of tumour cells, the proliferation of the latter is not included in the continuum model. The authors analyse numerically in one and two dimensions the impact of ECM gradients resulting from the destruction of ECM by MDE and the role of haptotaxis on cancer invasion. An extension of this model is presented in Chaplain and Anderson [34] who consider the role of oxygen as a nutrient for the tumour cells. The authors propose also a new model equation for endogenous inhibitors, such as tissue inhibiting MMPs, that can neutralize MDEs. We include this equation in our model (8), see Section 2 below. The model of Chaplain and Anderson [34] has been further developed by Lolas [37] and Chaplain and Lolas [35,36] who have considered terms describing chemotaxis, proliferation of cancer cells and reestablishment of the ECM. Lolas [37] examines a variety of continuum models, in particular incorporating the effects of just chemotaxis, and haptotaxis, and their combination, and so forth. One of the conclusions of the author is that the mechanism of chemotaxis without haptotaxis cannot lead to a successful cancer invasion if there is no proliferation of tumour cells and reestablishment of ECM. Further novel ordinary differential equations that describe the cancer cell proliferation and the remodeling of the extracellular matrix re-establishment function allowing the incorporation of the plasminogen activation cycle are included in the model of Chaplain and Lolas [36] that also investigates the role of the uPA system for the cancer invasion. uPA inhibitors and plasmin have also been investigated in the model by Chaplain and Lolas [35]. Clear and detailed description of the biological processes observed during the cancer invasion and metastasis is provided in [31,[34][35][36][37]. In particular, in these paper, the key stages of the metastatic cascade, the structure and functions of the major constituents of the ECM and the basic representatives of the MDEs participating in the interactions between the healthy and cancer cells are systematically presented on the basis of broad theoretical and experimental bibliography.
In our paper, we propose a different numerical approach than the approach used, for example, in [31,[34][35][36][37]. The goal of the paper is to obtain numerical results which are based on small amounts of spatial grid points applied to the model equations so that low-dimensional vectors of data are used to make the numerical computations fast. We construct a new algorithm for the systems [31,34,36,37] by using spectrally accurate approximations to the terms that model the tumour cell random motility, the haptotaxis, the MDE diffusion, and the diffusion of the endogenous inhibitors. Since the algorithm computes the solutions with spectral accuracy, it is based on smaller amounts of spatial grid points than the amounts of grid points used for the less accurate finite difference approximations (strategy applied, e.g., in [31,[34][35][36][37]), which consequently saves computational time. The idea of using small amounts of spatial grid point and saving time for computing one solution for one set of parameters, which has to be repeated many times for many sets, is important, for example, for the numerical experiments carrying the goal of estimating parameter values from laboratory data. This idea is applied in [44] to estimate parameter values of one of the models presented in [36,37] from the in vivo experimental data [45] developed by using transgenic mouse models. The numerical approach from [44] is based on a different approximation to the haptotactic term than the approximations used in this paper and our numerical schemes are constructed for systems which are various variants and generalizations of the model investigated in [44]. Furthermore, because of considering different variants of boundary conditions the schemes in this paper differ from that of the paper [44].
Additionally to the model presented in [36,37] and applied in [44], in this paper, we investigate other models, which are presented in [34] or are combinations of the model equations from [34,36,37]. Moreover, in [44], the parameter values are evaluated quantitatively from the laboratory data [45] so that the solutions of the model equations correlate with the data. Contrarily to [44], in this paper, we choose the parameter values qualitatively in order to observe and compare solutions computed with different parameters. This comparison allows to analyse the influence of the parameters on the shape of the solutions and we conclude that complicated interactions between tumour cells, ECM, MDEs, and endogenous inhibitors can be directed by choosing the parameter values. Our sequence of numerical simulations is initiated from the solutions obtained with the parameter values chosen in [34] (for comparison) and next we gradually change the values and analyse their influence on the solutions. Animated graphical Computational and Mathematical Methods in Medicine 3 visualization of the solutions and how they change according to the parameters is helpful in observing the influence of the parameters on the shape of the solutions. The idea of using small amounts of spatial grid points and saving time for computing solutions of the model equations is crucial in the effective utilization of, for example, animated simulations of tumours, which can be used as a predictive and visualized tool in clinical applications. Decreasing the amounts of spatial grid points used for such visualizations saves not only the time of demonstrations but also the computer memory. It is not possible to demonstrate the animated simulations in papers and we only note that they are interesting and help in visualization of the complicated biological processes. Instead of the animated simulations we include snapshots at different stages in time.
The contents of this paper is as follows: the model equations are described in Section 2, the algorithm is introduced in Section 3, the results of numerical experiments and simulations are presented in Section 4, and Section 5 includes our concluding remarks and future research work.
Mathematical Model
In this section, we investigate various models of tissue invasion by cancerous cells. In Section 2.1, we investigate the Chaplain and Anderson model [34] focusing on interactions between ECM and cancer tumour and metastatic abilities of cancer cells. In Section 2.2, we investigate further expansions of the model and its different versions with additional terms connected with proliferation of tumour cells, ECM renewal, and different functions modelling the production of MDEs by the tumour cells. Section 2.3 deals with a more general model with an additional equation, which describes evolution of endogenous inhibitors.
Cell-Matrix Interactions and Cell Migration.
In the next section, we construct a numerical scheme for the following model of tissue invasion: with the space variable x belonging to the scaled domain [0, 1] of tissue, and time t. The model equations (1) describe interactions between tumour cells, MDEs, and ECM. The interacting variables are n-tumour cell density, f -ECM density, and m-MDEs concentration. The system (1) is derived in detail in [34] and is a part of a more general system consisting of (1) with an additional fourth equation for endogenous inhibitor concentration denoted by u. In [34], it is assumed that the tumour cells, the MDEs, and the inhibitors remain within the space domain and zeroflux boundary conditions are imposed. The fourth equation for the endogenous inhibitor concentration is dropped under the additional assumption that negative effect of the endogenous inhibitors is overcame by the MDEs in an actively invading tumour. This assumption implies that u = 0 and the general system of four equations is reduced to (1). In Section 2.3, we investigate the model with all four equations.
Migration and Proliferation of Cancer Cells, ECM
Renewal, and MDE Production. We also investigate further expansions of the model (1), which are introduced, for example, in [36,37]. After adding the proliferation term μ 1 n(1 − n − f ) to the right-hand side of the equation governing tumour cell motion (the first equation in (1)) and the ECM renewal term μ 2 f (1 − n − f ) to the right-hand side of the equation for the ECM (the second equation in (1)), we obtain the following model: where μ 1 is the proliferation rate of the tumour cells and μ 2 is the growth rate of the ECM. We also make experiments with the following modification of (2): where the MDE production is modeled by αn(1 − n). The motivation for choosing such form of the MDE production in [36,37] follows from experimental observations of polarized expression of MDEs at the invading leading edge of tumour, see, for example, Estreicher et al. [46]. We investigate the model equations (1), (2), and (3) supplemented by the zero-flux boundary conditions at x = 0 and either the Dirichlet conditions or the zero-flux boundary condition at x = 1. As in [34], we assume that the initial tumour is centered at x = 0, the initial MDE concentration is proportional to the initial tumour cell density with 1/2 as the constant of the proportionality, and the MDE has already degraded the ECM, thus we consider the same initial conditions as in [34], which are the following: for x ∈ [0, 1]. The parameter values for the model equations are specified in Section 4.
Production of Endogenous Inhibitors.
We additionally consider the general model where the last equation describes evolution of endogenous inhibitors (concentration of which is denoted by u). This equation is the fourth equation in the model (10.5) proposed by Chaplain and Anderson in [34], where it is assumed that endogenous inhibitors are produced by ECM as a response to the MDEs and the function F(m, f ) models the inhibitor production. The term θum models neutralization of the MDEs and ρu models decay of the inhibitors. We assume that the initial inhibitor concentration is and impose the zero-flux boundary conditions Our goal is to construct a new efficient algorithm for solving the models (1), (2), (3), and (8) and investigate the ability of cancer cells to produce and secrete the MDE, which then degrade the ECM, and allow the cells to start their migration towards healthy parts of the tissue.
Construction of Numerical Approximations to Tumour Cells, ECM, and MDEs
In this section, we construct numerical solutions to the model equations (1), (2), and (3) supplemented by the initial conditions (7) and the boundary conditions (4) and (5). For the numerical solutions, we consider the Chebyshev-Gauss-Lobatto points with i = 0, 1, . . . , N + 1, in the scaled domain [0, 1] of tissue. Our goal is to construct approximations to n( and m(x i , t), for i = 0, 1, . . . , N; the values of the solutions at x N+1 are known from (5). Let and we use similar notations for f and m. We shall replace the spatial derivatives in (1) by numerical approximations constructed for the vectors n xx (t), n x (t), f x (t), f xx (t) in the first equation and for m xx (t) in the third equation. For n x (t), we apply the following spectrally accurate approximations with i = 0, 1, . . . , N + 1, where is the first-order differentiation matrix based on the points (11), see [47,48]. We also apply the analogous spectrally accurate approximations for f x (t) and m x (t), respectively. Since the exact value of (∂n/∂x)(x 0 , t) is given by (4), the approximation (13) is not needed at the first point x 0 , that is, for the first component of n x (t). Therefore, from (13), the first approximation in (15) where From (13) and (16) we obtain the following approximation for the second-order derivatives where We now construct approximations to f x (t), m x (t), f xx (t), and m xx (t). From the spectrally accurate approximations (15) and from (4) and (5) we obtain According to (20), we have the following approximation for the second-order derivative of f with the notation From (21) we have with the similar notation for m We now replace the spatial derivatives in the model (1) by their corresponding approximations. We apply (18), (16), (20), and (22) to the first equation in (1) and obtain its discrete version written in the following form where stands for the component-wise multiplication between two vectors. The discrete version of the second equation in (1) is written in the form and from (24) we obtain the following discrete form of the third equation in (1) dm The resulting system (26)-(28) is composed of 3N + 3 ordinary differential equations and is a semidiscrete version of (1). Note that since the spatial derivatives in (1) are approximated with the spectral accuracy, much smaller numbers of grid-points x i are needed for (26)-(28) than for finite difference schemes, and time integration of the smaller systems is more robust and more efficient than time integration of the finite difference systems. For the models (2) and (3) supplemented with (4) and the right-hand side boundary condition (6), which is different than (5), we need to apply different approximations than (16), (18), (20), and (22) as they include (5) instead of (6). For this problem, from (13), instead of (16), we obtain where Further, from (29), we obtain Computational and Mathematical Methods in Medicine 7 As in (13) and (15), for the vector of approximations to the haptotactic term we obtain and instead of (24), from (6), we obtain From (29), (31), (33), and (34) we obtain the following scheme for the problem (3), (4), and (6) dn where e is a vector entries of which are all 1-s. For (2), the component αn(t) (e − n(t)) needs to be replaced by αn(t) in the last equation of (35). From the boundary conditions (10), we obtain the approximation for the diffusion of the endogenous inhibitors u xx (t) ≈ D (1) 00 u(t) (36) and the semi-discrete version for the last equation in (8) is written in the following form where we assume that the inhibitor production is modelled by F(m, f ) = ξ f . The semi-discrete equations have to be closed by initial conditions chosen according to (7) and (9).
Numerical Experiments
We apply the approximations introduced in Section 3 and begin our series of numerical simulations from (26)- (28), which correspond to model (1). Results of our numerical experiments are presented in Figures 1-6 (1): random motility d n Δ 2 n and haptotaxis −γΔ· (nΔ f ), respectively. Since γ is greater for Figure 3 than for Figure 1, because of larger haptotactic migration in Figure 3 than in Figure 1, the two clusters seen in Figure 3 are more separated from each other than the two clusters in Figure 1. The pictures show the effect of haptotaxis. The small clusters of cells, which break away from the main body of the tumour, illustrate the potential for the cancer cells to degrade the surrounding tissue, migrate, and start the metastatic cascade. The migrations of the small clusters may not be detected during the processes of medical treatments, and even after resections of the main tumours, the new small clusters may initiate recurrences of the disease. A new cluster of tumour cells broken away from the main body of the tumour is also observed in Figures 5 and 6, which present numerical data computed with γ = 0.02 and η = 20.
The next part of our numerical experiments concerns the models (2) and (3) supplemented by (4), (6), and (7). The results for model (2) are presented in Figures 7, 8, 11, and 12 and for model (3) in Figures 9, 10, 13, and 14. These experiments start from the initial condition (7) corresponding to the snapshot in time t = 0 in Figure 1. We observe that the small clusters of cancer cells separated from the main tumours are better formed at t = 2 than at t = 1 and as time evolves the haptotactic migration together with the production of new cancer cells spread the shapes of the tumours over the x-domain. We also observe that the snapshots in time t = 1 in Figures 1, 7, and 8 look similar to each other and the models (1), (2), and (3) give similar results for t ∈ [0, 1] although they are supplemented by the different boundary conditions, either (5) or (6), and solved with different parameters μ 1 , μ 2 ∈ {0, 0.1, 0.5} and β ∈ {0, 0.07}. However, these similarities are observed only for t ∈ [0, 1] and as time evolves the corresponding solutions of the models (1), (2), and (3) differ from each other. For example, already at t = 2, Figure 8 shows greater production of tumour cells than Figure 7. Moreover, at t = 10 and t = 20, Figure 7 shows weaker MDE production and greater production of the tumour cells and the ECM than in Figure 1 due to the fact that μ 1 , μ 2 , and β are greater in Figure 7 than in Figure 1. On the other hand, also at t = 10 and t = 20, Figure 8 shows greater MDE production than Figure 1. Although β = 0.07 for Figure 8 and β = 0 for Figure 1, since the MDE production is greater in Figure 8 than in Figure 1, the MDE concentration is greater in Figure 8 than in Figure 1 and consequently the ECM degradation is more progressive in Figure 8 than in Figure 1. It can also be observed that although the parameters d m , α, and β in the third equation of (1) and (2) are the same for Figures 1, 7, and 8, the MDE curves show different MDE concentrations in all of these figures.
In Figures 7-10, we observe differences due to the MDE production terms αn and αn(1 − n) in (2) and (3), respectively. The parameter values for Figure 9 are the same as for Figure 7 and the parameter values for Figure 10 are the same as for Figure 8 but the MDE production is weaker in while Figures 13 and 14 illustrate numerical solutions to (3). Comparison of Figures 11-14 confirms that the term αn(1 − n) in (3) models lower MDE production than the term αn in (2) (with the same parameter values). We also observe that since MDE production is lower in Figures 13 and 14 than in Figures 11 and 12, the ECM degradation is smaller in Figures 13 and 14 than in Figures 11 and 12. Figure 15 shows snapshots in time of the four solutions to the more general model (8) describing the interactions between the tumour cells, ECM, MDEs, and endogenous inhibitors. All four profiles show that, as time evolves, the ECM produces endogenous inhibitors, concentration of which increases in time and their higher concentration is located in the regions where ECM is not yet entirely degraded rather than in the regions where the degradation is already effectively developed. The inhibitor profile shows that the ECM responds to the MDEs by producing the endogenous inhibitors.
Concluding Remarks and Future Directions
We have constructed a new numerical algorithm for fast computations of the solutions of the mathematical models proposed by Chaplain et al. in [34,36,37], which consist of systems of nonlinear partial differential equations describing interactions between tumour cells, ECM, and MDEs. The algorithm is based on spectrally accurate approximations and small amounts of grid points, which results in ordinary differential systems of small dimensions and fast computations. We have applied the algorithm and presented and compared the numerical simulations with a variety of model equations. The simulations demonstrate that the models describe important features of the interactions between tumour cells and the surrounding tissue, and in particular the initiation of a new colony of cells and metastasis.
Our future research work will address the question for which parameter values and domains the model [34] and the kinetic type model proposed in [49] are equivalent. We will also address numerical methods with spectrally accurate approximations for the models with two-dimensional spatial domain and with different kinds of the function F(m, f ) modelling the inhibitor production [34]. | 5,798.8 | 2010-12-30T00:00:00.000 | [
"Mathematics",
"Medicine"
] |
Predicted Membrane-Associated Domains in Proteins Encoded by Novel Monopartite Plant RNA Viruses Related to Members of the Family Benyviridae
As a continuation of our previous work, in this paper, we examine in greater detail the genome organization and some protein properties of the members of a potential group named Reclovirids and belonging to Benyviridae-related viruses. It can be proposed that the single-component Reclovirid genomes encode previously undiscovered transport genes. Indeed, analysis of the coding potential of these novel viral genomes reveals one or more cistrons ranging in size from 40 to 80 to about 600 codons, located in the 3′-terminal region of the genomic RNA, encoding proteins with predicted hydrophobic segments that are structurally diverse among Reclovirids and have no analogues in other plant RNA viruses. Additionally, in many cases, the possible methyltransferase domain of Reclovirid replicases is preceded by membrane-embedded protein segments that are not present in annotated members of the Benyviridae family. These observations suggest a general association of most Reclovirid proteins with cell membranes.
Introduction
Some recent papers have highlighted a significant diversity of monopartite plant, fungal, and insect viruses encoding replicases related to those encoded by the multipartite RNA genomes of members of the genus Benyvirus [1][2][3][4][5][6].
However, RNA-dependent RNA polymerase (RdRp) sequences from viruses of the genus Benyvirus are most closely related to the proteins encoded by VLRAs of land plants only, and form a distinct branch of the corresponding phylogenetic tree (see Figure 2 in [2] and Figure 3 in [6]). This cluster of the phylogenetic tree includes two sub-branches, namely those representing viruses with multipartite and monopartite genomes. In addition to members of the genus Benyvirus, the former branch includes several members of the Benyviridae family, particularly, Wheat stripe mosaic virus and Fern benyvirus. It also contains a recently described group of "Tetra-cistron movement block (TCMB)-containing viruses" (Tecimovirids) coding for the TCMB movement gene module instead of the triple gene block (TGB) of movement genes found in members of the family Benyviridae. [2,7] (Figure 1). The latter branch contains exclusively monopartite viruses and, in particular, includes Goji berry chlorosis virus (GBCV), which encodes six polypeptides and has no close relatives with similar genome organization [8]. Four additional benyvirus-like members of this brunch are predicted to encode a binary movement block (BMB), which is also found in multipartite viruses of the family Kitaviridae [7,[9][10][11] (Figure 1). In the course of recent studies on the phylogeny of benyvirus-like RNA polymerases, we found that phylogenetically close monopartite viruses include a new, previously In the course of recent studies on the phylogeny of benyvirus-like RNA polymerases, we found that phylogenetically close monopartite viruses include a new, previously un-described group of viruses related to members of the family Benyviridae. These viruses contain polyadenylated, single-component RNA genomes with a maximum size of up to 10,000 nucleotide residues, include several species annotated in the NCBI database, and have been named Reclovirids after Red clover virus 1 [2]. In contrast to the vast majority of annotated plant viruses in the family Benyviridae, the viral genomes of Reclovirids do not encode any previously characterized transport gene blocks. Our initial assumption that these viral genomes belong to two-or multi-component viruses has not been confirmed [2]. Thus, it would be proposed that the single-component Reclovirid genomes encode previously undiscovered transport genes. The coding potential of these novel viral genomes has revealed one or more cistrons ranging in size from 40 to 80 to about 600 codons, located in the 3 -terminal region of the genomic RNA. The encoding proteins with predicted hydrophobic segments have no analogues in other plant RNA viruses [2]. These data allowed us to hypothesize that flowering plants can be infected with the novel viruses (Reclovirids), which are related to the members of family Benyviridae and belong to a new taxon in the order Hepelivirales (probably a subfamily, or even a family). In this paper, we comparatively analyzed the gene organization of Relovirid genomes and the structural properties of the encoded non-replicative "orphan" hydrophobic polypeptides as well as replicative polypeptides.
Results
The NCBI non-redundant nucleotide sequence library currently contains the only four annotated Reclovirids. These viruses include Red clover virus 1, Dactylorhiza hatagirea beny-like virus, Carrot associated RNA virus 1, and Arceuthobium sichuanense virus 3 [2]. Public databases of plant transcriptomes proved to be a good source for identifying previously unknown viruses. Therefore, we searched the NCBI transcriptome shotgun assembly (TSA), short read assembly (SRA) and 1KP libraries for Reclovirus-like transcripts in order to identify novel viral sequences. We have discovered a total of 27 viruses ( Figure 1) infecting 78 species from 12 families of monocotyledonous and dicotyledonous flowering plants (Table 1).
Genome Organization of Reclovirids
Most Reclovirid genomes contain two ORFs, namely, the 5 -terminal long replicase ORF (approximate size of 6300-7000 nucleotides in the nearly full-length genomic RNAs) and the shorter downstream ORF with sizes varying from 330 to 2000 nts [2] (Supplementary Table S1). In some cases, Reclovirid genomes encode two proteins in addition to replicase, and their ORFs often overlap. An exception is Scutellaria montana VLRA, which encodes four small proteins in addition to replicase ( Figure 2) (Supplementary Table S1). The gene organization of the 3 -terminal regions in Reclovirid RNAs is rather variable. For example, two highly similar VLRAs found in Gymnadenia rhellicani (family Orchidaceae) ( Table 1)
Genome Organization of Reclovirids
Most Reclovirid genomes contain two ORFs, namely, the 5′-terminal long replicase ORF (approximate size of 6300-7000 nucleotides in the nearly full-length genomic RNAs) and the shorter downstream ORF with sizes varying from 330 to 2000 nts [2] (Supplementary Table S1). In some cases, Reclovirid genomes encode two proteins in addition to replicase, and their ORFs often overlap. An exception is Scutellaria montana VLRA, which encodes four small proteins in addition to replicase ( Figure 2) (Supplementary Table S1). The gene organization of the 3′-terminal regions in Reclovirid RNAs is rather variable. For example, two highly similar VLRAs found in Gymnadenia rhellicani (family Orchidaceae) ( Table 1) Table S1).
Protein Domains in the Replicases of Reclovirids
We have previously reported that the RNA polymerases of Red clover RNA virus 1 and other Reclovirids have two characteristic domains, namely, a viral helicase 1 domain (HEL, pfam01443) and an RdRp core motif (pfam00978). In addition, analysis of the NCBI Conserved Domain Database (CDD) has shown no viral methyltransferase domain (MTR, pfam01660) in the replicases of Reclovirids as well as members of the genus Benyvirus [2]. Nevertheless, the current comprehensive CDD analysis of benyvirus-like replicases distantly related to those of Reclovirids (see Figure 2 in [2]) has shown that the insect Hubei beny-like virus 1 (HBLV1) encodes an MTR domain (pfam01660, e-value 6e−16) in the Nterminal part of the replicase protein. The BLASTP search revealed that Reclovirid full-
Protein Domains in the Replicases of Reclovirids
We have previously reported that the RNA polymerases of Red clover RNA virus 1 and other Reclovirids have two characteristic domains, namely, a viral helicase 1 domain (HEL, pfam01443) and an RdRp core motif (pfam00978). In addition, analysis of the NCBI Conserved Domain Database (CDD) has shown no viral methyltransferase domain (MTR, pfam01660) in the replicases of Reclovirids as well as members of the genus Benyvirus [2]. Nevertheless, the current comprehensive CDD analysis of benyvirus-like replicases dis-tantly related to those of Reclovirids (see Figure 2 in [2]) has shown that the insect Hubei beny-like virus 1 (HBLV1) encodes an MTR domain (pfam01660, e-value 6e−16) in the N-terminal part of the replicase protein. The BLASTP search revealed that Reclovirid fulllength replicases contain moderately comparable domains of 340-350 amino acids (Table 2 and Figure 2), which yield negative CDD results. Pairwise comparisons revealed that all Reclovirid MTR sequences contain a number of gaps when compared to the HBLV1 protein domain ( Table 2). This fact may lead to negative results in CDD searches for Reclovirid replicases. Importantly, a recent general comparative analysis of Riboviria-encoded methyltransferases [12] clearly indicates that the methyltransferase domains are characteristic for replicases of Benyviridae and particularly the genus Benyvirus. Interestingly, in some cases, the possible methyltransferase domain of Reclovirid replicases is preceded by membrane-embedded protein segments that are not present in annotated members of the family Benyviridae (Supplementary Figure S1). To date, the presence and function of membrane-spanning segments in viral RNA replicases has only been well-characterized for coronaviruses [13,14].
Protein Domains and Motifs in Non-Replicative Proteins of Reclovirids
It has been shown that most non-replicative proteins of Reclovirids possess predicted membrane-embedded segments [2] (Supplementary Table S1 and Figure S2). The length of these proteins varies from 40 to nearly 670 amino acids (Supplementary Table S1). All of these proteins represent "orphan" viral proteins with membrane-spanning domains [9]. Previously, pairwise sequence comparisons have revealed a group of hydrophobic nonreplicative proteins with apparent overall sequence conservation among Reclovirids infecting the family Orchidaceae (see Figure 9 in [2]). All these proteins contain a characteristic CX(3)CX(10)CX(3)C motif (putative zinc-finger domain) in the N-terminal region ( Figure 3) and hydrophobic segments in the C-terminal half (Supplementary Figure S2). Currently, we have found putative zinc-finger domains with slightly different motifs in non-replicative proteins of some other Reclovirids ( Figure 3A). Our analysis revealed that a number Reclovirids encode non-replicative proteins with other types of putative zinc-finger motifs. In particular, the ORF2 protein of Atriplex prostrata VLRA (Supplementary Table S1) contains the hexa-cysteine motif C(X6)C(X)C(XX)CXC(X12)C, which resembles unconventional hexa-cysteine motifs like those found in proteins of Hepatitis virus E and viruses of the genus Pestivirus [16,17], and in the small protein encoded by the ORF preceding TCMB in Colobanthus quitensis VLRA [7].
Discussion
The discovery of multiple plant-specific, capsidless Reclovirid VLRAs by highthroughput sequencing raises a number of questions about the evolution of the Benyviridae-like viruses and, in particular, the directions of evolution of plant virus-movement protein systems [2]. We have proposed that Reclovirids are likely to use novel, as yet undescribed, movement protein systems. The closely related plant Benyviridae-like viruses use movement gene blocks (BMB, TGB, or TCMB) that encode one, two or three small proteins with hydrophobic membrane-spanning segments as well as RNA helicase. On the other hand, the Reclovirid cell-to-cell movement may be carried out by a wide variety of the hydrophobic "orphan" proteins (usually a single protein per virus) [2].
What characteristics of these proteins make them suitable candidates for viral RNA transmission from cell to cell? What are the specific features that make these proteins suitable candidates for performing cell-to-cell trafficking of virus RNA? Considering single MP-based transport systems exemplified by that of Tobacco mosaic virus, several features can be distinguished, namely, the ability to bind RNA, interact with ER/actin/microtubules, modify plasmodesmata (PD), move from cell to cell, and direct virus replication complexes to PD [18,19]. It should be noted that in case of Reclovirids, the latter function can be performed directly by the replication protein due to the presence of the N-terminal membrane-spanning segments, which can potentially target the ER at the PD entrance. Computer prediction methods have indicated that non-replicative proteins of Reclovirids can perform at least two of the other MP functions. First, most Reclovirid "orphan" proteins contain membrane-spanning segments at their N-termini and internal trans-membrane regions that may direct these proteins to the ER membranes. Second, these proteins possess putative zinc-finger motifs, which in many cases are known to participate in nucleic acid binding (including ssRNA binding) [20][21][22]. Interestingly, the Rice yellow mottle virus protein with movement and silencing suppression functions has been shown to Quite interestingly, CDD analysis of non-replicative proteins of Reclovirids revealed that the largest of these proteins (668 aa in length), the Astragalus canadensis VLRA ORF2 protein, contains a domain of the Mpp10 protein family (COG5384) (positions 205-465, e-value 2.84e−03). This family includes proteins related to Mpp10, which is part of the U3 small nucleolar ribonucleoprotein in yeast [15]. Outside the Mpp10-like domain, this protein contains two putative membrane-embedded segments (Supplementary Figure S2) and a region of distant similarity (identity 24%, e-value 2.84e−03) to the non-replicative protein of two other Reclovirids (namely, 2063722-Leontopodium_alpinum VLRA and Vicia faba VLRA) (Supplementary Table S1). This region contains a putative zinc-finger domain with the signature CX(3)CX(7)CX(3)C ( Figure 3B).
Our analysis revealed that a number Reclovirids encode non-replicative proteins with other types of putative zinc-finger motifs. In particular, the ORF2 protein of Atriplex prostrata VLRA (Supplementary Table S1) contains the hexa-cysteine motif C(X6)C(X)C(XX)CXC(X12)C, which resembles unconventional hexa-cysteine motifs like those found in proteins of Hepatitis virus E and viruses of the genus Pestivirus [16,17], and in the small protein encoded by the ORF preceding TCMB in Colobanthus quitensis VLRA [7].
Discussion
The discovery of multiple plant-specific, capsidless Reclovirid VLRAs by highthroughput sequencing raises a number of questions about the evolution of the Benyviridae-like viruses and, in particular, the directions of evolution of plant virus-movement protein systems [2]. We have proposed that Reclovirids are likely to use novel, as yet undescribed, movement protein systems. The closely related plant Benyviridae-like viruses use movement gene blocks (BMB, TGB, or TCMB) that encode one, two or three small proteins with hydrophobic membrane-spanning segments as well as RNA helicase. On the other hand, the Reclovirid cell-to-cell movement may be carried out by a wide variety of the hydrophobic "orphan" proteins (usually a single protein per virus) [2].
What characteristics of these proteins make them suitable candidates for viral RNA transmission from cell to cell? What are the specific features that make these proteins suitable candidates for performing cell-to-cell trafficking of virus RNA? Considering single MP-based transport systems exemplified by that of Tobacco mosaic virus, several features can be distinguished, namely, the ability to bind RNA, interact with ER/actin/microtubules, modify plasmodesmata (PD), move from cell to cell, and direct virus replication complexes to PD [18,19]. It should be noted that in case of Reclovirids, the latter function can be per-formed directly by the replication protein due to the presence of the N-terminal membranespanning segments, which can potentially target the ER at the PD entrance. Computer prediction methods have indicated that non-replicative proteins of Reclovirids can perform at least two of the other MP functions. First, most Reclovirid "orphan" proteins contain membrane-spanning segments at their N-termini and internal trans-membrane regions that may direct these proteins to the ER membranes. Second, these proteins possess putative zinc-finger motifs, which in many cases are known to participate in nucleic acid binding (including ssRNA binding) [20][21][22]. Interestingly, the Rice yellow mottle virus protein with movement and silencing suppression functions has been shown to contain two essential zinc-finger motifs [23][24][25]. One of these motifs belongs to the C4 type, as those found in most Reclovirid "orphan" proteins, and performs an unknown function in cell-to-cell movement [23]. Thus, it can be proposed that the hydrophobic Reclovirid proteins with the zinc-finger motif may be responsible for genomic RNA binding and the interaction with the ER tubule, as well as increasing the PD permeability. Obviously, these hypotheses require more experimental validation. In evolutionary terms, this type of putative movement protein may represent an alternative for adaptation of beny-like viruses to colonize multicellular land plants.
Materials and Methods
Reclovirid-related nucleotide and protein sequences were collected from the NCBI plant transcriptome database. Sequence comparisons were carried out using the BLAST algorithm (TBLASTn and BLASTp) at the National Center for Biotechnology Information (NCBI). The nucleotide raw sequence reads from each analyzed SRA experiment linked to the virus nucleotide sequence and TSA projects returning Reclovirid-like hits were downloaded and subjected to bulk local BLASTX searches (e-value ≤ 1e−105) against a Refseq virus protein database available at ftp://ftp.ncbi.nlm.nih.gov/refseq/release/viral/ viral.1.protein.faa.gz (accessed on 20 May 2023). The resulting viral sequence hits of each SRA read were then visually explored. Tentative virus contigs were extended by iterative mapping of each SRA library's raw reads. This strategy employed re-iterative BLAST to extract a subset of reads related to the query contig, and these retrieved reads were used to extend the contig, and then, the process was repeated iteratively using as the query the extended sequence.
Conclusions
In conclusion, it should be noted that a number of the recently annotated plant tymovirus-like genomes may have small "orphan" hydrophobic protein ORFs and lack the well characterized CPs and MP systems. In particular, these viruses include Broom forkmoss associated tymo-like virus, Yellow horn associated tymo-like virus, Badge moss associated tymo-like virus, Polish wheat virus 1, Kava virus 1, Agave tequiliana deltaflexivirus 1, and Sesame deltaflexivirus 1 [6,27]. Thus, it appears that Reclovirids represent members of a class of plant viruses with missing or as yet unknown cell-to-cell movement systems. Indeed, it is known that persistent plant viruses lack cell-to-cell movement systems and do not cause visible symptoms; accordingly, they are transmitted only vertically via gametes. Persistent plant viruses represent a few virus families such as Endornaviridae [28]. Among non-persistent plant viruses, several viruses named umbravirus-like associated RNAs (ulaRNAs) that lack both CP and MP genes have recently been discovered [29,30]. It is proposed that ulaRNAs could be transmitted by arthropods and spread within the plant body with the help of other co-infecting viruses from the family Tombusviridae. After the initial vector transmission, the helper virus could be lost during progress of infection, for example, because of high temperatures [29]. | 3,984.4 | 2023-07-29T00:00:00.000 | [
"Biology"
] |
Small Target Detection in Refractive Panorama Surveillance Based on Improved YOLOv8
Panoramic imaging is increasingly critical in UAVs and high-altitude surveillance applications. In addressing the challenges of detecting small targets within wide-area, high-resolution panoramic images, particularly issues concerning accuracy and real-time performance, we have proposed an improved lightweight network model based on YOLOv8. This model maintains the original detection speed, while enhancing precision, and reducing the model size and parameter count by 10.6% and 11.69%, respectively. It achieves a 2.9% increase in the overall<EMAIL_ADDRESS>and a 20% improvement in small target detection accuracy. Furthermore, to address the scarcity of reflective panoramic image training samples, we have introduced a panorama copy–paste data augmentation technique, significantly boosting the detection of small targets, with a 0.6% increase in the overall<EMAIL_ADDRESS>and a 21.3% rise in small target detection accuracy. By implementing an unfolding, cutting, and stitching process for panoramic images, we further enhanced the detection accuracy, evidenced by a 4.2% increase in the<EMAIL_ADDRESS>and a 12.3% decrease in the box loss value, validating the efficacy of our approach for detecting small targets in complex panoramic scenarios.
Introduction
The field of view in conventional optical system cameras is constrained.As a result, they can only record a limited set of data.However, when covering a vast area for observation, people's observational needs frequently exceed these restrictions.With their 360-degree field of vision, omnidirectional imaging or panoramic cameras [1] offer an efficient option.A far more comprehensive range can be seen by mounting panoramic cameras from great heights for overhead views, or using drones for aerial patrols.This quality of panoramic cameras gives them a broad range of potential applications, such as simultaneous localization and mapping (SLAM), missile tracking systems, security surveillance for large outdoor warehouses, visual tracking of drones, visual surveillance of border defense equipment, and other fields [2].
Panoramic cameras achieve this vast field of view through designed optical components like fisheye, annular, reflecting lenses, or multi-camera image stitching.Since the first reflective panorama system's introduction in 1997 [3], panoramic imagery has seen significant advancements, especially in producing high-quality, real-time images encompassing a full 360-degree horizontal view.Despite these advancements, detecting small targets in panoramic images poses unique challenges.Compared to regular photographs, panoramic images have lower resolution per object due to their wide field of view, increasing the complexity of the background and making small target detection more difficult.Addressing these challenges is crucial, especially as limited public datasets are available for catadioptric panoramic images.Our study focuses on enhancing the precision of small target detection in panoramic images and meeting real-time detection requirements.
Current methods for panoramic target identification predominantly focus on prominent targets, leaving a significant gap in the research that explicitly targets the panoramic detection of small objects.Our study aims to address this need and enhance the precision of small target detection in panoramic images, while meeting real-time detection requirements.We have made several key advancements in this area.The main contributions of this study are as follows: (1) We propose a novel data augmentation method for panoramic images involving copy-paste data.This technique significantly increases the number of training set samples for small targets, ultimately enhancing the model's ability to detect small targets within panoramic photographs.(2) We advanced the original YOLOv8s [4,5] network by integrating a specialized tiny target detection layer and replacing the C2f layer in the backbone network with a C2f_EAM layer that incorporates the EAM attention mechanism.We specifically applied these enhancements to the YOLOv8 network.(3) We upgraded the eighth layer in the network to the FasterNext configuration.This development improves the model's accuracy and convergence rate and reduces the computational load required.(4) Our study incorporated a preprocessing step for panoramic images involving planar unfolding, truncation, and stitching.This technique significantly improved small target detection in panoramic images.
Related Work
Over the past few decades, significant research has focused on small target detection, a field challenged by the limited size, texture, and shape information these targets provide, often ranging from just a few to dozens of pixels.This has led researchers to continually refine detection algorithms for small targets, aiming to strike a balance between detection speed and model accuracy.
In the early stages of target detection, methods often relied on manually designed features combined with detection classifiers.However, since 2012, the rapid development of convolutional neural networks (CNN) has spurred a shift towards deep learning-based object detection algorithms.The two-stage target detection methodology, starting with the region-based convolutional neural network (R-CNN) in 2014 and evolving to Fast R-CNN and SPPnet [6,7], generates candidate regions and classifies them.Though offering high detection accuracy, these methods tend to be slower.In contrast, single-stage target detection algorithms like YOLO (You Only Look Once), SSD (Single-Shot Detector) [8,9], and DETR (Detection Transformer) [10], prioritize speed but often at the expense of accuracy.
Researchers have proposed various methodologies to enhance the detection of small targets.For instance, they have introduced deformable convolution [11] to improve the process of sampling the object size and shape, which addresses the limitations of fixed convolution in extracting spatial information.The feature pyramid network (FPN) [12] development aimed to preserve richer scale features, thus bolstering small target detection.Gong Y and colleagues devised a fusion factor to balance the information transfer between deep and shallow layers of the image pyramid, enhancing the learning efficiency for small targets [13].Additionally, they proposed the Normalized Wasserstein Distance (NWD) as a scale-insensitive method for measuring similarity [14].They developed feature superresolution (FSR) [15] as an alternative to traditional image super-resolution approaches.
Several researchers have addressed the unique challenges in the context of panoramic imaging.Pengyu Zhao [16] and the team proposed the reprojected R-CNN for rapid and precise detection in 360 • panoramic images.Junjie Wu [17] and his team developed a method using the Sample Adaptive View Transformer (SAVT) module, enhancing notable target detection in 360 • panoramic images.Pengfei Jia [18] and others improved the YOLOv4 algorithm for panoramic video target detection.At the same time, Olfa Haggui [19] and colleagues developed a color histogram comparison model based on the Bhattacharyya distance for fisheye image tracking.Cai Chengtao's team improved the YOLO algorithm for real-time multi-object monitoring in catadioptric panoramic images.Hongzhen Xu [20] and others used the Lucas-Kanade optical flow algorithm for moving obstacle detection in panoramic images.Ma Ziling and colleagues utilized a differential block elimination model for motion target detection.Yuhang He [21] introduced a multi-modal target tracking framework that leverages 2D panoramic images and 3D point clouds.
Methodology
Figure 1 depicts the workflow utilized in this study for detecting small targets within custom-made panoramic datasets.The diagram conveys the end-to-end process, from the original panoramic image acquisition to the final detection output.The annotation of raw dataset images is conducted using the LabelImg tool, preparing them for compatibility with the YOLOv8 model's requirements.Following this, data augmentation strategies are implemented to correct sample imbalances and to enrich the dataset post-panoramic expansion.The YOLOv8 detector is then trained on these annotated and processed images.Finally, the fully trained model is applied to previously unseen images for detecting small targets, culminating in a comprehensive visualization of the detected targets.⃝ the acquired images are annotated using the LabelImg tool and then cropped to conform to the YOLOv8 model's input format; 2 ⃝ data augmentation is utilized to address sample imbalance and to expand the dataset after panoramic expansion; 3 ⃝ these processed images are used to train the YOLOv8 detector; 4 ⃝ the trained model is deployed to detect small targets in panoramic images, producing the final visualization of the detection results.
YOLOv8s Network Architecture
The YOLOv8 network [5,22], which debuted in 2023 and is the most recent iteration of the YOLO architecture series [4], does away with the bounding box operation that YOLOv5 [23][24][25] utilized.Its primary input is 640 × 640, and depending on the scaling factor, the network offers different size models with N, S, M, L, and X dimensions to accommodate the requirements of a wide variety of use cases.When the YOLOv8s model was put through its paces using the official COCO dataset, the mAP@50-90 accuracy value reached 44.9, and the CPU inference speed was measured at 128.4 milliseconds.
The high detection accuracy of YOLOv8 stands as one of its most notable features, with YOLOv8s demonstrating a 7.5% improvement in model accuracy compared to YOLOv5s.While the inference speed of the YOLOv8s model is marginally lower than that of YOLOv5s, it significantly excels in regard to accuracy, especially for non-small objects.However, it was found that the standard YOLOv8 model is less effective in accurately detecting small targets.This limitation led us to focus our research on enhancing small target detection.The structure of the YOLOv8s network, which our research is based on, is depicted in Figure 2. Our enhancements, while not visually represented in Figure 2, aim to optimize the YOLOv8s architecture to improve the precision, accuracy, and recall of small target detection in wide-area panoramas, with minimal impact on the detection speed.These advancements have been instrumental in refining the network's capability for small target detection in complex panoramic scenes.The YOLOv8s architecture comprises three main components: the backbone network, neck network, and detection network.For the backbone network, YOLOv8s specifically utilizes DenseNet-53 for its backbone network, chosen for its robust feature extraction capabilities.This architecture incorporates the C2f module instead of the standard C3 module, as depicted in Figure 2. The C2f module, inspired by CSPNet's bypass extraction concept, integrates with the residual structure and consists of three convolution modules (Conv+BN+SiLU) and n bottlenecks.Such configuration allows YOLOv8s to ensure a lighter network, while facilitating richer gradient flow information.
The YOLOv8 model employs a bounding box-free operation.As shown in Figure 2, the structure of the head section visibly draws inspiration from the design concepts of YOLOX and YOLOv6 [26,27] and implements a decoupled-head structure.This structure uses two convolution layers to perform classification and regression operations.Specifically, the upper branch is responsible for the convolution regression of the bounding box, while the lower limb handles the convolution classification of the categories.Worth noting is that, compared to YOLOv5, the YOLOv8 model eliminates the obj_loss branch, thereby negating the need for a grid confidence calculation.Additionally, the number of channels in the regression head is set to be four times the maximum regression value (reg_max).
The YOLOv8 model has adopted a novel strategy for sample matching.It discards the traditional IOU matching or unilateral ratio allocation method, instead using the taskaligned assigner for positive and negative sample matching, and introducing Distribution Focal Loss (DFL) [28] (as shown in Equation ( 1)).
Equation ( 1) modifies the conventional focal loss by considering the distribution of the predicted value around the target class y.Here, S i and S i+1 represent the scores of the adjacent classes of the actual class y.The term y i+1 − ylog(S i ) weighs the log likelihood of the lower adjacent class S j by the difference between the upper adjacent class y i+1 and the actual value y, while (y − y i ) log(S i+1 ) similarly weighs the likelihood of the higher adjacent class S j+1 .Through this, DFL allows the network to adaptively focus on the prediction errors closer to the actual value, thereby improving the accuracy of bounding box localization.
For the classification loss function, YOLOv8 incorporates Versatile Focal Loss (VFL), as formulated in Equation ( 2): VFL applies an asymmetric weighting operation to adjust the loss based on the label q.When q > 0, indicative of a positive sample, the loss is calculated using a weighted combination of the predicted probability p and the adaptive IOU weighting q, emphasizing the agreement between the predicted box and the ground truth; for negative samples where q = 0, the term −αp γ log(1 − p) reduces the loss for well-classified negative samples, preventing them from dominating the gradient.This selective focus on positive samples and their alignment with the ground truth aids in refining the classification accuracy.
Addition of Small Object Detection Layer
To enhance the detection of small objects in panoramic images, we can understand from the network architecture of YOLOv8s that the input image size is 640 × 640, and the output feature size is 80 × 80, 40 × 40, 20 × 20.When the size of the target object is 30 × 30, the corresponding size of the feature map is 3.75, 1.875, 0.9375.If it is less than 1, the feature does not exist.Therefore, the smaller the target size, the weaker the detection capability.
Processing panoramic images presents a unique challenge: the expansive field of view necessitates the detection of smaller targets.We have enhanced our network to integrate small target detection methodologies more deeply to address this. Figure 3 illustrates that a new branch has been added after the second layer, C2f.The area highlighted in red signifies our optimization, introducing a 160 × 160 feature map to the output.This advancement facilitates more accurate detection of smaller targets by preserving extensive feature information during training.As the network deepens, however, there is a potential loss of valuable small target information, and the model's size may increase, slowing the detection speed.To counter this, we incorporated the multiscale attention (EMA) mechanism and the FasterNext module in subsequent steps.This allows the network to concentrate more effectively on small target information, enhancing the overall detection accuracy and efficiency, while reducing the operational speed.
Integration into the FasterNext Network
To enhance model training efficiency and minimize the number of floating-point operations (FLOPs), Jay Chen et al. [29] introduced the faster neural network (FNN), a novel network that significantly bolsters the operational speed of the network without compromising the accuracy of visual tasks.Inspired by their work, we incorporated FNN into YOLOv8.We further improved the faster neural network by replacing the C2f network structure at layer 8 in the YOLOv8s model with an enhanced FasterNext module.The design of the FasterNext neural network proceeds as follows.Due to frequent memory access, traditional convolution often needs to be more efficient in terms of computational speed and floating-point operations (FLOPs).Despite being widely adopted as a fundamental building block for neural networks, depthwise convolution (DWConv), a variant of Conv, also presents significant challenges.We have utilized the FasterNext network to address these issues, which employs parametric convolution (PConv) [30].This approach allows more efficient and effective use of the device's computational power to extract spatial features.
As shown in Figure 4, regular Conv is applied for spatial feature extraction only on the part of the input channel, without affecting the rest of the channels; for sequential or regular memory access, we consider the first or the last consecutive c_p channel as a representation of the whole feature map used for the computation; the input and the output feature maps have the same number of channels: h × w × k 2 × c p 2 and the ratio of Conv to r c = 0.25, and the ratio of r FLOPs = 0.0625.The PConv memory is only 1/4 of the Conv.The FasterNext block consists of a single PConv layer, which, together with the two Convs, forms an inverted residual block, where the intermediate layer has an extended number of channels, and shortcut connections are placed in order to reuse the input features.Conv layers are placed between the normalization and activation layers to maintain feature diversity and achieve lower latency.
Integration into the Multiscale Attention (EMA) Module
The channel or spatial attention mechanism can be used to achieve good feature representation, but the channel dimensionality reduction it brings will have some side effects when extracting the depth visual representation.Daliang Ouyang et al. [31] proposed a new efficient multiscale attention module that reduces the operations, while preserving the information of each channel.We add the EMA mechanism to the C2f module in the backbone network, and C2f_EMA replaces the original backbone C2f module.
As detailed in Figure 5, the EMA module adopts a parallel structure, which can be divided into three branches, two 1 × 1 and 3 × 3 branches, to extract the attention weight descriptors of the grouped feature maps, where the 1 × 1 branch references the revisit coordinate (CA) [32] module.The CA input tensor is decomposed into two parallel one-dimensional feature encoding vectors, along the horizontal and vertical dimensional directions.Complete set average pooling, using spatial location information, is used to model cross-channel correlations, no dimensionality reduction convolution in 1 × 1; 2D binary distribution on linear convolution is fitted with a Sigmoid function after decomposing the output of the 1 × 1 convolution into two vectors.Only a 3 × 3 kernel is stacked in the 3 × 3 branch to capture multiscale feature representations to expand the feature space.Global spatial information is then encoded in the output of the 1 × 1 branch using group normalization and 2D global pooling operations.The natural nonlinear function Softmax, using 2D Gaussian mapping, is used at the back of the 2D global pooled output.The processed output is then multiplied with the matrix dot product operation after convolution in the 3 × 3 branch to obtain the first spatial attention map.The global spatial information from the 3 × 3 branches is encoded using 2D global average pooling, and the output of the smallest branch will be converted into the corresponding dimensional shape directly before the joint activation mechanism of the channel features, in order to derive the second spatial attention map that preserves the entire precise spatial location information.Finally, the output feature maps within each group are computed as an aggregation of the two generated spatial attention weight values, followed by a Sigmoid function.It captures pixel-level pairwise relationships and highlights the global context of all the pixels.
Figure 6 illustrates the schematic structure of the added small object detection layer in the YOLOv8s architecture.In this structure, we have optimized explicitly for detecting small objects within panoramic images.Traditional feature map sizes may fail to capture smaller target objects, so we introduced a higher-resolution feature map output to preserve more detailed feature information.As shown in Figure 6, a new branch has been added following the second layer C2f, with the area highlighted in red indicating our optimization strategy, introducing a 160 × 160 feature map to the output.This approach allows for more accurate detection of small targets during training.As the network delves into deeper layers, there is a potential risk of losing crucial information about small targets, which could also lead to an increase in the model size and possibly a decrease in the detection speed.We have strategically integrated the multiscale attention (EMA) mechanism to address these challenges at the 2nd, 4th, and 6th C2F layers.Additionally, we replaced the eighth layer C2f with the FasterNext module.These modifications enable the network to focus more effectively on small target information, thereby improving detection accuracy and optimizing computational efficiency.
Experiments 4.1. Panoramic Image Acquisition
There is a growing need for larger, more complex, high resolution, and accurately annotated datasets for object detection in reflective panoramic images [1].To address this need, this study concentrates on small targets within retroreflective panoramic images and has established a dedicated dataset repository specifically for these types of images.The dataset was gathered at a large steel warehouse located in the Tiexi district of Shanghai.Figure 7 displays the physical image of the hyperbolic retroreflective panoramic camera used for data collection.The panoramic camera was mounted on a 20 m tower crane, covering an area of at least 7000 m 2 .Two scenes were captured with five different pixel panoramic images.One set of images had a resolution of 2596 × 1920 pixels, with small targets occupying only 3427 pixels, accounting for only 0.018% of the entire image.The other set of images had a resolution of 1920 × 1080 pixels, with people occupying only 23 × 11 pixels in the image.The retroreflective panoramic image collection process intermittently collected images for one month, from morning to night, capturing one image every half an hour, including in sunny, cloudy, and rainy weather.A total of 4382 panoramic images were obtained with 80 pixels, with people occupying only 23 × 11 pixels in the image.A total of 4382 panoramic images were obtained using the BMP image format.
Figure 8a-h shows images captured at different periods of the same scene, while Figure 8i-l shows panoramic images captured of different scenes.Observing the data, we found three significant problems in retroreflective panoramas: 1. various forms of distortion; 2. discontinuous edge effects; 3. changes in object scale.Therefore, we must improve the object detection algorithm to improve the model's tolerance for distortion, edge effects, and object scale features.In addition to people (small targets), we labeled seven other objects with different scales during the panoramic image labeling process, namely trucks, ships, steel piles, cars, loaded and unloaded truck compartments, and forklifts.Due to the uneven distribution of the collected dataset, there is more data on steel piles and ships, with 10,721 and 7906 objects labeled, respectively.In contrast, the datasets on cars and people are relatively small, with 1062 and 477 objects labeled, respectively.The edges of the panoramic images have significant deformation, and the ships in the dataset are located on the edges of the images and are easily occluded.The size of the people also changes according to their distance from the camera, leading to changes in scale.We randomly selected 3560 images from the 4382 panoramic images as the training set and 866 as the validation set.The distribution of the data labels in the training and validation sets is shown in Figure 9.
Image Preprocessing
Because deep learning requires a large amount of datasets to obtain the feature information of images, the categories of the collected panoramic datasets are unevenly distributed, and there are insufficient small target samples.Therefore, we augmented the existing datasets.Due to the limited data samples featuring humans, we used an improved image copy-paste method [33] to expand the tiny target (human) sample labels to obtain more comprehensive feature information, thereby enhancing detection accuracy.We first extract the small target model (human) through the SAM (Grounded-Segment-Anything) [34] model, randomly rotate it from 0-45 • , and scale it according to the distance from the center point of the panoramic image, randomly generating 716 small targets, and randomly integrating them into different panoramic image training sets.In this way, the training set of small targets is expanded.If too many are copied, this random copying may block the original data, and the overall detection accuracy will decrease.Moreover, the integrated samples are only present in the training set, not the test set.Figure 10a is a schematic diagram of the copy-paste method for panoramic images.The person circled in Figure 10b is integrated into the panoramic image through the copy-paste method, expanding the sample labels of the small targets.To further expand the dataset and enhance the robustness of our model, we leveraged the data augmentation technology built into YOLO [4].Given the spherical nature of panoramic images, we employed a range of image processing techniques to augment the data.These techniques include image rotation, mirroring, HSV space channel enhancement, scaling, multiscale transformation, blending, cropping, and segment copy-paste.These modifications enable the model to better adapt to diverse scenarios.
For a single viewpoint, complete reflection panoramic image, each pixel corresponds to a light ray that intersects the viewpoint.By employing the parameters of the panoramic imaging system, we project these pixels onto a cylindrical surface or a plane, unfolding the image.This unfolded representation articulates 360 • of the horizontal spatial information, which significantly enhances the efficacy of the subsequent processes, such as small target detection and other computational tasks.Our research predominantly utilizes the plane unfolding method for data preprocessing of the complete reflection panoramic image.Figure 11 shows a two-dimensional rectangular, cylindrical panorama from the original image.The center of this panorama is denoted, with R representing the radius.There exists a precise mathematical relationship between the coordinates of a point, M(x,y), on the panoramic image and its corresponding point, M(x ′ ,y ′ ), on the unfolded image.
The radial distance from the center to the point M(x,y) is determined by Equation (3).The radian, which is the projection angle onto the unfolded plane, is calculated by Equation ( 4).
The coordinates of the expanded image point M(x ′ ,y ′ ) are determined by the transformation of Equations ( 5) and ( 6): Equation ( 5) represents the transformation of the x coordinate from the original panoramic image to the unfolded image.The angle formed by the radial line from the center to M(x, y) in the panoramic image concerning the horizontal axis.R is the radius of the cylindrical panorama, which is a constant for a given panoramic system.Where x ′ is the new x coordinate on the unfolded image and is the radial distance from the center to the point M(x, y), as calculated in Equation (6).And y ′ is the new y coordinate on the unfolded image.
Most panoramic video detections rely on unfolded panoramic images.Since the aspect ratio of the image after unfolding is 2πR/(R − r) ≈ 23.14970/(970 − 228) = 8.2, if the training is carried out directly, the image will be compressed into 640 × 640, resulting in a severe imbalance in the aspect ratio and serious image distortion.We cut the image into two parts and stitched them together after the panoramic view was unfolded to minimize distortion due to image compression.To streamline the creation of an unfolded panoramic dataset, we automated labeling objects in the unfolded images.This was achieved by utilizing the mapping relationship between the unfolded and the original panoramic images.This approach enabled us to efficiently produce processed panoramic datasets with minimized distortion and enhanced accuracy.
Training Platforms
This experiment relies on an ASUS TUFz690 model computer (Intel(R)Core(TM) I7-12700k CPU; 3187 Mhz, 32 GB; NVIDIA Geforce RTX3080 GPU, 12 GB video memory), using the Windows 10 system, Python 3.9, CUDA 11.7, the deep learning framework PyTorch 2.0.0, and OpenCV 4.7.0 to implement the folded reflection panoramic small target detection model training and testing in real-time.In this study, an end-to-end approach is used to train the improved YOLOv8s network using stochastic gradient descent (SGD); the batch size of the model training is set to 8, the learning rate lr0/f = 0.01, and the SGD momentum factor is set to 0.937.The data enhancement coefficients for the image left-right flip (probabilistic), image splicing, image obfuscation, and segmented copy-paste (probabilistic) are 0.5, 1.0, 0.1, 0.1, and 0.1; the training batch is set to 300 rounds, and the resulting recognition model weights file is saved after the training is completed, where early abort is set.
Key Evaluation Indicators
In assessing the performance of the enhanced YOLOv8s network, we adopted standard metrics that are essential for evaluating object detection models.As defined by Equation ( 7), precision is the ratio of correctly identified positive detections to the total number of detections made.This metric is crucial in understanding how often our model's predictions are accurate when it claims to have detected a panoramic target.× 100% = TP ALL Detection (7) Here, a true positive (TP) occurs when a predicted panoramic tiny target corresponds accurately to a labeled panoramic target.In contrast, a false positive (FP) occurs when the model incorrectly identifies the background as a panoramic target.
Recall, given by Equation ( 8), measures the proportion of actual panoramic targets the model correctly detects out of all the ground truths.This metric provides insight into the model's ability to find all the relevant instances in the dataset.
Moreover, to encapsulate the overall prediction accuracy across various intersection over union (IOU) thresholds, we calculate the mean average precision (mAP) as formulated by Equation ( 9).This comprehensive metric considers the precision-recall balance across different detection thresholds, providing a holistic view of the model's performance using our panoramic dataset.
In Equation ( 9), q indicates the IOU threshold, N represents the total number of IOU thresholds considered, and Q R is the set of detected targets across all the thresholds.
Effects of Improved Network Modeling
To evaluate the effectiveness of the small target detection layer, the FasterNext module, and the EMA module in detecting small targets in panoramic images, we initially test the panoramic images enhanced through data augmentation, but not those unfolded for stitching.Ablation experiments were performed using the YOLOv8 model, with an input image size of 640 × 640 pixels, targeting eight different types of objects.Table 1 shows the detection results for all these target types, demonstrating the improved model's enhanced performance in detecting small targets across various panoramic scenes.According to Table 1, we have an<EMAIL_ADDRESS>of 87.2% for YOLOv8s using the same set of folded-reflection panorama datasets with the same conditions, and the grid optimization of the number of parameters and the model size for the<EMAIL_ADDRESS>value and the network structure results in a significant improvement.
By replacing the C2f module in layer 8 of the original backbone network with the FasterNext module, PConv replaces Conv due to the addition of FasterNext, reducing the floating-point operations by 1.7GFLOPs, the number of parameters by 1.03 M, and the model size is reduced by 2.1 MB.
With the addition of the small target detection header to the YOLOv8s network, the overall<EMAIL_ADDRESS>value increases by 2.3%, and the<EMAIL_ADDRESS>of the tiny target (human) increases by 17.7% to 53.9%, which effectively captures the features of the tiny target and enhances the detection of the small target.However, due to the addition of the tiny target detection layer, the number of floating-point operations is increased by 0.8GFLOPs compared to the YOLOv8s model, with the consequent increase in the number of floatingpoint operations.Adding a small target detection layer to the YOLOv8_fasterNext network increases the overall accuracy of the<EMAIL_ADDRESS>by 2.9% and the detection accuracy of small targets of the<EMAIL_ADDRESS>by 24.7%.
We incorporate the EMA mechanism into layers 2, 4, and 6 of C2f in the backbone network of YOLOv8.We find that the YOLOv8 network incorporating the multiscale attention EMA mechanism helps to capture cross-dimensional interactions and establish dimensional dependencies efficiently, and that it highlights the global context of all the pixels, and improves the<EMAIL_ADDRESS>of the target detection in panoramic images by 0.4%, and the<EMAIL_ADDRESS>of small target detection goes up by 24.7%.YOLOv8s_small increases the overall<EMAIL_ADDRESS>by 0.5% and the detection rate of small targets (people) by 4.0%, after incorporating the EMA module.
When the YOLOv8s_small_fasterNext_EMA model is compared with the YOLOv8s model, the overall accuracy is improved by 2.9%, in which the<EMAIL_ADDRESS>of the small targets is improved by 20.0%, the model parameter is 1.3 M less, and the size of the model is reduced by 2.4 MB.
We used Grad-CAM as a visualization tool for our network, and the region of most interest to the network is the darker color in the region map.The improved YOLOv8s_small_ fasterNext_EMA network outperforms the original YOLOv8s model under the same lighting and environmental conditions.
Figure 12a-d illustrates the heat maps generated by the original YOLOv8s model, and Figure 12e-h depicts the heat maps from the YOLOv8s_small_fasterNext_EMA model.This side-by-side arrangement of the heat maps allows for a direct comparison between the models regarding the focal intensity on the detected objects.Notably, Figure 12e-h reveals a deeper color intensity within the red-circled regions, signifying a heightened focus of the YOLOv8s_small_fasterNext_EMA model on areas containing small objects.This suggests an enhancement in the model's capacity to detect and home in on smaller objects within panoramic images.When comparing corresponding Figure 12a with Figure 12e,b with Figure 12f,c with Figure 12g,d with Figure 12h, the increased attention to relevant target areas by the improved model becomes evident, highlighting its superior detection performance.The discernible difference in color depth between the original and the enhanced model's heat maps accentuates the progress made in pinpointing and delineating smaller objects amidst complex panoramic scenes.
Comparison with Other Algorithms in the YOLO Series using the Panoramic Dataset
We compared the currently available models and our improved models using the panoramic dataset without data enhancement and any processing, and we can draw the following conclusions from Table 2. Compared with YOLOv8s and YOLOv5s [24,35,36], the overall accuracy<EMAIL_ADDRESS>of YOLOv8s is 1.4% smaller than that of YOLOv5s, the detection accuracy capability related to the small target (person) by YOLOv8s is 16.7% smaller than that of YOLOv5s, and the detection accuracy related to the rest of the categories is better than that of YOLOv5.We improve the model based on the basic model of YOLOv8.The improved model has the most significant improvement in the small target (person) accuracy compared to the YOLOv5s model, which is 21.2%, and the performance related to all the rest of the categories is better than YOLOv5s.The overall detection accuracy is improved by 1.7%.
While YOLOv7 [37,38] is recognized for its improved detection of small targets in standard image datasets, this advantage is less pronounced in panoramic datasets, particularly when comparing models of equal size.This distinction is reflected in Table 2, where YOLOv7's performance in detecting the 'person' category is notably less effective than expected.It is crucial to highlight that such comparisons are context dependent, and the efficacy of a model can vary significantly with different dataset characteristics.Consequently, we have focused on refining the YOLOv8s model to suit the intricacies of panoramic image analysis better.The enhancements have resulted in a model that is not only more lightweight, but also demonstrates increased accuracy in detecting small targets within panoramic scenes, surpassing the performance of the original YOLOv8s model.
Panoramic Copy-Paste Data-Enhanced Ablation Experiments
For cases with insufficient samples of small targets, we introduced the copy-paste method for panoramic images to expand the labels of small targets.Comparison experiments were conducted under the same conditions.
After we made the panoramic dataset, we prepared some small target object pools according to their distance from the center of the panoramic image, scaled them, and randomly rotated them; we randomly chose the objects from the object pools and randomly copied the small targets in the images in the training set, and experimented on the improved model uniformly, as shown in Figure 13.After the data was subject to the copy-paste method, the model<EMAIL_ADDRESS>went up by 0.6%, the recall went up by 1.1%, and the detection precision was improved.Table 3 shows that the precision of detecting small targets in the panorama image after the improved model and copy-paste data enhancement is still lower than 60%.The small target dataset are not easy to collect, and it takes time to make the dataset.To further improve the accuracy of the small target model detection, we unfold the panoramic image in the same dataset and then bisect and splice it; we achieve the unfolded bisect and spliced image through the mapping relationship between the panoramic image and the unfolded image.The resulting dataset has a better detection effect compared to the unprocessed one, and we have conducted experiments in regard to this.The experimental results are shown in Table 3.The panoramic copy-paste expand is the dataset resulting from the panoramic unfolded bisection stitching.Based on the data in Table 3 and Figure 14, comparing the final results, it can be found that the<EMAIL_ADDRESS>of the processed dataset compared to the panoramic copy-paste expand dataset rises by 4.2% for all target mAP@0.5,21.3% for the<EMAIL_ADDRESS>of the small targets, and 27.3% for the recall of the small targets.The box_loss is also reduced by 0.0916.
Visualization and Verification
To verify the visual effect of the algorithms in this paper using large-field panoramic images, we collected some panoramic images mounted at a height of 20 m overlooking the ground scene for testing.Figure 15 illustrates the detections using the improved model in four panoramas labeled a, b, c, and d, where eight types of objects are detected.Notably, the small target people in scenes a and b are successfully detected.However, in Figure 16, which displays two small targets in scene c, only one is detected, and the single small target in scene d is missed.Figure 16 also reveals the detection results using the unimproved YOLOv8s model.The red circles highlight objects that are either missed or incorrectly detected, demonstrating that all small target individuals in scenes b, c, and d remain undetected by the unimproved model.
Figures 17-19 present a side-by-side analysis of the same scene captured under different weather conditions, allowing for an assessment of how varying lighting and environmental factors impact detection efficacy.Figure 17 depicts the scene on a cloudy day, demonstrating the robustness of the copy-paste data enhancement method in conditions with diffuse lighting.Figure 18 showcases the scene in sunny conditions, where the stark shadows and bright light typically pose challenges for detection algorithms.However, our method continues to identify each individual accurately.Lastly, Figure 19 captures the scene amidst rainfall, where the reduced visibility and reflective surfaces from wet conditions are present, illustrating that our preprocessing technique significantly enhances the accuracy of small target detection across all weather scenarios.The comparative results show that processed panoramic images, adjusted for each specific weather condition, provide more reliable detection outcomes than those achieved by direct detection using the original panoramic images.
Conclusions
In this paper, for the study of small target detection tasks in wide-area panoramic images, an algorithm YOLOv8s_small_fasterNext_EMA is proposed for small target detection in panoramic images, based on the YOLOv8 network.
In order to resolve the insufficient ability of YOLOv8s in regard to small target detection, a small target detection layer is added to the network to improve the network's ability to perceive small targets.The EMA mechanism is incorporated into the YOLOv8s backbone network for feature enhancement, suppressing ineffective information in the input features and focusing more on the features of small targets, and the depth of the network is deepened in the network structure, incorporating the attention mechanism and the small-in-small target layer.The parameters are increased and the floating-point arithmetic increases; for this reason, we changed the eighth layer in the YOLOv8s network to the FasterNext module and replaced part of the traditional convolutional CONV with PCONV at the same time, without affecting the detection accuracy, to improve the efficiency of the detection, compared to the YOLOv8s network which is more lightweight.The algorithm in this paper has apparent advantages in regard to wide-area panoramic datasets, and the value of<EMAIL_ADDRESS>is significantly improved compared with the standard model of YOLOv8s and YOLOv7, and the YOLOv5s model with the same network width and depth.The ablation studies on the improved model design reveal that the components of this design significantly enhance detection accuracy, reduce the model's size, and increase the speed of detection.The visualization experiments show that the wide-area panoramic image of various scenarios resulted in good detection for small targets.However, there is still room for improvement in the presence of occlusion concerning the situation panoramic image detection effect.
Furthermore, this paper introduces a copy-paste data enhancement method for panoramic images, addressing the scarcity of small target sample labels.This method improves the detection of small targets, enhancing both recall and precision.Instead of directly detecting targets in unfolded panoramas, we crop and splice them, resulting in better detection compared to the original panoramic images.
Figure 1 .
Figure 1.The general training and generalization process for panoramic image small target detection involves four steps: 1⃝ the acquired images are annotated using the LabelImg tool and then cropped to conform to the YOLOv8 model's input format;2 ⃝ data augmentation is utilized to address sample imbalance and to expand the dataset after panoramic expansion;3 ⃝ these processed images are used to train the YOLOv8 detector;4 ⃝ the trained model is deployed to detect small targets in panoramic images, producing the final visualization of the detection results.
Figure 2 .
Figure 2. The network structure of YOLOv8s and the CSPLayer_2Conv (C2f) structure in the YOLOv8 network, and the detection header structure in YOLOv8.
Figure 3 .
Figure 3. Schematic structure of the small object detection layer added to YOLOv8s.
Figure 7 .
Figure 7. Physical image of the hyperbolic refractive panoramic camera, where reflector 1 and reflector 2 are shown on the upper and lower surfaces.
Figure 8 .
Figure 8. Images captured by a refractive panoramic camera of different scenes under varying light and weather conditions.(a) Clear day with moderate lighting, (b) Sunny day with strong shadows, (c) Overcast conditions, (d) Overcast evening, (e) Morning with fog and dew, (f) Foggy condition with artificial lighting, (g) Overcast with diffuse illumination, (h) Bright sunny day, (i) Overcast day with heavy traffic, (j) Sunny day with reflective surfaces, (k) Rainy conditions with wet surfaces, (l) Green outdoor setting with natural light.
Figure 9 .
Figure 9. Number of tags per category in the panorama dataset.
Figure 10 .
Figure 10.Schematic of the panoramic copy-paste data enhancement approach.(a) Schematic of copy-paste method in panoramic view; (b) the result of the copy-paste model in the panorama is realistic.
Figure 13 .
Figure 13.(a) Shows the<EMAIL_ADDRESS>comparison between the improved model before and after copy-paste data enhancement, while (b) details the corresponding recall comparison.These metrics effectively demonstrate the enhancements in model performance post-data augmentation.
Figure 14 .
Figure 14.(a-d) Shows the comparison between the panoramic image dataset and the panoramic processed dataset detecting various evaluation metrics, respectively.
Figures 15 and 16
Figures 15 and 16 present our tests on panoramic images across various scenes.Figure15illustrates the detections using the improved model in four panoramas labeled a, b, c, and d, where eight types of objects are detected.Notably, the small target people in scenes a and b are successfully detected.However, in Figure16, which displays two small targets in scene c, only one is detected, and the single small target in scene d is missed.Figure16also reveals the detection results using the unimproved YOLOv8s model.The red circles highlight objects that are either missed or incorrectly detected, demonstrating that all small target individuals in scenes b, c, and d remain undetected by the unimproved model.Figures17-19present a side-by-side analysis of the same scene captured under different weather conditions, allowing for an assessment of how varying lighting and environmental factors impact detection efficacy.Figure17depicts the scene on a cloudy day, demonstrating the robustness of the copy-paste data enhancement method in conditions with diffuse lighting.Figure18showcases the scene in sunny conditions, where the stark shadows and bright light typically pose challenges for detection algorithms.However, our method continues to identify each individual accurately.Lastly, Figure19captures the scene amidst rainfall, where the reduced visibility and reflective surfaces from wet conditions are present, illustrating that our preprocessing technique significantly enhances the accuracy of small target detection across all weather scenarios.The comparative results show that processed panoramic images, adjusted for each specific weather condition, provide more reliable detection outcomes than those achieved by direct detection using the original panoramic images.
Figure 15 .
Figure 15.(a-d) Shows the detection results for the improved model using panoramic images of different scenes.
Figure 16 .
Figure 16.(a-d) Shows the detection results for the YOLOv8s model using panoramic images of different scenes.
Figure 17 .
Figure 17.Detection results for the improved model using a preprocessed panoramic image (cloudy).
Figure 18 .
Figure 18.Detection results for the improved model using a preprocessed panoramic image (sunny).
Figure 19 .
Figure 19.Detection results for the improved model using a preprocessed panoramic image (rainy).
Table 1 .
Improved network for ablation experiments.
Table 2 .
Comparison of the performance of other YOLO models and YOLOv8s_ours using the panorama dataset, where a person is the smallest target (tested on RTX3080).
Table 3 .
Comparison between target detection results for our dataset before and after using data processing. | 9,966.6 | 2024-01-26T00:00:00.000 | [
"Computer Science",
"Engineering"
] |
Precipitation Hardening in Ferroelectric Ceramics
Domain wall motion in ferroics, similar to dislocation motion in metals, can be tuned by well‐concepted microstructural elements. In demanding high‐power applications of piezoelectric materials, the domain wall motion is considered as a lossy hysteretic mechanism that should be restricted. Current applications for so‐called hard piezoelectrics are abundant and hinge on the use of an acceptor‐doping scheme. However, this mechanism features severe limitations due to enhanced mobility of oxygen vacancies at moderate temperatures. By analogy with metal technology, the authors present here a new solution for electroceramics, where precipitates are utilized to pin domain walls and improve piezoelectric properties. Through a sequence of sintering, nucleation, and precipitate growth, intragranular precipitates leading to a fine domain structure are developed as shown by transmission electron microscopy, piezoresponse force microscopy, and phase‐field simulation. This structure impedes the domain wall motion as elucidated by electromechanical characterization. As a result, the mechanical quality factor is increased by ≈50% and the hysteresis in electrostrain is suppressed considerably. This is even achieved with slightly increased piezoelectric coefficient and electromechanical coupling factor. This novel process can be smoothly implemented in industrial production processes and is accessible to simple laboratory experimentation for microstructure optimization and implementation in various ferroelectric systems.
Introduction
Piezoelectricity is an important feature of poled ferroelectrics, which enables conversion between electrical and mechanical signals. Ferroelectric materials are therefore widely utilized in actuators, transducers, sensors, etc. [1] With the development of technology, ferroelectric materials are becoming increasingly important in novel and highly demanding application fields, for example, photon-electronic communication [2,3] and energy storage. [4] The macroscopic electromechanical response of piezoelectrics relies on an interplay of intrinsic and extrinsic contributions, where the intrinsic effect utilizes reversible lattice extension/contraction and the extrinsic effect is facilitated by irreversible, hysteretic domain wall motion and possibly phase transition. [5] High-power applications in ultrasonic motors, transducers, and transformers demand so-called hard ferroelectrics with low energy losses. [6] Therefore, a stringent reduction of all lossy electrical and mechanical mechanisms, in particular an effective ferroelectric domain wall immobilization, is required. The state-of-the-art concept for hardening of ferroelectrics relies on doping with acceptor elements. [7,8] The effectiveness of this approach is quantified by the mechanical quality factor Q m , which is the reciprocal of mechanical loss and places a stark requirement on resonance applications. [1,9,10] The mechanism of acceptor doping relies on oxygen vacancies, which become mobile at moderate temperature with the consequence that the market-dominating material, lead zirconate titanate (PZT), heats up under high vibration velocity [8] and loses 50% of its electromechanical quality factor already at a moderate usage temperature of 79 °C. [11] This effect decisively limits the operational range of piezoceramics.
Ferroelectric hardening as a process to reduce hysteretic movement of domain walls (2D carriers of deformation) is suggested to bear strong resemblance to hardening of metals, where the mobility of dislocations (1D carriers of deformation) is reduced by multidimensional defects. Hardening or strengthening of metals is achieved by point defects (0D defects), dislocations (1D defects), grain boundaries (2D defects), and secondary phases (3D defects in form of precipitates or added secondary phases). [12] Precipitation hardening in metals is particularly appealing as it affords high homogeneity and efficient industry-scale processing, both for applications as structural materials [13][14][15] and as ferromagnetic materials. [16][17][18] Domain wall motion in ferroics, similar to dislocation motion in metals, can be tuned by well-concepted microstructural elements. In demanding high-power applications of piezoelectric materials, the domain wall motion is considered as a lossy hysteretic mechanism that should be restricted. Current applications for so-called hard piezoelectrics are abundant and hinge on the use of an acceptordoping scheme. However, this mechanism features severe limitations due to enhanced mobility of oxygen vacancies at moderate temperatures. By analogy with metal technology, the authors present here a new solution for electroceramics, where precipitates are utilized to pin domain walls and improve piezoelectric properties. Through a sequence of sintering, nucleation, and precipitate growth, intragranular precipitates leading to a fine domain structure are developed as shown by transmission electron microscopy, piezoresponse force microscopy, and phase-field simulation. This structure impedes the domain wall motion as elucidated by electromechanical characterization. As a result, the mechanical quality factor is increased by ≈50% and the hysteresis in electrostrain is suppressed considerably. This is even achieved with slightly increased piezoelectric coefficient and electromechanical coupling factor. This novel process can be smoothly implemented in industrial production processes and is accessible to simple laboratory experimentation for microstructure optimization and implementation in various ferroelectric systems.
Recently, secondary phase ferroelectric hardening was demonstrated in Na 1/2 Bi 1/2 TiO 3 -based (NBT-based) piezoceramics by forming 0-3 type composites of 0.94Na 1/2 Bi 1/2 TiO 3 -0.06BaTiO 3 (NBT-6BT) matrix and added ZnO grains, located at the grain boundaries. [19][20][21][22][23] A lower dielectric loss and a nearly twofold increase in Q m were found in a NBT-6BT:0.1ZnO composite. [19] The hardening mechanism has recently been rationalized through a mechanical interaction between the secondary-phase particles and the matrix. [19,21] Riemer et al. pointed out that the difference in the thermal expansion coefficient of ZnO and NBT-BT grains induces deviatoric stresses in the matrix, which stabilize the ferroelectric phase. [21] The hardening effect through this composite approach was also observed in other piezoelectric systems such as 0.83(Na 1/2 Bi 1/2 ) TiO 3 -0.17(K 1/2 Bi 1/2 )TiO 3 , [22] Bi 3 TaTiO 9 :40 wt%BiFeO 3 , [24] and 0.2Pb(Zn 1/3 Nb 2/3 )O 3 -0.8Pb(Zr 0.5 Ti 0.5 )O 3 . [25] An apparent shortcoming of the NBT-BT:ZnO composites lies in the limited flexibility of tuning microstructure. Due to the nature of the composite processing, the ZnO grains are predominantly located at grain boundaries and triple junctions, which limit their interaction with ferroelectric domains located inside the matrix grains. Inspired by precipitation hardening in metals, we hypothesize that precipitation can be used as a means to homogeneously distribute secondary-phase particles into ferroelectric grains, in order to alter the domain structure, pin domain walls, and suppress their motion. Despite the extensive investigations on precipitation hardening on metals, the studies on ceramic materials are quite limited. One successful case of precipitate-tuned ceramic material is Mg partially stabilized ZrO 2 (Mg-PSZ), in which the fracture toughness has a fourfold increase and achieves 10 MPa m 1/2 . [26] Besides, this concept has also been previously applied in mere exploratory fashion to Al 2 O 3 -Fe 2 O 3 [27] and MgO-Cr 2 O 3 [28] solid solutions, both focused on tuning the mechanical properties. To the best of our knowledge, precipitation has not been utilized in electroceramics so far.
Here, we demonstrate that secondary-phase precipitates can be applied to tune the domain size and to pin domain walls, which effectively hardens the electromechanical response-a mechanism hereafter referred to as "precipitation hardening." The concept is demonstrated using the model pseudo-binary system BaTiO 3 -CaTiO 3 , (barium calcium titanate, BCT), with a curved line of solid solubility-a precondition for precipitation in solid solutions. Non-ferroelectric CaTiO 3 -rich precipitates were successfully introduced in the ferroelectric BaTiO 3 -rich matrix. Through various structural and microstructural characterizations and electrical property measurements, it was found that the precipitates have an influence on their vicinal domain structures and suppress domain switching during the application of electric field, leading to lower saturated polarization, strain, permittivity at a poled state, and a higher mechanical quality factor.
After drying, calcination was performed at 1100 °C for 4 h. To ensure complete chemical reaction, powders were then crushed and ball milled again at 250 r min −1 for 12 h and calcined for a second time with the same condition as the first calcination. Then, the twice-calcined powders were cold-isostatically pressed into pellets with a dimension of ≈Ø 10 mm × 1 mm under a hydrostatic pressure of 357 MPa. The pellets were then sintered at 1500 °C, which is in the single-phase region of the BCT phase diagram (Figure 1a), for 8 h using a tube furnace. When sintering was completed, the samples were air-quenched, that is, they were directly taken out of the tube furnace from 1500 °C to room temperature, to kinetically suppress the formation of the Ca-rich secondary phase. Those as-quenched samples (unaged samples) were denoted as S u . For the aging treatment, the asquenched samples were annealed at 1200 °C for 72 h and then cooled with 5 K min −1 and were denoted as S o . Some of the 1200 °C-annealed samples were further annealed at 1300 °C for 24 h and were denoted as S t .
Microstructure Characterization
X-ray diffraction (XRD) measurements were conducted using a laboratory XRD (Bruker D8 Advanced, Germany) with Cu Kα radiation. Bulk samples were ground with a 1200-mesh sand paper followed by annealing at 300 °C for 2 h before investigation. Bragg-Brentano geometry was adopted. The two-theta range from 10° to 90° with a step size of 0.02° was considered. The scanning electron microscopy (SEM) images were taken by (XL30FEG, Philips, Amsterdam, Netherlands). Samples were ground and polished with diamond polishing paste down to 0.25 μm particle size before the measurement. The back-scattered electron (BSE) detector was used to distinguish between Ba-rich and Ca-rich phases. A relatively low electron energy of 8 keV was selected to visualize grain boundaries by the contrast difference of grains with different orientations.
The transmission electron microscopy (TEM) images and energy dispersive spectroscopy (EDS) mappings were taken by a JEM-2100F TEM. To prepare the TEM sample, ceramic pellet was first ground to a thickness of 250 μm. Then, a disk with 3 mm in diameter was prepared by ultrasonic cutting for further polishing. Both the top and bottom surfaces of the disk were polished using diamond lapping films with grain size of 9, 6, 3, 1, and 0.25 μm in turn, to gradually reduce the thickness of the disk down to 20 μm. The polished 3 mm-disk was then annealed at 400 °C for 0.5 h with both slow heating and cooling rate of 1 °C min −1 to release the accumulated strain during polishing. Afterwards, the disk was glued to a supporting molybdenum grid and finally thinned to obtain electron transparent areas by ion milling (Gatan Model 600 dual ion mill).
Piezoresponse force microscopy (PFM) data was recorded using a NT-MDT (NTEGRA, Apeldoorn, The Netherlands) atomic force microscope. A conductive Ti/Ir coated tip (Asyelec.01-R2, Oxford Instruments, USA) was used for scanning in contact mode. For domain imaging, a sinusoidal alternating current excitation voltage of 10 V was applied to the back electrode at a frequency of 40.13 kHz. The deflection of the laser signal was read out as the amplitude, R, and the phase, ϑ, of the piezoresponse using a lock-in amplifier (SR830, Stanford Research Systems, USA). Spatial resolution of Rcosϑ enabled to qualitatively distinguish domains with different orientation.
Phase-Field Simulation
Phase-field simulations were performed on a 2D system consisting of a circular CaTiO 3 particle with a diameter of 130 nm inside a BaTiO 3 matrix. The ferroelectric domain structure was simulated by solving the time-dependent Ginzburg-Landau equation [29] for the evolution of the polarization field P(x), that is, A periodic boundary condition was employed. The free energy, F, was formulated as the sum of the Landau free energy, the electrostatic free energy, the elastic energy, and the gradient energy, that is, F = F Landau + F electrostatic + F elastic + F gradient . The Landau free energy was given by F aP a P P a P P P a P P P P x where a i , a ij , a ijk , and a ijkl were the Landau coefficients. An Einstein summation convention of automatic summation over repeated indices i, j, k, l = 1, 2, 3 was employed herein. The electrostatic energy was written as where κ 0 was the vacuum permittivity, κ b was the background dielectric constant, and E(x) was the electric field, which was obtained by solving the electrostatic equilibrium equation with a periodic boundary condition. The elastic energy was given by Here, C was the elastic stiffness tensor, ε(x) was the strain field, and ε 0 (x) was the eigenstrain field given by ij 0 ijkl k l Q P P ε = , with Q being the electrostrictive coefficient. The strain field was obtained by solving the elastic equilibrium equation where σ (x) was the stress field. A periodic boundary condition with a zero homogeneous stress was employed. The gradient energy was expressed as where g was the gradient energy coefficient. The material constants of BaTiO 3 used in the present work, including the Landau coefficients, the background dielectric constant, the elastic stiffness, the electrostrictive coefficient, and [37] BT ss and CT ss represent the Ba-rich and Ca-rich solid solution phases, respectively. b) Temperature profile of the precipitate-formation process: sintering, quenching, and aging, where T s , T a , and T r represent the sintering temperature, aging temperature, and room temperature, respectively, and t s and t a represent the sintering duration and aging duration, respectively. Schematics of the microstructures after quenching and aging are depicted in the insets. c) XRD patterns of the as-quenched and of the sample aged at 1325 °C for 8 h. The reflections arising from the CT ss precipitates are marked by asterisks. d,e) SEM images of unaged (S u ) and two-stage aged (S t ) samples. Dark features in (d) were identified as triple-point pores, while dark grey areas in (e) were identified to be the CT ss precipitates (see the purple arrows). the gradient energy coefficient are listed in Table 1. The CaTiO 3 precipitate was considered as a dielectric particle with a dielectric constant of κ r = 168 [30] and a corresponding Landau coefficient given by a 1 = (2κ 0 κ r ) −1 = 3.36 × 10 8 J m C −2 ; higher-order Landau coefficients were neglected. The background dielectric constant, the elastic stiffness, the electrostrictive coefficient, and the gradient energy coefficient of CaTiO 3 were taken the same as those of BaTiO 3 for simplicity. The system was discretized into square meshes with a mesh size of 0.8 nm × 0.8 nm for the numerical simulation.
Electrical Characterization
The samples were made in disk shape with a dimension of ≈Ø 8.00 mm × 0.50 mm. Both top and bottom surfaces were fully covered with Pt electrodes. For the measurements of unpoled samples, the samples were annealed at 300 °C for 2 h to release the polarized state and then were kept at room temperature for 24 h before the measurements. Poling was conducted at 4 kV mm −1 for 15 min at room temperature, followed by a 24 h period before investigation. Polarization and strain hysteresis loops were obtained by a modified Sawyer-Tower circuit and an optical displacement sensor (D63, Philtec Inc., USA). A triangular electric field with a maximum 2 kV mm −1 and frequency of 1 Hz for bipolar measurements and 2 Hz for unipolar measurements (to ensure the same ramping rate of the applied field) was applied. The permittivity frequency spectra were recorder by a broadband dielectric analyzer (Novocontrol, Germany). A sinusoidal alternating current signal with peak-to-peak voltage V pp of 2.88 V was applied with frequency ranging from 10 −1 to 10 5 Hz. The longitudinal piezoelectric coefficient d 33 was determined using a commercial Berlincourt meter (Piezotest PM300, Singapore), with a static clamping force of 2 N, dynamic driving force of 0.25 N, and driving frequency of 110 Hz. Planar coupling factor, k p , and mechanical quality factor, Q m , were quantified using the impedance spectrum as function of frequency near the resonance frequency. The impedance spectra were also obtained by the broadband dielectric analyzer (Novocontrol, Germany) and the driving voltage was set as V pp = 0.288 V. Q m was obtained by the following equation: [36] 1 2 m r a 1kHz eff where f r and f a were the resonance and antiresonance frequencies, respectively; C 1kHz was the free capacitance of the sample at 1 kHz, which was far away from the resonance frequency; and k eff was the effective electromechanical coupling factor, which could be obtained by: The planar electromechanical coupling factor, k p , could be determined graphically from the relationship between k p /k eff and k eff . [36] All the electromechanical parameters (d 33 , k p , and Q m ) were measured on three samples for each aging condition, and the error bars in Figure 2c denote the standard deviation of the measured values for the same aging condition.
Results and Discussion
The equilibrium pseudo-binary phase diagram of the (1−x) BaTiO 3 -xCaTiO 3 system (x represents the weight percentage of CaTiO 3 ) (Figure 1a) [37] indicates a strong temperature dependence in solubility of CaTiO 3 between 1200 °C and 1500 °C. A subsolidus line divides the phase diagram into two regions: single-phase region with a Ba-rich solid solution (BT ss ) and two-phase region with the coexistence of the BT ss phase and a Ca-rich solid solution (CT ss ). The processing of precipitation-hardened materials can be divided into three stages: sintering, quenching, and aging ( Figure 1b). The sintering process achieves densification within the single-phase region with homogeneous distribution of elements. A quenching process follows in order to kinetically hinder the uncontrolled formation of the thermodynamically-stable CT ss phase. This results in a supersaturated solid solution, which is metastable at room temperature. [14,38] The temperature for the aging process [39] in the two-phase region is chosen based on both, thermodynamic as well as kinetic considerations.
The ideal microstructure for precipitation hardening is characterized by a high density of precipitates inside the grain while the size of them is in a range from several tens of nanometers to hundreds of nanometers, so that a large fraction of domain walls can be effectively hindered. According to the theory of diffusional phase transitions in solids, a reduced aging temperature leads to a larger driving force for precipitation, since the difference between the solubility and the actual solute concentration is larger. This avails the nucleation process. On the other hand, a higher temperature facilitates precipitate growth as atomic diffusion is more significant. The nucleation rate and the growth rate of precipitates can be expressed by the following equations, [38] Adv. Mater. 2021, 33,2102421 where N hom is the number density of the homogeneous nuclei; ΔG m is the activation energy for atomic migration; ΔG * is the nucleation energy barrier, which in general decreases with decreasing temperature; k is the Boltzmann constant and T is the temperature in K; r is the mean radius of the precipitates; X 0 , X e , and X b represent the solute concentrations in the matrix, equilibrium state, and precipitates, respectively; D is the interdiffusion coefficient and t is the duration of aging. The relationship between nucleation/growth rate and aging temperature is schematically depicted in Figure S1, Supporting Information. In addition, the aging temperature also has an influence on the nucleation on different sites. The energy barrier of heterogeneous nucleation, for example, nucleation at grain boundaries and dislocations, is reduced by these defects, and is usually lower than that of homogeneous nucleation (i.e., nucleation at defect-free sites within grains). The homogeneous/heterogeneous nucleation ratio can be expressed by: [38] · exp het hom 1 0 hom h et where N het is the number density of the heterogeneous nuclei; C 0 and C 1 represent the number density of the sites for homogeneous and heterogeneous nucleation, respectively; G het ∆ * and G hom ∆ * represent the nucleation energy barrier for homogeneous and heterogeneous nucleation, respectively. A reduction in aging temperature can lead to a smaller difference between G het ∆ * and G hom ∆ * and therefore lower heterogeneous/homogeneous nucleation ratio.
According to the abovementioned theory, the aging temperature plays an important role in the precipitate formation. An examination of this theory is provided in Figure S2, Supporting Information, where the number density, mean size, total amount, and size distribution of the precipitates in 80 wt%BaTiO 3 -20 wt%CaTiO 3 (BCT20) highlight the strong impact of aging temperature. The relevance of nucleation versus precipitate growth was assessed in BCT20 using three different conditions: I) as-quenched sample without any other heat treatment (unaged sample, S u ); II) sample aged at 1200 °C for 72 h (one-stage aged sample, S o ); III) sample aged at 1200 °C for 72 h and then at 1300 °C for 24 h (two-stage aged sample, S t ). The purpose of the two-stage aging is to first increase the number of the precipitates within grains at the lower temperature and then to grow the precipitates at a higher temperature. The detailed temperature profile of the S u , S o , and S t samples is displayed in Figure S3, Supporting Information. Figure 1c depicts the XRD patterns of the unaged sample S u and an aged sample. In order to clearly indicate the reflections of the secondary phase, an aging condition with higher aging temperature (1325 °C for 8 h, denoted by S o ′) is selected for this comparison, which led to the highest secondary phase amount. The comparison of XRD patterns of the S u and S t samples are depicted in Figure S4, Supporting Information. The star symbols mark the reflections arising from the CT ss phase, which is absent in the unaged sample and present in the aged sample. This confirms that the single-phase state is achieved after quenching (within detection limits) and the CT ss phase is successfully formed during the aging process. The {110} PC reflections of the BT ss phase shift to lower angles after aging, indicating an expansion of the matrix lattice. This can be attributed to the lower Ca content in the BT ss phase due to the formation of the Ca-rich phase ( Figure S4, Supporting Information). Since Ca 2+ has a smaller ionic radius than Ba 2+ , a larger lattice constant for the BT ss with lower Ca content can be rationalized, as also evidenced in ref. [40]. Figure 1d,e provides the SEM images in BSE mode of the S u and the S t samples, respectively. Larger areas are represented in Figure S5, Supporting Information. The CT ss particles can be visualized with dark grey contrast in the SEM images of the aged sample. The size of the CT ss particles varies from submicron to a few microns. Both the XRD and SEM results suggest that a homogeneous single phase has been achieved in the unaged sample and the precipitates with CT ss phase emerge thereafter. Figure 3 presents a focus on local chemistry and microstructural details around specific precipitates. A bright-field TEM image of an intragranular precipitate and the corresponding EDS mapping of Ca and Ba are featured in Figure 3a-c. Enhancement of the Ca concentration and Ba deficiency can be observed in the precipitate, compared to the surrounding matrix, which confirms that the precipitate is of CT ss phase. The electron diffraction (SAED) pattern along the [111] PC zone axis of a selected area in the precipitate (Figure 3d) exhibits strong dominant reflections accompanied with weak superlattice reflections, as indicated by the arrows and circles, respectively. The strong reflections confirm that CT ss has a perovskite structure. The weak superlattice reflections present at ½(101), ½(110), and ½(011) indicate the existence of in-phase tilting in the CT ss precipitate. [41,42] Dark field TEM images of the CT ss precipitate obtained using ½(101), ½(110), and ½(011) superlattice reflections are displayed in Figure S6, Supporting Information.
Adv. Mater. 2021, 33, Several structural features related to precipitates are highlighted in TEM images in Figure 3e-g: I) Change of the local domain structure: the domain pattern terminates in wedge-shaped ends in the vicinity of the precipitate (Figure 3e,f ). Regions with high domain wall density (i.e., finer domain structure) are formed near precipitates ( Figure 3e); II) Emergence of dislocations: dislocation loops emerge near the precipitate (Figure 3f ), [43] and some dislocations are found at the precipitate/matrix interface (Figure 3g). For the fine domains near the precipitates, a similar phenomenon was observed near the grain boundaries in polycrystalline PZT and was related to increased microstrain. [44] A decrease in domain size adjacent to Ag intragranular nanoparticles has also been observed in Pb(Zn 1/3 Nb 2/3 ) 0.20 (Zr 0.50 Ti 0.50 ) 0.80 O 3 /6 vol% Ag composites, [45] while ZrO 2 inclusions have been reported to introduce internal stresses and microcracks in PZT matrix, which inhibited domain wall movement. [46] Similarly, the regions with fine domains can be attributed to the misfit strain at the precipitate/matrix interface, which arises from the difference in lattice parameters, spontaneous strain, and thermal expansion coefficients of the CT ss and BT ss phases. [47][48][49] The dislocation loops appearing in the vicinity of the precipitates may be attributed to local Ca deficiency, since Ca ions have been depleted in the matrix in order to form the precipitates.
Complementary PFM images of the S t sample are presented in Figure 3h,i. Precipitates were identified in PFM by the absence of a piezoelectric response. Analogous to the TEM measurements, an enhanced density of domain walls is resolved in the vicinity of a precipitate (red circles in Figure 3e) and the termination of domains at the precipitate/matrix interface (red squares in Figure 3f) can be observed in the PFM images. Please note that the domain structure also depends on the grain orientation and viewing direction, therefore a fine domain structure is not observed for all precipitates using this 2D imaging methodology.
The fine domain structure near a precipitate was further evidenced by phase-field simulation, as highlighted in Figure 3j. Similar to the experimental observation, the precipitates alter the local structure with concurrent local refinement of the domain pattern. The simulation also indicates that the domain refinement is caused by a relaxation of the local electrostatic free energy. The spatial electrostatic free energy distributions in the refined domains near the precipitate is depicted in Figure S7b, Supporting Information, as contrasted to the case without domain refinement in Figure S7a, Supporting Information, obtained by first simulating the domain structure in a pure BaTiO 3 and then adding a CaTiO 3 precipitate. The electrostatic free energy density is extremely high (≈4 MJ m −3 ) around the precipitate/matrix interface without the fine domain structure due to the net bound charges at the interface, while it is substantially reduced by the formation of the fine domain structure. The average electrostatic free energy densities without and with the fine domain structure are 0.24 and 0.19 MJ m −3 , respectively. The simulated domain configuration in a larger region around a precipitate is provided in Figure S7c, Supporting Information. We approximated the area of the fine-domain region induced by the presence of a precipitate with a diameter of 130 nm to be 4 μm 2 , consistent with that observed in the PFM measurement.
The introduction of precipitates inevitably alters the static domain structure of the ferroelectric ceramic. This strong modification is expected to impact in a similar fashion the dynamic properties, which are highlighted next. The largesignal properties of the S u , S o , and S t samples are characterized by the polarization and strain hysteresis loops. The bipolar polarization and strain hysteresis loops of the S u , S o , and S t samples (Figure 4a,b) are featured next to the unipolar strain hysteresis loops in Figure 4c. It can be found that the saturated polarization, remanent polarization, and maximum strain under both bipolar and unipolar electric fields are consistently decreased from unaged (S u ) to one-stage aged (S o ) to two-stage aged (S t ) samples. For the aged samples, two factors should be considered regarding the mechanisms of their macroscopic property change. On the one hand is the compositional change in the matrix phase (intrinsic effect). The matrix phase dominates the dielectric, ferroelectric, and piezoelectric properties of the composite rather than the precipitate phase, since the matrix usually represents a volume fraction in excess of 90% and the precipitates are not ferroelectric. On the other hand, the effect of the precipitates on the domain wall movement (extrinsic effect) should also be addressed.
In order to quantify the effect of the matrix with reduced content of Ca due to precipitation, specimens with varying Ca content in the composition range between 16 and 20 wt% CaTiO 3 ( Figure S8, Supporting Information) were prepared. Results for polarization and strain hysteresis loops for the respective single-phase BCT solid solution demonstrate that saturated polarization and maximum strain increase with decreasing Ca content. The S t sample has the largest amount of precipitates, hence the lowest Ca content in the matrix. However, it possesses the lowest remanent polarization and strain, indicating that the reduction of polarization and strain is not due to the compositional change in the matrix, but is triggered by the influence of the precipitates.
The bipolar strain hysteresis loops (Figure 4b) allow quantification of negative strain as a signature for non-180° domain wall motion. [50] The negative strain in the S t sample is the smallest. As the negative strain is consistently reduced with increasing aging treatment, this measure suggests that non-180° domain wall motions are suppressed in the aged samples.
The electromechanical loss can be quantified from unipolar loops in Figure 4c by assessing the normalized hysteresis area, A h , of the unipolar strain loops. [51] This quantity is obtained by normalizing the strain loops with their maximum strain values and then calculating the closed area of the loops. The normalized hysteresis areas follow the trend of S u > S o > S t (Figure 4d), indicating that aged samples have lower electromechanical loss under a low-frequency, large-signal electric field.
The empirical squareness parameter, R s , describes the shape of the loops and estimates both switching as well as backswitching characteristics: [52,53] where P r , P s , and P 1.1Ec denote the remanent polarization, saturated polarization, and the polarization at 1.1 times of the coercive field, respectively. The R s values of different samples are reduced with increasing aging treatment (Figure 4d). The backswitching appears to be affected to a smaller degree. The more slanted curve of the aged samples is suggested to arise from a wider distribution of local switching electric fields, which is due to the broad spatial distribution of precipitates where the domain walls are pinned. The microstrain can act as a restoring force for switched domains and the high-domain-wall-density regions affected by the microstrain can have higher switching fields, which was evidenced in PZT samples by a previous PFM study. [44] In contrast to the properties featured in Figure 4d, aging treatment has only little influence on the coercive field, E c , as outlined in Table S1, Supporting Information. The small-signal properties of the S u , S o , and S t samples are displayed in Figure 2. The real part of the relative permittivity ε′ r as a function of frequency at room temperature is revealed in Figure 2a. For the unpoled state, the aged samples exhibit a slight enhancement in relative permittivity over the whole measured frequency range. Moreover, a significant increase in the low-frequency relative permittivity can be observed in the S t sample. For the poled state, in addition to the increase at low frequency, the high-frequency relative permittivity decreases with increasing aging degree (i.e., S u > S o > S t ).
In general, the increase in the low-frequency relative permittivity can be attributed to two mechanisms: one is an increase in conductivity of the material and the other is the Maxwell-Wagner effect (i.e., interfacial polarization), which is related to an increase in the number of interfaces between phases with different conductivity. [8] For the former mechanism, the conductivity increase in ferroelectric ceramics is usually related to an introduction of charged defects, which can act as additional mobile charge carriers. In order to examine whether the enhanced low-frequency relative permittivity is due to the longterm aging process, a supplemental experiment was conducted: the aged sample was heated to 1500 °C and kept for 8 h followed by air quenching (requenched sample). The ε′ r -f spectra of the unaged, aged, and requenched samples are contrasted in Figure S9, Supporting Information. It can be found that the ε′ r -f behavior of the requenched sample is mostly reversed to the unaged state, suggesting that the long-term aging treatment does not contribute to noticed salient effects in tuning the properties. Therefore, it is demonstrated that the rapid increase in the low-frequency permittivity is dominantly attributed to the Maxwell-Wagner effect (space charge polarization) arising from the precipitates. The high-frequency permittivities (kHz-MHz range) can be contributed by both the intrinsic (lattice response) and extrinsic (domain wall motion) effects. The ε′ r -f spectra of unaged and unpoled BCT samples with different Ca content are depicted in Figure S10, Supporting Information, from which it can be seen that the compositional change has negligible influence on the ε′ r -f behavior. Thus, it is reasonable that the high-frequency permittivities of the S u , S o , and S t samples are comparable and the differences are less than 7%. In addition, a difference in the dielectric response of the unpoled and poled states can be noticed from the permittivity frequency spectra in the high-frequency range. The aging treatment has led to a decrease in the high-frequency permittivity at poled state, while this influence is negligible at unpoled state. | 7,851.2 | 2021-07-24T00:00:00.000 | [
"Materials Science"
] |
Best Time Domain Features for Early Detection of Faults in Rotary Machines Using RAT and ANN
Bearing failure, the most frequent failure mode in rotating machinery is the typical mechanical fault. Such a failure might result in substantial financial losses at the workplace. One of the approaches made possible by other signal processing techniques is the early identification of various faults in rotating machinery; including bearing failures, misalignment, and others. This fault is associated with many features used to diagnose different faults; thus, the Diagnostic Features (DF) is estimated at limited cyclic frequencies that refer to machine faults. Two methods are used to extract the DF. The first one depends on time-domain features. The second is based on an advanced representation of the frequency domain, which depends on spectral coherence (SCoh) data over the spectral frequency domain using a center frequency and frequency range determined by a 1/3 binary tree structure. The calculated DFs are represented by a 2D map against the center frequency and frequency resolution. The maps from different fault features are collected to form the diagnostic patterns. The best characteristics connected to these various flaws can be found using statistical techniques like reverse arrangement tests (RAT). Artificial neural networks (ANN) may be trained and auto-diagnosed using the results from the best characteristics. Using RAT is considered very important to summarize features. This method is given good results in training and diagnosis. Additionally, ANN and RAT provide a detection result of 100% based on the description of the machine's operating situation, whether it functioned commonly or incorrectly.
Introduction
Features were employed in early failure detection techniques in the recent time domain. Depending on how well these elements can identify problems, some temporal domain (TD) characteristics are derived TAKÁCS [1]. By estimating the degradation trend in features over time with a cumulative approach, Kosasih et al. [2] were able to determine how the slewing bearing was degrading. An autoregressive model was employed by James and Walter [3]. The retrieved characteristics are established to acquire a deterioration trend; the vibration signals could be changed when faults occur in rotating machinery Liu et al. [4]. The TD signals' amplitude and distribution may differ from the standard conditions. It is feasible to assess whether rotating machine damage is occurring using the TD statistical characteristics, according to Sreejith et al. [5]. The time and frequency domain can be used in feature extraction techniques by Chen et al. [6]. Bansal et al. [7] has presented a study comparing the performance of these three feature extraction methods (SIFT, SURF, and ORB), particularly when combined to recognize an object. The authors presented a comparative study of various feature descriptor algorithms and classification models for 2D object recognition. Bansal et al. [8] has extended their work to Combining in-depth features extracted using 1 3 VGG19 with handcrafted feature extraction methods, such as SIFT, SURF, ORB, and Shi-Tomasi corner detector algorithm, improving image classification performance.
Tsang [9] revealed the best way to discover the trend in data via a review. These data can be delivered separately and similarly. Additionally, the null hypothesis is promised these data. By integrating convolutional neural networks (CNNs) with variational mode decomposition (VMD) algorithms, processing raw vibration data directly without artificial intelligence or manual labor allowed Xu et al. [10] to complete the end-to-end defect detection of rolling bearings. Zhang et al. and Zhao et al. [11,12]. They demonstrated how to leverage the CM and FDD to identify defects and failures in components. Beck et al. [13] review the statistical test procedures for determining the stationary of the surface electromyography (EMG) signal. A rat is used to symbolize one of these statistical techniques. Watson et al. [14] use rat to test the randomization hypothesis for each data set. Further, Murray et al. [15] also introduced rat for pattern recognition issues. They showed that the rat performs better than the support vector machine. In this work, we will put all of our efforts into using the rat approach to estimate the trend in features and identify the gradients in the data caused by a specific machine issue. Next, this cutting-edge technique finds and isolates spinning machine flaws.
The best machine fault detection techniques continue to be learning techniques like ANN. Using machine learning techniques, Attaran [16] looked into the issue of automated bearing defect diagnosis. These techniques focus on the TD statistical characteristic and encompass feature extraction, selection, and classification. Vibration signature analysis also uses different development methodologies. Sar et al. [17] introduced a review of current methods for developing vibration signatures. This paper aims to use a numerical method to extract features issued with features in either a time or frequency domain depending on transforming signals to image representation, which is used to detect rotating machinery faults.
Time Domain Features
The following TD characteristics may be extracted from a raw vibration signal:
Mean Value
In most cases, vibration analysis does not make much use of a signal's mean value. However, it showed how the vibration signal behaved.
Root Mean Square (RMS)
This function performs well in monitoring the signal's total vibration intensity Sar et al. [17]. Calculating the RMS value from the formula by Cuc et al. [18]:
Standard Deviation, SD
It is a measurement of the mean's dispersion. This is how the standard deviation is described:
Kurtosis
It indicates the extent of the distribution tail and identifies the most prominent peaks in the data set. Kurtosis is given by Cho et al. [19], Bhende et al. [20]:
Skewness
When studying dynamic signals, the term skewness (SK) is frequently utilized; SK is defined by Cho et al. [16], The value of x is increased to the third power because the SK reading is more sensitive to asymmetry in high x v readings than the mean.
Peak Value
It is unquestionably a sign of component degeneration The PV is defined by Bhende et al. [20] and Benkedjouh et al. [21] as:
Shape Factor
The ratio of the RMS to the mean value is known as the shape factor (SF). It illustrates adjustments under misalignment and imbalance Benkedjouh et al. [21].
Impulse Factor
The ratio of the peak value to the mean value of the time signal is known as the impulse factor (IMF), which is also a bearing defect indication Bhende et al. [20].
Clearance Factor (CF)
Another time domain characteristic that Bhende et al. [20] can estimate is this one,
Crest Factor
An impact estimate in a waveform can be calculated quickly using the CF formula. Wear-on gear teeth, cavitations, and roller bearings are frequently linked to impacting. One of the most important characteristics used to trend machine conditions is the CF Bendat and Peirsol [22]. CF is given below by Cho et al. [19]: x(i) is a signal series for i = 1, 2… N, and N is the number of data points.
Signal Processing
Signal processing involves the processing, amplification, and interpretation of signals. Raw vibration signal always contains contamination components ("noise") and frequently some basic components that may partially obscure other components that are the important part of the measurements taken. Several choices can have the noise or other uninterested signal parts removed. One is the time domain signal averaging to attenuate the unsynchronized signal. The other option is to pass the raw signal through a filter and analyze the signal in the time or frequency domain. Another powerful indicator commonly used to measure the strength of correlation between the signal and its time-shifted version is the spectral coherence given in Eq. (11) Antoni [23]: where S 2X (f , ) is the spectral correlation function that represents the power distribution of the signal concerning its spectral frequency f, which can estimate from: Moreover, cyclic frequency α and X(f) represent the Fourier transform of the signal blocks. The SC function can be calculated using several techniques, such as the FFT accumulation method (FAM) Roberts et al. [24], averaged cyclic periodogram (ACP) Antoni [25], or fast spectral correlation (FSC), Antoni et al. [26], which can lead to producing the image representation form, as shown in Fig. 1?
Processing of Image using SIFT
The resulting image from spectral coherence analysis now deals with Scale Invariant Feature Transform (SIFT), a feature detection algorithm. SIFT also assists in finding the local features in an image, usually known as the 'key points of the image. These key points are scale & rotation invariants used for numerous pc vision applications. They can also be utilized in the fault designation method, which may be used in this text in machine fault diagnosis supported the image ensuing from reworking the time domain signal to image matching. We can also use the key points generated using SIFT as features for the image during model training. The significant advantage of SIFT features, over edge features or hog features, is that they are not affected by the size or orientation of the image (Fig. 2).
RAT in Features Trend and Selection
The machine's healthy state is changed according to the features. As a result, changes in time-domain properties are frequently used to audit bearing deterioration. The incipient of the defect is found to be more accurate based on the extracted features than the retrieved characteristics from the original vibration signal Kosasih et al. [2]. Defect bearing displays a gradual upward trend and increases in variance over healthy bearing, and its magnitude is larger than healthy bearing Watson et al. [14]. In this part can be used a formalist statistical test. The test is excellent at detecting monotonic (i.e., gradual and continuing) trends in data. The existence of these trends also shows nonrandomness. The existence of these trends also indicates non-randomness. The created test, a well-known accurate randomness test, was given by Bendat and Peirsol [22]. The rat considers a hypothesis test that looks for trends in the observed parameter Bendat and Peirsol [22]. Following is the test procedure: First, consider the sampled time sequence of signal where N is the number of samples in each measurement. There are N segments in the series. The parameters are calculated using these segments. To examine their trend, several metrics are employed. The signal segment might be identified as. According to what was stated in part before. This technique tests the TD parameters that were previously described in the section that represented those parameters. The anticipated value owing to sample variations is then determined by computing a new function, h ij , as Bendat and Peirsol [22] suggest. This is done to test the sequence of integers y i associated with each parameter for changes outside: Calculating A i factor for each number as: The variable Several reverse arrangements can be estimated by: The two relevant equations Brandt [27] and Tsang [9] provide the mean and variance of an N-set of independent observations of a stationary random variable: Under this test's null hypothesis, the signal data points are shown as independent observations from random variables Beck et al. [13]. However, the alternate or renewal hypothesis demonstrated a relationship between the data points that make up the signal and an underlying pattern in a series of observations. If A falls inside Bendat and Peirsol's [22] range, the stationery hypothesis is accepted at the significance level of = 0.05: The hypothesis is rejected at the α level of significance if the observed runs are outside the interval.
Artificial Neural Network
ANNs attempt to emulate their biological counterparts. ANNs have been developed in parallel distributed network models based on the human brain's biological learning process of the human brain Murray et al. [15]. Different methods can be used to deal with data in ANN applications. Multilayer receptor (MLP) neural networks are employed in this work since they are regarded as universal by the many forms of artificial neural networks Fig. 3 depicts the traditional three-layer feed-forward NN design.
For a network with N input nodes, H hidden nodes, and M output nodes, the mapping from the input vector (I1,.…, IN) to the output vector (O1, ….., ON) is given by: where q = 1,…, M, V jq is the weight from the hidden node (j) to the output node (q), and (g) is the activation function. The reading of hidden layer node h j is given by: w ij is the input weight, b j is the threshold weight, and σ is the activation function which was chosen as the sigmoid function; that is very popular because it is monotonous, bounded, and has a simple derivative. Three steps comprise the general back propagation training: feed-forward of the input training pattern, back propagation of the related error, and weight modification.
In neural networks, sigmoid activation functions are commonly used. As it is monotonic and differentiable, each is needed for the back-propagation algorithm. Karlik and Olgac [28] give the equation for sigmoid functions: where s i refers to the weighted sum.
Fault Detection Architecture
The suggested technique is depicted in a flow chart in Fig. 4. Its foundation is the integration of data from several sensors. Under two circumstances, the signals are taken from MFS. The first represents the machine's normal functioning situation, while the second is its broken state. A data capture system is used to transfer this data. The measured signal from four sensors is used to extract TD characteristics. Four places have these sensors installed (vertical and horizontal inboard and vertical and horizontal outboard). Then, rat examines the trend connected to each defect for the characteristics derived from the four sensors. The best feature is designated as one that affects any responsibility, while the rejected feature is designated as one with no impact on fault change. To train the network, the best characteristics are sent into the system. Following the training process, each fault type's associated characteristics are categorized. Then, it is looked at how well the features pass the test and can detect errors. This technique is employed to find faults. The characteristics are diverse in their influence on each type of fault; hence all TD domain features should be analyzed to discover the optimum features connected with each fault.
Experimental Work
The suggested approach is validated using the Machinery Fault Simulator (MFS). method. 5. Study has been done on the bearing problems. These faults include ball and bearing wear and exterior and inner race faults. First-hand information about high-quality inboard and outboard rotor bearings may be gathered using the MFS. The device has to be correctly positioned. One at a time, a faulty bearing will be installed, and data will be gathered. The objective is to contrast the outcomes of suitable bearings with bad bearings. At 10 kHz, data is gathered. At 2700 RPM, the machine revolves (Fig. 5).
SIFT
The ensuing image from Spectral Coherence analysis for roller bearing of rotating machine as shown in Fig. 6 currently upset Scale Invariant Feature remodel (SIFT) as established in Fig. 6 to introduce SIFT operator to extract fault features from the recursive graphs. Following that, the most compelling feature of this fault is identified in rat for coaching or determining the fault; another case study involves outer race bearings, shown in Fig. 7 improves the image with entirely different features compared to either a traditional or faulty state.
As mentioned above, another case study on ball-bearing faults, as shown in Fig. 8, represents the Spectral Coherence of bearing faults. The work now yields the SIFT Fig. 9, representing the critical points generated using SIFT as features for the image during model training.
Then these produce the image that represents the key features associated with the traditional or faulty state.
The basic steps of the SIFT image registration algorithm are as follows: (1) Feature point extraction, (2) Generating feature descriptors, and (3) feature point matching. The feature point extraction mainly includes generating a difference of Gaussian (DOG) scale space, finding local extremum points, screening feature points, and determining the direction of feature points Wang et al. [29] RAT The best characteristics required to define the various bearing defects have been completed. The fault's most important features are selected to capture a data set at a rotating speed of 2700 RPM. Time-domain characteristics are determined for 10 reading sections for each rotation speed in the four directions (vertical Inboard VI, horizontal inboard HI, vertical outboard VO, and horizontal outboard HO). 4096 samples are taken per second at a sampling rate of 40,960 samples per reading. These data' rat is computed, as shown in Table 1. The rat indicator's acceptance values are AC, and its unacceptance values are UA. Time-domain Factors were shown to be less than ideal as a ball-bearing failure indicator, according to the results of rat. For this reason, it is essential to look for a new strategy to remedy this problem.
Nevertheless, would like to emphasize that Kurtosis, Shape, and Skewness still have an impact on fault identification since they provide indicators for condition prediction and fault diagnosis. Additionally, the horizontal and vertical directions, ranked by priority, are more effective in detecting ball faults. Kurt, form factor, and vertical order skewness were the more valuable criteria in identifying the inner race defect. These characteristics include kurtosis, skewness, impulsivity, and crest factors in the horizontal direction. As indicated in Table 2, these characteristics are assessed on the outboard or cage of the defective bearing, while the inboard cage continues to relate to the vibration signal. Here, we demonstrated how these characteristics might serve as an intelligent system's inputs. As a result of applying the rat method to identify the outer race fault, Table 3 confirmed the characteristics of Kurt, form factor, impulsiveness, and skewness on each vertical and horizontal axis, except for that within the horizontal path, the characteristics of the far crest component dominated so that the horizontal direction emerged as a result of the appropriate method for identifying the outer race fault. It is possible to deduce the rat compound fault table by looking at the three tables for each bearing fault.
Based on the vibration information of compound faults, as shown in Table 4, the rat of it determines the ruled path to be used to detect compound faults; Thus, the functions connected to this direction became RMS, well-known deviations, Kurt, shape, impulsive, clearance, and crest factors, as well as the form factor in the vertical course, which should be examined in the intelligent machine to enhance fault detection capabilities. One may infer from the analysis of bearing faults that the fault impacts both inboard and outboard supports, which is evident from the tables. The study confined flaws, and the effect of a bearing problem did not cross over to other supports. This information will help with fault diagnostics.
ANN
The characteristics obtained from time domains are employed as input nodes in the input layer of the conventional three-layer NN. When the data corresponds to a normal operating condition (NOC), one node in the output layers has a value set to zero, while the output value refers to the defective operation situation (FOC). The primary software used to determine the NN weights is graphically written. The best NN design is [20-20-1], as shown in Table 5. The mean square error of the training technique is lowered by utilizing a varied number of nodes in the hidden layer (NHN), starting in training from 10 to 23 NHN. All of the characteristics that were retrieved from the time domain should be used since they have varied associations with various defect types. So that we may use those traits previously discovered and associated with various failures, the fault detection procedure cANNot make assumptions about a particular type of defect. To obtain disclosure problems for different types from a wide variety of data that could have specific deficiencies, the fault detection technique must be employed collectively. Identifying the rotating machine condition is beneficial even though the neural network training capacity may, at most, not reach 95%. As stated in Table 6, the NN approach for fault diagnosis is employed for validation with either NOC or FOC. Depending on the number of nodes in the hidden layer, their values may be evaluated using FOC using various examples. The machine being tested is shown in the test case and in Fig. 10a, which is a training program component. The second Fig. 10b is for the scenario in which the machine breaks down. This program component, which will be used later to identify and forecast the kind of problem in rotary machines, is installed during the training process and tests a variety of data. It is part of the training process and not the finished program. This is a step in verifying the program's ability to find errors and install the appropriate fixes, ensuring that it will function correctly during the fault detection process. The likelihood of identifying distinct defects utilizing this portion of the program has demonstrated encouraging results and achieved 100% for specific errors. James and Walter assess the error's value [3]:
Comparison of Actual Value and ANN
The neural network operation prediction method outperformed the overall defect prognosis approach. The neural network is effectively used in differentiating defects because the difference between the real values and the neural network is relatively tiny. Figure 13 depicts the progression in data deterioration for bearings from the desired operating condition to a severe operation index for this bearing. These data cover the time period from 23/10 to 2/12, with a variety of bearings operating at 2700 RPM. The loading condition is applied by screw tightening until a deep state is attained, in accordance with the ISO 10816-3 Vibration Severity Chart.
Conclusions
This work uses a novel method to detect the trend in features used in the bearing fault detection process. rat method has been influential in the extraction well and the necessary features associated with each failure independently and given sufficient accuracy in distinguishing different types of defects. The TD features are chosen to identify the different types of faults in the rotating machine. The rat determines the relationship between fault and feature by examining the data trend for those alternatives. By analyzing the data associated with each failure, the rat technique has demonstrated that this data can be used to discover flaws. The machine's deterioration has been located. The rat discovered that not all fault kinds share the same criteria for identification. In other words, as indicated in all tables, each defect has specific characteristics. The best and most important approach for fault detection is considered to be ANN. The statement of machine operation status is detected by ANN with a detection rate of 100%, whether it is operating normally or not. Springer Nature or its licensor holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law. | 5,328.4 | 2022-09-02T00:00:00.000 | [
"Engineering",
"Computer Science"
] |
Gauge coupling unification in a classically scale invariant model
There are a lot of works within a class of classically scale invariant model, which is motivated by solving the gauge hierarchy problem. In this context, the Higgs mass vanishes at the UV scale due to the classically scale invariance, and is generated via the Coleman-Weinberg mechanism. Since the mass generation should occur not so far from the electroweak scale, we extend the standard model only around the TeV scale. We construct a model which can achieve the gauge coupling unification at the UV scale. In the same way, the model can realize the vacuum stability, smallness of active neutrino masses, baryon asymmetry of the universe, and dark matter relic abundance. The model predicts the existence vector-like fermions charged under SU(3)C with masses lower than 1 TeV, and the SM singlet Majorana dark matter with mass lower than 2.6 TeV.
Introduction
The Higgs mass parameter m 2 h is only a dimensionful parameter in the standard model (SM), and its value is estimated by the observed Higgs mass as −2m 2 h = M h = 125.09 ± 0.21 (stat.) ± 0.11 (syst.) GeV [1]. Then, a running of the Higgs quartic coupling becomes negative below the Planck scale within the SM. If the SM can be valid up to a high energy scale such as a breaking scale of a gauge symmetry in the grand unification theory (GUT), the electroweak (EW) scale should be stabilized against radiative corrections coming from the high energy physics. To solve the gauge hierarchy problem, there are a lot of works motivated by a classically scale invariance [2]- [29]. The scale invariance prohibits dimensionful parameters at a classical level, while it can be radiatively broken by the Coleman-Weinberg (CW) mechanism [30]. In addition to the classically scale invariance, with an additional U(1) X gauge symmetry, e.g., U(1) B−L gauge symmetry, it is possible to naturally realize experimentally observed values of the Higgs mass. When the U(1) X symmetry is broken by the CW mechanism, the EW symmetry could be also broken through the scalar mixing term. If the U(1) X breaking scale is not far from the EW scale, the Higgs mass corrections would be sufficiently small, and then the hierarchy problem can be solved. Note that these statements are based on the Bardeen's argument [31], and we consider only logarithmic divergences in this paper (see ref. [7] for more detailed discussions).
In this paper, we assume the classically scale invariance at the UV scale, where the SM gauge couplings are unified. We expect that some unknown mechanism, such as a string theory, realizes the classically scale invariance and the gauge coupling unification (GCU). Actually, the GCU can be realized at 3 × 10 16 GeV in our model, and the scale is near the typical string scale (∼ 10 17 GeV). To realize the GCU, some additional particles with the JHEP02(2016)058 SM gauge charges are needed. Conditions of the GCU can be systematically obtained by an analysis of renormalization group equations (RGEs) [32,33]. When all additional particles are vector-like fermions with the TeV scale masses, the GCU scale can be realized between 10 16 GeV and 10 17 GeV, and there are a lot of possibilities to realize the GCU at the scale. 1 For example, vector-like pairs of quark doublet Q L,R and down-type quark singlet D L,R can achieve the GCU [34,35]. When there are additional fermions charged under the SM gauge symmetries, the gauge couplings and the top Yukawa coupling respectively become larger and smaller compared to the SM case, and then, both changes make the β function of the Higgs quartic coupling become larger. Therefore, the vacuum can become stable when the GCU is realized.
To solve the gauge hierarchy problem, there should be no intermediate scale between the EW and the GCU scales except an energy scale, which is not so far from the EW scale, i.e., the TeV scale. Then, phenomenological and cosmological problems (e.g., smallness of active neutrino masses, baryon asymmetry of the universe, and dark matter (DM)) should be explained with sufficiently small Higgs mass corrections. The first two problems can be explained by the right-handed neutrinos, which are naturally introduced to cancel the anomalies accompanied with the U(1) X gauge symmetry, via type-I seesaw mechanism [36][37][38][39][40] and resonant leptogenesis [41], respectively. In our model, the DM is identified with the SM singlet Majorana fermions, and its stability can be guaranteed by an additional Z 2 symmetry [42]. In this paper, we will show that our model can explain the above problems as well as realizing the GCU without affecting the hierarchy problem. 2 In the next section, we will define our model, and explain the U(1) X gauge symmetry breaking as well as the EW symmetry breaking via the CW mechanism. We also obtain the upper bound on the U(1) X breaking scale from the naturalness. In section 3, we will discuss the GCU, vacuum stability, smallness of active neutrino masses, baryon asymmetry of the universe, and the DM relic abundance. Our model predicts the existence vector-like fermions charged under SU(3) C with masses lower than 1 TeV, and the SM singlet Majorana dark matter with mass lower than 2.6 TeV. We summarize our results in section 4.
Symmetry breaking mechanism
We consider the U(1) X gauge extension of the SM with three generations of the righthanded neutrinos ν R i (i = 1, 2, 3), six vector-like fermions (Q L , Q R , D L , D R , N L , and N R ), and two SM singlet scalars (Φ and S). Charge assignments of the particles are shown in table 1. The U(1) X charge are given by B − L + 2x H Y , where x H , B, L, and Y denote a 1 For example, we can consider the origin of the vector-like fermions as the string theory, in which a number of vector-like fermions should appear above the compact scale, which is expected to be the GCU scale in our model. Some of them might have the TeV scale masses due to the fine-tuning of moduli (or Wilson line, extra-dimensional component of anti-symmetric tensor field, and so on).
2 From theoretical point of view, there are some papers constructing a model which realizes classically scale invariance and gauge coupling unification at the same scale [43]- [45]. Furthermore, asymptotic safety of gravity [46] leads vanishing couplings at the UV scale, which suggests vanishing quartic couplings and gauge coupling unification around the Planck scale [see figure 1 in ref. [47] for example]. In this paper, we simply expect such a situation comes from unknown UV physics. real number, the baryon and lepton numbers, and the U(1) Y hypercharge, respectively. In particular, x H = 0, −1 and −2/5 correspond to U(1) B−L , U(1) R and U(1) χ , respectively. The vector-like fermions Q L,R , D L,R , and N L,R respectively have the same charges as the SM quark doublet, the SM down-quark singlet, and the right-handed neutrino, while only the vector-like fermions are odd under an additional Z 2 symmetry. Four of the vector-like fermions (Q L,R and D L,R ) play a role for achieving the GCU, and the others (N L,R ) are the DM candidates, whose stability is guaranteed by the Z 2 symmetry. These particles are not necessary for the realization of GCU and DM. We choose them for the simplest extension. The relevant Lagrangian is given by
JHEP02(2016)058
where L SM is the SM Lagrangian except for the Higgs sector, L kinetic includes kinetic terms of the Higgs and new particles, and V (H, Φ, S) is a scalar potential of the model. Without the Z 2 symmetry, there are also additional Yukawa interactions between the SM particles and the new particles, e.g., y 1 Q L H c u R , y 2 Q L Hd R , and y 3 q L HD R . However, these coupling constants have to be very small due to constraints from the precision electroweak data [48].
To forbid these terms, we have imposed odd parity to only the vector-like fermions under the Z 2 symmetry. Since there are two U(1) gauge symmetry, U(1) kinetic mixing generally arises in the model. We can take covariant derivative as
JHEP02(2016)058
where g's are gauge couplings, T α and T a are generators of SU(3) C and SU(2) L , respectively, and V µ (V = G α , W a , B, Z ) are gauge bosons. The coupling constant g mix denotes the kinetic mixing between the U(1) Y and the U(1) X gauge symmetries, and we will take g mix = 0 at the GCU scale. This boundary condition naturally arises from breaking a simple unified gauge group into SU We impose the classically scale invariance at the GCU scale, and hence, the scalar potential V (H, Φ, S) is given by where there is no dimensionful parameter. In the model, a complex scalar singlet Φ spontaneously breaks the U(1) X gauge symmetry due to radiative corrections, i.e. the CW mechanism. Since the complex scalar field obtains the nonzero vacuum expectation value (VEV), the SM singlet scalar Φ, the U(1) X gauge boson Z , the right-handed neutrinos and the vector-like fermion N L,R become massive. After the U(1) X symmetry breaking, negative mass terms of a real scalar singlet S and the SM Higgs doublet H are generated, which induces the EW symmetry breaking. Then, S, the vector-like fermions and the SM particles become massive, and typically their masses are lighter than those obtained by the U(1) X symmetry breaking.
Let us explain the symmetry breaking mechanism more explicitly. We consider the CW potential for a classical field of the singlet scalar φ as where we have taken Φ = φ/ √ 2 without loss of generality, and φ = v Φ is the VEV of φ. β functions of Φ, β λ Φ , almost depends on quartic terms of g X , Y M and Y N L,R for λ Φ 0. (β functions of the model parameters are given in appendix.) The effective potential (2.4) satisfies the following renormalization conditions 5) and the minimization condition of V Φ induces where we have assumed that the scalar quartic couplings are negligibly small in the righthand side. When this relation is satisfied, the U(1) X symmetry is broken, and Φ and Z become massive as respectively. Since the right-hand side of eq. (2.6) should be positive, is required, and hence, M φ < M Z is generally expected. In addition, the quartic terms of Majorana Yukawa couplings (Y M and Y N L,R ) are smaller than the quartic terms of g X JHEP02(2016)058 The masses of right-handed neutrinos and N L,R will be discussed in section 3.3.
After the U(1) X symmetry breaking, the effective potentials for s and h are approximately given by Here, we have assumed that λ HS are negligibly small compared to λ ΦS and λ HΦ for simplicity. For κ 1,2 0, λ HS is always negligibly small during renormalization group evolution [see eq. (A. 16)]. When λ ΦS and λ HΦ are negative, the nonzero VEVs s = v S and h = v H are obtained as Note that v S and v H is typically lower than v Φ , because the ratios of quartic couplings (λ ΦS /(2λ S ) and λ HΦ /(2λ H )) should be lower than unity to avoid the vacuum instability. The vector-like fermions and the SM particles become massive, while the masses of vectorlike fermions (Q L,R and D L,R ) have to be lower than 1 TeV to realize the GCU as we will show in section 3.1.
In the end of this section, we mention the U(1) X breaking scale, which is described by v Φ . Since M Z /g X > 6.9 TeV is required from the LEP-II experiments [49], we obtain the lower bound v Φ 3.5 TeV. On the other hand, the naturalness of the Higgs mass suggests a relatively small v Φ . A major correction to the Higgs mass is given by Z intermediating diagrams, and one-loop and two-loop corrections are approximately written as where we have taken y t ≈ 1. For |x H | < 0.1, the two-loop correction gives stronger bound than one-loop correction. In the following, we will use the stronger bound for fixed x H . Note that the mass correction from Φ is always negligible because of a small mixing coupling λ HΦ .
Phenomenological and cosmological aspects
In this section, we will discuss phenomenological and cosmological aspects of the model: the GCU, vacuum stability and triviality, smallness of active neutrino masses, baryon JHEP02(2016)058 asymmetry of the universe, and dark matter. We will also restrict the model parameters from the naturalness of the Higgs mass.
Gauge coupling unification
First, we discuss the possibility of the GCU at a high energy scale. Since four additional vector-like fermions (Q L,R and D L,R ) have gauge charges under the SM gauge groups as shown in table 1, runnings of the SM gauge couplings are modified from the SM. Then, β functions of gauge coupling constants are given by at 1-loop level. Figure 1 shows runnings of gauge couplings α −1 i ≡ 4π/g 2 i , where U(1) Y gauge coupling is normalized as g 1 ≡ 5/3g Y . The calculation has been done for x H = 0 with using 2-loop RGEs. We note that the running of gauge couplings are almost independent of x H . In the figure, the horizontal axis is the renormalization scale and the vertical axis indicates value of α −1 i . The red, green, and blue lines show α −1 1 , α −1 2 , and α −1 3 , respectively. The dashed and solid lines correspond to the SM and our model, respectively. The left vertical line stands for a typical scale of vector-like fermions, which has been taken as M V = 800 GeV in figure 1. For µ < M V , the β functions are the SM ones, and we take boundary conditions for the gauge couplings such that experimental values of the Weinberg angle, the fine structure constant, and the strong coupling can be reproduced [50]. The GCU can be achieved at Λ GCU = (2-4) × 10 16 GeV, and the unified gauge coupling is α are added into the SM. As the vector-like fermion masses become larger, the precision of the GCU becomes worse. Thus, the masses of Q L,R and D L,R should be lighter than 1 TeV, while vector-like fermion masses are constrained by the LHC experiments [53][54][55]. Since the lower bound of vector-like quark lies around 700 GeV, the possibility of the GCU can be testable in the near future. We note that the proton lifetime in a GUT model. The proton lifetime is roughly derived from a four-fermion approximation for the decay channel p → e + + π 0 , which is given by where m p is the proton mass. For Λ GCU = 3 × 10 16 GeV and α −1 GCU = 35.6, we can estimate τ p ∼ 10 37 yrs, which is much longer than the experimental lower bound τ p > 8.2 × 10 33 yrs [56]. Thus, the model are free from the constraint of the proton decay.
Vacuum stability and triviality
Next, we discuss the vacuum stability. However, it is difficult to investigate exact vacuum stability conditions, since there are three scalar fields and each of them has nonzero VEVs. Therefore, we simply investigate three necessary conditions: λ H > 0, λ Φ > 0 and λ S > 0.
The condition λ H > 0 depends on additional contributions to β λ H , i.e., κ 1,2 , g X and scalar mixing couplings. 4 If their contributions to β λ H are negligible, since the SM gauge couplings are larger compared to the SM case, running of λ H is raised and always positive. figure 2, where β λ H is independent of g X up to the one-loop level, and contributions of g X can be negligible. The red and blue lines correspond to κ = 0 and κ = 0.33, respectively. The black dashed line shows running of λ H in the SM. Thus, κ < 0.33 is required to realize the vacuum stability. The Higgs mass corrections from Q L,R and D L,R loops are given by where we have taken κ = κ 1 = κ 2 , which naturally arises from L ↔ R symmetry for the vector-like particles, and 2) for simplicity. Then, the naturalness requires κ < 0.1 for M V ∼ 1 TeV. Although κv H is a contribution to the vector-like fermion masses from the Higgs, it can be ignored because of κv H M V . Since the contribution of κ to β λ H , i.e., 24λ H κ 2 − 12κ 4 , is always positive for κ < 0.1, the naturalness condition also guarantees the vacuum stability. Note that κ 0 guarantees λ HS 0 at any energy scale, which is required to justify our potential analysis for eq. (2.8).
Here, we check contributions of vector-like fermions to the S and T parameters, which are approximately given by [58,59] δS ≈ 43 30π where θ W and M W are the Weinberg angle and the W boson mass, respectively. For κ < 0.1, the parameters are estimated as δS < 3 × 10 −4 and δT < 2 × 10 −5 , which are consistent with the precision EW data S = 0.00 ± 0.08 and T = 0.05 ± 0.07 [56]. The condition λ Φ > 0 is almost always satisfied when g X is dominant in the right-hand side of eq. (2.6), i.e., λ Φ (v Φ ) ∼ g 4 X (v Φ ). In this case, β λ Φ is positive up to the GCU scale, and then λ Φ is also positive up to the GCU scale. It is also possible to realize the critical condition λ Φ (Λ GCU ) = 0 as well as λ Φ > 0, where the running of λ Φ is curved upward as in the so-called flatland scenario [9,14,16,21,24]. Then, both g X and Majorana Yukawa couplings are dominant in β λ Φ , while λ Φ is much smaller than them. This means that there is a fine-tuning to satisfy eq. (2.6).
When λ S is negligible in its β function, a solution of its RGE is approximately given by where µ is a renormalization scale. Once v S is fixed, f Q and f D are determined to realize the GCU, while f N remains a free parameter. To estimate the condition of λ S > 0, we assume f N = f Q = f D at µ = v S for simplicity. Then, we can find that λ S is positive up to the GCU scale for λ S (v S ) 0.01. This lower bound of λ S (v S ) is almost unchanged for different values of v S , because v S dependence is logarithmic. On the other hand, when λ S is dominant in β λ S , the Landau pole might exist, at which the theory is not valid from the point of view of perturbativity (triviality). The energy scale where the Landau pole appears is approximately estimated as where M s = 2λ S (v S )v S is a mass of the real singlet scalar field. Figure 3 shows v S dependence on the upper (red) and lower (blue) bonds of M s , which correspond to the Landau pole and vacuum stability conditions, respectively. Since the both bounds are almost proportional to v S , allowed values of λ S (v S ) are almost unchanged for different v S . We can find a strong constraint for λ S as 0.01 λ S (v S ) 0.05.
In the same way, the Landau pole also exists when g X (v Φ ) is sufficiently large. The energy scale where the Landau pole appears is approximately estimated by the one-loop JHEP02(2016)058 where M Z is given in eq. (2.7). Figure 4 shows [60,61]. When we define the triviality bound as Λ GCU < Λ LP , it prohibits the regions above the solid lines. One can see that the bound leads g X (v Φ ) 0.5 from eq. (2.7), which is almost independent of v Φ . Since the naturalness requires the stronger constraints than the triviality bound in almost all parameter space, we can say that the naturalness guarantees no Landau pole below the GCU scale. Note that the both bounds are almost the same for v Φ = 10 TeV, and they exclude M Z > 10 TeV.
Neutrino masses and baryon asymmetry of the universe
From the Lagrangian (2.1), the neutrino mass terms are given by There is no mixing term between ν L,R and N L,R due to the Z 2 symmetry. The active neutrino masses can be obtained by the usual type-I seesaw mechanism [36][37][38][39][40], i.e., m ν ≈ m D M −1 M m T D . The heavier mass eigenvalue is nearly equal to M M , whose upper bound is given by the naturalness of the Higgs mass. Neutrino one-loop diagram contributes the Higgs mass as where we have used the seesaw relation. For m ν ∼ 0.1 eV, the naturalness requires M M 10 7 GeV. We mention the baryon asymmetry of the universe. In the normal thermal leptogenesis [62], there is a lower bound on the right-handed neutrino mass as M M 10 9 GeV [63]. However, the resonant leptogenesis can work even at the TeV scale, where two right-handed neutrino masses are well-degenerated [41]. In our model, additional U(1) X gauge interactions make the right-handed neutrinos be in thermal equilibrium with the SM particles [64]. A large efficiency factor can be easily obtained, and the sufficient baryon asymmetry of the JHEP02(2016)058 N a (a = 1, 2). universe can be generated by the right-handed neutrinos with a few TeV masses. Since the neutrino Yukawa coupling Y N and Y M almost do not depend on the other phenomenological problems, we can do the same analysis as in ref. [64], and hence, the result is also the same as in ref. [64].
For the vector-like neutrinos (N L,R ), we consider M N = M N L = M N R , which naturally arises from L ↔ R symmetry for the vector-like fermions. Then, the mass eigenvalues are respectively The lighter mass eigenstate N 1 is a DM candidate, because its stability is guaranteed by the Z 2 symmetry. In the limit of m N → 0 (M N 1 = M N 2 ), N 1 and N 2 are degenerate, and N 2 is also effective for a calculation of the DM relic abundance. In the next subsection, we will investigate the degenerate N 1,2 case.
In our model, the U(1) X gauge symmetry is successfully achieved via the CW mechanism. It requires λ Φ (v Φ ) > 0 in eq. (2.6), that is, where n ν is a relevant number of right-handed neutrinos, which is defined as Thus, the Majorana masses must be lighter than the Z boson mass. We have made sure that this constraint is always satisfied when N 1,2 explain the DM relic abundance.
Dark matter
To calculate the DM relic abundance, we use the same formula for the DM annihilation cross sections as in ref. [19], where a new vector-like fermion is only N L,R (or N 1,2 ), and the SM fermions do not have U(1) X charges. The annihilation processes are t-channel N N → φφ, tchannel N N → Z φ, and Z mediated s-channel N N → Z φ. The corresponding diagrams are shown in figure 5. Although our model has other contributions to the annihilation cross sections, they are all negligible in the following setup. We consider the degenerate case for simplicity, in which there is no vector-like mass term of N . Thus, t-channel N N → ss process and s mediated s-channel N N → ν R ν R process does not occur at tree level. From eq. (3.10), (2M N ) 2 < M 2 Z is always required. Then, the annihilation cross section σ(N N → Z * → ff ), where f is some U(1) X charged fermion, is suppressed by 1/M 2 Z . As a result, we can use the same formula for the DM annihilation cross sections as in ref. [19]. The spin independent cross section for the direct detection is almost dominated by t-channel exchange of scalars h and φ, which has been considered in ref. [19]. However, our model has an additional contribution due to Z exchange diagrams, which is given by [65] σ SI = where m n is the nucleon mass, and µ n = m n M N /(m n + M N ) is the reduced nucleon mass. For the DM with the masses of 100 GeV and 1 TeV, the small v Φ regions such as v Φ < 11 TeV and v Φ < 6 TeV are excluded by the LUX experiment, respectively [67]. These bound are stronger than the LEP bound, where v Φ < 3.5 TeV is excluded. In the following, we consider x H = 0 (U(1) B−L ) case. There are six new parameters in the model: the U(1) B−L gauge coupling g X , the two Majorana Yukawa coupling Y N L , Y N R , the two quartic couplings λ Φ , λ HΦ , and the VEV of the complex scalar field v Φ . On the other hand, there are two conditions Y N L = Y N R and eq. (2.9), and we require that N explains the DM relic abundance Ω DM h 2 = 0.1187 [66]. Thus, we have three free parameters for the DM analysis.
Conclusion
To solve the gauge hierarchy problem, we have constructed a classically scale invariant model with a U(1) X gauge extension. We have assumed the classical scale invariance at the GCU scale, where the Higgs mass completely vanishes even with some quantum corrections. The scale invariance is violated around the TeV scale by the CW mechanism, and the Higgs mass can be naturally generated through the scalar mixing term. The GCU is realized by vector-like fermions Q L,R and D L,R , which respectively have the same quantum number as the SM quark doublet and down-type quark singlet but distinguished by the additional Z 2 symmetry, and their masses lie in 800 GeV M V 1 TeV. The GCU scale is Λ GCU = 3 × 10 16 GeV with α −1 GCU = 35.6, and the proton life time is estimated as τ p ∼ 10 37 yrs, which is much longer than the experimental lower bound τ p > 8.2 × 10 33 yrs.
In addition, we have shown that the model can explain the vacuum stability, smallness of active neutrino masses, baryon asymmetry of the universe, and dark matter relic abundance without inducing large Higgs mass corrections. Since there are additional fermions with the SM gauge charges, the SM gauge couplings become larger than the SM case, which leads smaller top Yukawa couplings. Then, the β function of the Higgs quartic coupling becomes larger, and hence the EW vacuum becomes stable. The smallness of active neutrino masses and the baryon asymmetry of the universe can be explained by the righthanded neutrinos via the type-I seesaw mechanism and resonant leptogenesis, respectively. The DM candidate is the SM singlet Majorana fermions N 1,2 , and stability of the DM is guaranteed by the additional Z 2 symmetry. We have analyzed the DM relic abundance in the degenerate case (M N 1 = M N 2 ), and found the upper bound on the DM mass as M N 2.6 TeV. | 6,619.2 | 2016-02-01T00:00:00.000 | [
"Physics"
] |
A New Analytical Procedure to solve Two phase Flow in Tubes
A new formulation for a proposed solution to the 3D Navier-Stokes Equations in cylindrical 1 co-ordinates coupled to the continuity and level set convection equation is presented in terms of an 2 additive solution of the three principle directions in the radial, azimuthal and z directions of flow and 3 a connection between the level set function and composite velocity vector for the additive solution is 4 shown. For the case of a vertical tube configuration with small inclination angle, results are obtained 5 for the level set function defining the interface in both the radial and azimuthal directions. It is found 6 that the curvature dependent part of the problem alone induces sinusoidal azimuthal interfacial 7 waves wheras when the curvature is small oscillating radial interfacial waves occur. The implications 8 of two extremes indicate the importance of looking at the full equations including curvature. 9
Introduction
Annular two phase flow is frequently experienced in industrial applications, for example, in power generation plants and heat removal devices.Applications include transfer lines in pipes (gas-liquid oil wells), evaporators and condensers.This flow regime involves a liquid phase flow in the form of a thin film along the pipe wall and a core region of a gas phase where there are droplets entrained within the gas.An important feature of annular pipe flows is the formation of waves at the core/film interface.Thus the Navier Stokes equations which describe the flow problem must be addressed inorder to determine if the 3-D velocity components exhibit wave-like characteristics.The Navier Stokes equations have been dealt with extensively in the literature for both analytical [1,2] and numerical solutions [3,4].The level set method, has been used originally as a numerical technique for tracking interfaces and shapes [5,6] and has been increasingly applied to various areas of engineering and applied mathematics.In the level set method, contours or surfaces are represented as the zero level set of a higher dimensional function called a level set function.This can be the distance from the particular phase of material to the interface.For example in fracture mechanics level set methods have been used to track the shape around a crack in two and three dimensions that is propagating with a sharp kink [7].In addition, various applications in image segmentation have been used with corresponding active curve evolution algorithms [8,9].Reachability analysis is frequently used to study the safety of control systems.Using exact reachability operators for nonlinear hybrid systems is presented in [10].An algorithm for determining reachable sets and synthesizing control laws is implemented using level set methods in [10].Various models used to compute the interaction of 3D incompressible fluids with elastic membranes or bodies, rely on the use of level set functions [11], to capture the fluid-solid interfaces and to measure elastic stresses that have been used.In [11] the computation of equilibrium shapes of biological vesicles is presented and numerical simulations of spontaneous cardiomoyocyte contractions is presented.A conservative method of level set type for moving interfaces in divergence free velocity fields is presented in [12,13].The method in [13] was coupled to a Navier-Stokes solver for incompressible two phase flow with surface tension.Wave phenomena is known to exist at the interface of two phase immiscible flows [14].In the present paper, we present a level set method for moving interfaces for such velocity fields which are coupled to Navier-Stokes equations for two phase flows in tubes.The need for a cylindrical co-ordinate system is apparent as most of the industrial applications involve assemblies of round pipes and therefore a simple model using a straight tube is necessary to obtain insight into the more complex problem found there .The novelty of the present work is to reveal an analytical approach in solving the 3D cylindrical Navier-Stokes equations where the three principle directions of flow, in radial, azimuthal and longitudinal directions are summed to form a new composite vector velocity expression.In this light, we propose to solve a curvature only formulation of the governing equation for the level set function and one in which curvature is added to the governing equation in the large time evolution for the level set function and hence composite velocity formulation.The importance of these two extremes is to get some insight into the nature of and existence of interfacial waves and level set function for the interface and velocity profiles in annular flow.Also it is hoped that the method outlined in this work will lead to eventually solve the full problem analytically for the composite velocity with full curvature expression in cylindrical coordinates included.To the best of our findings this remains an open problem on the subject of two phase flows.Finally, the method is new in the sense that having an analytical solution allows us to solve for velocity and coupled temperature fields easier than employing numerical methods for certain cases.Our results for the limiting cases demonstrate how the velocity and level set function exhibits wave like phenomea in both the radial and azimuthal directions of flow.The problem is solved by assuming the level set function is a product of a decreasing exponential term with a function of radial and azimuthal coordinates.It is shown that when the rate of decrease of time dependent term is very high the level set function exhibits oscillations with higher frequency and lower amplitude and eventually approaches a hyperbolic tan function which is a steady state level set function.Further results showing composite velocity are also shown.
Level Sets in Cylindrical Co-Ordinates
Let φ be a level set function [5,15].The gradient of the level set function in cylindrical co-ordinates is defined as: The mean curvature, κ, of the interface defined by the zero isocontour of the level set function φ [5,15], is the divergence of the normal to the interface given by Thus it can be expressed as: The mean curvature κ of a dynamic surface φ(r, θ, z, t) = ψ(r, θ, t) + z in cylindrical co-ordinates is, where The geometric trace of a dynamic closed surface that bounds an open set can be represented implicitly as and for two phase flow has two separate regions where φ > 0 and φ < 0 respectively [15].The surface evolution is determined by: In this work the function φ is what is called a filter which has an effective width and will be defined on a small region near the interface.The form of the Navier-Stokes Equations that are not coupled to the interface of the two phases are: where for cylindrical (r, θ, z) coordinate system, Laplace operator has the form the gradient is given by Equation ( 1) and u is the three dimensional velocity of the flow field and is described further below.
A New Composite Velocity Formulation
The 3D cylindrical incompressible unsteady Navier-Stokes equations coupled to interface convection equation are written in expanded form, for each component, u r , u θ and u z , where the remaining non linear terms appearing in full Navier-Stokes equations are suppressed for the time being: and where u r is the radial component of velocity, u θ is the azimuthal component and u z is the component along the pipe in the direction of fully developed upward flow, ρ is density, µ is dynamic viscosity, Fg r , Fg θ , Fg z are body forces on fluid and σ is surface interfacial tension.Introducing the following definition, Multiplying Equations ( 8)-( 10) by unit vectors e r , e θ and k respectively and adding Equations ( 8)- (10) gives the following equation, for L = u r e r + u θ e θ + u z k and L, The level set function φ is governed by Equation ( 7), which when expanded becomes in cylindrical co-ordinates: The continuity equation in cylindrical co-ordinates is Multiply Equation ( 11) by Multiply Level set function convection Equation ( 12) by ρ µ L: Adding Equations ( 14) and (15) gives By product rule we rewrite the previous equation as: It is noted that since L is a composite velocity term it must have units of length per time, and since φ is usually taken to be a distance function from the interface we can consider the following expression in terms of φ : where µ ρ has SI units of m 2 /s and L is defined previously and is of dimension 1/cm and φ is dimensionless, (See Appendix A for decomposition of µ ρ ) where ρ 1 , µ 1 are density and dynamic viscosity of fluid 1 and ρ 2 , µ 2 are density and dynamic viscosity of fluid 2 separated by interface and we write It is assumed that the density and viscosity of fluid 1 is negligible compared to fluid 2. From Level set Equation (7), From the first line of Equation ( 17) using Equation ( 18) we have a derivative term in t, first we write: and for ρ 1 relatively small in comparison to ρ 2 and for µ 1 relatively small to µ 2 we obtain and for derivative terms in r, θ, z , for ρ 1 small relative to ρ 2 and for µ 1 small relative to µ 2 from the first line of Equation ( 17) it can be proven that,
Special Case Solution
In this section, it is assumed that the tube is in a vertical configuration with a small inclination angle, with assumption that ρ 2 and µ 2 are nonzero terms for fluid 2 whereas fluid 1 has approximately zero density and viscosity.This is an idealization of the full problem which is amenable only to numerical treatment.It is worthy to notice that part of Equation ( 22) can be rewritten using the time dependent continuity equation Equation (13), Use of Equation ( 17), Equations ( 18)-( 24) gives a non-linear PDE Using Equation ( 4) and incorporating the curvature κ, and using Equation ( 18) and dot product in Equation ( 19), with the assumption of small inclination angle, Equation (25) for fully developed flow in the vertical z direction becomes, Fz is force of gravity in inclined tube.Incorporating Equation (18) and partial fraction decomposition for µ ρ found in Appendix A, Equation(26) becomes for ρ 1 and µ 1 negligibly small in comparison to ρ 2 and µ 2 respectively of fluid 2, where we have solved Equation (26) using the assumption of small inclination of tube, and have subsitituted L = ∇φ and then set φ = e αt F(r, θ) for α < 0 in Equation(26).This leads to Equation (27).
It can be proven that F(r, θ) is multiplicatively separable when term in t is dropped or t gets large since exponential term is in the denominator for α < 0 , thus, F(r, θ) = f (r)g(θ) and we obtain the following, It can be shown that g(θ) is an arbitrary function of θ, for which one chooses g(θ) to be constant and large inorder to be consistent with the omission of the term in e αt in Equation ( 27).
Consideration of Curvature Alone
Secondly we solve the curvature pde given in Equation ( 4).For small inclination angle of tube Equation (4) reduces to, where c 3 is sufficiently large.For constant c 2 negative there is a sinusoidal component of the azimuthal part of φ(r, θ, t) = f 1(r) f 2(θ) f 3(t).
Results and Discussion
An analytical solution is found to exist for Equation (28) and is given by the following where
Conclusions
It is worth noting that the interfacial oscillations occurring as extremes of two problems one for curvature coupled to composite velocity and the other for curvature alone presents the daunting problem of solving the full equation of Equation (25) without the assumption of small inclination angle.There is a plethora of results for computational multiphase flow using level set methods.The advantage of the present work lies in that analytical results are possible for the two extreme cases presented.It is conjectured at this point that the combination or full Equation (25) without the small inclination angle, i.e., approaching a horizontal tube configuration of flow, is non separable due to the inherent complexity of Equations ( 4) and (25) combined.We can expect that there be a very complex relationship between azimuthal and radial components of L. Work on the complete problem for this complex relationship is in progress for future studies.
Figure 2 .
Figure 2. f versus r for α = −1000 in level set function φ, R i = 0.85 cm is the radial value at the gas/liquid interface.
Figure 3 .
Figure 3. f versus r for α = −5000 in level set function φ, R i = 0.85 cm is the radial value at the gas/liquid interface.
Figure 4 .
Figure 4. f versus r for α = −10,000 in level set function φ, R i = 0.85 cm is the radial value at the gas/liquid interface.
Figure 5 .
Figure 5. f versus r for α = −100,000 in level set function φ, R i = 0.85 cm is the radial value at the gas/liquid interface.
Figure 6 .
Figure 6.f versus r for α = −1,000,000 in level set function φ, R i = 0.85 cm is the radial value at the gas/liquid interface. | 3,098.2 | 2018-05-23T00:00:00.000 | [
"Engineering",
"Physics"
] |
Preparation of Porous Materials by Magnesium Phosphate Cement with High Permeability
High permeability and strength magnesium phosphate cement (MPC) with porosity, average pore size, and compressive strength varied from 63.2% to 74%, 138.7 μm to 284.7 μm and 2.3MPa to 4.7MPa, respectively, were successfully prepared by combining the physical foaming method and chemically entrained gas method at room temperature. )e effects of borax content, chemical foaming agent content, zinc powder content and W/S ratio on the porosity, pore size distribution, compressive strength, and permeability of theMPCwere investigated.)e results indicate that the chemical foaming agent content tends to have little impact on the porosity and compressive strength, and the zinc powder content has the most significant influence on the average pore size of MPC.)e air pores distribution and connectivity of MPCwere mainly controlled by the borax content, W/S ratio, and chemical foaming agent content. Zinc powder played a destructive role in the pores formed by the early physical foaming and led to an increase in pore size and a large number of through pores, which increased the permeability of the materials.
Introduction
Cement-based foam material has high heat capacity, excellent fire resistance, and low cost and is usually used in building energy efficient materials because of its lightweight and thermal insulation properties [1,2].It has also been widely applied in acoustic insulation [3,4], electromagnetic wave absorbing material [5], and safety block material [6].In recent years, magnesium phosphate cement has also been developed to prepare porous materials to obtain foamed concrete for cast-in-situ construction and high-temperature resistance of cement-based porous materials [7].
Phosphate cement is fabricated by an acid-based solution reaction between a divalent or trivalent oxide and an acid phosphate or phosphoric acid [8][9][10].e phosphate used in this system is potassium dihydrogen phosphate, sodium dihydrogen phosphate, or ammonium dihydrogen phosphate.
e metal oxides often used are magnesium, aluminum, zinc, and calcium oxides, and magnesium oxide is the most common oxide to prepare magnesium phosphate cement (MPC) [11].In this reaction, a phosphate gel was formed as a precursor of the ceramics since the metal oxide dissolved to release cations reacted with the hydrolytic phosphate ions, and the main final phase of hydration product is struvite, but many other phases also existed during or after hydration reaction, such as dittmarite (NH 4 MgPO 4 •H 2 O), schertelite ((NH 4 ) 2 Mg(HPO 4 ) 2 •4H 2 O), newberyite (MgHPO 4 •3H 2 O), and magnesium phosphate hydrate Mg 3 (PO 4 ) 2 •4H 2 O [12,13].With the in-depth research of MPC, MPC has received an astonishing amount of attention since it was discovered in the nineteenth century [14,15].Due to its series of advantages, such as rapid setting, high early strength, strong bonding strength, excellent biocompatibility, and low drying shrinkage, it has been widely used in biomedical fields, civil engineering repair, and stabilization of nuclear or heavy metal [16][17][18][19][20][21].Recently, the preparation of porous MPC was reported in some studies such as that of Li and Chen, who prepared a new type of MPC with a density ranging from 210 to 380 kg/m 3 and compressive strength ranging from 1.0 to 2.8 MPa by a prefoaming method [22].Liu et al. also fabricated MPC with a maximum compressive strength of 0.30 ± 0.05 MPa and porosity of 83.75% by the physical foaming process [23].Ma and Chen [7] also obtained a novel foamed concrete with the characteristics of quick setting and high early-strength by using sodium bicarbonate as a foaming agent.However, imperviousness and low strength are the dominant properties of MPC fabricated by the preforming method mentioned above.Recently, the application of cement-based porous materials for hazardous wastewater, such as heavy metals and the radioactive nucleus has attracted more and more attention [24,25].Unlike the applications mentioned above, adsorbent materials require good permeability and high strength to form self-supporting systems, thus achieving good service performance.
In this study, magnesium phosphate cement is used as the base material, and porous materials with high water permeability and strength will be prepared by combining the physical foaming method and chemically entrained gas method, using a chemical foaming agent (CF) and zinc powder as the compound foaming agent.Additionally, the effects of borax content, CF content, zinc powder content, and the water to solid ratio (W/S) on the pore size distribution, connectivity, and compressive strength of the MPC were systematically investigated.
Materials and Experiments
In the current work, the raw materials for the MPC formation were dead-burnt magnesia (MgO, Liaoning Xinrong Mining Co., Ltd., China) obtained by calcinating magnesium carbonate at 1700 °C, and analytically pure ammonium dihydrogen phosphate (ADP, AR-grade, Jinshan Chemical Reagent Co., Ltd., Chengdu, China) and quartz powders were prepared by ball-milling quartz for 30 min.e chemical foaming agent (CF, the primary chemical composition is sodium dodecyl sulfate, alkyl amido betaine, and citric acid) having weak acidity was prepared by our lab, which was combined with zinc powder (Zn) for use as a compound foaming agent.Borax (Na 2 B 4 O 7 •10H 2 O) and analytically pure citric acid monohydrate (CAM) were used as a retarder in this study.As characterized by a laser particle size analyzer, the mean particle diameters of the MgO, ADP, and quartz powders are 29 μm, 60 μm, and 20 μm, respectively.e formulas for fabricating MPC are listed in Table 1, where M/P, M/Q, and Zn/CAM represent the mass ratio of MgO to ADP, MgO to quartz powder, and zinc powder to CAM, and the value is 1, 0.5, and 1, respectively.e CF, zinc powder, and CAM were weighed by the mass of all the solid powders, and the mass of borax weighed the MgO.Additionally, all the water used in this work was tap water from the laboratory.
First, the raw materials including MgO powder, ADP powder, quartz powder, and borax were mixed for 1 min in a vertical-axis planetary mixer according to Table 1.Secondly, the CF, CAM, and water were added and stirred for 1 min, and then zinc powder was mixed with them and stirred rapidly for 90 s. irdly, the slurry after mixing was cast into steel moulds with a size of 40 mm × 40 mm × 160 mm.Finally, the specimens of MPC were demolded after 2 h and cured in the lab at a temperature of 20 ± 2 °C and a relative humidity of 60 ± 5%.It should be pointed out that the CF and CAM were dissolved in water in advance and then added into the mixer.
e sample for SEM is prepared by cast slurry into Φ50 mm × 150 mm column steel moulds, cured at a temperature of 20 ± 2 °C and a relative humidity of 60 ± 5% for 28 d, and cut horizontally into five test blocks for analyzing the pore distribution by SEM pictures.
Before testing porosity, the samples should be dried under the temperature of 60 °C for 24 h to obtain a constant weight.e porosity of material was calculated using P � (1 − (ρ apparent /ρ th )) × 100% with an apparent density of the dried material and theoretical density [26].
e compressive strength of 28 d was applied by a microcomputer control universal testing machine (CMT5105, Shenzhen SANS Testing Machine Co., Ltd., China) with a loading rate of 2.4 kN per second according to standard Chinese GB17671-1999, each sample taken six specimens tested, and the compressive strength is the average value of six samples.Crystal phases were determined by X-ray diffraction using a D/Max-RB (Rigaku, Japan) powder X-ray diffractometer using CuKα irradiation generated at 60 mA and 35 kV; the scanning rate was 8 °/min from 3 to 80 °. e microstructure of the MPC was observed by scanning electron microscopy (TM1000, Hitachi, Japan), and the distribution of the pore size was determined by analyzing SEM pictures using a soft image analyzer (Nano measurer, China).
Results and Discussion
e porosity, average pore size, and compressive strength of the porous MPC fabricated according to Table 1 were measured and the results are listed in Table 2.It can be seen from Table 2 that the porosity increased from 66% to 71.6%, the compressive strength reduced from 4.7 MPa to 2.6 MPa, and the average pore size ranged from 138.7 μm to 160.01 μm as the borax content increased from 4% (M I-1 ) to 10% (M I-3 ).
is result can be attributed to the retarding effect of borax, which can delay the setting time of MPC slurry.As a result, there is enough time for the air bubbles to expand and migrate in the slurry.As the content of CF increased from As the content of zinc powder and CAM increased from 0.5% to 1.5%, the porosity and average pore size increased from 65.2% to 74% and from 151.9 μm to 284.7 μm, respectively, while the compressive strength decreased from 4.6 MPa to 2.3 MPa.ese results may be explained by the reaction between the zinc powder and hydrogen ions which were jointly supplied by the ADP and CAM. e more zinc powder and CAM were added, the more air bubbles were obtained, which led to the higher porosity and lower compressive strength.Furthermore, as the W/S ratio increased from 0.14 to 0.18, the compressive strength reduced from 4.3 MPa to 2.5 MPa due to the porosity and pore diameter size rising from 63.2% to 72%, and from 141.8 μm to 200.4 μm, respectively.e higher W/S ratio makes the consistency of the slurry and foaming resistance lower, which led to a large number of air pores left in the samples and the high porosity obtained.
Figure 1 shows the effect of borax content, CF content, zinc powder content (or CAM content), and W/S ratio on the pore size distribution of the MPC.As shown in Figure 1(a), with the borax content (B%), increased from 4%, 7% to 10%, the pore distribution became wider, ranging from 40 μm to 550 μm, from 80 μm to 1150 μm, and from 70 μm to 1280 μm, respectively.Besides, the majority of pores uniformly ranged from 40 to 300 μm, and the accumulative frequency exceeded 70%.As the content of CF (CF %) increased from 1.0%, 1.5% to 2.0%, the air pore size distribution varied about from 80 μm to 780 μm, 80 μm to 950 μm, and 90 μm to 1280 μm, respectively, as observed from Figure 1(b).However, the CF content has little influence on the smaller air pores, which range in size from 0 to 150 μm.It can be seen from Figure 1(c) that the air pore size distribution ranged from about 0 to 700 μm, from 50 μm to 1300 μm, and from 100 μm to 1350 μm, respectively, as the zinc powder (Zn%) increased from 0.5%, 1.0% to 1.5%.Additionally, a distinct difference can be found that there are about 25% air pores whose size distribution varied from 300 μm to 1350 μm with 1.5% zinc powder in the samples.As seen in Figure 1(d), as the W/S increased from 0.14, 0.16 to 0.18, the size distribution of the air pores varied from 40 to 600 μm, from 50 μm to 1300 μm and from 40 to 750 μm, respectively.Besides, the size of the pores distribution became wider for the specimens with the W/S ratio of 0.16 compared to the other two ratios.
By analyzing the data in Table 2 and Figure 1, there is a certain relationship between the strength of the phosphate cement porous materials and the porosity, pore size of asprepared materials.In general, the greater the porosity, the lower the strength, and the larger the average pore size, the smaller the strength.But comparing the porosity and average pore size, the effect of porosity is higher than the effect of pore size on the strength; for M II-1 , M II-2 , and M II-3 , the porosity is almost the same, and the average pore size increases from 140 μm to 180 μm, but their strength is maintained at about 3.2 MPa; for M II-1 and M IV-1 with almost the same average pore size of 140 μm, the porosity of M II-1 and M IV-1 is 70.4% and 63.2%, the strength of M II-1 and M IV-1 is 3.2 MPa and 4.3 MPa, and the results indicate that the porosity has a huge influence on the strength.Of course, the foaming agent will also affect the strength.In this paper, zinc powder is used as a chemical foaming material, and the strength of different zinc powders is analyzed, such as M III-1 , M III-2 , and M III-3 in Table 1, and the strength decreased from 4.6 MPa to 2.3 MPa with increasing amount of Zn. is is due to the increase in the amount of zinc powder, the increase in porosity and the average pore size, and also due to the side effects of zinc on the hydration process [27].
erefore, for the strength of phosphate cement porous material, the effects are multifaceted, and the influence of porosity plays a leading role than other factors.
Figure 2 shows SEM micrographs of the MPC formulated with different borax content, CF content, zinc powder content, and W/S ratio.As shown in Figures 2(a)-2(c), the size of air pores and the number of connected pores increased obviously when the borax content increased from 4%, 7% to 10%, and the air pores were distributed more uniformly.e pore distribution uniformity and the size of the samples increased slightly as the content of CF increased from 1.0%, 1.5% to 2.0%, but the specimen having the most connected macropores was that with the foaming agent content of 1.5%, as presented in Figures 2(d)-2(f ).As can be seen in Figures 2(g)-2(i), the size and connectivity of the air pores increased significantly as the zinc powder content increased from 0.5 wt.% to 1.5 wt.%.However, the pores were distributed less uniformly as higher contents of zinc powder were used in the samples.As seen in Figures 2(j)-2(l), the size and uniformity of the pores increased obviously when the W/S ratio increased from 0.14 to 0.18.As shown by the consequence of Figure 2, the size and uniformity of the pores were mainly controlled by the borax content, zinc powder content, and W/S ratio, while the CF content mainly controlled the connectivity of the air pores.
We selected the sample M III-2 for XRD analysis shown in Figure 3. Figure 3 shows that the primary phase is SiO 2 , MgO, and MgNH 4 •6H 2 O, of which the SiO 2 peak is the strongest, MgO second, and MgNH 4 •6H 2 O the weakest.e main reason is that SiO 2 mainly acts as a filler and does not participate in the hydration process.Due to the large amount of unreactive SiO 2 , the di raction peaks are sharp.Of course, the incorporation of silica also has more e ect.On the one hand, it can prolong the setting time, which is bene cial to the foaming e ect of Zn powder and forms more pores, on the other hand, silica can also reduce the hydration heat.e reduction of hydration heat avoids the release of ammonia gas during the reaction, while also avoiding the decomposition of hydration products which leads to the strength loss of the materials [28].
Figure 4 shows the water permeability of MPC fabricated by combining the physical foaming and chemically entrained gas methods.It can be seen that water can pass quickly from the surface to underlying layers of the MPC through the abundant connected pore channels that exist in its interior.In general, permeability is mainly a ected by pore size and connectivity of samples.e porous materials obtained by composite foaming technology have a large number of connected pores and large pore size, which can ensure wastewater to easily pass through when used as adsorbent materials.
is paper mainly uses zinc powders as the foaming agent.e previous study also showed that the use of zinc powder could form a better closed-hole structure which has been discussed more detailed, and zinc powder content, water to cement ratio, and borax content play a role on the formation and growth of foams [29].e chemical medium used in this paper is a weak acid foaming agent, which is to reduce its e ect on the slurry of magnesium phosphate cement due to the acidic environment of the initial Advances in Materials Science and Engineering hydration and to ensure that its strength has no significant loss.In the initial stage of molding, a large number of micropores were formed during the stirring process due to the action of the foaming agent CF.At this time, the effect of zinc powder foaming was not apparent.In the static stage, due to the acidic environment at the initial stage of the phosphate cement reaction, together with the low viscosity in the early stage of hydration, the silica fillers prolonged the hydration hardening time, and the zinc foaming effect was exerted.However, at this time, the zinc powder is mainly located on the pore wall formed in the earlier stage.erefore, during the foaming process, the pores formed in the early stage were damaged and enlarged, and a large number of through-holes are formed in the structure so that the water permeability of the material is significantly increased.
Conclusion
In the present work, magnesium phosphate cement (MPC) with high permeability and strength was successfully fabricated by combining the physical foaming method and chemically entrained gas method at room temperature.e porosity, average pore size, and compressive strength of little influence on the porosity and compressive strength of specimens.e pore size distribution is significantly influenced by the zinc powder content and W/S ratio.e compressive strength is affected by the borax content, zinc powder content, and W/S ratio.Additionally, MPC with higher porosity and larger pore diameter shows high permeability due to the mass of through pores existing in the samples.As a result, the porous MPC has enormous potential for filtration applications, such as heavy metals, radioactive waste, exhaust fumes, and could also be used for bio-scaffolds in tissue engineering.
Figure 1 :
Figure 1: Pore size distribution of porous MPC fabricated with di erent borax content in (a), varying foaming agent content in (b), varying zinc powder content in (c), and W/S ratios in (d).
Table 1 :
e formulas for fabricating MPC.
II-1 ) to 2.0% (M II-3 ), the average pore size rose from 140.6 μm to 180.11 μm, but the porosity and compressive strength remained at about 69% and 3.2 MPa, respectively.
Table 2 :
Porosity, average pore size, and compressive strength of porous MPC. | 4,293.8 | 2018-09-09T00:00:00.000 | [
"Materials Science"
] |
Peptidyl-prolyl isomerase-B is involved in Mycobacterium tuberculosis biofilm formation and a generic target for drug repurposing-based intervention
Tuberculosis (TB), a disease caused by Mycobacterium tuberculosis (M.tb), takes one human life every 15 s globally. Disease relapse occurs due to incomplete clearance of the pathogen and reactivation of the antibiotic tolerant bacilli. M.tb, like other bacterial pathogens, creates an ecosystem of biofilm formed by several proteins including the cyclophilins. We show that the M.tb cyclophilin peptidyl-prolyl isomerase (PpiB), an essential gene, is involved in biofilm formation and tolerance to anti-mycobacterial drugs. We predicted interaction between PpiB and US FDA approved drugs (cyclosporine-A and acarbose) by in-silico docking studies and this was confirmed by surface plasmon resonance (SPR) spectroscopy. While all these drugs inhibited growth of Mycobacterium smegmatis (M.smegmatis) when cultured in vitro, acarbose and cyclosporine-A showed bacteriostatic effect while gallium nanoparticle (GaNP) exhibited bactericidal effect. Cyclosporine-A and GaNP additionally disrupted M.tb H37Rv biofilm formation. Co-culturing M.tb in their presence resulted in significant (2–4 fold) decrease in dosage of anti-tubercular drugs- isoniazid and ethambutol. Comparison of the cyclosporine-A and acarbose binding sites in PpiB homologues of other biofilm forming infectious pathogens revealed that these have largely remained unaltered across bacterial species. Targeting bacterial biofilms could be a generic strategy for intervention against bacterial pathogens.
INTRODUCTION
Biofilm associated diseases cause nearly 80% of the recalcitrant hospital infections. 1 Several non-pathogenic and pathogenic species of microorganism including mycobacteria make biofilm as one of the generic mechanisms to overcome stress. The matrix of the biofilm is composed of extracellular components consisting of biopolymers that are essentially secreted by the microorganisms and act as a physical barrier to drugs or against immune surveillance. In addition to the rapid emergence of drug resistance in several strains of mycobacteria, the growing menace of drug tolerance has led to the requirement of higher doses of drugs for effective management of diseases such as tuberculosis (TB). 2 There is acute shortage of drugs that can be used against biofilm forming pathogens. The inherent ability of the pathogen to evolve under selective drug pressure outpaces the rate of development of new drugs, and diminishes the efficacy of the drug by the time it is commercially available. The eminent solution is to expedite new arsenal of drugs against classical and non-classical targets in M.tb proteome or establish new roles for currently available drugs. Drugs licensed for other known disorders in humans related to mental illness, diabetes, malaria etc. target cellular pathways which are also utilised by M.tb for survival. Drug repurposing offers a viable option to fast track new therapies against other diseases. 3 Previous studies showed the involvement of biofilm formation by Mycobacterium abscessus and Pseudomonas aeruginosa in cystic fibrosis and also as a virulence determinant in uropathogenic Escherichia coli isolates. [4][5][6] The presence of extracellular M.tb within biofilm like structure inside lung lesions of M.tb infected guinea pigs undergoing antibiotic treatment points to the possibility of biofilm formation within the host tissues. 7 Peptidylprolyl isomerases (PPIase), popularly known as cyclophilins, are ubiquitously expressed protein foldases which aid in protein folding or refolding by accelerating the rate-limiting cis-trans and trans-cis-conformational changes at Xaa-Pro bonds. 8 M.tb Ppiases is also involved in chaperonic activity, chromatin remodelling, regulatory processes in the cell, RNA-mediated gene expression, modulating of infections etc. 9,10 Several FDA approved drugs and nanoparticle based therapies are being repurposed against biofilms and have shown promising results. Anti-helminth drug, niclosamide, has shown inhibitory effects against biofilms formed by P.aeruginosa. 11 Nanoparticles, by virtue of their small size and charge have also been effective as antimicrobial agents. Silver nanoparticles have shown promising results as an alternative agent to inhibit bacterial biofilms. 12 The biology of M.tb biofilm formation and its clinical relevance is scant in the literature. Pellicles formed at the liquid-air interface of a static culture are working model for in vitro studies on biofilms. 13 In the present study we elucidate the role of M.tb PpiB and identify drug repurposing-based biofilm inhibitors. Recombinant M.smegmatis cells carrying M.tb PpiB gene under anhydrotetracycline inducible promoter, was used as a model for biofilm studies. We show that heterologous expression of M.tb PpiB in M. smegmatis exhibited enhanced biofilm formation as compared to wild type M.smegmatis, pointing to its likely role in developing drug tolerance. Previous studies 14 pointed to the possible interaction of PpiB with cyclosporine-A rendering it a possible candidate among US FDA approved drug for inhibition of biofilms. Recent reports suggest that gallium, a FDA approved agent used in cancer related hypercalcemia and cancer diagnostics, has been repurposed for antimicrobial therapies. 15,16 In-silico studies supported by SPR data showed that acarbose, a FDA approved drug against diabetes, and cyclosporine-A, a FDA approved immunosuppressant used in patients undergoing organ transplantation, interact with PpiB and inhibit biofilm forming activity of PpiB. A comparison of PpiB homologues in different groups of biofilm forming pathogens reveals that the binding residues that interact with cyclosporine-A or acarbose have largely remained conserved, thereby pointing to its efficacy as a putative candidate for targeting biofilms across a wide genre of microorganisms. To our knowledge, the present study proves that PpiB is a suitable candidate to target biofilm forming organisms. We also demonstrate that cyclosporine-A, acarbose or GaNP can reduce the dosage of anti-TB drugs and can be used as conjunct drug/agent for targeting biofilm associated diseases involving other bacteria.
RESULTS
Recombinant M.smegmatis expressing M.tb PpiB show increased biofilm formation in vitro M.smegmatis vector control (Ms_VC) lacking either M.tb_PpiA or M. tb_PpiB genes were used as control to examine the role of M.tb PpiA and PpiB in biofilm formation. Ms_VC and recombinant M. smegmatis (Ms_PpiA and Ms_PpiB) were induced by culturing cells in the absence and presence of anhydrotetracycline, as described in methods. Results in Fig. 1a show that Ms_VC and Ms_PpiA express basal level of biofilm, indicating that M.tb PpiA gene does not contribute to biofilm formation. It could be seen (Fig. 1b) that Ms_PpiB exhibited nearly 1.5-fold increase (p < 0.005) in biofilm formation as compared to either Ms_VC or Ms_PpiA cells. These results demonstrate the involvement of M.tb PpiB in biofilm formation, thereby modulating the cell surface properties of the pathogen.
Cyclosporine-A, acarbose, or GaNP binding sites in PpiB homologues in biofilm forming bacterial species have largely remained unaltered; Evidence of physical interaction A comparison of amino acid sequences using BLAST showed that M.tb PpiB exhibits at least 30% similarity with PpiB homologues in biofilm causing bacteria (supplementary Fig. S1). Results in Fig. 2a show the multiple sequence alignment of amino acid groups in the docking site of PpiB in biofilm forming bacteria that have remained conserved and could putatively interact with acarbose, cyclosporine-A or dimeric atomic gallium. 17 Pro162 and Arg184 in M.tb PpiB are conserved for acarbose and cyclosporine-A binding, respectively and also present in other biofilm forming bacteria. Similarly for binding of dimer of atomic gallium, 17 Gly203 in M.tb PpiB is conserved in all pathogens and the adjacent Thr204, present in the binding groove, is conserved in most of the pathogens. Homologues of M.tb PpiB are also present in several well known pathogenic bacteria such as Staphylococcus aureus, Staphylococcus epidermidis, Staphylococcus intermedius, Streptococcus mutans, Staphylococcus saprophyticus, Streptococcus constellatus, Pseudomonas aeruginosa that are known to make biofilm. It was therefore, investigated whether the amino acid residues present in the active site of PpiB, involved in interaction with acarbose or cyclosporine-A and dimeric atomic gallium, are common to other PpiB homologues in biofilm forming bacteria. The modelled structure of PpiB was found to have overall 98% residues in the allowed regions. Our model scored −1.23 in the MolProbity Clashscore was greater than the recommended Global Z-score values of −3, suggestive of being an adequate model. High throughput virtual screeing (HTVS), as described in methods, was done to study the probable interaction of modelled PpiB sturucture with US FDA approved drugs. These drugs were ranked in order of their docking score with PpiB (supplementary Table 1). Acarbose, with the highest docking score (−13.3), was selected for inhibition studies. Cyclosporine-A, despite having a lesser score (−5.2), was also selected as putative drug in view of its known function as cyclophilin inhibitor. 18 Prokaryotic cyclophilins bind with cyclosporine-A with weak affinity, 14 although not much is known about PpiB cyclosporine-A interaction. 19 The binding site in PpiB used for docking analysis of cyclosporine-A was taken from the homologous structure of PpiB, which showed conserved docking site residue in the catalytic site. 20 We used the Arg184 conserved residue from the catalytic centre of M.tb PpiBcyclosporine-A docked complex which generated high potential energy and therefore we assumed that this protein-drug complex may represent a real entity. Multiple interactions between M.tb PpiB and cyclosporine-A can be seen (Fig. 2b). Molecular docking Table 2). While the interaction of gallium to PpiB is based on dimeric nature of bonding of gallium, 17 it remains to be shown whether the same will be true in a preparation of nanoparticles. Reports exist where atomic/ molecule level docking involving specific amino acids in the target groove have been extrapolated to aggregation/complex/nanoparticle of the same ( 21,22 and references therein). These results, based on in-silico docking and high binding capacity with PpiB, point to the possibility of acarbose, cyclosporine-A and gallium in acting as inhibitors of M.tb PpiB. Drug docking and molecular simulations studies of PpiB with homologues proteins present in biofilm forming bacteria, annotated as WP_061736025.1, WP_049374178.1, WP_019168288.1, WP_019320573.1, WP_048792681.1, WP_006270079.1, CRQ97127.1, were also performed with acarbose, cyclosporine A and dimeric atomic gallium. Our results (supplementary Table 2) show that amino acid residues of Pro and Arg that interact with acarbose and cyclosporine-A, respectively are largely conserved across the PpiB homologous proteins, in some cases it is present at different positions. Presence of conserved amino acids at the cyclosporine-A, acarbose and gallium binding site in PpiB homologues of several biofilm forming bacteria indicate that these have also largely remained unaltered and hence could prove to be an excellent putative target across bacterial species.
Having shown in-silico binding of cyclosporine A or acarbose to PpiB, the actual physical interaction between purified recombinant M.tb PpiB and cyclosporine-A or acarbose, was tested using SPR spectroscopy. SPR analyses show that cyclosporine-A (Fig. 2c), acarbose ( Fig. 2e) or GaNP (Fig. 2g) interact with M.tb PpiB in a dose dependent manner and bind with high affinity. These results suggest that cyclosporine-A, acarbose or GaNP, by virtue of their ability to bind to PpiB, could modulate the activity of PpiB.
M.smegmatis expressing M.tb PpiB show reduced biofilm formation in presence of cyclosporine-A or acarbose or GaNP Given the earlier observation ( Fig. 1) that M.tb PpiB activity is essential for biofilm formation, we speculated that modulation of PpiB activity upon binding with cyclosporine-A, acarbose or GaNP could affect biofilm formation. A threshold concentration of 100, 1000, 50 µg/ml of cyclosporine-A, acarbose and GaNP, respectively at which viability of PpiB expressing M.smegmatis was not significantly affected (supplementary Fig. S4) and did not showed bactericidal effect (supplementary Fig. S5) was validated using alamar blue assay. A decrease in biofilm formation may be a result of decreased cell number per-se, so it was important to ascertain tb PpiB (Rv2582) exhibits homology with proteins from other biofilm forming bacteria and possesses similar amino acids Arg and Pro at the binding site of cyclosporine-A (highlighted in green box) and acarbose (highlighted in red box), respectively. A dimer of atomic gallium 17 similarly binds to Gly residue (highlighted in black box), which is conserved within the PpiB binding site of all biofilm-forming bacteria. b, d, f Interaction of cyclosporine-A, acarbose and dimer of atomic gallium with PpiB was tested by molecular docking analysis. b The docked complex of cyclosporine-A and PpiB. The protein (pink) is shown in surface view whereas interacting residues (grey) and ligand (green) is represented in stick model. Hydrogen bond (yellow) is shown in dotted lines. d Interactions of PpiB with acarbose showing various hydrogen and hydrophobic interactions. f The docked complex of dimer of atomic gallium and PpiB. The protein (pink) is shown in surface view whereas interacting residues (green) and ligand (red) is represented in stick model. Hydrogen bond (black) is shown in dotted lines. c, e, g SPR analysis was performed as described in methods. Response units (RU) of the interaction of PpiB with cyclosporine-A (c) or acrabose e, or GaNP g from representative experiment are shown ) PpiB tet+] of anhydrotetracycline, as described in methods. Cells were treated with cyclosporine-A (0, 10, 100, 1000 μg/ml) a or acarbose (0, 1, 10, 100, 500, 1000 μg/ml) b or GaNP (0, 10, 50, 100, 1000 μg/ml) c and incubated for 7 days. At the end point, biofilm was quantified as described in methods. Values shown from a representative experiment are means [±s.e.m] of biofilm formed.*p < 0.05, **p < 0.01, ***p < 0.005 (Student's t test) the dose of cyclosporine-A, acarbose and GaNP that does not affect the overall growth of M.smegmatis. We accordingly assessed their effect on biofilm formation by M.smegmatis expressing PpiB, as described in methods. It is evident (Fig. 3) that in the absence of anhydrotetracycline induction, Ms_PpiB cells (PpiB tet-) do not develop significant levels of biofilm. Upon induction with anhydrotetracycline, Ms_PpiB cells (PpiB tet+) developed biofilm. It is apparent (Fig. 3a) that 100 µg/ml of cyclosporine-A resulted in significant decrease (p < 0.05) in biofilm formation while complete inhibition of biofilm formation (p < 0.005) was evident at a concentration of 1000 µg/ml, and this was comparable to basal levels of biofilm formation in Ms_VC cells (VC tet-or VC tet+). As shown in Fig. 3b there was a significant inhibition in the biofilm formation in the presence of 500 and 1000 µg/ml of acarbose in Ms_PpiB cells, compared to no acarbose. However, at lower concentrations of acarbose treatment, a transient increase in biofilm formation, attributed due to "Hormetic effect", was noted in VC tet-and PpiB tet-cells. [23][24][25] Significant reduction in biofilm formation compared to control (no treatment) was observed in presence of GaNP at 50 µg/ml (Fig. 3c) (p < 0.05). As expected, in the absence of ppib gene in Ms_VC cells, only basal levels of biofilm formation occurred in either presence or absence of cyclosporine-A, acarbose or GaNP. However, it was intriguing to observe that the levels of biofilm formation in Ms_PpiB cells, uninduced by anhydrotetracycline (PpiB tet-), did not exhibit any significant change to either cyclosporine-A, acarbose or GaNP administration. In Ms_PpiB cells, induction with anhydrotetracycline (PpiB tet+) showed significant decrease in biofilm formation at 100, 1000, and 50 µg/ml of cyclosporine-A, acarbose or GaNP, respectively. These in vitro biofilm inhibition results lend support to the earlier in-silico SPR results, thereby demonstrating that cyclosporine-A, acarbose or GaNP physically interact with M.tb PpiB in a dose dependent manner and suppress the activity of PpiB protein resulting in inhibition of biofilm formation. Cyclosporine-A or acarbose or GaNP co-treatment with anti-TB drugs increases susceptibility of mycobacteria to these drugs We next investigated the impact of reduced biofilm formation in terms of susceptibility to anti-TB drugs. Isoniazid and ethambutol are front line antibiotics that are normally effective at a dosage of 16 µg/ml and 1 µg/ml, respectively. Since M.tb PpiB expression results in enhanced biofilm formation, it could abrogate drug sensitivity of M.smegmatis and hence the MIC of drugs is altered. In the absence of cyclosporine-A or acarbose or GaNP, PpiB tet+ cultures, induced to form biofilm, develop physical barrier over cells and prevent exposure to anti-TB drugs. PpiB tet-cultures, that were not induced to express PpiB proteins, are unaffected by the inhibitory action of cyclosporine-A or acarbose of GaNP on biofilms and hence are exposed directly to anti-TB drugs. Results in Fig. 4a show that in the presence of cyclosporine-A (100 µg/ml), dosage of isoniazid was reduced from 64 to 32 μg/ml. Similarly, the dosage of isoniazid in the presence of acarbose (500 µg/ml), was reduced (Fig. 4b) from 64 to 32 μg/ml for PpiB (tet+). Likewise, dosage of isoniazid in presence of GaNP (50 µg/ml), was 16 µg/ml (Fig. 4c), a four-fold decrease as compared to control.
Similar experiments were carried out for ethambutol. Results (Fig. 4d) show that in the absence of cyclosporine-A, PpiB tet+ cultures exhibited dosage of 16 μg/ml for ethambutol but this decreased four-fold (4 μg/ml, p < 0.05) in the presence of cyclosporine-A (100 µg/ml). Results (Fig. 4e) show that dosage of ethambutol in the presence of GaNP (50 µg/ml), decreased from 16 to 1 µg/ml, as compared to control. The efficacy of acarbose in decreasing the dosage of ethambutol was insignificant (data not shown). These results clearly demonstrate that cyclosporine-A (100 μg/ml) or GaNP (50 µg/ml) inhibit the activity of PpiB protein which in turn negatively impacts the ability of the bacterium to form biofilm efficiently, resulting in reduced percent viability thereby enabling greater access of anti-TB drugs to cells. The consequent reduced viability of Mycobacterium in the presence of cyclosporine-A, acarbose, or GaNP points to their potential use as adjunct therapy.
Cyclosporine-A or GaNP inhibit biofilm formation in M.tb While the experiments described so far were carried out on a nonpathogenic strain of M.smegmatis, the eventual objective of our study was to examine if PPIase is involved in biofilm formation in virulent M.tb strains. Static culture of H 37 Rv cells were incubated in absence/presence of cyclosporine-A or GaNP in a BSL-3 containment facility. Results show that while untreated control H 37 Rv cells formed pellicle, treatment with cyclosporine-A (100 µg/ml) resulted in significant reduction (Fig. 5a) in biofilm formation, reduction was more pronounced in the presence of 50 µg/ml GaNP (Fig. 5b). The role of GaNP in suppressing biofilm formation was further examined in two clinically relevant scenarios. H 37 Rv cells, pretreated for 6 and 24 h with GaNP (25, 50 µg/ml), were incubated to allow biofilm formation. In another set of experiments, H 37 Rv cells were allowed to form biofilm and GaNP (25, 50 µg/ml) treatment was carried out post-biofilm formation. Pretreatment of H 37 Rv cells with GaNP (Fig. 5c) resulted in dose dependent inhibition of biofilm formation that correlated with the duration for which the cells were pre-treated, the suppression in biofilm formation being enhanced when H 37 Rv cells were pretreated with GaNP for 24 h as compared to 6 h. Treatment with GaNP post-biofilm formation resulted in disintegration of pellicle at the liquid-air interface.
DISCUSSION
The current study, directed to address the problem of biofilms using Mycobacterium as model organism, explores proteins that may aid in biofilm formation and their putative inhibitors from the databank of US FDA approved drugs. Biofilm formation involves a complex process that exhibits heterogeneity in terms of the key pathways or mechanisms among different groups of microorganisms. While factors such as PrfA and SinR regulate biofilm formation in Listeria and Bacillus subtilis respectively, several other factors such as RNA regulatory proteins RsmA in P. aeruginosa, chaperonic proteins GroEL-1 in M.smegmatis, cell wall proteins PE11 in M.tb have also been shown to play key role in biofilm formation. 26 There is little consensus on a single protein or factor that may act as a master molecule for biofilm formation. We therefore, started to identify unique protein that could act as putative candidate affecting biofilm formation across species. Except in Mycoplasma genitilium and some members of archaea, all microorganisms possess a highly conserved and ubiquitously expressed group of proteins known as cyclophilins. Cyclophilins, such as peptidyl-prolyl isomerases (PPIases; EC 5.2.1.8), catalyse the cis/trans isomerization of peptidyl-prolyl bonds and are therefore important for correct folding or refolding of nascent proteins that in turn regulate interacting partner proteins to form complexes. 27 The role of several PPIases in biofilm formation, stress tolerance and pathogenesis of bacteria are already known. [28][29][30] M.tb possesses two types of cyclophilins, PpiA and PpiB, of these only PpiB is essential for the survival of the pathogen as knockout variants fail to survive. 31 Previous studies 9 showed that M.tb PpiB possess chaperonic activity and aid in intracellular survival of M.tb. M.tb PpiA or PpiB genes under the control of anhydrotetracycline inducible promoter were cloned in M. smegmatis. Our results (Fig. 1) demonstrated that M.smegmatis overexpressing M.tb PpiB, not PpiA, developed significantly greater biomass of pellicle as compared to a basal expression in control cells. It is interesting to mention that M.smegmatis PpiB displays 64% homology with M.tb PpiB (supplementary Fig. S6). A distinct increase in biofilm formation, when compared with vector control, was expectedly seen in M.smegmatis over expressing M.tb PpiB. Glycopeptidolipids, like PpiB, are component of the membrane fraction and are also part of the secretome and are known to play important role in biofilm formation. [32][33][34] Our results clearly demonstrate a direct involvement of M.tb PpiB in biofilm formation. That PpiB also acts as a chaperone 9 is in agreement with reports of staphylococcus trigger factor having roles in stress tolerance and biofilm formation. 32 Our next step was to identify suitable drug(s) that could act as an inhibitor of PpiB protein. Developing new drugs is a long process taking about 10-15 years. Drug repurposing is gaining popularity as it allows bypassing of the cumbersome clinical trial of drugs for which the parameters of toxicity and effectiveness have already been tested and approved. The effectiveness of osteoarthritis drug Celebrex in decreasing polyp formation in colon cancer patients, anti-malarial drug chloroquine in improving outcome of cancer drug Erlotinib, anti-diabetic drug metformin in lowering morbidity of TB patients are some examples of drug. 35 An inhibitor of human phosphodiesterase, sildenafil originally used in case of erectile dysfunction, has shown encouraging results in animal studies and is now being deliberated as an adjuvant host directed therapy to curtail the duration of TB drug regimen. 36 Nanoparticles are also emerging as key modulators against several human pathogens. FDA approved gallium has shown promising efficacy against M.tb due to similar charge as Fe, thereby allosterically competing with Fe to bind Fesiderophores. 37 This results in disruption of iron metabolism, leading to failure of microbial cells to grow in presence of gallium. 38 Cyclophilin inhibitors, as a unique tool in therapeutic biology, are showing promising results in several diseases. 39 Previous studies 19 pointed that PpiA is a Cyclosporine-A binding cyclophilin and treatment with cyclosporine-A sensitises drug-tolerant biofilm of Candida albicans to various antifungal drugs. 40,41 We used M. smegmatis over expressing M.tb PpiB as a model to evaluate the effect of known and unknown inhibitors of cyclophilins. In-silico docking analysis of interaction of cyclosporine-A with M.tb PpiB homologue revealed that PipB possesses conserved amino acid groups in the binding pocket. These molecular docking studies, in sync with previous studies, show that Cyclosporine-A can stereochemically bind with PpiB. Among the FDA approved drugs, acarbose exhibited greatest docking score and could potentially interact with PpiB. The physical interaction of cyclosporine-A, acarbose and GaNP with PpiB was experimentally confirmed through SPR studies. Consistent with these results, we showed that cyclosporine-A, acarbose or GaNP could suppress biofilm formation. A biphasic dose response for biofilms has been reported for many inhibitors/antibiotics/chemicals/drugs/ligands etc. Such a response, termed as "Hormetic response", is characterised by stimulation of biofilm formation at lower dose and inhibition of the same at higher dose. Some reports have also pointed out to antibiotic acting as antagonist of biofilm formation at low levels, agonists at higher levels and once again antagonist at still higher level. [23][24][25] This is exactly what we have observed: acarbose at lower concentration (upto 100 µg/ml) showed increase in biofilm formation and at 500 µg/ml and above inhibited biofilm formation. While cyclosporine A and acarbose exhibit bacteriostatic activity at the concentration reported (supplementary Fig. S5), only at higher concentration GaNP exhibits bactericidal effect.
A comparison of the amino acid domains in the binding pocket of PpiB homologues expressed in biofilm forming microorganisms interestingly showed that PpiB possess similar amino acids that can interact with either cyclosporine-A or acarbose or GaNP. The structure of gallium nanoparticle has not been reported so far, we therefore used dimeric form of atomic gallium 17 that could act as the building block of gallium nanoparticle. This clearly positions PpiB as a unique protein that can be targeted to inhibit biofilm formation across bacterial species, more so when several mixed species of microorganisms exist in the biofilm. Each of these heterogenous species develop biofilms using varying cellular pathways. Although some antibiotics act as anti-biofilm agents, however, such a drug that may be effective against a putative protein involved in biofilm formation may not be as effective in other organisms either due to the absence of the protein target or redundancy in the metabolic pathway. The presence of conserved amino acid Arg, Pro, Gly at the binding site of Cyclosporine-A, acarbose and GaNP, respectively highlights that PpiB could prove to be a unique target in controlling biofilms, thereby providing a possible generic mechanism for treatment of infections caused by other biofilm producing pathogens.
While there are global efforts to develop new drugs against TB, efforts are needed to reduce the duration of the drug regimen. Using first line anti-TB drugs (ethambutol and isoniazid), we have shown that the reduced mycobacterial biofilm formation in the presence of cyclosporine-A, acarbose or GaNP (Fig. 3) results in dosage reductions for these anti-TB drugs (Fig. 4). While increased dosage of anti-TB drugs results in drug tolerance of the pathogen, it also has a negative impact on patients in terms of toxicity. Our results show that treatment with cyclosporine-A or acarbose help in reducing the dosage of anti-TB drugs by at least two-fold. This has wide implications as it provides proof of principle that cyclosporine-A, a known immunosuppressant that affects T cells, can be repurposed as a conjunct therapy against biofilm associated diseases (Fig. 6). One can argue that treatment with cyclosporine-A may activate latent TB by suppressing immunity. It is conceivable that the concentration of cyclosporine-A at which it A. Kumar et al. inhibits biofilm formation can be reduced further to minimal concentrations by using this drug with suitable adjuvants, thereby reducing chances of its immunosuppressive effects to prevail over its efficacy as biofilm inhibitor. Our results related to the use and efficacy of cyclosporine-A is in line with previous reports that suggest that it acts synergistically to improve the efficacy of antifungals against C.parapsilosis. 42 Cyclosporine-A in combination with azole antifungal flucanizole has been shown to be effective against biofilms formed by C.albicans 43 and also imparts sensitivity to C.albicans towards fluconazole by involving multiple pathways. 44 GaNP is known to facilitate phagosome maturation, inhibit growth of M.tb in macrophages, inhibit HIV infection through release of interferons and can be targeted to human macrophages infected with both M.tb and HIV. 45,46 These studies support our results and point to the possibility that GaNP would be an effective intervention against bacterial biofilms. It has not escaped our attention that cyclosporine-A as an adjunct to existing antitubercular drugs could be a potential strategy to address the problem of eradicating latent TB by first activating the bacterium, by virtue of its immune-suppressive action, followed by the biofilm inhibition reported in our study.
We also show that acarbose treatment resulted in a reduction of dosage of anti-TB drugs such as isoniazid. Acarbose is widely used for the management of type 2 Diabetes mellitus as well. The efficacy of acarbose to block maltose importer and consequently suppress growth of E. coli is known. 47 Given the fact that TB and Fig. 6 Schematic overview of the effect of repurposed drugs on biofilm and its outcome on tuberculosis treatment: Under stress like conditions Mycobacteria secrete exogenous layer of matrix that forms a physical barrier for entry of drugs. The cells within the matrix continuously secrete to develop a biomass of biofilm that enables the cells to withstand high minimum inhibitory concentration (MIC) of drugs. As a result, higher dosage of drugs is required to kill the cells. Cells at the core of the biofilm matrix are least affected by drugs and evolve in due time so as to withstand even higher concentration of drugs. This confers drug tolerance and leads to drug toxicity, increased treatment cost and mortality. Cyclosporine-A, acarbose and GaNP inhibit the activity of PpiB that play crucial role in biofilm formation. Treatment with these drugs suppresses formation of biofilm and the bacterium is exposed directly to the drugs. As a result the drug is effective at low MIC values. Treatment with these drugs also reduces the MIC of existing anti-tubercular drugs resulting in decreased toxicity. The end result is that patient mortality and treatment cost may be reduced significantly. Regular and dotted arrows in the figure denote confirmed and putative roles respectively A. Kumar et al. diabetes exhibit distinct correlation in patients and synergistically affect the clinical outcome of each other in patients, the efficacy of acarbose as a medicament in reducing the dosage of anti-TB drugs could prove to be beneficial. There are several bacteria that are involved in the biofilm of cystic fibrosis, wounds, contact lenses, orthopaedic implants, breast implants, dental biofilm, pacemakers, prosthetic heart valves etc. and have similar protein to M.tb PpiB. 48 Our results can be extrapolated to test the efficacy of cyclosporine-A or acarbose or GaNP in reducing the dosage of other drugs or for diseases caused by biofilm-forming microorganisms as well.
Taken together, these results conclusively demonstrate that PpiB is a potential drug target involved in Mycobacterium biofilm formation and cyclosporine-A, acarbose and GaNP directly bind to PpiB and disrupt biofilm formation. The consequent reduction in dosage values for anti-TB drugs ethambutol and isoniazid points to their ability to act as therapeutic interventions to counter drug tolerance and also possibly reduce dosage of existing antitubercular drugs with implications in reducing drug-induced toxicity and also treatment duration. It will also be interesting to evaluate PpiB as a drug target given the fact that PpiB is not only an essential gene of M.tb, but is involved in biofilm formation. The conservation of drug binding sites within PpiB across pathogenic bacteria biofilm tempts us to suggest that PpiB-targeted biofilm disruption could prove to be a masterstroke for combatting biofilm-related infections across microbial species.
METHODS Reagents
M.smegmatis mc 2 155, initially obtained from ATCC, was maintained in our laboratory as glycerol stocks. M.tb H 37 Rv (gift from Prof R. K. Bhatnagar, Jawaharlal Nehru University, New Delhi, India), was cultured in BSL-3 facility. Growth media (Middlebrook 7H9) and OADC supplement were obtained from BD, USA. Glycerol, Tween 80, acetic acid, acarbose and cyclosporine-A were procured from Sigma-Aldrich, India. GaNP (purity > 99.9%) was obtained from Nanoshel, India. All other reagents such as Alamar Blue, Crystal Violet, Isoniazid, and Ethambutol were of analytical grade and obtained from Himedia, India.
Constructs and recombinant strains used in the study
Recombinant strains of M.smegmatis expressing M.tb Ppiases were generated using E. coli-mycobacterium shuttle vector pST2K and specific oligonucleotide primers, as detailed elsewhere. 10 Briefly, pST_ppiA, pST_ppiB and pST2K vector containing anhydrotetracycline inducibe promoter were electroporated in wild type M.smegmatis (Ms_WT) and transformed strains were designated as Ms_PpiA, Ms_PpiB, and Ms_VC, respectively.
Cell culture and biofilm formation M.smegmatis and M.tb H 37 Rv were maintained in growth media supplemented with 10% OADC, 0.001% glycerol, and 0.05% Tween-80. Cultures were incubated at 37°C in a shaker incubator and diluted to OD of 0.08 in growth media prior to sub-culturing in 96-wells or in test tubes for induction of biofilms. 20 ng/ml anhydrotetracycline was used to induce PPIase expression in recombinant M.smegmatis, resulting in biofilm formation. Recombinant M.smegmatis and M.tb H 37 Rv were cultured in static phase in Tween 80-free growth media for 7 days and 4 weeks, respectively to allow formation of pellicle at the liquid air interface.
Crystal violet assay
Cyclosporine-A, acarbose, and GaNP effect on biofilm formation was assessed by quantifying the pellicle formed at the liquid air interface, using crystal violet. 49 Anhydrotetracycline induced Ms_PpiB and Ms_VC cells were cultured in the presence of various concentrations of cyclosporine-A (0, 10, 100, and 1000 µg/ml), acarbose (0, 1, 10, 100, 500, 1000 µg/ml) or GaNP (0, 10, 50, 100, 1000 µg/ml) in sterile flat bottom 96-well microtiter plate (Thermo Scientific, India). At the end of 7 days of static phase culture, the media beneath the pellicle were aspirated out and remaining solid pellicle was stained by adding 125 µl (w/v) 0.1% Crystal Violet solution. The stained pellicle was washed thrice with water followed by addition of 30% acetic acid. The samples were subsequently incubated for 10-15 min at room temperature to dissolve the stain and absorbance was spectrophotometrically recorded at 550 nm.
Percent viability of biofilm induced culture
Biofilm formation by M.smegmatis Ms_VC and Ms_PpiB induced with or without anhydrotetracycline was performed in presence of FDA approved agents, as described above. At the end of 7 days of incubation period isoniazid (8,16,32, 64 µg/ml) or ethambutol (0.25, 1, 4, 16 µg/ml) was added to the wells of the microtiter plate and the plate was incubated further for 68 h. Cell viability in presence of isoniazid and ethambutol were assessed using Alamar Blue assay. 50 Briefly, 0.01% alamar blue reagent was added to each well of microtiter plate and the plates were further incubated for 3-4 h. Conversion of resazurin (blue) to resorufin (pink) was monitored at 570 nm and 600 nm, respectively to score the viability of the cells. All assays were performed in triplicate.
In-silico amino acid sequence alignment and similarity search The amino acids sequences of M.tb PpiA (GenBank accession number: CCP42731.1), M.tb PpiB (GenBank accession number: CCE38048.1 were downloaded from the NCBI. The sequence homology search of M.tb PpiB was done using BLASTp in known biofilm forming bacteria on NCBI website. M.tb PpiB amino acids sequence was used as queries in BLASTp analyses against the NCBI non-redundant protein database of the specific bacteria to find their similar homologues.
Modelling of PpiB structure and molecular dynamics (MD) simulations
Crystallographic structure of PpiB, being unavailable at Protein Data Bank, homology modelling techniques involving multiple bioinformatics tools or servers such as the MODELLER version 9.11 or Phyre2, respectively were used to generate PpiB model structure. Protein sequence of PpiB from M.tb (strain ATCC 25618/H 37 Rv) was obtained from UniprotKB database [P9WHW1]. Protein structure model validation was carried out using protein structure validation software suite (PSVS). Molecular docking analysis of M.tb cyclophilin (PpiB) in complex with cyclosporine-A, acarbose, or gallium Molecular docking analysis of cyclosporine-A was carried out to study the interactions and affinity with the PpiB protein using AutoDock Tools 1.5.6 (open access). 3D structure of cyclosporine-A was obtained from chemical structure database ChemSpider. AutoDockVina 1.1.2 program was used for docking of cyclosporine-A at the docking site of PpiB protein. The location of the catalytic site was mapped and deduced from the structure-based alignment of related proteins reported earlier. 20 Alternatively, Glide module of Schrodinger was used to screen other compounds from FDA library as described previously. 51 Briefly, Drug library was prepared using Ligprep module applying OPLS 2005 force field and docking was performed using HTVS and XP (extra precision) docking to filter out the compounds with low binding energy. Compounds having a docking score greater than −5 in HTVS were used for XP docking protocol. An XP score greater than −8 was scored as strong binding. The dynamic nature of interaction between PpiB and acarbose was studied using GROMACS version 4.6.5 and above assigning GROMOS96 43a1 force field as per standard protocols.
The chemical structure of elemental gallium was obtained from PubChem (CID 5464084). Using the elemental gallium, a dimer structure for atomic gallium was created using Maestro interface available from Schrodinger. Molecular orbital analysis of gallium in dimeric state showed that gallium dimer is the essential building block for the formation of gallium clusters. 17 Further molecular docking study of dimeric atomic gallium was carried out with PpiB protein and its homologues using PatchDock algorithm. 52 Ligplot was used for visualisation of the interactions between protein-ligand complex in 2D schematic representations. PyMol and Chimera were used for preparing cartoon representations of the structures.
A. Kumar et al. | 8,384.6 | 2019-01-15T00:00:00.000 | [
"Biology",
"Medicine"
] |
Thymic Stromal Lymphopoietin Induction Suppresses Lung Cancer Development
Simple Summary The recurrence rate for lung cancer is high after the removal of the primary tumor. Herein, we demonstrate the potential of immunotherapy against lung cancer by examining the impact of Thymic Stromal Lymphopoietin (TSLP) cytokine induction on early lung cancer development. TSLP induction suppresses the development of invasive lung tumors in a mouse model of spontaneous lung cancer. This cancer suppression is dependent on CD4+ T cells, which highlights the role of adaptive immune response in protection against lung cancer progression. Abstract Lung cancer is the leading cause of cancer deaths in the United States and across the world. Immunotherapies, which activate tumor-infiltrating cytotoxic T lymphocytes, have demonstrated efficacy for the treatment of advanced-stage lung cancer. However, the potential for harnessing the immune system against the early stages of lung carcinogenesis to prevent cancer development and recurrence remains unexplored. Using a mouse model of lung adenocarcinoma, we investigated the effects of thymic stromal lymphopoietin (TSLP) induction on early cancer development in the lungs. Herein, we demonstrate that systemic TSLP induction suppressed spontaneous lung cancer development in KrasG12D mice. TSLP drove a significant CD4+ T cell response to block lung cancer progression from atypical alveolar hyperplasia to adenocarcinoma. Our findings suggest that TSLP can be used in the early stages of lung cancer development to trigger a lasting immunity in the tissue and prevent the development of advanced disease.
Introduction
Lung cancer is the primary cause of cancer-related deaths in the US and the world [1,2]. Non-Small Cell Lung Cancer (NSCLC) comprises 85% of lung cancer diagnoses, and 40% of those are adenocarcinomas, a glandular neoplasm [3]. Surgery for the early and locally advanced disease remains the main treatment option for patients diagnosed with NSCLC. However, despite curative resection, recurrence rates remain high with 30-75% of NSCLC patients developing lung cancer after surgical removal of their first tumor, and with up to 80% developing recurrence within 2 years after diagnosis [4]. Thus, novel strategies to block lung cancer development and recurrence are urgently needed.
Among the potential therapeutic approaches for early lung cancer, current immunotherapies have several limitations. Apart from immune-related adverse events [5], immunotherapies that depend on high tumor-infiltrating lymphocytes at baseline may have minimal efficacy against early malignancies, which are largely non-immunogenic "cold" tumors [6].
Early malignant lesions in the lungs have a low mutational burden, and a less inflamed microenvironment, with low infiltration of CD8 + T cells, which may lead to poor responses to checkpoint blockade therapy, neoantigen-based vaccines and engineered T cells [7,8]. Thus, identifying novel immune inductive pathways that can target tumors at the early stages of their development is imperative to generate effective therapeutic and preventative approaches against poorly inflamed early malignant lesions. In addition to suppressing cancer development, such strategies used in the neoadjuvant and adjuvant settings will create an opportunity for a lasting antitumor immunity that prevents cancer recurrence.
We have previously demonstrated that the induction of Thymic Stromal Lymphopoietin (TSLP) leads to a robust antitumor CD4 + T helper 2 (Th2) cell immunity that suppresses skin and breast cancer development [9,10]. TSLP is an epithelium-derived alarmin cytokine that is a master regulator of allergic inflammation at barrier organs like the skin and lungs [11]. Interestingly, epidemiological data suggest an inverse relationship between allergic disease and lung cancer risk [12,13], which could be explained by higher serum TSLP levels in the patients. Accordingly, we hypothesize that TSLP induction in patients with early lung adenocarcinoma may prevent cancer progression and recurrence. TSLP can promote differentiation of CD4 + T cells in the Th2 cells either directly by binding to TSLP receptor (TSLPR) expressed on CD4 + T cells or through activation of dendritic cell and other antigen-presenting cells, which can then polarize CD4 + T cells to differentiate into the Th2 phenotype [14]. TSLP-activated CD4 + Th2 cells create an immune milieu rich in IL-4, IL-5, IL-13 as well as TNF-alpha [15]. Although baseline TSLP expression in the tumor microenvironment has been associated with tumor promotion [16][17][18][19][20], harnessing the potential of TSLP induction to trigger a strong CD4 + T cell response, which has been previously shown to be cancer suppressive [9,10,21], may provide a powerful therapeutic approach against lung cancer.
To determine the effects of TSLP induction on early carcinogenesis in the lung, we crossed a mouse model of spontaneous lung adenocarcinoma, Kras +/G12D (Kras G12D ), with K14-TSLP tg (Tslp tg ) mice. Tslp tg Kras G12D mice developed a significantly lower lung tumor burden compared with Kras G12D mice. Tslp tg Kras G12D lung tumors were mainly composed of lower-grade atypical alveolar hyperplasia and adenoma compared with adenocarcinoma in Kras G12D lung. We found increased numbers of lymphoid aggregates in Tslp tg Kras G12D lungs and increased T cell infiltration in Tslp tg Kras G12D lung tumors compared with Kras G12D lungs and lung tumors, respectively. Finally, CD4 + T cell depletion reversed the protective impact of TSLP against lung carcinogenesis in TSLP overexpressing mice. Our results indicate that TSLP induction can generate a strong, long-lasting cancer suppressive immune response in the lung and provide a potential therapeutic approach for the prevention and treatment of lung cancer.
Mice
All mice were housed under pathogen-free conditions in an animal facility at Massachusetts General Hospital in accordance with all animal care and relevant ethical regulations. Kras +/G12D (B6.129S-Krastm3Tyj/Nci or K-rasLA2) mice were obtained from NCI Mouse Repository. K14-TSLP tg mouse strain was a gift from Dr. Andrew Farr (University of Washington, Seattle, WA, USA).
Statistical Analysis
A two-tailed Mann-Whitney U test was used as the test of significance for tumor counts, T cell counts and other quantitative measurements. A two-tailed Fisher's exact test was used to compare tumor grade distributions. For multiple group comparisons, the Kruskal-Wallis test with Dunn's multiple comparison post-hoc test was used. A p value < 0.05 was considered significant. All bar graphs show mean + SD.
Study Approval
Animal studies were approved by Massachusetts General Hospital Institutional Animal Care and Use Committee (IACUC).
Spontaneous Lung Cancer Studies
All mice were harvested at postnatal day 100. Lungs and other tissues were collected for histological analysis.
Histology and Immunofluorescence
Mouse lung was fixed in 4% paraformaldehyde (Sigma-Aldrich, Darmstadt, Germany; P6148) and incubated at 4 • C overnight on a shaker. The next day, tissues were washed in 1x PBS and dehydrated in ethanol. Next, lungs were processed, embedded in paraffin, and cut into 5 µm sections. Then lungs were deparaffinized and stained with Hematoxylin (Sigma Aldrich, St. Louis, MO, USA; GHS132) and Eosin (Leica Biosystems; Buffalo Grove, IL, USA; 3801619). For Immunofluorescence assays, tissue slides were deparaffinized and rehydrated. Then tissues were permeated with 1x PBS supplemented with 0.2% v/v Triton X-100 (Thermo Fisher Scientific, Waltham, MA, USA; BP151). For all washing steps, slides were dipped in three, three-minute rounds of 1x PBS supplemented with 0.1% v/v Tween 20 (Sigma-Aldrich; P1379). Antigen retrieval was performed in antigen unmasking solution (Vector Laboratories, Burlingame, CA, USA; H-3300) using a Cuisinart pressure cooker for 20 min at high pressure. Slides were then washed, and incubated in a blocking buffer composed of 5% v/v goat serum (Sigma-Aldrich; G9023), and 5% m/v bovine serum albumin (Thermo Fisher Scientific; BP1600) for 1 h at room temperature (RT). Then slides were incubated overnight at 4 • C with primary antibodies (Table S1). The next day, slides were washed and incubated at RT with fluorochrome-conjugated secondary antibody diluted in blocking buffer for 2 h. After washing, slides were counterstained with 1:4000 DAPI (Thermo Fisher Scientific; D3571) at RT for 5 min. Then slides were washed and mounted with Prolong Gold Antifade Reagent (Thermo Fisher Scientific; P36930).
For imaging, slides were scanned using Nanozoomer s60 digital scanner (Hamamatsu Corp. Bridgewater, NJ, USA) or Axio Scanner (Axio Scan.Z1, Zeiss, Jena, Germany). H&E images were analyzed using NDP.view 2+ software or the Zeiss ZEN Image processing software. Quantification of cell population in tumor and lymphoid aggregates were performed manually using Cell Counter on NDP.view 2+ or with HALO Digital Pathology Software for Image Analysis (HPF, Indica Labs, Albuquerque, NM, USA). Positive cells were quantified within 20× frame by comparing fluorescent intensity of double-(DAPI + CD3 + /Ki67 + ) and/or triple-(DAPI + CD3 + CD4 + ) positive cells to the background (Table S1).
CD4 + T Cell Depletion
After weaning (postnatal day 30), Tslp tg Kras G12D and Kras G12D mice were injected intraperitoneally with 750 µg in 200 µL of sterile PBS with anti-CD4 (αCD4, Table S1) or IgG (control; Sigma-Aldrich, I4131) antibody for the first dose. One week later, the mice were then injected with 250 µg of αCD4 or IgG antibody in 200 µL of sterile PBS weekly until postnatal day 100. To determine whether CD4 + T cells were systemically depleted, we performed flow cytometry on mice spleen, lymph node and blood at the time of harvest.
TSLP Induction Reduces Tumor Burden in a Mouse Model of Spontaneous Lung Carcinogenesis
To study the impact of systemic TSLP induction on early lung carcinogenesis, we used Kras G12D mice, which develop a spontaneous mutation at position 12 of the Kristen Rat Sarcoma oncogene [22]. This somatic activation of Kras oncogene recapitulates spontaneous oncogene activation in humans [22]. We crossed Kras G12D mice with Tslp tg mice that overexpress TSLP in the skin leading to high circulating TSLP levels [10]. Kras G12D mice develop atypical alveolar hyperplasia as early as two weeks after birth and these lesions progress to adenocarcinomas by postnatal day thirty [22]. Tslp tg Kras G12D (test) and Kras G12D (control) mice were harvested at postnatal day 100, and average lung tumor size, tumor counts, and tumor surface area as a percentage of total lung surface area were recorded. Tslp tg Kras G12D lungs showed a marked reduction in tumor burden with betterpreserved lung architecture compared with Kras G12D lungs ( Figure 1a). Although TSLP's effect on suppressing tumor initiation in the lung was marginal (reflected in the number of tumor foci developed in the lung), we found TSLP induction to strongly suppress lung tumor promotion reflected in average tumor size and percentage of the lung surface area covered by tumors (Figure 1b-d). Therefore, systemic TSLP induction substantially suppressed Kras-driven lung tumor promotion.
TSLP Induction Reduces Tumor Burden in a Mouse Model of Spontaneous Lung Carcinogenesis
To study the impact of systemic TSLP induction on early lung carcinogenesis, we used Kras G12D mice, which develop a spontaneous mutation at position 12 of the Kristen Rat Sarcoma oncogene [22]. This somatic activation of Kras oncogene recapitulates spon taneous oncogene activation in humans [22]. We crossed Kras G12D mice with Tslp tg mice that overexpress TSLP in the skin leading to high circulating TSLP levels [10]. Kras G12D mice develop atypical alveolar hyperplasia as early as two weeks after birth and these lesions progress to adenocarcinomas by postnatal day thirty [22]. Tslp tg Kras G12D (test) and Kras G12D (control) mice were harvested at postnatal day 100, and average lung tumor size tumor counts, and tumor surface area as a percentage of total lung surface area were rec orded. Tslp tg Kras G12D lungs showed a marked reduction in tumor burden with better preserved lung architecture compared with Kras G12D lungs ( Figure 1a). Although TSLP's effect on suppressing tumor initiation in the lung was marginal (reflected in the number of tumor foci developed in the lung), we found TSLP induction to strongly suppress lung tumor promotion reflected in average tumor size and percentage of the lung surface area covered by tumors (Figure 1b-d). Therefore, systemic TSLP induction substantially sup pressed Kras-driven lung tumor promotion.
TSLP Induction Arrests Tumor Development at an Early Adenoma-like Stage
Spontaneous activation of an oncogenic allele of Kras results in early-onset lung cancer in Kras G12D mice, with tumors harboring similar histopathological features as human non-small cell lung cancer (NSCLC) [22]. These mice develop hyperplastic lesions in the alveolar epithelium composed of mild dysplasia without disruption of the surrounding lung architecture and closely resemble atypical alveolar hyperplasia (AAH), which have been suggested to be precursors of lung adenocarcinoma in humans [23,24]. AAH progresses to small tumors termed alveolar adenoma, with more densely packed atypical cells and smaller gaps between the basement membrane of type II pneumocytes [22]. Some alveolar adenomas develop into larger, less differentiated tumors, referred to as adenocarcinomas [22]. The slow growth kinetics and human-like morphological progression of the lung tumors in Kras G12D mice provided a suitable model to investigate the mechanism by which TSLP suppressed the early stages of lung carcinogenesis.
To determine the effects of TSLP on the histopathological progression of lung tumors, we utilized a grading system based on the defined Kras G12D tumor progression in the lung ( Figure S1). Interestingly, we observed that the majority of the lung tumors in Tslp tg Kras G12D mice were of AAH and adenoma types while the majority of lung tumors in Kras G12D mice were high-grade adenocarcinoma (Figure 2a,b). To further examine the mechanism of TSLP-mediated cancer suppression in the lung, we stained the Tslp tg Kras G12D and Kras G12D lungs with TUNEL (a marker for cell death) and Ki67 (a marker for cell proliferation). Although we did not detect any increase in tumor cell death in Tslp tg Kras G12D lungs (data not shown), we detected a significant reduction in tumor cell proliferation in Tslp tg Kras G12D compared with Kras G12D lungs (Figure 2c,d). These results suggest that TSLP induction leads to lung cancer suppression by blocking tumor cell proliferation and cancer progression instead of cytotoxicity.
Tslp tg Kras G12D Lungs and Tumors Are Highly Infiltrated by CD4 + T Cells
Based on our previous findings on the critical role of CD4 + T cells in mediating the antitumor effects of TSLP in the skin and breast [9,10], we examined the status of T cell infiltration into the lung and tumors of Tslp tg Kras G12D versus Kras G12D mice. Although Kras G12D lungs showed a trend toward increased number of lymphoid aggregates compared with WT lungs ( Figure S2), Tslp tg Kras G12D lungs contained a significantly higher number of T cell-rich lymphoid aggregates compared with Kras G12D lungs ( Figure 3a). Notably, Tslp tg lungs had a similar number of lymphoid aggregates compared with WT lungs ( Figure S2). In addition, lymphoid aggregates in Tslp tg Kras G12D lungs contained significantly higher T cell proliferation, a higher percentage of CD4 + T cells, and were larger compared with lymphoid aggregates in Kras G12D lungs (Figure 3b-d). Consistent with the increased lymphoid aggregate formation in Tslp tg Kras G12D lungs, Tslp tg Kras G12D tumors contained CD4 + T cell-rich lymphocytic infiltrates (Figure 3e,f). These results suggest a critical role for TSLP-induced CD4 + T cell immunity in blocking lung cancer development.
TSLP-Activated CD4 + T Cells Are Required for Suppressing Lung Carcinogenesis
To determine whether CD4 + T cells were required for the decrease in lung tumor burden observed in Tslp tg Kras G12D mice, we depleted CD4 + T cells in Tslp tg Kras G12D and Kras G12D mice. Although long-term αCD4 antibody treatment failed to completely deplete CD4 + T cells in mice ( Figure S3), the marked reduction in CD4 + T cells in Tslp tg Kras G12D mice erased any differences between tumor size and percentage of lung surface area with tumor in Tslp tg Kras G12D compared with Kras G12D mice (Figure 4a-c). Interestingly, CD4 + T cell depletion resulted in similar lung tumor grades in Tslp tg Kras G12D compared with Kras G12D mice (Figure 4d,e). Importantly, we observed that, in the absence of TSLP induction, Kras G12D mice treated with αCD4 + T cell antibody developed lower grade lung tumors compared with Kras G12D mice treated with IgG control (Figure 4d,e). These results demonstrate that TSLP-activated CD4 + T cells play an essential role in blocking lung cancer Cancers 2022, 14, 2173 6 of 12 progression in Tslp tg Kras G12D mice. However, baseline CD4 + T cells present in Kras G12D lungs may have a tumor-promoting function [16,25,26]. mechanism of TSLP-mediated cancer suppression in the lung, we stai Kras G12D and Kras G12D lungs with TUNEL (a marker for cell death) and Ki6 cell proliferation). Although we did not detect any increase in tumor cell Kras G12D lungs (data not shown), we detected a significant reduction in tum eration in Tslp tg Kras G12D compared with Kras G12D lungs (Figure 2c,d). These that TSLP induction leads to lung cancer suppression by blocking tumor ce and cancer progression instead of cytotoxicity. cantly higher T cell proliferation, a higher percentage of CD4 + T cells, and were larger compared with lymphoid aggregates in Kras G12D lungs (Figure 3b-d). Consistent with the increased lymphoid aggregate formation in Tslp tg Kras G12D lungs, Tslp tg Kras G12D tumors contained CD4 + T cell-rich lymphocytic infiltrates (Figure 3e,f). These results suggest a critical role for TSLP-induced CD4 + T cell immunity in blocking lung cancer development. cell depletion resulted in similar lung tumor grades in Tslp tg Kras G12D compared with Kras G12D mice (Figure 4d,e). Importantly, we observed that, in the absence of TSLP induction, Kras G12D mice treated with αCD4 + T cell antibody developed lower grade lung tumors compared with Kras G12D mice treated with IgG control (Figure 4d,e). These results demonstrate that TSLP-activated CD4 + T cells play an essential role in blocking lung cancer progression in Tslp tg Kras G12D mice. However, baseline CD4 + T cells present in Kras G12D lungs may have a tumor-promoting function [16,25,26].
Discussion
Immunomodulatory agents that can block early stages of cancer development provide a novel strategy to enable cancer immunoprevention. Due to reduced antigen presentation and low co-stimulatory molecule expression, lung cancer cells evade cytotoxic immunity [27]. KRAS G12D and TP53 co-mutation are associated with reduced immune cell infiltrates in the human lung adenocarcinoma [28]. In addition, patients with lung adenocarcinoma lack antitumor innate immunity at baseline including tumor-infiltrating NK cells, which are less cytolytic due to low expression of granzyme B and interferon-gamma [29]. In this report, we have shown that TSLP cytokine induction can mount a strong antitumor immunity in an oncogene-driven in vivo model of spontaneous lung adenocarcinoma. Our findings demonstrate that cytokine induction, in the context of oncogene-driven early carcinogenesis, can lead to T cell infiltration into the tumor and the suppression of cancer progression from premalignancy to invasive disease. This is also significant because current cancer immunotherapies for lung cancer are not effective against early disease due to their poor immune infiltration at baseline. Our results indicate that cytokines can trigger a potent antitumor adaptive immune response during the early stages of lung cancer development. When delivered between the time of cancer diagnosis and surgical resection, the cytokine inductive treatment may provide an effective strategy for treating the malignant lesion as well as preventing its recurrence.
We find that TSLP induction not only suppresses lung tumor initiation as demonstrated by lower tumor counts in mice expressing the Tslp transgene but also results in the development of smaller, more differentiated tumors. This is of particular relevance because lower histological grades in lung cancers are associated with lower recurrence rates [30] and better survival outcomes [31]. We also show higher CD4 + T cell infiltrates in lung tumors and lymphoid aggregate formation in the lungs of Tslp tg Kras G12D compared with Kras G12D mice that do not overexpress TSLP. This is consistent with TSLP's role in the induction of Th2 cell proliferation and activation in barrier organs [14]. As a cardinal cytokine in the pathogenesis of allergic diseases, the cancer-protective effect of TSLP in the lung provides an explanation for the epidemiological observation of an inverse relationship between allergic disease and lung cancer risk [12,13]. Importantly, the significantly higher lymphoid aggregate counts in Tslp tg Kras G12D compared with Tslp tg mice suggests that TSLP-activated T cells in the lungs are responding to the tumor antigens as opposed to a nonspecific allergic inflammation caused by TSLP overexpression alone. Our results put forward TSLP induction as a potential therapeutic approach that can trigger a lasting antitumor immune response capable of bringing the adaptive immune cells to recognize and suppress early stages of lung carcinogenesis. Considering a significant reduction in tumor cell proliferation in Tslp tg Kras G12D tumors compared with Kras G12D tumors, cytotoxicity is unlikely to be the main mode of antitumor immunity downstream of TSLP induction in the lung. Future studies are warranted to investigate the precise nature of antitumor CD4 + T cell immunity induced by TSLP and determine the impact of TSLP on other immune cells including CD8 + T cells in suppressing lung carcinogenesis.
It is important to note the emerging role of CD4 + T cells in antitumor immunity, and in particular, their functional versatility in the context of the tumor immune microenvironment. CD4 + T cells have been shown to initiate a strong antitumor response in tumors by enhancing the clonal expansion of cytotoxic T lymphocytes as well as directly through the secretion of TNF-α and other tumor-suppressive factors [32]. Additionally, CD4 + T cells have been shown to promote the differentiation of effector CD8 + T cells in a mouse melanoma model known to be immunogenically cold and poorly responsive to immune checkpoint blockade [33,34]. CD4 + T cells are found to be required for immune checkpoint blockade response in the context of a poorly immunogenic T3 sarcoma cell line engineered to express MHCII antigens [35]. Of note, CD4 + T cells have been demonstrated to have anti-cancer functions outside their role in stimulating cytotoxic immune cells in cancer. TGF-β blockade leads to CD4 + Th2 cell-induced cancer cell hypoxia through the remodeling of the tissue vasculature in an MMTV-PyMt murine breast cancer model [36]. Clinically, the presence of CD4 + T cells in the peripheral blood of lung cancer patients before immune checkpoint blockade therapy is associated with increased tumor-infiltrating CD8 + T cells and better response to PD-1 therapy [37,38]. We provide evidence to support the critical role of CD4 + T cells in antitumor immunity and highlight the need for future investigation into the role of type 2 immunity in the context of early carcinogenesis. We also find induction of broad T cell immunity by TSLP in the lung including CD8 + T cells with known antitumor effects. Future studies are warranted to address the role of TSLP-stimulated CD8 + T cells in lung carcinogenesis. In addition, understanding the role of TSLP signaling to innate immune cells and Kras mutant tumor cells in regulating lung cancer development is an important area for future investigation. Furthermore, we aim to study the effects of TSLP receptor expression in lung tumor cells, lymphocytes and other antigen presenting cells to understand the role of TSLP signaling to these cell types in the lung tumor microenvironment.
Conclusions
In conclusion, we have shown that TSLP induction reduces tumor burden and blocks cancer progression in a mouse model of spontaneous lung adenocarcinoma. TSLP induction leads to a massive T cell infiltration into and lymphoid aggregate formation in the tumor-bearing lungs. Finally, we have demonstrated that TSLP-activated CD4 + T cells are required to suppress the progression of Kras-driven lung tumors to invasive adenocarcinoma in mice.
Supplementary Materials:
The following are available online at https://www.mdpi.com/article/ 10.3390/cancers14092173/s1, Figure S1: Representative images of lung tumor histological grades, Figure S2: Lymphoid aggregate numbers in mice lacking Kras G12D oncogene, Figure S3: CD4 + T cell depletion platform and the efficacy of long-term CD4 + T cell depletion in mice, Table S1: Antibodies used in the study. | 5,517.4 | 2022-04-27T00:00:00.000 | [
"Biology",
"Medicine"
] |
Superalloy—Steel Joint in Microstructural and Mechanical Characterisation for Manufacturing Rotor Components
The structure of energy rotor components includes different structural materials in the sections, which are subjected to varying levels of thermal loading. The first component section has to include a precipitation-hardened nickel-based alloy, while the second one may be manufactured from other materials. Due to the installation cost, the use of expensive nickel-based materials is not recommended for applications in sections with a lower degree of thermal loading. Therefore, this aspect is still actually from an engineering point of view and is discussed in the paper by means of manufacturing and experimental approaches. The paper follows the welding problems related to a hybrid joint made of superalloy (Alloy 59) and hard rusting steel (S355J2W+N steel). The problem is solved using the MIG process at various parameters. With respect to the joint quality, microstructural features and mechanical parameters of the examined zone are presented. In the case of microstructure analysis, the dendritic and cellular natures of austenite were dominant elements of the joint. Mechanical tests have expressed a 50% reduction in elongation of the steel and alloy steel weld and lowering mechanical parameters. Mechanical parameters of the joint were on the level of their values observed for the steel, while the hardening coefficient followed the hardening curve of the alloy. Decohesion of the steel and mixed weld has reflected the constant proportion of values of axial and shear stress components up to the total separation. It is noted the tensile curves of the alloy and alloy steel joint follow a very similar shape, reporting the same response on the monotonic tension. The materials can be analysed by applying constitutive equations at very similar values of their coefficients. The obtained results enabled elaborating and examining the MIG welding process for thick-walled structures (not smaller than 8 mm) in detail giving all parameters required for technology. Finally, the technology for producing a hybrid joint using difficult-to-weld materials with different physical and mechanical properties, such as nickel alloys and low-alloy steels, is proposed. Results have shown it possible to develop a technology for producing of hybrid joints (supper alloy + hard rusting steel) with assumed physical and mechanical properties for rotors applied in the power boiler. This solution was proposed instead of previously used elements of rotors from expensive materials. It was assumed that the newly proposed and utilised method of welding will allow for obtaining good properties in terms of energy devices.
Introduction
Forged rotors, pipes and cast casings used in structures of high-temperature steam power plants as well as in elements of gas turbines or other rotating machines with operating temperatures >700 • C, should be made of nickel-based alloys with the required mechanical parameter and resistance on creep [1][2][3]. At temperatures exceeding the application range of high-temperature-resistant steels, nickel-based alloys are recommended
•
The lack of appropriate production equipment concerning a crack occurrence during the production process [7]; • The wide range of solidification of these alloy types leads to cracks [8]; • Welding of fully cured material promotes the formation of cracks as an effect of the relative inability of the material to compensate for the differential expansion [9].
For significant components such as impellers and housings, which are exposed to high and very high values of temperature, there are often regions with the most elevated and medium operating conditions [10]. In these cases, it is proposed to install elements in multiple subsections, where each area consists of materials with different mechanical, physical and microstructural properties. This structure creates welding problems [5,11].
Rotary air heaters (called rotors) are indispensable elements in the technological system of power boilers. They improve the functioning of the power installation because each degree of heat removed from the flue gas brings measurable economic benefits. The use of an air heater increase, the boiler efficiency by 1% for each 15-25 • C air temperature [5,12].
The durability of rotor welded joints determines the level of boiler efficiency. Structural elements require a welded connection between the section of thermally loaded components. They include a plurality of component sections which, during operation, are exposed to different temperature levels, a first component region being designed for temperatures of >750 • C, and a second component section being designed for temperatures of approximately 600 • C [6]. The first component section includes a precipitation-hardened nickel-based alloy (e.g., Alloy C-276 and Alloy 625 (2.4856) [13][14][15], and the second one represents a low-alloy steel S355J2W+N [16][17][18]. These elements are responsible for the increase in mechanical resistance on high-temperature, lowering cracks occurrence and good resistance to high-temperature corrosion [19].
Unfortunately, during the welding process, the crystallizing metal of the mixed welded joint becomes affected by low values of ultimate tensile stress, which are the result of weldfree shrinkage and cooling of not uniform heated base materials. It leads to cracks and final fractures [20,21].
The observation is in line with Schaeffler's graph, presented in [22], when welding nickel alloys with steels, an analysis of the Ni-Fe equilibrium graph is insufficient. According to the graph the verification of other elements, especially ferrite-forming ones: chromium and molybdenum are possible. Assuming that molybdenum is the equivalent of chromium ( [22], Figure 1), the Fe-Ni-Cr ternary equilibrium system can be analysed ( Figure 2). There is no developed simple method of joining various materials of complex nickel alloys with other steels [10,11,[24][25][26].
It can be found in patent JP4206216B2 [1]; the solution makes it possible to join steel with a nickel-based alloy only under the condition of producing the so-called interlayers. This approach can only be applied to flow machines in the power industry.
The hybrid joint of high-alloy, heat-resistant martensitic-ferritic steels and nickelbased superalloys requires several preparatory steps [27]. The joined parts, e.g., IN706 with high-alloy martensitic-ferritic steel St13TNiEL, are first coated by SG-NiCr20Nb surfacing in the area of material joining, and then the material (clad/surfaced) is subjected to high-quality heat treatment (stabilizing annealing at 820 ± 15 °C, cooling to RT, precipitation hardening at 730 ± 15 °C, cooling to RT). During welding, the root layers are applied using the TIG method, and the reinforcement layers are applied using submerged arc welding.
The excellent weldability of the superalloy is a result of its microstructure. According to the diffraction pattern, the microstructure of Inconel 718 contains a dispersion of the γ′ and the γ″ of precipitates in the γ matrix [28][29][30]. Nickel forms solid solutions with copper and saturated solid solutions with the most important alloying elements, including chromium, molybdenum, iron and copper [31,32]. Of these elements, chromium dissolves best in the nickel A1 network (up to 29%). When the concentration of chromium is higher, the α phase that mainly contains chromium, which does not dissolve nickel, becomes present [33,34]. The welding elements from the same material (especially Monel and Inconel) are relatively well-weldable by the most popular arc processes, i.e., MIG, TIG at coated electrodes, covered arc and laser welding [26,35]. There is no developed simple method of joining various materials of complex nickel alloys with other steels [10,11,[24][25][26].
It can be found in patent JP4206216B2 [1]; the solution makes it possible to join steel with a nickel-based alloy only under the condition of producing the so-called interlayers. This approach can only be applied to flow machines in the power industry.
The hybrid joint of high-alloy, heat-resistant martensitic-ferritic steels and nickel-based superalloys requires several preparatory steps [27]. The joined parts, e.g., IN706 with highalloy martensitic-ferritic steel St13TNiEL, are first coated by SG-NiCr20Nb surfacing in the area of material joining, and then the material (clad/surfaced) is subjected to high-quality heat treatment (stabilizing annealing at 820 ± 15 • C, cooling to RT, precipitation hardening at 730 ± 15 • C, cooling to RT). During welding, the root layers are applied using the TIG method, and the reinforcement layers are applied using submerged arc welding.
The excellent weldability of the superalloy is a result of its microstructure. According to the diffraction pattern, the microstructure of Inconel 718 contains a dispersion of the γ and the γ" of precipitates in the γ matrix [28][29][30]. Nickel forms solid solutions with copper and saturated solid solutions with the most important alloying elements, including chromium, molybdenum, iron and copper [31,32]. Of these elements, chromium dissolves best in the nickel A1 network (up to 29%). When the concentration of chromium is higher, the α phase that mainly contains chromium, which does not dissolve nickel, becomes present [33,34]. The welding elements from the same material (especially Monel and Inconel) are relatively well-weldable by the most popular arc processes, i.e., MIG, TIG at coated electrodes, covered arc and laser welding [26,35].
The highest tendency to crack is those welds in which cell microstructure is formed during clotting [36][37][38][39]. The cracking of these welds is favoured by relatively smooth surfaces of grain boundaries, with strong segregation of low-melting components. Authors K. Rajasekhar, C. S. Harendranath and R. Raman indicate that to avoid cracking in the joint, cell-dendritic microstructure should be formed during the solidification of the weld [38].
The authors: Yang Li Shuangming et al. demonstrated the concentration of lowmelting phases per unit area decreases, reducing the tendency to crack [37]. The data presented in [40][41][42] exhibits that the joint including 40% of nickel content has a tendency to crack.
The last assumption implies another problem in developing a technology for joining the two materials because of enormously different values of thermal coefficients. The average coefficient of Alloy 59 thermal expansion is about 13.1 × 10 −6 1/K, whereas for hard-rusting steels, it is around 10.3 × 10 −6 1/K. As a result, in thick-walled joints, a risk of cracks appearing during the solidification of the weld (crystallisation cracks) or during reheating of base material and weld (segregation cracks) occurs.
It is important to approach the welding of two-phase materials dominated by austenite and δ-ferrites distributed at austenite grain boundaries. The authors of [64] point out that during stainless steel laser arc hybrid welding, welding cracks may occur due to the presence of δ-ferrite. The same phases play a significant role in DWJ welding non-alloy steel with nickel alloy.
In turn, the authors of [62] investigated the behaviour of bimetallic joints CP-Ti/Q235 bimetallic sheets.
Precisely changed various process parameters will allow for the production of a joint of the best quality and properties.
The analysis of the literature and published patents in the field of joining difficult-toweld nickel alloys with steels clearly shows the lack of guidelines for the welding process and the use of relatively simple methods to obtain correct welds, which can be considered a research gap.
Therefore, this article aims to propose the welding process (MIG) for hybrid joints represented by superalloy and rust resistance low alloy steel. It was assumed that the newly proposed and used method of welding will allow for obtaining good properties in terms of energy devices.
The TIG process is well recognised for the considered problem, but in the article, the authors focused on the MIG welding process. The MIG process is more efficient. The authors presented the role of varying electrode wires and shielding gas mixtures. They have attempted to prove that the MIG method is correct for welding dissimilar joints based on the steel-nickel alloy. Serious attention is paid to the method of bevelling both sheets due to a different heat transfer coefficient, much higher for nickel alloys (90 W/(m·K).) than for low-alloy steel (60 W/(m·K) [65,66].
Materials, Technology and Methods
Manufacturing the correct DWJ (dissimilar welded joint) is very often a big problem, due to the different materials' microstructure and their mechanical and physical properties, such as density and thermal conductivity. In the case of the DWJ low-alloy steel and nickel alloy joint, it is important to consider into account the significant difference in heat conduction (1.5 greater for nickel alloys) between the two materials, which may result in high welding stresses, which may provoke cracks. So, as a part of this article, the investigation schema including metallurgical and technological analyses was realised.
In the metallurgical approach, various wires and shielding gases were selected, and as part of the technological approach, the focus was on the role of bevelling and basic welding parameters: arc voltage, current intensity and welding speed. The welding scheme is shown in Figures 3 and 4. ials 2023, 16, x FOR PEER REVIEW In the metallurgical approach, various wires and shielding g as part of the technological approach, the focus was on the role welding parameters: arc voltage, current intensity and weldin scheme is shown in Figures 3 and 4. The tested object for the study was a mixed welded j method, Figures 3 and 4. The joints were made from an 8 mm ufactured through the two electrode wires, i.e., NiCr23Mo1 mate tensile strength = 700 MPa [48]), and G19-9NbSi (yield sile strength = 630 MPa [46]). The NiCr23Mo16 wire is used i ments. It is recommended for joining duplex and super dupl nickel-based alloys [34]. The G19-9NbSi wire is successfully loyed steels with the superalloy [20,30].
The preparation of the three-stitched joints, and the b sheet, are presented in Figures 3 and 4.
The chamfering at the angle of 45° is intended to redu crack. In the welding process, an additional distance of 2 mm allowed for limiting the degree of mixing in the weld of base tion and welding wire during the welding process. The stitch in Figure 4.
In this case worth to notice various methods of bevell bevel did not give incorrect results. The one-sided bevelling ditions of heat distribution during welding.
The welding parameters were as follows: the electrode the arc voltage U = 21 V and the welding current was diffe layers, Figure 3. In the lower stitch arrangement, the I3 curre while the I2 current in the two upper layers ranged from 120 had dimensions of 800 mm × 200 mm × 8 mm, and the weld h In the MIG process, the following mixtures were used as sh and 90% Ar-10% He. The shielding gas flow rate was at the was made with variable tested speed V3: 200-230 mm/min 270 mm/min (upper layer). MIG welding method (131) in selected according to the requirements of EN 15614-1 norm. R The tested object for the study was a mixed welded joint made by the MIG (131) method, Figures 3 and 4. The joints were made from an 8 mm thick metal sheet and manufactured through the two electrode wires, i.e., NiCr23Mo16 (yield stress 450 MPa, ultimate tensile strength = 700 MPa [48]), and G19-9NbSi (yield stress 460 MPa, ultimate tensile strength = 630 MPa [46]). The NiCr23Mo16 wire is used in highly aggressive environments. It is recommended for joining duplex and super duplex steels, stainless steels and nickel-based alloys [34]. The G19-9NbSi wire is successfully employed for welding unalloyed steels with the superalloy [20,30].
The preparation of the three-stitched joints, and the bevelling method of the steel sheet, are presented in Figures 3 and 4.
The chamfering at the angle of 45 • is intended to reduce the tendency of a joint to crack. In the welding process, an additional distance of 2 mm was assumed, Figure 3. This allowed for limiting the degree of mixing in the weld of base material chemical composition and welding wire during the welding process. The stitches laying order is presented in Figure 4.
In this case worth to notice various methods of bevelling were examined. Double bevel did not give incorrect results. The one-sided bevelling was used to equalise the conditions of heat distribution during welding.
The welding parameters were as follows: the electrode wire diameter was 1.2 mm, the arc voltage U = 21 V and the welding current was different in the root and the face layers, Figure 3. In the lower stitch arrangement, the I 3 current ranged from 130 to 150 A, while the I 2 current in the two upper layers ranged from 120 to 150 A. The welded sheets had dimensions of 800 mm × 200 mm × 8 mm, and the weld had a three-stitched character. In the MIG process, the following mixtures were used as shielding gases: 95% Ar-5% He and 90% Ar-10% He. The shielding gas flow rate was at the level of 15 L/min. The joint was made with variable tested speed V 3 : 200-230 mm/min (bottom layer) and V 1,2 : 220-270 mm/min (upper layer). MIG welding method (131) in the down position (PA) was selected according to the requirements of EN 15614-1 norm. Rotor joints were welded with direct current with a positive polarity on the electrode. Preheating was not applied.
Tests Details
It was decided to verify the weldability of the joint made of Alloy 59 and S355J2W+N low-alloy steel. After producing welded joints using different parameters (two different electrode wires and two different shielding mixtures Ar-He), visual tests were carried out (PN-EN 970: 1999 standard [67]).
The tests aimed to assess the correctness of joints, identify incompatibilities in the form of cracks, and eliminate any incorrectly made connections. The analysis was expanded, including the results of non-destructive tests: penetration (PN-EN 571: 1999 standard [68]) and ultrasonic (PN-EN 1714: 2002 standard [69]). The test results were documented as macroscopic images (PN-EN 1321: 2000 standard [70]).
Connections that did not present macroscopic changes in the form of cold, hot and lamellar cracks were qualified for additional mechanical (bending, hardness, tensile tests) and microscopic observations. The joints presenting the best mechanical properties were subjected to microscopic inspection to examine the influence of welding parameters on the microstructure. The proposed evaluation results were the prerequisite for selecting parameters and processes required for manufacturing a mixed butt welded joint of a rotor structure.
The geometry of the hourglass specimen allows us to follow the requirements for fatigue tests, shown in American standards, i.e., ASTM-E468 [71] and ASTM-E466 [72]. This geometry is expressed by the shape of the measurement section and the value of the radius. These details were employed because the middle zone of the measurement section can be easily used for a weld, and this region can be easily subjected to loading without influencing the other specimen sections on results.
The specimens for the mechanical test were randomly selected from a pool of 20 specimens previously inspected for defects. In practical terms, these specimens were defect-free, i.e., identical in quality.
Tensile tests were carried out considering the requirements of PN-EN ISO 6892-1: 2020 standard [73], at room temperature employing hourglass specimens and servo-hydraulic testing machines denoted by 8802 Instron, Figure 5. The testing machine was equipped with an alignment system to avoid the bending moment. Bluehill Instron software was used to elaborate stages of the tensile tests up to fracture. Specimens were directly mounted in hydraulic grips. Nominal dimensions of the specimens in the minimum cross-section were represented by the following values: 5 × 5 [mm], Figure 5a. A hybrid joint was located in the middle section of the measurement region and was directly subjected to loading. The testing machine was tuned at a close-loop feedback signal for the displacement velocity of 2 mm/min. Measurements of an axial strain were conducted using the extensometer technique using the 2620-601 Instron sensor, Figure 5b,c. This device has enabled capturing values of the axial strain at the gauge length equal to 50 mm. Concerning the range of the strain measurement, the extensometer has allowed collection results up to 10% strain, i.e., 5 mm. Therefore, the experimental programme has contained the stages for the extensometer removal at the value of elongation mentioned. The tested materials' behaviour under a tensile force was not only recorded in the form of digital results but also it was expressed by photos from macro-photography techniques, collecting details of the measurement sections and fracture zones.
Once the joints were welded with the use of various parameters (two different electrode wires, three different shielding mixtures, different linear energy of the process), the visual form of cracks, and eliminate any incorrectly made connections. The analysis was e panded, including the results of non-destructive tests: penetration (PN-EN 571: 19 standard [68]) and ultrasonic (PN-EN 1714: 2002 standard [69]). The test results were do umented as macroscopic images (PN-EN 1321: 2000 standard [70]).
Connections that did not present macroscopic changes in the form of cold, hot a lamellar cracks were qualified for additional mechanical (bending, hardness, tensile tes and microscopic observations. The joints presenting the best mechanical properties we subjected to microscopic inspection to examine the influence of welding parameters the microstructure. The proposed evaluation results were the prerequisite for selecti parameters and processes required for manufacturing a mixed butt welded joint of a ro structure.
The geometry of the hourglass specimen allows us to follow the requirements f fatigue tests, shown in American standards, i.e., ASTM-E468 [71] and ASTM-E466 [7 This geometry is expressed by the shape of the measurement section and the value of t radius. These details were employed because the middle zone of the measurement secti can be easily used for a weld, and this region can be easily subjected to loading witho influencing the other specimen sections on results.
The specimens for the mechanical test were randomly selected from a pool of 20 sp imens previously inspected for defects. In practical terms, these specimens were defe free, i.e., identical in quality.
Tensile tests were carried out considering the requirements of PN-EN ISO 689 1:2020 standard [73], at room temperature employing hourglass specimens and servo-h draulic testing machines denoted by 8802 Instron, Figure 5. The testing machine w equipped with an alignment system to avoid the bending moment. Bluehill Instron so ware was used to elaborate stages of the tensile tests up to fracture. Specimens were rectly mounted in hydraulic grips. Nominal dimensions of the specimens in the minimu cross-section were represented by the following values: 5 × 5 [mm], Figure 5a. A hybr joint was located in the middle section of the measurement region and was directly su jected to loading. The testing machine was tuned at a close-loop feedback signal for t displacement velocity of 2 mm/min. Measurements of an axial strain were conducted u ing the extensometer technique using the 2620-601 Instron sensor, Figure 5b,c. This dev has enabled capturing values of the axial strain at the gauge length equal to 50 mm. Co cerning the range of the strain measurement, the extensometer has allowed collection sults up to 10% strain, i.e., 5 mm. Therefore, the experimental programme has contain the stages for the extensometer removal at the value of elongation mentioned. The test materials' behaviour under a tensile force was not only recorded in the form of digi results but also it was expressed by photos from macro-photography techniques, colle ing details of the measurement sections and fracture zones.
Weld Inspection, Hardness and Microstructure
Based on the inspection, it was found that:
•
There is an occurrence of small cracks in joints made with G19-9NbSi austenitic wire using the following shielding gases: argon and Ar-10% He; For the further approaches (bending test), only joints without defects and incompatibilities were selected. The joints manufactured at G19-9NbSi austenitic electrode wire and the Ar-10% He shielding mixture were not satisfactory for the experimental procedure. The use of austenitic wire during the MIG welding method (131) did not produce connections characterised by the desired level of quality B (according to PN-EN ISO 5817: 2005). Due to unsatisfactory results of the external examination of joints and cracks, connections made with austenitic wire were rejected. For the other joints examined, a bending test was carried out by the PN-EN ISO 5173: 2010 standard [75]. For the tests, a specimen with thickness a = 8 mm, width b = 10 mm, mandrel d = 32 mm, roller distance 54 mm and bending angle of 180° was used. Five measurements of the bending test were made from the face and the root side of the weld. Only when the G19 -9NbSi austenitic electrode wire and the Ar-5% He mixture were used cracks in the weld were observed at a bending angle above 130°. As a result, it can be concluded that the adopted welding parameters made with austenitic wire will have lower performance properties than welds made with the same welding current-voltage parameters and a different welding wire.
Weld Inspection, Hardness and Microstructure
Based on the inspection, it was found that:
•
There is an occurrence of small cracks in joints made with G19-9NbSi austenitic wire using the following shielding gases: argon and Ar-10% He; For the further approaches (bending test), only joints without defects and incompatibilities were selected. The joints manufactured at G19-9NbSi austenitic electrode wire and the Ar-10% He shielding mixture were not satisfactory for the experimental procedure. The use of austenitic wire during the MIG welding method (131) did not produce connections characterised by the desired level of quality B (according to PN-EN ISO 5817: 2005). Due to unsatisfactory results of the external examination of joints and cracks, connections made with austenitic wire were rejected. For the other joints examined, a bending test was carried out by the PN-EN ISO 5173: 2010 standard [75]. For the tests, a specimen with thickness a = 8 mm, width b = 10 mm, mandrel d = 32 mm, roller distance 54 mm and bending angle of 180 • was used. Five measurements of the bending test were made from the face and the root side of the weld. Only when the G19 -9NbSi austenitic electrode wire and the Ar-5% He mixture were used cracks in the weld were observed at a bending angle above 130 • . As a result, it can be concluded that the adopted welding parameters made with austenitic wire will have lower performance properties than welds made with the same welding current-voltage parameters and a different welding wire.
When a nickel-based electrode wire NiCr23Mo16 was used (together with the two tested shielding gases), no cracks or other incompatibilities were found in the tested specimens. Both the macroscopic observations and the bending test results showed that the welded joints were made accurately, the selected welding parameters were correct, and the adopted bevelling angle and NiCr23Mo16 wire allowed us to obtain joints with the required quality level.
Due to the occurring defects in the joint made with the use of austenitic electrode wire G19-9NbSi, in the further part of the study, it was decided to analyse only those joints made with the nickel-based wire NiCr23Mo16. Tests of immediate tensile strength were carried out. They were performed on a ZWICK 100N5A strength-testing machine.
Data analysis (Table 2) shows that the welds are made correctly. All joints have comparable mechanical properties (YS above 320 MPa; UTS above 520 MPa). The table data also expresses that welds made with lesser linear energy have higher YS values (lower current, higher speed). Joint yield stress should be above 355 MPa (in accordance with the symbol and requirements of S355JR+N steel). Such a high value of yield stress can be obtained when welding with lower linear energy and the use of a shielding compound Ar-5% He. It can also be observed that the shielding mixture of 95% Ar-5% He is the most advantageous due to the highest value of ultimate tensile strength (553 MPa). The introduction of helium into the mixture in a small amount affects the shape of the weld, increasing its concavity, which according to the literature data [22], is very beneficial. Helium has a higher heat transfer coefficient than argon. As a result, adding more helium to an argon mix may affect the grinding of grain in the weld. The helium content in the argon mixes up to 10% might be considered ineffective, as it does not guarantee yield stress of 355 MPa and, at the same time, does provide a further increase in joint strength. It was observed that in all the studied cases, welding with lower linear energy is more advantageous. The use of NiCr23Mo16 electrode wire together with the Ar-5% He gas mixture is the most appropriate, as it allows us to obtain the highest values of yield stress (365 MPa) and ultimate tensile strength (550 MPa) of the joint.
It is worth noticing that, with respect to the paper's aim, the SEM method was not used, but the macro-photography technique ( Figure 6) was applied to collect details of the joint manufactured. This has enabled us to follow the weld quality and qualified it for the mechanical tests and microstructural observations. Vickers steel hardness amounted to 185 MPa, whereas Alloy 59 hardness was equal to 375 MPa. Table 3 presents the results of hardness tests in the heat-affected zone from both welded sides and the weld hardness in the six tested joints (electrode wire based on NiCr23Mo16 nickel alloy, two gas shielding mixtures and two different welding linear energies). The table data shows that an increase in the helium content in the Ar-He mix has a direct influence on the increase in hardness value. The most favourable results were Vickers steel hardness amounted to 185 MPa, whereas Alloy 59 hardness was equal to 375 MPa. Table 3 presents the results of hardness tests in the heat-affected zone from both welded sides and the weld hardness in the six tested joints (electrode wire based on NiCr23Mo16 nickel alloy, two gas shielding mixtures and two different welding linear energies). The table data shows that an increase in the helium content in the Ar-He mix has a direct influence on the increase in hardness value. The most favourable results were obtained for joints made using an Ar-5% He mixer. An increase in the welding linear energy does not cause any noticeable changes in the hardness of the weld, Table 3. Table 3. Hardness test results on Vickers method, HAZ-heat-affected zone.
Parameters of:
Hardness in Point: Vickers steel hardness amounted to 185 MPa, whereas Alloy 59 hardness was equal to 375 MPa. Table 3 presents the results of hardness tests in the heat-affected zone from both welded sides and the weld hardness in the six tested joints (electrode wire based on NiCr23Mo16 nickel alloy, two gas shielding mixtures and two different welding linear energies). The table data shows that an increase in the helium content in the Ar-He mix has a direct influence on the increase in hardness value. The most favourable results were obtained for joints made using an Ar-5% He mixer. An increase in the welding linear energy does not cause any noticeable changes in the hardness of the weld, Table 3. The analysis of Table 3 shows that the joint is made correctly. Hardness test results on the Vickers method show that obtained joints using gas mixture Ar-5% He is more beneficial (because the hardness difference in tested join zones is the lowest), Figures 3 and 6.
Based on the analysis of table data, the following welding parameters were finally selected for making joints for all further tests: The analysis of Table 3 shows that the joint is made correctly. Hardness test results on the Vickers method show that obtained joints using gas mixture Ar-5% He is more beneficial (because the hardness difference in tested join zones is the lowest), Figures 3 and 6.
Based on the analysis of table data, the following welding parameters were finally selected for making joints for all further tests:
The fuse was complete with a clear fusion line. This indicated that both the bevelling method and the welding parameters were selected properly. Production of a welded joint is not easy because of the different microstructures of both materials and other heat transfer coefficients. The thermal conductivity of steel amounts to 60 W/(m·K), whereas the thermal conductivity of the alloy is at the level of 90 W/(m·K). Steel S355J2W+N has a ferritic-pearlitic microstructure (Figure 7), while Alloy 59 has a single-phase austenitic microstructure with good solubility of alloying elements in the FCC nickel network, Figure 8. the thermal conductivity of the alloy is at the level of 90 W/(m·K). Steel S355J2W+N has a ferritic-pearlitic microstructure (Figure 7), while Alloy 59 has a single-phase austenitic microstructure with good solubility of alloying elements in the FCC nickel network, Figure 8. The microstructure observations were performed on the LM (light microscopy) observation under various magnifications. The specimen was digested in Adler's A11 reagent. Figure 8 shows the ferritic-pearlitic microstructure of the base material. the thermal conductivity of the alloy is at the level of 90 W/(m·K). Steel S355J2W+N has a ferritic-pearlitic microstructure (Figure 7), while Alloy 59 has a single-phase austenitic microstructure with good solubility of alloying elements in the FCC nickel network, Figure 8. Figure 8 shows the ferritic-pearlitic microstructure of the base material. The microstructure observations were performed on the LM (light microscopy) observation under various magnifications. The specimen was digested in Adler's A11 reagent. Figure 8 shows the ferritic-pearlitic microstructure of the base material. Figure 9a shows the microstructure of the joint from the S355J2W+N steel side, whereas Figure 9b shows the section of the fusion line from the Alloy 59 side. coarse ferrite and MAC phases (martensite, residual austenite, carbides), whereas in the superalloy base material, a two-phase austenitic-ferritic microstructure appeared. The observation of the weld on a micro-scale confirms that the joint was manufactured correctly. In addition to the comment on the fusion line, it was decided to verify the joint microstructure (under various magnifications) in the central part of the weld, Figure 9. Figure 8 shows a single-phase, semi-similar austenitic microstructure of Alloy 59. The microstructure of both base materials changes when approaching the weld. The analysis shows the base material adjacent to the fusion line has a high content of coarse ferrite and MAC phases (martensite, residual austenite, carbides), whereas in the superalloy base material, a two-phase austenitic-ferritic microstructure appeared. The observation of the weld on a micro-scale confirms that the joint was manufactured correctly. In addition to the comment on the fusion line, it was decided to verify the joint microstructure (under various magnifications) in the central part of the weld, Figure 9. Figure 8 shows a single-phase, semi-similar austenitic microstructure of Alloy 59. The microstructure of both base materials changes when approaching the weld. Figure 10 shows an austenitic-ferritic microstructure with a favourable dendritic cell formation. Based on the calculations of iron (about 30%), nickel (40%) and chromium equivalents, as well as the equilibrium graph (Figure 1), it can be concluded that the weld microstructure should contain about 95% of austenite and about 5% of delta ferrite. The dendritic and cellular natures of austenite translate into the excellent plastic properties of a joint. Figure 9a shows the microstructure of the joint from the S355J2W+N steel side, whereas Figure 9b shows the section of the fusion line from the Alloy 59 side.
The analysis shows the base material adjacent to the fusion line has a high content of coarse ferrite and MAC phases (martensite, residual austenite, carbides), whereas in the superalloy base material, a two-phase austenitic-ferritic microstructure appeared. The observation of the weld on a micro-scale confirms that the joint was manufactured correctly. In addition to the comment on the fusion line, it was decided to verify the joint microstructure (under various magnifications) in the central part of the weld, Figure 9. Figure 8 shows a single-phase, semi-similar austenitic microstructure of Alloy 59. The microstructure of both base materials changes when approaching the weld. The obtained results have allowed us to select the process parameters with respect to the high quality of the joint. It was decided to perform the mechanical resistance of the hybrid joint in detailed tests.
Joints should be made of low-alloy steel with low-alloy steel and Alloy 59 with Alloy 59 to determine whether the hybrid joints will have mechanical properties (as a result or similar) to one of the joints. The same parameters of the welding process were used to make all the joints in that part of the investigation:
The Base Metal and MIG Hybrid Weld under Tensile Force
The behaviour of the base metals and weld was expressed by the stress-strain relationship up to fracture and mechanical parameters from an elastic and elastic-plastic state, Figures 11-15. In the case of Alloy 59 (as the base metal), the value of axial strain was the biggest one, and it exceeded its limited value related to the extensometer used, Figure 11b. Therefore, for the experimental way, values of strain were calculated based on the values of strain from the extensometer and displacement from a linear sensor of the testing machine, comparing the stress-strain relationship collected at both measurement sensor activities.
The obtained results have allowed us to select the process parameters with respect to the high quality of the joint. It was decided to perform the mechanical resistance of the hybrid joint in detailed tests.
Joints should be made of low-alloy steel with low-alloy steel and Alloy 59 with Alloy 59 to determine whether the hybrid joints will have mechanical properties (as a result or similar) to one of the joints. The same parameters of the welding process were used to make all the joints in that part of the investigation:
The Base Metal and MIG Hybrid Weld under Tensile Force
The behaviour of the base metals and weld was expressed by the stress-strain relationship up to fracture and mechanical parameters from an elastic and elastic-plastic state, Figures 11-15. In the case of Alloy 59 (as the base metal), the value of axial strain was the biggest one, and it exceeded its limited value related to the extensometer used, Figure 11b. Therefore, for the experimental way, values of strain were calculated based on the values of strain from the extensometer and displacement from a linear sensor of the testing machine, comparing the stress-strain relationship collected at both measurement sensor activities.
(a) (b) Figure 11. Alloy 59 specimen before the tensile test (a); a part of the stress-strain characteristic of the alloy joint up to the extensometer removing (b). Figure 11. Alloy 59 specimen before the tensile test (a); a part of the stress-strain characteristic of the alloy joint up to the extensometer removing (b). The final results of this approach are presented in the form of the tensile curve and mechanical parameters of Alloy 59, Figure 12. It can be noticed that the superalloy has obtained a wide range of elastic regions between values of proportional limit and elastic limit, i.e., represented by almost 100 MPa. This value follows the 50% value of the proportional limit and takes a significant meaning in the elastic behaviour of the joint tested. The other section of the tensile curve is dominant, expressing a wide range of hardening represented by 405 MPa. This hardening can be directly taken to the engineering approach concerning construction safety because of the operation state with plastic deformation, the final fracture can be noticed by many measurement techniques, enabling to avoid unexpected failure. Analysing the fracture zone allows for distinguishing the stress component directly connected with the weld degradation. In this case, the shear stress created the final cracking, Figure 12.
In the case of the S355J2W+N steel, the elastic region was limited by a value close to 290 MPa, while the elastic-plastic with hardening is differenced by 250 MPa, Figure 13. These values have enabled us to write as follows: the elastic and elastic-plastic sections of the tensile curve are significant in the weld behaviour. The last region of the characteristic directly expressed the unstable behaviour of the joint, reflecting the neck effect at an extensive range of strain and stress, i.e., 6% (66% of the final plastic deformation) and 454 MPa, respectively. It means the inspection of a component having this type of weld should be carried out carefully and more often than in the case of a typical one because avoiding the stage related to the necking.
The behaviour of the mixed Alloy 59 and S355J2W+N steel weld ( Figure 14) reflected that in the comparison to the proportional section, the elastic region was represented by smaller values compared to data of the welded base metals: Figures 12 and 13.
The elastic-plastic part of the curve up to the ultimate tensile curve was the critical section of the characteristic considered. This was denoted by the value of a stress range equal to 311 MPa and 80% of the plastic deformation. It can be concluded the tested weld response with respect to plastic features is very similar to the MIG joint made of steel. The fracturing of the hybrid weld was mixed, containing brittle-ductile features created by both stress components, i.e., axial and shear, Figure 15.
Comparing data for the weld types, represented by ultimate tensile strength and yield stress, it can be observed the proportion of the mechanical parameters considered for the alloy and hybrid welds is very similar, while in the case of the steel, it is 30% lower, Figure 16. This indicates the hardening curves can have very similar features, besides the The final results of this approach are presented in the form of the tensile curve and mechanical parameters of Alloy 59, Figure 12. It can be noticed that the superalloy has obtained a wide range of elastic regions between values of proportional limit and elastic limit, i.e., represented by almost 100 MPa. This value follows the 50% value of the proportional limit and takes a significant meaning in the elastic behaviour of the joint tested. The other section of the tensile curve is dominant, expressing a wide range of hardening represented by 405 MPa. This hardening can be directly taken to the engineering approach concerning construction safety because of the operation state with plastic deformation, the final fracture can be noticed by many measurement techniques, enabling to avoid unexpected failure. Analysing the fracture zone allows for distinguishing the stress component directly connected with the weld degradation. In this case, the shear stress created the final cracking, Figure 12.
In the case of the S355J2W+N steel, the elastic region was limited by a value close to 290 MPa, while the elastic-plastic with hardening is differenced by 250 MPa, Figure 13. These values have enabled us to write as follows: the elastic and elastic-plastic sections of the tensile curve are significant in the weld behaviour. The last region of the characteristic directly expressed the unstable behaviour of the joint, reflecting the neck effect at an extensive range of strain and stress, i.e., 6% (66% of the final plastic deformation) and 454 MPa, respectively. It means the inspection of a component having this type of weld should be carried out carefully and more often than in the case of a typical one because avoiding the stage related to the necking.
The behaviour of the mixed Alloy 59 and S355J2W+N steel weld ( Figure 14) reflected that in the comparison to the proportional section, the elastic region was represented by smaller values compared to data of the welded base metals: Figures 12 and 13.
The elastic-plastic part of the curve up to the ultimate tensile curve was the critical section of the characteristic considered. This was denoted by the value of a stress range equal to 311 MPa and 80% of the plastic deformation. It can be concluded the tested weld response with respect to plastic features is very similar to the MIG joint made of steel. The fracturing of the hybrid weld was mixed, containing brittle-ductile features created by both stress components, i.e., axial and shear, Figure 15.
Comparing data for the weld types, represented by ultimate tensile strength and yield stress, it can be observed the proportion of the mechanical parameters considered for the alloy and hybrid welds is very similar, while in the case of the steel, it is 30% lower, Figure 16. This indicates the hardening curves can have very similar features, besides the weld types being enormously different. It was checked and confirmed using a power law (σ = Kε n , K-strength coefficient, n-strain hardening exponent, [41,42], calculating all coefficients of the equation at the true stress-true strain relationship ranged by yield stress (YS) and ultimate tensile strength (UTS), Table 4, Figure 17. As can be noticed, in the case of the alloy and the welds for alloy steel the values of the coefficients are very similar. It means the stress-strain relationship for the Alloy 59 and the weld being its combination with the S355J2W+N steel has a similar path for its shape, indicating the examined regions express almost the same response on the tensile and by this, they can be analysed using constitutive equations having very similar values of coefficients. weld types being enormously different. It was checked and confirmed using a power law ( , K-strength coefficient, n-strain hardening exponent, [41,42], calculating all coefficients of the equation at the true stress-true strain relationship ranged by yield stress (YS) and ultimate tensile strength (UTS), Table 4, Figure 17. As can be noticed, in the case of the alloy and the welds for alloy steel the values of the coefficients are very similar. It means the stress-strain relationship for the Alloy 59 and the weld being its combination with the S355J2W+N steel has a similar path for its shape, indicating the examined regions express almost the same response on the tensile and by this, they can be analysed using constitutive equations having very similar values of coefficients. Some conclusions on the behaviour of the weld tested can also be captured based on the engineering tensile curves as data for the typical approaches for analysis of mechanical resistance of joints and welded components under various types of loading using theoretical and numerical stages, Figure 18. They are expressed by values of a relative weld types being enormously different. It was checked and confirmed using a power law ( , K-strength coefficient, n-strain hardening exponent, [41,42], calculating all coefficients of the equation at the true stress-true strain relationship ranged by yield stress (YS) and ultimate tensile strength (UTS), Table 4, Figure 17. As can be noticed, in the case of the alloy and the welds for alloy steel the values of the coefficients are very similar. It means the stress-strain relationship for the Alloy 59 and the weld being its combination with the S355J2W+N steel has a similar path for its shape, indicating the examined regions express almost the same response on the tensile and by this, they can be analysed using constitutive equations having very similar values of coefficients. Some conclusions on the behaviour of the weld tested can also be captured based on the engineering tensile curves as data for the typical approaches for analysis of mechanical resistance of joints and welded components under various types of loading using theoretical and numerical stages, Figure 18. They are expressed by values of a relative Some conclusions on the behaviour of the weld tested can also be captured based on the engineering tensile curves as data for the typical approaches for analysis of mechanical resistance of joints and welded components under various types of loading using theoretical and numerical stages, Figure 18. They are expressed by values of a relative strain as well as values of ultimate tensile strength. As it can be noticed, in the case of the steel and its connection with alloy, a reduction of elongation was expressed by 50% compared to data for the alloy, while the ultimate tensile strength was only lowered by 30%. Moreover, the hybrid joint can be called the weakest weld because this zone has reached the smallest values of proportional limit, elastic limit, yield stress and ultimate tensile strength. This result is confirmed in values of energy related to elastic limit and yield stress, Figure 19a, while the energy values at the ultimate tensile strength of the hybrid joint were not the smallest ones, Figure 19b. Nevertheless, in engineering practice, a material behaviour at an elastic state plays an essential role in modelling, designing and operating, therefore at the smallest values of data from the state considered and the 33% difference in energy value for the ultimate tensile strength of the hybrid joint. strain as well as values of ultimate tensile strength. As it can be noticed, in the case of the steel and its connection with alloy, a reduction of elongation was expressed by 50% compared to data for the alloy, while the ultimate tensile strength was only lowered by 30%. Moreover, the hybrid joint can be called the weakest weld because this zone has reached the smallest values of proportional limit, elastic limit, yield stress and ultimate tensile strength. This result is confirmed in values of energy related to elastic limit and yield stress, Figure 19a, while the energy values at the ultimate tensile strength of the hybrid joint were not the smallest ones, Figure 19b. Nevertheless, in engineering practice, a material behaviour at an elastic state plays an essential role in modelling, designing and operating, therefore at the smallest values of data from the state considered and the 33% difference in energy value for the ultimate tensile strength of the hybrid joint. Other details on the behaviour of the tested regions can be captured in the analysis of fracture regions, Figure 20. They are connected with the geometrical features of the zones considered because changes in fracturing sections reflect variations in stress state strain as well as values of ultimate tensile strength. As it can be noticed, in the case of the steel and its connection with alloy, a reduction of elongation was expressed by 50% compared to data for the alloy, while the ultimate tensile strength was only lowered by 30%. Moreover, the hybrid joint can be called the weakest weld because this zone has reached the smallest values of proportional limit, elastic limit, yield stress and ultimate tensile strength. This result is confirmed in values of energy related to elastic limit and yield stress, Figure 19a, while the energy values at the ultimate tensile strength of the hybrid joint were not the smallest ones, Figure 19b. Nevertheless, in engineering practice, a material behaviour at an elastic state plays an essential role in modelling, designing and operating, therefore at the smallest values of data from the state considered and the 33% difference in energy value for the ultimate tensile strength of the hybrid joint. Other details on the behaviour of the tested regions can be captured in the analysis of fracture regions, Figure 20. They are connected with the geometrical features of the zones considered because changes in fracturing sections reflect variations in stress state Other details on the behaviour of the tested regions can be captured in the analysis of fracture regions, Figure 20. They are connected with the geometrical features of the zones considered because changes in fracturing sections reflect variations in stress state components and enable formulating the conclusion on the role of stress type in the zones' degradation. Looking at the fracture region of Alloy 59 (Figure 20a), the multi-planar degradation can be indicated as the main feature due to the loading type used. In contrast, in the case of S355J2W+N steel (Figure 20b) and the steel alloy (Figure 20c) joint, the region is represented by a one-fracture plane. Therefore, the following sentences can be formulated: (a) Reorientation of axial and shear stress components follow the degradation of the Alloy 59 as well as differences in their values as stress state components; (b) In the case of the weld manufactured by means of Alloy 59 and S355J2W+N steel, the proportion between axial and shear stress can be indicated as a constant because the fracturing is represented by one fundamental region. components and enable formulating the conclusion on the role of stress type in the zones' degradation. Looking at the fracture region of Alloy 59 (Figure 20a), the multi-planar degradation can be indicated as the main feature due to the loading type used. In contrast, in the case of S355J2W+N steel (Figure 20b) and the steel alloy (Figure 20c) joint, the region is represented by a one-fracture plane. Therefore, the following sentences can be formulated: (a) Reorientation of axial and shear stress components follow the degradation of the Alloy 59 as well as differences in their values as stress state components; (b) In the case of the weld manufactured by means of Alloy 59 and S355J2W+N steel, the proportion between axial and shear stress can be indicated as a constant because the fracturing is represented by one fundamental region. For further experiments, it is worth focusing on SEM approaches to the fracture region because more details on the weld degradation can be collected. These details on the weld can be directly used for damage mechanics for the weld behaviour description concerning damages due to monotonic tensile. Moreover, the degradation mechanism can be more clearly presented, and a scheme for the damage-type features is possibly easily presented.
Another important stage for the further examination of the proposed weld technology and the structural materials can be connected with quantitative determination because it enables us to follow an extended measurement uncertainty employing a calliper and testing machine accuracy as well as an error related to a tested object mounting. Taking those details, the uncertainty of the individual components of the experiment can be covered. Next, expanded and complex uncertainties can be resolved. This approach will avoid significant mistakes, and following the quality of the welds manufactured using different material types can be discussed very precisely.
Novelty and Application
The novelty and application of the welding technology and the testing method can be presented as follows: Novelty
•
The MIG process at the determined parameters can be directly used for mixed joint manufacturing; • The welding process does not require additional devices or systems, i.e., cooling or heating; • The hourglass specimen with a weld in its middle region of a measurement section is very useful for determining the joint quality; • For mixed joint quality, the fundamental features of the joint such as stress-strain characteristics, mechanical parameters and hardening curves for analytical and FEA approaches are determined.
Application For further experiments, it is worth focusing on SEM approaches to the fracture region because more details on the weld degradation can be collected. These details on the weld can be directly used for damage mechanics for the weld behaviour description concerning damages due to monotonic tensile. Moreover, the degradation mechanism can be more clearly presented, and a scheme for the damage-type features is possibly easily presented.
Another important stage for the further examination of the proposed weld technology and the structural materials can be connected with quantitative determination because it enables us to follow an extended measurement uncertainty employing a calliper and testing machine accuracy as well as an error related to a tested object mounting. Taking those details, the uncertainty of the individual components of the experiment can be covered. Next, expanded and complex uncertainties can be resolved. This approach will avoid significant mistakes, and following the quality of the welds manufactured using different material types can be discussed very precisely.
Novelty and Application
The novelty and application of the welding technology and the testing method can be presented as follows: Novelty
•
The MIG process at the determined parameters can be directly used for mixed joint manufacturing; • The welding process does not require additional devices or systems, i.e., cooling or heating; • The hourglass specimen with a weld in its middle region of a measurement section is very useful for determining the joint quality; • For mixed joint quality, the fundamental features of the joint such as stress-strain characteristics, mechanical parameters and hardening curves for analytical and FEA approaches are determined.
Application
• Improvement of welding technology for other mixed metal joints; • Power plant industry for operational conditions at elevated temperatures and inspections for replacing selected components due to failure; • Analytical and numerical approaches for superalloy and steel welding using the collected results; • Forecasting service life using the determined mechanical parameters of the joint.
Summary
The analysis of the obtained test results enabled us to formulate as below: • It is possible to make a correct mixed joint made of S355J2W + N steel and Alloy 59 using the MIG process without welding defects and incompatibilities.
•
The MIG connection technology with one-side bevelling and using NiCr23Mo16 nickel-based electrode wire and Ar-5% He shielding gas is the correct choice.
•
The mixed weld had excellent mechanical properties: yield stress (248 MPa) and ultimate tensile strength (518 MPa) values, which means the joint can be applied to rotor structural elements.
•
In the case of the superalloy and mixed joint, the hardening sections of the tensile curves were very similar in shape, and digital results represented almost the same values of power law coefficients.
•
The fracturing of the steel and mixed weld was expressed by the one fundamental decohesion region, which has reflected the constant proportion of values of axial and shear stress components up to the separation. | 13,246.8 | 2023-04-01T00:00:00.000 | [
"Engineering",
"Materials Science"
] |
Coexistence Mechanism between eMBB and uRLLC in 5G Wireless Networks
uRLLC and eMBB are two influential services of the emerging 5G cellular network. Latency and reliability are major concerns for uRLLC applications, whereas eMBB services claim for the maximum data rates. Owing to the trade-off among latency, reliability and spectral efficiency, sharing of radio resources between eMBB and uRLLC services, heads to a challenging scheduling dilemma. In this paper, we study the co-scheduling problem of eMBB and uRLLC traffic based upon the puncturing technique. Precisely, we formulate an optimization problem aiming to maximize the MEAR of eMBB UEs while fulfilling the provisions of the uRLLC traffic. We decompose the original problem into two sub-problems, namely scheduling problem of eMBB UEs and uRLLC UEs while prevailing objective unchanged. Radio resources are scheduled among the eMBB UEs on a time slot basis, whereas it is handled for uRLLC UEs on a mini-slot basis. Moreover, for resolving the scheduling issue of eMBB UEs, we use PSUM based algorithm, whereas the optimal TM is adopted for solving the same problem of uRLLC UEs. Furthermore, a heuristic algorithm is also provided to solve the first sub-problem with lower complexity. Finally, the significance of the proposed approach over other baseline approaches is established through numerical analysis in terms of the MEAR and fairness scores of the eMBB UEs.
I. INTRODUCTION
The wireless industries are going through different kinds of emerging applications and services, e.g., high-resolution video streaming, virtual reality (VR), augmented reality (AR), autonomous cars, smart cities and factories, smart grids, remote medical diagnosis, unmanned aerial vehicles (UAV), artificial intelligence (AI) based personal assistants, sensing, metering, monitoring etc, along with the explosive trends of mobile traffic [1]. It is foreseen that the mobile application market will flourish in a CAGR of 29.1% during 2015 − 2020 [2]. Energy efficiency, latency, reliability, data rate, etc are distinct for separate applications and services. To handle these diversified requirements, International Telecommunication Union (ITU) has already classified 5G services into uRLLC, mMTC, and eMBB categories [3]. Gigabit per second (Gbps) level data rates are required for eMBB users, whereas connection density and energy efficiency are the major concern for mMTC, and uRLLC traffic focuses on extremely high reliability (99.999%) and remarkably low latency (0.25 ∼ 0.30 ms/packet) [4].
Generally, the lions' share of wireless traffic is produced by eMBB UEs. uRLLC traffic is naturally infrequent and needs to be addressed spontaneously. The easiest way to settle this matter is to allocate some resources for uRLLC. However, under-utilization of radio resources may emerge from this approach, and generally, effective multiplexing of traffics is required.
Though the short-TTI mechanism is straightforward for implementation, it degrades spectral efficiency because of the massive overhead in the control channel. On the contrary, the puncturing strategy decreases the above overhead, although it necessitates an adequate mechanism for recognizing and healing the punctured case. Slot (1 ms) and mini-slot (0.125 ms) are proposed as time units for meeting the latency requirement of uRLLC traffic in the 5G NR. At the outset of a slot, eMBB traffic is scheduled and continues unchanged throughout the slot. If the same physical resources are used, uRLLC traffic is overridden upon the scheduled eMBB transmission.
Currently, much attention has been paid to resource sharing for offering QoS or QoE to the users. Studies [6] and [7] investigate the sharing of an unlicensed spectrum between LTE and WiFi networks, however, the study [8] con sider LTE-A and NB-IoT services for sharing the same resources. Study [9] solves user association and resource allocation problems. The study [9] consider the downlink of fog network to support QoS provisions of the uRLLC and eMBB. Some other studies, however, investigates and/or analyzes the influence of uRLLC traffic on eMBB [10]- [15] or presents architecture and/or framework for co-scheduling of eMBB and uRLLC traffic [16]- [19]. Moreover, some authors consider eMBB and uRLLC traffic in their coexisting/multiplexing proposals [20]- [27] where they apply puncturing technique.
As per our knowledge, concrete mathematical models and solutions, however, are lacking in most of these coexistence mechanisms. Most of the studies mainly focus on analysis, system-level design or framework. Thus, effective coexistence proposals between eMBB and uRLLC traffic are wanting in literature. So, to enable eMBB and uRLLC services in 5G wireless networks, we propose an effective coexistence mechanism in this paper. Our preliminary work has been published in [24] where we have used a one-sided matching and heuristic algorithm, respectively, for resolving resource allocation problems of eMBB and uRLLC users. The major difference between [24] and current work is the involvement of PSUM and TM for solving similar problems. This paper mainly focuses on the followings: • First, we formulate an optimization problem for eMBB UEs with some constraints, where the objective is to maximize the minimum expected rate of eMBB UEs over time.
• Second, to solve the optimization problem effectively, we decompose it into two subproblems: resource scheduling for eMBB UEs, and resource scheduling of uRLLC UEs.
PSUM is used to solve the first sub-problem, whereas the TM is employed to solve the second one.
• Third, we redefine the first sub-problem into a minimization problem for each slot and provide an algorithm based upon PSUM to obtain near-optimal solutions.
• Fourth, we redefine the second sub-problem as a minimization problem for each mini-slot within every slot and present the algorithm based upon MCC and MODI methods of the transportation model to find an optimal solution of the second sub-problem.
• Fifth, we also present a cost-effective heuristic algorithm for resolving the first sub-problem.
• Finally, we perform a comprehensive experimental analysis for the proposed scheduling approach and compare the results, MEAR and fairness [41] of the eMBB UEs, with the PS [22], MUPS [26], RS, EDS, and MBS approaches.
The remainder of the paper is systematized as follows. In Section II, we present the literature review. We explain the system model and present the problem formulation in Section III. The proposed solution approach of the above-mentioned problem is addressed in Section IV. In Section V, we provide experimental investigation, discussion, and comparison concerning the proposed solution. Finally, we conclude the paper in Section VI.
II. LITERATURE REVIEW
Recently, both industry and academia focus on the study of multiplexing between eMBB traffic and uRLLC traffic on the same physical resources. Information-theoretic arguments-based performance analysis for eMBB and uRLLC traffic has performed in [10]. The authors consider both OMA and NOMA for uplink in C-RAN framework. An insight into the performance tradeoffs among the eMBB and uRLLC traffic is explained in [10]. In [11], authors have introduced eMBB influenced minimization problem to protect the uRLLC traffic from the dominant eMBB services. This paper explores their proposal for the mobile front-haul environment. In [12], the authors present an effective solution for multiplexing different traffics on a shared resource.
Particularly, they propose an effective radio resource distribution method between the uRLLC and eMBB service classes following trade-offs among the reliability, latency and spectral efficiency.
Moreover, they investigate the uRLLC and eMBB performance adopting different conditions. In order to 5G service provisioning (i.e., eMBB, mMTC and uRLLC services), the authors of [13] have studied radio resources slicing mechanism, where the performance of both orthogonal and non-orthogonal are analyzed. They have proposed a communication-theoretic model by considering the heterogeneity of 5G services. They also found that the non-orthogonal slicing is significantly better to perform instead of orthogonal slicing for those 5G service multiplexing.
Recently, for 5G NR physical layer challenges and solution mechanisms of uRLLC traffic communications has been presented in [14], where they pay attention to the structure of packet and frame. Additionally, they focus on the improvement of scheduling and reliability mechanism for uRLLC traffic communication such that the coexistence of uRLLC with eMBB is established.
In [15], the authors have been analyzed the designing principle of the 5G wireless network by employing low-latency and high-reliability for uRLLC traffic. To do this, they consider varying requirements of uRLLC services such as variation of delay, packet size, and reliability. To an extent, they explore different topology network architecture under the uncertainty. proposed for fulfilling uRLLC traffic demand in [19], where they exhibit that the static bandwidth partitioning is inefficient for eMBB and uRLLC traffic. Thus, the authors of [19] have illustrated a dynamic mechanism for multiplexing of eMBB and uRLLC traffic and apply this in both frequency and time domain.
The efficient way of network resource sharing for the eMBB and uRLLC is studied in [20] and [21]. A dynamic puncturing mechanism is proposed for uRLLC traffic in [20] within eMBB resources to increase the overall resource utilization in the network. To enhance the performance for decoding of eMBB traffic, a joint signal space diversity and dynamic puncturing schemes have proposed, where they improve the performance of component interleaving as well as rotation modulation. In [21], a joint scheduling problem is formulated for eMBB and uRLLC traffic in the goal of maximizing eMBB users'utility while satisfying stochastic demand for the uRLLC UEs. Specifically, they measure the loss of eMBB users for superposition/puncturing by introducing three models, which include linear, convex and threshold-based schemes. For reducing the queuing delay of the uRLLC traffic, the authors introduce punctured scheduling (PS) in [22]. In case of insufficient radio resource availability, the scheduler promptly overwrites a portion of the eMBB transmission by the uRLLC traffic. The scheduler improves the uRLLC latency performance; however, the performance of the eMBB users are profoundly deteriorated.
The authors of [23] and [24] manifest the coexistence technique for enabling 5G wireless services like eMBB and uRLLC based upon a punctured scheme. The authors present an enhanced PS (EPS) scheduler to enable an improved ergodic capacity of the eMBB users in [25]. EPS is capable of recovering the lost information due to puncturing and partially. eMBB users are supposed to be cognizant about the corresponding resource that is being penetrated by uRLLC.
Therefore, the victim eMBB users ignore the punctured resources from the erroneous chase condensing HARQ process. The authors of [26] propose a MUPS, where they discretize the trade-off among network system capacity and uRLLC performance. to reliability concerns. The authors of [27] propose a null-space-based preemptive scheduler (NSBPS) for jointly serving uRLLC and eMBB traffic in a densely populated 5G arrangement.
The proposed approach ensures on-the-spot scheduling for the sporadic uRLLC traffic, while makes a minimal shock on the overall system outcome. The approach employs the system spatial degrees of freedom (SDoF) for uRLLC traffic for spontaneously providing a noise-free subspace. In [28], the authors present a risk-sensitive approach for allocating RBs to uRLLC traffic in the goal of minimizing the uncertainty of eMBB transmission. Particularly, they launch the Conditional Value at Risk (CVaR) for estimating the uncertainty of eMBB traffic in [28].
III. SYSTEM MODEL AND PROBLEM FORMULATION
In this work, we consider a 5G network scenario with one gNB which supports a group of user equipment (UE) E requiring eMBB service, and a set of user equipment U demanding uRLLC service. The system operates in downlink mode for the UEs and the overall system diagram is shown in Fig. 1. gNB supports the UEs using licensed RBs K each with equal bandwidth of B. Every time slot, with a length ∆, is split into M mini-slots of duration δ for managing low latency services. For supporting eMBB UEs, we consider T s LTE time slots and denoted by T = {1, 2, · · · , T s }. uRLLC traffic arrive at gNB (any mini-slot m of time slot t) follows Gaussian distribution, i.e., U ∼ N (µ, σ 2 ). Here, µ and σ 2 denote the mean and variance of U .
Each uRLLC UE u ∈ U request for a payload of size L m,t u (varying from 32 to 200 Bytes [29]).
gNB allots the RBs to the eMBB UEs at the commencement of any time slot t ∈ T . The achievable rate of e ∈ E for RB k ∈ K is as follows: where γ t e,k = Peh 2 e N 0 B presents SNR. P e is the transmission power of gNB for e ∈ E and h e denotes the gain of e ∈ E from the gNB, and N 0 represnts the noise spectral density. eMBB UEs require more than one RB for satisfying their QoS. Therefore, the achievable rate of eMBB UE e ∈ E in time slot t as follows: where α denotes the resource allocation vector for E at any time slot t, and each element is as follows: uRLLC traffic can arrive at some moment (i.e. mini-slot) inside any time slot t and requires to be attended quickly. Any uRLLC traffic needs to be completed within a mini-slot period for its' latency and reliability constraints. Normally, the payload size of uRLLC traffic is really short, and therefore, we cannot straightforwardly adopt Shannon's data rate formulation [10].
The achievable rate of a uRLLC UE u ∈ U in RB k ∈ K, when its' traffic is overlapped with eMBB traffic, can properly be approximated by employing [30] as follows: where γ m,t u = h 2 u Pu N 0 B+h 2 u Pe represents the SINR for u ∈ U at mini-slot m of t. Here, h 2 u P e indicates the interference generated from serving e ∈ E in the same RB, depicts the channel dispersion, and meaning of other symbols are shown in II. However, the reliability of uRLLC traffic fall into vulnerability due to the interference. Hence, superposition mechanism is not a suitable for serving uRLLC UE [11]. Thus, for serving uRLLC UEs, we concentrate on the puncturing technique . In the punctured mini-slot, gNB allots zero power for eMBB UE, and therefore, the interference cannot affect the uRLLC traffic. At that time, The achieved rate of u ∈ U, when it uses multiple RBs, is as follows: where β is the resource allocation vector for U at m of t, and each of its' element follows: 0, otherwise. All the uRLLC request in any m of t needs to be served for sure, and hence, where φ denotes a vector for the serving uRLLC UEs, and thus, Within the stipulated period δ, the payload L m,t u of u ∈ U needs to be transferred, and hence, satisfy the following: Hence, the reliability and latency concerns of uRLLC traffic are simultaneously shielded by (7) and (9). Besides, e ∈ E loses some throughput at t if uRLLC traffic is punctured within its' RBs. We utilize the linear model of [21] for estimating the throughput-losses of eMBB UE.
Therefore, the throughput-losses e ∈ E looks like as follows: r t e,loss = k∈K r t e,k m∈M u∈U I(α t e,k = β m,t u,k ).
So, the actual achievable rate of e ∈ E in any t is as follows: r t e,actual = r t e − r t e,loss .
We see that β affects on α, and hence, impact negatively to the eMBB throughput in each t ∈ T . At the start of any t ∈ T , gNB allocates the RBs K among the E in an orthogonal fashion as shown in Fig. 2. These characteristics of α are shown mathematically as follows: e∈E k∈K α t e,k ≤ |K|.
Within each t ∈ T , gNB allows uRLLC UEs to get some RBs immediately on a mini-slot basis. Therefore, uRLLC traffic overlaps with eMBB traffic at m and also shown in Fig. 2.
Accordingly, β satisfy the following conditions on each m: u∈U k∈K φ m,t u β m,t u,k ≤ |K|.
Finally, our objective is to maximize the actual achievable rate of each eMBB UE across T while entertaining nearly every uRLLC request within its' speculated latency. We apply Max-Min fairness doctrine for this mission, and it contributes stationary service quality, enhances spectral efficiency and makes UEs more pleasant in the network. Hence, the maximization problem is formulated as follows: In (18) shows that every item of α, β and φ are binary. The formulation (18) is a Combinatorial Programming (CP) problem having chance constraint, and NP-hard due to its nature.
IV. DECOMPOSITION AS A SOLUTION APPROACH FOR PROBLEM (18)
We assume that eMBB UEs are data-hungry over the considered period. Thus, at the commencement of a time slot t ∈ T , gNB schedules all of its' RBs among the eMBB UEs and stay unchanged over t. If uRLLC traffic requests come in any m of t, the scheduler tries to serve the requests in the next m + 1. Hence, the overlapping of uRLLC traffic over eMBB traffic happens as shown in Fig. 2. Usually, a portion of all RBs is required for serving such uRLLC traffic.
However, the challenge is to find the victimized eMBB UE(s) following the aspiration of the problem (18).
On the other hand, the second sub-problem (with α t , ∀t as the solution of 19) is manifested as follows: β m,t u,k , φ m,t u ∈ {0, 1}, ∀u, k, m, t. A. PSUM as a Solution of the Sub-Problem (19) Problem (19) is still is computationally expensive to reach a globally optimal solution due to its' NP-hardness. In this sub-section, we propose the PSUM algorithm to solve (19) approximately with low complexity. Relaxation of the binary variable and the addition of a penalty term to the objective function is the main philosophy of our proposed PSUM algorithm. We redefine (19) as follows: α t e,k ∈ [0, 1], ∀e, k, t.
Now according to Theorem 2 of [31], if |K| is sufficiently large then original sub-problem (19) and (21) are equivalent. Moreover, we add a penalty term L p to the objective function to get binary soltion of relaxed variable from (21). Let α t k = {α t e,k } e∈E and we can rewrite (19a) as α t k 1 ≤ 1, ∀t, k. The penalized problem is as follows: where σ > 0 is the penalty parameter, with p ∈ (0, 1), and ε is any non-negative constant. Following the fact of [32] which is further described in [31], the optimal value is as follows: Generally, the parameter σ should big enough to make the values of {α t e,k } near zero or one. Then, we achieve a feasible solution of (22) by applying the rounding process.
It is not easy to solve (22) directly. However, by utilizing the successive upper bound minimization (SUM) technique [33], [34], we can efficiently resolve (22). This method tries to secure the lower bound of the actual objective function by determining a sequence of approximation of the objective functions. As P ε (α t ) is concave in nature and hence, Algorithm 1 Solution of (19) for each t based on PSUM where α t,i is the value of current allocation of iteration i. At the (i + 1)-th iteration of t, we solve the following problem: In each iteration, we can get a globally optimal solution for sub-problem (26) by using the solver. Algorithm 1 shows the proposed mechanism for solving (19). In this Algorithm, 0 < η < 1 < ζ where ζ and η represent two constants defined previously.
B. Solution of Sub-Problem (20) through TM
Due to the existence of chance constraint (20a) and also the combinatorial variable, β, (20) is still difficult to resolve by using traditional optimizer. Now, we need to transmute (20a) into deterministic form for solving (20). Moreover, let us assume g(φ, U ) = u∈U φ m,t u − U , U ∈ R and U ∼ N (µ, σ 2 ), ∀m, t and hence, Here, F U is the cumulative distribution function (CDF) of random variable U . Thus, from constraint (20a), we can rewrite as follows: Now, (28d) and (20a) are identical. Hence, the renewed form of (20) looks like as follows: r t e ,loss − r t e,loss , ∀e. As gNB engages OFDMA for uRLLC UEs, constraint (20c) holds. Moreover, depending on U , constraints (20d), (20e), and (20f) also hold. Constraint (29c) can be used as a basic block to build a cost matrix C = (c u,e ), u ∈ U , e ∈ E. As K are held by eMBB UEs E in any time slot t ∈ T , we can find a vector s = [s 1 , s 2 , · · · , s |E| ]. Now redefine problem (29) as follows: The goal of (30) is to find a matrix χ ∈ Z |U |×|E| = (χ ue ), ∀u ∈ U , e ∈ E that will minimize the cost/loss of eMBB UEs. This is a linear programming problem equivalent to the Hitchcock problem [35] with inequities, which contributed to unbalanced transportation model. Introducing slack variables χ |U |+1,e , ∀e ∈ E and d |U |+1 in the constraints (30b) and (30c), respectively, which convert them into equality, we have: Now the modified problem in (30) is a BTM. Moreover, we have to add d |U |+1 = e∈E s e − u∈U d u to the demand vector d as d = d ∪ {d |U |+1 } and a row [0] 1×|E| to cost matrix C as C = C ∪ {[0] 1×|E| }. BTM can be solved by the simplex method [36]. The solution matrix χ will be in the form of Z (|U |+1)×|E| . NWC [37], MCC [37], and VAM [37], [38] are some of the popular methods for obtaining initial feasible solution of BTM. We can use the stepping-stone [39] or MODI [40] method to get an optimal solution of the BTM. In the following sub-section, we use the combination of the MCC and MODI for acquaring the optimal result from the BTM. 1) Determining Initial Feasible Solution by MCC Method: MCC method allots to those cells of χ considering the lowest cost from C. Firstly, the method allows the maximum permissible to the cell with the lowest per RB cost. Secondly, the amount of quantity and need is synthesized while crossing out the satisfied row(s) or column(s). Either row or column is ruled out if both of them are satisfied concurrently. Thirdly, we inquire into the uncrossed-out cells which have the least unit cost and continue it till there is specifically one row or column is left uncrossed.
The primary steps of the MCC method are compiled as follows: Step 1: Distribute maximum permissible to the worthwhile cell of χ which have the minimum cost found from C, and update the supply (s) and demand (d).
Step 2: Continue Step 1 till there is any demand that needs to be satisfied.
2) MODI Method for Finding an Optimal Solution:
The initial solution found from section IV-B1 is used as input in the MODI method for finding an optimal solution. We need to augment an extra left-hand column and the top row (indicated by x u and y e respectively) with C whose values require to be calculated. The values are measured for all cells which have the corresponding allocation in χ and shown as follows: x u + y e = c u,e , ∀χ u,e = ∅.
Now we solve (33) to obtain all x u and y e . If necessary then assign zero to one of the unknowns toward finding the solution. Next, evaluate for all the empty cells of χ as follows: Now select k u,e corresponding to the most negative value and determine the stepping-stone path for that cell to know the reallocation amount to the cell. Next, allocate the maximum permissible to the empty cell of χ corresponding to the selected k u,e . x u and y e values for C and χ must be recomputed with the help of (33) and a cost change for the empty cells of χ need to be figured out using (34). A corresponding reallocation takes place just like the previous step and the process continues till there is a negative k u,e . At the end of this repetitive process, we get the optimal allocation (χ). The MODI method described above can be summed as follows: Step 1: Develop a preliminary solution (χ) applying the MCC method.
Step 2: For every row and column of C, measure x u and y e by applying (33) to each cell of χ that has an allocation.
Step 3: For every corresponding empty cell of χ, calculate k u,e by applying (34).
Step 4: Determine the stepping-stone path [39] from χ corresponding to minimum k u,e that found in Step 3.
Step 5: Based on the stepping-stone path found in Step 4, allocate the highest possible to the free cell of χ.
C. Low-Complexity Heuristic Algorithm for Solving Sub-Problem (19) Though Algorithm 1 can solve the sub-problem (19) optimally, but computation time requires to solve it grows much faster as the size of the problem increase. Besides, the number of eMBB UEs is large in reality, and we have a short period to resolve this kind of problem. Therefore, we need a faster and efficient heuristic algorithm, which may sacrifice optimality, to solve (19).
Thus, we propose Algorithm 2 for solving (19). At t = 1, Algorithm 2 allocate resources equally to the eMBB UEs. But, it allocates resources to eMBB UEs in the rest of the time slots depending on the proportional loss of the previous time slot. In this way, Algorithm 2 can accommodate the EAR of eMBB UEs in the long-run. The complexity of Algorithm 2 depends on T and E.
V. NUMERICAL ANALYSIS AND DISCUSSIONS
In this section, we assess the proposed approach using comprehensive experimental analyses.
Here, we compare our results with the results of the following state-of-the-art schedulers: for each e ∈ E do 7: for each k = 1 · · · N RB do 8: Determine r t−1 e,loss and r t−1 e,actual for all e ∈ E by using (10) and (11) The main performance parameters are MEAR and fairness [41] of the eMBB UEs and defined In our scenario, we consider an area with a radius of 200 m and gNB resides in the middle of the considered area. eMBB and uRLLC UEs are disseminated randomly in the coverage space. gNB works on a 10 MHz licensed band for supporting the UEs in downlink mode. Every uRLLC UE needs a single PRB for its service. Furthermore, gNB estimates path-loss for both eMBB and uRLLC UEs using a free space propagation model amidst Rayleigh fading. Table III exhibits the significant parameters for this experiment. We use similar PSUM parameters as of [31]. We realize the results of every approaches after taking 1, 000 runs.
A comparison of MEAR and fairness scores are presented in Fig. 4 and Fig. 5, respectively, between the proposed (PSUM+TM) and the optimal value for a small network. Fig. 4 shows and PS method reduces with the increased arrival of uRLLC traffic, as the PS scheme gets more chance to adjust the users with the higher expected achieved rate.
We compare the fairness scores among various methods with different values of σ which is shown in Fig. 7. The scores originating from the proposed method are greater than or similar to that of others as indicated in Fig. 7. Fig. 7 Fig. 7(b) and 7(c), respectively. Moreover, the fairness scores increase for the Proposed, MBS and PS methods with the increasing value of σ as it gets more chance to maximize the minimum achieved rate, whereas the same scores decrease with the increasing value of σ for RS, EDS and MUPS as eMBB UEs have more opportunity to be affected by the uRLLC UEs. Bytes.
emerging from our method is bigger than or similar to other comparing methods for different values of σ and shown in Fig. 9. Fig. 9 also reveals that the σ value has a negligible impact on the average score of the fairness in the Proposed, RS, EDS, MBS, PS methods, but it impacts inversely to the MUPS method more and more uRLLC traffic choose same eMBB UE for the PRBs. Moreover, the average fairness scores of the proposed method are similar to both MBS and PS methods. However, the proposed method treats eMBB UEs 0.92%, 0.92%, and 1.92% fairly than RS, EDS, and MUPS methods , respectively, when σ = 1, whereas, the similar scores are 1.23%, 1.23%, and 12.21%, respectively, during σ = 10.
In Fig. 10, we compare the average MEAR of eMBB UEs for considering varying uRLLC load (L) and uRLLC traffic (σ). The MEAR value of our method surpasses other concerned methods in every circumstance as revealed from Fig. 10. The same figure also explicates that these values degrade when L increases for varying σ as the system needs to allocate more PRBs to the uRLLC UEs. Moreover, these values decrease with the increasing value of σ for a fixed L, and also the same for increasing the value of L with a fixed σ. In Fig. 11, we compare the average fairness score of eMBB UEs for the different methods for changing the uRLLC load (L) and uRLLC traffic (σ). Fig. 11 exposes that the fairness scores of our method are better than or at least similar to that of its' rivals. The figure also reveals that these scores decrease with an increasing L for the lower value of σ. However, these scores increase with the increasing L when σ value is high. Moreover, for the MUPS method, these values decrease with the increasing value of σ and L.
VI. CONCLUSIONS
In this paper, we have introduced a novel approach for coexisting uRLLC and eMBB traffic in the same radio resource for enabling 5G wireless systems. We have expressed the coexisting dilemma as a maximizing problem of the MEAR value of eMBB UEs meanwhile attending the uRLLC traffic. We handle the problem with the help of the decomposition strategy. In every time slot, we resolve the resource scheduling sub-problem of eMBB UEs using a PSUM based algorithm, whereas the similar sub-problem of uRLLC UEs is unraveled through optimal transportation model, namely MCC and MODI methods. For the efficient scheduling of PRBs among eMBB UEs, we also present a heuristic algorithm. Our extensive simulation outcomes demonstrate a notable performance gain of the proposed approach over the baseline approaches in the considered indicators. | 7,106.4 | 2020-03-10T00:00:00.000 | [
"Computer Science"
] |
Regulation of De Novo Lipid Synthesis by the Small GTPase Rac1 in the Adipogenic Differentiation of Progenitor Cells from Mouse White Adipose Tissue
White adipocytes act as lipid storage, and play an important role in energy homeostasis. The small GTPase Rac1 has been implicated in the regulation of insulin-stimulated glucose uptake in white adipocytes. Adipocyte-specific rac1-knockout (adipo-rac1-KO) mice exhibit atrophy of subcutaneous and epididymal white adipose tissue (WAT); white adipocytes in these mice are significantly smaller than controls. Here, we aimed to investigate the mechanisms underlying the aberrations in the development of Rac1-deficient white adipocytes by employing in vitro differentiation systems. Cell fractions containing adipose progenitor cells were obtained from WAT and subjected to treatments that induced differentiation into adipocytes. In concordance with observations in vivo, the generation of lipid droplets was significantly attenuated in Rac1-deficient adipocytes. Notably, the induction of various enzymes responsible for de novo synthesis of fatty acids and triacylglycerol in the late stage of adipogenic differentiation was almost completely suppressed in Rac1-deficient adipocytes. Furthermore, the expression and activation of transcription factors, such as the CCAAT/enhancer-binding protein (C/EBP) β, which is required for the induction of lipogenic enzymes, were largely inhibited in Rac1-deficient cells in both early and late stages of differentiation. Altogether, Rac1 is responsible for adipogenic differentiation, including lipogenesis, through the regulation of differentiation-related transcription.
Introduction
WAT is a type of mammalian adipose tissue serving as storage of lipids, and plays a pivotal role in energy and glucose homeostasis [1]. Triacylglycerol is the main constituent of lipid droplets in white adipocytes, and is synthesized from glucose and fatty acids through multiple metabolic reactions. On the other hand, the transport of glucose and fatty acids into adipocytes from the blood is mediated by various specific transporters. Insulin is known to enhance the synthesis and accumulation of triacylglycerol, as well as glucose and fatty acid uptake in adipocytes.
The glucose transporter GLUT4 is responsible for insulin-stimulated glucose uptake in adipocytes [2]. The increase in the level of plasma membrane-localized GLUT4 in response to insulin results in the enhanced uptake of glucose into the cell [2,3]. Signaltransducing pathways downstream of the insulin receptor leading to GLUT4 translocation to the plasma membrane in adipocytes have been investigated extensively for decades, and two major signaling pathways are well characterized [3,4]. One signaling cascade comprises phosphoinositide 3-kinase (PI3K) and the serine/threonine protein kinases PDK1 and Akt2. This signaling cascade plays a crucial role in both adipocytes and skeletal muscle [3][4][5]. Downstream of Akt2, the Akt substrate of 160 kDa, also termed TBC1D4, has been implicated in the regulation of GLUT4 vesicle trafficking in response to insulin, acting as a GTPase-activating protein for the Rab family small GTPase Rab10 in adipocytes [6][7][8]. The other cascade is thought to be specific to adipocytes and independent of PI3K [9].
The expression of the enzymes for the synthesis of fatty acids and triacylglycerol is known to be regulated by a variety of transcription factors, such as the nuclear receptor peroxisome proliferator-activated receptor γ (PPARγ) [24] and CCAAT/enhancer-binding protein (C/EBP) family transcription factors [25], in response to insulin. Thus, it is important to examine expression levels of these transcription factors to clarify the role of Rac1 in the regulation of de novo lipid synthesis in white adipocytes.
In this study, we aimed to further reveal the mechanisms underlying atrophy of WAT in adipo-rac1-KO mice, employing in vitro differentiation systems of mouse progenitor cells isolated from WAT and the 3T3-L1 cell line. We show that Rac1 plays a pivotal role in the induction of differentiation into adipocytes, regulating not only glucose uptake but also the expression of diverse enzymes for de novo lipid synthesis.
Establishment and Characterization of an In Vitro Differentiation Assay Using Adipose Progenitor Cells Obtained from Mouse WAT
We established an in vitro differentiation assay as a first step to clarify the mechanisms for atrophy of WAT observed in adipo-rac1-KO mice [13]. Collagenase-treated mouse subcutaneous WAT was centrifuged and mature adipocytes as floating cells were removed. We then confirmed that the precipitated stromal vascular fraction (SVF) contained CD34positive adipose progenitor cells, but not perilipin 1-positive mature adipocytes, by reversetranscriptase polymerase chain reaction (RT-PCR) analysis [26,27] ( Figure 1A).
The SVF containing adipose progenitor cells was cultured until cells reached confluence in a culture medium optimized for growth of adipose progenitor cells (KBM ADSC-1) ( Figure 1B). The day when cells reached confluence was referred to as day 0. Confluent cells were further cultured for two days in Dulbecco's modified Eagle's medium (DMEM)-based growth medium and then subjected to treatment with reagents, such as insulin, which are required for differentiation into adipocytes, as described in Figure 1B. Expression levels of genes for the Cre recombinase and Rac1 during adipogenic differentiation in vitro from progenitor cells were monitored by quantitative RT-PCR analysis ( Figure 1C,D). The Cre recombinase transgene is expressed under the control of the adiponectin (Adipoq) promoter, which is specifically activated in adipocytes [28]. Therefore, the expression of the Cre recombinase transgene was expected to increase after the initiation of adipogenic differentiation. In fact, the expression of the Cre recombinase transgene in cells derived from control (adipo-Cre) and adipo-rac1-KO mice was stimulated after 3-day induction of adipogenic differentiation, and reached a plateau at day 4 ( Figure 1C). Consistently with these observations, the expression level of the rac1 gene was significantly reduced after 4-day induction of adipogenic differentiation in cells derived from adipo-rac1-KO mice ( Figure 1D). The expression of the rac1 gene was enhanced after day 5 in control cells, suggesting a significant role of Rac1 in differentiated adipocytes.
After 4-day induction of adipogenic differentiation, small lipid droplets emerged in cells derived from control mice ( Figure 1E). In contrast, virtually no cells derived from adipo-rac1-KO mice contained lipid droplets in this stage ( Figure 1E). At day 7, a large quantity of lipid droplets was detected in cells from control mice, whereas only a limited number of cells from adipo-rac1-KO mice contained lipid droplets ( Figure 1E). Differentiated adipocytes harboring lipid droplets were collected by centrifugation from the cell cultures at day 7, and seeded onto chamber slides. The Rac1 protein and lipid droplets were then stained with an anti-Rac1 antibody and a fluorescent dye for lipid droplets, respectively ( Figure 1F). The average size of cells from adipo-rac1-KO mice was significantly less than that of control cells ( Figure 1G). The size of lipid droplets in cells from adipo-rac1-KO mice was also measured, and found to be largely reduced compared with that in control cells ( Figure 1H,I). Taken together, these results demonstrate that knockdown of the rac1 gene during the process of adipogenic differentiation severely affects the accumulation of lipid droplets in adipocytes. These results also show that the effect of Rac1 deficiency on adipogenic differentiation in vivo is well reproduced by the aforementioned in vitro differentiation assay using adipose progenitor cells.
The Expression of Enzymes for De Novo Synthesis of Fatty Acids and Triacylglycerol during Differentiation of Adipose Progenitor Cells into Adipocytes In Vitro
We previously demonstrated that mRNA levels of various enzymes for de novo synthesis of fatty acids and triacylglycerol, such as ACLY, ACC, FASN, SCD1, and GPAT1, were significantly lowered in subcutaneous WAT in adipo-rac1-KO mice [13]. These results suggest that Rac1 is responsible for the regulation of de novo lipid synthesis [13]. To further explore this possibility, we next assessed expression levels of the above enzymes during adipogenic differentiation in vitro. In the control cell culture, the expression of the genes that encode the above enzymes was highly stimulated during adipogenic differentiation ( Figure 2). The increase in the expression level of the gene encoding ACLY was observed at day 5, followed by further increase until day 7 ( Figure 2A). Likewise, expression levels of genes encoding ACC, FASN and GPAT1 started to increase at day 5, and continued to increase up to the maximal level around day 6 ( Figure 2B,C,E). The expression level of the scd1 gene was increased after day 2, and reached a maximum at day 5 ( Figure 2D). In marked contrast, virtually no increase in expression levels of all of the above genes was detected in cells derived from adipo-rac1-KO mice ( Figure 2). These results provide evidence that Rac1 is intimately involved in the induction of genes that encode enzymes for de novo synthesis of fatty acids and triacylglycerol during adipogenic differentiation. The expression of the scd1 gene was increased until day 3 not only in control cells but also in cells from adipo-rac1-KO mice, because the rac1 gene was not disrupted at this time point ( Figure 2D).
The Expression and Phosphorylation of Transcription Factors during Differentiation of Adipose Progenitor Cells into Adipocytes In Vitro
To further explore the role of Rac1 in the induction of adipogenic differentiation, we next examined the expression of various transcription factors that have been implicated in the upregulation of differentiation-related genes. PPARγ activates a variety of target genes, regulating adipocyte differentiation and function [24]. The expression level of the gene encoding PPARγ started to increase at day 2, and reached a maximum at day 6 in control cells ( Figure 3A). In cells derived from adipo-rac1-KO mice, this gene was induced until day 3, but virtually no additional increase was observed after day 4, at which point the expression of the rac1 gene was suppressed ( Figure 3A). Therefore, it is likely that Rac1 is involved in the induction of the gene encoding PPARγ during adipogenic differentiation.
Another family of transcription factors, the C/EBP family, is composed of six members, in which the basic leucine zipper (bZIP) domain is conserved at the C terminus [25]. Among them, C/EBPα has been implicated in the control of differentiation into adipocytes through the induction of diverse target genes. The mRNA level of C/EBPα in control cells was increased as adipogenic differentiation proceeded, whereas no induction was observed after day 4 in cells isolated from adipo-rac1-KO mice, similarly to the case of PPARγ ( Figure 3B). The expression of genes for C/EBPβ and C/EBPδ isoforms is known to be promoted in the early stage of adipogenic differentiation, and these two isoforms are involved in the induction of PPARγ and C/EBPα [25]. C/EBPβ and C/EBPδ mRNAs were increased to near-maximal levels within two days in cells from both control and adipo-rac1-KO mice ( Figure 3C,D). Expression levels of C/EBPβ and C/EBPδ in cells from adipo-rac1-KO mice were largely decreased after 5-day induction of differentiation, in contrast to control cells, in which the expression levels were sustained ( Figure 3C,D).
Sterol regulatory element-binding protein 1c (SREBP-1c) is a transcription factor implicated in the regulation of fatty acid and triacylglycerol synthesis in the liver and adipose tissue [23,29]. We next examined mRNA levels of SREBP-1c in cells from control and adipo-rac1-KO mice, because SREBP-1c activates genes encoding enzymes for de novo lipid synthesis described above in response to insulin [23,29] ( Figure 3E). The mRNA level of SREBP-1c was rapidly increased at day 5 and sustained until day 7 in control cells. In contrast, virtually no increase in the mRNA level was observed in cells derived from adipo-rac1-KO mice.
Three different-sized polypeptides, named LAP*, LAP, and LIP, are produced from the C/EBPβ mRNA molecule by alternative use of translation initiation codons [25]. Both transcriptional activation and bZIP domains are present in LAP* and LAP, whereas LIP contains the bZIP domain, but not the transcriptional activation domain. Therefore, LIP is thought to act as a dominant-negative form by dimerizing with LAP* or LAP [25]. Furthermore, sequential phosphorylation of C/EBPβ by mitogen-activated protein kinase, cyclin-dependent kinase 2/cyclin A, and glycogen synthase kinase 3β increases the DNAbinding activity [30]. We then examined protein and phosphorylation levels of LAP*, LAP, and LIP by immunoblot analysis in the late stage of differentiation ( Figure 4). Protein levels of LAP* and LAP were sustained from day 5 to day 7 in control cells, whereas the levels in cells derived from adipo-rac1-KO mice were markedly reduced similarly to the mRNA levels ( Figure 4A,C,D). The protein level of LIP in cells from control mice was increased at day 6, but rapidly decreased at day 7 ( Figure 4A,E). Although the mechanisms for this change remain unclear, the rapid decrease at day 7 may contribute to further induction of target genes. In cells from adipo-rac1-KO mice, the protein level of LIP was also largely suppressed, but its effect on the induction of target genes may be limited due to the low levels of LAP* and LAP ( Figure 4A,E). Phosphorylation levels of the above three variants of C/EBPβ were also evaluated by using a phospho-specific antibody. Phosphorylation levels of LAP* and LAP in control cells rapidly decreased at day 7, suggesting lowered transcription activities of these proteins at this point ( Figure 4B,F,G). The phosphorylation level of LAP* was largely suppressed in adipo-rac1-KO mice-derived cells from day 5 to day 6, suggesting a role of Rac1 in the regulation of phosphorylation ( Figure 4B,F). In contrast, the phosphorylation level of LAP in cells from adipo-rac1-KO mice was similar to that in control cells ( Figure 4B,G). The phosphorylation level of LIP was also significantly reduced in cells from adipo-rac1-KO mice ( Figure 4B,H).
Protein and Phosphorylation Levels of C/EBPβ in the Early Stage of Differentiation of Adipose Progenitor Cells into Adipocytes In Vitro
The induced expression of the rac1 gene after day 5 in control cells suggested an important role of Rac1 in the late stage of differentiation ( Figure 1D), and indeed Rac1 was involved in the regulation of lipogenesis in this stage, as described above. It is also important to clarify the function of Rac1 in the early stage of differentiation, because the rac1 gene was significantly expressed (approximately 30% of the maximal level) even before day 3 ( Figure 1D), and the activity of Rac1 is generally enhanced through GDP/GTP exchange (GTP-binding) of preexisting Rac1 molecules rather than induced expression [10]. Therefore, we next tested the effect of functional inactivation of Rac1 on the expression and activation of C/EBPβ in the early stage. We cannot examine whether Rac1 is required for adipogenic differentiation before day 3 by Cre-mediated knockdown of the rac1 gene in the in vitro differentiation system using adipose progenitor cells from mouse WAT, because knockdown of Rac1 was initiated at day 4 ( Figure 1D). Thus, we employed two types of specific chemical inhibitors of Rac1, RI-II and NSC23766, to address the role of Rac1 in the early stage of adipogenic differentiation. SVF cultures derived from control mouse WAT were treated with RI-II or NSC23766 for 24 h prior to the induction of differentiation, and further treated during differentiation. At day 2, the effect of Rac1 inhibitors on protein and phosphorylation levels of C/EBPβ was examined by immunoblot analysis (Figures 5 and 6). Neither RI-II nor NSC23766 affected cell shape and the formation of lipid droplets in the cell (Figures 5A and 6A). In the absence of the Rac1 inhibitor, protein levels of three C/EBPβ variants were markedly increased at day 2 ( Figures 5 and 6). Considering that the increase in the C/EBPβ mRNA level at day 2 was approximately twofold (Figure 3C), it is likely that the rapid increase in the protein level is ascribed to translational upregulation or inhibition of protein degradation. Both RI-II and NSC23766 almost completely inhibited differentiation-associated increase in protein levels of C/EBPβ (Figures 5 and 6). Similarly to the results in the late stage of differentiation (Figure 4), Rac1 inhibition negatively affected the phosphorylation level of LAP*, but not LAP, in the early stage ( Figures 5 and 6). On the other hand, Rac1 inhibitors exerted almost no effect on the phosphorylation level of LIP in the early stage ( Figures 5 and 6), whereas knockdown of Rac1 resulted in the decreased phosphorylation level of LIP in the late stage ( Figure 4).
Role of Rac1 in Differentiation of 3T3-L1 Cells into Adipocytes In Vitro
To further confirm that Rac1 plays an important role in adipogenic differentiation, we next examined the effect of Rac1 knockdown in another in vitro differentiation system using the 3T3-L1 preadipocyte line. Adipogenic differentiation of 3T3-L1 cells was induced according to a standard protocol as described previously [11,12] ( Figure 7A). The day when cells reached confluence was referred to as day 0. Confluent cells were further cultured for two days in DMEM-based growth medium. At day 2, reagents, such as insulin, were added to the culture medium, inducing adipogenic differentiation ( Figure 7A).
We infected 3T3-L1 cells with lentivirus expressing control or Rac1-targeting small hairpin RNA (shRNA), and puromycin-resistant cells, which expressed respective shRNAs, were selected. The expression level of Rac1 as determined by immunofluorescent microscopy was actually suppressed in 3T3-L1 cells that expressed Rac1-targeting shRNA ( Figure 7B,C). Those 3T3-L1 cells that expressed control or Rac1-targeting shRNA were then subjected to the induction of adipogenic differentiation, as described above. At day 8, a large population of control shRNA-expressing cells harbored lipid droplets, which are characteristic of adipocytes ( Figure 7D,E). In marked contrast, lipid droplets were detected in only a small population of cells that expressed Rac1-targeting shRNA ( Figure 7D,E). Therefore, it is plausible that Rac1 is required for adipogenic differentiation of 3T3-L1 cells as well.
Discussion
In this study, we investigated the role of Rac1 in adipogenic differentiation using two different in vitro cell systems: adipose progenitor cells obtained from mouse WAT and the 3T3-L1 preadipocyte line. In both systems, functional deficiency of Rac1 led to the attenuated formation of lipid droplets within the cell, a characteristic feature of adipogenic differentiation. Therefore, it is likely that Rac1 is critically involved in the induction of adipogenic differentiation. This conclusion is consistent with our previous findings that subcutaneous and epididymal WAT in adipo-rac1-KO mice is significantly smaller than in control mice, showing severe atrophy [13]. Furthermore, the size of white adipocytes was reduced in adipo-rac1-KO mice compared with those in control mice [13].
A significant observation that needs to be considered is a difference in the adipocyte type between in vivo and in vitro experiments. Although the adipose progenitor cells used in this study were derived from WAT, adipocytes differentiated from these progenitor cells in vitro contained a number of small lipid droplets, but not a single large lipid droplet: these cells were similar in appearance to brown or beige adipocytes, rather than white adipocytes. Further characterization of adipocytes differentiated from the progenitor cells in vitro will be performed in the future. On the other hand, our results lead to the possibility that Rac1 is implicated in the differentiation into brown adipocytes as well as white adipocytes. We are currently investigating this possibility in vivo and in vitro.
We have provided evidence that the induction of various enzymes for de novo synthesis of fatty acids and triacylglycerol at the mRNA level during adipogenic differentiation largely depends on Rac1 (Figure 2). This notion was also supported by reduced mRNA levels of these enzymes observed in white adipocytes of adipo-rac1-KO mice in our recent study [13]. These findings are important because defects in de novo lipid synthesis due to the insufficient induction of various enzymes may account, at least in part, for the reduced size of white adipocytes and atrophy of WAT in adipo-rac1-KO mice.
Furthermore, we demonstrated that Rac1 contributes to the induction of two transcription factors, PPARγ and C/EBPα, which act as master switches of adipogenic differentiation [24,25] (Figure 3A,B). Towards understanding the mechanisms underlying Rac1-dependent induction of these transcription factors, we then examined the induction of upstream transcription factors-C/EBPβ and C/EBPδ ( Figure 3C,D). Moreover, protein and phosphorylation levels of three variants of C/EBPβ were evaluated by immunoblot analysis both in early and late stages (Figures 4-6).
In the early stage of differentiation (day 2), Rac1 was involved in the rapid increase in the protein level of LAP*, LAP, and LIP. This rapid increase in the protein level is likely to be due to the upregulation of translation or downregulation of protein degradation, given that the increase in the C/EBPβ mRNA level at day 2 was approximately twofold ( Figure 3C). The precise roles of Rac1 in the regulation of translation and protein degradation of C/EBPβ remain incompletely understood, and are currently under investigation. In addition, Rac1 was responsible for the induction of C/EBPβ in the late stage (day 5-day 7) (Figures 3 and 4). Rac1 may be involved mainly in transcriptional regulation in this stage, and the detailed mechanisms need to be clarified in future studies.
We demonstrated that Rac1 was involved in the induction of the mRNA level of the transcription factor SREBP-1c in the late stage of adipogenic differentiation ( Figure 3E). This may also provide an explanation for the decreased expression of various enzymes for the synthesis of fatty acids and triacylglycerol (Figure 2), although the role of SREBP-1c in de novo lipogenesis in white adipocytes remains controversial [23].
Carbohydrate response element-binding proteins (ChREBPs) are also identified as major transcription factors that induce lipogenic enzymes in response to glucose in adipocytes [23]. We did not test the effect of Rac1 knockdown on the induction of ChREBPs in this study, considering that ChREBPs are mostly induced by glucose rather than insulin. However, it is possible that Rac1 knockdown causes insufficient activation of ChREBPs in vivo, because insulin-stimulated glucose uptake in white adipocytes is severely impaired in adipo-rac1-KO mice [13]. This possibility will be tested in our future studies.
Rac1 has been implicated in the regulation of insulin-stimulated glucose uptake in white adipocytes [10][11][12][13]. Considering that glucose is utilized for fatty acid synthesis as well as the production of ATP, defects in glucose uptake may be another major cause of the reduced size of white adipocytes in adipo-rac1-KO mice [13]. On the other hand, fatty acid transport from the circulation into adipocytes is also regulated by insulin [31]. Rac1 may be implicated in insulin-stimulated fatty acid uptake because insulin regulates glucose and fatty acid transport across the plasma membrane by similar mechanisms. In this case, defects in fatty acid uptake may be another cause of the impaired accumulation of lipids in white adipocytes in adipo-rac1-KO mice. The expression level of GPAT1, which is responsible for the synthesis of lysophosphatidic acid from glycerol-3 phosphate and fatty acids, was significantly reduced in cells derived from adipo-rac1-KO mice ( Figure 2E). Therefore, the synthesis of triacylglycerol is expected to be impaired, at least in part, if sufficient amounts of glucose and fatty acids are incorporated from the blood. Further studies will be needed to better understand the mechanisms.
A recent study using a mouse-dedifferentiated fat cell line showed that Rac1 is involved in actin depolymerization-induced differentiation into adipocytes [32]. In particular, insulin-activated Rac1 is thought to regulate the formation of adipocyte-associated cortical actin structures, which is required for the completion of adipogenic differentiation [32]. Therefore, it is likely that Rac1 exerts multiple functions, including the regulation of glucose uptake, the expression of enzymes for lipid synthesis, and cortical actin cytoskeletal rearrangements, in developing adipocytes, and its loss may cause aberrations in these cells.
In contrast to our findings, Rac1 has been implicated in negative regulation of the expression of PPARγ and C/EBPα and the accumulation of lipid droplets in 3T3-L1 cells in response to the activation of integrins [33,34]. Thus, Rac1 may exert multiple functions in response to different stimulations in the different processes of adipogenic differentiation. It is important that the results obtained from the analysis of in vitro differentiation systems are interpreted in terms of their relevance to in vivo observations. In the present study, we revealed novel functions of Rac1 that may account for atrophy of WAT in adipo-rac1-KO mice, and further investigations are required to understand the mechanisms in detail.
Animal Experiments
All animal experiments were approved by the Ethics Committee for Animal Experiments at Osaka Metropolitan University (approval codes 20-74, 20-75, 21-81, 21-82, 22-101, and 22-102) and carried out according to the institutional guidelines of Osaka Metropolitan University. All mice used in this study had a C57BL/6 genetic background. We routinely crossbred rac1 flox/flox mice [35] with adipo-rac1-KO mice to obtain adipo-rac1-KO mice for experiments. Adipoq-Cre transgenic mice [28] were used as controls throughout this study. Mice were fed a normal chow diet and adult (22-to 26-week-old) male mice were used for all experiments.
Induction of Differentiation of Adipose Progenitor Cells in the SVF into Adipocytes In Vitro
The protocol for differentiation of adipose progenitor cells into adipocytes in vitro is also shown in Figure 1B. The day when cells reached confluence was referred to as day 0. At day 0, the culture medium was changed to DMEM (043-30085, Fujifilm Wako) supplemented with 10% (v/v) fetal bovine serum (FBS) (Corning, NY, USA), 2500 IU/mL penicillin, and 2500 µg/mL streptomycin, and cells were cultured for two days. The culture medium was changed to DMEM supplemented with 10% (v/v) FBS, 100 nM insulin, 1 µM dexamethasone (Dex), 500 µM 3-isobutyl-1-methylxanthine (IBMX), 2 µM rosiglitazone, 2500 IU/mL penicillin, and 2500 µg/mL streptomycin at day 2. After two days, the culture medium was changed to DMEM supplemented with 10% (v/v) FBS, 100 nM insulin, 2500 IU/mL penicillin, and 2500 µg/mL streptomycin, and cells were further cultured for two days. At day 6, the culture medium was changed again to the same medium, and cells were cultured for one more day. In some experiments, cells were treated with 25 µM RI-II or 100 µM NSC23766 from day −1.
Immunofluorescent Microscopy
Immunofluorescent microscopy was carried out essentially as described in [13]. Cells were fixed with 40 mg/mL paraformaldehyde in phosphate-buffered saline (PBS) for 30 min. Rac1 was detected with anti-Rac1 and fluoresceinated secondary antibodies. Lipid droplets and nuclei were stained with LipiDye and 4 ,6-diamidino-2-phenylindole, respectively. Images were obtained and analyzed using a confocal laser-scanning microscope (FV1200, Olympus, Tokyo, Japan). Fluorescent intensities were quantified using ImageJ software.
Immunoblot Analysis
Immunoblot analysis was carried out essentially as described in [13]. Proteins separated by sodium dodecyl sulfate-polyacrylamide gel electrophoresis were transferred onto a 0.45 µm-pore polyvinylidene difluoride membrane (Cytiva, Shanghai, China). Membranes were incubated with primary antibodies, and then horseradish peroxidase-conjugated secondary antibodies. Specific proteins were visualized by Chemi-Lumi One Ultra (Nacalai tesque). Images were captured, and densitometric analysis was carried out using a chemiluminescence imaging system (Ez-Capture MG, Atto, Tokyo, Japan).
Induction of Differentiation of 3T3-L1 Cells into Adipocytes In Vitro
Induction of differentiation of 3T3-L1 cells into adipocytes in vitro was carried out essentially as described in [12]. The protocol for differentiation of 3T3-L1 cells into adipocytes in vitro is also shown in Figure 7A. Undifferentiated 3T3-L1 cells were cultured in DMEM supplemented with 10% (v/v) FBS, 100 IU/mL penicillin, and 100 µg/mL streptomycin. The day when cells reached confluence was referred to as day 0. At day 0, the culture medium was changed to the same medium, and cells were cultured for two more days. The culture medium was changed to DMEM supplemented with 10% (v/v) FBS, 100 nM insulin, 1 µM Dex, 500 µM IBMX, 2 µM rosiglitazone, 100 IU/mL penicillin, and 100 µg/mL streptomycin at day 2. After two days, the culture medium was changed to DMEM supplemented with 10% (v/v) FBS, 100 nM insulin, 100 IU/mL penicillin, and 100 µg/mL streptomycin, and cells were further cultured for two days. At day 6, the culture medium was changed again to the same medium, and cells were cultured for two more days.
shRNA-Mediated Knockdown of Rac1 in 3T3-L1 Cells
The TRC2-pLKO1-puro plasmid containing shRNA for mouse Rac1 (GGAGACG-GAGCTGTTGGTAAA, TRCN0000310888) and the nonmammalian shRNA control plasmid (TRC2-pLKO.5-puro non-target shRNA #1) (SHC202) were purchased from Sigma-Aldrich. Either one of these shRNA expression lentiviral plasmids was introduced into HEK-293TN cells with lentiviral packaging plasmids (pMISSION GAG POL and pMISSION VSV-G) using the TransIT-293 Reagent (Takara Bio). Forty-eight hours later, the culture medium containing lentiviruses was collected and then filter-sterilized. 3T3-L1 cells were infected with the lentiviruses at a multiplicity of infection of 4000 in the culture medium supplemented with 7 µg/mL polybrene. Those 3T3-L1 cells that stably expressed shRNA were selected with 2 µg/mL puromycin for three days.
Informed Consent Statement: Not applicable.
Data Availability Statement: The data presented in this study are available on request.
Conflicts of Interest:
The authors declare no conflict of interest. The funders had no role in the design of the study, collection, analyses, or interpretation of data, writing of the manuscript, or decision to publish the results. | 6,613.6 | 2023-02-27T00:00:00.000 | [
"Biology"
] |
Direct measurements of phonon–phonon scattering in liquid 4He
We have measured the scattering between two phonon-beams, at different angles, in HeII. When the angle between their propagation directions is 30°, we detect scattering between the low-energy phonons (ϵ/kB ∼ 1 K) in the two beams. No scattering is seen between low-energy phonon beams at 40° due to the low three phonon scattering rate at large angles. At 30° and 40°, we detect scattering between the high-energy phonons (ϵ/kB > 10 K) in the two beams for the first time. This is due to four phonon scattering. There is no detectable scattering between the low- and high-energy phonons at these small angles. When beams collide nearly head-on, there is scattering between the low- and high-energy phonons, and also scattering between the high-energy phonons and R+-rotons. The power and pulse length dependencies of these interactions are presented and discussed. The results are analysed to obtain the lifetime of the probe phonons in the scattering beam of phonons.
DEUTSCHE PHYSIKALISCHE GESELLSCHAFT
the fastest phonon scattering process and plays the dominant role in phonon creation and decay at high momenta.
The interplay between 3pp, 1-to-n and 4pp can be seen clearly when a short (100 ns) heat pulse is injected into pure and cold (50 mK) liquid 4 He. The heat is initially in the form of low-energy phonons which rapidly come into a quasi-equilibrium though 3pp in a time, ∼10 −8 s [2]. This is an order of magnitude shorter than the duration of the heat pulse. These phonons, in quasi-equilibrium, occupy a very limited angular segment of momentum space: to a first approximation they occupy a cone with a cone angle typically 10 • and cone axis along p z where z is the direction of propagation of the pulse in real space. So these phonons form a highly anisotropic phonon system.
Within this pulse, phonons with momentum p > p c are readily created by 4pp. The process may be represented as l 1 + l 2 → h + l 3 , where l i are low momentum phonons with p < p c , and h is a high momentum phonon with p > p c . We shall refer to these types as l-phonons and h-phonons. Once the h-phonon has been created it has a relatively long lifetime within the pulse, typically ∼10 −6 s [3], and before it has time to decay, it is left behind by the pulse because its group velocity is considerably less than the velocity of the l-phonon pulse, i.e., 189 and 238 ms −1 respectively. Once the h-phonon has been left behind, it is in helium with an insignificant number of thermal phonons and so its lifetime is only limited by the size of the container.
The result of the processes described above means that a single heat pulse creates two propagating phonon pulses in the liquid helium; a fast non dispersed l-phonon pulse that travels at the velocity of sound [14], and a slower dispersed h-phonon pulse [15]. It is dispersed because higher momentum phonons have lower group velocities and are created at all distance in the liquid. Although the behaviour of this system depends on 3pp and 4pp, the details of these processes are difficult to extract from the collective behaviour. This motivated the present study; we wanted to make measurements where the details of the scattering processes could be determined.
Much is already known about 3pp scattering. The fact that phonons spontaneously decay was established by measuring the widening of a collimated beam, as it propagated [16]. There are several good indications that 3pp spontaneous decay rates are very fast. The neutron scattering linewidth was measured using neutron spin echo [17]. It was found that the scattering rate increased rapidly with momentum and reached a peak value of 4 × 10 9 s −1 at wavevector q = 0.4 Å (or cp/k B = 7.2 K). According to theory, the spontaneous decay rate varies as q 5 [4,5], and at q = 0.4 Å the scattering rate is 1.45 × 10 10 s −1 , which is much higher than the measured value.
When phonons are in a strongly interacting group, the scattering rates are much higher than the spontaneous decay rate. For example, for a phonon with cp/k B ∼ 1.5 K in the isotropic system at T = 1K, the scattering rate is ∼ 2 × 10 9 s −1 , see figure 3(b) in [2] and in a typical anisotropic system the scattering rate is ∼2 × 10 8 s −1 see figure 5 in [2], compared with the theoretical spontaneous decay rate of ∼7 × 10 6 s −1 [4,5].
In [18], it was shown that the l-phonon pulse has an area, typically >1 mm 2 , where the energy density is constant. This is due to h-phonon creation which cools the l-phonons to ∼0.7 K where the h-phonon creation rate is very low. The constant energy area is termed a 'phonon sheet' as its transverse dimensions are much greater than its thickness. At typical distances from the heater the sheet is nearly flat.
3pp scattering involves small angles between the two phonons created or annihilated. This was first shown by [16]. Recently, this has been seen in more detail by colliding two phonon sheets together [19]. The strongest interaction between the sheets occurs when the normals to the plane 4 DEUTSCHE PHYSIKALISCHE GESELLSCHAFT of the sheets make an angle equal to the angle between typical phonons in a 3pp interaction, i.e., 12 • to each other [20]. For two beams crossing at α ∼ 30 • , the scattering lifetime of the l-phonons is very long compared with the lifetime within the beams, because between the beams, most of the angles between phonons are larger than the maximum angle for three phonon scattering.
4pp scattering rates have been determined by measuring the attenuation of a beam of highenergy phonons in liquid helium at different temperatures [7]. The h-phonons are scattered by the ambient phonons whose energy and density increase with temperature. The measured scattering rates agreed with rates calculated with the corrected theory [7,6]. In anisotropic phonon systems, the scattering rates have been calculated [21,22].
A h-phonon is created by 4pp scattering. The rate for h-phonon creation, in a pulse of l-phonons, is 2 × 10 4 s −1 , as we know h-phonons are created in less than the propagation time 4.2 × 10 −5 s in experiments [15]. Theory predicts it to be of order 10 4 s −1 [3], but such rates depend strongly on the h-phonon momentum, and the energy and momentum of the l-phonon pulse [3]. The measured h-phonon signal has been successfully modelled at low pulse energies [23,24].
Specific predictions for the angular distribution of phonons in equilibrium have been made, it is the Bose distribution which is a function of two parameters, the temperature T and the drift velocity of the pulse u [25]. The number of phonons, in unit volume, with energy ε(p), and momentum p in the range p → p + dp, at an angle θ to the symmetry axis in the range θ → dθ, is given by [25] Equation (1) shows that phonon states are most strongly occupied near θ = 0, the propagation direction, and the distribution falls off rapidly as θ increases. However, the occupation is not zero at any value of θ. The data reported here will test the validity this distribution function.
The decay rate of h-phonons in the l-phonon pulse, is strongly curtailed by the l-phonons only occupying a narrow range of angles in momentum space, see equation (1). In contrast, this narrow angular occupation has only a small effect on the creation rate. This asymmetry between creation and decay rates is due to the asymmetry in the angles between phonons in the 4pp; the angles between phonons of similar momentum are smaller than the angles between phonons of dissimilar momentum. For example consider 4pp scattering where phonons with energy 7 and 4 K combine and give two phonons with energy 10 and 1 K: for angles between the 7 and 4 K phonons in the range 0-10 • , the angles between the 10 and 1 K phonons are in the range 53-54 • . This follows directly from the measured dispersion curve and conservation of momentum and energy. Hence, h-phonon creation involves small angles and h-phonon decay involves large angles. As the l-phonons in the pulse are mostly in a narrow cone, and the h-phonons are mainly along the cone axis, the angles between the h-and l-phonons are mostly small. Hence, there are no l-phonons at the angle needed to scatter the h-phonons and so their decay rate is correspondingly low.
The experiments described in this paper are designed to investigate 3pp and 4pp between phonons at different angles, in a simple and well-defined experimental arrangement. This means that it should be possible to theoretically model the interactions and hence calculate the scattering rates in some detail. The paper is organized as follows. In section 2, the design of the experiment is discussed, in sections 3 and 4 the results of scattering at small and large angles respectively, shows in the top panel, a signal from a pulse which has been through a scattering beam, and in the bottom panel, a signal from an unscattered pulse. There are two cross-talk spikes in the top panel, from the two heater pulses. The first at t = 0 is from the scattering pulse and the second at t = 14.3 µs is from the probe pulse. In the top panel, the l-phonon signal at t ∼ 65 µs is clearly smaller than the l-phonon peak in the bottom panel. The peak of the h-phonon signal at t ∼ 78 µs in the top panel, is smaller than the h-phonon peak in the bottom panel, but this is due to the attenuation of the tail l-phonons, due to the cooling of the substrate, which extends under the h-phonons. There is negligible attenuation of the h-phonons due to h-h scattering at this delay. are presented. In section 5, expressions for the scattering lifetimes are derived and some values calculated, and in section 6, we draw conclusions.
Design of the experiment
The experimental method consists of crossing two beams of phonons and measuring the attenuation of one beam due to the other. The detected beam is called the probe beam and the other is called the scattering beam. A schematic of the experiment is shown in figure 1(a). The cylindrical collimators are machined from one piece of brass. They have internal radii of 3.0 and 6.5 mm and are 0.5 mm thick. The collimation holes are 1 mm diameter. Between heaters H0 and H4 the angle is 30 • , and between heaters H1 and H2 the angle is 40 • . The support for heaters H1, H2 and H4 has a radius 10.0 mm. The heater substrates are glass cover slips 0.12 mm thick. Heater H0 is glued on the inside of the outer collimator and H5 is glued on the inside of the inner collimator directly below the collimation hole. Bolometer B0 is glued on the inner surface of the outer collimator, and the other bolometers, B1, B2 and B4 are on a 5 mm radius and in line with their heater and collimation.
The attenuation of the probe beam is defined by where s 0 and s 1 are the phonon signals in the probe beam without any scattering and after passing through the scattering beam, respectively, and n 0 and n 1 are the corresponding phonon fluxes. The beams are pulsed so that the times of flight determine which groups of phonons are interacting. The time that a volume element of the probe pulse overlaps the scattering pulse is called the crossing time [25] and it is determined by the lengths and widths of the pulses in the liquid helium. The l-phonon pulse has a length ct p where t p is the duration of the current pulse in the heater, typically 10 −7 < t p < 10 −6 s. The h-phonon pulse is much longer and has a typical width at half height of 10 −5 s, see figure 1, which is independent of t p for t p 10 −6 . The transverse dimensions of the l-and h-phonon pulses are typically 1 mm, and are determined by the collimation. For there to be scattering, the crossing time must be longer than the inverse of the scattering rate.
As the scattering processes vary with the angle between the propagation directions of the pulses, the experiment is designed to cross beams at several angles. The arrangement is shown figure 1(a). The smallest angle between two beams is 30 • ; it is limited by the path length and the diameter of the collimating holes. Increasing the path length and decreasing the diameter of the collimating holes, increases the angular resolution but reduces the detected signal, so the geometry in figure 1(a) is a compromise between these effects. The largest angle is ∼160 • , so the beams are then meeting nearly head-on.
At small angles, we delay one pulse with respect to the other so that scattering between different phonons is selected, i.e. l-l, l-h and h-h phonon scattering. Delay is not useful at large angles as the pulses pass through each other over a wide range of delays.
The heaters are thin films of gold evaporated onto glass cover slips with area 1 mm 2 . These are glued onto the brass with GE 7031 varnish. The phonon pulses are detected with superconducting zinc bolometers made from a zinc film 1 × 1 mm, cut into a serpentine track. The zinc film was in a constant magnetic field, parallel to its plane, and held at its superconducting transition edge by a feedback circuit [26]- [28], which maintains a constant bolometer resistance at ∼0.1 of its normal value at 4.2 K. The corresponding bolometer temperature is ∼0.35 K. The cell was cooled to ∼50 mK with a dilution refrigerator and filled with isotopically pure 4 He [29].
Heaters were pulsed in pairs with currents from a LeCroy 9210 pulse generator with two independent but synchronous outputs, one of which could be delayed with respect to the other. Heater powers in the range 3.125-25 mW were used. The signals from the bolometer were amplified with an EG&G PAR 5113 preamplifier and digitally recorded with a Tektronix DSA 601A. Many signals were averaged to obtain a good signal to noise ratio.
Results and discussion at small angles
In the first experiment, the probe beam was collimated with a hole 1 mm diameter in a shield 0.5 mm thick, which was 3 mm in front of the heater H0, see figure 1(a), and a similar collimator 3 mm in front of the bolometer B0. The total path length was 12.8 mm. The scattering beam had two similar collimators 3 and 6.5 mm in front of its heater H4. The two lines through the centres of the collimators crossed at 6.4 mm in front of the probe heater and 9.9 mm in front of the scattering heater. The times for the various phonon interactions were calculated from these distances and the phonon velocities; 238 ms −1 for l-phonons and 189 ms −1 for h-phonons. The times for scattering between the l-and h-phonon peaks, i.e., l p -l s , l p -h s , h p -l s and h p -h s scattering, where the subscripts denote the probe and scattering beams, are respectively 14.7, 25.5, 7.7 and 18.5 µs. A guide to the results is shown in table 1. Figure 1(b) shows the probe signals with and without the scattering pulse at a delay of 14.3 µs. We see that the l-phonon signal is decreased by the scattering beam. Figure 2 shows the attenuation of the peak of the l p -phonons as a function of the delay of the probe beam relative to the scattering beam. The most significant attenuation of the probe beam is at 14.3 µs. From this it is clear that the strongest interaction, at an angle of 30 • , is between the l-phonons of both beams.
There is a weaker attenuation of the h p -phonons at 19µs which is shown in figure 3(a). This delay indicates that there is an interaction between the h p -and h s -phonons at this angle. The h p -phonon attenuation occurs over a wider range of delays than the l p -phonon attenuation because the h-phonons are more dispersed than the l p -phonons. There is no interaction between the l-phonons of one pulse and the h-phonons of the other pulse at this angle.
The beams can be interchanged by detecting the pulse from heater H4 and scattering it with the pulse from heater H0. The general behaviour is the same as before, there is strong l-l scattering and weak h-h scattering. The attenuation of the h p -phonons is shown as a function of delay in figure 3(b). Using pulses from heaters H1 and H2, at 40 • to each other, there is only h-h attenuation, as shown in figure 3(c); there is no measurable l-l attenuation at this angle. This is to be expected because the 3pp scattering rate is very low at 40 • because of the small number of phonons at the right angle for 3pp scattering. Again, there was no sign of l-h or h-l scattering. Figure 4 shows the effect of varying the pulse length of the scattering pulse in the first arrangement, while the probe pulse is constant at 6.25 mW and 100 ns. The delay is constant at 14.3 µs, so the scattering is between l-phonons in the two pulses. We see that the attenuation initially increases linearly with pulse length, then increases more slowly and saturates when the scattering pulse is more than ∼1 µs. The value of the saturation attenuation is only at 0.8 and not 1, which shows that, due to the angular spread of phonon momenta, some of the l-phonons in the probe beam are at too large an angle to scatter with phonons in the other beam. Figure 5 shows the effect of varying the power in the scattering pulse. The phonon density increases with increasing pulse power which leads to more scattering and a higher attenuation. The power in the probe beam is constant at 12.5 mW, and both pulses are 100 ns duration. The delay is again 14.3 µs so selecting l-l scattering. The attenuation initially increases linearly with scattering pulse power and then increases more slowly. With the powers used in this experiment, we do not see the attenuation saturate as it did with pulse duration.
When two beams collide, both beams lose the same number of phonons. Usually the probe beam is made much weaker than the scattering beam so the probe beam is significantly attenuated but the scattering beam remains nearly constant. If both beams are equal, then they both are attenuated the same amount, and if the probe beam is stronger than the scattering beam then the attenuation of the probe beam saturates when all the scattering beam has been scattered away. In figure 6, the power in both beams is increased together but, because of the different distances and collimation, the scattering beam is weaker than the probe beam. The pulse duration of both beams is 100 ns and the delay is 14.3 µs so that the scattering is l-l, and the angle between the beams is 30 • . The attenuation of the probe beam is low at low powers because the phonon densities are low which makes the scattering rate low. It rapidly increases with power and at 12.5 mW it has saturated at an attenuation of 0.32. This indicates that the probe beam has 1/0.32 ∼ 3 times the phonon density of the scattering beam.
Scattering at large angles
When the angle between the probe and scattering beams is large, the l-l scattering will be zero because the angle is too large for 3pp scattering. Hence, the attenuation of h-phonons can be by either h-l or h-h scattering. In this section, we present results where the angle between the beams is ∼160 • . At such a large angle, the beams overlap for such a long time that different interactions cannot be well separated by times of flight. However, the delay run shows that the attenuation of the h p -phonons is constant for delays between 7 and 20 µs and then the attenuation decreases linearly to zero at a delay of ∼47µs. At this delay, the h p -phonons reach the heater H5 at the start of the scattering pulse (9.4 mm/189 ms −1 = 50 µs), so we would expect no attenuation at this delay. A delay of 20 µs is consistent with the h p -phonons meeting the l s -phonons at 6.4 mm from heater H0: the l s -phonons have travelled 3.2 mm from heater H5. The h s -phonons are strongly concentrated in the direction normal to the heater H5 [3,30] this direction is parallel to, but off-set, from the h-phonons in the probe beam and so little interaction between these h-phonons and the h-phonons of the probe beam is expected. The concentration of the h s -phonons along direction normal to heater H5 also explains the observation that the l-phonons of the probe beam are not noticeably attenuated by the h-phonons from heater H5. All this suggests that the attenuation of the h-phonons in the probe beam is due to h p -l s scattering rather than h p -h s scattering. Further evidence for this conclusion is given by figure 7. Figure 7 shows the attenuation of the peak of the h-phonon probe signal as a function of the pulse length of the scattering beam. It is measured at a delay of 17 µs where the attenuation is independent of delay. The probe beam is 12.5 mW and 100 ns, and the scattering beam power is 25 mW, the highest power used in these experiments. The figure shows that the attenuation rises with pulse length. Previously, we have established that the h-phonon signal saturates at a heater pulse duration ∼0.3µs [31], where the h-phonons are scattered within the l-phonon pulse and so do not escape the pulse, i.e., ∼0.3µs is the boundary between short-and long-pulse regimes. In contrast, the l-phonon density increases linearly with pulse length see figure 6 of [18] which is similar to the behaviour of the attenuation shown in figure 7, and there is no sign of the attenuation saturating at a scattering pulse length of 0.3 µs where no more h-phonons are created. This supports the conclusion that the h-phonons of the probe beam scatter with the l-phonons of the scattering beam. Figure 8 shows the attenuation of the peak of the h-phonon probe signal as a function of the power of the scattering beam. The probe beam is 12.5 mW and 100 ns, and the scattering beam has a pulse length of 1 µs. The attenuation rises with pulse power in a similar way to the rise in the l-phonon signal, see figure 7 of [18]. Again this power dependence supports the idea that these l-phonons are the cause of the scattering. (The attenuation does not appear to go through zero power but this offset is within the random error of ∼2%, it stems from the reference probe being a little too small when there is no scattering pulse.) When the probe pulse length is 100 ns and the scattering beam is 25 mW and 500 ns, the attenuation of the h-phonons in the probe beam is 17%, independent of probe power in the measured range 3 < W p < 25 mW. As the probe beam is always much smaller than the scattering beam then the scattering beam is unaffected by the probe beam and so the fractional decrease in the h-phonon probe signal is independent of probe beam power, as is found. Figure 9 shows the effect of varying the pulse width of the probe pulse. The power of the probe beam is 12.5 mW and the scattering beam is 25 mW and 500 ns. The attenuation decreases as the pulse width increases which is to be expected when the probe pulse scatters away a significant part of the scattering beam; the later part of the probe beam is attenuated less than the leading part, and so a long pulse, as a whole, is attenuated less than a short pulse. Figure 9 shows that this happens when the probe pulse is longer than ∼200 ns. For very long probe pulses the attenuation should asymptotically approach zero. Finally at this angle the scattering beam was set to a low power, 3 mW, and very long-pulse lengths, up to 10 µs. Under these conditions the heater creates mainly R + -rotons and very few h-phonons [31]. Figure 10 shows that, for a probe beam of 12.5 mW and 100 ns, the h-phonons are scattered by rotons. The scattering increases with scattering beam pulse length which just increases the number of rotons in the dispersed pulse of rotons.
Analysis
A probe beam passing through a scattering beam, where the probe beam density is much smaller than the density of the scattering beam so the scattering beam density is not altered by the scattering, decays exponentially with time t. The probe phonon density after scattering, n(t), is given by where τ is the lifetime of a probe beam phonon in the scattering beam and n 0 is the initial probe density. The time that an element of the probe pulse is in the scattering pulse is given by the crossing time, t x , [25]. For two sheet-like pulses, where the dimension in the propagation direction is much smaller than the transverse dimensions, this is given by [25] where t p,s is the duration of the scattering pulse and α is the angle between their propagation directions.
For long and collimated pulses, the phonons are in a box-shaped volume where the dimension of the pulses in their propagation directions is larger than their transverse dimension w in the plane of the heater normals. The crossing time for these box-shaped pulses, at small values of α, can be derived from their relative velocity 2v sin (α/2) where v is the speed of the pulses, and the length of the path, w/cos(α/2), of the probe pulse through the scattering pulse in the direction of their relative velocity. t x is given by For these long pulses, the ends of the pulses experience a shorter crossing path than w/cos(α/2), so the crossing time will be overestimated by equation (5), as it is also for cylindrically shaped pulses. From equations (2) and (3) and putting t = t x , the attenuation is given by For t x τ the attenuation is small and is given by Combining equations (4) and (7), we obtain an expression for the scattering lifetime, for sheet-like pulses and with t x τ Now A/t p,s is just the initial gradient of the attenuation versus pulse length plot, see for example figure 4, hence equation (8) enables τ to be calculated. For box-like pulses when t x τ, we combine equations (5) and (7) and obtain another expression for the scattering lifetime τ = w Av sin(α) .
For l-l scattering with the pulses at α = 30 • , the initial gradient of the graph in figure 4 is 2.25 × 10 6 s −1 for the scattering pulse power of 12.5 mW. This gives, using equation (8), the lifetime τ = 3.3 × 10 −6 s. This lifetime is some two orders of magnitude longer than for phonons in equilibrium at 1 K [2]. The reason for the long lifetime in the crossed beams experiment is because there are very few phonons in the two beams which have a small enough angle between them, to interact. This is also the reason why a hot line is not formed by two pulses at this angle [20].
For h-h scattering, we use equation (9) because the h-phonon pulses are very dispersed. Taking w = 1 mm, v = 189 ms −1 and A = 0.04 from figure 3(a), we find the lifetime of a h-phonon in the scattering h-phonon pulse, created by a heater pulse of 12.5 mW and 100 ns, is τ = 2.6 × 10 −4 s.
For h-l scattering, we can use equation (8). From figure 7, we find A/t ps = 3.5 × 10 5 s −1 which gives the lifetime of a h-phonon in the l-phonon pulse, τ = 1.5 × 10 −6 for scattering pulse power of 25 mW.
Conclusions
From the attenuation of various phonon probe pulses as they pass through pulses of other phonons, we have extracted the lifetime of probe phonons due to scattering with phonons in the scattering pulse. This lifetime is proportional to the density of phonons in the scattering pulse, which is determined by the pulse power, the collimation, the transverse expansion of the l-phonon sheet, the temperature and drift velocity of the l-phonons, and the details of the creation of h-phonons. Further, theoretical study is required to estimate the densities. However, the conditions of the experiment are well defined so we hope that in the future a detailed comparison with theory can be made.
When pulses cross at 30 • , l-phonons in the probe pulse scatter with the l-phonons in the scattering pulse. The actual angles between the phonons are given by the angular spread of momenta in each pulse, see equation (1), as well as the angle between the heater normals. The l-l scattering rate at these relatively large angles, which are comparable with the maximum angle of 27 • for 3pp [2], is much slower than the l-l scattering rate in liquid helium at 1 K. This is due to the relatively small number of phonons at an angle where 3pp are possible. At 40 • between the heater normals there is no sign of l-l scattering, as then there are even fewer phonons at angles for 3pp scattering.
There are h-h interactions at 30 • and 40 • , see figure 3, but the attenuation is small. However, this is the first time that scattering between phonons, both with energy greater than 10 K, has been detected and there is no theoretical calculation of the 4pp scattering rate for such phonon interactions. No h-l scattering is seen with pulses intersecting at these angles. This is probably because a larger angle between the h-and l-phonon is needed to satisfy both energy and momentum conservation in 4pp scattering.
When the beams intersect nearly head-on, the h-phonons of the probe beam are attenuated. We have argued that this is due to these phonons scattering with the l-phonons of the scattering pulse, because only these phonons intersect the h-phonons in the probe pulse, as the h-phonons of the scattering pulse are narrowly beamed along the heater normal, so they are parallel to, but offset from, the h-phonons in the probe pulse. The lifetime of a h-phonon scattering with low-energy phonons is around 10 −6 s. When the pulse to the scattering heater is long and low power, R + -rotons are created, and we see the h-phonons of the probe beam attenuated by these R + -rotons. This is the first observation of such scattering.
We hope these results will stimulate the development of calculations of 3pp and 4pp lifetimes for the conditions in these experiments. | 7,421.4 | 2007-03-01T00:00:00.000 | [
"Physics"
] |
Atmospheric Measurement Techniques On the improvement of NO 2 satellite retrievals – aerosol impact on the airmass factors
The accurate determination of nitrogen dioxide (NO2) tropospheric vertical columns from satellite measurements depends strongly on the airmass factor (AMF) used. A sensitivity study was performed with the radiative transfer model SCIATRAN to better understand the impact of aerosols on the calculation of NO 2 AMFs. This influence was studied by varying the NO 2 and aerosol vertical distributions, as well as physical and optical properties of the particles. In terms of aerosol definitions, the key factors for these calculations were identified as the relation between trace gas and aerosol vertical profiles, the optical depth of the aerosol layer, and single scattering albedo. In addition, surface albedo also has a large impact on the calculations. Overall it was found that particles mixed with the trace gas increases the measurements’ sensitivity, but only when the aerosol is not very absorbing. The largest change, a factor of ∼2 relative to the situation without aerosols, was found when a low layer of aerosol (600 m) was combined with a homogenous NO2 layer of 1.0 km. A layer of aerosol above the NO 2 usually reduces the sensitivity of the satellite measurement. This situation is found mostly for runs with discrete elevated aerosol layers (representative for long-range transport) that can generate a decrease of the AMF values of up to 70%. The use of measured aerosol profiles and modelled NO 2 resulted, generally, in much smaller changes of AMF relative to the pure Rayleigh case. Exceptions are some events of elevated layers with high aerosol optical depth that lead to a strong decrease of the AMF values. These results highlight the importance of aerosols in the retrieval of tropospheric NO2 columns from space and indicate the need for detailed information on aerosol properties and vertical distribution. Correspondence to: J. Leit̃ao<EMAIL_ADDRESS>
Introduction
The possibility of measuring trace gases (e.g.ozone (O 3 ), nitrogen dioxide (NO 2 ), sulphur dioxide (SO 2 ), among others) from space provides a unique opportunity to observe the Earth and its atmosphere from above and, consequently, monitor the air quality in remote places with low density of in situ measurements.Remote sensing of atmospheric pollution is currently performed by several instruments flying on satellites.One common technique is the use of backscattered solar radiation from which information can be retrieved on the amounts of aerosols and trace gases in the atmosphere.Some instruments were mainly developed for trace gas observations, as it is the case for GOME (Burrows et al., 1999) flying on ERS-2, SCIAMACHY (Burrows et al., 1995;Bovensmann et al., 1999) on the ENVISAT platform, OMI (Levelt et al., 2006) on EOS-AURA and more recently GOME-2 (Callies et al., 2000) launched on MetOp-A.These also provide information on aerosols (mainly aerosol optical depth (AOD) but also, e.g., aerosol size distribution) albeit at low spatial resolution.Other instruments such as MODIS (King et al., 1992) flying on Terra and Aqua, MISR (Diner et al., 1998) also on Terra measuring with multi-angle viewing directions, or MERIS (Bézy et al., 2000) on the ENVISAT platform, are better suited for aerosol retrievals since they provide high spatial resolution and, in some cases, multiple viewing directions.More recently, the active lidar system CALIOP (Winker et al., 2003) flying on the CALIPSO satellite has become available which for the first time can resolve aerosol vertical distributions with high resolution.
Nitrogen oxides (NO x =NO+NO 2 ) can be considered one of the main pollutants present in urban and industrialized areas, originating mainly from fossil fuel combustion processes.NO x is also emitted from biomass burning events and via natural processes mostly during lightning events and as result of microbial processes in soils.NO x play a key role in the basic tropospheric chemistry acting as a precursor Published by Copernicus Publications on behalf of the European Geosciences Union.
for photochemical ozone production and acidification of the atmosphere via nitric acid (Seinfeld and Pandis, 1998).Furthermore, they also contribute to global climate change by interfering, directly and indirectly, with the Earth's radiative budget (Solomon et al., 1999;IPCC, 2007;Vasilkov et al., 2009).According to the recent Fourth Assessment Report (AR4, IPCC, 2007 and references therein), anthropogenic NO x emissions have increased drastically since preindustrial times.While emissions in some industrialized countries have decreased over the last decade in response to emission reduction measures and use of cleaner fuels, emissions in the rapidly developing economies in Asia are expected to continue their increase.In the mid-1990s NO x emission rates for Asia were reported by Akimoto (2003) to exceed the amount emitted in North America and Europe, and in 2005, Richter et al. found a high enhancement of NO 2 columns over China measured by GOME and SCIAMACHY.While the global budget of anthropogenic sources is relatively well constrained, the natural emissions are still rather uncertain.Ground-based measurements and model studies aim on assessing pollution levels and analyse evolution trends.To estimate accurate NO 2 concentrations, the satellite datasets offer advantages both on the time and spatial scales with a nearly global coverage that can often be achieved with high temporal resolution (approx. 1 to 6 days depending on the instrument coverage).
Although a great fraction of aerosols is part of the natural components of the Earth's atmosphere, they can still be harmful for human health and contribute to visibility degradation when present in high amounts (Chang et al., 2009 and references therein;Wang et al., 2009).In addition to their relevance as pollutants, they play a major role in climate change by their direct and indirect impact on radiative forcing (e.g.IPCC, 2007).Aerosols vary strongly in size and composition, and their proper characterization, in particular using remote sensing, is still a challenge.In recent years numerous experimental studies focused on aerosols.Still, the formation processes, transport and transformation of aerosols are not completely understood.Their high spatial and temporal variability (especially for tropospheric aerosol) represents a complication to the process of identifying and quantifying its sources and types.
Aerosol present in the atmosphere interacts with radiation consequently influencing the remote sensing measurements of atmospheric trace gases.Depending on the particles' optical properties, the amount of aerosol and its vertical distribution relative to that of the trace gas of interest, the sensitivity of the satellite measurements can either be increased or decreased.As the anthropogenic sources of aerosols and other pollutants are often collocated, a proper characterization of the aerosols' impact on the retrieval is needed to accurately quantify trace gas amounts derived from satellite observations.This is of particular importance if long-term trends of, for example, tropospheric NO 2 are studied which are accompanied by large changes also in the aerosol loading.
Several aspects contributing to the total error in the determination of tropospheric NO 2 columns from the satellite measurements were studied by Boersma et al. (2004).In that analysis it was reported that including realistic aerosol in the radiative transfer calculations would increase the airmass factors by up to 40% depending on aerosol type and aerosol optical depth.The vertical profile of the aerosols was assumed to be exponential with a scale height of 2.0 km and the vertical NO 2 profile was not specified.The conclusion from this study was that the correction for the aerosol impact cannot be simply separated from the effect of clouds and, therefore, if a cloud retrieval scheme is adopted, it will account for a large part of the aerosol effect by retrieving a different cloud fraction and height.Martin et al. (2003) also analysed the aerosol impact on the airmass factor applied in the retrieval process.Monthly aerosol properties derived with the GOCART model were used for that study.The authors found that biomass burning aerosol and desert dust would reduce the AMF by 10-20% while over industrial regions an increase of 5-10% was observed.
A comparable sensitivity study to the one presented here was carried out by Gloudemans et al. (2008) to analyse, among other aspects, the impact of aerosols in the retrieval of CH 4 and CO (in the IR region) from the SCIAMACHY instrument.One of the main findings from this study was that, depending on the location of the plume and type of aerosol, the omission of aerosol influence in the retrieval process can lead to significant errors in the total column of CH 4 .Thomas et al. (2005), with a similar study for SO 2 (retrieved in the UV region) concluded that aerosols are relevant mainly for optical thickness above 0.3 and in the presence of desert dust plumes in the boundary layer (BL).If these two conditions were realised at the same time the authors estimated that the column would be underestimated by 5-10%.For the TOMS SO 2 retrieval, Krueger et al. (1995) showed that neglecting a rather thin aerosol layer may result in a systematic overestimation of the retrieved total SO 2 content.Focusing on HCHO retrieval, Fu et al. (2007) have also analysed the sensitivity of the AMF to aerosol definitions.From the results, the relative vertical distribution of the trace gas and aerosol was identified as major factor influencing the AMF.A strong enhancement of the AMF was observed for the case of an aerosol layer standing below the HCHO.
Furthermore, recently, the aerosol impact on ground-based zenith-sky DOAS measurements was also investigated by Chen et al. (2009).The aerosol effect was studied by changing the vertical distribution of aerosol and NO 2 layers, together and independently of each other, and varying also the single scattering albedo (SSA).From this analysis an error of 10% was determined.Nevertheless, for these measurements, the uncertainties caused by unknown aerosol properties and vertical profiles of both aerosol and NO 2 tended to cancel each other.
The present study assesses the importance of aerosol for the retrieval of tropospheric NO 2 columns from satellite observations in clear sky cases.The effects of clouds in the retrieval were already analysed in detail in many previous studies such as Boersma et al. (2004), Wang et al. (2005) and Kokhanovsky and Rozanov (2009).Therefore, this study focussed mainly on the aerosol effect and, to this end, a number of different scenarios were defined with varying aerosol settings and NO 2 distributions.Those were used for radiative transfer calculations with the SCIATRAN model (Rozanov et al., 2005) in order to evaluate changes in the measurements' sensitivity relative to a scenario without aerosols.The results show to which parameters the measurements are most sensitive and to which extent the modification of aerosol properties affects the results.As far we are aware, a comprehensive study as ours has not been conducted before.In this analysis the impact of clouds was not taken into account in the calculations but will be discussed briefly after the main results.
The paper starts with an introduction of aerosol effects on radiative transfer and a short description of the satellite retrieval and the AMF concept.In addition, a description is provided for the selection process of data used and definition of the case scenarios considered.The following section focuses on the results of the sensitivity study, presenting the variations of NO 2 airmass factors in response to changes in the surface albedo values, the boundary layer height, variation in the aerosol layer distribution and use of different single scattering albedo values.A discussion on the possible cloud effects and current algorithms is presented in Sect.5, followed by general conclusions in the final section of this manuscript.
The effect of aerosols on the radiative transfer
Satellite measurements of tropospheric trace gases using scattered solar light are based on detection of the absorption along the light path from the sun through the atmosphere to the instrument.For an ensemble of photons, scattering is regarded as a statistical process, where many different light paths contribute to the signal observed at the top of the atmosphere.For an optically thin absorber, the overall absorption signal is determined by the amount of absorption along the individual light paths weighted with their relative contributions to the total radiance measured.In comparison to a pure Rayleigh atmosphere, the presence of aerosols can change both the individual light path lengths and their contributions to total radiance observed at the satellite.
Qualitatively, the effects of an aerosol layer on tropospheric measurements using scattered sun light can be separated into four groups: -light path enhancement within the aerosol layer as result of multiple scattering, leading to an increase in absorption signal from the path between scattering events; -increased sensitivity within and above the aerosol layer as result of larger scattering probability and therefore a larger contributions of these paths to the radiance observed at the satellite (albedo effect); -decreased sensitivity below the aerosol layer as more photons are scattered back to the satellite before they can reach these altitudes (shielding effect); -decreased sensitivity within and below the aerosol layer in cases of strongly absorbing aerosols as the number of photons returning from this region is reduced.
With the exception of the last point, the effects of aerosols listed above are very similar to the considerations made for clouds (e.g.Hild et al., 2002;Beirle et al., 2009;Kokhanovsky and Rozanov, 2009).The overall impact of aerosols on a measurement will depend on the relative importance of the above mentioned effects which depends mainly on aerosol properties, aerosol amounts, surface reflectance and the vertical distribution of aerosol and trace gas of interest, but also on the solar zenith angle (SZA) and satellite viewing angle.The results can be both, an increase or a decrease in observed absorption signal depending on the specific conditions.
The AMF
One way of expressing the sensitivity of the measurement is to calculate the airmass factor (AMF) which is defined as the ratio between the apparent (slant) column (SC) of the absorber retrieved from a measurement and the vertical atmospheric column (VC, Solomon et al., 1987): The definition can be generalised by applying it to discrete layers in different altitudes, defining the block airmass factor (BAMF) for a layer i: which describes the change of airmass factor with altitude.
The AMF can be computed from the BAMF by weighting it with the atmospheric absorber profile: where VC (i) is the vertical column of the absorber in layer i and N is the total number of layers.More detailed discussions of the airmass factor concept can be found in Wagner et al. (2007) and Rozanov and Rozanov (2010).
The larger the airmass factor, the higher the sensitivity of the measurement.The aerosol effects discussed in the previous section are illustrated in Fig. 1 where the NO 2 block airmass factor is shown as function of altitude for three different scenarios, highlighting the increased sensitivity in the presence of aerosols in the upper part of the layer and above it as well as the reduced sensitivity close to the surface.
The AMF is computed using radiative transfer calculations that require information on measurement conditions (such as, observation geometry and wavelength) and atmospheric characteristics (e.g., vertical distribution of the chemical species, surface albedo, aerosol loading and clouds).Hence, an appropriate selection of the a priori assumptions used is essential to obtain the correct values of the AMF and thus reduce the uncertainties of the NO 2 columns.Selecting an AMF that is too high will result in an underestimation of the VC.Likewise, the determined NO 2 VC will be too large if the value of the AMF used for the conversion of the SC is too small.
Satellite retrieval
The retrieval of tropospheric NO 2 columns from space is performed in several steps.First, NO 2 slant column densities (SC) are retrieved with the DOAS (Differential Optical Absorption Spectroscopy) technique (Platt and Stutz, 2008) in the UV/visible wavelength range.The SC corresponds to the amount of absorber present along the average light path through the atmosphere to the satellite sensor.The tropospheric SC is calculated by eliminating the stratospheric contribution from the total columns retrieved (further details regarding the retrieval can be found, for example in Leue et al. (2001), Richter and Burrows (2002), Martin et al. (2002) and Boersma et al. (2004)).As explained above, when dividing this tropospheric SC by an appropriate airmass factor a NO 2 tropospheric vertical column is obtained.
In current retrieval methods the presence of aerosols in the atmosphere is included in many different ways.Some retrieval processes do not explicitly correct for aerosol impact, arguing that the cloud correction scheme also accounts for a large part of the aerosol effect (Boersma et al., 2004(Boersma et al., , 2007)).Another approach is to use a static a priori profile for aerosol loading and type (Nüß, 2005;Richter et al., 2005).Finally, some retrievals include full aerosol treatment in the radiative transfer using aerosol fields from models (Martin et al., 2003;Lee et al., 2009).
Radiative transfer settings
Numerous factors need to be accounted for when considering the effect of aerosols on satellite NO 2 measurements.In the present study, these effects were analysed in detail by considering multiple scenarios where the aerosol vertical distribution was varied together with its load and optical properties.In addition, as it will be explained below, the surface albedo value and the NO 2 vertical profile were also varied.Airmass factors were calculated with the radiative transfer model SCIATRAN 2.2 (Rozanov et al., 2005).The calculations were performed on a vertical grid of 200 m from surface to the top of the trace gas and aerosol layers, at four different wavelengths (425, 437.5, 440 and 450 nm), in nadir observation, and at six solar zenith angles (SZA, from 20 • to 70 • in steps of 10 • ).SCIATRAN was operated using the discrete ordinate method for solution of the radiative transfer equation, in plane parallel geometry, accounting for full multiple scattering effects, but without including polarization.At the viewing geometry used here, the effects of Earth's curvature and refraction can be neglected.As atmospheric scenario, the US standard atmospheric pressure and temperature were used.The surface albedo was set to 0.03 assuming that this is, for the spectral range used, an average value for urban areas.Nevertheless, this value was also varied to 0.01, 0.07 and 0.1 so that the effect of the surface albedo on the calculations could be determined.
In a first step, simplified scenarios were investigated with, for example, NO 2 and aerosol vertical distribution as box profiles and considering purely scattering or partly absorbing aerosols.The results from these simulations provide insight into the direction and magnitude of the effects of different parameters on the satellite sensitivity.Furthermore, a second phase of the study included measured aerosol profiles from different locations and NO 2 profiles from model simulations for urban and rural conditions.In both cases, the interference exerted by aerosols can be analysed by comparing these scenarios with a reference scenario where no aerosol is considered in the radiative transfer calculations.In following sections the scenario assumptions will be described in more detail and these settings are also summarized in Tables 1 and 2.
NO 2 profile
For the initial analysis, the NO 2 profiles were considered to be homogenously distributed within a boundary layer of 0.6, 1.0 or 2.0 km height (box profiles).This homogenous distribution is, most probably, not often found close to the major anthropogenic sources, i.e., in urban locations.For that reason, in other case scenarios simulated, two different NO 2 profiles, from surface to 5.5 km, were used: average urban and rural.These profiles are an average from CHIMERE (Schmidt et al., 2001;Honoré et al., 2008) model runs with a 9×9 km 2 resolution, at 10 LT and for the period from 23 May to 11 June 2007 (randomly selected).The NO 2 volume mixing ratio above 5.5 km (the top of model simulations) decreases slowly to a value of 1.5×10 −6 ppm at 100 km.The model results are based on a simulation for Paris downtown, a location in the close vicinity of Paris at 15 km East and a rural region at 100 km East of Paris.Since the first two sites are very similar, and, considering that the typical size of a satellite pixel would include both of these measurements, their average is defined as average urban -"Avg Urb".Near surface NO 2 levels in this profile are close to the climatological average of urban background surface NO 2 within in the Paris town and its near suburbs of above 20 ppb (http://www.airparif.asso.fr/).While the model does not consider lightning explicitly, NO x (background) from this source are included via the boundary conditions for the domain.In Fig. 2 the different NO 2 mixing ratio profiles are shown from surface up to 10 km.As it can be seen in the profiles presented in Fig. 2, the NO 2 profile determined by the model for the urban conditions is not at all similar to a homogenous distribution over a boundary layer of 1.0 km or even 600 m.This has a significant impact on the results as it will be further discussed in Sect.3.3.In addition, the large difference between the urban and rural profiles illustrates how the NO 2 vertical distribution can show significant variations over a short distance.This might be a crucial point for satellite retrievals where, at the spatial resolution of current a priori data, both urban and rural scenes are often contained in one model grid cell (Heckel et al., 2010).
Aerosol settings
Currently, several datasets of aerosol characteristics are available from records of either ground-based or satellite instruments (e.g.MODIS, MISR, MERIS, CALIPSO, etc).Worldwide ground-based networks offer the possibility to obtain crucial information to better characterize aerosols and reduce the current uncertainties on the definition of aerosol optical properties.This is the case of, for instance, the Aerosol Robotic Network (AERONET, Holben et al., 1998), the European Aerosol Research Lidar Network (EARLINET, Mattis et al., 2002), or the Asian dust network (AD-net, Murayama et al., 2001).In the present study data from these networks was applied in the definition of the aerosol optical properties and its vertical distribution.The optical properties and size distributions, at 440 nm, were mainly taken from records of 12 worldwide AERONET stations presented in Dubovik et al. (2002).This dataset is representative for the usual classification of four different aerosol types that have distinctive physicochemical, optical and radiative properties: urban/industrial, biomass burning, desert dust and oceanic.The precision of the AERONET dataset is discussed in detail in several publications but this subject will not be explored here because the accuracy of these measurements and representativeness of the dataset is not central to the conclusions to be drawn.
Size distribution and phase function
Aerosols emitted in urban areas and from open vegetation fires are, on average, dominated by small particles (Seinfeld and Pandis, 1998).Yet, the size distribution of this aerosol, especially that of biomass burning cases, is not constant and varies in time and space.The dimension of the particles is mostly dependent on the type of fuel, the combustion phase and the age of smoke (e.g.Dubovik et al., 2002).This last factor can be important when considering fire plumes that are transported for some days away from the source.On the other hand, mostly coarse particles are found in desert dust scenes or oceanic aerosol (Seinfeld and Pandis, 1998).The selection of results presented here will follow these assumptions, i.e., not every simulated case will be shown but mostly those which are more representative of the aerosol type in consideration.In order to facilitate the interpretation of results, a mixing of fine and coarse aerosol, as it would happen in reality, was not simulated.
Within the radiative transfer model, angular distributions of scattered light are required to simulate the interaction of particles and light.The details of the angular distribution of the phase function vary with the aerosol composition and the size of the particles relative to the wavelength of the radiation and also depend on particle shape and internal structure.For each of the aerosol types, the phase functions (see Fig. 3) for both fine and coarse particles were considered.These were determined with a FORTRAN program developed by Michael Mishchenko and freely available at http://www.giss.nasa.gov/staff/mmishchenko/brf/.
Single scattering albedo
The scattering efficiency of aerosols strongly depends on their concentration, size and shape, as well as on their refractive index, determined by their chemical composition.Aerosol scattering also depends on the scattering angle and usually has a pronounced maximum for forward scattering.Likewise, the absorbing properties of aerosols are usually expressed via the single scattering albedo (SSA) which is defined as the ratio of scattering to extinction and depends on particle composition and wavelength.The SSA differs according to type and source of aerosol and, therefore, is in part dependent on the location of measurement (see, for example, Hu et al., 2007).For the majority of the scenarios considered in this analysis, the impact of aerosol absorption was simply investigated by comparing the AMF determined with runs where ω 0 was set to 0.93 (average from all the SSA values given at 440 nm in Dubovik et al., 2002) and others where ω 0 =1.0.This allowed determining the maximum effect in the results when reducing the absorbing ability of aerosol.Furthermore, when evaluating the impact of the SSA in the radiative transfer calculations, this parameter was also set to 0.8 and 0.95 (see Sect. 3.5).In the second stage of the analysis, when considering measurements of aerosol profiles, the ω 0 values required were either taken from the corresponding records or based on typical values available from other studies that focused specifically on each of the aerosol types.
Vertical distribution
For remote sensing applications, the total amount of aerosols present in the atmosphere is often specified by an aerosol optical depth (AOD) which is the vertical integral of the extinction by aerosols from the top of the atmosphere to the ground.In the first phase of the study, the aerosol vertical distribution was defined as a box shaped profile.These well defined layers with homogenously distributed aerosol had variable top height.Three cases were set with extinction coefficients representative for three aerosol loads: 0.1 (low pollution level), 0.5 (moderate pollution) and 0.9 (polluted scene) aerosol optical depths.Alternatively, the aerosol's vertical distribution was defined in different ways: following the NO 2 profile; starting at surface level and with the top of the layer lower or higher than that of the NO 2 profile; furthermore, discrete elevated aerosol layers above the NO 2 layer (assumed to be in the BL) were also taken into account.The scenarios A to H (Table 1) will probably not be realised for all types of aerosols.Normally, the urban aerosol is assumed to be either in homogenous layers extending from the surface to the top of BL or, often, following an exponential decrease with height.In general, one can assume that the majority of anthropogenic sources are the same for both NO 2 and aerosols and, therefore, they would have similar spatial distributions.However, depending on the source location and transport processes, the aerosol layer can extend to a higher altitude, whereas NO 2 will be in general more concentrated closer to the source region and at lower levels, due to a shorter life time.For that reason, the extension of each layer was also varied independently so that different scenarios could be Table 1.Scenarios considered for the SCIATRAN runs, defined by the combination of a NO 2 and an aerosol layer, as in box profiles (e.g., Scenario B: NO 2 layer 0-1.0 km and aerosol layer 0-0.6 km).
NO 2 layer (km) 0-0.6 0-1.0 0-2.0 Aerosol layer (km) 0-0.6 0-0.6 0-1.0 0.6-1.01.0-2.00-2.0 2.0-3.0 0-2.0 analysed.The scenarios with elevated discrete aerosol layers are mostly adequate to illustrate plumes of biomass burning smoke and desert dust that are transported several hundreds to thousand kilometres away from the source and which can be lifted to higher altitudes during transport.These events can happen not only on a continental scale (e.g.smoke from fires in the African savanna that is transported across the Atlantic Ocean) but also on a regional scale, like the transport within Europe.These aerosol plumes often occur in the free troposphere, but they can also be part of the boundary layer either by intrusion processes or due to the low initial injection height (Müller et al., 2003;Labonne et al., 2007).Good examples of this case are the dust outbreaks from deserts that often can mix with urban type aerosol emitted within European or Asian cities (e.g.Zhou et al., 2002).
In a second stage, the definition of the case studies was based on data available from different measurements.Like this, situations as those described above could be simulated.The size distribution and corresponding phase functions were maintained from the initial stage.Nevertheless, the extinction coefficients used for the profile definition were based on lidar measurements performed at numerous locations at different times of the year (see representation of profiles I to P in Fig. 4 and further details in Table 2).Still, the profiles considered in this study are not the exact representation of the original ones.Often adjustments were required in order to obtain a profile from surface to the top of atmosphere of 100 km.Moreover, as these are meant to be examples for case studies their accuracy is not a subject of this analysis and does not influence the conclusions drawn.Because lidar measurements (both satellite and ground-based) are usually performed at 355 nm and/or 532 nm an Ångström exponent ( Ångström, 1929) was necessary to convert these values to the corresponding ones at 440 nm (within the wavelength region where NO 2 is retrieved).These values were also taken from the referred literature.The oceanic aerosol type was not included at this stage because this aerosol is normally only observed in very low concentrations at polluted sites and is usually mixed with other types of aerosol.Therefore, for simplicity of the analysis, it is assumed that its influence in the NO 2 retrieval is similar to that of the other types considered.2.
Results
A comprehensive analysis was performed with airmass factors of NO 2 calculated for many different case studies where different settings of the model calculations were changed.Here only the results obtained for 440 nm are analysed as this is the wavelength for which the AERONET aerosol optical properties are given.Extension of the calculations to the wavelength range often used for NO 2 retrieval revealed an average increase of AMF by 10% from 425 to 450 nm.This variation is relatively small and will largely cancel if it is linear with wavelength but might be relevant in some cases.For different solar zenith angles (SZA), the general trend shows that the AMF increases for higher sun, but for specific cases, this tendency can also be reverted.In some circumstances (not presented here), when considering fine aerosol, a decrease occurs with high sun, and in other cases, with coarse particles, a small increase is then followed by decay after 50 • or 60 • .The variation of the size parameters (mean radius and its standard deviation within the fine or coarse mode categories) of the different aerosol types representative for the locations considered in this study was rather small.The similarity in values resulted in nearly identical phase functions with noticeable differences only between the two general size distributions considered: fine and coarse (see Fig. 3).As a result, the NO 2 AMF determined within the various scenarios with fine particles are very similar, and the same occurs for those with coarse aerosol.When comparing to the scenario without aerosol it was found that fine particles have a higher impact on intensifying the changes on the AMF than the coarse ones.However, this effect depends on many fac-tors such as the vertical distribution or sun position (e.g., very low sun can favour the enhancement of signal by the coarse particles standing in a discrete layer above the trace gas).As expected the values of AMF determined with non-absorbing aerosol were the highest.This decrease in the measurements' sensitivity for absorbing aerosol is a consequence of the reduction of available light when such aerosol is present in the atmosphere.It is also important to note that at 440 nm the Rayleigh optical thickness is 0.32.Consequently, in scenarios I and N, the molecular scattering dominates.
Surface Albedo
The surface albedo selected for all scenarios included in this sensitivity study was 0.03.NO 2 is not measured only over urban areas but also in remote locations where the surface albedo can vary according to the different type of soil and vegetation.Knowing that the surface reflectivity is influencing the sensitivity of the satellite measurements, the impact of changes in surface albedo was also analysed here.As mentioned above, this value was set to 0.01, 0.07 and 0.1 in a scenario of NO 2 and aerosol homogenously mixed in a 1.0 km layer (scenario C).
From the results presented in Fig. 5 it is possible to see that the impact of the surface albedo can be quite high in the AMF calculations.An increase of the surface albedo value results always in an increase of the AMF.Brighter surfaces will more efficiently reflect the sun light back to the satellite and therefore contribute to the enhancement of the measured NO 2 columns.A change from 0.01 to 0.1 can result in an increase of the AMF on average of 90% (for different AODs).The maximum value obtained is in fact much higher and changes by a factor of 2.8 are registered for the case of high sun (SZA of 20 • and 30 • ), with coarse particles mixed with the trace gas and AOD=0.1.The dependence of this change on the aerosol amount present in the atmosphere is illustrated in panel (b) of Fig. 5 where the AMFs are plotted as a function of surface albedo, for one SZA (50 • ) and different AODs.The impact of aerosols is largest over dark surfaces and rapidly decreases as albedo increases as the increase in reflectivity resulting from aerosols has less effect if the surface is already bright.However, the situation changes when considering absorbing aerosol (results not shown).In that case, a decrease of the AMF is observed when a dark layer of aerosol (mixed with the trace gas) stands above a bright surface, i.e., surface albedo of 0.1 and above.
Changes in boundary layer height
The height of the boundary layer differs for different locations and is also dependent on seasonal variations.To investigate the effect of BL height changes alone, a study was carried out varying this height to 0.6, 1.0 and 2.0 km (scenarios A, C and H, respectively), and maintaining both the NO 2 and aerosol homogenously distributed in this layer.In Fig. 6 the results obtained for urban fine and coarse aerosol with different aerosol optical thickness are presented.
Before discussing the impact of the boundary layer height with trace gas and aerosol mixed in the atmosphere it is important to mention that the variations of the boundary layer influence the AMF calculations even when considering only a layer of NO 2 without aerosol present.When the top of the NO 2 layer expands from 0.6 to 2.0 km the AMF will increase in average by a factor of 1.4.This is related to the fact that the sensitivity of the measurements is smaller close to the surface.For every case (from scenarios A, C and H) it was found that the smallest AMFs are determined for the conditions without aerosol (not shown).Thus, one can conclude that, in these scenarios, the presence of aerosol results in an increase of the sensitivity of the measurements, even if quite small for coarse particles and low aerosol load.In practice, this indicates that, for the cases exemplified here, if the effect of aerosol scattering is not accounted for in the retrieval, the NO 2 VC will be overestimated.Furthermore, comparing the results for different boundary layer heights, the results for NO 2 mixed with aerosol (Fig. 6) follow the same pattern as in the calculations performed only with the NO 2 layer.If a too low BL height is assumed in the retrieval, the tropospheric columns of NO 2 will be overestimated.However, the changes in the AMF are smaller in this case than for the scenarios without aerosol.The AMF increases, on average, by about 20% when the boundary layer top changes from 0.6 to 2.0 km.The largest effect was found for the simulation with coarse aerosol and optical thickness of 0.1 with 37% variation, at 440 nm.Interestingly, the effect seems to decrease with growing aerosol load.Such a variation is possibly a result of the increase in scattering (and therefore the effective albedo) which will improve the sensitivity the satellite measurements to the lower atmosphere.
The results presented in Fig. 6 also reveal, as mentioned above, the difference in behaviour between fine and coarse particles.The AMFs resulting from the simulations with fine aerosol mixed with the trace gas are higher, i.e., at the same AOD, fine aerosol increases the sensitivity to the NO 2 more than coarse particles, and this difference of results increases with AOD.This is most likely related to the less pronounced forward peak in scattering on fine particles (see phase function in Fig. 3) which increases the ratio of photons scattered towards the satellite under this observation geometry and therefore improves the sensitivity.
Box aerosol profiles
In the previous section, NO 2 and aerosol had the same vertical profiles representing a situation where both are well mixed.In the following scenarios, the vertical extension of the aerosol layer was varied to 0.6 and 2.0 km (scenarios B and F, respectively) while the NO 2 profile was kept constant.This was done for two NO 2 profiles, a simple 1.0 km box profile and the more realistic urban profile as modelled by CHIMERE. Figure 7 shows the results side by side for different AODs.
As it can be observed, in general, any aerosol mixed with the trace gas tends to enhance the NO 2 signal, indicating that an overestimation of the NO 2 VC will likely occur when effects caused by aerosol presence are neglected in the retrieval.However, the magnitude of the influence does vary as it depends on the relative position of trace gas and aerosol, in particular the aerosol load above the trace gas.In addition, the size of the particles also plays a role in the calculations.As for the previous scenarios discussed above, at the same AOD, fine particles have a larger influence on the airmass factors, due to the generally higher backscattering (see Fig. 3).In the simulations with box profiles, the interplay between reduction and enhancement of sensitivity can explain the observed variations: if the aerosol layer is to the surface, i.e., with its top at 600 m, below the top of the trace gas layer, the sensitivity will be enhanced due to higher reflectivity and multiple scattering.An increase of the AMF by 11% on average is found when the top of aerosol layer lowers from 1.0 km to the 600 m and, in the case of highly polluted scenes with AOD = 0.9, the difference between the values can be as high as 25%.Compared to the simulation without aerosol, the sensitivity can be enhanced by up to a factor of two.On the other hand, when the aerosol layer extends higher than the layer of NO 2 , the AMF is lower (by 5 to 45%) than in the case when both aerosol and NO 2 have the top layer at 1.0 km.This results from the elevated part of the layer of aerosol that acts as a shield and thereby partly cancels the enhancement of sensitivity in the lower part.Still, compared to the AMF values obtained without aerosol, the fine particles will slightly increase the NO 2 signal, with the exception of high solar zenith angles.In comparison, the coarse particles have smaller influence on the measurements.
The differences found in the AMFs calculated with AOD=0.1 and higher values indicate the importance of using the right AOD in the retrieval.An underestimation of the AOD will lead to an overestimation of the VC.The scenario F (aerosol layer extending to 2.0 km) is an exception to this statement as the AMF values do not vary much for different AODs.
In qualitative terms, the interpretation of the scenarios with the urban NO 2 profile is quite similar.However, for the latter, the AMF values are smaller as the NO 2 is more concentrated at the surface where the satellite sensitivity is the smallest.This is directly observed in the "no aerosol" case where the AMF decreases from 0.81 (at 440 nm and SZA=40 • ) to 0.70 (due to the shielding effect related to Rayleigh scattering).In addition, shielding effect of aerosols is also more pronounced for the NO 2 urban profile than for the 1.0 km box one, leading to an overall reduced effect of aerosols.In the case with an aerosol layer of coarse particles above the NO 2 , even a slight decrease in the AMF is observed.Thus, the importance of aerosols is reduced if a more realistic NO 2 profile is assumed.
Measured aerosol profiles
In addition to what was described above, the model NO 2 profiles were also combined with aerosol profiles derived from measurements in rural areas (scenario I) and urban environments (scenarios J and K).The results are shown in Fig. 8 for calculations assuming fine and coarse particles separately.Clearly, in these particular circumstances, the aerosol effect is much smaller than before, and very close to zero in the case of typical background profiles for both the NO 2 and aerosol.Independently of its detailed shape, the presence of an aerosol layer tends to cover the NO 2 layer below thereby decreasing the sensitivity of the measurements to trace gas amounts close to the surface.Depending on the sun position and the aerosol profile, small enhancements as well as reductions in sensitivity can occur.This emphasises the point that the sensitivity of the measurements does not only depend on the vertical distribution or total load of the aerosol but the combined effect of both aerosol and NO 2 distribution.For coarse particles all the AMFs were smaller than for the case without aerosol, indicating that the aerosol might be pre-Fig.8. NO 2 airmass factors for no aerosol (red) cases (rural -Rurand urban -"Avg urb" -NO 2 profiles from CHIMERE) and for the scenarios I (background -Rur -NO 2 and aerosol vertical profiles), J and K (urban -Urb -NO 2 and aerosol vertical profiles) calculated with the phase functions determined for coarse (CR) and fine (F) particles (optical properties taken from Creteil/Paris AERONET station).AMFs determined at 440 nm, with surface albedo = 0.03, ω 0 = 0.93 (I, K) and 0.87 (J), and AOD = 0.07 (I), 0.40 (J) and 0.62 (K) (see Table 2).
venting light from reaching down lower into the NO 2 layer close to the surface (or back from this layer to the satellite instrument).
Box aerosol profiles
The transport of dust and smoke plumes into European and certain Asian cities is not a rare event.These plumes are not only observed in the free troposphere but can, sporadically, also make a large contribution to the aerosol load measured in the boundary layer.Scenarios D, E and G (elevated aerosol layers from 0.6 to 1.0 km, from 1.0 to 2.0 km and 2.0 to 3.0 km, respectively) are simplified representations of such events with aerosol mostly being concentrated at higher altitudes.The results from these runs lead to the same conclusions as before, i.e., an aerosol layer standing above the trace gas obstructs the observation from space (see Fig. 9).A decrease of 6% to ∼70% is observed when comparing the AMFs obtained for the scenario without aerosol to that with aerosol distributed from 1.0 to 2.0 km.This reduction is higher for larger aerosol load, i.e., optical thickness of 0.9.If such plumes, standing in high altitudes, are not accounted for in the retrieval process, the tropospheric VCs are underestimated.The differences of the results for the layers 1.0 to 2.0 km and 2.0 to 3.0 km (not presented here) were not significant.This indicates that the height of the aerosol layer is not so relevant for the sensitivity of the measurements when there is no overlap of the trace gas and aerosol layers.Contrary to this, in the case of aerosol mixed with NO 2 at the top of the layer (from 0.6 to 1.0 km), it was possible to notice (Fig. 9) that the particles do not interfere much with the measurements of the trace gas (cancelling of albedo and shielding effect).In fact, a slight enhancement (∼10% maximum for 440 nm) of the columns is registered only when small particles are present.It should be however, that this is not the case for lower SSA (see next section).In the presence of highly absorbing aerosol, the shielding effect will be dominant and a decrease of the AMF is found.Therefore, the cancelling between the two effects caused by the aerosol verified for these circumstances is naturally related to the definition of the aerosol properties.
Furthermore it is important to refer that the effect of aerosol on measurements of NO 2 present within a biomass burning plume will be quite different than in the case of NO 2 located in the boundary layer as discussed here.
Measured aerosol profiles
In a more realistic scenario, aerosols are also present close to surface in urban areas.Therefore, profiles have been defined to include both the local plumes and those of long-range transport from biomass burning smoke or desert dust (e.g.scenarios L and P from Table 2).An example of these results is presented in Fig. 10 for desert dust layers and fire plumes measured over different cities across the globe.As it can be seen from these findings, the effect of the aerosol layers transported above polluted areas can be quite different.Once more, the reduction in the sensitivity of the measurements, when compared with the "no aerosol" case, can be negligible or as large as ∼62% (for scenario O).This pronounced reduction is caused by the combination of several factors: the large aerosol optical depth (AOD=1.05);its absorbing nature (ω 0 =0.92); and the small fraction of particles that are mixed with the trace gas.This distribution of aerosol is the main difference between scenario L and O.The aerosol close to the surface present in scenario L may contribute to the cancelling of the shielding effect and therefore explain the large discrepancy between the results of the scenarios.In the case of simulations M, N and P the AMFs are not so reduced mainly because of the lower aerosol loads.Fig. 10.NO 2 airmass factors for urban NO 2 profile from CHIMERE using no aerosol and also for scenarios L to P (measured aerosol profiles) calculated with the phase functions determined for desert dust (DD) coarse (CR) particles (optical properties taken from Saudi Arabia AERONET station) and for biomass burning (BB) fine (F) particles (optical properties taken from Amazonian Forest/Brazil AERONET station).AMFs determined at 440 nm, with surface albedo = 0.03, and ω 0 = 0.92 (L, O, P) and 0.93 (N) (in scenario M ω 0 varies in height from 0.80 to 0.95), and AOD = 1.05 (L, O), 0.66 (M), 0.16 (N) and 0.42 (P) (see Table 2).
For the desert dust cases, only coarse aerosol was considered in the radiative transfer calculations but both fine and coarse (not presented here) particles were used for the biomass burning situations.The difference in the AMF calculated with each of the aerosol types is in the order of 20-25% with the higher values obtained for the runs with fine aerosol.
Changes in single scattering albedo
As mentioned above, for all the scenarios including box profiles, the AMFs were calculated both for a single scattering albedo (SSA) of 0.93 and 1 (not presented here).However, the SSA varies in time and space.Thus, the effect of deviations in the SSA in the radiative transfer calculations was also tested by changing this parameter to 0.80 and 0.95.These results are presented for both the simulations performed with the box profiles in scenario C (Fig. 11) and scenarios J and O (Fig. 12), where the NO 2 modelled profiles and measured aerosol vertical distribution were considered (see Tables 1 and 2 for more details on scenarios definitions).As expected, the SSA can have a great impact on the calculation of the AMF.An increase in the absorbing properties of the aerosol (SSA decreases from 0.95 to 0.80) results in a general decrease of the AMF.While, for low aerosol load (in the scenarios with box profiles) this variation of SSA values results in a difference of the AMF on the order of 5-10%, in the a more polluted atmosphere with AOD=0.9, the effect of SSA on the AMF can be as high as 77%.Still, the variation of the AMF values is not only dependent on the aerosol amount but also on the profiles considered.The variation of the AMF caused by the changes in the SSA values was found to vary for different aerosol vertical distributions.
Effect of clouds
The sensitivity analysis described here does not take into account the influence and interference of clouds on the measurements.In real data however, most measurements are affected by clouds in the satellite field of view, at least to some extent.Therefore, cloud correction algorithms are applied in the satellite retrievals to account for the effect of clouds, usually by assuming optically thick clouds and computing cloud fraction from reflectance and cloud top height from absorption of O 2 , O 4 or the amount of Raman scattering (e.g.Joiner and Bhartia, 1995;Koelemeijer et al., 2001;Acarreta et al., 2002;Kokhanovsky et al., 2003).As the radiative effects of clouds and aerosols have large similarities, it was suggested by Boersma et al. (2004) that cloud correcting algorithms also account, even if only partly, for aerosol effects.
The presence of non-absorbing particles will increase the retrieved cloud fraction.This type of aerosol is rather comparable to thin clouds.Therefore, as a result of this similarity, if no other form of cloud is present in the field of view, the cloud correction algorithms will perform as they do on thin clouds and in fact correct for part of the aerosol effect.However, if parts of the satellite pixel are also covered by meteorological clouds, the situation changes.While the retrieved cloud fraction will again increase in the presence of aerosol, the cloud top altitude will be close to that representative for the much brighter cloud and not for that of the aerosol layer.Therefore, if the cloud is higher than the aerosol layer (which will often be the case), the cloud correction algorithm will over-compensate the shielding effect of the cloud while neglecting the enhancing effect of the low aerosol layer.As a result, cloud correction algorithms cannot compensate aerosol effects in these situations and will lead to an overestimation of any NO 2 below the cloud.
On the other hand, if the aerosol is absorbing, the retrieved effective cloud fraction will be too small, as the reflectance is smaller than that of a non-absorbing cloud.Also, as discussed above, the airmass factors are reduced for absorbing aerosols and this cannot be accounted for by assuming a non-absorbing cloud.An additional and more subtle difference between clouds and aerosols might be introduced by the phase function which, as mentioned before, depends on the composition and size distribution of the aerosol.This might lead to different top of atmosphere reflectance for a layer having the same optical thickness of scattering aerosol or cloud particles.
In summary, cloud retrievals can be expected to compensate aerosol effects under some conditions, but may well enhance them in other situations.As the results depend on the details of the cloud correction algorithm used, this should be investigated for each of the products in use separately.
Conclusions
Aerosols can have a significant impact on the retrieval of tropospheric trace gases using UV/visible nadir measurements from space.In order to identify and quantify this impact, the effects of different aerosol parameters were investigated using both idealised and realistic scenarios.Overall, a large variability in the results was observed with examples of both increasing and decreasing sensitivity.The most important factors for the satellite sensitivity are not only related to aerosol assumptions, but also with the definition of surface albedo.For the latter, on average, changes of 90% of the AMF values were registered when the surface albedo was increased from 0.01 to 0.1.This illustrates how important it is to have accurate knowledge of the surface properties.Regarding the aerosol definitions, the key factors in the determination of NO 2 columns were identified as the relative vertical distribution of aerosol and NO 2 , the AOD and the SSA.In addition, differences in the airmass factors were found when applying either coarse or fine aerosol size distribution.However, no large differences were evident when considering small variations of those main types.
Variations of the vertical extension of a well mixed boundary layer of 1.0 km containing both NO 2 and aerosols can result in large differences (max.26%) of the airmass factors calculated, especially when the aerosol load is low and in low sun conditions.However, even larger effects (up to 55%) are found in the case without aerosols.The boundary layer height has strong seasonal, daily, and diurnal variations and not accounting for these changes will contribute to the inaccuracy of the calculated columns.The determined AMFs indicate that, if the boundary layer height is underestimated in the a priori assumptions, the tropospheric NO 2 column will be overestimated (and vice-versa).
Aerosol mixed with the trace gas, even if not on the full extension of the layer, will, by means of increased effective albedo and multiple scattering, enhance the NO 2 signal.In contrast, any aerosol layer that lies above the trace gas will act as a shield, decreasing the sensitivity of the measurements.If an elevated aerosol layer is not accounted for, the computed NO 2 columns will be too small, and this underestimation can be quite high.It is important to mention that these findings hold only for the SSA considered here (0.93), and that a dominant shielding effect is found in the event of highly absorbing aerosol mixed with the NO 2 .In any case, the magnitude of these effects will be determined by the relative vertical distribution of aerosol and NO 2 .A balance between enhancement and reduction of the signal will occur when the aerosol is both mixed with and above the NO 2 layer as might often be the case.As two examples, the AMF for a 1.0 km layer of NO 2 increases by a factor of 2 when mixed with a 600 m high from surface (Scenario B) aerosol layer of AOD 0.9, while for the case with an aerosol layer of same optical thickness between 2.0 and 3.0 km (Scenario G) the AMF is reduced by ∼78%.
The absorption properties of the particles also play an important role in the retrieval of the trace gas.The largest airmass factors were always obtained for the purely scattering aerosol (ω 0 = 1.0).A decreasing SSA reduces the measurement sensitivity.For highly polluted scenes (AOD > 0.9) the airmass factor can increase by a factor of 1.5 and more when the single scattering albedo is modified from 0.80 to 0.95.
When more realistic vertical profiles were applied for both NO 2 and aerosols, a much smaller effect of aerosol was observed.Large decreases of the sensitivity of the measurements were found only for aerosol layers that are elevated or expand to higher altitudes in the atmosphere.These situa-tions usually correspond to cases of biomass burning events or desert dust storms.For urban scenes, the changes in the airmass factors were rather small.This indicates that in these circumstances, the uncertainties introduced by neglecting the aerosol impact in the retrieval are moderate, i.e., the AMFs vary only by ∼7% on average.Situations of highly polluted scenes, as those of megacities, were not fully represented here (AOD, e.g., can be much higher than 0.9).Thus, in order to allow a better understanding of the aerosol influence in the measured NO 2 columns, in these circumstances, further analysis is still required.
In the present study, only clear sky cases have been considered.For partially cloudy scenes the results would differ in particular if the data are corrected for cloud effects.The presence of aerosols will also impact on the retrieved cloud properties which in part can compensate the aerosol effects in the absence of real clouds.The details of the interplay between aerosol effects and cloud correction algorithms are complex and should be investigated in more detail.
The continuing use of fossil fuels and biomass burning in a changing climate will result in changes in the amounts and distribution of NO x which is one of the key precursors for tropospheric ozone.To accurately assess these changes and to efficiently allocate efforts to mitigate pollution, precise knowledge of the global tropospheric column of NO 2 is essential.This study shows that to improve our current knowledge of the global distributions of tropospheric NO 2 and its evolution, improved knowledge of the aerosol properties are required.These include the vertical profile, AOD, size distribution and also the scattering/absorption properties of the particles.Simultaneous measurements of trace gas and aerosol properties from space would be the ideal answer to solve this issue.Some instruments have the potential to retrieve both required quantities.But while this is not done, a synergistic approach can be the alternative by combining data from two instruments, e.g., using AOD from MERIS in the retrieval of NO 2 from SCIAMACHY (both instruments flying on the ENVISAT platform).Another promising approach is the extension of what was done in this study: a combination of satellite (e.g.MODIS, MISR, CALIPSO) and groundbased measurements (e.g., from AERONET and EARLINET networks) with model predictions, when those are available in a suitable resolution.Furthermore, not only aerosol data is required.As it was demonstrated in this analysis, the relative vertical distribution of NO 2 and aerosols has a large impact on the calculations.The exact shape of the NO 2 profile in different locations is still rather unknown.Very recently, data from ground-based measurements as, for example, those of MAX-DOAS instruments (Wagner et al., 2004) show potential to provide simultaneous measurements of trace gas and aerosol profiles in the lower troposphere.Such measurements, together with model results could be used for improved a priori data sets in the near future.Static climatological assumptions that are often used can be replaced by more up to date data that is more suitable to describe the measurements conditions.In this way, spatial and temporal variability can be accounted for, improving the retrieval algorithm for tropospheric NO 2 columns.
Fig. 2 .
Fig. 2. NO 2 profiles from surface to 10.0 km used in the SCIA-TRAN settings for the airmass factor calculations: box profile of 1.0 km (red); average urban ("Avg Urb", green) and rural (blue) based on CHIMERE model results.
Fig. 4 .
Fig. 4. Aerosol extinction profiles from surface level to 10.0 km used in the SCIATRAN settings for the airmass factor calculations for: (a) rural (Rur) and urban (Urb) locations; and (b) desert dust (DD) events and biomass burning (BB) plumes.These profiles are based on measurements performed at different locations as it is explained in Table2.
Fig. 7 .
Fig. 7. (a)-(c) NO 2 airmass factors for a 1.0 km box NO 2 profile using no aerosol (red) and for the scenarios B, C and F (extension of aerosol layer (AL) from surface to 0.6, 1.0 and 2.0 km, respectively) calculated with the phase functions determined for coarse (CR) and fine (F) particles (optical properties taken from Creteil/Paris AERONET station).AMFs determined at 440 nm, with surface albedo = 0.03, ω 0 = 0.93 and different AODs: 0.1, 0.5 and 0.9.(e)-(g) Same as (a)-(c) for the aerosol settings but using the average of modelled urban NO 2 profile ("Avg Urb").AMFs results in panel (d) and (h) are presented for SZA = 50 • .
Fig. 9 .
Fig. 9. NO 2 airmass factors for a 1.0 km box NO 2 profile using no aerosol (red) and also for scenarios D (a), (b) and E (c), (d) (elevated aerosol layers (AL) from 0.6 to 1.0 km and 1.0 to 2.0 km, respectively) calculated with the phase functions determined for (a), (c) coarse (CR) and (b), (d) fine (F) particles (optical properties taken from Amazonian Forest/Brazil and from Saudi Arabia AERONET stations, respectively for the biomass burning (BB) and desert dust (DD) cases).AMF determined at 440 nm, with surface albedo = 0.03, ω 0 = 0.93 and different AODs: 0.1, 0.5 and 0.9.
Fig. 12 .
Fig. 12. NO 2 airmass factors for different single scattering albedo (SSA, ω 0 = 0.80, 0.95 and 1.00) for scenario (a) J and (b) O (urban NO 2 profile from CHIMERE with urban (Urb) and biomass burning (BB) aerosol, respectively) calculated with the phase functions determined for coarse (CR) and fine (F) particles (optical properties taken from Creteil/Paris and Amazonian Forest/Brazil AERONET stations for scenario J and O, respectively).AMFs determined at 440 nm, with surface albedo = 0.03, and the AOD = 0.40 (J) and 1.05 (O).
Table 2 .
Aerosol parameters (single scattering albedo (ω 0 ), Ångström exponent (α) and aerosol optical depth, AOD) that were used to define the aerosol vertical profile (with extinction coefficients) for the SCIATRAN scenarios -taken from each of the references mentioned.These are representative of different aerosol types: Urban (Urb), Desert Dust (DD), and biomass burning (BB) scenes. | 14,127.4 | 2009-12-11T00:00:00.000 | [
"Environmental Science",
"Physics"
] |
Single-shot measurement of longitudinal phase space beam profile in an electron storage ring
A novel scheme to measure the longitudinal emittance and phase space profile in an electron storage ring by using correlations between time and the vertical coordinate, and between energy and the horizontal coordinate, is proposed.
Introduction
Transverse bunch crabbing using a two-frequency crab cavity scheme (Zholents, 2015;Huang et al., 2019) provides the optimal solution to produce short-pulse (1-10 ps full width at half-maximum) X-rays in a storage ring. This scheme would enable increased precision of timing-mode studies of a large number of dynamic processes in materials as they function. A crab cavity can also be used as an injection kicker in a new onaxis injection scheme (Kim et al., 2019) that uses a transverse deflecting radio-frequency (RF) cavity to kick the incoming beam into an already populated bucket but with a timing offset from the synchronous phase.
A crab cavity can also be used to measure bunch length (Loew & Altenmueller, 1965;Emma et al., 2000). A crab cavity couples the y-z or x-z planes, so bunch length is projected in the y or x dimension, respectively. The resolution of this bunch length measurement is limited by transverse emittances " x and " y unless a specialized beam transport line such as a chicane (Xiang & Ding, 2010) is used. Regardless of using a crab cavity, the distribution of the energy deviation = dp=p of a bunch can be projected in the x dimension at which horizontal dispersion x is large. The resolution of this is limited by the ratio " x / x . Combining these two principles enables measurement of the longitudinal profiles (z-) of a bunch in the x-y plane.
A fourth-generation storage ring (4GSR) is an accelerator that provides small emittance so that the measurement can be made with high resolution. A 4GSR adopts a multi-bend achromat (MBA), that effectively suppresses natural emittance. The MAX IV 4GSR is currently operating; others, including APS-U, SPring-8-II, SLS-II, ALS-U, SIRIUS, ESRF-EBS and Korea-4GSR, are being designed, constructed or commissioned [Streun, 2017;Steier et al., 2016;Liu et al., 2013; see also design reports for MAX IV (https://www. maxiv.lu.se/acceleratorsbeamlines/accelerators/accelerator documentation/max-iv-ddr), APS Upgrade (https://www. aps.anl.gov/APSUpgrade), SPring-8-II (http://rsc.riken.jp/ eng/index.html) and ESRF-EBS (https://indico.psi.ch/event/ 5589/)]. All of these rings use the MBA concept. A single cell of an MBA lattice has M-1 dispersion maxima with similar amplitudes, and one of these maxima can be used as a watching point for the measurement using a crab cavity. A hybrid MBA lattice such as APS-U, ESRF-EBS or Korea-4GSR provides two large dispersion bumps at the edge of a cell to enable effective correction of chromaticity, and the region of the large dispersion bumps is a desirable choice as a watching point.
In this study, we show that in a storage ring that uses a hybrid MBA lattice the measurement on longitudinal bunch profiles gives good resolution without the need for other additional magnets. In Section 2, we recall the matrix formalism between two arbitrary points of a storage ring, and projection of longitudinal (z-) beam profiles on the transverse (x-y) plane by using a crab cavity. In Section 3, we briefly introduce PAL-4GSR, which is used as an example lattice for the novel measurement scheme. We also calculated the intrinsic resolution of the measurement in PAL-4GSR. In Section 4, we present a numerical simulation of the measurement of the longitudinal profile by using the PAL-4GSR lattice. We also examine the resolution of the measurement when wakefield data are included.
Coupling of the y-z plane
Projection of the z-beam profile into the x-y plane can be described by linear optics theory. A thin crab cavity of TM110 mode has a linearized function as follows (Huang, 2016), where = eVk=E 0 , in which e is the electron charge, V [V] is the maximum voltage of a crab cavity, k is the angular wavenumber and E 0 [eV] is the nominal energy of a beam. In a storage ring, a linear matrix from arbitrary position 1 to arbitrary position 2 is given as (Chao, 2002) Definitions of each matrix element can be found in Appendix A. When initial beam coordinates at the position 1 are given as X 1 = (x 1 , x 0 1 , y 1 , y 0 1 , z 1 , 1 ), its continuous mapping via T c and M 12 becomes Let us assume that position 1 is located at one of the achromats of a storage ring ( 1 = 0), and position 2 is at a point that has large dispersion. If x ' (2n + 1) and y ' ½ð2n þ 1Þ=2 , contributions from R 12 and R 33 are negligible. The magnitude of R 11 is approximately ð x2 = x1 Þ 1=2 when x ' (2n + 1), and it is conservatively less than 3 as the ratio of x2 over x1 is less than 10 for most lattices. Specifically, when emittance ' 100 pm, 1 ' 0.001 and 2 ' 0.10 m, the contribution from 2 1 at x 2 is one order larger than that from x 1 (i.e. x 1 ' 10 À5 m and 2 1 ' 10 À4 ), which means we can expect projection between x 2 and 2 1 . Likewise, y 2 is dominated by R 34 z 1 as y 0 1 is of the order of 10 À6 but z 1 is of the order of 10 À3 and has order 10 À2 . Clearly, smaller emittance will lead to better resolution.
Hence, x and y coordinates at position 2 are expressed as which shows the projection of the z-plane onto the x-y plane. Its resolution is mainly dependent on whether the contributions from x 1 , x 0 1 , y 1 and y 0 1 are negligible and whether x ' (2n + 1) and y ' ½ð2n þ 1Þ=2 . If they are, then R 34 has sin( y ) dependence, so large 2 leads to increased magnification for the x plane, and y close to ½ð2n þ 1Þ=2 leads to increased magnification for the y plane.
A hybrid MBA lattice such as ESRF-EBS, APS-U, PAL-4GSR satisfies the above-described conditions. Due to the common ( x , y ) = (3, ) phase advance between two dispersion bumps in a cell of a hybrid MBA lattice, each cell has a similar phase advance. Specifically, a hybrid MBA lattice has ' x ' 0.9 and ' y ' 0.4 from the center of the long straight section to the nearest dispersion bump. Hence, choosing the position of a crab cavity satisfying ' y = =2 is accompanied by ' x ' .
PAL-4GSR and the intrinsic resolution
The PAL-4GSR storage ring is a hybrid seven-bend achromat (H7BA) lattice with a horizontal emittance of 90 pm. The research papers ring has a circumference of 570 m, and is composed of 20 symmetrical cells. From experience on PLS-II (Shin et al., 2013), the length of the straight section is considered to be 6.5 m to accommodate two SCRF modules in one straight section. The PAL-4GSR lattice (Table 1, Fig. 1) contains a 2 T super-bend in the central dipole to produce radiation with a critical energy of 12 keV.
The concepts of the ESRF-EBS and APS-U lattices were adopted in the PAL-4GSR lattice. The dispersion was deliberately enlarged between the first and second dipoles and between the sixth and seventh dipoles, then three chromatic sextupoles were located in this dispersion bump region to reduce the strength required to control the chromaticity. The betatron phase advances between the two dispersion bumps were set to Á' x ' 3 in the horizontal plane and Á' y ' in the vertical plane; as a result, nonchromatic effects of the sextupoles are canceled out naturally. To minimize natural emittance, five-step longitudinal gradient dipoles and reverse bending magnets were considered (Streun & Wrulich, 2015;Nagaoka & Wrulich, 2007;Delahaye & Potier, 1989;Streun, 2014).
We performed tracking simulation to examine the resolution of the measurement using a transverse deflecting cavity (TDC) on PAL-4GSR. For tracking simulation, we used elegant software (Borland, 2000). A total of 100 000 particles were generated at the position of the TDC (Fig. 1). They have a 58 pm (standard deviation) Gaussian distribution with matched Twiss functions in the horizontal and vertical planes, and ten lines of uniform density with maximum values of AE 0.3% and AE 24 mm in the longitudinal plane (Fig. 2). A watching point located at the left dispersion bump has a vertical phase advance of =2 from the position of the TDC. For a TDC, we set a voltage of 4.5 MV and a frequency of 750 MHz, which yield = 0.02358. When the TDC was off, we observed five resolved distributions on the x axis at the watching point due to the correlation Figure 1 Twiss functions of PAL-4GSR (two cells) and the position of the transverse deflecting cavity and watching point.
Figure 2
Longitudinal profiles of 100 000 particles at the position of the crab cavity for examination of intrinsic resolution. Particles are distributed with a delta function in longitudinal phase space.
Figure 3
Beam distribution at the watching point with 100 000 particles prepared for examination of intrinsic resolution. (a) TDC off; (b) TDC on. 0.037 mm on the x axis and 0.022 mm on the y axis, which are explained by x 2 ' R 11 x 1 + R 12 x 0 1 and y 2 ' R 33 y 1 + R 34 y 0 1 . When the TDC was on, the x axis was not affected, but five resolved distributions appeared on the y axis [ Fig. 3(b)]. The effective length on the y axis increased from 0.13 mm to 5.95 mm due to y 2 ' R 34 z 1 whereas the standard deviations of the Gaussian peaks did not change. These processes allow clear resolution of the five Gaussian peaks [ Fig. 4(c)]. From the revealed Gaussian peaks with the use of a line distribution on the longitudinal plane, we define the intrinsic resolution on an axis as R I = (distance between adjacent Gaussian peak) / (standard deviation of a Gaussian peak). R I increases with increase in the clarity of distinction of the Gaussian peak on an axis. PAL-4GSR has R I = 6.69 on the x axis and R I = 64.68 on the y axis. The dispersion bump (for -x correlation) increases and the voltage of the TDC increases (for z-y correlation), so the resolution can be increased. Adjustment of dispersion bumps is constrained by lattice requirements, but the voltage of the TDC can be increased further without affecting lattice requirements and we can expect much better resolution than is currently achieved.
Application on PAL-4GSR
A Gaussian bunch of 100 000 particles in an equilibrium state was generated for the six-dimensional phase space with the round-beam mode. It had 58 pm emittance in both horizontal and vertical phase spaces. It should be mentioned that, for simplicity, distributed coupling errors are considered along the ring except for the region from the TDC and watching point to generate round beam. With a main RF voltage of 2.15 MV, its longitudinal distribution had = 0.00108 and z = 7.677 mm ( Table 2). For a TDC, a voltage of 4.5 MV and a frequency of 750 MHz were also used.
Longitudinal emittance measurement
The longitudinal emittance can be written with the standard deviation as Here, using equation (4), the standard deviations z and can be approximately converted to and respectively. Therefore, measurements of x and y directly lead to estimates of longitudinal emittance. The generated Gaussian bunch is deflected at the TDC. After a =2 phase advance (at a watching point), the bunch's projection on the x-y plane has a clear Gaussian distribution [Fig. 5(b)], with x = 0.197 mm and y = 0.922 mm. From these values, reprojection using equation (4) gives = 1.096 Â 10 À3 and z = 7.680 mm, which have 1.46% and 0.04%, respectively, error relative to the original values. As a result, the longitudinal emittance from the measurement is calculated to be " z = 8.417 Â 10 À6 m, which is a 1.52% error relative to the original value. The original longitudinal profile at the TDC and the profile (Fig. 6) were obtained from re-projection using equation (4).
Measurement of the longitudinal phase space profile with instability
Beam manipulation in the longitudinal phase can be used to generate coherent radiation (Ratner & Chao, 2010) and beam dynamics in the longitudinal phase space to enable exploration of a new injection scheme (Kim et al., 2019;Aiba et al., 2015;Jiang et al., 2016;Jiang & Xu, 2018;Kuske et al., 2020). These goals invoke a need to measure the longitudinal beam profile. To generate a special shape in the longitudinal phase space, instabilities were deliberately induced by using wakefield data -here the impedance data of APS (Chae, 2003(Chae, , 2007Lindberg & Blednykh, 2015). However, PAL-4GSR and APS have different characteristics such as circumference (PAL- Figure 4 Density distribution on the x and y axis for the beam distribution shown in Fig. 3. (a) Density on the x axis when the TDC is on or off, (b) density on the y axis when the TDC is off, and (c) density on the y axis when the TDC is on. 4GSR: 570 m; APS: 1100 m), nominal energy (PAL-4GSR: 3 GeV; APS: 7 GeV), so use of these data can invoke instabilities in PAL-4GSR. We first estimated the current threshold I threshold of the strong instability by following Boussard's criterion (Boussard, 1975), where c is the momentum compaction, z is the equilibrium bunch length, is the equilibrium relative energy spread, c is the speed of light, T 0 is the revolution time, and Z k =n is the effective longitudinal impedance. This conservative estimate gives I threshold ' 0.37 mA; therefore we used a single bunch current of 7.7 mA (or single bunch charge of 15 nC) which is sufficiently high above the estimated threshold. Including the impedance data, a tracking simulation was conducted using a Gaussian bunch of 100 000 particles in the equilibrium state.
Oscillations of r.m.s. bunch length and r.m.s. energy spread along a number of turns show that both quantities increased rapidly after tracking started, reached their maximum within 2000 turns and eventually converged to a new equilibrium at $ 10000 turns (Fig. 7). The new equilibrium with the impedance data has 1.3 times larger bunch length and 1.8 times larger energy spread, compared with the original equilibrium. Those results imply that strong instability was driven well. We next examined longitudinal beam profiles at 10 000 turns (Fig. 8). After the kick of the TDC was applied, the longitudinal phase space was projected to the x-y plane. The central beam distribution split up and the outer beam distribution seems to have a spiral shape. Also, the center of the bunch was moved to $ 0.3 mm on the x axis. The overlap of the original longitudinal profile [ Fig. 8(a)] and re-projection from the x-y plane [ Fig. 8(b)] using equation (4) agreed well (Fig. 9). Oscillation of (a) r.m.s. bunch length and (b) r.m.s. energy spread versus turns when impedance data are included. Comparison of the original longitudinal profile at the crab cavity position (blue dots) and re-projection from equation (5) (red dots). No wakefield included.
Conclusion
We have described a novel scheme to measure the emittance and phase space profile in the longitudinal plane by using correlations between time and the vertical coordinate, and between energy and the horizontal coordinate. A large dispersion bump has a strong correlation with energy and the horizontal coordinate, and the crab cavity has a strong correlation with time and the vertical coordinate. As a result, longitudinal emittance was estimated with < 1.52% error in the PAL-4GSR lattice and micro-bunching instability was observed at the synchrotron radiation source point. This longitudinal profile measurement scheme will help to guide the manipulation of beams in longitudinal phase space.
Figure 9
Comparison of the original longitudinal profile at the crab cavity position (blue dots) and re-projection of the x-y profile at the monitor position using equation (4) (red dots). Wakefield is included for a single bunch of 15 nC. | 3,815 | 2022-01-01T00:00:00.000 | [
"Physics",
"Engineering"
] |