diff --git a/samples/texts/102047/page_1.md b/samples/texts/102047/page_1.md new file mode 100644 index 0000000000000000000000000000000000000000..9b8c7f6b9cae5c0d4aa19d420fcc475494b04a46 --- /dev/null +++ b/samples/texts/102047/page_1.md @@ -0,0 +1,12 @@ +ON TWO POLYNOMIAL SPACES +ASSOCIATED WITH A BOX SPLINE + +CARL DE BOOR, NIRA DYN, AND AMOS RON + +The polynomial space $\mathcal{H}$ spanned by the integer translates of a box spline $M$ admits a well-known characterization as the joint kernel of a set of homogeneous differential operators with constant coefficients. The dual space $\mathcal{H}^*$ has a convenient representation by a polynomial space $\mathcal{P}$, explicitly known, which plays an important role in box spline theory as well as in multivariate polynomial interpolation. + +In this paper we characterize the dual space $\mathcal{P}$ as the joint kernel of simple differential operators, each one a power of a directional derivative. Various applications of this result to multivariate polynomial interpolation, multivariate splines and duality between polynomial and exponential spaces are discussed. + +**1. Introduction.** The space $H(\phi)$ of all exponentials in the linear span $S(\phi)$ of the integer translates of a compactly supported distribution $\phi$ is of basic importance in multivariate spline theory since, in principle, it allows the construction of good approximation maps to $S(\phi)$ from spaces containing $S(\phi)$. Generically, $H(\phi)$ is $D$-invariant (i.e., closed under differentiation), hence is the joint kernel for a set $\{p(D): p \in I_{H(\phi)}\}$ of differential operators with constant coefficients, with $I_{H(\phi)}$ a polynomial ideal of finite codimension (in the space $\Pi$ of all multivariate polynomials, i.e., an ideal of transcendental dimension 0, hence with finite variety). An understanding of the interplay between the space $H(\phi)$ and its associated ideal $I_{H(\phi)}$ is useful in the determination of the basic properties of $H(\phi)$ such as its spectrum, its dimension, and its local approximation order. + +For the important special case when $\phi$ is a polynomial box spline (and $\mathcal{H} := H(\phi)$ is thus a polynomial space), an explicit set of generators for the ideal $I_\mathcal{H}$ is known [BH1], but nevertheless, the construction of their joint kernel was found to be very difficult. At the same time, a polynomial space $\mathcal{P}$ (of very simple structure) is known which serves as a natural dual for $\mathcal{H}$ and is of substantial use in the analysis of $\mathcal{H}$. Specifically, the duality between $\mathcal{H}$ and $\mathcal{P}$ has been used in [DR1] in the investigation of the local approximation order of some exponential spaces, in [DR1,2] in the solution of an \ No newline at end of file diff --git a/samples/texts/102047/page_10.md b/samples/texts/102047/page_10.md new file mode 100644 index 0000000000000000000000000000000000000000..ebd4cb76616c01547a5a050a156ce405aad8f08a --- /dev/null +++ b/samples/texts/102047/page_10.md @@ -0,0 +1,34 @@ +On the other hand, by (4.4) and (4.6), + +$$ \dim(\Pi_j \cap \mathcal{P}(X)) = \dim(\Pi_j \cap \mathcal{H}(X)), \quad \forall j, $$ + +and (5.3) now follows from the fact that $\Pi(M_X) = \mathcal{H}(X)$. $\square$ + +We note that the result is no longer valid if we drop the unimodularity assumption (cf. [R; Ex. 4.1]). + +**6. A remark on (2.11) Lemma.** A careful examination of the proof of (2.11) Lemma shows that the details of the connection between the ideal $I^X$ and its kernel $I^X \perp$ enter into the argument in only a minor way. The only facts used are: (i) the map $\Pi \to L(\Pi): p \mapsto p(D)$ is a ring homomorphism; and (ii) for any basis $B$ in $X$, $\dim I^B \perp = 1$. +This suggests the following result. + +(6.1) PROPOSITION. Let $M: \Pi \to L(V)$ be a ring-homomorphism into the ring of linear maps on the linear space $V$. For a given multiset $X$ of directions, define + +$$ I_M^X \perp := \bigcap_{p \in I^X} \ker M(p). $$ + +Then + +$$ \dim I_M^X \perp \leq \sum_{B \in \mathfrak{B}(X)} \dim I_B^X \perp. $$ + +The proof of the proposition follows entirely that of (2.11) Lemma, with the obvious modifications whenever the induction hypothesis is applied. + +It would be nice to identify other settings rather than the one utilized in this paper, where the above proposition is of use. + +REFERENCES + +[BH1] C. de Boor and K. Höllig, *B-splines from parallelepipeds*, J. d'Anal. Math., **42** (1982/3), 99-115. + +[BH2] __________, *Bivariate box splines and smooth pp functions on a three-direction mesh*, J. Comput. Applied Math., **9** (1983), 13-28. + +[BR1] C. de Boor and A. Ron, *On multivariate polynomial interpolation, Constructive Approx.*, **6** (1990), 287-302. + +[BR2] __________, *On polynomial ideals of finite codimension with application to box spline theory*, J. Math. Anal. Appl., to appear. + +[DM1] W. Dahmen and C. A. Micchelli, *Translates of multivariate splines*, Linear Algebra and Appl., **52/3** (1983), 217-234. \ No newline at end of file diff --git a/samples/texts/102047/page_12.md b/samples/texts/102047/page_12.md new file mode 100644 index 0000000000000000000000000000000000000000..8414215172b0a504e0f028b408820c0905fb8de4 --- /dev/null +++ b/samples/texts/102047/page_12.md @@ -0,0 +1,40 @@ +interpolation problem induced by $\mathcal{H}$, in [J] in the construction of +linear projectors onto a box spline space, and in [DR1] in the com- +putation of the homogeneous degrees of $\mathcal{H}$ (which is equivalent to +computing the Hilbert function of $I_{\mathcal{H}}$). See also [DM3, § 3]. + +It is the purpose of this paper to establish the surprising result that +$\mathcal{P}$, too, is the joint kernel of a rather simple set of constant coefficient +differential operators, each being just a power of a directional deriva- +tive. This allows us to characterize $\mathcal{P}$ in terms of the degrees of its +polynomials when restricted to certain linear manifolds. Various ap- +plications of this result to multivariate polynomial interpolation, box +spline theory, and duality between polynomial and exponential spaces +are discussed as well. + +In §2, after defining the space $\mathcal{P}$ and its associated differential op- +erators, we prove that $\mathcal{P}$ is indeed the joint kernel of these operators. +In §3, we identify $\mathcal{P}$ as a space of least degree among all polyno- +mial spaces that interpolate correctly on certain subsets of the integer +lattice. As a matter of fact, the discussion in that section may have +an independent value: this discussion illustrates how the interpolating +space of least degree from [BR1] may be computed using the tech- +nique from [BR2] of “perturbing the generators of a homogeneous +ideal”, hence in a computationally painless way. + +Section 4 is devoted to the more general discussion of duality be- +tween finite-dimensional polynomial and exponential spaces, a dis- +cussion which improves proofs and results from [DM2] and [DR1]. +Finally, we discuss in §5 the construction of piecewise polynomials on +the support of a box spline and improve thereby an observation in +[R]. + +**2. The main result.** Let $X$ be a multiset of vectors in $\mathbb{R}^s \setminus \{0\}$. We will at times think of $X$, equivalently, as a real matrix, of size $(s \times |X|)$. Let $\mathbb{H}(X)$ denote the collection of all hyperplanes (i.e., linear subspaces of codimension 1) which are spanned by some columns of $X$. We associate with each $h \in \mathbb{H}(X)$ a nontrivial linear polynomial which vanishes on $h$, and write this polynomial + +$$ (h^\perp, \cdot), $$ + +thus using $h^\perp$ to stand for any particular nonzero vector normal to $h$. We are interested in the ideal $I^X$ generated by all polynomials of the form + +$$ (2.1) \qquad p_h := p_{h,X} := \langle h^\perp, \cdot \rangle^{(X\setminus h)}, \quad h \in \mathbb{H}(X), $$ + +where $X\setminus h := \{x \in X \mid x \notin h\}$. \ No newline at end of file diff --git a/samples/texts/102047/page_15.md b/samples/texts/102047/page_15.md new file mode 100644 index 0000000000000000000000000000000000000000..5ef5a13b002b9e2db27dddab612a8ae8d01ef028 --- /dev/null +++ b/samples/texts/102047/page_15.md @@ -0,0 +1,62 @@ +(2.9) LEMMA. + +$$ +b(X) \leq \dim \mathcal{P}(X). +$$ + +(2.10) LEMMA. + +$$ +\mathcal{P}(X) \subset I^{\times}_{\perp}. +$$ + +(2.11) LEMMA. + +$$ +\dim I^{\mathcal{X}} \perp \leq b(X). +$$ + +Proof [DR1] of (2.9) Lemma. Every polynomial + +$$ +q_B := \prod_{x \in X \setminus B} (\langle x, \cdot \rangle - \lambda_x), +$$ + +with $B \in \mathbb{B}(X)$ and $(\lambda_x)$ arbitrary constants, is in $\mathcal{P}(X)$, as follows +readily by multiplying out. For each $B \in \mathbb{B}(X)$, there is a unique +common zero of the $s$ linear factors $\langle x, \cdot \rangle - \lambda_x$, $x \in B$, which do +not occur in the associated $q_B$; call that point $\theta_B$. Choose now, as +we may, the constants $(\lambda_x)$ in such a way that $\theta_B \neq \theta_{B'}$ whenever +$B \neq B'$. (In fact, almost every choice of the $\lambda_x$ would satisfy this +condition.) It then follows that + +$$ +q_B(\theta_{B'}) = 0 \Leftrightarrow B \neq B', +$$ + +proving the linear independence of the collection $(q_B)_{B \in \mathbb{B}(X)}$ in +$\mathcal{P}(X)$. $\square$ + +*Proof of (2.10) Lemma.* We have to prove that, for each $h \in \mathbb{H}(X)$, +$p_h(D) = (D_{h^\perp})^{#(X\setminus h)}$ annihilates $\mathcal{P}(X)$, i.e., that $p_h(D)p_V = 0$ for +every $V \subset X$ for which $X\setminus V$ contains a basis. For this, we note that +$D_{h^\perp}p_{V \cap h} = 0$; hence + +$$ +p_h(D)p_V = (p_{V \cap h}) p_h(D) p_{V \setminus h}. +$$ + +On the other hand, since $X\setminus V$ contains a basis, $V\setminus h$ cannot co- +incide with $X\setminus h$, hence $\deg p_{V\setminus h} < \#X\setminus h = \deg p_h$ and therefore +$p_h(D)p_{V\setminus h} = 0$. $\square$ + +*Proof of (2.11) Lemma.* We prove the lemma by induction on #X. +For the case #X = s we observe that I^X is generated by s linearly in- +dependent linear homogeneous polynomials, consequently I^X ⊥ con- +tains only constants, and so dim I^X ⊥ = 1 = b(X). Assume now that +#X > s. We follow the argument used in the proof of [DM3; Thm. +3.1], decompose X as + +$$ +X := X' \cup X, +$$ \ No newline at end of file diff --git a/samples/texts/102047/page_2.md b/samples/texts/102047/page_2.md new file mode 100644 index 0000000000000000000000000000000000000000..62bd443f889efba77e9210ec5b1aaf59923ab649 --- /dev/null +++ b/samples/texts/102047/page_2.md @@ -0,0 +1,49 @@ +this notion of the least term $f_{\perp}$ of $f$, we then define + +$$ (3.2) \qquad H_{\perp} := \operatorname{span}\{f_{\perp} : f \in H\}, $$ + +and have [BR1] + +$$ (3.3) \qquad \dim H_{\perp} = \dim H. $$ + +We say that a pointset $\Theta \subset \mathbb{R}^s$ is correct for the polynomial space $P$ +if the restriction map $P \to C^\Theta: p \mapsto p|_\Theta$ is invertible. Equivalently, $\Theta$ +is correct for $P$ if, for every data $(d_0)_{\theta \in \Theta}$, there is exactly one $p \in P$ +for which $p(\theta) = d_0$ for all $\theta \in \Theta$. In other words, interpolation +from $P$ at the points of $\Theta$ is correct. + +**RESULT [BR1].** $\Theta$ is correct for $(\exp_{\Theta})_{\downarrow}$. Moreover, among all polynomial spaces $P$ for which $\Theta$ is correct, $(\exp_{\Theta})_{\downarrow}$ is "of least degree" in the sense that + +$$ (3.4) \qquad \dim(P \cap \Pi_j) \le \dim((\exp_{\Theta})_{\downarrow} \cap \Pi_j), \quad \forall j. $$ + +Here, + +$$ (3.5) \qquad \exp_{\Theta} := \operatorname{span}\{e_{\theta}\}_{\theta \in \Theta}, $$ + +where $e_{\theta}: x \mapsto e^{<\theta, x>}$ is the exponential function (with frequency $\theta$). + +In view of this result, it is useful to be able to identify the “least” +space $(\exp_{\Theta})_{\downarrow}$ for given $\Theta$. This we now do for certain pointsets +$\Theta = \nu_X$ associated with the box splines $M_X$. Our tool is the following +result + +(3.6) **RESULT [BR2].** + +$$ H_{\perp} = (I_{H_{\perp}})^{\perp}, $$ + +which obtains $H_{\perp}$ as the kernel of the ideal generated by the leading +terms of the annihilator of $H$. Precisely, [BR2; (4.3) Theorem (b)] +provides the statement that, for any polynomial ideal $I$ of finite codi- +mension, + +$$ (I^{\perp})_{\downarrow} = (I_{\uparrow})^{\perp}, $$ + +and (3.6) Result is obtained by applying this to + +$$ (3.7) \qquad I = I_H := \{p \in \Pi : p(D)H = 0\}, $$ + +which is an ideal since $H$ is closed under differentiation. In fact, with +$H = \exp_{\Theta}$, $I_H$ consists of all polynomials which vanish at $\Theta$; hence +$I_H$ has finite codimension and + +$$ H = I_H^{\perp}, $$ \ No newline at end of file diff --git a/samples/texts/102047/page_3.md b/samples/texts/102047/page_3.md new file mode 100644 index 0000000000000000000000000000000000000000..99c4f6b77c7f4a668aea4e01475b1ddb23fdce05 --- /dev/null +++ b/samples/texts/102047/page_3.md @@ -0,0 +1,25 @@ +(cf. [BR2, §3]). From $I_H$, we obtain its homogeneous counterpart $I_{H\uparrow}$ as + +$$ (3.8) \qquad I_{H\uparrow} := \operatorname{span}\{p_\uparrow : p \in I_H\}, $$ + +where $p_\uparrow$ is the leading term of the polynomial $p$, namely the homogeneous polynomial satisfying + +$$ \deg(p - p_\uparrow) < \deg p. $$ + +The result (3.6) is of interest here since it is easy to identify elements of $I_{H\uparrow}$ in case $H = \exp_\Theta$: If $(p_j)$ are linear homogeneous polynomials for which the union of the corresponding hyperplanes $\{x \in \mathbb{R}^s : p_j(x) = c_j\}$ (for suitable choices of the constants $c_j$) contains $\Theta$, then $p := \prod_j (p_j - c_j) \in I_H$; hence $\prod_j p_j \in (I_{H\uparrow})$. If we obtain enough of these $p$ to generate all of $I_{H\uparrow}$, then we know by (3.6) Result that $H_↓$ is the joint kernel of all the corresponding differential operators $p(D)$. In fact, since we know from (3.3) that $\dim H_↓ = \dim H$, we can already reach this conclusion when we only know that the $p$ so constructed generate an ideal $J$ of codimension $\le \dim H$. + +(3.9) RESULT [BR2]. If the ideal $J$ generated from the leading terms of some polynomials in $I_{H\uparrow}$ has codimension $\le \dim H$, then $J = I_{H\uparrow}$; therefore + +$$ H_↓ = J \perp. $$ + +In our case, we have identified (in (2.7) Theorem) $\mathcal{P}(X)$ as the joint kernel of the differential operators $(D_h)^{#\{X\setminus h\}}$ (with $h$ running over $\mathbb{H}(X)$), hence are entitled to conclude that $(\exp_\Theta)_↓ = \mathcal{P}(X)$ whenever we can find, for each such $h$, constants $c_{j,h}$ so that + +$$ \prod_{j=1}^{#\{X\setminus h\}} ((h^\perp, \cdot) - c_{j,h}) $$ + +vanishes on $\Theta$, and know additionally that $\# \Theta \ge \dim \mathcal{P}(X)$. + +Just such a pointset is, under certain assumptions on $X$, provided by + +$$ (3.10) \quad \nu_X := \nu_X(z) := \left\{ \alpha \in \mathbb{Z}^s : z - \alpha = \sum_{x \in X} t_x x, 0 < t_x < 1, \forall x \right\}, $$ + +with $z \in \mathbb{R}^s \setminus \bigcup_{h \in \mathbb{H}(X)} (h + \mathbb{Z}^s)$. The set $-\nu_X$ comprises the integer points in the support of the shifted box spline $M_X(\cdot + z)$ (cf. [DM2]). \ No newline at end of file diff --git a/samples/texts/102047/page_5.md b/samples/texts/102047/page_5.md new file mode 100644 index 0000000000000000000000000000000000000000..7976bdda6c8db1cadc94771544d3fe8f380d8016 --- /dev/null +++ b/samples/texts/102047/page_5.md @@ -0,0 +1,31 @@ +Let $I_X$ be the ideal generated by all polynomials of the form $p_V = \prod_{v \in V} (v, \cdot)$ with $V \subset X$ and $\text{span}(X \setminus V) \neq \mathbb{R}^s$. This means that + +$$ (4.1) \qquad I_X := \text{ideal}\{p_{X\setminus h} : h \in \mathbb{H}(X)\}. $$ + +Let $\mathcal{H}(X)$ denote the kernel of $I_X$, i.e., + +$$ \mathcal{H}(X) := I_X \perp . $$ + +It is known [BH1], [DM1] that $\mathcal{H}(X)$ is a finite-dimensional polynomial space and [DM2] that + +$$ (4.2) \qquad \dim \mathcal{H}(X) = b(X). $$ + +The spaces $\mathcal{P}(X)$ and $\mathcal{H}(X)$ are dual to each other in the following sense. Each $p \in \Pi$ gives rise to a linear functional $p^*$ on $\Pi$ (and even on a larger space of smooth functions), viz. the linear functional + +$$ p^*: q \mapsto p(D)q(0). $$ + +This allows us to consider, for any two finite-dimensional linear polynomial spaces $Q$ and $R$, the map + +$$ M: Q \to R^* : p \mapsto p^*|_R. $$ + +If $M$ is invertible, we say that $Q$ is *dual to* $R$ (in the sense that we can then use the elements of $Q$ in this fashion to represent uniquely the dual of $R$). Note that the dual to $M$ carries $R^{**} = R$ in the same way to $Q^*$; hence $Q$ is dual to $R$ iff $R$ is dual to $Q$. + +A necessary and sufficient condition for such duality is that $M$ be 1-1 and $\dim R \le \dim Q$ (since then $M$ is necessarily onto). In particular, if $\dim Q = \dim R$, then such duality is assured as soon as we know that, for every $q \in Q \setminus 0$, there is an $r \in R$ for which $q^*(r) \ne 0$. By the duality already mentioned, this is equivalent to having, for every $r \in R \setminus 0$, a $q \in Q$ for which $q^*(r) \ne 0$. + +For $q \in Q$, the linear functional $q^*$ cannot tell the difference between $f$ and $T_k f := f_0 + \cdots + f_{k-1} := \text{the power expansion of } f$ up to order $k$, if $k$ is sufficiently large. This allows us to extend this notion of duality to pairs $Q, R$ in which $R$ is a finite-dimensional space of smooth functions. + +We will eventually make use of the following observation: + +(4.3) **THEOREM.** Let $P$ be an $n$-dimensional homogeneous polynomial space. Let $H$ be an $n$-dimensional space of entire functions. If $P$ is dual to $H_\downarrow$, then $P$ is dual to $H$. + +*Proof.* For any $f \in H\setminus 0$, $f_\downarrow \in H_\downarrow\setminus 0$; hence, by assumption, $p^*(f_\downarrow) \ne 0$ for some $p \in P$. Further, since $P$ is spanned by homogeneous \ No newline at end of file diff --git a/samples/texts/102047/page_7.md b/samples/texts/102047/page_7.md new file mode 100644 index 0000000000000000000000000000000000000000..62f3f27f598d37979341f94280bb13e3a87776b8 --- /dev/null +++ b/samples/texts/102047/page_7.md @@ -0,0 +1,50 @@ +ideal (viz. $I_X$) whose kernel is $\mathcal{H}(X)$. Specifically, given any map +$\Gamma: X \to \mathbb{C}: x \mapsto \Gamma_x$, we consider the ideal + +$$ +(4.7) \qquad I_{\Gamma} := \operatorname{ideal}\{q_h : h \in \mathbb{H}(X)\}, +$$ + +with + +$$ +q_h := \prod_{x \in X \setminus h} (\langle x, \cdot \rangle - \Gamma_x). +$$ + +Then $\mathcal{H}(\Gamma) := I_{\Gamma} \perp$ is an exponential space (i.e., a space which is spanned by certain products of exponentials with polynomials), and [BR2] + +$$ +(4.8) \qquad \mathcal{H}(\Gamma)_\downarrow = \mathcal{H}(X). +$$ + +The diagram below illustrates for a unimodular $X$ the various connections established so far between the ideals $I_X^x$, $I_X$, their kernels and the associated exponential spaces. + +$$ +\begin{tikzcd}[column sep=2.5em, row sep=2.5em] +(\exp_{\nu_X})_\downarrow \arrow[r] & \mathcal{P}(X) \arrow[r] & I_X^\perp \\ +& \arrow[uur]^{dual} & \\ +\mathcal{H}(\Gamma)_\downarrow & \mathcal{H}(X) & I_X^\perp +\end{tikzcd} +$$ + +Our first corollary improves [DM2; Thm. 4.1]: + +(4.9) COROLLARY. Let $X$ be unimodular. Then $\nu_X$ is correct for $\mathcal{H}(X)$, and $\mathcal{H}'(X)$ is of least degree among all polynomial spaces for which $\nu_X$ is correct. + +*Proof*. We apply (4.3) Theorem with $H = \exp_{\nu_X}$ and $P = \mathcal{H}(X)$. +By (3.12) Theorem, $H_\downarrow = \mathcal{P}(X)$, while by (4.6) Result, $\mathcal{P}(X)$ is +dual to $\mathcal{H}(X)$. Since also $\mathcal{H}(X)$ is homogeneous (as the kernel of +a homogeneous ideal), (4.3) Theorem implies that $\mathcal{H}(X)$ is dual to +$\exp_{\nu_X}$, which is equivalent (cf. [BR1; §4]) to the correctness of $\nu_X$ for +$\mathcal{H}(X')$. Moreover, since $\mathcal{P}(X)$ and $\mathcal{H}(X)$ are both homogeneous +and, by (4.6) Result, dual to each other, we must have + +$$ +(4.10) \qquad \dim(\Pi_j \cap \mathcal{H}(X)) = \dim(\Pi_j \cap \mathcal{P}(X)) +$$ + +for every $j$, by (4.4) Proposition. Since we already know by (3.12) +Theorem that $\mathcal{P}(X)$ is of least degree among all polynomial spaces +for which $\nu_X$ is correct, it follows from (4.10) that $\mathcal{H}(X)$ has the +same property. +$\square$ \ No newline at end of file diff --git a/samples/texts/102047/page_8.md b/samples/texts/102047/page_8.md new file mode 100644 index 0000000000000000000000000000000000000000..a3ccfae65d8fc8758ba026348e26f6f73ddbf312 --- /dev/null +++ b/samples/texts/102047/page_8.md @@ -0,0 +1,29 @@ +(4.11) COROLLARY [DR1]. The spaces $\mathcal{P}(X)$ and $\mathcal{H}(\Gamma)$ are dual to each other. + +*Proof*. Take $P = \mathcal{P}(X)$ and $H = \mathcal{H}(\Gamma)$ in (4.3) Theorem. Since, by (4.8), $\mathcal{H}(\Gamma)_\downarrow = \mathcal{H}(X)$, and, by (4.6) Result, $\mathcal{H}(X)$ is dual to $\mathcal{P}(X)$, (4.3) Theorem provides the desired result. $\square$ + +We refer to [DR1; §7] for a discussion of the interpolation conditions induced by $\mathcal{H}(\Gamma)$. + +5. **Application to box splines.** In this section we point out some connections between the results of the previous sections and the theory of multivariate splines. In the discussion here the (polynomial) box spline $M_X$ associated with a set of directions $X$ plays a central role. For our purposes, it is sufficient to note that $M_X$ is a piecewise-polynomial function supported on + +$$ \Omega_X := \{Xt : t \in [0, 1]^{\#X}\} $$ + +and satisfies + +$$ \Pi(M_X) = \mathcal{H}(X), $$ + +where, for a general compactly supported $\phi$, the notation $\Pi(\phi)$ stands for the space of polynomials spanned by the integer translates of $\phi$. + +We make use of the following result, which is a special case of [R; Thm.1.1]: + +(5.1) **RESULT.** Let $P$ be a translation-invariant space of polynomials, and $\Omega$ a compact subset of $\mathbb{R}^s$ with boundary $\partial\Omega$. Then the following conditions are equivalent: + +(a) There exists a function $\phi$ supported in $\Omega$ and satisfying $\Pi(\phi) = P$, $\tilde{\phi}(0) \neq 0$. + +(b) For every $z \in \mathbb{R}^s \setminus \bigcup_{\alpha \in Z^s} \alpha + \partial\Omega$, the set + +$$ v_{\Omega} := v_{\Omega}(z) := \{\alpha \in \mathbb{Z}^s : z - \alpha \in \Omega\} $$ + +is total for $P$, i.e., no element of $P\backslash 0$ vanishes on this set. + +Note that when we take $\Omega$ to be $\Omega_X = \text{supp } M_X$, the sets $v_\Omega$ are identical with the sets $v_X$ from (3.10). Thus, by appealing to (3.12) Theorem, we deduce from the implication (b) ⇒ (a) of (5.1) Result \ No newline at end of file diff --git a/samples/texts/1072043/page_13.md b/samples/texts/1072043/page_13.md new file mode 100644 index 0000000000000000000000000000000000000000..7880a5468e380e07e8393082aa342d5d18d20bb7 --- /dev/null +++ b/samples/texts/1072043/page_13.md @@ -0,0 +1,175 @@ +Criterion (AIC), as shown in equation (16), is used to calculate a comparison of models with different structures. + +$$ +AIC = \log V + \frac{2d}{N} \tag{16} +$$ + +Waveforms of input and output from the experimental setup consist of DC voltage, DC output current, AC voltage and AC output current. Model properties, estimators, percentage of accuracy, Final Prediction Error - FPE and Akaikae Information Criterion - AIC of the model are shown in Table 3. Examples of voltage and current output waveforms of multi input-multi output (MIMO) model in steady state condition (FVMC) having accuracy 97.03% and 91.7 % are shown in Fig 21. + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + +
TypeI/PO/PLinear model parameters
[nb1 nb2 nb3 nb4] poles
[nf1 nf2 nf3 nf4] zeros
[nk1 nk2 nk3 nk4] delays
% fit Voltage CurrentFPEAIC
Steady state conditions
FCLVDZDZ[4 4 3 5];87.33,080.9010.9
[5 5 3 6];85.7
[3 4 4 2]
FCMVPWPW[5 2 4 4];84.5729.036.59
[4 2 3 4];86.4
[2 2 4 3];
FCHVSTST[2 2 3 4];89.526.273.26
[1 2 1 2];88.7
[2 1 3 2];
FVLCSNSN[3 6 3 2];56.80.072.57
[8 5 4 3];60.5
[2 4 3 5];
FVMCWNWN[3 4 2 5];97.03254.457.89
[4 2 3 4];91.7
[2 3 2 4];
FVHCWNWN[1 4 3 5];883,079.810.33
[5 2 3 5];94
[1 3 2 4];
Transient conditions
Step UpDZDZ[3 4 2 4];91.753,2307.40
[4 5 4 3];87.20
[2 3 5 5];
[4 5 2 2];
Step DownPWPW[3 5 5 3];85.993,23310.0
[3 5 4 3];85.12
[3 5 5 4];
[4 4 4 1];
+ +Table 3. Results of a PV inverter modeling using a Hammerstein-Wiener model \ No newline at end of file diff --git a/samples/texts/1072043/page_15.md b/samples/texts/1072043/page_15.md new file mode 100644 index 0000000000000000000000000000000000000000..fe092699622fe4bf8c1741d295e591fdad10ce85 --- /dev/null +++ b/samples/texts/1072043/page_15.md @@ -0,0 +1,17 @@ +## 6. Applications: Power quality problem analysis + +A power quality analysis from the model follows the Standard IEEE 1159 Recommend Practice for Monitoring Electric Power Quality [59]. In this Standard, the definition of power quality problem is defined. In summary, a procedure of this Standard when applied to operating systems can be divided into 3 stages (i) Measurement Transducer, (ii) Measurement Unit and (iii) Evaluation Unit. In comparing operating systems and modeling, modeling is more advantageous because of its predictive power, requiring no actual monitoring. Based on proposed modeling, the measurement part is replaced by model prediction outputs, electrical values such as RMS and peak values, frequency and power are calculated, rather than measured. The actual evaluation is replaced by power quality analysis. The concept representation is shown in Fig.22. + +Fig. 22. Diagram of power quality analysis from IEEE 1159 and application to modeling + +### 6.1 Model output prediction + +In this stage, the model output prediction is demonstrated. From the 8 operation conditions selected in experimental, we choose two representative case. One is the steady state Fix Voltage High Current (FVHC) condition, the other the transient step down condition. To illustrate model predictive power, Fig.23 shows an actual and predictive output current waveforms of the transient step down condition. We see good agreement between experimental results and modeling results. + +### 6.2 Electrical parameter calculation + +In this stage, output waveforms are used to calculate RMS, peak and per unit (p.u.) values, period, frequency, phase angle, power factor, complex power (real, reactive and apparent power), Total Harmonic Distortion - THD. + +#### 6.2.1 Root mean square + +RMS values of voltage and current can be calculated from the following equations: \ No newline at end of file diff --git a/samples/texts/1072043/page_16.md b/samples/texts/1072043/page_16.md new file mode 100644 index 0000000000000000000000000000000000000000..f980340912278d451db712bc70c52eb539cd345b --- /dev/null +++ b/samples/texts/1072043/page_16.md @@ -0,0 +1,15 @@ +$$V_{rms} = \sqrt{\frac{1}{T} \int_{0}^{T} v^2(t) dt} \qquad (17)$$ + +$$I_{rms} = \sqrt{\frac{1}{T} \int_{0}^{T} v^2(t) dt} \qquad (18)$$ + +Fig. 23. Prediction and experiment results of AC output current under a transient step down condition + +### 6.2.2 Period, frequency and phase angle + +We calculate a phase shift between voltage and current from the equation (19), and the frequency (f) from equation (20). + +$$\phi = \frac{\Delta t(\text{ms}) \cdot 360^\circ}{T \text{ ms}} \qquad (19)$$ + +$$f = \frac{1}{T} \qquad (20)$$ + +$\Delta t$ is time lagging or leading between voltage and current (ms), T is the waveform period. \ No newline at end of file diff --git a/samples/texts/1072043/page_2.md b/samples/texts/1072043/page_2.md new file mode 100644 index 0000000000000000000000000000000000000000..aa9b76f2a63d74dc1bbe06a7e81799c71f1241e4 --- /dev/null +++ b/samples/texts/1072043/page_2.md @@ -0,0 +1,33 @@ +## 3.2 Hammerstein-Wiener (HW) nonlinear model + +In this section, a combination of the Wiener model and the Hammerstein model called the Hammerstein-Wiener model is introduced, shown in Fig. 9. In the Wiener model, the front part being a dynamic linear block, representing the system, is cascaded with a static nonlinear block, being a sensor. In the Hammerstein model, the front block is a static nonlinear actuator, in cascading with a dynamic linear block, being the system. This model enables combination of a system, sensors and actuators in one model. The described dynamic system incorporates a static nonlinear input block, a linear output-error model and an output static nonlinear block. + +Fig. 9. Structure of Hammerstein-Weiner Model + +General equations describing the Hammerstein-Wiener structure are written as the Equation (1) + +$$ +\left. +\begin{aligned} +w(t) &= f(u(t)) \\ +x(t) &= \sum_{i=1}^{n_u} \frac{B_i(q)}{F_i(q)} w(t-n_k) \\ +y(t) &= h(x(t)) +\end{aligned} +\right\} \qquad (1) +$$ + +Which $u(t)$ and $y(t)$ are the inputs and outputs for the system. Where $w(t)$ and $x(t)$ are internal variables that define the input and output of the linear block. + +### 3.2.1 Linear subsystem + +The linear block is similar to an output error polynomial model, whose structure is shown in the Equation (2). The number of coefficients in the numerator polynomials $B(q)$ is equal to the number of zeros plus 1, $b_n$ is the number of zeros. The number of coefficients in denominator polynomials $F(q)$ is equal to the number of poles, $f_n$ is the number of poles. The polynomials B and F contain the time-shift operator $q$, essentially the z-transform which can be expanded as in the Equation (3). $n_u$ is the total number of inputs. $n_k$ is the delay from an input to an output in terms of the number of samples. The order of the model is the sum of $b_n$ and $f_n$. This should be minimum for the best model. + +$$ x(t) = \sum_{i=1}^{n_u} \frac{B_i(q)}{F_i(q)} w(t-n_k) \qquad (2) $$ + +$$ +\begin{aligned} +B(q) &= b_1 + b_2 q^{-1} + \dots + b_n q^{-b_n+1} \\ +F(q) &= 1 + f_1 q^{-1} + \dots + f_n q^{-f_n} +\end{aligned} +\qquad (3) +$$ \ No newline at end of file diff --git a/samples/texts/1072043/page_3.md b/samples/texts/1072043/page_3.md new file mode 100644 index 0000000000000000000000000000000000000000..d67b5b07bf1f38735952280435f058e851530b62 --- /dev/null +++ b/samples/texts/1072043/page_3.md @@ -0,0 +1,17 @@ +### 3.2.2 Nonlinear subsystem + +The Hammerstein-Wiener Model composes of input and output nonlinear blocks which contain nonlinear functions $f(\bullet)$ and $h(\bullet)$ that corresponding to the input and output nonlinearities. The both nonlinear blocks are implemented using nonlinearity estimators. Inside nonlinear blocks, simple nonlinear estimators such deadzone or saturation functions are contained. + +i. The dead zone (DZ) function generates zero output within a specified region, called its dead zone or zero interval which shown in Fig. 10. The lower and upper limits of the dead zone are specified as the start of dead zone and the end of dead zone parameters. Deadzone can define a nonlinear function $y = f(x)$, where $f$ is a function of $x$. It composes of three intervals as following in the equation (4) + +$$ \left. \begin{array}{ll} x \le a & f(x) = x - a \\ a < x < b & f(x) = 0 \\ x \ge b & f(x) = x - b \end{array} \right\} \qquad (4) $$ + +when $x$ has a value between $a$ and $b$, when an output of the function equal to $F(x)=0$, this zone is called as a "zero interval". + +Fig. 10. Deadzone function + +ii. **Saturation (ST) function** can define a nonlinear function $y = f(x)$, where $f$ is a function of $x$. It composes of three interval as the following characteristics in the equation (5) and Fig. 11. The function is determined between $a$ and $b$ values. This interval is known as a "linear interval" + +$$ \left. \begin{array}{ll} x > a & f(x) = a \\ a < x < b & f(x) = x \\ x \le b & f(x) = b \end{array} \right\} \qquad (5) $$ + +Fig. 11. Saturation function \ No newline at end of file diff --git a/samples/texts/1072043/page_5.md b/samples/texts/1072043/page_5.md new file mode 100644 index 0000000000000000000000000000000000000000..2b20d7a0e5d750336c3ffa365fbaa358b86fd96e --- /dev/null +++ b/samples/texts/1072043/page_5.md @@ -0,0 +1,18 @@ +wavelet dilation coefficient, cs and cw are scaling translation and wavelet translation coefficients. The scaling function f(.) and the wavelet function g(.) are both radial functions, and can be written as the equation (9) + +$$ +\begin{aligned} +f(u) &= \exp(-0.5 * u' * u) \\ +g(u) &= (\dim(u) - u'*u) * \exp(-0.5 * u'*u) +\end{aligned} +\tag{9} $$ + +In a system identification process, the wavelet coefficient (a), the dilation coefficient (b) and the translation coefficient (c) are optimized during model learning steps to obtain the best performance model. + +### 3.3 MIMO Hammerstein-Wiener system identification + +The voltage and current are two basic signals considered as input/output of PV grid connected systems. The measured electrical input and output waveforms of a system are collected and transmitted to the system identification process. In Fig. 13 show a PV based inverter system which are considered as SISO (single input-single output) or MIMO (multi input-multi output), depending on the relation of input-output under study [57]. In this paper, the MIMO nonlinear model of power inverters of PV systems is emphasized because this model gives us both voltage and current output prediction simultaneously. + +Fig. 13. Block diagram of nonlinear SISO and MIMO inverter model + +For one SISO model, there is only one corresponding set of nonlinear estimators for input and output, and one set of linear parameters, i.e. pole $b_n$, zero $f_n$ and delay $n_k$, as written in the equation (9). For SIMO, MISO and MIMO models, there would be more than one set of \ No newline at end of file diff --git a/samples/texts/1072043/page_6.md b/samples/texts/1072043/page_6.md new file mode 100644 index 0000000000000000000000000000000000000000..2f9a96bf627cb11ad5e956f97736f927cf02f4ca --- /dev/null +++ b/samples/texts/1072043/page_6.md @@ -0,0 +1,37 @@ +nonlinear estimators and linear parameters. The relationships between input-output of the MIMO model have been written in the equation (10) whereas vdc is DC voltage, idc DC current, vac AC voltage, iac AC current. q is shift operator as equivalent to z transform. f(•) and h(•) are input and output nonlinear estimators. In this case a deadzone and saturation are selected into the model. In the MIMO model the relation between output and input has four relations as follows (i) DC voltage (vdc) - AC voltage (vac), (ii) DC voltage (vdc) - AC current (iac), (iii) DC current (idc) - AC voltage (vac) and (iv) DC current(vdc)-AC voltage (vac). + +$$ +\left. +\begin{aligned} +v_{ac}(t) &= \frac{B(q)}{F(q)} f(v_{dc}(t-n_k)) + e(t) \\ +i_{ac}(t) &= \frac{B(q)}{F(q)} f(i_{dc}(t-n_k)) + e(t) +\end{aligned} +\right\} \tag{10} +$$ + +$$ +\begin{equation} +\begin{split} +V_{ac}(t) &= h\left(\frac{B_1(q)}{F_1(q)}f(v_{dc}(t-n_{k1})) + e(t)\right) \otimes h\left(\frac{B_2(q)}{F_2(q)}f(i_{dc}(t-n_{k2})) + e(t)\right) \\ +I_{ac}(t) &= h\left(\frac{B_3(q)}{F_3(q)}f(v_{dc}(t-n_{k3})) + e(t)\right) \otimes h\left(\frac{B_4(q)}{F_4(q)}f(i_{dc}(t-n_{k4})) + e(t)\right) +\end{split} +\tag{11} +\end{equation} +$$ + +$$ +B_i(q) = b_1 + b_2 + \dots + b_{n_{bi}} q^{-n_{bi}+1} +$$ + +$$ +F_i(q) = f_1 + f_2 + \dots + f_{n_{fi}} q^{-n_{fi} + 1} \quad (12) +$$ + +Where $n_{bi}$, $n_{fi}$ and $n_{ki}$ are pole, zero and delay of linear model. Where as number of subscript i are 1,2,3 and 4 which stand for relation between DC voltage-AC voltage, DC current-AC voltage, DC voltage-AC current and DC current-AC current respectively. The output voltage and output current are key components for expanding to the other electrical values of a system such power, harmonic, power factor, etc. The linear parameters, zeros, poles and delays are used to represent properties and relation between the system input and output. There are two important steps to identify a MIMO system. The first step is to obtain experimental data from the MIMO system. According to different types of experimental data, the second step is to select corresponding identification methods and mathematical models to estimate model coefficients from the experimental data. The model is validated until obtaining a suitable model to represent the system. The obtained model provides properties of systems. State-space equations, polynomial equations as well as transfer functions are used to describe linear systems. Nonlinear systems can be described by the above linear equations, but linearization of the nonlinear systems has to be carried out. Nonlinear estimators explain nonlinear behaviors of nonlinear system. Linear and nonlinear graphical tools are used to describe behaviors of systems regarding controllability, stability and so on. + +**4. Experimental** + +In this work, we model one type of a commercial grid connected single phase inverters, +rating at 5,000 W. The experimental system setup composes of the inverter, a variable DC +power supply (representing DC output from a PV array), real and complex loads, a digital +power meter, a digital oscilloscope, , a AC power system and a computer, shown \ No newline at end of file diff --git a/samples/texts/1072043/page_8.md b/samples/texts/1072043/page_8.md new file mode 100644 index 0000000000000000000000000000000000000000..617092ba01afb85d87df87d96896cc00f55139af --- /dev/null +++ b/samples/texts/1072043/page_8.md @@ -0,0 +1,9 @@ +The system identification scheme is shown in Fig.15. Good accuracy of models are achieved by selecting model structures and adjusting the model order of linear terms and nonlinear estimators of nonlinear systems. Finally, output voltage and current waveforms for any type of loads and operating conditions are then constructed from the models. This allows us to study power quality as required. + +## 4.1 Steady state conditions + +To emulate working conditions of PVGCS systems under environment changes (irradiance and temperature) affecting voltage and current inputs of inverters, six conditions of DC voltage variations and DC current variations. The six conditions are listed as Table 1. They are 3 conditions of a fixed DC current with DC low, medium and high voltage, i.e., FCLV (Fixed Current Low Voltage), FCMV (Fixed Current Medium Voltage) and FCHV (Fixed Current High Voltage) which shown in Fig. 16. The other three corresponding conditions are a DC fixed voltage with DC low, medium and high current, i.e., FVLC (Fixed Voltage Low Current), FVMC (Fixed Voltage Medium Current), and FVHC (Fixed Voltage High Current) as shown in Fig.17. + +Fig. 16. AC voltage and current waveforms corresponding to FCLV, FCMV and FCHV conditions + +Fig. 17. AC voltage and current waveforms corresponding to FVLC, FVMC and FVHC conditions \ No newline at end of file diff --git a/samples/texts/1072043/page_9.md b/samples/texts/1072043/page_9.md new file mode 100644 index 0000000000000000000000000000000000000000..4b492c51653f3abe9c518dbeda5fbdb1b97c3587 --- /dev/null +++ b/samples/texts/1072043/page_9.md @@ -0,0 +1,9 @@ +
No.CaseIdc (A)Vdc (V)Pdc (W)Iac (A)Vac (A)Pac (VA)
1FCLV122102520112202420
2FCMV122402880132202860
3FCHV122803360152203300
4FVLC22354702220440
5FVMC102402,400102202,200
6FVHC212455,145232205,060
+ +Table 1. DC and AC parameters of an inverter under changing operating conditions + +## 4.2 Transient conditions + +Transient conditions are studied under two cases which composed of step up power transient and step down power transient. The step up condition is done by increasing power output from 440 to 1,540 W, and the step down condition from 1,540 to 440 W, shown in Table 2. Power waveform data of the two conditions are divided in two groups, the first group is used to estimate model, the second group to validate model. Examples of captured voltage and current waveforms under the step-up power transient condition (440 W or 2 A) to 1540 W or 7 A) and the step-down power transient condition (1540 W or 7 A) to 440 W (2 A) are shown in Fig. 18 and 19, respectively. + +Fig. 18. AC voltage and current waveforms under the step up transient condition \ No newline at end of file diff --git a/samples/texts/1084779/page_1.md b/samples/texts/1084779/page_1.md new file mode 100644 index 0000000000000000000000000000000000000000..0b7307b90235d9e5a82fe33bcdc7efe719452ced --- /dev/null +++ b/samples/texts/1084779/page_1.md @@ -0,0 +1,33 @@ +# OPEN CHALLENGE '06 + +## 1. MAGICAL MAZE + +$$ \begin{array}{ccc} \boxed{19} & \times & \boxed{91} & \times & \boxed{95} & \times & \boxed{38} & \times & \boxed{179} = & \boxed{1117262510} \\ 1419 = \text{IN} & \rightarrow & I = 14 & N = 19 & & & & & \\ 202625 = \text{OUT} & \rightarrow & O = 20 & U = 26 & T = 25 & & & & \end{array} $$ + +Code number is alphabet position + 5 +$1117262510$ = FLUTE +Thus the password is FLUTE. + +## 2. A LITTLE LIGHT MOVEMENT + +The least possible labour involves 18 moves as follows: + +Move to empty room Wardrobe; Bookcase; Piano; Wardrobe; Bookcase; Chest of Drawers; Cabinet; +Bookcase; Wardrobe; Piano; Chest of Drawers; Wardrobe; Bookcase; Cabinet; Wardrobe; Chest of +Drawers; Piano; Bookcase. + +## 3. MIDDLE C + +The clarinettist is at the centre of a circle of radius r. + +$$ \text{Let } \angle OFP = \theta $$ + +$$ \begin{aligned} \cos\theta &= \frac{14^2 + 15^2 - 13^2}{2 \times 14 \times 15} \\ &= \frac{3}{5} \end{aligned} $$ + +$$ \text{Thus } \sin\theta = \frac{4}{5} $$ + +$$ \text{Now } \angle OCP = 2\theta $$ + +$$ \begin{aligned} \text{Thus } r &= \frac{13}{2\sin\theta} = \frac{65}{8} \\ &= 8\frac{1}{8} \text{ feet.} \end{aligned} $$ + +Thus the clarinettist is $8 \text{ feet} 1\frac{1}{2} \text{ inches}$ from the other three. \ No newline at end of file diff --git a/samples/texts/1084779/page_2.md b/samples/texts/1084779/page_2.md new file mode 100644 index 0000000000000000000000000000000000000000..8a1748344539c66254cfa9747624c1ff7c501bf8 --- /dev/null +++ b/samples/texts/1084779/page_2.md @@ -0,0 +1,42 @@ +### 4. THE TRIANO + +The arrangement of notes +within the basic pattern is: + +### 5. CANDLIGHT SONATA + +The lengths of candles A and B are both functions of time such that: + +$$A(t) = \alpha - \frac{\alpha t}{4} \qquad B(t) = \alpha - \frac{\alpha t}{5}, \text{ where } \alpha \text{ is the original length of both candles.}$$ + +At time T + +$$A(T) = \frac{1}{4} B(T)$$ + +$$\alpha - \frac{\alpha T}{4} = \frac{1}{4} \left( \alpha - \frac{\alpha T}{5} \right)$$ + +$$4\alpha - \alpha T = \alpha - \frac{\alpha T}{5}$$ + +Thus + +$$\frac{4}{5}T = 3$$ + +$$T = 3\frac{3}{4} \text{ hours}$$ + +Thus Wolfgang practised for 3hours 45 minutes. + +### 6. SALIERI'S CONFESSION + +NOW AT MY PASSING I ADMIT A GRAVE SIN. CONSUMED BY JEALOUSY I +ADMINISTERED THE POISON ARSENIC TO WOLFGANG AMADEUS MOZART AND +THUS PROCURED HIS END AND MY ADVANCEMENT. +ANTONIO SALIERI + +The poison was arsenic. + +The substitution code gives: + +A B C D E F G H I J K L M N O P Q R S T U V W X Y Z +T C H A I K O V S Y Z X W U R Q P N M L J G F E D B + +Thus the code word used was TCHAIKOVSKY but as he was not born until 1840 his name would have been unknown to Salieri who died in 1825. Hence researchers are convinced that this is not genuine. \ No newline at end of file diff --git a/samples/texts/1469251/page_6.md b/samples/texts/1469251/page_6.md new file mode 100644 index 0000000000000000000000000000000000000000..f708ff040d0721e7c2dbcb6e519c8fdf737bc4c2 --- /dev/null +++ b/samples/texts/1469251/page_6.md @@ -0,0 +1,23 @@ +and rational functions $f_i$ on $U_i$ such that $f_{ij} = f_i - f_j$ is a regular function on $U_i \cap U_j$. We can define, as for (multiplicative) divisors, the notions of definition functions of an additive divisor, equivalence between two additive divisors, etc. We find, for example, that $H^1(X, \mathcal{O})$ is equal to the group of classes of additive divisors on $X$. + +Let $D$ be a (multiplicative) divisor on $X$. We define $\mathrm{supp}D$ to be the set of points $x \in X$ such that $D(x)$ is not the identity element in $\mathcal{R}^{\times}/\mathcal{O}_x^{\times}$, i.e. such that every definition function of $D$ at $x$ is either not defined at $x$ or takes the value 0 at $x$. + +**Proposition 7.** *The support of a divisor *D* on a variety *X* is a closed subset ≠ *X* of *X*, and *D* = 0 if and only if the support is empty.* + +*Proof.* The latter claim is trivial. For the former, we prove that the set $E$ of points $x \in X$ +such that every definition function of $D$ at $x$ belongs to $\mathcal{O}_x^\times$ is a non-empty open subset; +indeed, if we take a definition function $g$ of $D$ at $x$, then it is also a definition function of +$D$ in an open subset $U$ that contains $x$. By hypothesis, if $x \in E$, then $g$ is regular at $x$ and +$g(x) \neq 0$, and we can choose $U$ such that $g$ is regular and invertible on $U$, which proves +that $E$ is open. +$\square$ + +**Proposition 8.** If *D* is a divisor on a normal variety *X*, then *supp*(*D* is a union of hyper-surfaces (i.e. of closed subvarieties of codimension 1). + +*Proof.* If $f$ is a function on a normal variety $Y$, we know that, if $f$ is not defined at $x \in Y$, then $x$ belongs to a variety of poles or zeros of $f$ (i.e. to an irreducible component of the closure of the set of points $x \in Y$ such that $f(x) \in \{0, \infty\}$). So, if we take $f$ to be a definition function for $D$ in some open subset $U \subset X$, then $\text{supp} D \cap U$ is the union of the pole and zero varieties of $f$ in $U$, and we know that these varieties are of codimension 1 ([2, chapitre III]). + +**Remark.** If *X* is not normal, then the support of a divisor *D* on *X* is not necessarily of codimension 1. It is easy to define an affine variety *X* of dimension > 1 that is normal everywhere except at a single point *x*₀ (for example, the point (*a*,*ab*,*b*², *b*³) in the four-dimensional space *K*⁴). There exists a function *u* that is everywhere defined on *X*, and which is entire on the local ring of *x*₀, but which is not contained in this ring; by adding, if necessary, a constant, we can assume that *x*₀ is not a zero of *u*. There is then an open neighbourhood *X*' of *x*₀ such that the divisor of the function induced by *u* on *X*' has support equal to the single point *x*₀. + +Suppose that X is a normal variety, and D is a divisor on X. Let S be a hypersurface of X. If f is a definition function of D at x ∈ S, then the order of f on S ([2]) does not depend on the choice of f, nor on x ∈ S. We denote this integer by ord_S D. It is easy to see that ord_S D = 0 if and only if S ⊂ supp D. If we now take the formal combination C = ∑_S (ord_S D) ⋅ S, where S runs over the set of all hypersurfaces of X, then C is a cycle of codimension 1, and we call it the associated cycle of the divisor D. + +**Proposition 9.** Let X be a normal variety. The map that sends a divisor D to its associated cycle of codimension 1 is an injective homomorphism from the group of divisors on X to the group of cycles of codimension 1. \ No newline at end of file diff --git a/samples/texts/1469251/page_7.md b/samples/texts/1469251/page_7.md new file mode 100644 index 0000000000000000000000000000000000000000..b02d222e8fd43b2f022a6b2716bd59b2ef2939f4 --- /dev/null +++ b/samples/texts/1469251/page_7.md @@ -0,0 +1,25 @@ +*Proof.* The proof is trivial. + +**Proposition 10.** If X is further a non-singular variety, then the homomorphism that sends a divisor to its associated cycle is bijective. + +*Proof.* It suffices to show that, for every hypersurface $S$, there exists a divisor $D$ such that the cycle $1 \cdot S$ is the cycle associated to $D$. Since $X$ is non-singular, for every $x \in X$, the local ring $\mathcal{O}_x$ is factorial [3]; thus, for every $x \in S$, $S$ is defined by one single equation in a neighbourhood of $x$. So there exists a cover $\{U_i\}_{i=1,\dots,p}$ of $S$ by open subsets $U_i$ of $X$, and, for each $i$, a regular function $f_i$ on $U_i$ that is non-zero outside of $U_i \cap S$ in $U_i$ with $\text{ord}_S f_i = 1$. It then follows that $f_i/f_j$ is an invertible regular function in $U_i \cap U_j$. Now take the cover $\{U_i\}_{i=0,\dots,p}$, where $U_0 = CS$, and take $f_0 = 1$. IT is easy to see that the divisor $D$ for which $f_i$ is a definition function of $D$ on $U_i$ is such that the cycle associated to $D$ is $1 \cdot S$. So the proposition is proven. $\square$ + +p. 4-09 + +**Remark.** The proposition is not necessarily true if $X$ is non-singular. For example, for the cone $xy-zw = 0$ in $K^4$, the cycle defined by $x = z = 0$ is not a cycle associated to any divisor. + +## References + +[1] Cartier, P. Diviseurs et dérivations en géométrie algébrique. (Thèse Sc. math. Paris. 1958) (To appear in *Bull. Soc. math. France*). + +[2] Chevalley, C. Fondements de la Géométrie algébrique Paris, Secrétariat mathématique, 1958, multigraphed. (Class taught at the Sorbonne in 1957–58). + +[3] Godement, R. Propriétés analytiques des localités In *Séminarie Cartan-Chevalley*, vol. **8**, 1955–56. (Talk number 19). + +[4] Grothendieck, A. Sur les faisceaux algébriques et les faisceaux analytiques cohérents. In *Séminaire H. Cartan*, vol. **9**. (Talk number 2). + +[5] Serre, J.-P. Faisceaux algébriques cohérents. Ann. Math. **61** (1955), 197–279. + +[6] Serre, J.-P. Sur la cohomologie des variétés algébriques. *J. Math. pures et appl.* **36** (1957), 1–16. + +[7] Weil, A. Fibre spaces in algebraic geometry (Notes taken by A. Wallace, 1952) Chicago, University of Chicago, 1955. \ No newline at end of file diff --git a/samples/texts/1754951/page_1.md b/samples/texts/1754951/page_1.md new file mode 100644 index 0000000000000000000000000000000000000000..b0e323a6bcc9c2bd6dbe1395e087128b6992697d --- /dev/null +++ b/samples/texts/1754951/page_1.md @@ -0,0 +1,24 @@ +# An Exponential Time Parameterized Algorithm for Planar Disjoint Paths* + +Daniel Lokshtanov† + +Pranabendu Misra‡ + +Michal Pilipczuk§ + +Saket Saurabh¶ + +Meirav Zehavi|| + +August 30, 2021 + +## Abstract + +In the Disjoint Paths problem, the input is an undirected graph $G$ on $n$ vertices and a set of $k$ vertex pairs, $\{s_i, t_i\}_{i=1}^k$, and the task is to find $k$ pairwise vertex-disjoint paths such that the $i$'th path connects $s_i$ to $t_i$. In this paper, we give a parameterized algorithm with running time $2^{\mathcal{O}(k^2)} n^{\mathcal{O}(1)}$ for Planar Disjoint Paths, the variant of the problem where the input graph is required to be planar. Our algorithm is based on the unique linkage/treewidth reduction theorem for planar graphs by Adler et al. [JCTB 2017], the algebraic cohomology based technique of Schrijver [SICOMP 1994] and one of the key combinatorial insights developed by Cygan et al. [FOCS 2013] in their algorithm for Disjoint Paths on directed planar graphs. To the best of our knowledge our algorithm is the first parameterized algorithm to exploit that the treewidth of the input graph is small in a way completely different from the use of dynamic programming. + +* A preliminary version of this paper appeared in the proceedings of STOC 2020. +† University of California, Santa Barbara, USA. daniello@ucsb.edu +‡ Max Planck Institute for Informatics, Saarbrucken, Germany. pmisra@mpi-inf.mpg.de +§ Institute of Informatics, University of Warsaw, Poland. michal.pilipczuk@mimuw.edu.pl +¶ The Institute of Mathematical Sciences, HBNI, Chennai, India. saket@imsc.res.in +|| Ben-Gurion University, Beersheba, Israel. meiravze@bgu.ac.il \ No newline at end of file diff --git a/samples/texts/1754951/page_10.md b/samples/texts/1754951/page_10.md new file mode 100644 index 0000000000000000000000000000000000000000..319e5bdf255a6d3e0405f348a8964aa273598c28 --- /dev/null +++ b/samples/texts/1754951/page_10.md @@ -0,0 +1,21 @@ +Figure 12: Crossing (left) and non-crossing (right) walks. + +# 5 Discrete Homotopy + +The purpose of this section is to assert that rather than working with homology (Definition 3.2) or the standard notion of homotopy, to obtain our algorithm it will suffice to work with a notion called *discrete homotopy*. Working with discrete homotopy will substantially shorten and simplify our proof, particularly Section 8. Translation from discrete homotopy to homology is straightforward, thus readers are invited to skip the proofs in this section when reading the paper for the first time. We begin by defining the notion of a weak linkage. This notion is a generalization of a linkage (see Section 3) that concerns walks rather than paths, and which permits the walks to intersect one another in vertices. Here, we only deal with walks that may repeat vertices but which do not repeat edges. Moreover, weak linkages concern walks that are *non-crossing*, a property defined as follows (see Fig. 12). + +**Definition 5.1 (Non-Crossing Walks).** Let $G$ be a plane graph, and let $W$ and $W'$ be two edge-disjoint walks in $G$. A crossing of $W$ and $W'$ is a tuple $(v, e, \hat{e}, e', \tilde{e}')$ where $e, \hat{e}$ are consecutive in $W$, $e', \tilde{e}'$ are consecutive in $W'$, $v \in V(G)$ is an endpoint of $e, \hat{e}, e'$ and $\tilde{e}',$ and when the edges incident to $v$ are enumerated in clockwise order, then exactly one edge in $\{e', \tilde{e}'\}$ occurs between $e$ and $\hat{e}$. We say that $W$ is self-crossing if, either it has a repeated edge, or it has two edge-disjoint subwalks that are crossing. + +We remark that when we say that a collection of edge-disjoint walks is non-crossing, we mean that none of its walks is self-crossing and no pair of its walks has a crossing. + +**Definition 5.2 (Weak Linkage).** Let $G$ be a plane graph. A weak linkage in $G$ of order $k$ is an ordered family of $k$ edge-disjoint non-crossing walks in $G$. Two weak linkages $W = (W_1, \dots, W_k)$ and $Q = (Q_1, \dots, Q_k)$ are aligned if for all $i \in \{1, \dots, k\}$, $W_i$ and $Q_i$ have the same endpoints. Given an instance $(G, S, T, g, k)$ of Planar Disjoint Paths, a weak linkage $W$ in $G$ (or $H_G$) is sensible if its order is $k$ and for every terminal $s \in S$, $W$ has a walk with endpoints $s$ and $g(s)$. + +The following observation is clear from Definitions 5.1 and 5.2. + +**Observation 5.1.** Let $G$ be a plane graph, and let $W$ be a weak linkage in $G$. Let $e_1, e_2$ and $e_3, e_4$ be two pairs of edges in $E(W)$ that tare all distinct and incident on a vertex $v$, and there is some walk in $W$ where $e_1, e_2$ are consecutive, and likewise for $e_3, e_4$. Then, in a clockwise enumeration of edges incident to $v$, the pairs $e_1, e_2$ and $e_3, e_4$ do not cross, that is, they do not occur as $e_1, e_3, e_2, e_4$ in clockwise order (including cyclic shifts). + +We now define the collection of operations applied to transform one weak linkage into another weak linkage aligned with it (see Fig. 13). We remark that the face push operation is not required for our arguments, but we present it here to ensure that discrete homotopy defines an equivalence relation (in case it will find other applications that need this property). + +**Definition 5.3 (Operations in Discrete Homotopy).** Let $G$ be a triangulated plane graph with a weak linkage $W$, and a face $f$ that is not the outer face with boundary cycle $C$. Let $W \in W$. + +* **Face Move.** Applicable to $(W, f)$ if there exists a subpath $P$ of $C$ such that (i) $P$ is a subwalk of $W$, (ii) $1 \le |E(P)| \le |E(f)| - 1$, and (iii) no edge in $E(C) \setminus E(P)$ belongs to \ No newline at end of file diff --git a/samples/texts/1754951/page_11.md b/samples/texts/1754951/page_11.md new file mode 100644 index 0000000000000000000000000000000000000000..5b11d864f93ee48c2d6a8797cd448bf74e961521 --- /dev/null +++ b/samples/texts/1754951/page_11.md @@ -0,0 +1,21 @@ +Figure 13: Face operations. + +any walk in W. Then, the face move operation replaces P in W by the unique subpath of C between the endpoints of P that is edge-disjoint from P. + +* **Face Pull.** Applicable to (W, f) if C is a subwalk Q of W. Then, the face pull operation replaces Q in W by a single occurrence of the first vertex in Q. + +* **Face Push.** Applicable to (W, f) if (i) no edge in E(C) belongs to any walk in W, and (ii) there exist two consecutive edges e, e' in W with common vertex v ∈ V(C) (where W visits e first, and v is visited between e and e') and an order (clockwise or counter-clockwise) to enumerate the edges incident to v starting at e such that the two edges of E(C) incident to v are enumerate between e and e', and for any pair of consecutive edges of W' for all W' ∈ W incident to v, it does not hold that one is enumerated between e and the two edges of E(C) while the other is enumerated between e' and the two edges of E(C). Let $\tilde{e}$ be the first among the two edges of E(C) that is enumerated. Then, the face push operation replaces the occurrence of v between e and e' in W by the traversal of C starting at $\tilde{e}$. + +We verify that the application of a single operation results in a weak linkage. + +**Observation 5.2.** Let $G$ be a triangulated plane graph with a weak linkage $W$, and a face $f$ that is not the outer face. Let $W \in W$ with a discrete homotopy operation applicable to $(W, f)$. Then, the result of the application is another weak linkage aligned to $W$. + +Then, discrete homotopy is defined as follows. + +**Definition 5.4 (Discrete Homotopy).** Let $G$ be a triangulated plane graph with weak linkages $W$ and $W'$. Then, $W$ is discretely homotopic to $W'$ if there exists a finite sequence of discrete homotopy operations such that when we start with $W$ and apply the operations in the sequence one after another, every operation is applicable, and the final result is $W'$. + +We verify that discrete homotopy gives rise to an equivalence relation. + +**Lemma 5.1.** Let $G$ be a triangulated plane graph with weak linkages $W, W'$ and $W''$. Then, (i) $W$ is discretely homotopic to itself, (ii) if $W$ is discretely homotopic to $W'$, then $W'$ is discretely homotopic to $W$, and (iii) if $W$ is discretely homotopic to $W'$ and $W'$ is discretely homotopic to $W'',$ then $W$ is discretely homotopic to $W''$. + +*Proof.* Statement (i) is trivially true. The proof of statement (ii) is immediate from the observation that each discrete homotopy operation has a distinct inverse. Indeed, every face move operation is invertible by a face move operation (applied to the same walk and cycle). Additionally, every face pull operation is invertible by a face push operation (applied to the same walk and cycle), and vice versa. Hence, given the sequence of operations to transform $W$ to $W'$, say $\phi$, the sequence of operations to transform $W'$ to $W$ is obtained by first writing the \ No newline at end of file diff --git a/samples/texts/1754951/page_12.md b/samples/texts/1754951/page_12.md new file mode 100644 index 0000000000000000000000000000000000000000..69029c6c355cd91edf0219d950df4a46e43d0c2d --- /dev/null +++ b/samples/texts/1754951/page_12.md @@ -0,0 +1,28 @@ +# 1 Introduction + +In the Disjoint Paths problem, the input is an undirected graph $G$ on $n$ vertices and a set of $k$ pairwise disjoint vertex pairs, $\{s_i, t_i\}_{i=1}^k$, and the task is to find $k$ pairwise vertex-disjoint paths connecting $s_i$ to $t_i$ for each $i \in \{1, \dots, k\}$. The Disjoint Paths problem is a fundamental routing problem that finds applications in VLSI layout and virtual circuit routing, and has a central role in Robertson and Seymour's Graph Minors series. We refer to surveys such as [21, 43] for a detailed overview. The Disjoint Paths problem was shown to be NP-complete by Karp (who attributed it to Knuth) in a followup paper [25] to his initial list of 21 NP-complete problems [24]. It remains NP-complete even if $G$ is restricted to be a grid [33, 30]. On directed graphs, the problem remains NP-hard even for $k = 2$ [20]. For undirected graphs, Perl and Shiloach [35] designed a polynomial time algorithm for the case where $k = 2$. Then, the seminal work of Robertson and Seymour [37] showed that the problem is polynomial time solvable for every fixed $k$. In fact, they showed that it is fixed parameter tractable (FPT) by designing an algorithm with running time $f(k)n^3$. The currently fastest parameterized algorithm for Disjoint Paths has running time $h(k)n^2$ [26]. However, all we know about $h$ and $f$ is that they are computable functions. That is, we still have no idea about what the running time dependence on $k$ really is. Similarly, the problem appears difficult in the realm of approximation, where one considers the optimization variant of the problem where the aim is to find disjoint paths connecting as many of the $\{s_i, t_i\}$ pairs as possible. Despite substantial efforts, the currently best known approximation algorithm remains a simple greedy algorithm that achieves approximation ratio $O(\sqrt{n})$. + +The Disjoint Paths problem has received particular attention when the input graph is re- +stricted to be planar [2, 17, 42, 14]. Adler et al. [2] gave an algorithm for Disjoint Paths on +planar graphs (Planar Disjoint Paths) with running time $2^{2^{\mathcal{O}(k)}} n^2$, giving at least a concrete form +for the dependence of the running time on $k$ for planar graphs. Schrijver [42] gave an algo- +rithm for Disjoint Paths on directed planar graphs with running time $n^{\mathcal{O}(k)}$, in contrast to the +NP-hardness for $k=2$ on general directed graphs. Almost 20 years later, Cygan et al. [14] im- +proved over the algorithm of Schrijver and showed that Disjoint Paths on directed planar graphs +is FPT by giving an algorithm with running time $2^{2^{\mathcal{O}(k^2)}} n^{\mathcal{O}(1)}$. The Planar Disjoint Paths prob- +lem is well-studied also from the perspective of approximation algorithms, with a recent burst +of activity [7, 8, 9, 10, 11]. Highlights of this work include an approximation algorithm with +factor $\mathcal{O}(n^{9/19} \log^{\mathcal{O}(1)} n)$ [8] and, under reasonable complexity-theoretic assumptions, hardness +of approximating the problem within a factor of $2^{\Omega(\frac{1}{(\log \log n)^2})}$ [10]. + +In this paper, we consider the parameterized complexity of **Planar Disjoint Paths**. Prior to our work, the fastest known algorithm was the $2^{2^{\mathcal{O}(k)}} n^2$ time algorithm of Adler et al. [2]. Double exponential dependence on *k* for a natural problem on planar graphs is something of an outlier–the majority of problems that are FPT on planar graphs enjoy running times of the form $2^{\mathcal{O}(\sqrt{k} \text{ polylog } k)} n^{\mathcal{O}(1)}$ (see, e.g., [15, 18, 19, 29, 36]). This, among other reasons (discussed below), led Adler [1] to pose as an open problem in GROW 2013¹ whether **Planar Disjoint Paths** admits an algorithm with running time $2^{k^{\mathcal{O}(1)}} n^{\mathcal{O}(1)}$. By integrating tools with origins in algebra and topology, we resolve this problem in the affirmative. In particular, we prove the following. + +**Theorem 1.1.** *The Planar Disjoint Paths problem is solvable in time 2*$^{{\mathcal{O}(k^2)}}$*n*^{{\mathcal{O}}(1)}.²* + +In addition to its value as a stand-alone result, our algorithm should be viewed as a piece of an +on-going effort of many researchers to make the Graph Minor Theory of Robertson and Seymour +algorithmically efficient. The graph minors project is abound with powerful algorithmic and + +¹The conference version of [2] appeared in 2011, before [1]. The document [1] erroneously states the open problem for Disjoint Paths instead of for Planar Disjoint Paths—that Planar Disjoint Paths is meant is evident from the statement that a $2^{2^{\mathcal{O}(k)}} n^{\mathcal{O}(1)}$ time algorithm is known. + +²In fact, towards this we implicitly design a w*$^{\mathcal{O}(k)}$-time algorithm, where w is the treewidth of the input graph. \ No newline at end of file diff --git a/samples/texts/1754951/page_13.md b/samples/texts/1754951/page_13.md new file mode 100644 index 0000000000000000000000000000000000000000..594d0a801010261b54b4407d3f4d130f33447cd7 --- /dev/null +++ b/samples/texts/1754951/page_13.md @@ -0,0 +1,19 @@ +operations of $\phi$ in reverse order and then inverting each of them. Finally, Statement (iii) follows by considering the sequence of discrete homotopy operations obtained by concatenating the sequence of operations to transform $W$ to $W'$ and the sequence of operations to transform $W'$ to $W''$. $\square$ + +Towards the translation of discrete homotopy to homology, we need to associate a flow with every weak linkage and thereby extend Definition 3.4. + +**Definition 5.5.** Let $(D, S, T, g, k)$ be an instance of Directed Planar Disjoint Paths. Let $W$ be a sensible weak linkage in $D$. The flow $\phi: A(D) \to \text{RW}(T)$ associated with $W$ is defined as follows. For every arc $e \in A(D)$, define $\phi(e) = 1$ if there is no walk in $W$ that traverses $e$, and $\phi(e) = t$ otherwise where $t \in T$ is the end-vertex of the (unique) walk in $W$ that traverses $e$. + +Additionally, because homology concerns directed graphs, we need the following notation. Given a graph $G$, we let $\vec{G}$ denote the directed graph obtained by replacing every edge $e \in E(G)$ by two arcs of opposite orientations with the same endpoints as $e$. Notice that $G$ and the underlying graph of $\vec{G}$ are not equal (in particular, the latter graph contains twice as many edges as the first one). Given a weak linkage $W$ in $G$, the weak linkage in $\vec{G}$ that corresponds to $W$ is the weak linkage obtained by replacing each edge $e$ in each walk in $W$, traversed from $u$ to $v$, by the copy of $e$ in $\vec{G}$ oriented from $u$ to $v$. + +Now, we are ready to translate discrete homotopy to homology. + +**Lemma 5.2.** Let $(G, S, T, g, k)$ be an instance of Planar Disjoint Paths where $G$ is triangulated. Let $W$ be a sensible weak linkage in $G$. Let $W'$ be a weak linkage discretely homotopic to $W'$. Let $\tilde{W}$ and $\tilde{W}'$ be the weak linkages corresponding to $W$ and $W'$ in $\vec{G}$, respectively. Then, the flow associated with $\tilde{W}$ is homologous to the flow associated with $\tilde{W}'$. + +*Proof.* Let $\phi$ and $\psi$ be the flows associated with $\tilde{W}$ and $\tilde{W}'$, respectively, in $\vec{G}$. Consider a sequence $O_1, O_2, \dots, O_\ell$ of discrete homotopy operations that, starting from $W$, result in $W'$. We prove the lemma by induction on $\ell$. Consider the case when $\ell = 1$. Then, the sequence contains only one discrete homotopy operation, which a face move, face pull or face push operation. Let this operation be applied to a face $f$ and a walk $W \in W$, where the walk $W$ goes from $s \in S$ to $g(s) \in T$. Let $C$ be the boundary cycle of $f$ in $G$, and let $\vec{C}$ denote the collection of arcs in $G$ obtained from the edges of $C$. After this discrete homotopy operation, we obtain a walk $W' \in W'$, which differs from $W$ only in the subset of edges $C$. All other walks are identical between $W$ and $W'$. Hence, $\tilde{W}$ and $\tilde{W}'$ differ in $\vec{G}$ only in a subset of $\vec{C}$. Then observe that the flows $\phi$ and $\psi$ are identical everywhere in $A(\vec{G})$ except for a subset of $\vec{C}$. More precisely, Let $P = \vec{C} \cap \tilde{W}$ and $P' = \vec{C} \cap \tilde{W}'$. Then $\phi(e) = g(s)$ if $e \in P$ and $\phi(e) = 1$ if $e \in \vec{C} - P$; a similar statement holds for $\psi$ and $P'$. Furthermore, it is clear from the description of each of the discrete homotopy operations that $P$ and $P'$ have no common edges and $P \cup P'$ is the (undirected)¹¹ cycle $\vec{C}$ in $\vec{G}$. + +It only remains to describe the homology between the flows $\phi$ and $\psi$, which is exhibited by a function $h$ on the faces of $\vec{G}$. Then $h$ assigns 1 to all faces of $\vec{G}$ that lie in the exterior of $\vec{C}$, and $g(s)$ to all the faces that lie in the interior of $\vec{C}$. Note that $h$ assigns 1 to the outer face of $\vec{G}$. It is easy to verify that $h$ is indeed a homology between $\psi$ and $\phi$, that is, for any edge $e \in A(\vec{G})$ it holds that $\psi(e) = h(f_1)^{-1} \cdot \phi(e) \cdot h(f_2)$, where $f_1$ and $f_2$ are the faces on the left and the right of $e$ with respect to its orientation. This proves the case where $\ell = 1$. + +Now for $\ell > 1$, consider the weak linkage $\tilde{W}^*$ obtained from $\tilde{W}$ after applying the sequence $O_1, O_2, \dots, O_{\ell-1}$. Then by the induction hypothesis, we can assume that the flow associated with $\tilde{W}^*$, say $\psi^*$ is homologous to $\phi$. Further, applying $O_\ell$ to $\tilde{W}^*$ gives us $\tilde{W}'$, and hence the flows $\psi^*$ and $\phi$ are also homologous. Hence, by Observation 3.1 the flows $\phi$ and $\psi$ are homologous. $\square$ + +¹¹That is, the underlying undirected graph of $\vec{C}$ is a cycle. \ No newline at end of file diff --git a/samples/texts/1754951/page_14.md b/samples/texts/1754951/page_14.md new file mode 100644 index 0000000000000000000000000000000000000000..fbf5c73f9ae5de7deaf306cf74df60973b9a7705 --- /dev/null +++ b/samples/texts/1754951/page_14.md @@ -0,0 +1,19 @@ +Having Corollary 3.1 and Lemma 5.2 at hand, we prove the following theorem. + +**Lemma 5.3.** There exists a polynomial-time algorithm that, given an instance $(G, S, T, g, k)$ of Planar Disjoint Paths where $G$ is triangulated, a sensible weak linkage $\mathcal{W}$ in $G$ and a subset $X \subseteq E(G)$, either finds a solution of $(G - X, S, T, g, k)$ or determines that no solution of $(G - X, S, T, g, k)$ is discretely homotopic to $\mathcal{W}$ in $G$. + +*Proof.* We first convert the given instance of Planar Disjoint Paths into an instance of Directed Planar Disjoint Paths as follows. We convert the graph $G$ into the digraph $\tilde{G}$, as described earlier. Then we construct $\tilde{X}$ from $X$ by picking the two arcs of opposite orientation for each edge in $X$. Then we convert the sensible weak linkage $\mathcal{W}$ into a weak linkage $\tilde{\mathcal{W}}$ in $\tilde{G}$. Finally, we obtain the flow $\phi$ in $\tilde{G}$ associated with $\tilde{\mathcal{W}}$. Next, we apply Corollary 3.1 to the instance $(\tilde{G}, S, T, g, k)$, $\tilde{X}$ and $\phi$. Then either it returns a solution $\hat{\mathcal{P}}$ that is disjoint from $\tilde{X}$, or that there is no solution that is homologous to $\phi$ and disjoint from $\tilde{X}$. In the first case, $\hat{\mathcal{P}}$ can be easily turned into a solution $\mathcal{P}$ for the undirected input instance that is disjoint from $X$. In the second case, we can conclude that the undirected input instance has no solution that is discretely homotopic to $\mathcal{W}$. Indeed, if this were not the case, then consider a solution $\mathcal{P}$ to $(G - X, S, T, g, k)$ that is discretely homotopic to $\mathcal{W}$. Then we have a solution $\hat{\mathcal{P}}$ to the directed instance that is disjoint to $\tilde{X}$. Hence, by Lemma 5.2, the flow associated with $\hat{\mathcal{P}}$ is homologous to $\phi$, the flow associated with $\tilde{\mathcal{W}}$. Hence, $\hat{\mathcal{P}}$ is a solution to the instance $(\tilde{G}, S, T, g, k)$ that is disjoint from $\tilde{X}$ and whose flow is homologous to $\phi$. But this is a contradiction to Corollary 3.1. □ + +As a corollary to this lemma, we derive the following result. + +**Corollary 5.1.** There exists a polynomial-time algorithm that, given an instance $(G, S, T, g, k)$ of Planar Disjoint Paths and a sensible weak linkage $\mathcal{W}$ in $H_G$, either finds a solution of $(G, S, T, g, k)$ or decides that no solution of $(G, S, T, g, k)$ is discretely homotopic to $\mathcal{W}$ in $H_G$. + +*Proof.* Consider the instance $(H_G, S, T, g, k)$ along with the set $X = E(H_G) \setminus E(G)$ of forbidden edges. We then apply Lemma 5.3 to $(H_G, S, T, g, k)$, $X$ and $\mathcal{W}$ (note that $H_G$ is triangulated). If we obtain a solution to this instance, then it is also a solution in $G$ since it traverses only edges in $E(H_G) \setminus X = E(G)$. Else, we correctly conclude that there is no solution of $(H_G - X, S, T, g, k)$ (and hence also of $(G, S, T, g, k)$) that is discretely homotopic to $\mathcal{W}$ in $H_G$. □ + +# 6 Construction of the Backbone Steiner Tree + +In this section, we construct a tree that we call a backbone Steiner tree $R = R^3$ in $H_G$. Recall that $H_G$ is the radial completion of $G$ enriched with $4|V(G)| + 1$ parallel copies for each edge. These parallel copies will not be required during the construction of $R$, and therefore we will treat $H_G$ as having just one copy of each edge. Hence, we can assume that $H_G$ is a simple planar graph, and then $E(H_G) = O(n)$ where $n$ is the number of vertices in $G$. We denote $H = H_G$ when $G$ is clear from context. The tree $R$ will be proven to admit the following property: if the input instance is a Yes-instance, then it admits a solution $\mathcal{P} = (P_1, \dots, P_k)$ that is discretely homotopic to a weak linkage $\mathcal{W} = (W_1, \dots, W_k)$ in $H$ aligned with $\mathcal{P}$ that uses at most $2^{\mathcal{O}(k)}$ edges parallel to those in $R$, and none of the edges not parallel to those in $R$. We use the term Steiner tree to refer to any subtree of $H$ whose set of leaves is precisely $S \cup T$. To construct the backbone Steiner tree $R = R^3$, we start with an arbitrary Steiner tree $R^1$ in $H$. Then over several steps, we modify the tree to satisfy several useful properties. + +## 6.1 Step I: Initialization + +We initialize $R^1$ to be an arbitrarily chosen Steiner tree. Thus, $R^1$ is a subtree of $H$ such that $V_{=1}(R^1) = S \cup T$. The following observation is immediate from the definition of a Steiner tree. \ No newline at end of file diff --git a/samples/texts/1754951/page_15.md b/samples/texts/1754951/page_15.md new file mode 100644 index 0000000000000000000000000000000000000000..7f69066fed0b8184e6841053b0fbec2fd7cf5ba5 --- /dev/null +++ b/samples/texts/1754951/page_15.md @@ -0,0 +1,25 @@ +**Observation 6.1.** Let $(G, S, T, g, k)$ be an instance of Planar Disjoint Paths. Let $R'$ be a Steiner tree. Then, $|V_1(R')| = 2k$ and $|V_{\ge 3}(R')| \le 2k - 1$. + +Before we proceed to the next step, we claim that every vertex of $H$ is, in fact, “close” to the vertex set of $R_1$. For this purpose, we need the following proposition by Jansen et al. [23]. + +**Proposition 6.1 (Proposition 2.1 in [23zia]).** Let $G$ be a plane graph and with disjoint subsets $X, Y \subseteq V(G)$ such that $G[X]$ and $G[Y]$ are connected graphs and $\mathrm{rdist}_G(X, Y) = d \ge 2$. For any $r \in \{0, 1, \dots, d-1\}$, there is a cycle $C$ in $G$ such that all vertices $u \in V(C)$ satisfy $\mathrm{rdist}_G(X, \{u\}) = r$, and such that $V(C)$ separates $X$ and $Y$ in $G$. + +Additionally, we need the following simple observation. + +**Observation 6.2.** Let $G$ be a triangulated plane graph. Then, for any pair of vertices $u, v \in V(G)$, $\mathrm{dist}_G(u, v) = \mathrm{rdist}_G(u, v)$. + +*Proof.* Let $\mathrm{rdist}_G(u, v) = t$, and consider a sequence of vertices $u = x_1, x_2, \dots, x_{t+1} = v$ that witnesses this fact—then, every two consecutive vertices in this sequence have a common face. Since $G$ is triangulated, we have that $\{x_i, x_{i+1}\} \in E(G)$ for every two consecutive vertices $x_i, x_{i+1}$, $1 \le i \le t$. Hence, $x_1, x_2, \dots, x_{t+1}$ is a walk from $u$ to $v$ in $G$ with $t$ edges, and therefore $\mathrm{dist}_G(u, v) \le \mathrm{rdist}_G(u, v)$. Conversely, let $\mathrm{dist}_G(u, v) = \ell$; then, there is a path with $\ell$ edges from $u$ to $v$ in $G$, which gives us a sequence of vertices $u = y_1, y_2, \dots, y_{\ell+1} = v$ where each pair of consecutive vertices forms an edge in $G$. Since $G$ is planar, each such pair of consecutive vertices $y_i, y_{i+1}$, $1 \le i \le \ell$, must have a common face. Therefore, $\mathrm{rdist}_G(u, v) \le \mathrm{dist}_G(u, v)$. $\square$ + +It is easy to see that Observation 6.2 is not true for general plane graph. However, this +observation will be useful for us because the graph $H$, where we construct the backbone Steiner +tree, is triangulated. We now present the promised claim, whose proof is based on Proposition +6.1, Observation 6.2 and the absence of long sequences of $S \cup T$-free concentric cycles in good +instances. Here, recall that $c$ is the fixed constant in Corollary 4.1. We remark that, for the +sake of clarity, throughout the paper we denote some natural numbers whose value depends on +$k$ by notations of the form $\alpha_{\text{subscript}}(k)$ where the subscript of $\alpha$ hints at the use of the value. + +**Lemma 6.1.** Let $(G, S, T, g, k)$ be a good instance of Planar Disjoint Paths. Let $R'$ be a Steiner tree. For every vertex $v \in V(H)$, it holds that $\mathrm{dist}_H(v, V(R')) \le \alpha_{\mathrm{dist}}(k) := 4 \cdot 2^{ck}$. + +*Proof.* Suppose, by way of contradiction, that $\mathrm{dist}_H(v^*, V(R')) > \alpha_{\mathrm{dist}}(k)$ for some vertex $v^* \in V(H)$. Since $H$ is the (enriched) radial completion of $G$, it is triangulated. By Observation 6.2, $\mathrm{rdist}_H(u,v) = \mathrm{dist}_H(u,v)$ for any pair of vertices $u,v \in V(H)$. Thus, $\mathrm{rdist}_H(v^*, V(R')) > \alpha_{\mathrm{dist}}(k)$. By Proposition 6.1, for any $r \in \{0, 1, \dots, \alpha_{\mathrm{dist}}(k)\}$, there is a cycle $C_r$ in $H$ such that all vertices $u \in V(C_r)$ satisfy $\mathrm{rdist}_H(v^*, u) = \mathrm{dist}_H(v^*, u) = r$, and such that $V(C_r)$ separates $\{v^*\}$ and $V(R')$ in $H$. In particular, these cycles must be pairwise vertex-disjoint, and each one of them contains either (i) $v^*$ in its interior (including the boundary) and $V(R')$ in its exterior (including the boundary), or (ii) $v^*$ in its exterior (including the boundary) and $V(R')$ in its interior (including the boundary). We claim that only case (i) is possible. Indeed, suppose by way of contradiction that $C_i$, for some $r \in \{0, 1, \dots, \alpha_{\mathrm{dist}}(k)\}$, contains $v^*$ in its exterior and $V(R')$ in its interior. Because the outer face of $H$ contains a terminal $t^* \in T$ and $t^* \in V(R')$, we derive that $t^* \in V(C_i)$. Thus, $\mathrm{rdist}_H(v^*, t^*) = i \le \alpha_{\mathrm{dist}}(k)$. However, because $t^* \in V(R')$, this is a contradiction to the supposition that $\mathrm{dist}_H(v^*, V(R')) > \alpha_{\mathrm{dist}}(k)$. Thus, our claim holds true. From this, we conclude that $C = (C_0, C_1, \dots, C_{\alpha_{\mathrm{dist}}(k)})$ is a $V(R')$-free sequence of concentric cycles in $H$. Since $S \cup T \subseteq V(R')$, it is also $S \cup T$-free. + +Consider some odd integer $r \in \{1, 2, \dots, \alpha_{\text{dist}(k)}\}$. Note that every vertex $u \in V(C_r)$ that does not belong to $V(G)$ lies in some face $f$ of $G$, and that the two neighbors of $u$ in $C_r$ must belong to the boundary of $f$ (by the definition of radial completion). Moreover, each of the \ No newline at end of file diff --git a/samples/texts/1754951/page_16.md b/samples/texts/1754951/page_16.md new file mode 100644 index 0000000000000000000000000000000000000000..09d9d07671cc2fe2f0dda6b7b1f6a98146fad4a1 --- /dev/null +++ b/samples/texts/1754951/page_16.md @@ -0,0 +1,29 @@ +vertices on the boundary of $f$ is at distance (in $H$) from $u$ that is the same, larger by one or +smaller by one, than the distance of $f$ from $u$, and hence none of these vertices can belong to +any $C_i$ for $i > r + 1$ as well as $i < r + 1$. For every $r \in \{1, 2, \dots, \alpha_{\text{dist}}(k) - 1\}$ such that $r$ +mod $3 = 1$, define $C'_r$ as some cycle contained in the closed walk obtained from $C_r$ by replacing +every vertex $u \in V(C_r) \setminus V(G)$, with neighbors $x, y$ on $C_r$, by a path from $x$ to $y$ on the boundary +of the face of $G$ that corresponds to $u$. In this manner, we obtain an $S \cup T$-free sequence of +concentric cycles in $G$ whose length is at least $2^{ck}$. However, this contradicts the supposition +that $(G, S, T, g, k)$ is good. +□ + +## 6.2 Step II: Removing Detours + +In this step, we modify the Steiner tree to ensure that there exist no “shortcuts” via vertices outside the Steiner tree. This property will be required in subsequent steps to derive additional properties of the Steiner tree. To formulate this, we need the following definition (see Fig. 6). + +**Definition 6.1 (Detours in Trees).** A subtree *T* of a graph *G* has a detour if there exist two vertices *u*, *v* ∈ *V*≥3(*T*) ∪ *V*=1(*T*) that are near each other, and a path *P* in *G*, such that + +1. *P* is shorter than $\mathbf{path}_T(u, v)$, and + +2. one endpoint of *P* belongs to the connected component of $T - V(\mathbf{path}_T(u,v)) \setminus \{u,v\}$ that contains *u*, and the other endpoint of *P* belongs to the connected component of $T - V(\mathbf{path}_T(u,v)) \setminus \{u,v\}$ that contains *v*. + +Such vertices *u*, *v* and path *P* are said to witness the detour. Moreover, if *P* has no internal vertex from (*V*(*T*) \ *V*(\*path*T*(*u*, *v*)) ) ∪ {*u*, *v*} and its endpoints do not belong to *V*=1(*T*) \ {*u*, *v*}, then *u*, *v* and *P* are said to witness the detour compactly. + +We compute a witness for a detour as follows. Note that this lemma also implies that, if there exists a detour, then there exists a compact witness rather than an arbitrary one. + +**Lemma 6.2.** There exists an algorithm that, given a good instance (G, S, T, g, k) of Planar Disjoint Paths and a Steiner tree R', determines in time O(k²·n) whether R' has a detour. In case the answer is positive, it returns u, v and P that witness the detour compactly. + +*Proof.* Let $Q = \mathbf{path}_{R'}(u, v) - \{u, v\}$ for some two vertices $u, v \in V_{\ge 3}(T) \cup V_{=1}(T)$ that are near each other. Then, $R' - V(Q)$ contains precisely two connected components: $R'_u$ and $R'_v$ that contain $u$ and $v$, respectively. Consider a path $P$ of minimum length between vertices $x \in V(R'_u)$ and $y \in V(R'_v)$ in $H$, over all choices of $x$ and $y$. Further, we choose $P$ so that contains as few vertices of $(S \cup T) \setminus \{u, v\}$ as possible. Suppose that $|E(P)| \le |E(\mathbf{path}_{R'}(u, v))| - 1$. Then, we claim that $P$ is a compact detour witness. To prove this claim, we must show that (i) no internal vertex of $P$ lies in $(V(R') \setminus V(\mathbf{path}_{R'}(u, v))) \cup \{u, v\} = V(R'_u) \cup V(R'_v)$, and (ii) the endpoints of $P$ do not lie in $V_{=1}(R') \setminus \{u, v\} = (S \cup T) \setminus \{u, v\}$. The first property follows directly from the choice of $P$. Indeed, if $P$ were a path from $x \in V(R'_u)$ to $y \in V(R'_v)$, which contained an internal vertex $z \in V(R'_u)$, then the subpath $P'$ of $P$ with endpoints $z$ and $y$ is a strictly shorter path from $V(R'_u)$ to $V(R'_v)$ (the symmetric argument holds when $z \in V(R'_v)$). + +For the second property, we give a proof by contradiction. To this end, suppose that some terminal $w \in (S \cup T) \setminus \{u, v\}$ belongs to $P$. Necessarily, $w \in V(R'_u) \cup V(R'_v)$ (by the definition of a Steiner tree). Without loss of generality, suppose that $w \in V(R'_u)$. By the first property, $w$ must be an endpoint of $P$. Let $z \in V(R'_v)$ be the other endpoint of $P$. Because the given instance is good, $w$ has degree 1 in $G$, thus we can let $n(w)$ denote its unique neighbor in $G$. Observe that $w$ lies on only one face of $G$, which contains both $w$ and $n(w)$. Hence, $w$ is adjacent to exactly two vertices in $H: n(w)$ and a vertex $f(w) \in V(H) \setminus V(G)$. Furthermore, $\{n(w), f(w)\} \in E(H)$, i.e. $w, n(w)$ and $f(w)$ form a triangle in $H$. Thus, $P$ contains exactly one \ No newline at end of file diff --git a/samples/texts/1754951/page_17.md b/samples/texts/1754951/page_17.md new file mode 100644 index 0000000000000000000000000000000000000000..01e58345c3886584a63e0f3554e18d8f7defd5ef --- /dev/null +++ b/samples/texts/1754951/page_17.md @@ -0,0 +1,11 @@ +of $n(w)$ or $f(w)$ (otherwise, we can obtain a strictly shorter path connecting $w$ and the other endpoint of $P$ that contradicts the choice of $P$). Let $a(w) \in \{n(w), f(w)\}$ denote the neighbor of $w$ in $P$, and note that, by the first property, $a(w) \notin V(R'_u)$. Note that it may be the case that $a(w) = z$. Since $w$ is a leaf of $R'$, exactly one of $n(w)$ and $f(w)$ is adjacent to $w$ in $R'$, and we let $b(w)$ denote this vertex. Because $w \neq u$, we have that $V(R'_u)$ contains but is not equal to $\{w\}$, and therefore $b(w) \in V(R'_v)$. In turn, by the first property, this means that $a(w) \neq b(w)$ (because otherwise $a(w) \neq z$ and hence it is an internal vertex of $P$, which cannot belong to $V(R'_u)$). Because $w, a(w)$ and $b(w)$ form a triangle in $H$, we obtain a path $P' \neq P$ in $H$ by replacing $w$ with $b(w)$ in $P$. Observe that $P'$ connects the vertex $b(w) \in V(R'_u)$ to the vertex $z \in V(R'_v)$. Furthermore, because $|E(P')| = |E(P)|$, and $P'$ contains strictly fewer vertices of $(S \cup T) \setminus \{u, v\}$ compared to $P$, we contradict the choice of $P$. Therefore, $P$ also satisfies the second property, and we conclude that $u, v, P$ compactly witness a detour in $R'$. + +We now show that a compact detour in $R'$ can be computed in $\mathcal{O}(k^2 \cdot n)$ time. First, observe that if there is a detour witnessed by some $u, v$ and $P$, then $u, v \in V_{\ge 3}(R') \cup V_{=1}(R')$. By Observation 6.1, $|V_{\ge 3}(R') \cup V_{=1}(R')| \le 4k$. Therefore, there are at most $16k^2$ choices for the vertices $u$ and $v$. We consider each choice, and test if there is detour for it in linear time as follows. Fix a choice of distinct vertices $u, v \in V_{\ge 3}(R') \cup V_{=1}(R')$, and check if they are near each other in $R'$ in $\mathcal{O}(|V(R)|)$ time by validating that each internal vertex of $\textbf{path}_{R'}(u, v)$ has degree 2. If they are not near each other, move on to the next choice. Otherwise, consider the path $Q = \textbf{path}_{R'}(u, v) - \{u, v\}$, and the trees $R'_u$ and $R'_v$ of $R' - V(Q)$ that contain $u$ and $v$, respectively. Now, consider the graph $\tilde{H}$ derived from $H$ by first deleting $(V(Q) \cup S \cup T) \setminus \{u, v\}$ and then introducing a new vertex $r$ adjacent to all vertices in $V(R'_u)$. We now run a breadth first search (BFS) from $r$ in $\tilde{H}$. This step takes $\mathcal{O}(n)$ time since $|E(\tilde{H})| = \mathcal{O}(n)$ (because $H$ is planar). From the BFS-tree, we can easily compute a shortest path $P$ between a vertex $x \in V(R'_u)$ and a vertex $y \in V(R'_v)$. Observe that $V(P) \cap (S \cup T) \subseteq \{u, v\}$ by the construction of $\tilde{H}$. If $|E(P)| < |E(\textbf{path}_{R'}(u, v))|$, then we output $u, v, P$ as a compact witness of a detour in $R'$. Else, we move on to the next choice of $u$ and $v$. If we fail to find a witness for all choices of $u$ and $v$, then we output that $R'$ has no detour. Observe that the total running time of this process is bounded by $\mathcal{O}(k^2 \cdot n)$. This concludes the proof. $\square$ + +Accordingly, as long as $R^1$ has a detour, compactly witnessed by some vertices $u, v$ and a path $P$, we modify it as follows: we remove the edges and the internal vertices of $\textbf{path}_{R^1}(u, v)$, and add the edges and the internal vertices of $P$. We refer to a single application of this operation as *undetouring* $R^1$. For a single application, because we consider compact witnesses rather than arbitrary ones, we have the following observation. + +**Observation 6.3.** Let $(G, S, T, g, k)$ be a good instance of Planar Disjoint Paths with a Steiner tree $R'$. The result of undetouring $R'$ is another Steiner tree with fewer edges than $R'$. + +*Proof.* Consider a compact detour witness $(u,v,P)$ of $R'$. Then, $|E(P)| < |E(\textbf{path}_{R'}(u,v))|$. Let $Q = \textbf{path}_{R'}(u,v) - \{u,v\}$, and let $R'_u$ and $R'_v$ be the two trees of $R' - V(Q)$ that contain $u$ and $v$, respectively. Consider the graph $\tilde{R}$ obtained from $(R' - V(Q)) \cup P$ by iteratively removing any leaf vertex that does not lie in $S \cup T$. We claim that the graph $\tilde{R}$ that result from undetouring $R'$ (with respect to $(u,v,P)$) is a Steiner tree with strictly fewer edges than $R'$. Clearly, $\tilde{R}$ is connected because $P$ reconnects the two trees $R'_u$ and $R'_v$ of $R' - V(Q)$. Further, as $P$ contains no internal vertex from $(V(R') \setminus V(\textbf{path}_{R'}(u,v))) \cup \{u,v\} = V(R'_u) \cup V(R'_v)$, and $R'_u$ and $R'_v$ are trees, $\tilde{R}$ is cycle-free. Additionally, all the vertices in $S \cup T$ are present in $\tilde{R}$ by construction and they remain leaves due to the compactness of the witness. Hence, $\tilde{R}$ is a Steiner tree in $G$. Because $|E(P)| < |E(\textbf{path}_{R'}(u,v))|$, it follows that $\tilde{R}$ contains fewer edges than $R'$. $\square$ + +Initially, $R^1$ has at most $n-1$ edges. Since every iteration decreases the number of edges (by Observation 6.3) and can be performed in time $\mathcal{O}(k^2 \cdot n)$ (by Lemma 6.2), we obtain the following result. \ No newline at end of file diff --git a/samples/texts/1754951/page_18.md b/samples/texts/1754951/page_18.md new file mode 100644 index 0000000000000000000000000000000000000000..d4127d56ec0cabfea83c9e24a32e2519fdd5b534 --- /dev/null +++ b/samples/texts/1754951/page_18.md @@ -0,0 +1,23 @@ +Figure 14: Separators and flows for long degree-2 paths. + +**Lemma 6.3.** Let $(G, S, T, g, k)$ be a good instance of Planar Disjoint Paths with a Steiner tree $R'$. An exhaustive application of the operation undetouring $R'$ can be performed in time $O(k^2 \cdot n^2)$, and results in a Steiner tree that has no detour. + +We denote the Steiner tree obtained at the end of Step II by $R^2$. + +## 6.3 Step III: Small Separators for Long Paths + +We now show that any two parts of $R^2$ that are “far” from each other can be separated by small separators in $H$. This is an important property used in the following sections to show the existence of a “nice” solution for the input instance. Specifically, we consider a “long” maximal degree-2 path in $R^2$ (which has no short detours in $H$), and show that there are two separators of small cardinality, each “close” to one end-point of the path. The main idea behind the proof of this result is that, if it were false, then the graph $H$ would have had large treewidth (see Proposition 6.2), which contradicts that $H$ has bounded treewidth (by Corollary 4.1). We first define the threshold that determines whether a path is long or short. + +**Definition 6.2 (Long Paths in Trees).** Let $G$ be a graph with a subtree $T$. A subpath of $T$ is $k$-long if its length is at least $\alpha_{\text{long}}(k) := 10^4 \cdot 2^{ck}$, and $k$-short otherwise. + +As $k$ will be clear from context, we simply use the terms long and short. Towards the computation of two separators for each long path, we also need to define which subsets of $V(R^2)$ we would like to separate. + +**Definition 6.3 ($P'_u, P''_u, A_{R^2,P,u}$ and $B_{R^2,P,u}$).** Let $(G, S, T, g, k)$ be a good instance of Planar Disjoint Paths. Let $R^2$ be a Steiner tree that has no detour. For any long maximal degree-2 path $P$ of $R^2$ and for each endpoint $u$ of $P$, define $P'_u$, $P''_u$ and $A_{R^2,P,u}$, $B_{R^2,P,u} \subseteq V(R^2)$ as follows. + +* $P'_u$ (resp. $P''_u$) is the subpath of $P$ consisting of the $\alpha_{\text{pat}}(k) := 100 \cdot 2^{ck}$ (resp. $\alpha_{\text{pat}}(k)/2 = 50 \cdot 2^{ck}$) vertices of $P$ closest to $u$. + +* $A_{R^2,P,u}$ is the union of $V(P''_u)$ and the vertex set of the connected component of $R^2$ - $(V(P'_u) \setminus \{u\})$ containing $u$. + +* $B_{R^2,P,u} = V(R^2) \setminus (A_{R^2,P,u} \cup V(P'_u))$. + +For each long maximal degree-2 path $P$ of $R^2$ and for each endpoint $u$ of $P$, we compute a “small” separator $\text{Sep}_{R^2}(P, u)$ as follows. Let $A = A_{R^2,P,u}$ and $B = B_{R^2,P,u}$. Then, compute a subset of $V(H) \setminus (A \cup B)$ of minimum size that separates $A$ and $B$ in $H$, and denote it by $\text{Sep}_{R^2}(P, u)$ (see Fig. 14). Since $A \cap B = \emptyset$ and there is no edge between a vertex in $A$ and a vertex in $B$ (because $R^2$ has no detours), such a separator exists. Moreover, it can be computed \ No newline at end of file diff --git a/samples/texts/1754951/page_19.md b/samples/texts/1754951/page_19.md new file mode 100644 index 0000000000000000000000000000000000000000..7e5906878cb23ef3184238cb98c4ef70546027e1 --- /dev/null +++ b/samples/texts/1754951/page_19.md @@ -0,0 +1,31 @@ +in time $O(n|\mathbf{Sep}_{R^2}(P, u)|)$: contract each set among A and B into a single vertex and then obtain a minimum vertex $s-t$ cut by using Ford-Fulkerson algorithm. + +To argue that the size of $\mathbf{Sep}_{R^2}(P, u)$ is upper bounded by $2^{O(k)}$, we make use of the following proposition due to Bodlaender et al. [5]. + +**Proposition 6.2 (Lemma 6.11 in [5]).** Let $G$ be a plane graph, and let $H$ be its radial completion. Let $t \in \mathbb{N}$. Let $C, Z, C_1, Z_1$ be disjoint subsets of $V(H)$ such that + +1. $H[C]$ and $H[C_1]$ are connected graphs, + +2. $Z$ separates $C$ from $Z_1 \cup C_1$ and $Z_1$ separates $C \cup Z$ from $C_1$ in $H$, + +3. $\text{dist}_H(Z, Z_1) \geq 3t + 4$, and + +4. $G$ contains $t+2$ pairwise internally vertex-disjoint paths with one endpoint in $C \cap V(G)$ and the other endpoint in $C_1 \cap V(G)$. + +Then, the treewidth of $G[V(M) \cap V(G)]$ is larger than $t$ where $M$ is the union of all connected components of $H \setminus (Z \cup Z_1)$ having at least one neighbor in $Z$ and at least one neighbor in $Z_1$. + +Additionally, the following immediate observation will come in handy. + +**Observation 6.4.** Let $G$ be a plane graph. Let $H$ be the radial completion of $G$, and let $H'$ be the radial completion of $H$. Then, for all $u, v \in V(H)$, $\text{dist}_H(u, v) \leq \text{dist}_{H'}(u, v)$. + +*Proof.* Note that $H$ is triangulated. Thus, for all $u, v \in V(H)$ and path $P$ in $H'$ between $u$ and $v$, we can obtain a path between $u$ and $v$ whose length is not longer than the length of $P$ by replacing each vertex $w \in V(H') \setminus V(H)$ by at most one vertex of the boundary of the face in $H$ that $w$ represents. $\square$ + +We now argue that $\mathbf{Sep}_{R^2}(P, u)$ is small. + +**Lemma 6.4.** Let $(G, S, T, g, k)$ be a good instance of Planar Disjoint Paths. Let $\mathbb{R}^2$ be a Steiner tree that has no detour, $P$ be a long maximal degree-2 path of $\mathbb{R}^2$, and $u$ be an endpoint of $P$. Then, $|\mathbf{Sep}_{\mathbb{R}^2}(P, u)| \leq \alpha_{\text{sep}}(k) := \frac{7}{2} \cdot 2^{ck} + 2$. + +*Proof.* Denote $P' = P_u', P'' = P_u''$, $A = A_{R^2,P,u}$ and $B = B_{R^2,P,u}$. Recall that $H$ is the radial completion of $G$ (enriched with parallel edges), and let $H'$ denote the radial completion of $H$. Towards an application of Proposition 6.2, define $C = A$, $C_1 = B$, $Z = N_{H'}(C)$, $Z_1 = N_{H'}(C_1)$ and $t = \frac{7}{2} \cdot 2^{ck}$. Since $\mathbb{R}^2$ is a subtree of $H$, it holds that $H[C]$ and $H[C_1]$ are connected, and therefore $H'[C]$ and $H'[C_1]$ are connected as well. From the definition of $Z$ and $Z_1$, it is immediate that $Z$ separates $C$ from $Z_1 \cup C_1$ and $Z_1$ separates $C \cup Z$ from $C_1$ in $H$. Clearly, $C \cap C_1 = \emptyset$, $C \cap Z = \emptyset$, and $C_1 \cap Z_1 = \emptyset$. We claim that, in addition, $Z \cap C_1 = \emptyset$, $Z_1 \cap C = \emptyset$ and $Z \cap Z_1 = \emptyset$. To this end, it suffices to show that $\text{dist}_{H'}(Z, Z_1) \geq 3t + 4$. Indeed, because $Z = N_{H'}(C)$ and $Z_1 = N_{H'}(C_1)$, we have that each inequality among $Z \cap C_1 \neq \emptyset$, $Z_1 \cap C = \emptyset$ and $Z \cap Z_1 = \emptyset$, implies that $\text{dist}_{H'}(Z, Z_1) \leq 2$. + +Lastly, we show that $\text{dist}_{H'}(Z, Z_1) \geq 3t + 4$. As $\text{dist}_{H'}(C, C_1) \leq \text{dist}_H(Z, Z_1) + 2$, it suffices to show that $\text{dist}_{H'}(C, C_1) \geq 3t + 6$. Because $C \cup C_1 \subseteq V(H)$, Observation 6.4 implies that $\text{dist}_{H'}(C, C_1) \geq \text{dist}_H(C, C_1)$. Hence, it suffices to show that $\text{dist}_H(C, C_1) \geq 3t + 6$. However, $\text{dist}_H(C, C_1) \geq |E(P')| - |E(P'')|$ since otherwise we obtain a contradiction to the supposition that $\mathbb{R}^2$ has no detour. This means that $\text{dist}_H(C, C_1) \geq \alpha_{\text{pat}}(k)/2 - 1 \geq 3t + 6$ as required. + +Recall that $\mathbf{Sep}_{\mathbb{R}^2}(P, u)$ is a subset of $V(H) \setminus (C \cup C'_1)$ of minimum size that separates $C$ and $C_1$ in $H$. We claim that $\left|\mathbf{Sep}_{\mathbb{R}^2}(P, u)\right| \leq \alpha_{\text{sep}}(k)$. Suppose, by way of contradiction, that $\left|\mathbf{Sep}_{\mathbb{R}^2}(P, u)\right| > \alpha_{\text{sep}}(k) = t + 2$. By Menger's theorem, the inequality $\left|\mathbf{Sep}_{\mathbb{R}^2}(P, u)\right| > \alpha_{\text{sep}}(k)$ implies that $H$ contains $t+2$ pairwise internally vertex-disjoint paths with one endpoint in $C \subseteq V(H)$ and the other endpoint in $C_1 \subseteq V(H)$. From this, we conclude that all of the conditions in the premise of Proposition 6.2 are satisfied. Thus, the treewidth of $H[V(M) \cap V(H)]$ is larger \ No newline at end of file diff --git a/samples/texts/1754951/page_2.md b/samples/texts/1754951/page_2.md new file mode 100644 index 0000000000000000000000000000000000000000..db3feed9be1c24798a071d05e84049982f411b1d --- /dev/null +++ b/samples/texts/1754951/page_2.md @@ -0,0 +1,13 @@ +Figure 8: Rollback spirals. + +of times it “winds around” $P$ inside the ring (see Fig. 9). At least intuitively, it should be clear that winding numbers and non-rollback spirals are related. In particular, each ring can only have $2^{\mathcal{O}(k)}$ visitors and crossings subpaths (because the size of each separator is $2^{\mathcal{O}(k)}$), and we only have $\mathcal{O}(k)$ rings to deal with. Thus, it is possible to show that if the winding number of every crossing subpath is upper bounded by $2^{\mathcal{O}(k)}$, then the total number of non-rollback spirals is upper bounded by $2^{\mathcal{O}(k)}$ as well. The main tool we employ to bound the winding number of every crossing path is the following known result (rephrased to simplify the overview). + +**Proposition 2.2 ([14]).** Let $G$ be a graph embedded in a ring with a crossing path $P$. Let $\mathcal{P}$ and $Q$ be two collections of vertex-disjoint crossings paths of the same size. (A path in $\mathcal{P}$ can intersect a path in $Q$, but not another path in $\mathcal{P}$.) Then, $G$ has a collection of crossing paths $\mathcal{P}'$ such that (i) for every path in $\mathcal{P}$, there is a path in $\mathcal{P}'$ with the same endpoints and vice versa, and (ii) the maximum difference between (the absolute value of) the winding numbers with respect to $P$ of any path in $\mathcal{P}$ and any path in $Q$ is at most 6. + +To see the utility of Proposition 2.2, suppose momentarily that none of our rings has visitors. Then, if we could ensure that for each of our rings, there is a collection $Q$ of vertex-disjoint paths of maximum size such that the winding number of each path in $Q$ is a constant, Proposition 2.2 would have the following implication: if there is a solution, then we can modify it within each ring to obtain another solution such that each crossing subpath of each of its paths will have a constant winding number (under the supposition that the rings are disjoint, which we will deal with later in the overview), see Fig. 9. Our situation is more complicated due to the existence of visitors—we need to ensure that the replacement $\mathcal{P}'$ does not intersect them. On a high-level, this situation is dealt with by first showing how to ensure that visitors do not “go too deep” into the ring on either side of it. Then, we consider an “inner ring” where visitors do not exist, on which we can apply Proposition 2.2. Afterwards, we are able to bound the winding number of each crossing path by $2^{\mathcal{O}(k)}$ (but not by a constant) in the (normal) ring. + +**Modifying $R$ within Rings.** To ensure the existence of the aforementioned collection $Q$ for each ring, we need to modify $R$. To this end, consider a long path $P$ with separators $S_u$ and $S_v$, and let $\mathcal{P}'$ be the subpath of $P$ inside the ring defined by the two separators. We compute a maximum-sized collection of vertex-disjoint paths $\text{Flow}(u, v)$ such that each of them has one endpoint in $S_u$ and the other in $S_v$.³ Then, we prove a result that roughly states the following. + +**Lemma 2.2.** There is a path $\mathcal{P}^*$ in the ring defined by $S_u$ and $S_v$ with the same endpoints as $\mathcal{P}'$ crossing each path in $\text{Flow}(u, v)$ at most once. Moreover, $\mathcal{P}^*$ is computable in linear time. + +³This flow has an additional property: there is a tight collection of $\mathcal{C}(u, v)$ of concentric cycles separating $S_u$ and $S_v$ such that paths in $\text{Flow}(u, v)$ do not “oscillate” too much between any two cycles in the collection. Such a maximum flow is said to be *minimal* with respect to $\mathcal{C}(u, v)$. \ No newline at end of file diff --git a/samples/texts/1754951/page_20.md b/samples/texts/1754951/page_20.md new file mode 100644 index 0000000000000000000000000000000000000000..3ebc441abe5c4851a32c83e76c8c0dda17617b78 --- /dev/null +++ b/samples/texts/1754951/page_20.md @@ -0,0 +1,27 @@ +than $t$ where $M$ is the union of all connected components of $H' \setminus (Z \cup Z_1)$ having at least one +neighbor in $Z$ and at least one neighbor in $Z_1$. However, $H[V(M) \cap V(H)]$ is a subgraph of +$H$, which means that the treewidth of $H$ is also larger than $t$. By Proposition 3.2, this implies +that the treewidth of $G$ is larger than $2^{ck}$. This contradicts the supposition that $(G, S, T, g, k)$ +is good. From this, we conclude that $\lvert \mathrm{Sep}_{R^2}(P, u) \rvert \le \alpha_{\mathrm{sep}}(k)$. $\square$ + +Recall that $\mathbf{Sep}_{R^2}(P, u)$ is computable in time $\mathcal{O}(n|\mathbf{Sep}_{R^2}(P, u)|)$. Thus, by Lemma 6.4, we obtain the observation below. We remark that the reason we had to argue that the separator is small is not due to this observation, but because the size bound will be crucial in later sections. + +**Observation 6.5.** Let $(G, S, T, g, k)$ be a good instance of *Planar Disjoint Paths*. Let $R^2$ be a Steiner tree that has no detour, $P$ be a long maximal degree-2 path of $R^2$, and $u$ be an endpoint of $P$. Then, $\mathbf{Sep}_{R^2}(P, u)$ can be computed in time $2^{\mathcal{O}(k)n}$. + +Moreover, we have the following immediate consequence of Proposition 3.1. + +**Observation 6.6.** Let $(G, S, T, g, k)$ be a good instance of *Planar Disjoint Paths*. Let $R^2$ be a Steiner tree that has no detour, $P$ be a long maximal degree-2 path of $R^2$, and $u$ be an endpoint of $P$. Then, $H[\mathbf{Sep}_{R^2}(P, u)]$ is a cycle. + +## 6.4 Step IV: Internal Modification of Long Paths + +In this step, we replace the “middle” of each long maximal degree-2 path $P = \text{path}_{R^2}(u, v)$ of $R^2$ by a different path $P^*$. This “middle” is defined by the two separators obtained in the previous step. Let us informally explain the reason behind this modification. In Section 7 we will show that, if the given instance $(G, S, T, g, k)$ admits a solution (which is a collection of disjoint paths connecting $S$ and $T$), then it also admits a “nice” solution that “spirals” only a few times around parts of the constructed Steiner tree. This requirement is crucial, since it is only such solutions $\mathcal{P}$ that are discretely homotopic to weak linkages $\mathcal{W}$ in $H$ aligned with $\mathcal{P}$ that use at most $2^{\mathcal{O}(k)}$ edges parallel to those in $R$, and none of the edges not parallel to those in $R$. To ensure the existence of nice solutions, we show how an arbitrary solution can be rerouted to avoid too many spirals. This rerouting requires a collection of vertex-disjoint paths between $\mathbf{Sep}_{R^2}(P, u)$ and $\mathbf{Sep}_{R^2}(P, v)$ which itself does not spiral around the Steiner tree. The replacement of $P$ by $P^*$ in the Steiner tree, described below, will ensure this property. + +To describe this modification, we first need to assert the statement in the following simple lemma, which partitions every long maximal degree-2 path $P$ of $R^2$ into three parts (see Fig. 14). + +**Lemma 6.5.** Let $(G, S, T, g, k)$ be a good instance of *Planar Disjoint Paths*. Let $R^2$ be a Steiner tree with no detour, and $P$ be a long maximal degree-2 path of $R^2$ with endpoints $u$ and $v$. Then, there exist vertices $u' = u'_P \in \mathbf{Sep}_{R^2}(P, u) \cap V(P)$ and $v' = v'_P \in \mathbf{Sep}_{R^2}(P, v) \cap V(P)$ such that: + +1. The subpath $P_{u,u'}$ of $P$ with endpoints $u$ and $u'$ has no internal vertex from $\mathbf{Sep}_{R^2}(P, u) \cup \mathbf{Sep}_{R^2}(P, v)$, and $\alpha_{\text{pat}}(k)/2 \le |V(P_{u,u'})| \le \alpha_{\text{pat}}(k)$. Additionally, the subpath $P_{v,v'}$ of $P$ with endpoints $v$ and $v'$ has no internal vertex from $\mathbf{Sep}_{R^2}(P, u) \cup \mathbf{Sep}_{R^2}(P, v)$, and $\alpha_{\text{pat}}(k)/2 \le |V(P_{v,v'})| \le \alpha_{\text{pat}}(k)$. + +2. Let $P_{u',v'}$ be the subpath of $P$ with endpoints $u'$ and $v'$. Then, $P = P_{u,u'} - P_{u',v'} - P_{v',v}$. + +*Proof.* We first prove that there exists a vertex $u' \in \text{Sep}_{R^2}(P, u) \cap V(P)$ such that the subpath $P_{u,u'}$ of $P$ between $u$ and $u'$ has no internal vertex from $\text{Sep}_{R^2}(P, u) \cup \text{Sep}_{R^2}(P, v)$. To this end, let $P' = P'_u$, and let $\tilde{P}$ denote the subpath of $P$ that consists of the $\alpha_{\text{pat}}(k)+1$ vertices of $P$ that are closest to $u$. Let $A = A_{R^2,P,u}$ and $B = B_{R^2,P,u}$. Recall that $\text{Sep}_{R^2}(P, u) \subseteq V(H) \setminus (A \cup B)$ separates $A$ and $B$ in $H$. Since $\tilde{P}$ is a path with the endpoint $u$ in $A$ and the other endpoint in $B$, it follows that $\text{Sep}_{R^2}(P, u) \cap V(\tilde{P}) \neq \emptyset$. Accordingly, let $u'$ denote the vertex of $P'$ closest \ No newline at end of file diff --git a/samples/texts/1754951/page_21.md b/samples/texts/1754951/page_21.md new file mode 100644 index 0000000000000000000000000000000000000000..3dede23b7b94f6302023612df330c16bbbe7c4bf --- /dev/null +++ b/samples/texts/1754951/page_21.md @@ -0,0 +1,30 @@ +to $u$ that belongs to $\mathrm{Sep}_{R^2}(P, u)$. Then, $u' \in \mathrm{Sep}_{R^2}(P, u) \cap V(P)$ and the subpath $P_{u,u'}$ of $P$ +between $u$ and $u'$ has no internal vertex from $\mathrm{Sep}_{R^2}(P, u)$. As the number of vertices of $P_{u,u'}$ +is between those of $P''_u$ and $P'$, the inequalities $\alpha_{\mathrm{pat}}(k)/2 \le |V(P_{u,u'})| \le \alpha_{\mathrm{pat}}(k)$ follow. It +remains to argue that $P_{u,u'}$ has no internal vertex from $\mathrm{Sep}_{R^2}(P, v)$. Because $\mathrm{Sep}_{R^2}(P, v) \subseteq +V(H) \setminus (A_{R^2,P,v} \cup B_{R^2,P,v})$, the only vertices of $P$ that $\mathrm{Sep}_{R^2}(P, v)$ can possibly contain are the +$\alpha_{\mathrm{pat}}(k)$ vertices of $P$ that are closest to $v$. Since $P$ is long, none of these vertices belongs to $P'$, +and hence $P_{u,u'}$ (which is a subpath of $P'$) has no internal vertex from $\mathrm{Sep}_{R^2}(P, v)$. + +Symmetrically, we derive the existence of a vertex $v' \in \text{Sep}_{R^2}(P, v) \cap V(P)$ such that the subpath $P_{v,v'}$ of $P$ between $v$ and $v'$ has no internal vertex from $\text{Sep}_{R^2}(P, u) \cup \text{Sep}_{R^2}(P, v)$. + +Lastly, we prove that $P = P_{u,u'} - P_{u',v'} - P_{v',v}$. Since $P_{u,u'}, P_{u',v'}$ and $P_{v',v}$ are subpaths of $P$ +such that $V(P) = V(P_{u,u'}) \cup V(P_{u',v'}) \cup V(P_{v,v'})$, it suffices to show that (i) $V(P_{u,u'}) \cap V(P_{u',v'}) = \{u'\}$, (ii) $V(P_{v,v'}) \cap V(P_{u',v'}) = \{v'\}$, and (iii) $V(P_{u,u'}) \cap V(P_{v,v'}) = \emptyset$. Because $P$ is long and +$|V(P_{u,u'})|, |V(P_{v,v'})| \le \alpha_{\text{pat}}(k)$, it is immediate that item (iii) holds. For item (i), note that +$V(P_{u,u'}) \cap V(P_{u',v'})$ can be a strict superset of $\{u'\}$ only if $P_{u',v'}$ is a subpath of $P_{u,u'}$; then, +$v' \in V(P_{u,u'})$, which means that $P_{u,u'}$ has an internal vertex from $\text{Sep}_{R^2}(P, v)$ and results in a +contradiction. Thus, item (i) holds. Symmetrically, item (ii) holds as well. $\square$ + +In what follows, when we use the notation $u'_p$, we refer to the vertex in Lemma 6.5. Before +we describe the modification, we need to introduce another notation and make an immediate +observation based on this notation. + +**Definition 6.4** ($\tilde{A}_{R^2,P,u}$). Let $(G, S, T, g, k)$ be a good instance of Planar Disjoint Paths. Let $R^2$ be a Steiner tree that has no detour, $P$ be a long maximal degree-2 path of $R^2$, and $u$ be an endpoint of $P$. Then, $\tilde{A}_{R^2,P,u} = (V(P_{u,u'_p}) \setminus \{u'_p\}) \cup A_{R^2,p,u}$. + +**Observation 6.7.** Let $(G, S, T, g, k)$ be a good instance of Planar Disjoint Paths. Let $R^2$ be a Steiner tree that has no detour, and $P$ be a long maximal degree-2 path of $R^2$ with endpoints $u$ and $v$. Then, there exists a single connected component $C_{R^2,P,u}$ in $H - (\text{Sep}_{R^2}(u, P) \cup \text{Sep}_{R^2}(v, P))$ that contains $\tilde{A}_{R^2,P,u}$ and a different single connected component $C_{R^2,P,v}$ in $H - (\text{Sep}_{R^2}(u, P) \cup \text{Sep}_{R^2}(v, P))$ that contains $\tilde{A}_{R^2,P,v}$. + +We proceed to describe the modification. For brevity, let $S_u = \text{Sep}_{R^2}(P, u)$ and $S_v = \text{Sep}_{R^2}(P, v)$. Recall that there is a terminal $t^* \in T$ that lies on the outer face of $H$ (and $G$). By Observation 6.6, $S_v$ and $S_v$ induce two cycles in $H$, and $t^*$ lies in the exterior of both these cycles. Assume w.l.o.g. that $u$ lies in the interior of both $S_u$ and $S_v$, while $v$ lies in the exterior of both $S_u$ and $S_v$. Then, $S_u$ belongs to the strict interior of $S_v$. We construct a sequence of concentric cycles between $S_u$ and $S_v$ as follows. + +**Lemma 6.6.** Let $(G, S, T, g, k)$ be a good instance of Planar Disjoint Paths. Let $R^2$ be a Steiner tree with no detour, and $P$ be a long maximal degree-2 path of $R^2$ with endpoints $u$ and $v$. Let $S_u = \text{Sep}_{R^2}(P, u)$ and $S_v = \text{Sep}_{R^2}(P, v)$, where $S_u$ lies in the strict interior of $S_v$. Then, there is a sequence of concentric cycles $C(u, v) = (C_1, C_2, \dots, C_p)$ in $G$ of length $p \ge 100\alpha_{\text{sep}}(k)$ such that $S_u$ is in the strict interior of $C_1$ in $H$, $S_v$ is in the strict exterior of $C_p$ in $H$, and there is a path $\eta$ in $H$ with one endpoint $v_0 \in S_u$ and the other endpoint $v_{p+1} \in S_v$, such that the intersection of $V(\eta)$ with $V(G) \cup S_u \cup S_v$ is $\{v_0, v_1, \dots, v_{p+1}\}$ for some $v_i \in V(C_i)$ for every $i \in \{1, \dots, p\}$. Furthermore, $C(u, v)$ can be computed in linear time. + +*Proof.* Towards the computation of *C*(u, v), delete all vertices that lie in the strict interior of Su or in the strict exterior of Sv, as well as all vertices of V(H) \ (V(G) ∪ Su ∪ Sv). Denote the resulting graph by G+u,v, and note that it has a plane embedding in the “ring” defined by H[Su] and H[Sv]. Observe that Su, Sv ⊆ V(G+u,v), where Sv defines the outer face of the embedding of G+u,v. Thus, any cycle in this graph that separates Su and Sv must contain Su in its interior and Sv in its exterior. Furthermore, Gu,v = G+u,v − V(H) is an induced subgraph of G, which \ No newline at end of file diff --git a/samples/texts/1754951/page_22.md b/samples/texts/1754951/page_22.md new file mode 100644 index 0000000000000000000000000000000000000000..14a15d4aa08956a2da64a6a4cad32870d006bf82 --- /dev/null +++ b/samples/texts/1754951/page_22.md @@ -0,0 +1,17 @@ +consists of all vertices of $G$ that, in $H$, lie in the strict exterior of $S_u$ and in the strict interior of $S_v$ simultaneously, or lie in $S_v \cup S_v$. In particular, any cycle of $G_{u,v}$ is also a cycle in $G$. + +Now, $\mathcal{C}(u, v)$ is computed as follows. Start with an empty sequence, and the graph $G_{u,v} - (S_u \cup S_v)$. As long as there is a cycle in the current graph such that all vertices of $S_u$ are in the strict interior of $C$ with respect to $H$, remove vertices of degree at most 1 in the current graph until no such vertices remain, and append the outer face of the current graph as a cycle to the constructed sequence. It is clear that this process terminates in linear time, and that by the above discussion, it constructs a sequence of concentric cycles $\mathcal{C}(u, v) = (C_1, C_2, \dots, C_p)$ in $G$ such that $S_u$ is in the strict interior of $C_1$ in $H$, $S_v$ is in the strict exterior of $C_p$ in $H$. + +To assert the existence of a path $\eta$ in $H$ with one endpoint $v_0 \in S_u$ and the other endpoint $v_{p+1} \in S_v$, such that the intersection of $V(\eta)$ with $V(G) \cup S_u \cup S_v$ is $\{v_0, v_1, \dots, v_{p+1}\}$ for some $v_i \in V(C_i)$ for every $i \in \{1, \dots, p\}$, we require the following claim. + +**Claim 6.1.** Let $C_{p+1} = H[S_v]$. For every $i \in \{1, 2, \dots, p\}$ and every vertex $w \in V(C_i)$, there exists a vertex $w' \in V(C_{i+1})$ such that $w$ and $w'$ lie on a common face in $G_{u,v}^+$. Moreover, there exist vertices $w \in S_u$ and $w' \in V(C_1)$ that lie on a common face in $G_{u,v}^+$. + +*Proof of Claim 6.1.* Consider $i \in \{1, 2, \dots, p\}$ and a vertex $w \in V(C_i)$. We claim that $\mathrm{rdist}(w, V(C_{i+1})) \le 1$, i.e. there must be $w' \in V(C_{i+1})$ such that $w, w'$ have a common face in $G_{u,v}^+$. By way of contradiction, suppose that $\mathrm{rdist}(w, V(C_{i+1})) \ge 2$. Then, by Proposition 6.1, there is a cycle $C$ that separates $w$ and $V(C_{i+1})$ in $G_{u,v}^+$ such that $\mathrm{rdist}(w, w'') = 1$ for every vertex $w'' \in V(C)$. Here, $w$ lies in the strict interior of $C$, and $C$ lies in the strict interior of $C_{i+1}$. Further, $C$ is vertex disjoint from $C_{i+1}$, since $\mathrm{rdist}(w, w') \ge 2$ for every $w' \in V(C_{i+1})$. Now, consider the outer face of $G[V(C_i) \cup V(C)]$. By the construction of $C_i$, this outer face must be $C_i$. However, $w \in V(C_i)$ cannot belong to it, hence we reach a contradiction. + +For the second part, we claim that $\mathrm{rdist}(S_u, V(C_1)) \le 1$, i.e. there must be $w \in S_u$ and $w' \in V(C_1)$ such that $w, w'$ have a common face in $G_{u,v}^+$. By way of contradiction, suppose that $\mathrm{rdist}(S_u, V(C_1)) \ge 2$. Then, by Proposition 6.1, there is a cycle $C$ that separates $S_u$ and $V(C_1)$ in $G_{u,v}^+$ such that $\mathrm{rdist}(S_u, w'') = 1$ for every vertex $w'' \in V(C)$. Further, $C$ is vertex disjoint from $C_1$, since $\mathrm{rdist}(S_u, w') \ge 2$ for every $w' \in V(C_1)$. However, this is a contradiction to the termination condition of the construction of $\mathcal{C}(u,v). ◦$ + +Having this claim, we construct $\eta$ as follows. Pick vertices $v_0 \in S_u$ and $v_1 \in V(C_1)$ that lie on a common face in $G_{u,v}^+$. Then, for every $i \in \{2, \dots, p+1\}$, pick a vertex $v_i \in V(C_i)$ such that $v_{i-1}$ and $v_i$ lie on a common face in $G_{u,v}^+$. Thus, for every $i \in \{0, 1, \dots, p\}$, we have that $v_i$ and $v_{i+1}$ are either adjacent in $H$ or there exists a vertex $u_i \in V(H) \setminus V(G)$ such that $u_i$ is adjacent to both $v_i$ and $v_{i+1}$. Because $\mathcal{C}(u,v) = (C_1, C_2, \dots, C_p)$ is a sequence of concentric cycles in $G$ such that $S_u$ is in the strict interior of $C_1$ and $S_v$ is in the strict exterior of $C_p$, the $u_i$'s are distinct. Thus, $\eta = v_0 - u_0 - v_1 - u_1 - v_2 - u_2 - \dots - v_p - u_p - v_{p+1}$, where undefined $u_i$'s are dropped, is a path as required. + +Finally, we argue that $p \ge 100 \cdot \alpha_{\mathrm{sep}}(k)$. Note that $100\alpha_{\mathrm{sep}}(k) = 100(\frac{7}{2} \cdot 2^{ck} + 2) \le 400 \cdot 2^{ck}$, thus it suffices to show that $p \ge 400 \cdot 2^{ck}$. To this end, we obtain a lower bound on the radial distance between $S_u$ and $S_v$ in $G_{u,v}^+$. Recall that $|S_u|, |S_v| \le \alpha_{\mathrm{sep}}(k) = \frac{7}{2} \cdot 2^{ck} + 2 \le 4 \cdot 2^{ck}$. Let $P = \mathrm{path}_{R^2}(u,v)$, and recall that its length is at least $\alpha_{\mathrm{long}}(k) = 10^4 2^{ck}$. Since $R^2$ has not detour, $P$ is a shortest path in $H$ between $u$ and $v$, thus for any two vertices in $V(P)$, the subpath of $P$ between them is a shortest path between them. Now, recall the vertices $u' = u'_P$, $v' = v'_P$ (defined in Lemma 6.5), and denote the subpath between them by $P'$. By construction, $|E(P')| \ge \alpha_{\mathrm{long}}(k) - 2 \cdot \alpha_{\mathrm{pat}}(k) = (10^4 - 200) \cdot 2^{ck}$. + +We claim that the radial distance between $S_u$ and $S_v$ in $G_{u,v}^+$ is at least $|E(P')|/2 - |S_u| - |S_v|$. Suppose not, and consider a sequence of vertices in $G_{u,v}^+$ that witnesses this fact: $x_1, x_2, x_3, \dots, x_{p-1}, x_p \in V(G_{u,v}^+)$ where $x_1 \in S_u$, $x_p \in S_v$, $p < |E(P')|/2 - |S_v| - |S_u|$, and every two consecutive vertices lie on a common face. Consider a shortest such sequence, which visits each face of $G_{u,v}^+$ at most once. In particular, $x_1$ and $x_p$ are the only vertices of $S_u \cup S_v$. \ No newline at end of file diff --git a/samples/texts/1754951/page_23.md b/samples/texts/1754951/page_23.md new file mode 100644 index 0000000000000000000000000000000000000000..10bb6138adce897e98dbe3c1243efb416628de83 --- /dev/null +++ b/samples/texts/1754951/page_23.md @@ -0,0 +1,15 @@ +structural results, such as the algorithm for Disjoint Paths [37], Minor Testing [37] (given two undirected graphs, *G* and *H* on *n* and *k* vertices, respectively, the goal is to check whether *G* contains *H* as a minor), the structural decomposition [38] and the Excluded Grid Theorem [40]. Unfortunately, all of these results suffer from such bad hidden constants and dependence on the parameter *k* that they have gotten their own term–“galactic algorithms” [31]. + +It is the hope of many researchers that, in time, algorithms and structural results from Graph Minors can be more algorithmically efficient, perhaps even practically applicable. Substantial progress has been made in this direction, examples include the simpler decomposition theorem of Kawarabayashi and Wollan [28], the faster algorithm for computing the structural decomposition of Grohe et al. [22], the improved unique linkage theorem of Kawarabayashi and Wollan [27], the linear excluded grid theorem on minor free classes of Demaine and Hajiaghayi [16], paving the way for the theory of Bidimensionality [15], and the polynomial grid minor theorem of Chekuri and Chuzhoy [6]. The algorithm for Disjoint Paths is a cornerstone of the entire Graph Minor Theory, and a vital ingredient in the $g(k)n^3$-time algorithm for Minor Testing. Therefore, efficient algorithms for Disjoint Paths and Minor Testing are necessary and crucial ingredients in an algorithmically efficient Graph Minors theory. This makes obtaining $2^{\text{poly}(k)}n^{\mathcal{O}(1)}$ time algorithms for Disjoint Paths and Minor Testing a tantalizing and challenging goal. + +Theorem 1.1 is a necessary basic step towards achieving this goal—a $2^{\text{poly}(k)}n^{\mathcal{O}(1)}$ time algorithms for Disjoint Paths on general graphs also has to handle planar inputs, and it is easy to give a reduction from Planar Disjoint Paths to Minor Testing in such a way that a $2^{\text{poly}(k)}n^{\mathcal{O}(1)}$ time algorithm for Minor Testing would imply a $2^{\text{poly}(k)}n^{\mathcal{O}(1)}$ time algorithms for Planar Disjoint Paths. In addition to being a necessary step in the formal sense, there is strong evidence that an efficient algorithm for the planar case will be useful for the general case as well—indeed the algorithm for Disjoint Paths of Robertson and Seymour [37] relies on topology and essentially reduces the problem to surface-embedded graphs. Thus, an efficient algorithm for Planar Disjoint Paths represents a speed-up of the base case of the algorithm for Disjoint Paths of Robertson and Seymour. Coupled with the other recent advances [6, 15, 16, 22, 27, 28], this gives some hope that $2^{\text{poly}(k)}n^{\mathcal{O}(1)}$ time algorithms for Disjoint Paths and Minor Testing may be within reach. + +**Known Techniques and Obstacles in Designing a 2poly(k) Algorithm.** All known algorithms for both Disjoint Paths and Planar Disjoint Paths have the same high level structure. In particular, given a graph G we distinguish between the cases of G having “small” or “large” treewidth. In case the treewidth is large, we distinguish between two further cases: either G contains a “large” clique minor or it does not. This results in the following case distinctions. + +1. **Treewidth is small.** Let the treewidth of *G* be *w*. Then, we use the known dynamic programming algorithm with running time $2^{\mathcal{O}(w \log w)} n^{\mathcal{O}(1)}$ [41] to solve the problem. It is important to note that, assuming the Exponential Time Hypothesis (ETH), there is no algorithm for Disjoint Paths running in time $2^{\mathcal{O}(w \log w)} n^{\mathcal{O}(1)}$ [32], nor an algorithm for Planar Disjoint Paths running in time $2^{\mathcal{O}(w)} n^{\mathcal{O}(1)}$ [4]. + +2. **Treewidth is large and *G* has a large clique minor.** In this case, we use the good routing property of the clique to find an irrelevant vertex and delete it without changing the answer to the problem. Since this case will not arise for graphs embedded on a surface or for planar graphs, we do not discuss it in more detail. + +3. **Treewidth is large and *G* has no large clique minor.** Using a fundamental structure theorem for minors called the “flat wall theorem”, we can conclude that *G* contains a large planar piece of the graph and a vertex *v* that is sufficiently insulated in the middle of it. Applying the unique linkage theorem [39] to this vertex, we conclude that it is irrelevant and remove it. For planar graphs, one can use the unique linkage theorem of Adler et al. [2]. In particular, we use the following result: + +Any instance of Disjoint Paths consisting of a planar graph with treewidth at least $82k^{3/2}2^k$ and $k$ terminal pairs contains a vertex $v$ such that every solution \ No newline at end of file diff --git a/samples/texts/1754951/page_24.md b/samples/texts/1754951/page_24.md new file mode 100644 index 0000000000000000000000000000000000000000..20ab4d7af41d5c0783dac2149dab5ce29441e0e7 --- /dev/null +++ b/samples/texts/1754951/page_24.md @@ -0,0 +1,23 @@ +in this sequence. Then, we can extend this sequence on both sides to derive another sequence of vertices of $G_{u,v}^+$ starting at $u'$ and ending at $v'$ such that the prefix of the new sequence is a path in $G_{u,v}^+[S_u]$ from $u'$ to $x_1$, the midfix is $x_1, x_2, x_3, \dots, x_{p-1}, x_p$, and the suffix is a path in $G_{u,v}^+[S_v]$ from $x_p$ to $v'$. Further, the length of the new sequence of vertices is smaller than $|E(P')|/2$. Hence, the radial distance between $u'$ and $v'$ in $H' = H[V(G) \cup S_u \cup S_v]$ (the graph derived from $G_{u,v}^+$ by reintroducing the vertices of $G$ that lie inside $S_u$ or outside $S_v$) is smaller than $|E(P')|/2$, and let it be witnessed by a sequence $Q[u', v']$. As $H$ is the radial completion of $G$, observe that $Q[u', v']$ gives rise to a path $Q$ in $H$ between $u'$ and $v'$ of length smaller than $|E(P')|$. However, then $u', v'$ and $Q$ witness a detour in $\mathbb{R}^2$, which is a contradiction. Hence, the radial distance between $S_u$ and $S_v$ in $G_{u,v}^+$ is at least + +$$ +\begin{aligned} +& \frac{(10^4 - 200)}{2} \cdot 2^{ck} - (|S_u| + |S_v|) \\ +& \geq 4900 \cdot 2^{ck} - 2\alpha_{\text{sep}}(k) \\ +& = 4900 \cdot 2^{ck} - (7 \cdot 2^{ck} + 4) \geq 400 \cdot 2^{ck}. +\end{aligned} +$$ + +Now, observe that $S_u$ and $S_v$ are connected sets in $G_{u,v}^+$ and $S_v$ forms the outer-face of $G_{u,v}^+$. Then, by Proposition 6.1, we obtain a collection of at least $400 \cdot 2^{ck}$ disjoint cycles in $G_{u,v}^+$, where each cycle separates $S_u$ and $S_v$. Note that these cycles are disjoint from $S_u \cup S_v$, and hence they lie in $G$. Moreover, each of them contains $S_u$ in its strict interior, and $S_v$ in its strict exterior. Thus, it is clear that the sequence $\mathcal{C}(u,v)$ computed above must contain at least $400 \cdot 2^{ck} > 100\alpha_{\text{sep}}(k)$ cycles. $\square$ + +Recall the graph $G_{u,v}$, which is an induced subgraph of $G$, which consists of all vertices of $G$ that, in $H$, lie in the strict exterior of $S_u$ and in the strict interior of $S_v$ simultaneously, or lie in $S_u \cup S_v$. With $\mathcal{C}(u,v)$ at hand, we compute a maximum size collection of disjoint paths from $S_u$ to $S_v$ in $G_{u,v}$ that minimizes the number of edges it traverses outside $E(\mathcal{C}(u,v))$. In this observation, the implicit assumption that $\ell \le \alpha_{\text{sep}}(k)$ is justified by Lemma 6.4. + +**Observation 6.8.** Let the maximum flow between $S_u \cap V(G)$ and $S_v \cap V(G)$ in $G_{u,v}$ be $\ell \le \alpha_{\text{sep}}(k)$. Given the sequence $\mathcal{C}(u,v)$ of Lemma 6.6, a collection $\text{Flow}_{\mathbb{R}^2}(u,v)$ of $\ell$ vertex-disjoint paths in $G_{u,v}$ from $S_u \cap V(G)$ to $S_v \cap V(G)$ that minimizes $|E(\text{Flow}_{\mathbb{R}^2}(u,v)) \setminus E(\mathcal{C}(u,v))|$ is computable in time $\mathcal{O}(n^{3/2}\log^3 n)$. + +*Proof.* We determine $\ell \le \alpha_{\text{sep}}(k)$ in time $2^{\mathcal{O}(k)n}$ by using Ford-Fulkerson algorithm. Next, we define a weight function $\bar{w}$ on $E(G_{u,v})$ as follows: + +$$ w(e) = \begin{cases} 0 & \text{if } e \in E(\mathcal{C}(u,v)) \\ 1 & \text{otherwise} \end{cases} $$ + +We now compute a minimum cost flow between $S_u \cap V(G)$ and $S_v \cap V(G)$ of value $\ell$ in $G_{u,v}$ under the weight function $\bar{w}$. This can be done in time $\mathcal{O}(n^{3/2}\log^3 n)$ by [13, Theorem 1], as the cost of such a flow is bounded by $\mathcal{O}(n)$. Clearly, the result is a collection $\text{Flow}_{\mathbb{R}^2}(u,v)$ of $\ell$ vertex-disjoint paths from $S_u \cap V(G)$ to $S_v \in V(G)$ minimizing $|E(\text{Flow}_{\mathbb{R}^2}(u,v)) \setminus E(\mathcal{C}(u,v))|$. $\square$ + +Having $\text{Flow}_{\mathbb{R}^2}(u,v)$ at hand, we proceed to find a certain path between $u'_P$ and $v'_P$ that will be used to replace $P_{u'_P, v'_P}$. We remark that all vertices and edges of $\text{Flow}_{\mathbb{R}^2}(u,v)$ lie between $S_u$ and $S_v$ in the plane embedding of $H$. The definition of this path is given by the following lemma, and the construction will make it intuitively clear that the paths in $\text{Flow}_{\mathbb{R}^2}(u,v)$ do not “spiral around” the Steiner tree once we replace $P_{u'_P, v'_P}$ with $P_{u'_P, v'_P}^*$. Note that in the lemma, we consider a path $\bar{P}$ in $H$, while the paths in $\text{Flow}_{\mathbb{R}^2}(u,v)$ are in $G$. \ No newline at end of file diff --git a/samples/texts/1754951/page_25.md b/samples/texts/1754951/page_25.md new file mode 100644 index 0000000000000000000000000000000000000000..d21b4c285daaa515423de0d5ef0829ecc1413249 --- /dev/null +++ b/samples/texts/1754951/page_25.md @@ -0,0 +1,17 @@ +**Lemma 6.7.** Let $(G, S, T, g, k)$ be a good instance of Planar Disjoint Paths. Let $\mathbb{R}^2$ be a Steiner tree with no detour, and $P$ be a long maximal degree-2 path of $\mathbb{R}^2$, with endpoints $u$ and $v$. Let $u' = u'_P$ and $v' = v'_P$. Then, there exists a path $P_{u',v'}^*$ in $H - (V(C_{R^2,P,u}) \cup V(C_{R^2,P,u}))$ between $u'$ and $v'$ with the following property: there do not exist three vertices $x, y, z \in V(P_{u',v'}^*)$ such that (i) $\text{dist}_{P_{u',v'}^*}(u', x) < \text{dist}_{P_{u',v'}^*}(u', y) < \text{dist}_{P_{u',v'}^*}(u', z)$, and (ii) there exist a path in $\text{Flow}_{\mathbb{R}^2}(u, v)$ that contains $x$ and $z$ and a different path in $\text{Flow}_{\mathbb{R}^2}(u, v)$ that contains $y$. Moreover, such a path $P_{u',v'}^*$ can be computed in time $\mathcal{O}(n)$. + +*Proof.* Let $P_{u',v'}^*$ be a path in $H - (V(C_{\mathbb{R}^2,P,u}) \cup V(C_{\mathbb{R}^2,P,u}))$ between $u'$ and $v'$ that minimizes the number of paths $Q \in \text{Flow}_{\mathbb{R}^2}(u,v)$ for which there exist at least one triple $x, y, z \in V(P_{u',v'}^*)$ that has the two properties in the lemma and $x, z \in V(Q)$. Due to the existence of $P_{u',v'}$, such a path $P_{u',v'}^*$ exists. We claim that this path $P_{u',v'}^*$ has no triple $x, y, z \in V(P_{u',v'}^*)$ that has the two properties in the lemma. Suppose, by way of contradiction, that our claim is false, and let $x, y, z \in V(P_{u',v'}^*)$ be a triple that has the two properties in the lemma. Note that, when traversed from $u'$ to $v'$, $P_{u',v'}^*$ first visits $x$, then visits $y$ and afterwards visits $z$. Let $Q'$ be the path in $\text{Flow}_{\mathbb{R}^2}(u,v)$ that contains $x$ and $z$. Let $x'$ and $z'$ be the first and last vertices of $Q'$ that are visited by $P_{u',v'}^*$. Then, replace the subpath of $P_{u',v'}^*$ between $x'$ and $z'$ by the subpath of $Q'$ between $x'$ and $z'$. This way we obtain a path $P'$ in $H - (V(C_{\mathbb{R}^2,P,u}) \cup V(C_{\mathbb{R}^2,P,u}))$ between $u'$ and $v'$ for which there exist fewer paths $Q \in \text{Flow}_{\mathbb{R}^2}(u,v)$, when compared to $P_{u',v'}^*$, for which there exists at least one triple $x, y, z \in V(P')$ that has the two properties in the lemma and such that $x, z \in V(Q)$. As we have reached a contradiction, we conclude that our initial claim is correct. While the proof is existential, it can clearly be turned into a linear-time algorithm. $\square$ + +The following is a direct corollary of the above lemma. + +**Corollary 6.1.** For each path $Q \in \text{Flow}_{\mathbb{R}^2}(u,v)$, there are at most two edges in $P_{u',v'}^*$ such that one endpoint of the edge lies in $V(Q)$ and the other lies in $V(G) \setminus V(Q)$. + +Having Lemma 6.7 at hand, we modify $\mathbb{R}^2$ as follows: for every long maximal degree-2 path $P$ of $\mathbb{R}^2$ with endpoints $u$ and $v$, replace $P_{u'_P, v'_P}$ by $P_{u'_P, v'_P}^*$. Denote the result of this modification by $\mathbb{R} = \mathbb{R}^3$. We refer to $\mathbb{R}^3$ as a *backbone Steiner tree*. Let us remark that the backbone Steiner tree $\mathbb{R}^3$ is always accompanied by the separators $\text{Sep}_{\mathbb{R}^2}(P, u)$ and $\text{Sep}_{\mathbb{R}^2}(P, v)$, and the collection $\text{Flow}_{\mathbb{R}^2}(u, v)$ for every long maximal degree-2 path $\mathcal{P} = \text{path}_{\mathbb{R}^2}(u, v)$ of $\mathbb{R}^2$. These separators and flows will play crucial role in our algorithm. In the following subsection, we will prove that $\mathbb{R}^3$ is indeed a Steiner tree, and in particular it is a tree. Additionally and crucially, we will prove that the separators computed previously remain separators. Let us first conclude the computational part by stating the running time spent so far. From Lemma 6.3, Observations 6.5 and 6.8, and by Lemma 6.7, we have the following result. + +**Lemma 6.8.** Let $(G, S, T, g, k)$ be a good instance of Planar Disjoint Paths. Then, a backbone Steiner tree $\mathcal{R}$ can be computed in time $2^{\mathcal{O}(k)n^{3/2}\log^3 n}$. + +## 6.5 Analysis of $\mathbb{R}^3$ and the Separators $\text{Sep}_{\mathbb{R}^2}(P, u)$ + +Having constructed the backbone Steiner tree $\mathbb{R}^3$, we turn to analyse its properties. Among other properties, we show that useful properties of $\mathbb{R}^2$ also transfer to $\mathbb{R}^3$. We begin by proving that the two separators of each long maximal degree-2 path $P$ of $\mathbb{R}^2$ partition $V(\mathcal{H})$ into five “regions”, and that the vertices in each region are all close to the subtree of $\mathcal{R}$ that (roughly) belongs to that region. Specifically, the regions are $V(C_{\mathbb{R}^2,P,u})$, $\text{Sep}_{\mathbb{R}^2}(P, u)$, $V(C_{\mathbb{R}^2,P,v})$, $\text{Sep}_{\mathbb{R}^2}(P, v)$, and $V(\mathcal{H}) \setminus (V(C_{\mathbb{R}^2,P,u}) \cup V(C_{\mathbb{R}^2,P,v}) \cup \text{Sep}_{\mathbb{R}^2}(P, u) \cup \text{Sep}_{\mathbb{R}^2}(P, v))$, and our claim is as follows. + +**Lemma 6.9.** Let $(G, S, T, g, k)$ be a good instance of Planar Disjoint Paths. Let $\mathbb{R}^2$ be a Steiner tree that has no detour, $P$ be a long maximal degree-2 path of $\mathbb{R}^2$, and $u$ and $v$ be its endpoints. \ No newline at end of file diff --git a/samples/texts/1754951/page_26.md b/samples/texts/1754951/page_26.md new file mode 100644 index 0000000000000000000000000000000000000000..a242bf6799cc7cf906a9e51204f928597701abc1 --- /dev/null +++ b/samples/texts/1754951/page_26.md @@ -0,0 +1,29 @@ +1. For all $w \in \text{Sep}_{R^2}(P, u)$, it holds that $\text{dist}_H(w, u'_P) \le \alpha_{\text{sep}}(k)$. + +2. For all $w \in \text{Sep}_{R^2}(P, v)$, it holds that $\text{dist}_H(w, v'_P) \le \alpha_{\text{sep}}(k)$. + +3. For all $w \in V(C_{R^2,P,u})$, it holds that $\text{dist}_H(w, \tilde{A}_{R^2,P,u} \cup \{u'_P\}) \le \alpha_{\text{dist}}(k) + \alpha_{\text{sep}}(k)$. + +4. For all $w \in V(C_{R^2,P,v})$, it holds that $\text{dist}_H(w, \tilde{A}_{R^2,P,v} \cup \{v'_P\}) \le \alpha_{\text{dist}}(k) + \alpha_{\text{sep}}(k)$. + +5. For all $w \in V(H) \setminus (V(C_{R^2,P,u}) \cup V(C_{R^2,P,v}) \cup \text{Sep}_{R^2}(P, u) \cup \text{Sep}_{R^2}(P, v))$, it holds that $\text{dist}_H(w, V(R^2) \setminus (\tilde{A}_{R^2,P,u} \cup \tilde{A}_{R^2,P,v})) \le \alpha_{\text{dist}}(k) + \alpha_{\text{sep}}(k)$. + +*Proof.* First, note that Conditions 1 and 2 follow directly from Lemma 6.4 and Observation 6.6. + +For Condition 3, consider some vertex $w \in V(C_{R^2,P,u})$. By Lemma 6.1, $\text{dist}_H(w, V(R^2)) \le \alpha_{\text{dist}}(k)$. Thus, there exists a path $Q$ in $H$ with $w$ as one endpoint and the other endpoint $x$ in $V(R^2)$ such that the length of $Q$ is at most $\alpha_{\text{dist}}(k)$. In case $x \in \tilde{A}_{R^2,P,u} \cup \{u'_P\}$, we have that $\text{dist}_H(w, \tilde{A}_{R^2,P,u}) \le \alpha_{\text{dist}}(k)$, and hence the condition holds. Otherwise, by the definition of $\text{Sep}_{R^2}(P, u)$, the path $Q$ must traverse at least one vertex from $\text{Sep}_{R^2}(P, u)$. Thus, $\text{dist}_H(w, \text{Sep}_{R^2}(P, u)) \le \alpha_{\text{dist}}(k)$. Combined with Condition 1, we derive that $\text{dist}_H(w, \tilde{A}_{R^2,P,u} \cup \{u'_P\}) \le \alpha_{\text{dist}}(k) + \alpha_{\text{sep}}(k)$. The proof of Condition 4 is symmetric. + +The proof of Condition 5 is similar. Consider some vertex $w \in V(H) \setminus (V(C_{R^2,P,u}) \cup V(C_{R^2,P,v}) \cup \text{Sep}_{R^2}(P, u) \cup \text{Sep}_{R^2}(P, v))$. As before, there exists a path $Q$ in $H$ with $w$ as one endpoint and the other endpoint $x$ in $V(R^2)$ such that the length of $Q$ is at most $\alpha_{\text{dist}}(k)$. In case $x \in \tilde{A}_{R^2,P,u} \cup \{u'_P\}$, we are done. Otherwise, the path $Q$ must traverse at least one vertex from $\text{Sep}_{R^2}(P, u) \cup \text{Sep}_{R^2}(P, v)$. Specifically, if $x \in V(C_{R^2,P,u})$, then it must traverse at least one vertex from $\text{Sep}_{R^2}(P, u)$, and otherwise $x \in V(C_{R^2,P,v})$ and it must traverse at least one vertex from $\text{Sep}_{R^2}(P, v)$. Combined with Conditions 1 and 2, we derive that $\text{dist}_H(w, V(R^2) \setminus (\tilde{A}_{R^2,P,u} \cup \tilde{A}_{R^2,P,v})) \le \alpha_{\text{dist}}(k) + \alpha_{\text{sep}}(k)$. $\square$ + +An immediate corollary of Lemma 6.9 concerns the connectivity of the “middle region” as follows. (This corollary can also be easily proved directly.) + +**Corollary 6.2.** Let $(G, S, T, g, k)$ be a good instance of Planar Disjoint Paths. Let $R^2$ be a Steiner tree that has no detour, $P$ be a long maximal degree-2 path of $R^2$, and $u$ and $v$ be its endpoints. Then, $H[V(H) \setminus (V(C_{R^2,P,u}) \cup V(C_{R^2,P,v}))]$ is a connected graph. + +*Proof.* By Lemma 6.9 and the definition of $\text{Sep}_{R^2}(P, u)$ and $\text{Sep}_{R^2}(P, v)$, for every vertex in $V(H) \setminus (V(C_{R^2,P,u}) \cup V(C_{R^2,P,v}))$, the graph $H$ has a path from that vertex to some vertex in $\text{Sep}_{R^2}(P, u) \cup \text{Sep}_{R^2}(P, v)$ that lies entirely in $H[V(H) \setminus (V(C_{R^2,P,u}) \cup V(C_{R^2,P,v}))]$. Thus, the corollary follows from Observation 6.6. $\square$ + +Next, we utilize Lemma 6.9 and Corollary 6.2 to argue that the “middle regions” of different long maximal degree-2 paths of $R^2$ are distinct. Recall that we wish to reroute a given solution to be a solution that “spirals” only a few times around the Steiner tree. This lemma allows us to independently reroute the solution in each of these “middle regions”. In fact, we prove the following stronger statement concerning these regions. The idea behind the proof of this lemma is that if it were false, then $R^2$ admits a detour, which is contradiction. + +**Lemma 6.10.** Let $(G, S, T, g, k)$ be a good instance of Planar Disjoint Paths. Let $R^2$ be a Steiner tree that has no detour. Additionally, let $P$ and $\hat{P}$ be two distinct long maximal degree-2 paths of $R^2$. Let $u$ and $v$ be the endpoints of $P$, and $\hat{u}$ and $\hat{v}$ be the endpoints of $\hat{P}$. Then, one of the two following conditions holds: + +• $V(H) \setminus (V(C_{R^2,P,u}) \cup V(C_{R^2,P,v})) \subseteq V(C_{R^2,\hat{P},\hat{u}}).$ + +• $V(H) \setminus (V(C_{R^2,P,u}) \cup V(C_{R^2,P,v})) \subseteq V(C_{R^2,\hat{P},\hat{v}}).$ \ No newline at end of file diff --git a/samples/texts/1754951/page_27.md b/samples/texts/1754951/page_27.md new file mode 100644 index 0000000000000000000000000000000000000000..a4eec69c6f8549b2ef481da3ef093340fe7dfa97 --- /dev/null +++ b/samples/texts/1754951/page_27.md @@ -0,0 +1,33 @@ +Figure 15: Illustration of Lemma 6.10. + +*Proof.* We first prove that the intersection of $V(H) \setminus (V(C_{R^2,P,u}) \cup V(C_{R^2,P,v}))$ with $V(H) \setminus (V(C_{R^2,\hat{P},\hat{u}}) \cup V(C_{R^2,\hat{P},\hat{v}}))$ is empty. To this end, suppose by way of contradiction that there exists a vertex $w$ in this intersection. By Lemma 6.9, the inclusion of $w$ in both $V(H) \setminus (V(C_{R^2,P,u}) \cup V(C_{R^2,P,v}))$ and $V(H) \setminus (V(C_{R^2,\hat{P},\hat{u}}) \cup V(C_{R^2,\hat{P},\hat{v}}))$ implies that the two following inequalities are satisfied: + +• $\mathrm{dist}_H(w, V(R^2) \setminus (\tilde{A}_{R^2,P,u} \cup \tilde{A}_{R^2,P,v})) \leq \alpha_{\mathrm{dist}}(k) + \alpha_{\mathrm{sep}}(k).$ + +• $\mathrm{dist}_H(w, V(R^2) \setminus (\tilde{A}_{R^2,\hat{P},\hat{u}} \cup \tilde{A}_{R^2,\hat{P},\hat{v}})) \leq \alpha_{\mathrm{dist}}(k) + \alpha_{\mathrm{sep}}(k).$ + +From this, we derive the following inequality: + +$$ +\mathrm{dist}_H(V(R^2) \setminus (\tilde{A}_{R^2,P,u} \cup \tilde{A}_{R^2,P,v}), V(R^2) \setminus (\tilde{A}_{R^2,\hat{P},\hat{u}} \cup \tilde{A}_{R^2,\hat{P},\hat{v}})) \leq 2(\alpha_{\mathrm{dist}}(k) + \alpha_{\mathrm{sep}}(k)). +$$ + +In particular, this means that there exist vertices $x \in V(P_{u'_P, v'_P})$ and $y \in V(\hat{P}_{\hat{u}'_P, \hat{v}'_P})$ and a path $Q$ in $H$ between them whose length is at most $2(\alpha_{\text{dist}}(k) + \alpha_{\text{sep}}(k))$. Note that the unique path in $R^2$ between $x$ and $y$ traverses exactly one vertex in $\{u, v\}$. Suppose w.l.o.g. that this vertex is $u$. Then, consider the walk $W$ (which might be a path) obtained by traversing $P$ from $v$ to $x$ and then traversing $Q$ from $x$ to $y$. Now, notice that + +$$ +\begin{align*} +|E(P)| - |E(W)| &\geq |E(P'_{u,u'_P})| - 2(\alpha_{\text{dist}}(k) + \alpha_{\text{sep}}(k)) \\ +&\geq \alpha_{\text{pat}}(k)/2 - 2(\alpha_{\text{dist}}(k) + \alpha_{\text{sep}}(k)) \\ +&= 50 \cdot 2^{ck} - 2(4 \cdot 2^{ck} + \frac{7}{2} \cdot 2^{ck} + 2) > 0. +\end{align*} +$$ + +Here, the inequality $|E(P'_{u,u'_P})| \geq \alpha_{\text{pat}}(k)/2$ followed from Lemma 6.5. As $|E(P)| - |E(W)| > 0$, we have that $u, v$ and any subpath of the walk $W$ between $u$ and $v$ witness that $R^2$ has a detour (see Fig. 15). This is a contradiction, and hence we conclude that the intersection of $V(H) \setminus (V(C_{R^2,P,u}) \cup V(C_{R^2,P,v}))$ with $V(H) \setminus (V(C_{R^2,\hat{P},\hat{u}}) \cup V(C_{R^2,\hat{P},\hat{v}}))$ is empty. + +Having proved that the intersection is empty, we know that + +$$ +V(H) \setminus (V(C_{R^2,P,u}) \cup V(C_{R^2,P,v})) \subseteq V(C_{R^2,\hat{P},\hat{u}}) \cup V(C_{R^2,\hat{P},\hat{v}}). +$$ + +Thus, it remains to show that $V(H)\setminus(V(C_{R^2,P,u})\cup V(C_{R^2,P,v}))$ cannot contain vertices from both $V(C_{R^2,\hat{P},\hat{u}})$ and $V(C_{R^2,\hat{P},\hat{v}})$. Since $H[V(H) \setminus (V(C_{R^2,P,u}) \cup V(C_{R^2,P,v}))]$ is a connected graph (by Corollary 6.2), if $V(H) \setminus (V(C_{R^2,P,u}) \cup V(C_{R^2,P,v}))$ contains vertices from both $V(C_{R^2,\hat{P},\hat{u}})$ and $V(C_{R^2,\hat{P},\hat{v}})$, then it must also contain at least one vertex from $\mathrm{Sep}_{R^2}(\hat{P}, \hat{u}) \cup \mathrm{Sep}_{R^2}(\hat{P}, \hat{v}) \subseteq V(H) \setminus (V(C_{R^2,\hat{P},\hat{u}}) \cup V(C_{R^2,\hat{P},\hat{v}}))$, which we have already shown to be impossible. Thus, the proof is complete. □ \ No newline at end of file diff --git a/samples/texts/1754951/page_28.md b/samples/texts/1754951/page_28.md new file mode 100644 index 0000000000000000000000000000000000000000..c03d09f134d66b5a1df08a8a2d9c5b250e3e33db --- /dev/null +++ b/samples/texts/1754951/page_28.md @@ -0,0 +1,36 @@ +We are now ready to prove that $R^3$ is a Steiner tree. + +**Lemma 6.11.** Let $(G, S, T, g, k)$ be a good instance of Planar Disjoint Paths. Let $R^2$ be a Steiner tree that has no detour, and $R^3$ be the subgraph constructed from $R^2$ in Step IV. Then, $R^3$ is a Steiner tree with the following properties. + +* $R^3$ has the same set of vertices of degree at least 3 as $R^2$. + +* Every short maximal degree-2 path $P$ of $R^2$ is a short maximal degree-2 path of $R^3$. + +* For every long maximal degree-2 path $P$ of $R^2$ with endpoints $u$ and $v$, the paths $P_{u,u'_P}$ and $P_{v,v'_P}$ are subpaths of the maximal degree-2 path of $R^3$ with endpoints $u$ and $v$. + +*Proof.* To prove that $R^3$ is a Steiner tree, we only need to show that $R^3$ is acyclic. Indeed, the construction of $R^3$ immediately implies that it is connected and has the same set of degree-1 vertices as $R^2$, which together with an assertion that $R^3$ is acyclic, will imply that it is a Steiner tree. The other properties in the lemma are immediate consequences of the construction of $R^3$. By its construction, to show that $R^3$ is acyclic, it suffices to prove two conditions: + +* For every long maximal degree-2 path $P$ of $R^2$ with endpoints $u$ and $v$, it holds that + $$V(P_{u'_{P}, v'_{P}}^{*}) \cap (V(R^2) \setminus V(P_{u'_{P}, v'_{P}})) = \emptyset.$$ + +* For every two distinct long maximal degree-2 paths $P$ and $\hat{P}$ of $R^2$, it holds that + $$V(P_{u'_{P}, v'_{P}}^{*}) \cap V(\hat{P}_{\hat{u}'', \hat{v}''}) = \emptyset,$$ + where $u$ and $v$ are the endpoints of $P$, $\hat{u}$ and $\hat{v}$ are the endpoints of $\hat{P}$, + $$u' = u'_{P}, v' = v'_{P}, \hat{u}' = \hat{u}'_{\hat{P}}, \text{ and } \hat{v}' = \hat{v}'_{\hat{P}}.$$ + +The first condition follows directly from the fact that $P_{u'_{P}, v'_{P}}^{*}$ is a path in $H - (V(C_{R^2,P,u}) \cup V(C_{R^2,P,u}))$ while $V(R^2) \setminus V(P_{u'_{P}, v'_{P}})$ $\subseteq$ $V(C_{R^2,P,u}) \cup V(C_{R^2,P,u})$. + +For the second condition, note that Lemma 6.10 implies that $V(H) \setminus (V(C_{R^2,P,u}) \cup V(C_{R^2,P,v})) \subseteq V(C_{R^2,\hat{P},\hat{u}}) \cup V(C_{R^2,\hat{P},\hat{v}})$. Thus, we have that $V(P_{u'_{P},v'_{P}}^{*}) \subseteq V(C_{R^2,\hat{P},\hat{u}}) \cup V(C_{R^2,\hat{P},\hat{v}})$. However, +$$V(\hat{P}_{\hat{u}',\hat{v}'}) \cap (V(C_{R^2,\hat{P},\hat{u}}) \cup V(C_{R^2,\hat{P},\hat{v}})) = \emptyset, \text{ and hence } V(P_{u'_{P},v'_{P}}^{*}) \cap V(\hat{P}_{\hat{u}',\hat{v}'}) = \emptyset. \quad \square$$ + +We remark that $R^3$ might have detours. (These detours are restricted to $P_{u'_{P}, v'_{P}}$ for some long path $P = \text{path}_{R^3}(u, v)$ in $R^3$.) However, what is important for us is that we can still use the same small separators as before. To this end, we first define the appropriate notations, in particular since later we will like to address objects corresponding to $R^3$ directly (without referring to $R^2$). The validity of these notations follows from Lemma 6.11. Recall that $u'_{P}$ and $P_{u,u'_{P}}$ refer to the vertex and path in Lemma 6.5. + +**Definition 6.5 (Translating Notations of $R^2$ to $R^3$).** Let $(G, S, T, g, k)$ be a good instance of Planar Disjoint Paths. Let $R^2$ and $R^3$ be the Steiner trees constructed in Steps II and IV. For any long maximal degree-2 path $\hat{P}$ of $R^3$ and for each endpoint $u$ of $\hat{P}$: + +* Define $\hat{P}_{R^2}$ as the unique (long maximal degree-2) path in $R^2$ with the same endpoints as $\hat{P}$. + +* Let $P = \hat{P}_{R^2}$. Then, denote $u'_{\hat{P}} = u'_{P}$, $\hat{P}_{u,u'_{\hat{P}}} = P_{u,u'_{P}}$ and $\text{Sep}_{R^3}(\hat{P}, u) = \text{Sep}_{R^2}(P, u)$. + +* Define $A_{R^3, \hat{P}, u}^{*}$ as the union of $V(\hat{P}_{u, u'_{\hat{P}}})$ and the vertex set of the connected component of $R^3 - (V(\hat{P}_{u, u'_{\hat{P}}}) \setminus \{u\})$ containing $u$. + +* Define $B_{R^3, \hat{P}, u}^{*} = V(R^3) \setminus A_{R^3, \hat{P}, u}^{*}$. \ No newline at end of file diff --git a/samples/texts/1754951/page_29.md b/samples/texts/1754951/page_29.md new file mode 100644 index 0000000000000000000000000000000000000000..64d7dff7e02c8c56a961aee899dab09aa68c1194 --- /dev/null +++ b/samples/texts/1754951/page_29.md @@ -0,0 +1,18 @@ +Figure 16: Illustration of Lemma 6.12. + +In the context of Definition 6.5, note that by Lemma 6.5, $u' \in \text{Sep}_{R^3}(\hat{P}, u) \cap V(\hat{P})$ where $u' = u'_{\hat{P}}$, $\hat{P}_{u,u'}$ is the subpath of $\hat{P}$ between $u$ and $u'$, $\hat{P}_{u,u'}$ has no internal vertex from $\text{Sep}_{R^3}(\hat{P}, u) \cup \text{Sep}_{R^3}(\hat{P}, v)$, and $\alpha_{\text{pat}}(k)/2 \le |V(\hat{P}_{u,u'})| \le \alpha_{\text{pat}}(k)$. Additionally, note that $A^*_{R^3, \hat{P}, u}$ might not be equal to $A_{R^2, P, u}$ where $P = \hat{P}_{R^2}$. When $R^3$ is clear from context, we omit it from the subscripts. + +Now, let us argue why, in a sense, we can still use the same small separators as before. +Recall that a backbone Steiner tree is a Steiner tree constructed in Step IV. + +**Lemma 6.12.** Let $(G, S, T, g, k)$ be a good instance of Planar Disjoint Paths. Let $R^3$ be a backbone Steiner tree. Additionally, let $\tilde{P}$ be a long maximal degree-2 path of $R^3$, and $u$ be an endpoint of $\tilde{P}$. Then, $\text{Sep}(\tilde{P}, u)$ separates $A^*_{\tilde{P},u}$ and $B^*_{\tilde{P},u}$ in $H$. + +*Proof*. Denote $P = \hat{P}_{R^2}$ where $R^2$ is the Steiner tree computed in Step II to construct $R^3$. Then, $\text{Sep}(\hat{P}, u)$ separates $V(C_{R^2,P,v})$ and $V(C_{R^2,P,u})$ in $H$. Thus, to prove that $\text{Sep}(\hat{P}, u)$ separates $A^*_{\hat{P},u}$ and $B^*_{\hat{P},u}$ in $H$, it suffices to show that $A^*_{\hat{P},u} \subseteq V(C_{R^2,P,u})$ and $A^*_{\hat{P},v} \subseteq V(C_{R^2,P,v})$ (because $B^*_{\hat{P},u} \setminus A^*_{\hat{P},v} \subseteq V(P^*_{u'_P,v'_P})$ and $P^*_{u'_P,v'_P} \cap V(C_{R^2,P,u} = \emptyset$ by the construction of $P^*_{u'_P,v'_P}$). We only prove that $A^*_{\hat{P},u} \subseteq V(C_{R^2,P,u})$. The proof of the other containment is symmetric. + +Clearly, $A_{R^2,P,u} \cap A^*_{\hat{P},v} \subseteq V(C_{R^2,P,u})$. Thus, due to Lemma 6.11, to show that $A^*_{\hat{P},\tilde{u}} \subseteq V(C_{R^2,P,u})$, it suffices to show the following claim: For every long maximal degree-2 path $\tilde{P}$ of $R^2$ whose vertex set is contained in $V(C_{R^2,P,u})$, it holds that the vertex set of $\tilde{P}_{\tilde{u}'\tilde{P},\tilde{v}'\tilde{P}}$ (computed by Lemma 6.7) is contained in $V(C_{R^2,P,u})$ as well, where $\tilde{u}$ and $\tilde{v}$ are the endpoints of $\tilde{P}$. We refer the reader to Fig. 16 for an illustration of this statement. For the purpose of proving it, consider some long maximal degree-2 path $\tilde{P}$ of $R^2$ whose vertex set is contained in $V(C_{R^2,P,u})$. + +By Lemma 6.10, we know that either $V(H) \setminus (V(C_{R^2,\tilde{P},\tilde{u}}) \cup V(C_{R^2,\tilde{P},\tilde{v}})) \subseteq V(C_{R^2,P,u})$ or $V(H) \setminus (V(C_{R^2,\tilde{P},\tilde{u}}) \cup V(C_{R^2,\tilde{P},\tilde{v}})) \subseteq V(C_{R^2,P,v})$. Moreover, by the definition of $\tilde{P}_{\tilde{u}'\tilde{P},\tilde{v}'\tilde{P}}$, its vertex set is contained in $V(H) \setminus (V(C_{R^2,\tilde{P},\tilde{u}}) \cup V(C_{R^2,\tilde{P},\tilde{v}}))$. Thus, to conclude the proof, it remains to rule out the possibility that $V(H) \setminus (V(C_{R^2,\tilde{P},\tilde{u}}) \cup V(C_{R^2,\tilde{P},\tilde{v}})) \subseteq V(C_{R^2,P,v})$. For this purpose, recall that we chose $\tilde{P}$ such that $V(\tilde{P}) \subseteq V(C_{R^2,P,u})$, and that $V(\tilde{P}) \cap V(C_{R^2,P,u}) \neq \emptyset$. Because $V(C_{R^2,P,u}) \cap V(C_{R^2,P,v}) = \emptyset$, we derive that the containment $V(H) \setminus (V(C_{R^2,\tilde{P},\tilde{u}}) \cup V(C_{R^2,\tilde{P},\tilde{v}})) \subseteq V(C_{R^2,P,v})$ is indeed impossible. □ + +## 6.6 Enumerating Parallel Edges with Respect to $R^3$ + +Recall that $H$ is enriched with $4n + 1$ parallel copies of each edge of the (standard) radial completion of $G$. While the copies did not play a role in the construction of $R$, they will be \ No newline at end of file diff --git a/samples/texts/1754951/page_3.md b/samples/texts/1754951/page_3.md new file mode 100644 index 0000000000000000000000000000000000000000..49ef5b9a4282095b05ae0fee3eb6da12cf032973 --- /dev/null +++ b/samples/texts/1754951/page_3.md @@ -0,0 +1,7 @@ +Figure 9: A solution winding in a ring (top), and the “unwinding” or it (bottom). + +Having $P^*$ at hand, we replace $P'$ by $P^*$. This is done for every maximal degree-2 path, and thus we complete the construction of $R$. However, at this point, it is not clear why after we perform these replacements, the separators considered earlier remain separators, or that we even still have a tree. Roughly speaking, a scenario as depicted in Fig. 10 can potentially happen. To show that this is not the case, it suffices to prove that there cannot exist a vertex that belongs to two different rings. Towards that, we apply another preprocessing operation: we ensure that the radial completion of $G$ does not have $2^{ck}$ (for some constant $c$) concentric cycles that contain no vertex in $S \cup T$ by using another result by Adler et al. [2]. Informally, a sequence of concentric cycles is a sequence of vertex-disjoint cycles where each one of them is contained inside the next one in the sequence. Having no such sequences, we prove the following. + +**Lemma 2.3.** Let $R'$ be any Steiner tree. For every vertex $v$, there exists a vertex in $V(R')$ whose distance to $v$ (in the radial completion of $G$) is $2^{ck}$ for some constant $c$. + +To see why intuitively this lemma is correct, note that if $v$ was “far” from $R'$ in the radial completion of $G$, then in $G$ itself $v$ is surrounded by a large sequence of concentric cycles that contain no vertex in $S \cup T$. Having Lemma 2.3 at hand, we show that if a vertex belongs to a certain ring, then it is “close” to at least one vertex of the restriction of $R$ to that ring. In turn, that means that if a vertex belongs to two rings, it can be used to exhibit a “short” path between one vertex in the restriction of $R$ to one ring and another vertex in the restriction of $R$ to the second ring. By choosing constants properly, this path is shown to exhibit a detour in $R$, and hence we reach a contradiction. (In this argument, we use the fact that for every vertex $u$, towards the computation of the separator, we considered a vertex $u'$ of distance $2^{c_1k}$ from $u$—this subpath between $u$ and $u'$ is precisely that subpath that we will shortcut.) \ No newline at end of file diff --git a/samples/texts/1754951/page_31.md b/samples/texts/1754951/page_31.md new file mode 100644 index 0000000000000000000000000000000000000000..89328149ddc243cdd51950485a007764b5222eaa --- /dev/null +++ b/samples/texts/1754951/page_31.md @@ -0,0 +1,25 @@ +**Observation 6.10.** Let $(G, S, T, g, k)$ be a good instance of Planar Disjoint Paths with a backbone Steiner tree $R$. For every $v \in V(H)$, $\text{order}_v$ is an enumeration of $\hat{E}_R(v)$ in either clockwise or counter-clockwise order around $v$ (with a fixed start). Further, for any pair $e, e' \in E_R(v)$ such that $e$ occurs before $e'$ in $\text{order}_v$, the edges $e_0, e_1, \dots, e_{2n}$ occur before $e'_0, e'_1, \dots, e'_{2n}$. + +# 7 Existence of a Solution with Small Winding Number + +In this section we show that if the given instance admits a solution, then it admits a “nice solution”. The precise definition of nice will be in terms of “winding number” of the solution, which counts the number of times the solution spirals around the backbone steiner tree. Our goal is to show that there is a solution of small winding number. + +## 7.1 Rings and Winding Numbers + +Towards the definition of a ring, let us remind that $H$ is the triangulated plane multigraph obtained by introducing $4n + 1$ parallel copies of each edge to the radial completion of the input graph $G$. Hence, each face of $H$ is either a triangle or a 2-cycle. + +**Definition 7.1 (Ring).** Let $I_{\text{in}}$, $I_{\text{out}}$ be two disjoint cycles in $H$ such that the cycle $I_{\text{in}}$ is drawn in the strict interior of the cycle $I_{\text{out}}$. Then, $\text{Ring}(I_{\text{in}}, I_{\text{out}})$ is the plane subgraph of $H$ induced by the set of vertices that are either in $V(I_{\text{in}}) \cup V(I_{\text{out}})$ or drawn between $I_{\text{in}}$ and $I_{\text{out}}$ (i.e. belong to the exterior of $I_{\text{in}}$ and the interior of $I_{\text{out}}$). + +We call $I_{\text{in}}$ and $I_{\text{out}}$ are the *inner* and *outer interfaces* of $\text{Ring}(I_{\text{in}}, I_{\text{out}})$. We also say that this ring is induced by $I_{\text{in}}$ and $I_{\text{out}}$. Recall the notion of self-crossing walks defined in Section 5. Unless stated otherwise, all walks considered here are *not self-crossing*. A walk $\alpha$ in $\text{Ring}(I_{\text{in}}, I_{\text{out}})$ is *traversing* the ring if one of its endpoints lies in $I_{\text{in}}$ and the other lies in $I_{\text{out}}$. A walk $\alpha$ is *visiting* the ring if both its endpoints together lie in either $I_{\text{in}}$ or in $I_{\text{out}}$; moreover $\alpha$ is an *inner visitor* if both its endpoints lie in $I_{\text{in}}$, and otherwise it is an *outer visitor*. + +**Definition 7.2 (Orienting Walks).** Fix an arbitrary ordering of all vertices in $I_{\text{in}}$ and another one for all vertices in $I_{\text{out}}$. Then for a walk $\alpha$ in $\text{Ring}(I_{\text{in}}, I_{\text{out}})$ with endpoints in $V(I_{\text{in}}) \cup V(I_{\text{out}})$, orient $\alpha$ from one endpoint to another as follows. If $\alpha$ is a traversing walk, then orient it from its endpoint in $I_{\text{in}}$ to its endpoint in $I_{\text{out}}$. If $\alpha$ is a visiting walk, then both its endpoints lie either in $I_{\text{in}}$ or in $I_{\text{out}}$; then, orient $\alpha$ from its smaller endpoint to its greater endpoint. + +Observe that if $\alpha$ is a traversing path in the ring, then the orientation of $\alpha$ also defines its left-side and right-side. These are required for the following definition. + +**Definition 7.3 (Winding Number of a Walk w.r.t. a Traversing Path).** Let $\alpha$ be an a walk in $\text{Ring}(I_{\text{in}}, I_{\text{out}})$ with endpoints in $V(I_{\text{in}}) \cup V(I_{\text{out}})$, and let $\beta$ be a traversing path in this ring, such that $\alpha$ and $\beta$ are edge disjoint. The winding number, $\overline{\text{WindNum}}(\alpha, \beta)$, of $\alpha$ with respect to $\beta$ is the signed number of crossings of $\alpha$ with respect to $\beta$. That is, while walking along $\alpha$ (according to the orientation in Definition 7.2, for each intersection of $\alpha$ and $\beta$ record +1 if $\alpha$ crosses $\beta$ from left to right, -1 if $\alpha$ crosses $\beta$ from right to left, and 0 if it does not cross $\beta$. Then, the winding number $\overline{\text{WindNum}}(\alpha, \beta)$ is the sum of the recorded numbers. + +Observe that if $\alpha$ and $\beta$ are edge-disjoint traversing paths, then both $\overline{\text{WindNum}}(\alpha, \beta)$ and $\overline{\text{WindNum}}(\beta, \alpha)$ are well defined. We now state some well-known properties of the winding number. We sketch a proof of these properties in Appendix A, using homotopy. + +**Proposition 7.1.** Let $\alpha, \beta$ and $\gamma$ be three edge-disjoint paths traversing $\text{Ring}(I_{\text{in}}, I_{\text{out}})$. Then, + +$$ (i) \overline{\text{WindNum}}(\beta, \gamma) = -\overline{\text{WindNum}}(\gamma, \beta). $$ \ No newline at end of file diff --git a/samples/texts/1754951/page_32.md b/samples/texts/1754951/page_32.md new file mode 100644 index 0000000000000000000000000000000000000000..72a7a373e670b22238d09306c4da7a775b930038 --- /dev/null +++ b/samples/texts/1754951/page_32.md @@ -0,0 +1,34 @@ +$$ +(ii) \quad \left| \overline{\text{WindNum}}(\alpha, \beta) - \overline{\text{WindNum}}(\alpha, \gamma) \right| - \left| \overline{\text{WindNum}}(\beta, \gamma) \right| \le 1. +$$ + +We say that $\mathrm{Ring}(I_{\mathrm{in}}, I_{\mathrm{out}})$ is *rooted* if it is equipped with some fixed path $\eta$ that is traversing it, called the *reference path* of this ring. In a rooted ring $\mathrm{Ring}(I_{\mathrm{in}}, I_{\mathrm{out}})$, we measure all winding numbers with respect to $\eta$, hence we shall use the shorthand $\overline{\mathrm{WindNum}}(\alpha) = \overline{\mathrm{WindNum}}(\alpha, \eta)$ when $\eta$ is implicit or clear from context. Here, we implicitly assume that the walk $\alpha$ is edge disjoint from $\eta$. This requirement will always be met by the following assumptions: (i) $H$ is a plane multigraph where we have $4n + 1$ parallel copies of every edge, and we assume that the reference path $\eta$ consists of only the 0-th copy $e_0$; and (ii) whenever we consider the winding number of a walk $\alpha$, it will edge-disjoint from the reference curve $\eta$ as it will not contain the 0-th copy of any edge. (In particular, the walks of the (weak) linkages that we consider will always satisfy this property.) + +Note that any visitor walk in $\mathrm{Ring}(I_{\mathrm{in}}, I_{\mathrm{out}})$ with both endpoints in $I_{\mathrm{in}}$ is discretely homotopic to a segment of $I_{\mathrm{in}}$, and similarly for $I_{\mathrm{out}}$. Thus, we derive the following observation. + +**Observation 7.1.** Let $\alpha$ be a visitor in $\mathrm{Ring}(I_{\mathrm{in}}, I_{\mathrm{out}})$. Then, $|\overline{\mathrm{WindNum}}(\alpha)| \le 1$. + +Recall the notion of a weak linkage defined in Section 5, which is a collection of edge-disjoint non-crossing walks. When we use the term *weak linkage of order k* in $\mathrm{Ring}(I_{\mathrm{in}}, I_{\mathrm{out}})$, we refer to a weak linkage such that each walk has both endpoints in $V(I_{\mathrm{in}}) \cup V(I_{\mathrm{out}})$. For brevity, we abuse the term ‘weak linkage’ to mean a weak linkage in a ring when it is clear from context. Note that every walk in a weak linkage $\mathcal{P}$ is an inner visitor, or an outer visitor, or a traversing walk. This partitions $\mathcal{P}$ into $P_{\mathrm{in}}, P_{\mathrm{out}}, P_{\mathrm{traverse}}$. A weak linkage is *traversing* if it consists only of traversing walks. Assuming that $\mathrm{Ring}(I_{\mathrm{in}}, I_{\mathrm{out}})$ is rooted, we define the *winding number* of a traversing weak linkage $\mathcal{P}$ as $\overline{\mathrm{WindNum}}(\mathcal{P}) = \overline{\mathrm{WindNum}}(P_1)$. Recall that any two walks in a weak linkage are non-crossing. Then as observed in [14, Observation 4.4],¹² + +$$ +|\overline{\text{WindNum}}(P_i) - \overline{\text{WindNum}}(\mathcal{P})| \le 1 \quad \text{for all } i = 1, \dots, k. +$$ + +The above definition is extended to any weak linkage $\mathcal{P}$ in the ring as follows: if there is no walk in $\mathcal{P}$ that traverses the ring, then $\overline{\text{WindNum}}(\mathcal{P}) = 0$, otherwise $\overline{\text{WindNum}}(\mathcal{P}) = \overline{\text{WindNum}}(\mathcal{P}_{\text{traverse}})$. Note that, two aligned weak linkages $\mathcal{P}$ and $\mathcal{Q}$ in the ring may have different winding numbers (with respect to any reference path). Replacing a linkage $\mathcal{P}$ with an aligned linkage $\mathcal{Q}$ having a “small” winding number will be the main focus of this section. + +Lastly, we define a labeling of the edges based on the winding number of a walk (this relation +is made explicit in the observation that follows). + +**Definition 7.4.** Let $(G, S, T, g, k)$ be a good instance of Planar Disjoint Paths, and $H$ be the radial completion of $G$. Let $\alpha$ be a (not self-crossing) walk in $H$, and let $\beta$ be a path in $H$ such that $\alpha$ and $\beta$ are edge disjoint. Let us fix (arbitrary) orientations of $\alpha$ and $\beta$, and define the left and right side of the path $\beta$ with respect to its orientation. The labeling $\text{label}_\beta^\alpha$ of each ordered pair of consecutive edges, $(e, e') \in E_H(\alpha) \times E_H(\alpha)$ by $\{-1, 0, +1\}$ with respect to $\beta$, where $e$ occurs before $e'$ when traversing $\alpha$ according to its orientation is defined as follows. + +• The pair $(e, e')$ is labeled $+1$ if $e$ is on the left of $\beta$ while $e'$ is on the right of $\beta$. + +• else, $(e, e')$ is labeled $-1$ if $e$ is on the right while $e'$ is on the left of $\beta$; + +* otherwise *e* and *e'* are on the same side of *β* and (*e*, *e*') is labeled 0. + +Note that in the above labeling only pairs of consecutive edges may get a non-zero label, +depending on how they cross the reference path. For the ease of notation, we extend the above + +12This inequality also follows from the second property of Proposition 7.1 by setting α to be the reference path, +β = P1 and γ = Pi and noting that WindNum(β, γ) = 0. \ No newline at end of file diff --git a/samples/texts/1754951/page_33.md b/samples/texts/1754951/page_33.md new file mode 100644 index 0000000000000000000000000000000000000000..d336db89df5a0270345bb704e23a386f3bf4155f --- /dev/null +++ b/samples/texts/1754951/page_33.md @@ -0,0 +1,24 @@ +labeling function to all ordered pairs of edges in $\alpha$ (including pairs of non-consecutive edges), by +labeling them 0. Then we have the following observation, when we restrict $\alpha$ to $\text{Ring}(I_{\text{in}}, I_{\text{out}})$ +and set $\beta$ to be the reference path of this ring. + +**Observation 7.2.** Let $\alpha$ be a (not self-crossing) walk in $\text{Ring}(I_{\text{in}}, I_{\text{out}})$ with reference path $\eta$. Then $|\overline{\text{WindNum}}(\alpha, \eta)| = |\sum_{(e,e')\in E(\alpha)\times E(\alpha)} \text{label}_{\eta}^{\alpha}(e, e')|.$ + +## 7.2 Rerouting in a Ring + +We now address the question of rerouting a solution to reduce its winding number with respect to the backbone Steiner tree. As a solution is linkage in the graph $G$ (i.e. a collection of vertex disjoint paths), we first show how to reroute linkages within a ring. In the later subsections, we will apply this to reroute a solution in the entire plane graph. We remark that from now onwards, our results are stated and proved only for linkages (rather than weak linkages). Further, define a *linkage of order k in a Ring($I_{in}, I_{out}$)* as a collection of *k* vertex-disjoint paths in $G$ such that each of these paths belongs to *Ring($I_{in}, I_{out}$)* and its endpoints belong to $V(I_{in}) \cup V(I_{out})$. As before, we simply use the term 'linkage' when the ring is clear from context. We will use the following proposition proved by Cygan et al. [14] using earlier results of Ding et al. [17]. Its statement has been rephrased to be compatible with our notation. + +**Proposition 7.2 (Lemma 4.8 in [14]).** Let $\text{Ring}(I_{\text{in}}, I_{\text{out}})$ be a rooted ring in $H$ and let $\mathcal{P}$ and $\mathcal{Q}$ be two traversing linkages of the same order in this ring. Then, there exists a traversing linkage $\mathcal{P}'$ in this ring that is aligned with $\mathcal{P}$ and such that $|\overline{\text{WindNum}}(\mathcal{P}') - \overline{\text{WindNum}}(\mathcal{Q})| \le 6$. + +The formulation of [14] concerns directed paths in directed graphs and assumes a fixed pat- +tern of in/out orientations of paths that is shared by the linkages $\mathcal{P}, \mathcal{Q}$ and $\mathcal{P}'$. The undirected +case (as expressed above) can be reduced to the directed one by replacing every undirected edge +in the graph by two oppositely-oriented arcs with same endpoints, and asking for any orienta- +tion pattern (say, all paths should go from $I_{\text{in}}$ to $I_{\text{out}}$). Moreover, the setting itself is somewhat +more general, where rings and reference paths are defined by curves and (general) homotopy. + +**Rings with Concentric Cycles.** Let $C = (C_1, C_2, \dots, C_p)$ concentric sequence of cycles in $\text{Ring}(I_{\text{in}}, I_{\text{out}})$ (then, $C_i$ is in the strict interior of $C_{i+1}$ for $i \in \{1, 2, \dots, p-1\}$). If $I_{\text{in}}$ is in the strict interior of $C_1$ and $C_p$ is in the strict interior of $I_{\text{out}}$, then we say that $C$ is *encircling*. An encircling concentric sequence $C$ in $\text{Ring}(I_{\text{in}}, I_{\text{out}})$ is *tight* if every $C \in C$ is a cycle in $G$, and there exists a path $\eta$ in $H$ traversing $\text{Ring}(I_{\text{in}}, I_{\text{out}})$ such that the set of internal vertices of $\eta$ contain exactly $|C|$ vertices of $V(G)$, one on each each cycle in $C$. Let us fix one such encircling tight sequence in the ring $\text{Ring}(I_{\text{in}}, I_{\text{out}})$ along with the path $\eta$ witnessing the tightness. Then, we set the path $\eta$ as the reference path of the ring. Here, we assume w.l.o.g. that $\eta$ contains only the 0-th copy of each of the edges comprising it. Any paths or linkages that we subsequently consider will not use the 0-th copy of any edge, and hence their winding numbers (with respect to $\eta$) will be well-defined. This is because that they arise from $G$, and when we consider them in $H$, we choose a 'non-0-th' copy out of the $4n+1$ copies of any (required) edge. + +A linkage $\mathcal{P}$ in $\text{Ring}(I_{\text{in}}, I_{\text{out}})$ is minimal with respect to $\mathcal{C}$ if among the linkages aligned with $\mathcal{P}$, it minimizes the total number of edges traversed that do not lie on the cycles of $\mathcal{C}$. The following proposition is essentially Lemma 3.7 of [14]. + +**Proposition 7.3.** Let $G$ be a plane graph, and with radial completion $H$. Let $\text{Ring}(I_{\text{in}}, I_{\text{out}})$ be a rooted ring in $H$. Suppose $|I_{\text{in}}|, |I_{\text{out}}| \le l$, for some integer $l$. Further, let $C = (C_1, \dots, C_p)$ be an encircling tight concentric sequence of cycles in $\text{Ring}(I_{\text{in}}, I_{\text{out}})$. Finally, let $\mathcal{P}$ be a linkage in $\text{Ring}(I_{\text{in}}, I_{\text{out}})$ that is minimal with respect to $C$. Then, every inner visitor of $\mathcal{P}$ intersects less than $10l$ of the first cycles in the sequence $(C_1, \dots, C_p)$, while every outer visitor of $\mathcal{P}$ intersects less than $10l$ of the last cycles in this sequence. \ No newline at end of file diff --git a/samples/texts/1754951/page_35.md b/samples/texts/1754951/page_35.md new file mode 100644 index 0000000000000000000000000000000000000000..a04a5a40ec83f91194934607e3acd7d998566926 --- /dev/null +++ b/samples/texts/1754951/page_35.md @@ -0,0 +1,17 @@ +A proof of this proposition can be obtained by first ordering the collection of inner and outer visitors by their 'distance' from the inner and outer interfaces, respectively, and the 'containment' relation between the cycles formed by them with the interfaces. This gives a partial order on the set of inner visitors and the set of outer visitors. Then if the proposition does not hold for $\mathcal{P}$, then the above ordering and containment relation can be used to reroute these paths along a suitable cycle. This will contradict the minimality of $\mathcal{P}$, since the rerouted linkage is aligned with it but uses strictly fewer edges outside of $\mathcal{C}$. The main result of this section can be now formulated as follows. (Its formulation and proof idea are based on Lemma 8.31 and Theorem 6.45 of [14].) + +**Lemma 7.1.** Let $G$ be a plane graph with radial completion $H$. Let $\text{Ring}(I_{\text{in}}, I_{\text{out}})$ be a ring in $H$. Suppose that $|I_{\text{in}}|, |I_{\text{out}}| \le l$ for some integer $l$, and that in $\text{Ring}(I_{\text{in}}, I_{\text{out}})$ there is an encircling tight concentric sequence of cycles $\mathcal{C}$ of size larger than $40l$. Let $\eta$ be a traversing path in the ring witnessing the tightness of $\mathcal{C}$, and fix $\eta$ as the reference path. Finally, let $\mathcal{P} = \mathcal{P}_{\text{traverse}} \uplus \mathcal{P}_{\text{visitor}}$ be a linkage in $G$, where $\mathcal{P}_{\text{traverse}}$ is a traversing linkage comprising the paths of $\mathcal{P}$ traversing $\text{Ring}(I_{\text{in}}, I_{\text{out}})$, while $\mathcal{P}_{\text{visitor}} = \mathcal{P} \setminus \mathcal{P}_{\text{traverse}}$ consists of the paths whose both endpoints lie in either $V(I_{\text{in}})$ or $V(I_{\text{out}})$. Further, suppose that $\mathcal{P}$ is minimal with respect to $\mathcal{C}$. Then, for every traversing linkage $\mathcal{Q}$ in $G$ that is minimal with respect to $\mathcal{C}$ such that every path in $\mathcal{Q}$ is disjoint from $\eta$ and $|\mathcal{Q}| \ge |\mathcal{P}_{\text{traverse}}|$, there is a traversing linkage $\mathcal{P}'_{\text{traverse}}$ in $G$ such that + +(a) $\mathcal{P}'_{\text{traverse}}$ is aligned with $\mathcal{P}_{\text{traverse}}$, + +(b) the paths of $\mathcal{P}'_{\text{traverse}}$ are disjoint from the paths of $\mathcal{P}_{\text{visitor}}$, and + +(c) $|\overline{\text{WindNum}}(\mathcal{P}'_{\text{traverse}}) - \overline{\text{WindNum}}(\mathcal{Q})| \le 60l + 6.$ + +*Proof.* Let $\mathcal{C} = (C_1, \dots, C_p)$, where $p > 40l$. Recall that $\mathcal{C}$ is a collection of cycles in $G$, and the path $\eta$ that witnesses the tightness of $\mathcal{C}$ contains $|\mathcal{C}|$ vertices of $V(G)$, one on each cycle of $\mathcal{C}$. Let $v_i$ denote the vertex where $\eta$ intersects the cycle $C_i \in \mathcal{C}$ for all $i \in \{1, 2, \dots, p\}$. Since $\mathcal{P}$ is minimal with respect to $\mathcal{C}$, Proposition 7.3 implies that the paths in $\mathcal{P}_{\text{visitor}}$ do not intersect any of the cycles $C_{10l}, C_{10l+1}, \dots, C_{p-10l+1}$ (note that since $p > 40l$, this sequence of cycles is non-empty). Call a vertex $x \in V(\text{Ring}(I_{\text{in}}, I_{\text{out}}))$ in the ring *non-separated* if there exists a path from $x$ to $C_{10l}$ whose set of internal vertices is disjoint from $\bigcup_{P \in \mathcal{P}_{\text{visitor}}} V(P)$. Otherwise, we say that the vertex $x$ is *separated*. Observe that every path in $\mathcal{P}_{\text{traverse}}$ is disjoint from the paths in $\mathcal{P}_{\text{visitor}}$ and intersects $C_{10l}$, hence all vertices on the paths of $\mathcal{P}_{\text{traverse}}$ are non-separated. Let $X$ denote the set of all non-separated vertices in the ring, and consider the graph $H[X]$. Observe that $H[X]$ is an induced subgraph of $\text{Ring}(I_{\text{in}}, I_{\text{out}})$, since it is obtained from $\text{Ring}(I_{\text{in}}, I_{\text{out}})$ by deleting the separated vertices. Further, observe that $H[X]$ is a ring of $H$. Indeed, the inner interface of $H[X]$ is the cycle $\hat{I}_{\text{in}}$ obtained as follows: let $\mathcal{P}_{\text{in}}$ be the set of inner visitors in $\mathcal{P}$; then, $\hat{I}_{\text{in}}$ is the outer face of the plane graph $H[V(I_{\text{in}}) \cup \bigcup_{P \in \mathcal{P}_{\text{in}}} V(P)]$. It is easy to verify that all vertices on $\hat{I}_{\text{in}}$ are non-separated, and any vertex of $\text{Ring}(I_{\text{in}}, I_{\text{out}})$ that lies in the strict interior of this cycle is separated. We then symmetrically obtain the outer interface $\hat{I}_{\text{out}}$ of $H[X]$ from the set $\mathcal{P}_{\text{out}}$ of outer visitors of $\mathcal{P}$. Then, $H[X] = \text{Ring}(\hat{I}_{\text{in}}, \hat{I}_{\text{out}})$. Here, $\hat{I}_{\text{in}}$ is composed alternately of subpaths of $I_{\text{in}}$ and inner visitors from $\mathcal{P}_{\text{visitor}}$, and symmetrically for $\hat{I}_{\text{out}}$. + +Note that the paths of $\mathcal{P}_{\text{traverse}}$ are completely contained in $\text{Ring}(\hat{I}_{\text{in}}, \hat{I}_{\text{out}})$ and they all traverse this ring. Thus, $\mathcal{P}_{\text{traverse}}$ can be regarded also as a traversing linkage in $\text{Ring}(\hat{I}_{\text{in}}, \hat{I}_{\text{out}})$. While $\mathcal{P}_{\text{traverse}}$ may have a different winding number in $\text{Ring}(\hat{I}_{\text{in}}, \hat{I}_{\text{out}})$ than in $\text{Ring}(I_{\text{in}}, I_{\text{out}})$, the difference is “small” as we show below. (Note that the two winding numbers in the following claim are computed in two different rings.) + +**Claim 7.1.** Let $P$ be a path in $G$ that is disjoint from all paths in $\mathcal{P}_{\text{visitor}}$, such that $P$ belongs to $\text{Ring}(I_{\text{in}}, I_{\text{out}})$ and traverses it. Then, $P$ also belongs to $\text{Ring}(\hat{I}_{\text{in}}, \hat{I}_{\text{out}})$, and $|\overline{\text{WindNum}}(P, \eta)| \le 20l$. ¹³ + +¹³Here $\hat{\eta}$ is the reference path of $\text{Ring}(\hat{I}_{\text{in}}, \hat{I}_{\text{out}})$, which is a subpath of $\eta$ in this ring. \ No newline at end of file diff --git a/samples/texts/1754951/page_36.md b/samples/texts/1754951/page_36.md new file mode 100644 index 0000000000000000000000000000000000000000..bc0e3d644aa7583b46e5c8e7e3223b459f3628d2 --- /dev/null +++ b/samples/texts/1754951/page_36.md @@ -0,0 +1,17 @@ +Figure 18: Illustration of Claim 7.2. + +*Proof.* Since $P$ traverses $\text{Ring}(I_{\text{in}}, I_{\text{out}})$, it must intersect the cycle $C_{10\ell}$. Therefore, as $P$ is disjoint from $\mathcal{P}_{\text{visitor}}$, all vertices in $V(P)$ are non-separated. Hence, $P$ is present in $\text{Ring}(\hat{I}_{\text{in}}, \hat{I}_{\text{out}})$. Next, observe that there are at most $20\ell$ vertices of $G$ that are visited by $\eta$ but not visited by $\hat{\eta}$; indeed, these are vertices in the intersection of $\eta$ with $I_{\text{in}}$, $I_{\text{out}}$ and the first and last $10\ell - 1$ cycles of $\mathcal{C}$. It follows that any path in $G$ has at most $20\ell$ more crossings with $\eta$ than with $\hat{\eta}$. Since each such crossing contributes +1 or -1 to the winding number of $P$ with respect to $\eta$, the winding numbers of $P$ with respect to $\eta$ and $\hat{\eta}$ differ by at most $20\ell$. $\diamond$ + +We now turn our attention to the linkage $Q$. In essence, our goal is show that every path in $Q$ can be “trimmed” to a path traversing $\text{Ring}(\hat{I}_{\text{in}}, \hat{I}_{\text{out}})$ such that their winding numbers are not significantly different. First, however, we prove that the paths in $Q$ cannot “oscillate” too much in $\text{Ring}(I_{\text{in}}, I_{\text{out}})$, based on the supposition that $Q$ is minimal with respect to $\mathcal{C}$. + +**Claim 7.2.** Let $Q \in \mathcal{Q}$, and let $u \in V(Q)$ such that it also lies on an inner visitor from $\mathcal{P}_{\text{visitor}}$. Then, the prefix of $Q$ between its endpoint on $I_{\text{in}}$ and $u$ does not intersect the cycle $C_{20\ell}$. + +*Proof.* Suppose, for the sake of contradiction, that the considered prefix contains some vertex $v$ that lies on $C_{20\ell}$. Since $u$ lies on an inner visitor $P \in \mathcal{P}_{\text{visitor}}$ and Proposition 7.3 states that an inner visitor cannot intersect $C_{10\ell}$, we infer that on the infix of $Q$ between $v$ and $u$ there exists a vertex that lies on the intersection of $Q$ and $C_{10\ell}$. Let $a$ be the first such vertex. Similarly, on the prefix of $Q$ from its endpoint on $I_{\text{in}}$ to $v$ there exists a vertex that lies on the intersection of $Q$ and $C_{10\ell}$. Let $b$ be the last such vertex. Then the whole infix of $Q$ between $a$ and $b$ does not intersect $C_{10\ell}$ internally (see Fig. 18), and hence, apart from endpoints, completely lies in the exterior of $C_{10\ell}$. Call this infix $Q^*$. + +Now consider $\text{Ring}(C_{10\ell}, I_{\text{out}})$, the ring induced by $I_{\text{out}}$ and $C_{10\ell}$. Moreover, consider the graph $G'$ obtained from $G$ by removing all vertices that are not in $\text{Ring}(C_{10\ell}, I_{\text{out}})$ and edges that are not in the strict interior of $\text{Ring}(C_{10\ell}, I_{\text{out}})$; in particular, the edges of $C_{10\ell}$ are removed, but the vertices are not. Note that $G'$ is a subgraph of $\text{Ring}(C_{10\ell}, I_{\text{out}})$. Finally, let $C' = C \setminus \{C_1, C_2, \dots, C_{10\ell}\}$; then, $C'$ is an encircling tight sequence of concentric cycles in $\text{Ring}(C_{10\ell}, I_{\text{out}})$. + +Let $Q'$ be the linkage in $G$ obtained by restricting paths of $Q$ to $G'$. Here, a path in $Q$ may break into several paths in $Q'$ (that are its maximal subpaths contained in $G'$). Since $Q$ is minimal with respect to $\mathcal{C}$, it follows that $Q'$ is minimal with respect to $\mathcal{C}'$. Now, observe that $Q^*$ belongs to $Q'$, hence it is an inner visitor of $\text{Ring}(C_{10\ell}, I_{\text{out}})$. However, $Q^*$ intersects the first $10\ell+1$ concentric cycles $C_{10\ell+1}, \dots, C_{20\ell}$ in the family $\mathcal{C}'$, which contradicts Proposition 7.3. $\diamond$ + +Clearly, an analogous claim holds for outer visitors and the cycle $C_{p-20\ell+1}$. We now proceed to our main claim about the restriction of $Q$ to $\text{Ring}(\hat{I}_{\text{in}}, \hat{I}_{\text{out}})$. + +**Claim 7.3.** For every path $Q \in \mathcal{Q}$, there exists a subpath $\tilde{Q}$ of $Q$ that traverses $\text{Ring}(\hat{I}_{\text{in}}, \hat{I}_{\text{out}})$ and such that $\left|\text{WindNum}(\tilde{Q}, \hat{\eta}) - \text{WindNum}(Q, \eta)\right| \le 40\ell$. \ No newline at end of file diff --git a/samples/texts/1754951/page_38.md b/samples/texts/1754951/page_38.md new file mode 100644 index 0000000000000000000000000000000000000000..4d1e09826886d7a1db3fe55753c6b0aecdd0b959 --- /dev/null +++ b/samples/texts/1754951/page_38.md @@ -0,0 +1,19 @@ +**Observation 7.3.** The path $\mathbf{path}_R(u', v')$ is contained in $\text{Ring}(S_u, S_v)$ and it traverses $\text{Ring}(S_u, S_v)$ from $u' \in S_u$ to $v' \in S_v$. Moreover, $A_{R,P,u}^* - \{\tilde{u}\}'$ is contained in the bounded region of $\mathbb{R}^2 - H[S_u]$, and similarly $A_{R,P,v}^* - \{\tilde{v}\}'$ is contained in the unbounded region of $\mathbb{R}^2 - H[S_v]$. + +Now, recall the encircling tight sequence of concentric cycles $C(u, v)$ and the path witnessing its tightness, which we denote by $\eta(u, v)$ (constructed in Lemma 6.6). We assume that $\eta(u, v)$ consists of only the 0-th copies of the edges comprising it, and fix $\eta(u, v)$ as the reference path of the ring $\text{Ring}(S_u, S_v)$. Moreover, observe that the subpath $\mathbf{path}_R(u', v')$ of $R$ also traverses $\text{Ring}(S_u, S_v)$. We assume w.l.o.g. that $\mathbf{path}_R(u', v')$ consists of only the 0-th copy of the edges comprising it. This will allow us to later define winding numbers with respect to $\mathbf{path}_R(u', v')$ in $\text{Ring}(S_u, S_v)$. Note that $\eta(u, v)$ and $\mathbf{path}_R(u', v')$ may be different paths with common edges, and further they may not be paths in $G$. + +Let us first argue that we can consider the rings corresponding to each of the long degree-2 paths in $R$ independently. To this end, consider another long maximal degree-2 path in $R$ different from $P$, denoted by $\mathbf{path}_R(\hat{u}, \hat{v})$, where $\hat{u}, \hat{v} \in V_{=1}(R) \cup V_{\ge 3}(R)$ and $t^*$ lies in the subtree containing $\hat{v}$ in $R - V(\mathbf{path}_R(\hat{u}, \hat{v}) - \{\hat{u}, \hat{v}\})$. Let $S_{\hat{u}} = \text{Sep}_R(\mathbf{path}_R(\hat{u}, \hat{v}), \hat{u})$ and $S_{\hat{v}} = \text{Sep}_R(\mathbf{path}_R(\hat{u}, \hat{v}), \hat{v})$. The $\text{Ring}(S_{\hat{u}}, S_{\hat{v}})$ is symmetric to $\text{Ring}(S_u, S_v)$ above. The following corollary follows from Lemma 6.10. + +**Corollary 7.1.** *The rings $\text{Ring}(S_u, S_v)$ and $\text{Ring}(S_{\hat{u}}, S_{\hat{v}})$ have no common vertices.* + +Let $t$ be the number of pairs of near vertices in $V_{=1}(R) \cup V_{\ge 3}(R)$ such that the path between them in $R$ is long. As above, we have a ring corresponding to each of these paths. Then, we partition the plane graph $H$ into $t+1$ regions, one for each of the $t$ rings of the long maximal degree-2 path in $R$, and the remainder of $H$ that is not contained in any of these rings. + +**Lemma 7.2.** Let $\{u_1, v_1\}, \{u_2, v_2\}, \dots, \{u_t, v_t\}$ denote pairs of near vertices in $V_{=1}(R) \cup V_{\ge 3}(R)$ such that $\mathbf{path}_R(u_i, v_i)$ is a long maximal degree-2 path in $R$ for all $i \in \{1, 2, \dots, t\}$. Then, the corresponding rings $\text{Ring}(S_{u_1}, S_{v_1}), \text{Ring}(S_{u_2}, S_{v_2}), \dots, \text{Ring}(S_{u_t}, S_{v_t})$ are pairwise disjoint. Further, the number of vertices of $R$ lying outside this collection of rings, $|V(R) \setminus \bigcup_{i=1}^t V(S_{u_i}, S_{v_i})|$, is upper bounded by $\alpha_{\text{nonRing}}(k) = 10^4 k \cdot 2^{ck}$. + +*Proof.* The first statement follows from Corollary 7.1 applied to each pair of rings. For the second statement, consider the vertices in $V(R) \setminus \bigcup_{i=1}^t V(S_{u_i}, S_{v_i})$. Each such vertex either belongs to a short degree-2 path between a pair of near vertices in $V_{=1}(R) \cup V_{\ge 3}(R)$, or it is a vertex that lies on a maximal degree-2 path of $R$ at distance at most $\alpha_{\text{pat}}(k) = 100 \cdot 2^{ck}$ in $R$ from $V_{=1}(R) \cup V_{\ge 3}(R)$. Recall that there are fewer than $4k$ vertices in $V_{=1}(R) \cup V_{\ge 3}(R)$ by Observation 6.1, and thus at most $4k-2$ maximal degree-2 paths in $R$. Therefore, the number of vertices in $V_{=1}(R) \cup V_{\ge 3}(R)$ is upper bounded by $(4k-2) \cdot \max\{2\alpha_{\text{pat}}(k), \alpha_{\text{long}}(k)\} \le 10^4 k \cdot 2^{ck}$. $\square$ + +## 7.4 Solutions with Small Winding Number + +Having established all the required definitions in the previous subsections, we are ready to exhibit a solution with a small winding number. Consider a ring, say $\text{Ring}(S_u, S_v)$ with an encircling tight concentric sequence of cycles $C(u, v)$ and reference path $\eta(u, v)$. Consider a linkage $\mathcal{P}$ in $G$, and let $\mathcal{P}(u, v) = \mathcal{P}[V(S_u, S_v) \cap V(\mathcal{P})]$. Observe that $\mathcal{P}(u, v)$ is a linkage in $\text{Ring}(S_u, S_v)$, and its endpoints lie in $(S_u \cup S_v) \cap V(G)$. Therefore, $\mathcal{P}(u, v)$ is a flow from $S_u \cap V(G)$ to $S_v \cap V(G)$ in $G_{u,v}$. Assume w.l.o.g. that the paths in $\mathcal{P}(u, v)$ use the 1-st copy of each edge in $H$. + +**Definition 7.5.** Let $u, v \in V_{=1}(R) \cup V_{\ge 3}(R)$ be a pair of near vertices such that $\mathbf{path}_R(u, v)$ is long maximal degree-2 path in $R$. Let $\mathcal{P}$ be a path system in $G$. Then, winding number of $\mathcal{P}$ in $\text{Ring}(S_u, S_v)$ is defined as $\text{WindNum}(\mathcal{P}, \text{Ring}(S_u, S_v)) = \max_{P \in \mathcal{P}(u,v)} |\text{WindNum}(P, \mathbf{path}_R(u', v'))|$ \ No newline at end of file diff --git a/samples/texts/1754951/page_39.md b/samples/texts/1754951/page_39.md new file mode 100644 index 0000000000000000000000000000000000000000..af222f5e53e9c9fdad5d0917a1ea79a4464507dc --- /dev/null +++ b/samples/texts/1754951/page_39.md @@ -0,0 +1,21 @@ +We remark that the notation above differs from our earlier use of $\overline{\text{WindNum}}(\cdot)$ in the choice of the reference path in $\text{Ring}(S_u, S_v)$. For $\overline{\text{WindNum}}(\cdot)$, the path $\eta(u, v)$ was the reference path, whereas for $\text{WindNum}(\cdot, \text{Ring}(S_u, S_v))$, we choose $\text{path}_R(u', v')$ as the reference path. + +We will now apply Lemma 7.1 in each ring of the form $\text{Ring}(S_{u_i}, S_{v_i})$ to obtain a solution of bounded winding number in that ring. Towards this, fix one pair $(u, v) := (u_i, v_i)$ for some $1 \le i \le t$. We argue that $\mathcal{C}(u, v)$ and $\text{Flow}_R(u, v)$ are suitable for the roles of $\mathcal{C}$ and $\mathcal{Q}$ in the premise of Lemma 7.1. Recall that $\text{Flow}_R(u, v)$ is a collection of vertex disjoint paths in the subgraph $G_{u_i v_i}$ of $G$, and assume w.l.o.g. that $\text{Flow}_R(u, v)$ uses only 2-nd copies of edges in $H$. + +**Observation 7.4.** Let $(G, S, T, g, k)$ be a good Yes-instance of Planar Disjoint Paths. Let $R$ be a backbone Steiner tree. Let $u, v$ be a pair of near vertices in $V_{=1}(R) \cup V_{\ge 3}(R)$ such that $\text{path}_R(u, v)$ is a long degree-2 path in $R$. Then, the following hold. + +(i) $\mathcal{C}(u, v)$ is an encircling tight sequence of concentric cycles in $\text{Ring}(S_u, S_v)$ where each cycle lies in $G_{u,v}$, $S_u$ is in its strict interior and $S_v$ is in its strict exterior. Further, $|\mathcal{C}(u, v)| \ge 40\alpha_{\text{sep}}(k)$ cycles. + +(ii) $\text{Flow}_R(u, v)$ is a maximum flow between $S_u \cap V(G)$ and $S_v \cap V(G)$ in $G_{u,v}$ that is minimal with respect to $\mathcal{C}(u, v)$. + +(iii) Each path $Q \in \text{Flow}_R(u, v)$ traverses $\text{Ring}(S_u, S_v)$, and $|\overline{\text{WindNum}}(\text{path}_R(u', v'), Q)| \le 1$. + +*Proof.* The first statement directly follows from the construction of $\mathcal{C}(u, v)$ (see Lemma 6.6). Similarly, the second statement directly follows from the construction of $\text{Flow}_R(u, v)$ using $\mathcal{C}(u, v)$ (see Observation 6.8). Additionally the construction implies that each path $Q \in \text{Flow}_R(u, v)$ traverses $\text{Ring}(S_u, S_v)$. For the second part of the third statement, first note any $Q \in \text{Flow}_R(u, v)$ and $\text{path}_R(u', v')$ are two edge-disjoint traversing paths in $\text{Ring}(S_u, S_v)$. Hence, $\overline{\text{WindNum}}(\text{path}_R(u', v'), P)$ is well defined. Since for any $Q \in \text{Flow}_R(u, v)$, there are at most two edges in $\text{path}_R(u', v')$ with only one endpoint in $V(Q)$ (by Corollary 6.1, the absolute value of the signed sum of crossing between these two paths is upper-bounded by 1, i.e. $|\overline{\text{WindNum}}(Q, \text{path}_R(u', v'))| \le 1$. $\square$ + +Finally we are ready to prove that the existence of a solution of small winding number. + +**Lemma 7.3.** Let $(G, S, T, g, k)$ be a good Yes-instance of Planar Disjoint Paths. Let $R$ be a backbone Steiner tree. Let $\{u_1, v_1\}, \{u_2, v_2\}, \dots, \{u_t, v_t\}$ be the pairs of near vertices in $V_{=1}(R) \cup V_{\ge 3}(R)$ such that $\text{path}_R(u_i, v_i)$ is a long maximal degree-2 path in $R$ for all $i \in \{1, 2, \dots, t\}$. Let $\text{Ring}(S_{u_1}, S_{v_1}), \text{Ring}(S_{u_2}, S_{v_2}), \dots, \text{Ring}(S_{u_t}, S_{v_t})$ be the corresponding rings. Then, there is a solution $\mathcal{P}^*$ to $(G, S, T, g, k)$ such that $|\overline{\text{WindNum}}(\mathcal{P}^*, \text{Ring}(S_{u_i}, S_{v_i}))| \le \alpha_{\text{winding}}(k)$ for all $i \in \{1, 2, \dots, t\}$, where $\alpha_{\text{winding}}(k) = 60\alpha_{\text{sep}}(k) + 11 < 300 \cdot 2^{ck}$. + +*Proof.* Consider a solution $\mathcal{P}$ to $(G, S, T, g, k)$. Fix a pair $(u, v) := (u_i, v_i)$ for some $i \in \{1, 2, \dots, t\}$. Recall that $|S_u|, |S_v| \le l$ where $l = \alpha_{\text{sep}}(k)$. Consider $\text{Ring}(S_u, S_v)$, and the linkage $\mathcal{P}(u, v)$ in $G$. Our goal is to modify $\mathcal{P}(u, v)$ to obtain another linkage $\mathcal{P}'(u, v)$ that is aligned with it and has a small winding number with respect to $\text{path}_R(u, v)$. + +Recall that by Observation 7.4, we have an encircling tight sequence of concentric cycles $\mathcal{C}(u,v)$ in $\text{Ring}(S_u,S_v)$ that contains at least $40l$ cycles. Let $\eta(u,v)$ be the path in this ring witnessing the tightness of $\mathcal{C}(u,v)$. Further, recall the linkage $\text{Flow}_R(u,v)$ between $S_u$ and $S_v$ in the subgraph $G_{u,v}$ of $G$ (the restriction of $G$ to $\text{Ring}(S_u,S_v)$), and that $\text{Flow}_R(u,v)$ is minimal with respect to $\mathcal{C}(u,v)$. We can assume w.l.o.g. that $\mathcal{P}(u,v)$ is minimal with respect to $\mathcal{C}(u,v)$. Otherwise, there is another solution $\hat{\mathcal{P}}$ such that it is identical to $\mathcal{P}$ in $G - V(S_u,S_v)$ and $\hat{\mathcal{P}}(u,v)$ is a minimal linkage with respect to $\mathcal{C}(u,v)$ (that is aligned with $\mathcal{P}(u,v)$). Then, we can consider $\hat{\mathcal{P}}$ instead of $\mathcal{P}$. Let $\mathcal{P}_{\text{traverse}}(u,v)$ be the set of traversing paths in $\mathcal{P}(u,v)$. Since $\mathcal{P}_{\text{traverse}}(u,v)$ is a flow between $S_u$ and $S_v$ in $G_{u,v}$ and $\text{Flow}_R(u,v)$ is a maximum flow, clearly $|\text{Flow}_R(u,v)| \ge |\mathcal{P}_{\text{traverse}}(u,v)|$. \ No newline at end of file diff --git a/samples/texts/1754951/page_4.md b/samples/texts/1754951/page_4.md new file mode 100644 index 0000000000000000000000000000000000000000..8864bab4cdb981567b9a00853fd31c4f62939435 --- /dev/null +++ b/samples/texts/1754951/page_4.md @@ -0,0 +1,17 @@ +Figure 10: The green and blue rings intersect, which can create cycles in $R$ when replacing paths. + +**Pushing a Solution Onto $R$.** So far, we have argued that if there is a solution, then there is also one such that the sum of the potential of all of the groups of all of its paths is at most $2^{\mathcal{O}(k)}$. Additionally, we discussed the intuition why this, in turn, implies the following result. + +**Lemma 2.4.** If there is a solution $\mathcal{P}$, then there is a weak linkage pushed onto $R$ that is discretely homotopic to $\mathcal{P}$ and uses at most $2^{\mathcal{O}(k)}$ copies of every edge. + +The formal proof of Lemma 2.4 (in Section 8) is quite technical. On a high level, it consists of three phases. First, we push onto $R$ all sequences of the solution—that is, maximal subpaths that touch (but not necessarily cross) $R$ only at their endpoints. Second, we eliminate some U-turns of the resulting weak linkage (see Fig. 11), as well as “move through” $R$ segments with both endpoints being internal vertices of the same maximal degree-2 path of $R$ and crossing it in opposing directions (called swollen segments). At this point, we are able to bound by $2^{\mathcal{O}(k)}$ the number of segments of the pushed weak linkage. Third, we eliminate all of the remaining U-turns, and show that then, the number of copies of each edge used must be at most $2^{\mathcal{O}(k)}$. We also modify the pushed weak linkage to be of a certain “canonical form” (see Section 8). + +**Generating a Collection of Pushed Weak Linkages.** In light of Lemma 2.4 and Proposition 2.1, it only remains to generate a collection of $2^{\mathcal{O}(k^2)}$ pushed weak linkages that includes all pushed weak linkages (of some canonical form) using at most $2^{\mathcal{O}(k)}$ copies of each edge. (This part, along with the preprocessing and construction of $R$, are the algorithmic parts of our proof.) + +This part of our proof is essentially a technical modification and adaptation of the work of Schrijver [42] (though we need to be more careful to obtain the bound $2^{\mathcal{O}(k^2)}$). Thus, we only give a brief description of it in the overview. Essentially, we generate pairs of a *pairing* and a *template*: a pairing assigns, to each vertex $v$ of $R$ of degree 1 or at least 3, a set of pairs of edges incident to $v$ to indicate that copies of these edges are to be visited consecutively (by at least one walk of the weak linkage under construction); a template further specifies, for each of the aforementioned pairs of edges, how many times copies of these edges are to be visited consecutively (but not which copies are paired-up). Clearly, there is a natural association of a pairing and a template to a pushed weak linkage. Further, we show that to generate all pairs of pairings and templates associated with the weak linkages we are interested in, we only need to consider pairings that in total have $\mathcal{O}(k)$ pairs and templates that assign numbers bounded by $2^{\mathcal{O}(k)}$ (because we deal with weak linkages using $2^{\mathcal{O}(k)}$ copies of each edge): + +**Lemma 2.5.** There is a collection of $2^{\mathcal{O}(k^2)}$ pairs of pairings and templates that, for any canonical pushed weak linkage $\mathcal{W}$ using only $2^{\mathcal{O}(k)}$ copies of each edge, contains a pair (of a pairing and a template) “compatible” with $\mathcal{W}$. Further, such a collection is computable in time $2^{\mathcal{O}(k^2)}$. + +Using somewhat more involved arguments (in Section 9), we also prove the following. + +**Lemma 2.6.** Any canonical pushed weak linkage is “compatible” with exactly one pair of a pairing and a template. Moreover, given a pair of a pairing and a template, if a canonical pushed weak linkage compatible with it exists, then it can be found in time polynomial in its size. \ No newline at end of file diff --git a/samples/texts/1754951/page_40.md b/samples/texts/1754951/page_40.md new file mode 100644 index 0000000000000000000000000000000000000000..1c5827715c53882a7e30c51a832e66cdb70706db --- /dev/null +++ b/samples/texts/1754951/page_40.md @@ -0,0 +1,17 @@ +Now, apply Lemma 7.1 to $\mathcal{P}(u, v)$, $\mathcal{C}(u, v)$ and $\text{Flow}_R(u, v)$ in $\text{Ring}(S_u, S_v)$. We thus obtain a linkage $\mathcal{P}'_{\text{traverse}}(u, v)$ disjoint from $\mathcal{P}_{\text{visitor}}(u, v)$ that is aligned with $\mathcal{P}_{\text{traverse}}(u, v)$. Hence, $\mathcal{P}'(u, v) = \mathcal{P}_{\text{visitor}}(u, v) \cup \mathcal{P}'_{\text{traverse}}(u, v)$ is a linkage in $G$ aligned with $\mathcal{P}(u, v)$. Assume w.l.o.g. that $\mathcal{P}'(u, v)$ uses the 3-rd copy of each edge in $H$. Let us now consider the winding number of $\mathcal{P}'_{\text{traverse}}(u, v)$ with respect to $\text{path}_R(u', v')$. By Lemma 7.1(c), $|\text{WindNum}(\mathcal{P}'_{\text{traverse}}(u, v)) - \text{WindNum}(\text{Flow}_R(u, v))| \le 60\ell + 6$. Now, note that for any path $P' \in \mathcal{P}'_{\text{traverse}}(u, v)$, $|\text{WindNum}(P') - \text{WindNum}(\mathcal{P}'_{\text{traverse}}(u, v))| \le 1$, (recall that the winding number of paths in a linkage differ by at most 1). Similarly, for any path $Q \in \text{Flow}_R(u, v)$, $|\text{WindNum}(\text{Flow}_R(u, v)) - \text{WindNum}(Q)| \le 1$. Therefore, it follows that for any two paths $P' \in \mathcal{P}'_{\text{traverse}}(u, v)$ and $Q \in \text{Flow}_R(u, v)$ $|\text{WindNum}(P') - \text{WindNum}(Q)| \le 60\ell + 8$. Recall that we chose $\eta(u, v)$ as the reference path of $\text{Ring}(S_u, s_v)$ in the above expression. Hence, we may rewrite it as $|\text{WindNum}(P', \eta(u, v)) - \text{WindNum}(Q, \eta(u, v))| \le 60\ell + 8$. Note that $P'$, $Q$ and $\eta(u, v)$ are three edge-disjoint paths traversing $\text{Ring}(S_u, S_v)$. Hence $\text{WindNum}$ is well defined in $\text{Ring}(S_u, S_v)$ for any pair of them. By Proposition 7.1, $|\text{WindNum}(P', Q)| \le |\text{WindNum}(P', \eta(u, v)) - \text{WindNum}(Q, \eta(u, v))| + 1$. We have so far established that for any $P' \in \mathcal{P}'_{\text{traverse}}(u, v)$ and $Q \in \text{Flow}_R(u, v)$, $|\text{WindNum}(P', Q)| \le 60\ell + 9$. Now, consider $\text{path}_R(u', v')$ and recall that for any $Q \in \text{Flow}_R(u, v)$, $|\text{WindNum}(\text{path}_R(u', v'), Q)| \le 1$ by Observation 7.4. Furthermore $\text{path}_R(u', v')$ uses the 0-th copies of edges in $H$. Hence, for each $P' \in \mathcal{P}'_{\text{traverse}}(u, v)$ $\text{WindNum}(P', \text{path}_R(u', v'))$ is well defined in $\text{Ring}(S_u, S_v)$, and by Proposition 7.1, $|\text{WindNum}(P', \text{path}_R(u', v'))| \le |\text{WindNum}(P', Q) - \text{WindNum}(\text{path}_R(u', v'), Q)| + 1 \le 60\ell + 11$. + +Finally, consider the paths in $\mathcal{P}'_{\text{visitor}} = \mathcal{P}_{\text{visitor}}$. For each $P \in \mathcal{P}_{\text{visitor}}$ the absolute value of the winding number of $\text{WindNum}(P, \text{path}_R(u', v'))$ is bounded by 1 (by Observation 7.1). Hence, we conclude that $\text{WindNum}(\mathcal{P}', \text{Ring}(S_u, S_v)) \le 60\alpha_{\text{sep}}(k) + 11 < 300 \cdot 2^{ck}$. $\square$ + +# 8 Pushing a Solution onto the Backbone Steiner Tree + +In this section, we push a linkage that is a solution of small winding number onto the backbone Steiner tree to construct a “pushed weak linkage” with several properties that will make its reconstruction (in Section 9) possible. Let us recall that we have an instance $(G, S, T, g, k)$ and $H$ is the radial completion of $G$ enriched with $4n + 1$ parallel copies of each edge. Then we construct a backbone Steiner tree $R$ (Section 6), which uses the 0-th copy of each edge. Formally, a pushed weak linkage is defined as follows. + +**Definition 8.1 (Pushed Weak Linkage).** Let $(G, S, T, g, k)$ be an instance of Planar Disjoint Paths, and $R$ be a Steiner tree. A weak linkage $W$ in $H$ is pushed (onto $R$) if $E(W) \cap E(R) = \emptyset$ and every edge in $E(W)$ is parallel to some edge of $R$. + +In what follows, we first define the properties we would like the pushed weak linkage to satisfy. Additionally, we partition any weak linkage into segments that give rise to a potential function whose maintenance will be crucial while pushing a solution of small winding number onto $R$. Afterwards, we show how to push the solution, simplify it, and analyze the result. We remark that the simplification (after having pushed the solution) will be done in two stages. + +## 8.1 Desired Properties and Potential of Weak Linkages + +The main property we would like a pushed weak linkage to satisfy is to use only “few” parallel copies of any edge. This quantity is captured by the following definition of multiplicity, which we will eventually like to upper bound by a small function of $k$. + +**Definition 8.2 (Multiplicity of Weak Linkage).** Let $H$ be a plane graph. Let $W$ be a weak linkage in $H$. Then, the multiplicity of $W$ is the maximum, across all edges $e$ in $H$, of the number of edges parallel to $e$ that belong to $E(W)$. \ No newline at end of file diff --git a/samples/texts/1754951/page_41.md b/samples/texts/1754951/page_41.md new file mode 100644 index 0000000000000000000000000000000000000000..dbfd6275c5ff4776f4c9a423d979ecd0eb2a1359 --- /dev/null +++ b/samples/texts/1754951/page_41.md @@ -0,0 +1,19 @@ +Figure 19: Segments, segment groups and their labeling. + +Towards bounding the multiplicity of the pushed weak linkage we construct, and also an important ingredient on its own in the reconstruction in Section 9, we need to repeatedly eliminate U-turns in the weak linkage we deal with. Here, U-turns are defined as follows. + +**Definition 8.3 (U-Turn in Weak Linkage).** Let $(G, S, T, g, k)$ be an instance of Planar Disjoint Paths, and $R$ be a Steiner tree. Let $\mathcal{W}$ be a weak linkage in $H$. Then, a U-turn in $\mathcal{W}$ is a pair of parallel edges $\{e, e'\}$ visited consecutively by some walk $W \in \mathcal{W}$ such that the strict interior of the cycle formed by $e$ and $e'$ does not contain the first or last edge of any walk in $\mathcal{W}$. We say that $\mathcal{W}$ is U-turn-free if it does not have any U-turn. + +Still, having a pushed weak linkage of low multiplicity and no U-turns does not suffice for faithful reconstruction due to ambiguity in which edge copies are being used. This ambiguity will be dealt with by the following definition. + +**Definition 8.4 (Canonical Weak Linkage).** Let $(G, S, T, g, k)$ be an instance of Planar Disjoint Paths, and $R$ be a Steiner tree. Let $\mathcal{W}$ be a weak linkage in $H$ pushed onto $R$. Then, $\mathcal{W}$ is canonical if (i) for every edge $e_i \in E(\mathcal{W})$, $i \ge 1$ (ii) if $e_i \in E(\mathcal{W})$, then all the parallel edges $e_j$, for $1 \le j < i$, are also in $E(\mathcal{W})$. + +For brevity, we say that a weak linkage $\mathcal{W}$ in $H$ is simplified if it is sensible, pushed onto $R$, canonical, U-turn-free and has multiplicity upper bounded by $\alpha_{\text{mul}}(k) := 2\alpha_{\text{potential}}(k)$ where $\alpha_{\text{potential}}(k) = 2^{\mathcal{O}(k)}$ will be defined precisely in Lemma 8.2. For the process that simplifies a pushed weak linkage, we will maintain a property that requires multiplicity at most $2n$ as well as a relaxation of canonicity. This property is defined as follows. + +**Definition 8.5 (Extremal Weak Linkage).** Let $(G, S, T, g, k)$ be an instance of Planar Disjoint Paths, and $R$ be a Steiner tree. Let $\mathcal{W}$ be a weak linkage in $H$ that is pushed onto $R$. Then, $\mathcal{W}$ is extremal if its multiplicity is at most $2n$ and for any two parallel edges $e_i, e_j \in E(\mathcal{W})$ where $i \ge 1$ and $j \le -1$, we have $(i-1) + |j+1| \ge 2n$. + +Additionally, we will maintain the following property. + +**Definition 8.6 (Outer-Terminal Weak Linkage).** Let $(G, S, T, g, k)$ be a nice instance of Planar Disjoint Paths, and $R$ be a Steiner tree. Let $\mathcal{W}$ be a weak linkage in $H$. Then, $\mathcal{W}$ is outer-terminal if it uses exactly one edge incident to $t^*$. + +**Segments, Segment Groups and Potential.** To analyze the “complexity” of a weak linkage, we partition it into segments and segment groups, and then associate a potential function with it based on this partition. Intuitively, a segment of a walk is a maximal subwalk that does not cross $R$ (see Fig. 19). Formally, it is defined as follows. \ No newline at end of file diff --git a/samples/texts/1754951/page_43.md b/samples/texts/1754951/page_43.md new file mode 100644 index 0000000000000000000000000000000000000000..052d7c796204086cdca340a6675547b1579ee164 --- /dev/null +++ b/samples/texts/1754951/page_43.md @@ -0,0 +1,23 @@ +**Lemma 8.1.** Let $(G, S, T, g, k)$ be a good Yes-instance of Planar Disjoint Paths, and $R$ be a backbone Steiner tree. Then, there exists a solution $\mathcal{P}$ to $(G, S, T, g, k)$ such that $|\text{SegGro}(\mathcal{P})| \le \alpha_{\text{segGro}}(k) := 10^5 k \cdot 2^{ck}$. + +*Proof.* First, notice that for every path $P \in \mathcal{P}$, every segment group of $P$ whose endpoints are both internal vertices of some maximal degree-2 path of $R$ has the following property: it is neither a prefix nor a suffix of $P$ (when $P$ is oriented arbitrarily, say, from its endpoint in $S$ to its endpoint in $T$), and the segment groups that appear immediately before and after it necessarily satisfy that each of them does not have both of its endpoints being internal vertices of some maximal degree-2 path of $R$. Let $s$ denote the number of segment groups of $\mathcal{P}$ that do not have both of their endpoints being internal vertices of some maximal degree-2 path of $R$. Then, we have that $|\text{SegGro}(\mathcal{P})| \le 2s - 1$, and therefore to complete the proof, it suffices to show that $s \le \alpha_{\text{segGro}}(k)/2$. + +Let $\{u_1, v_1\}, \{u_2, v_2\}, \dots, \{u_t, v_t\}$ denote the pairs of near vertices in $V_{=1}(R) \cup V_{\ge 3}(R)$ such that for each $i \in \{1, \dots, t\}$, $\text{path}_R(u_i, v_i)$ is a long maximal degree-2 path in $R$. Let $\text{Ring}(S_{u_1}, S_{v_1}), \text{Ring}(S_{u_2}, S_{v_2}), \dots, \text{Ring}(S_{u_t}, S_{v_t})$ be the corresponding rings. By Lemma 7.2, the number of vertices of $R$ lying outside these rings, $|V(R) \setminus \bigcup_{i=1}^t V(S_{u_i}, S_{v_i})|$, is upper bounded by $\alpha_{\text{nonRing}}(k) = 10^4 k \cdot 2^{ck}$. We further classify each segment group $S$ of $\mathcal{P}$ that does not have both of its endpoints being internal vertices of some maximal degree-2 path of $R$ as follows. (We remark that, by the definition of a segment group, $S$ must consist of a single segment.) + +* $S$ has at least one endpoint in $V(R) \setminus \bigcup_{i=1}^t V(S_{u_i}, S_{v_i})$. Denote the number of such segment groups by $s_1$. + +* $S$ has one endpoint in $V(S_{u_i}, S_{v_i})$ and another endpoint in $V(S_{u_j}, S_{v_j})$ for some $i, j \in \{1, \dots, t\}$ such that $i \neq j$. Denote the number of such segment groups by $s_2$. + +Then, $s_1 + s_2 = s$. Now, notice that the paths in $\mathcal{P}$ are pairwise vertex-disjoint, and the collection of segment groups of each path $P \in \mathcal{P}$ forms a partition of $P$. Thus, because $|V(R) \setminus \bigcup_{i=1}^t V(S_{u_i}, S_{v_i})| \le \alpha_{\text{nonRing}}(k)$ and each vertex in $V(R)$ is shared as an endpoint by at most two segment groups, we immediately derive that $s_1 \le \alpha_{\text{nonRing}}(k)$. To bound $s_2$, note that each segment group of the second type traverses at least one vertex in $S_{u_i} \cup S_{v_i}$ for some $i \in \{1, \dots, t\}$ (due to Lemma 6.12). By Lemma 6.4, for every $i \in \{1, \dots, t\}$, $|S_{u_i}|, |S_{v_i}| \le \alpha_{\text{sep}}(k) = \frac{7}{2} \cdot 2^{ck} + 2$. Moreover, by Observation 6.1, $t < 4k$. Thus, we have that $s_2 \le 4 \cdot 4k \cdot \alpha_{\text{sep}}(k)$, where the multiplication by 4 is done because two segment groups can share an endpoint and each maximal degree-2 path is associated with two separators. From this, we conclude that $s \le 10^4 k \cdot 2^{ck} + 16k (\frac{7}{2} \cdot 2^{ck} + 2) \le \alpha_{\text{segGro}}(k)/2$. $\square$ + +Now, based on Observation 7.2 and Lemmas 7.3 and 8.1, we derive the existence of a solution with low potential (if there exists a solution). + +**Lemma 8.2.** Let $(G, S, T, g, k)$ be a good Yes-instance of Planar Disjoint Paths, and $R$ be a backbone Steiner tree. Then, there exists a solution $\mathcal{P}$ to $(G, S, T, g, k)$ such that Potential$(\mathcal{P}) \le \alpha_{\text{potential}}(k) := (10^4 \cdot 4^{ck} + 1) \cdot \alpha_{\text{segGro}}(k)$. + +*Proof.* Let $\mathcal{Q}$ be a solution with the property in Lemma 7.3. Let $\mathcal{S}$ denote the segment groups in $\text{SegGro}(\mathcal{Q})$ whose both endpoints belong to the same maximal degree-2 path of $R$. For each segment group $\mathcal{S} \in \mathcal{S}$, denote $\text{labPot}(\mathcal{S}) = |\sum_{(e,e') \in E(\mathcal{S}) \times E(\mathcal{S})} \text{label}_{\mathcal{P}_\mathcal{S}}^{\mathcal{S}}(e, e')|$ where $\mathcal{P}_\mathcal{S}$ is the path in $\mathcal{Q}$ that has $\mathcal{S}$ as a segment group. Furthermore, denote $\mathcal{M} = \max_{\mathcal{S} \in \mathcal{S}} \text{labPot}(\mathcal{S})$. Now, by the definition of potential, + +$$ +\operatorname{Potential}(\mathcal{Q}) = |\operatorname{SegGro}(\mathcal{Q})| + \sum_{\mathcal{S} \in \mathcal{S}} \operatorname{labPot}(\mathcal{S}) \le (M+1)|\operatorname{SegGro}(\mathcal{Q})|. +$$ + +By Lemma 8.1, $|\text{SegGro}(\mathcal{Q})| \le \alpha_{\text{segGro}}(k)$. Thus, to complete the proof, it suffices to show that $M \le 10^4 \cdot 4^{ck}$. To this end, consider some segment group $\mathcal{S} \in \mathcal{S}$. Notice that $\text{labPot}(\mathcal{S})$ is \ No newline at end of file diff --git a/samples/texts/1754951/page_44.md b/samples/texts/1754951/page_44.md new file mode 100644 index 0000000000000000000000000000000000000000..08ac2624eccc6d65409de3940371e298309f21a5 --- /dev/null +++ b/samples/texts/1754951/page_44.md @@ -0,0 +1,31 @@ +upper bounded by the number of crossings of $S$ with $P_S$. Thus, if $P_S$ is short, it follows that $\text{labPot}(S) < \alpha_{\text{long}}(k) = 10^4 \cdot 2^{ck}$. + +Now, suppose that $P_S$ is long, and let $\text{Ring}(S_u, S_v)$ be the ring that corresponds to $P_S$. By the choice of $\mathcal{Q}$, $|\text{WindNum}(\mathcal{Q}, \text{Ring}(S_u, S_v))| \le \alpha_{\text{winding}}(k) < 300 \cdot 2^{ck}$. Let $\tilde{\mathcal{A}}$ be a collection of maximal subpaths of $S$ that are fully contained within $\text{Ring}(S_u, S_v)$, and let $\hat{P}_S$ be the maximal subpath of $P_S$ that is fully contained within $\text{Ring}(S_u, S_v)$. Then, by Observation 7.2, + +$$ \text{labPot}(S) \le |V(P_S) \setminus V(\hat{P}_S)| + \sum_{\hat{\mathcal{A}} \in \tilde{\mathcal{A}}} |\text{WindNum}(\hat{\mathcal{A}}, \hat{P}_S)|. $$ + +By the definition of WindNum, we have that $|\text{WindNum}(\hat{\mathcal{A}}, \hat{P}_S)| \le |\text{WindNum}(\mathcal{Q}, \text{Ring}(S_u, S_v))|$ for each $\hat{\mathcal{A}} \in \tilde{\mathcal{A}}$. Moreover, by Lemma 6.4, $|\tilde{\mathcal{A}}| \le |S_u| + |S_v| \le 2\alpha_{\text{sep}}(k) = 7 \cdot 2^{ck} + 4$. Additionally, by Lemmas 6.5 and 6.11, we have that $|V(P_S) \setminus V(\hat{P}_S)| \le 2\alpha_{\text{pat}}(k) = 200 \cdot 2^{ck}$. From this, + +$$ \text{labPot}(S) \le 200 \cdot 2^{ck} + (7 \cdot 2^{ck} + 4) \cdot 300 \cdot 2^{ck} \le 10^4 \cdot 4^{ck}. $$ + +Thus, because the choice of $S$ was arbitrary, we conclude that $M \le 10^4 \cdot 4^{ck}$. $\square$ + +## 8.2 Pushing a Solution onto R + +Let us now describe the process of pushing a solution onto $R$. To simplify this process, we define two “non-atomic” operations that encompass sequences of atomic operations in discrete homotopy. We remind that we only deal with walks that do not repeat edges. + +**Definition 8.11 (Non-Atomic Operations in Discrete Homotopy).** Let $G$ be a triangulated plane graph with a weak linkage $W$, and $C$ be a cycle¹⁷ in $G$. Let $W \in W$. + +* **Cycle Move.** Applicable to $(W, C)$ if there exists a subpath $Q$ of $C$ such that (i) $Q$ is a subpath of $W$, (ii) $1 \le |E(Q)| \le |E(C)| - 1$, (iii) no edge in $E(C) \setminus E(Q)$ belongs to any walk in $W$, and (iii) no edge drawn in the strict interior of $C$ belongs to any walk in $W$. Then, the cycle move operation replaces $Q$ in $W$ by the unique subpath of $C$ between the endpoints of $Q$ that is edge-disjoint from $Q$. + +* **Cycle Pull.** Applicable to $(W, C)$ if (i) $C$ is a subwalk of $W$, and (ii) no edge drawn in the strict interior of $C$ belongs to any walk in $W$. Then, the cycle pull operation replaces $Q$ in $W$ by a single occurrence of the first vertex in $Q$. + +We now prove that the operations above are compositions of atomic operations. + +**Lemma 8.3.** Let $G$ be a triangulated plane graph with a weak linkage $W$, and $C$ be a cycle in $G$. Let $W \in W$ with a non-atomic operation applicable to $(W, C)$. Then, the result of the application is a weak linkage that is discretely homotopic to $W$. + +*Proof.* We prove the claim by induction on the number of faces of $G$ in the interior of $C$. In the basis, where $C$ encloses only one face, then the cycle move and cycle pull operations are precisely the face move and face pull operations, respectively, and therefore the claim holds. Now, suppose that $C$ encloses $i \ge 2$ faces and that the claim is correct for cycles that enclose at most $i-1$ faces. We consider several cases as follows. + +First, suppose that $C$ has a path $P$ fully drawn in its interior whose endpoints are two (distinct) vertices $u, v \in V(C)$, and whose internal vertices and all of its edges do not belong to $C$. (We remark that $P$ might consist of a single edge, and that edge might be parallel to some edge of $C$.) Now, notice that $P$ partitions the interior of $C$ into the interior of two cycles $C_1$ and $C_2$ that share only $P$ in common as follows: one cycle $C_1$ consists of one subpath of $C$ + +¹⁷A pair of parallel edges is considered to be a cycle. \ No newline at end of file diff --git a/samples/texts/1754951/page_46.md b/samples/texts/1754951/page_46.md new file mode 100644 index 0000000000000000000000000000000000000000..a980f42c678334d5843c13c5843f6a7a5f3a6b52 --- /dev/null +++ b/samples/texts/1754951/page_46.md @@ -0,0 +1,32 @@ +Figure 20: A Sequence, its projecting cycle and a shrinking cycle. + +between $u$ and $v$ and the path $P$, and the other cycle $C_2$ consists of the second subpath of $C$ +between $u$ and $v$ and the path $P$. Notice that $C_1$ encloses less faces than $C$, and so does $C_2$. +At least one of these two cycles, say, $C_1$, contains at least one edge of $Q$. Then, the cycle move +operation is applicable to $(W, C_1)$. Indeed, let $\hat{Q}$ be the subpath of $Q$ that is a subpath of $C$, and +notice that $E(P) \subseteq E(C_1) \subseteq E(C) \cup E(P)$ and $E(P) \cap E(W) = \emptyset$ (because the cycle move/pull +operation is applicable to $(W, C)$). Therefore, $\hat{Q}$ is a subpath of $W$, $1 \le |E(\hat{Q})| \le |E(C_1)| - 1$, +and no edge in $E(C_1) \setminus E(\hat{Q})$ belongs to any walk in $W$. Moreover, because $C_1$ belongs to the +interior (including the boundary) of $C$, no edge drawn in the strict interior of $C$ belongs to +any walk in $W$. Now, notice that after the application of the cycle move operation for $(W, C_1)$, +$C_2$ also has at least one edge used by the walk $W'$ into which $W$ was modified—in particular, +$E(P) \subseteq E(W')$. Moreover, consider the subpath (or subwalk that is a cycle) $Q'$ of $W'$ that +results from the replacement of $\hat{Q}$ in $Q$ by the subpath of $C_1$ between the endpoints of $Q'$ that +does not belong to $W$. Then, $Q'$ traverses some subpath (possible empty) of $C_1$ or $C_2$, then +traverses $P$, and next traverses some other subpath of $C_1$ or $C_2$. So, the restriction of $Q'$ to $C_2$ is +a non-empty path or cycle $Q^*$ that is a subwalk of $W'$. Furthermore, because $C_2$ is drawn in the +interior of $C$ and the cycle move/pull operation is applicable to $(W, C)$, we have that no edge of +$E(C_2) \setminus E(Q^*)$ or the strict interior of $C_2$ belongs to $E(W)$. Thus, the cycle move/pull operation +is applicable to $(W', C_2)$. Now, the result of the application of this operation is precisely the +result of the application of the original cycle move or pull operation applicable to $(W, C)$. To +see this, observe that the edges of $E(C) \setminus E(W)$ that occur in $C_1$ along with $E(P)$ have replaced +the edges of $E(C) \cap E(W)$ that occur in $C_1$ in the first operation, and the edges of $E(C) \setminus E(W)$ +that occur in $C_2$ have replaced the edges of $E(C) \cap E(W)$ that occur in $C_1$ along with $E(P)$ in +the second operation. Thus, by the inductive hypothesis with respect to $(W, C_1)$ and $(W', C_2)$, +and because discrete homotopy is transitive, the claim follows. + +Thus, it remains to prove that *C* has a path *P* fully drawn in its interior whose endpoints are two (distinct) vertices *u*, *v* ∈ *V*(*C*), and whose internal vertices and all of its edges do not belong to *C*. In case *C* has a chord (that is, an edge in *G* between two vertices of *C* that does not belong to *C*), then the chord is such a path *P*. Therefore, we now suppose that this is not the case. Then, *C* does not contain in its interior an edge parallel to an edge of *C*. In turn, because *G* is triangulated), when we consider some face *f* in the interior of *C* that contains an edge *e* of *C*, this face must be a triangle. Moreover, the vertex of *f* that is not incident to *e* cannot belong to *C*, since otherwise we obtain a chord in *C*. Thus, the subpath (that consists of two edges) of *f* between the endpoints of *e* that does not contain *e* is a path *P* with the above mentioned properties. □ + +In the process of pushing a solution onto R, we push parts of the solution one-by-one. We refer to these parts as sequences, defined as follows (see Fig. 20). + +**Definition 8.12 (Sequence).** Let (G, S, T, g, k) be an instance of Planar Disjoint Paths, and R be a Steiner tree. Let W be a walk. Then, a sequence of W is a maximal subwalk of W whose \ No newline at end of file diff --git a/samples/texts/1754951/page_47.md b/samples/texts/1754951/page_47.md new file mode 100644 index 0000000000000000000000000000000000000000..7363b96926070f7e95844029dccb3cba89e1e66a --- /dev/null +++ b/samples/texts/1754951/page_47.md @@ -0,0 +1,27 @@ +internal vertices (if any exists) do not belong to $R$ and which contains at least one edge that is not parallel to an edge of $R$. The set of sequences of $W$ is denoted by $\text{Seq}(W)$. For a weak linkage $W$, the set of sequences of $W$ is defined as $\text{Seq}(W) = \bigcup_{W' \in W} \text{Seq}(W')$. + +Notice that the set of sequences of a walk does not necessarily form a partition of the walk because the walk can traverse edges parallel to the edges of $R$ and these edges do not belong to any sequence. Moreover, for sensible weak linkages, the endpoints of every sequence belong to $R$. To deal only with sequences that are paths, we need the following definition. + +**Definition 8.13 (Well-Behaved Weak Linkage).** Let $(G, S, T, g, k)$ be an instance of Planar Disjoint Paths, and $R$ be a Steiner tree. A weak linkage $W$ is well-behaved if every sequence in $\text{Seq}(W)$ is a path or a cycle. + +When we will push sequences one-by-one, we ensure that the current sequence to be pushed can be handled by the cycle move operation in Definition 8.11. To this end, we define the notion of an innermost sequence, based on another notion called a projecting cycle (see Fig. 20). We remark that this cycle will not necessarily be the one on which we apply a cycle move operation, since this cycle might contain in its interior edges of some walks of the weak linkage. + +**Definition 8.14 (Projecting Cycle).** Let $(G, S, T, g, k)$ be an instance of Planar Disjoint Paths, and $R$ be a Steiner tree. Let $W$ be a sensible well-behaved weak linkage, and $S \in \text{Seq}(W)$. The projecting cycle of $S$ is the cycle $C$ formed by $S$ and the subpath $P$ of $R$ between the endpoints of $S$. Additionally, $\text{Volume}(S)$ denotes the number of faces enclosed by the projecting cycle of $S$. + +Now, we define the notion of an innermost sequence. + +**Definition 8.15 (Innermost Sequence).** Let $(G, S, T, g, k)$ be an instance of Planar Disjoint Paths, and $R$ be a Steiner tree. Let $W$ be a sensible well-behaved weak linkage, and $S \in \text{Seq}(W)$. Then, $S$ is innermost if every edge in $E(W)$ that is drawn in the interior of its projecting cycle is parallel to some edge of $R$. + +We now argue that, unless the set of sequences is empty, there must exist an innermost one. + +**Lemma 8.4.** Let $(G, S, T, g, k)$ be an instance of Planar Disjoint Paths, and $R$ be a Steiner tree. Let $W$ be a sensible well-behaved weak linkage such that $\text{Seq}(W) \neq \emptyset$. Then, there exists an innermost sequence in $\text{Seq}(W)$. + +*Proof.* Let $S \in \text{Seq}(W)$ be a sequence that minimizes $\text{Volume}(S)$. We claim that $S$ is innermost. Suppose, by way of contradiction, that this claim is false. Then, there exists an edge $e \in E(W)$ that is drawn in the interior of the projecting cycle of $S$ and is not parallel to any edge of $R$. Thus, $e$ belongs to some sequence $S' \in \text{Seq}(W)$. Because $W$ is well-behaved, $S$ and $S'$ are vertex disjoint. This implies that the projecting cycle of $S'$ is contained in the interior of the projecting cycle of $S$. Because $H$ is triangulated, this means that $\text{Volume}(S') < \text{Volume}(S)$, which is a contradiction to the choice of $S$. $\square$ + +When we push the sequence onto $R$, we need to ensure that we have enough copies of each edge of $R$ to do so. To this end, we need the following definition. + +**Definition 8.16 (Shallow Weak Linkage).** Let $(G, S, T, g, k)$ be an instance of Planar Disjoint Paths, and $R$ be a Steiner tree. Let $W$ be a sensible well-behaved weak linkage. Then, $W$ is shallow if for every edge $e_0 \in E(R)$, the following condition holds. Let $\ell$ (resp. $h$) be the number of sequences $S \in \text{Seq}(W)$ whose projecting cycle encloses $e_1$ (resp. $e_{-1}$). Then, $e_i$ is not used by $W$ for every $i \in \{-n, -n+1, \dots, -n+\ell-1\} \cup \{0\} \cup \{n-h+1, n-h+2, \dots, n\}$. + +To ensure that we make only cycle moves/pulls as in Definition 8.11, we do not necessarily push the sequence at once, but gradually shrink the area enclosed by its projecting cycle.¹⁸ + +¹⁸Instead, we could have also always pushed a sequence at once by defining moves and pulls for closed walks, which we find somewhat more complicated to analyze formally. \ No newline at end of file diff --git a/samples/texts/1754951/page_48.md b/samples/texts/1754951/page_48.md new file mode 100644 index 0000000000000000000000000000000000000000..229589522d1674d23a61e9b844ca858d95defc99 --- /dev/null +++ b/samples/texts/1754951/page_48.md @@ -0,0 +1,15 @@ +Figure 21: An illustration of Lemma 8.5 + +**Definition 8.17 (Shrinking Cycle).** Let $(G, S, T, g, k)$ be an instance of Planar Disjoint Paths, and $R$ be a Steiner tree. Let $\mathcal{W}$ be a sensible well-behaved weak linkage, and $S \in \text{Seq}(\mathcal{W})$ with an endpoint $v \in V(R)$. Then, a cycle $C$ in $H$ is a shrinking cycle for $(S, v)$ if it has no edge of $R$ in its interior and it can be partitioned into three paths where the first has at least one edge and the last has at most one edge: (i) a subpath $P_1$ of $S$ with $v$ as an endpoint; (ii) a subpath $P_2$ from the other endpoint $u$ of $P_1$ to a vertex on $R$ that consists only of edges drawn in the strict interior of the projecting cycle of $S$ and no vertex of $S$ apart from $u$; (iii) a subpath $P_3$ that has $v$ as an endpoint and whose edge (if one exists) is either not parallel to any edge of $R$ or it is the $i$-th copy of an edge parallel to some edge of $R$ for some $i \in \{-n+\ell-1, n-h+1\}$, where $\ell$ and $h$ are as in Definition 8.16. + +With respect to shrinking cycles, we prove two claims. First, we assert their existence. + +**Lemma 8.5.** Let $(G, S, T, g, k)$ be an instance of Planar Disjoint Paths, and $R$ be a Steiner tree. Let $\mathcal{W}$ be a sensible well-behaved weak linkage, and $S \in \text{Seq}(\mathcal{W})$ with an endpoint $v \in V(R)$. Then, there exists a shrinking cycle for $(S, v)$. + +*Proof.* Let $e$ be an edge in $S$ incident to $v$ (if there are two such edges, when $S$ is a cycle, pick one of them arbitrarily), and denote the other endpoint of $e$ by $u$. Because $H$ is triangulated, $e$ belongs to the boundary $B$ of a face $f$ of $H$ in the interior of $C$ such that $B$ is a cycle that consists of only two or three edges. If $B$ does not contain any vertex of $V(R) \cup V(S)$ besides $u$ and $v$, then it is clearly a shrinking cycle (see Fig. 21). Thus, we now suppose that $B$ is a cycle on three edges whose third vertex, $w$, belongs to $V(R) \cup V(S)$. If $w \in V(S)$, then the cycle that consists of the subpath of $S$ from $v$ to $w$ and the edge in $E(B)$ between $v$ and $w$ is also clearly a shrinking cycle (see Fig. 21). Thus, we now also suppose that $w \in V(R)$. + +We further distinguish between two cases. First, suppose that $w$ is not adjacent to $v$ on $R$. In this case, $B$ does not enclose any edge of $R$ as well as any edge parallel to an edge of $R$. Moreover, $B$ can be partitioned into $P_1, P_2$ and $P_3$ that are each a single edge, where $P_1$ consists of the edge in $B$ between $v$ and $u$, $P_2$ consists of the edge in $B$ between $u$ and $w$, and $P_3$ consists of the edge in $B$ between $w$ and $v$, thereby complying with Definition 8.17. Thus, $B$ is a shrinking cycle for $(S, v)$. Now, suppose that $w$ is adjacent to $v$ on $R$. Then, define $P_1, P_2$ and $P_3$ similarly to before except that to $P_3$, we do not take the edge of $B$ between $v$ and $w$ but its parallel $i$-th copy where $i \in \{-n+\ell-1, n-h+1\}$ such that $\ell$ and $h$ are as in Definition 8.16. The choice of whether $i = -n+\ell-1$ or $i = n-h+1$ is made so that the cycle $B'$ consisting of $P_1, P_2$ and $P_3$ does not enclose any edge of $R$. (Such a choice necessarily exists, see Fig. 21). □ + +Now, we prove that making a cycle move/pull operation using a shrinking cycle is valid and maintains some properties of weak linkages required for our analysis. + +**Lemma 8.6.** Let $(G, S, T, g, k)$ be an instance of Planar Disjoint Paths, and $R$ be a Steiner tree. Let $\mathcal{W}$ be a sensible, well-behaved, shallow and outer-terminal weak linkage, and $S \in \text{Seq}(\mathcal{W})$ be innermost with an endpoint $v \in V(R) \setminus \{t^*\}$. Let $C$ be a shrinking cycle for $(S, v)$ that encloses as many faces as possible. Then, the cycle move/pull operation is applicable to $(W, C)$ where $W \in \mathcal{W}$ is the walk having $S$ as a sequence. Furthermore, the resulting weak linkage \ No newline at end of file diff --git a/samples/texts/1754951/page_5.md b/samples/texts/1754951/page_5.md new file mode 100644 index 0000000000000000000000000000000000000000..993507dfa6393a8a573a49ad5de1f420f0a5d4a1 --- /dev/null +++ b/samples/texts/1754951/page_5.md @@ -0,0 +1,15 @@ +Figure 11: A walk going back and forth along a path of $R$, which gives rise to $U$-turns. + +These two lemmas complete the proof: we can indeed generate a collection of $2^{\mathcal{O}(k^2)}$ pushed weak linkages containing all canonical pushed weak linkages using only $2^{\mathcal{O}(k)}$ copies of any edge. + +## 3 Preliminaries + +Let $A$ be a set of elements. A cyclic ordering $\prec$ on $A$ is an ordering $(a_0, a_1, \ldots, a_{|A|-1})$ of the elements in $A$ such that, by enumerating $A$ in clockwise order starting at $a_i \in A$, we refer to the ordering $a_i, a_{(i+1)} \bmod |A|, \ldots, a_{(i+|A|-1)} \bmod |A|$, and by enumerating $A$ in counter-clockwise order starting at $a_i \in A$, we refer to the ordering $a_i, a_{(i-1)} \bmod |A|, \ldots, a_{(i-|A|+1)} \bmod |A|$. We consider all cyclic orderings of $A$ that satisfy the following condition to be the equivalent (up to cyclic shifts): the enumeration of $A$ in cyclic clockwise order starting at $a_i$, for any $a_i \in A$, produces the same sequence. For a function $f : X \to Y$ and a subset $X' \subseteq X$, we denote the restriction of $f$ to $X'$ by $f|_{X'}$. + +**Graphs.** Given an undirected graph $G$, we let $V(G)$ and $E(G)$ denote the vertex set and edge set of $G$, respectively. Similarly, given a directed graph (digraph) $D$, we let $V(D)$ and $A(D)$ denote the vertex set and arc set of $D$, respectively. Throughout the paper, we deal with graphs without self-loops but with parallel edges. Whenever it is not explicitly written otherwise, we deal with undirected graphs. Moreover, whenever $G$ is clear from context, denote $n = |V(G)|$. + +For a graph $G$ and a subset of vertices $U \subseteq V(G)$, the subgraph of $G$ induced by $U$, denoted by $G[U]$, is the graph on vertex set $U$ and edge set $\{ \{u, v\} \in E(G) : u, v \in U \}$. Additionally, $G-U$ denotes the graph $G[V(G) \setminus U]$. For a subset of edges $F \subseteq E(G)$, $G-F$ denotes the graph on vertex set $V(G)$ and edge set $E(G) \setminus F$. For a vertex $v \in V(G)$, the set of neighbors of $v$ in $G$ is denoted by $N_G(v)$, and for a subset of vertices $U \subseteq V(G)$, the open neighborhood of $U$ in $G$ is defined as $N_G(U) = \bigcup_{v \in U} N_G(v) \setminus U$. Given three subsets of vertices $A, B, S \subseteq V(G)$, we say that *S separates* $A$ from $B$ if $G-S$ has no path with an endpoint in $A$ and an endpoint in $B$. For two vertices $u, v \in V(G)$, the distance between $u$ and $v$ in $G$ is the length (number of edges) of the shortest path between $u$ and $v$ in $G$ (if no such path exists, then the distance is $\infty$), and it is denoted by $\text{dist}_G(u, v)$; in case $u = v$, $\text{dist}_G(u, v) = 0$. For two subsets $A, B \subseteq V(G)$, define $\text{dist}_G(A, B) = \min_{u \in A, v \in B} \text{dist}_G(u, v)$. A linkage of order $k$ in $G$ is an ordered family $\mathcal{P}$ of $k$ vertex-disjoint paths in $G$. Two linkages $\mathcal{P} = (P_1, \ldots, P_k)$ and $\mathcal{Q} = (Q_1, \ldots, Q_k)$ are aligned if for all $i \in \{1, \ldots, k\}$, $P_i$ and $Q_i$ have the same endpoints. + +For a tree $T$ and $d \in \mathbb{N}$, let $V_{\ge d}(T)$ (resp. $V_d(T)$) denote the set of vertices of degree at least (resp. exactly $d$ in $T$. For two vertices $u, v \in V(T)$, the unique subpath of $T$ between $u$ and $v$ is denoted by $\textbf{path}_T(u, v)$. We say that two vertices $u, v \in V(T)$ are *near each other* if $\textbf{path}_T(u, v)$ has no internal vertex from $V_{\ge 3}(T)$, and call $\textbf{path}_T(u, v)$ a *degree-2 path*. In case $u, v \in V_{\ge 3}(T) \cup V_{=1}(T)$, $\textbf{path}_T(u, v)$ is called a *maximal degree-2 path*. + +**Planarity.** A planar graph is a graph that can be embedded in the Euclidean plane, that is, there exists a mapping from every vertex to a point on a plane, and from every edge to a plane curve on that plane, such that the extreme points of each curve are the points mapped to the endpoints of the corresponding edge, and all curves are disjoint except on their extreme points. A *plane graph* $G$ is a planar graph with a fixed embedding. Its faces are the regions bounded by the edges, including the outer infinitely large region. For every vertex $v \in V(G)$, we let $E_G(v) = (e_0, e_1, \ldots, e_{t-1})$ for $t \in \mathbb{N}$ where $e_0, e_1, \ldots, e_{t-1}$ are the edges incident to $v$ in clockwise order (the decision which edge is $e_0$ is arbitrary). A planar graph $G$ is *triangulated* if \ No newline at end of file diff --git a/samples/texts/1754951/page_50.md b/samples/texts/1754951/page_50.md new file mode 100644 index 0000000000000000000000000000000000000000..743bc4cdfc190fa96115fe00802a07e680107ef9 --- /dev/null +++ b/samples/texts/1754951/page_50.md @@ -0,0 +1,16 @@ +is pushed onto $R$, has multiplicity at most $2n$, discretely homotopic in $H$ to some solution of +$(G, S, T, g, k)$ and Potential($W$) $\le \alpha_{\text{potential}}(k)$. + +*Proof.* By Lemma 8.2, there exists a solution $\mathcal{P}$ to $(G, S, T, g, k)$ such that Potential($\mathcal{P}$) $\le \alpha_{\text{potential}}(k)$. Because the paths in $\mathcal{P}$ are pairwise vertex-disjoint and $\mathcal{P}$ has a path where $t^*$ is an endpoint, it is clear that $\mathcal{P}$ is a sensible, well-behaved, shallow and outer-terminal weak linkage. Since $\mathcal{P}$ is discretely homotopic to itself, it is well defined to let $\mathcal{W}$ a weak linkage that, among all sensible, well-behaved, shallow and outer-terminal weak linkages that are discretely homotopic to $\mathcal{P}$, minimizes $\sum_{S \in \text{Seq}(\mathcal{W})} \text{Volume}(S)$. Notice that shallowness is a stronger demand than having multiplicity at most $2n$, and that being pushed onto $R$ is equivalent to having an empty set of sequences. Thus, to conclude the proof, it suffices to argue that $\text{Seq}(\mathcal{W}) = \emptyset$. + +Suppose, by way of contradiction, that $\text{Seq}(\mathcal{W}) \neq \emptyset$. Then, by Lemmas 8.4 and 8.5, there exist an innermost sequence $S \in \text{Seq}(\mathcal{W})$ and a shrinking cycle $C$ for $(S, v)$ where we pick $v$ as an endpoint of $S$ that is not $t^*$ (because $\mathcal{W}$ is outer-terminal, not both endpoints of $S$ can be $t^*$), and we pick a shrinking cycle enclosing as many faces as possible. By Lemma 8.6, the cycle move/pull operation is applicable to $(\mathcal{W}, C)$ where $\mathcal{W} \in \mathcal{W}$ is the walk having $S$ as a sequence. Furthermore, the resulting weak linkage $\mathcal{W}'$ is sensible, well-behaved, shallow and outer-terminal having the same potential as $\mathcal{W}$, and $\sum_{\hat{S} \in \text{Seq}(\mathcal{W}')} \text{Volume}(\hat{S}) < \sum_{S \in \text{Seq}(\mathcal{W})} \text{Volume}(S)$. Since discrete homotopy is an equivalence relation, $\mathcal{W}'$ is discretely homotopic to $\mathcal{P}$. However, this contradicts the choice of $\mathcal{W}$. $\square$ + +## 8.3 Bounding the Total Number of Segments + +Having pushed the solution onto $R$, we further need to make the resulting weak linkage simplified, which requires to make it have low multiplicity, be U-turn-free and canonical. We first show that we can focus only on the first two properties, as being canonical can be easily derived using cycle move operations on cycles consisting of two parallel edges. + +**Lemma 8.8.** Let $(G, S, T, g, k)$ be an instance of Planar Disjoint Paths, and $R$ be a Steiner tree. Let $\mathcal{W}$ be a weak linkage in $H$ that is sensible and pushed onto $R$, whose multiplicity is at most $2n$. Then, there exists a weak linkage $\mathcal{W}'$ that is sensible, pushed onto $R$, canonical, discretely homotopic to $\mathcal{W}$, and whose multiplicity is upper bounded by the multiplicity of $\mathcal{W}$. + +*Proof.* Let us first argue that, there is a weak linkage $\mathcal{W}'$ that is sensible, pushed onto $R$, discretely homotopic to $\mathcal{W}$, and whose multiplicity is upper bounded by the multiplicity of $\mathcal{W}$ such that for all edges $e_i \in E(\mathcal{W}')$ $i \ge i$. In other words, $\mathcal{W}'$ contains only edges of positive subscript. Consider all weak linkages in $H$ that are sensible, pushed onto $R$, discretely homotopic to $\mathcal{W}$, and whose multiplicity is upper bounded by the multiplicity of $\mathcal{W}$. Among these weak linkages, let $\mathcal{W}'$ be one such that the sum of the subscripts of the edge copies in $E(\mathcal{W}')$ is maximized. Let us argue that for every $e_i \in E(\mathcal{W})$, $i \ge 1$ and for every $j > i$, $e_j \in E(\mathcal{W})$. Suppose not, and there exists an edge $e_i$ that is used by $\mathcal{W}'$ such that $(i)$ $i \le 0$ (in fact, $i = 0$ is not possible as $\mathcal{W}'$ is pushed onto $R$), or (ii) there exists an edge $e_j$ (parallel to $e_i$) for some $1 \le j < i$ that is not used by $E(\mathcal{W}')$. Since $\mathcal{W}'$ has multiplicity at most $2n$, the satisfaction of the first condition implies the satisfaction of the second one, thus there exists such an edge $e_j$. Let $e_t$ be the edge (parallel to $e_j$ and $e_i$) of largest $j$ that is used by $E(\mathcal{W}')$. Moreover, let $C$ be the cycle (which might be the boundary of a single face) that consists of two edges: $e_j$ and $e_t$. By the choice of $e_t$, the strict interior of $C$ does not contain any edge of $E(\mathcal{W}')$. Thus, the cycle move operation is applicable to $(\mathcal{W}, C)$ where $\mathcal{W}$ is the walk in $\mathcal{W}'$ that uses $e_t$. Let $\mathcal{W}^*$ be the result of the application of this operation. Then, the only difference between $\mathcal{W}^*$ and $\mathcal{W}'$ is the replacement of $e_t$ by $e_j$. + +Because $\mathcal{W}^*$ is discretely homotopic to $\mathcal{W}'$, $\mathcal{W}'$ is discretely homotopic to $\mathcal{W}$ and discrete homotopy is transitive, we derive that $\mathcal{W}^*$ is discrete Moreover, the endpoints of the walks in $\mathcal{W}'$ were not changed when the cycle move operation was applied. Thus, because $\mathcal{W}'$ is sensible, \ No newline at end of file diff --git a/samples/texts/1754951/page_51.md b/samples/texts/1754951/page_51.md new file mode 100644 index 0000000000000000000000000000000000000000..95aaa365397c7505b336a97f9b0d78803736e889 --- /dev/null +++ b/samples/texts/1754951/page_51.md @@ -0,0 +1,17 @@ +so is $\mathcal{W}^*$. Moreover, it is clear that $\mathcal{W}^*$ is pushed onto $R$ and has the same multiplicity as $\mathcal{W}'$. +However, as $j > t$, the sum of the subscripts of the edge copies in $E(\mathcal{W}^*)$ is larger than that of +$E(\mathcal{W}')$, which contradicts the choice of $\mathcal{W}'$. + +Hence, there exist weak linkages $\mathcal{W}'$ that are sensible, pushed onto $R$, discretely homotopic to $\mathcal{W}$, whose multiplicity is upper bounded by the multiplicity of $\mathcal{W}$, and for all edges $e_i \in E(\mathcal{W}')$ $i \ge i$. Consider the collection of all such weak linkages, and let $\mathcal{W}^*$ be the one maximizing $w(E(\mathcal{W}^*)) = \sum e \in E(\mathcal{W}^*)w(e)$ where Let us begin by defining a weight function $w: E(H) \to \mathbb{Z}$ on the parallel copies of edges in $H$ as follows. + +$$ w(e) = \begin{cases} -2n & e \text{ is not parallel to any edge in } E(R) \\ -2n & e = e_i \text{ is parallel to an edge in } E(R) \text{ and } i \le 0 \\ 2n-i & e = e_i \text{ is parallel to an edge in } E(R) \text{ and } i \ge 1 \end{cases} $$ + +We claim that $\mathcal{W}^*$ is canonical, i.e. for every edge $e_i \in E(\mathcal{W}^*)$ the subscript $i \ge 1$, and for every parallel edge $e_j$ where $i \le j < i$, $e_j \in E(\mathcal{W})$. The first property is ensured by the choice of $\mathcal{W}^*$. For the second property, we argue as before. Suppose not, then choose $i$ and $j$ such that $i-j$ is minimized. Then clearly $j = i-1$, since any parallel copy $e_t$ with $j < t < i$ is either in $E(\mathcal{W}^*)$ contradicting the choice of $i$, or not in $E(\mathcal{W}^*)$ contradicting the choice if $j$. Therefore, the edges $e_i$ and $e_j$ form a cycle $C$ such that the interior of $C$ contains no edge of any walk in $\mathcal{W}^*$. Let $\mathcal{W} \in \mathcal{W}^*$ be the walk containing $e_i$, and observe that the cycle move operation is applicable to $(\mathcal{W}, C)$. Let $\tilde{\mathcal{W}}$ be the result of this operation. Then observe that $w(E(\tilde{\mathcal{W}})) > w(E(\mathcal{W}^*))$, since $w(e_j) > w(e_i)$ and $E(\tilde{\mathcal{W}}) \setminus \{e_j\} = E(\mathcal{W}^*) \setminus \{e_i\}$. And, $\tilde{\mathcal{W}}$ is discretely homotopic to $\mathcal{W}'$ which is in turn discretely homotopic to $\mathcal{W}$, as before we can argue that $\tilde{\mathcal{W}}$ contradicts the choice of $\mathcal{W}^*$. Hence, $\mathcal{W}^*$ must be canonical. $\square$ + +In case we are interested only in extremality rather than canonicity, we can use the following lemma that does not increase potential. The proof is very similar to the proof of Lemma 8.8, except that now we can “move edges in either direction”, and hence avoid creating new crossings. + +**Lemma 8.9.** Let $(G, S, T, g, k)$ be an instance of Planar Disjoint Paths, and $R$ be a Steiner tree. Let $\mathcal{W}$ be a weak linkage in $H$ that is sensible and pushed onto $R$, whose multiplicity is at most $2n$. Then, there exists a weak linkage $\mathcal{W}'$ that is sensible, pushed onto $R$, extremal, discretely homotopic to $\mathcal{W}$, and whose potential is upper bounded by the potential of $\mathcal{W}$. + +*Proof.* Consider all weak linkages in $H$ that are sensible, pushed onto $R$, discretely homotopic to $\mathcal{W}$, and whose potential and multiplicity are upper bounded by the potential and multiplicity, respectively, of $\mathcal{W}$. Among these weak linkages, let $\mathcal{W}'$ be one such that the sum of the absolute values of the subscripts of the edge copies in $E(\mathcal{W}')$ is maximized. We claim that $\mathcal{W}'$ is extremal, which will prove the lemma. To this end, suppose by way of contradiction that $\mathcal{W}'$ is not extremal. Thus, there exist an edge $e_i, e_j \in \mathcal{W}'$ where $i \ge 1, j \le -1$ and $(i-1)+|j+1| \le 2n-1$. Because the multiplicity of $\mathcal{W}'$ is at most $2n$, this means that there exists an edge $e_p$ (parallel to $e_i$ and $e_j$) for some $p > i \ge 1$ that is not used by $E(\mathcal{W}')$. Let $e_t$ be the edge (parallel to $e_j$ and $e_i$) of largest subscript smaller than $p$ that is used by $E(\mathcal{W}')$. Moreover, let $C$ be the cycle (which might be the boundary of a single face) that consists of two edges: $e_p$ and $e_t$. By the choice of $e_t$, the strict interior of $C$ does not contain any edge of $E(\mathcal{W}')$. Thus, the cycle move operation is applicable to $(\mathcal{W}, C)$ where $\mathcal{W}$ is the walk in $\mathcal{W}'$ that uses $e_t$. Let $\mathcal{W}^*$ be the result of the application of this operation. Then, the only difference between $\mathcal{W}^*$ and $\mathcal{W}'$ is the replacement of $e_t$ by $e_p$. + +Because $\mathcal{W}^*$ is discretely homotopic to $\mathcal{W}'$, $\mathcal{W}'$ is discretely homotopic to $\mathcal{W}$ and discrete homotopy is transitive, we derive that $\mathcal{W}^*$ is discrete. Moreover, the endpoints of the walks in $\mathcal{W}'$ were not changed when the cycle move operation was applied. Thus, because $\mathcal{W}'$ is sensible, so is $\mathcal{W}^*$. Moreover, it is clear that $\mathcal{W}^*$ is pushed onto $R$, because $\mathcal{W}^*$ and $\mathcal{W}'$ cross $R$ exactly in \ No newline at end of file diff --git a/samples/texts/1754951/page_52.md b/samples/texts/1754951/page_52.md new file mode 100644 index 0000000000000000000000000000000000000000..4ae0a7e6a912c9d52458bd43f8f49bedab490ce1 --- /dev/null +++ b/samples/texts/1754951/page_52.md @@ -0,0 +1,21 @@ +Figure 22: Special U-Turns. + +the same vertices and in the same direction (indeed, we have only replaced one edge of positive subscript by another parallel edge of positive subscript), they have the same potential. However, as $p > t \ge 1$, the sum of the absolute values of the subscripts of the edge copies in $E(W^*)$ is larger than that of $E(W')$, which contradicts the choice of $W'$. + +To achieve the properties of having low multiplicity and being U-turn-free, we perform two stages. In the first stage, that is the focus of this subsection, we make modifications that bound the total number of segments (rather than only the number of segments groups). The second stage, where we conclude the two properties, will be performed in the next subsection. The first stage in itself is partitioned into two phases as follows. + +**Phase I: Eliminating Special U-Turns.** We eliminate some of the U-turns, but not all of them. Specifically, the elimination of some U-turns may result in too major changes in the segment groups, and hence we only deal with them after we bound the total number of segments, in which case classification into segment groups becomes immaterial. The U-turns we eliminate now are defined as follows (see Fig. 22). + +**Definition 8.18 (Special U-Turn).** Let $(G, S, T, g, k)$ be an instance of Planar Disjoint Paths, and $R$ be a Steiner tree. Let $\mathcal{W}$ be weak linkage that is pushed onto $R$. Let $U = \{e_i, e_j\}$ be a U-turn. Then, $U$ is special if at least one among the following conditions hold: (i) $i$ and $j$ have the same sign, i.e. they are on the same side of $R$; (ii) both endpoints of $e_i$ and $e_j$ do not belong to $V_{=1}(R) \cup V_{\ge 3}(R)$. + +We eliminate special U-turns one-by-one, where the U-turn chosen to eliminate at each step is an innermost one, defined as follows. + +**Definition 8.19 (Innermost U-Turn).** Let $(G, S, T, g, k)$ be an instance of Planar Disjoint Paths, and $R$ be a Steiner tree. Let $\mathcal{W}$ be weak linkage that is pushed onto $R$. Let $U = \{e_i, e_j\}$ be a U-turn. Then, $U$ is innermost if there does not exist a parallel edge $e_\ell \in E(\mathcal{W})$ such that $\min\{i, j\} \le \ell \le \max\{i, j\}$. We say that $U$ is crossing if the signs of $i$ and $j$ are different, i.e. $e_i$ and $e_j$ are on opposite sides of $R$. + +We argue that if there is a (special) U-turn, then there is also an innermost (special) one. + +**Lemma 8.10.** Let $(G, S, T, g, k)$ be an instance of Planar Disjoint Paths, and $R$ be a Steiner tree. Let $\mathcal{W}$ be a weak linkage pushed onto $R$ with at least one U-turn $U = \{e_i, e_j\}$. Then, $\mathcal{W}$ has at least one innermost U-turn $U' = \{e_x, e_y\}$ whose edges lie in the interior (including the boundary) of the cycle $C$ formed by $e_i$ and $e_j$. + +*Proof.* Denote the endpoints of $e_i$ and $e_j$ by $u$ and $v$. Among all U-turns whose edges lie in the interior (including the boundary) of the cycle $C$ formed by $e_i$ and $e_j$ (because $U$ satisfies these conditions, there exists at least one such U-turn), let $U' = \{e_x, e_y\}$ be one whose edges $e_x$ and $e_y$ form a cycle $C'$ that contains minimum number of edges of $H$ in its interior. Let $\mathcal{W}'$ be the walk in $\mathcal{W}$ that traverses $e_x$ and $e_y$ consecutively. Without loss of generality, suppose that when we traverse $\mathcal{W}'$ so that we visit $e_x$ and then $e_y$, we first visit $u$, then $v$, and then $u$ again. + +We claim that $U'$ is innermost. To this end, suppose by way of contradiction that $U'$ is not innermost. Thus, by Definition 8.19, this means that $C'$ contains an edge $e_\ell$ in its strict interior that belongs to some walk $\tilde{\mathcal{W}} \in \mathcal{W}$ (possibly $\tilde{\mathcal{W}} = W'$). Because $U'$ is a U-turn, $e_\ell$ is neither \ No newline at end of file diff --git a/samples/texts/1754951/page_53.md b/samples/texts/1754951/page_53.md new file mode 100644 index 0000000000000000000000000000000000000000..a2cfd8435a29f683e4bf8549c39f42b91b6276c2 --- /dev/null +++ b/samples/texts/1754951/page_53.md @@ -0,0 +1,11 @@ +Figure 23: U-Turn with $e_i$ and $e_j$ on the same side + +the first nor the last edge of $\tilde{W}$. Thus, when we traverse $\tilde{W}$ so that when we visit $e_\ell$, we first visit $u$ and then $v$, we next visit an edge $e'$. Because $\mathcal{W}$ is a weak linkage, this edge must belong to the strict interior of $C'$ (because otherwise we obtain that $(v, e_x, e_y, e_\ell)$ is a crossing or an edge is used more than once). However, this implies that $e'$ is parallel to the edges $e_\ell, e_x, e_y$, and $\tilde{U} = \{e_\ell, e'\}$ is a U-turn whose edges lie in the interior (including the boundary) of the cycle $C$ and which forms a cycle $\tilde{C}$ that contains fewer edge of $H$ than $C'$ in its interior. This is a contradiction to the choice of $U'$. $\square$ + +We now prove that an innermost U-turn corresponds to a cycle on which we can perform the cycle pull operation, and consider the result of its application. + +**Lemma 8.11.** Let $(G, S, T, g, k)$ be an instance of Planar Disjoint Paths, and $R$ be a Steiner tree. Let $\mathcal{W}$ be a sensible, outer-terminal, extremal weak linkage that is pushed onto $R$, and let $U = \{e_i, e_j\}$ be an innermost U-turn. Let $\mathcal{W}$ be the walk in $\mathcal{W}$ that uses $e_i$ and $e_j$, and $C$ be the cycle in $H$ that consists of $e_i$ and $e_j$. Then, the cycle pull operation is applicable to $(W, C)$. Furthermore, the resulting weak linkage $\mathcal{W}'$ is sensible, outer-terminal, extremal, pushed onto $R$, having fewer edges than $\mathcal{W}$, $|\text{Seg}(\mathcal{W}')| \le |\text{Seg}(\mathcal{W})|$, and if $U$ is special, then also its potential is upper bounded by the potential of $\mathcal{W}$. + +*Proof*. Because $U$ is innermost, there does not exist an edge in the strict interior of $C$ that belongs to $E(\mathcal{W})$. Therefore, the cycle pull operation is applicable to $(W, C)$. The only difference between $\mathcal{W}'$ and $\mathcal{W}$ is that $\mathcal{W}'$ does not use the edge $e_i$ and $e_j$ and hence the vertex, say, $v$, that $\mathcal{W}$ visits between them. Therefore, because $\mathcal{W}$ be a sensible, outer-terminal, extremal and pushed onto $R$, so is $\mathcal{W}'$. Moreover, the walks in $\mathcal{W}'$ have the same endpoints as their corresponding walks in $\mathcal{W}$, and thus because $\mathcal{W}$ is sensible, so is $\mathcal{W}'$. Let $u$ be the other endpoints of the edges $e_i$ and $e_j$, and let $\mathcal{W}'$ be the walk in $\mathcal{W}'$ that resulted from $\mathcal{W}$. Observe that $\mathcal{W}'$ has at most as many crossings with $R$ as $\mathcal{W}$ has—indeed, if the elimination of $e_i$ and $e_j$ created a new crossing at $u$ (this is the only new crossing that may be created), then $\mathcal{W}$ crosses $R$ between $e_i$ and $e_j$ and this crossing does not occur in $\mathcal{W}'$. Thus, $|\text{Seg}(\mathcal{W}')| \le |\text{Seg}(\mathcal{W})|$. + +Now, suppose that $U$ is special. Then, at least one among the following conditions holds: (i) $i$ and $j$ have the same sign; (ii) $u, v \notin V_{=1}(R) \cup V_{\ge 3}(R)$. We first consider the case where $i$ and $j$ have the same sign, say positive (without loss of generality; see Fig. 23). Let $e'_x, \hat{e}_y \in E(\mathcal{W})$ be such that $e'_x$ and $e_i$ are consecutive in $\mathcal{W}$ (if such an edge $e'_x$ exists) and denote the segment that contains $e'_x$ by $S_x$, and $\hat{e}_y$ and $e_j$ are consecutive in $\mathcal{W}$ (if such an edge $\hat{e}_y$ exists) and denote the segment that contains $\hat{e}_y$ by $S_y$. Possibly some edges among $e'_x, \hat{e}_y$ and $e_i$ are parallel. In case $e'_x$ does not exist (see Fig. 23(a)), then $u \in V_{=1}(R)$ and therefore $e_i$ and $e_j$ belong to a segment that is in a singleton segment group. When we remove $e_i$ and $e_j$, either this segment shrinks (and remains in a singleton group) or it is removed completely together with its segment group. If the segment shrinks, the potential clearly remains unchanged, and otherwise the reduction of segment groups makes the potential decrease by 1 (the potential of the consecutive segment group remains unchanged as it is a singleton segment group because it has an endpoint in $V_{=1}(R)$). The case where $\hat{e}_y$ does not exist is symmetric, thus we now assume that both $e'_x$ and $\hat{e}_y$ exist. In case both $x$ and $y$ are on the same side as $e_i$ and $e_j$ (see Fig. 23(b)), then the removal of $e_i$ and $e_j$ only shrinks the segment $S_x = S_y$ where all of the four edges $e_i, e_j, e'_x$ and \ No newline at end of file diff --git a/samples/texts/1754951/page_54.md b/samples/texts/1754951/page_54.md new file mode 100644 index 0000000000000000000000000000000000000000..6357e5c31d29ff79daeeff9e194366c94e4862c5 --- /dev/null +++ b/samples/texts/1754951/page_54.md @@ -0,0 +1,7 @@ +Figure 24: U-Turn with $e_i, e_j$ on one side and $e'_x, \hat{e}_y$ is on the opposite side. + +$\hat{e}_y$ lie, and thus does not change the potential. Similarly, if $e'_x$ is on the same side as $e_i, e_j$ and $\hat{e}_y$ is on the opposite side (see Fig. 23(c)), then we only shrink the segment where $e'_x, e_i$ and $e_j$ lie, and rather than crossing from $e_j$ to $\hat{e}_y$, we cross from $e'_x$ to $\hat{e}_y$ (which have the same label as both cross from the positive side to the negative side). The case where $e'_x$ is on the opposite side of $e_i, e_j$ and $\hat{e}_y$ is on the same side is not possible, since then W crosses itself. + +Now, consider the case where both $e'_x$ and $\hat{e}_y$ are both on the opposite side of $e_i, e_j$ (see Fig. 24) Then, $e_i$ and $e_j$ form a complete segment, which we call $S_{ij}$, and the segments $S_x, S_{ij}$ and $S_y$ are different. Notice that the two crossings with $R$, one consisting of $e'_x$ with $e_i$, and the other consisting of $e_j$ with $\hat{e}_y$, cross in different directions. If both $S_x, S_{ij}$ and $S_y$ belong to the same segment group (see Fig. 24(a)), then the removal of $S_{ij}$ removes the contributions of its two crossings (mentioned above), which sum to 0 as their direction is opposite. Then, the potential remains unchanged. Now, suppose that exactly one among $S_x$ and $S_y$ belongs to the same group as $S_{ij}$. Without loss of generality, suppose that it is $S_x$ (the other case is symmetric; see Fig. 24(b)). Then, when we remove $S_{ij}$, the segments $S_x$ and $S_y$ merge into one segment that has endpoints in different maximal degree-2 paths in $R$ (or on vertices of degree other than 2) and hence forms its own group. This group replaces the singleton group of $W$ that contained only $S_y$. Furthermore, the labeling of all crossings remain the same (as well as all other associations into segment groups), apart from the two crossings consisting of $e'_x$ with $e_i$, and of $e_j$ with $\hat{e}_y$, which are both eliminated but have previously contributed together 0 (as they cross in opposite directions). We remark that the size of the segment group that previously contained $S_x$ might become 1 or completely removed, but this does not increase potential. Thus, overall the potential does not increase. + +Lastly, suppose that $S_x, S_{ij}$ and $S_y$ belong to different segment groups. Then, $S_x$ and $S_y$ had endpoints in different maximal degree-2 paths in $R$ (or on vertices of degree other than 2), hence each one among $S_x, S_{ij}$ and $S_y$ belonged to a singleton segment group of $W$. The removal of $S_{ij}$ eliminates all of these three segment groups, which results in a decrease of 3 in the potential. However, now $S_x$ and $S_y$ belong to the same segment group. If they form a singleton segment group (see Fig. 24(c)), the overall potential decreases by 2. Else, they join an existing segment group, and we have several subcases as follows. In the first subcase, suppose that they join only the group that contains the segment $\tilde{S}_x$ of $W$ consecutive to $S_x$ (in the walk in $W$ to which $S_x, S_{ij}$ and $S_y$ belong; see Fig. 24(d)). Then, the crossing at the endpoint of $S_y$ is now contributing (1 or -1) to the sum of labels in the potential, and if $\tilde{S}_x$ was in a singleton group in $W$, then so are its crossings. Overall, this results in a contribution of at most 3, so in total the potential does not increase. The subcase where they join only the group that contains the segment $\tilde{S}_y$ of $W$ consecutive to $S_y$ is symmetric. Now, consider the subcase \ No newline at end of file diff --git a/samples/texts/1754951/page_55.md b/samples/texts/1754951/page_55.md new file mode 100644 index 0000000000000000000000000000000000000000..e685980107b24a3795c004bffcce98a523dd02fd --- /dev/null +++ b/samples/texts/1754951/page_55.md @@ -0,0 +1,9 @@ +Figure 25: U-Turn with $e_i$ on one side and $e_j$ on the opposite side. + +where they join both of these groups and hence merge them (see Fig. 24(e)). In this subcase, we have four new crossings that may contribute to the sum of labels, but we have also merged two groups which makes the potential decreased by at least 1, so overall the potential does not increase. + +Now, suppose that only case (ii) holds. That is, $u, v \notin V_{=1}(R) \cup V_{\ge 3}(R)$, and $i$ and $j$ have the different signs (see Fig. 25). Without loss of generality, suppose that $i \ge 1$ and $j \le -1$. Because $u, v \notin V_{=1}(R) \cup V_{\ge 3}(R)$ and $\mathcal{W}$ is sensible, $e'_x$ and $\hat{e}_y$ exist. The case where $e'_x$ is on the opposite side of $e_i$ and $\hat{e}_y$ is on the same side as $e_i$ cannot occur since then $\mathcal{W}$ crosses itself. Thus, we are left with three cases: (a) $e'_x$ is on the same side as $e_i$ and $\hat{e}_y$ is on the opposite side of $e_i$; (b) both $e'_x$ and $\hat{e}_y$ are on the opposite side of $e_i$; and (c) both $e'_x$, $\hat{e}_y$ are on the same side as $e_i$. The cases (b) and (c) are symmetric, therefore we will only consider cases (a) and (b). In case (a), $e'_x$ and $e_i$ belong to one segment, and $e_j$ and $\hat{e}_y$ belong to a different segment (see Fig. 25(a)). The removal of $e_i$ and $e_j$ only shirks these two segments by one edge each, and does not change the labeling of the crossings at their endpoints—previously, we crossed from $e_i$ to $e_j$, and now we cross from $e'_x$ to $\hat{e}_y$, which are both crossings from the side of $e_i$ to the opposite side. Thus, the potential does not increase. + +Lastly, consider case (b) (see Fig. 25(b)). In this case, $e_i$ belongs to a segment $S_i$ containing only $e_i$, and $e_j$ belongs to $S_y$. Further, when crossing from $e_x$ to $e_i$, we cross from the opposite side of $e_i$ to the same side, and when we cross from $e_i$ to $e_j$, we cross from the side of $e_i$ to the opposite side. Additionally, notice that the elimination of $e_i$ and $e_j$ results in the elimination of $S_i$, and the merge of $S_x$ and $S_y$ with $e_j$ removed. We consider several subcases as follows. In the subcase where $S_x$, $S_i$ and $S_y$ belong to the same group (see Fig. 25(c)), then the only possible effect with respect to the potential of this group is the cancellation of the two crossings (of $e'_x$ with $e_i$ and of $e_i$ with $e_j$), but these two crossings together contribute 0 to the sum of labels because they cross in opposite directions. Possibly the size of the segment group shrunk to 1, but this does not increase potential. Thus, in this subcase, the potential does not increase. + +Now, consider the subcase where $S_x$ and $S_i$ are in the same segment group, and $S_y$ is in a different segment group (see Fig. 25(d)). Then, $S_y$ is in a singleton segment group because its endpoints belong to different maximal degree-2 path of $\mathcal{R}$ (or vertices of degree other than 2). In $\mathcal{W}'$, the segment group that resulted from the merge of $S_x$ and $S_y$ is also a singleton segment group. Furthermore, the segment group of $\mathcal{W}$ that contained $S_x$ and $S_i$ does not change in terms of its labeled sum since the crossings at the endpoints of $S_i$ crossed in opposite directions. Possibly the size of the segment group shrunk to 1, but this does not increase potential. Thus, in this subcase, the potential does not increase. Next, we note that the \ No newline at end of file diff --git a/samples/texts/1754951/page_57.md b/samples/texts/1754951/page_57.md new file mode 100644 index 0000000000000000000000000000000000000000..4729d52f3ea6c5afd95ea7edfcda17ff283fae83 --- /dev/null +++ b/samples/texts/1754951/page_57.md @@ -0,0 +1,19 @@ +Figure 26: A swollen segment and the cycles in its move-through tuple. + +analysis of the subcase where $S_i$ and $S_y$ are in the same segment group, and $S_x$ is in a different segment group, is symmetric. Lastly, suppose that $S_x$, $S_i$ and $S_y$ belong to different segment groups (see Fig. 25(e)). The analysis of this case is the same as the analysis of the last subcase of case (i) (i.e., the subcase where $S_x$, $S_{ij}$ and $S_y$ belong to the same segment group, where now we have $S_i$ instead of $S_{ij}$). $\square$ + +We are now ready to assert that all special U-turns can be eliminated as follows. + +**Lemma 8.12.** Let $(G, S, T, g, k)$ be a good Yes-instance of Planar Disjoint Paths, and $R$ be a backbone Steiner tree. Then, there exists a sensible, outer-terminal, extremal weak linkage $W$ in $H$ that has no special U-turns, is pushed onto $R$ and discretely homotopic in $H$ to some solution of $(G, S, T, g, k)$ and Potential$(W) \le \alpha_{\text{potential}}(k)$. + +*Proof.* By Lemma 8.7, there exists a sensible outer-terminal weak linkage in $H$ that is pushed onto $R$, has multiplicity at most $2n$, discretely homotopic in $H$ to some solution of $(G, S, T, g, k)$ and has potential at most $\alpha_{\text{potential}}(k)$. By Lemma 8.9 and because discrete homotopy is an equivalence relation, there also exists such a weak linkage $W'$ that is extremal. Thus, there exists a weak linkage $W$ that among all sensible, outer-terminal, extremal weak linkages in $H$ that are pushed onto $R$, discretely homotopic in $H$ to some solution of $(G, S, T, g, k)$ and satisfy Potential$(W) \le \alpha_{\text{potential}}(k)$, the weak linkage $W$ is one that minimizes the number of edges that it uses. To conclude the proof, it suffices to argue that $W$ has no special U-turns. + +Suppose, by way of contradiction, that $W$ has at least one special U-turn. Then, by Lemma 8.10, $W$ has an innermost special U-turn $U = \{e_i, e_j\}$. Let $W$ be the walk in $W$ that uses $e_i$ and $e_j$, and $C$ be the cycle in $H$ that consists of $e_i$ and $e_j$. Then, by Lemma 8.11, the cycle pull operation is applicable to $(W, C)$. Furthermore, by Lemma 8.11, the resulting weak linkage $W'$ is sensible, outer-terminal, extremal, pushed onto $R$, has fewer edges than $W$, and its potential is upper bounded by the potential of $W$. Since discrete homotopy is an equivalence relation, $W'$ is discretely homotopic to some solution of $(G, S, T, g, k)$. However, this is a contradiction to the choice of $W$. $\square$ + +**Phase II: Eliminating Swollen Segments.** The goal of the second phase is to eliminate the existence crossings with opposing “signs” for each segment and thereby, as the potential is bounded, bound the number of segments (rather than only the number of segment groups). We remark that one can show, even without this step, that the multiplicity is bounded, however this complicates the analysis. Towards this, we eliminate “swollen” segments (see Fig. 26). + +**Definition 8.20 (Swollen Segment).** Let $(G, S, T, g, k)$ be an instance of Planar Disjoint Paths, and $R$ be a Steiner tree. Let $W$ be weak linkage that is pushed onto $R$. Consider segment $S \in \text{Seg}(W)$ for some $W \in W$ such that $S$ does not contain two the extreme edges of $W$. Let $e$ and $e'$ be the extreme edges of $S$ (possibly $e = e'$), and let $\hat{e}$ and $\tilde{e}$ be the edges of $E(W) \setminus E(S)$ that are consecutive to $e$ and $e'$ on $W$, respectively. Then, $S$ is swollen if its endpoints are internal vertices of the same maximal degree-2 path $P$ of $R$ and $\text{label}_P^W(e, e') \ne \text{label}_P^W(\hat{e}, \tilde{e})$. + +We show that due to the first phase, when we deal with outer-terminal weak linkages, the swollen segments have a “clean appearance” as stated in the following lemma. + +**Lemma 8.13.** Let $(G, S, T, g, k)$ be a nice instance of Planar Disjoint Paths, and $R$ be a Steiner tree. Let $W$ be an outer-terminal weak linkage that is pushed onto $R$ and has no special U-turns, and $S \in \text{Seg}(W)$ be swollen. Then, $S$ is parallel to a subpath of a maximal degree-2 path of $R$. \ No newline at end of file diff --git a/samples/texts/1754951/page_59.md b/samples/texts/1754951/page_59.md new file mode 100644 index 0000000000000000000000000000000000000000..90b7400b0c5efa744528f78abb1ff60ce6a562cb --- /dev/null +++ b/samples/texts/1754951/page_59.md @@ -0,0 +1,19 @@ +Figure 27: Illustration of Lemma 8.13. + +the index of sign opposite to $i_j$ that has the largest absolute value such that all indices $r$ of the same sign as $\ell$ and whose absolute value is upper bounded by $|\ell|$ satisfy that $e_r^j \notin E(W)$. + +The application of $T$ is done by applying the cycle move operation to $(W, C_i)$ for $i$ from 1 to $t$ (in this order) where $W$ is the walk that contains $S$ as a segment.²⁰ + +Now, we prove that application of a move-through tuple is valid. + +**Lemma 8.15.** Let $(G, S, T, g, k)$ be an instance of Planar Disjoint Paths, and $R$ be a Steiner tree. Let $W$ an outer-terminal extremal weak linkage that has no special U-turns and is pushed onto $R$, and let $S \in \text{Seg}(W)$ be an innermost swollen segment. Then, the move-through tuple $T = (C_1, \dots, C_t)$ of $S$ is well-defined, and the application of $T$ is valid (that is, the cycle move operation is applicable to $(W, C_i)$ when it is done). + +*Proof.* Let $e_{i_1}^1, e_{i_2}^2, \dots, e_{i_t}^t$, where $t = |E(S)|$, be the edges of $S$ in the order occurred when $S$ is traversed from one endpoint to another. To assert that $T$ is valid, consider some $j \in \{1, \dots, t\}$. We need to show that $e_b^j$, where $b \in \{-1, 1\}$ has sign opposite to the sign of $i_j$, does not belong to $E(W)$. Indeed, then there also exists an index $\ell_j$ of sign opposite to $i_j$ that has the largest absolute value such that all indices $r$ of the same sign as $\ell_j$ and whose absolute value is upper bounded by $|\ell_j|$ satisfy that $e_r^j \notin E(W)$. To this end, suppose by contradiction that $e_b^j \in E(W)$. Without loss of generality, suppose that $b = -1$ (the other case is symmetric). Then, we have $(i_j - 1) + |\ell_j - b| = i_j - 1 \le 2n - 1$, which contradicts the assumption that $W$ is extremal. Thus, $T$ is valid. + +Notice that for every $j \in \{1, \dots, t\}$, all of the edges parallel to $e_{i_j}^j$ (and $e_{\ell_j}^j$) whose index is of the same sign as $i_j$ and has absolute value is smaller than $|i_j|$ do not belong to $E(W)$ (because $S$ is innermost). Additionally for every $j \in \{1, \dots, t\}$, all of the edges parallel to $e_{i_j}^j$ (and $e_{\ell_j}^j$) whose index is of sign opposite to $i_j$ and has absolute value is smaller or equal to than $|\ell_j|$ do not belong to $E(W)$ (by the choice of $\ell_j$. Thus, for each $j \in \{1, \dots, t\}$ the interior of $C_j$ does not contain any edge of $W$, and in particular the cycle move operation is applicable to it. When we apply the cycle move operation to some cycle $C_j$, it replaces $e_{i_j}^j$ by $e_{i_j}^j$. By Lemma 8.14, these replacements are done on edges not parallel to one another—that is, for every pair of distinct $j, j' \in \{1, \dots, t\}$, the edges $e_{i_j}^j$ and $e_{i_{j'}}^{j'}$ are not parallel. Thus, the application of one cycle move operation in the application of $T$ does not effect the applicability of any other cycle move operation in the application of $T$. Therefore, the application of $T$ is valid. □ + +Now, we consider the properties of the weak linkage that results from the application of a move-through tuple. + +**Lemma 8.16.** Let $(G, S, T, g, k)$ be an instance of Planar Disjoint Paths, and $R$ be a Steiner tree. Let $W$ be a sensible, outer-terminal, extremal weak linkage that has no special U-turns and is pushed onto $R$, and let $S \in \text{Seg}(W)$ be an innermost swollen segment. Let $T$ be the move-through tuple of $S$, and let $W'$ be the weak linkage that results from the application of $T$. Then, $W'$ is a sensible, outer-terminal, extremal weak linkage that has no special U-turns and is pushed onto $R$, whose potential is upper bounded by the potential of $W$ and which has fewer segments than $W$. + +²⁰Note that $W$ changes in each application, thus by $W$ we mean the current walk with the same endpoints as the original walk in $W$ that had $S$ as a segment. \ No newline at end of file diff --git a/samples/texts/1754951/page_6.md b/samples/texts/1754951/page_6.md new file mode 100644 index 0000000000000000000000000000000000000000..d1560036cffff89d949fd5c073dc6c82f31c2552 --- /dev/null +++ b/samples/texts/1754951/page_6.md @@ -0,0 +1,31 @@ +the addition of any edge (not parallel to an existing edge) to G results in a non-planar graph. A plane graph G that is triangulated is 2-connected, and each of its faces is a simple cycle that is a triangle or a cycle that consists of two parallel edges (when the graph is not simple). As we will deal with triangulated graphs, the following proposition will come in handy. + +**Proposition 3.1** (Proposition 8.2.3 in [34]). Let $G$ be a triangulated plane graph. Let $A, B \subseteq V(G)$ be disjoint subsets such that $G[A]$ and $G[B]$ are connected graphs. Then, for any minimal subset $S \subseteq V(G) \setminus (A \cup B)$ that separates $A$ from $B$, it holds that $G[S]$ is a cycle.⁴ + +The radial graph (also known as the *face-vertex incidence graph*) of a plane graph $G$ is the planar graph $G'$ whose vertex set consists of $V(G)$ and a vertex $v_f$ for each face $f$ of $G$, and whose edge set consists of an edge $\{u, v_f\}$ for every vertex $u \in V(G)$ and face $f$ of $G$ such that $u$ is incident to (i.e. lies on the boundary of) $f$. The *radial completion* of $G$ is the graph $G'$ obtained by adding the edges of $G$ to the radial graph of $G$. The graph $G'$ is planar, and we draw it on the plane so that its drawing coincides with that of $G$ with respect to $V(G) \cup E(G)$. Moreover, $G'$ is triangulated and, under the assumption that $G$ had no self-loops, $G'$ also has no self-loops (since all new edges in $G'$ have one endpoint in $V(G)$ and the other endpoint in $V(G') \setminus V(G)$). For a plane graph $G$, the *radial distance* between two vertices $u$ and $v$ is one less than the minimum length of a sequence of vertices that starts at $u$ and ends at $v$, such that every two consecutive vertices in the sequence lie on a common face.⁵ We denote the radial distance by $rdist_G(u, v)$. This definition extends to subsets of vertices: for $X, Y \subseteq V(G)$, $rdist_G(X, Y)$ is the minimum radial distance over all pairs of vertices in $X \times Y$. + +For any $t \in \mathbb{N}$, a sequence $\mathcal{C} = (C_1, C_2, \dots, C_t)$ of $t$ cycles in a plane graph $G$ is said to be *concentric* if for all $i \in \{1, 2, \dots, t-1\}$, the cycle $C_i$ is drawn in the strict interior of $C_{i+1}$ (excluding the boundary, that is, $V(C_i) \cap V(C_{i+1}) = \emptyset$). The *length* of $\mathcal{C}$ is $t$. For a subset of vertices $U \subseteq V(G)$, we say that $\mathcal{C}$ is *U-free* if no vertex of $U$ is drawn in the strict interior of $C_t$. + +**Treewidth.** Treewidth is a measure of how “treelike” is a graph, formally defined as follows. + +**Definition 3.1 (Treewidth).** A tree decomposition of a graph $G$ is a pair $(T, \beta)$ of a tree $T$ and $\beta: V(T) \to 2^{V(G)}$, such that + +1. for any edge $\{x, y\} \in E(G)$ there exists a node $v \in V(T)$ such that $x, y \in \beta(v)$, and + +2. for any vertex $x \in V(G)$, the subgraph of $T$ induced by the set $T_x = \{v \in V(T): x \in \beta(v)\}$ is a non-empty tree. + +The width of $(T, \beta)$ is $\max_{v \in V(T)} |\{\beta(v)\}| - 1$. The treewidth of $G$, denoted by $\text{tw}(G)$, is the minimum width over all tree decompositions of $G$. + +The following proposition, due to Cohen-Addad et al. [12], relates the treewidth of a plane graph to the treewidth of its radial completion. + +**Proposition 3.2** (Lemma 1.5 in [12]). Let $G$ be a plane graph, and let $H$ be its radial completion. Then, $\text{tw}(H) \leq \frac{7}{2} \cdot \text{tw}(G)^6$. + +## 3.1 Homology and Flows + +For an alphabet $\Sigma$, denote $\Sigma^{-1} = \{\alpha^{-1} : \alpha \in \Sigma\}$. For a symbol $\alpha \in \Sigma$, define $(\alpha^{-1})^{-1} = \alpha$, and for a word $w = a_1a_2\cdots a_t$ over $\Sigma \cup \Sigma^{-1}$, define $w^{-1} = a_t^{-1}a_{t-1}^{-1}\cdots a_1^{-1}$ and $w^1 = w$. The empty + +⁴Here, the term cycle also refers to the degenerate case where $|S| = 1$. + +⁵We follow the definition of radial distance that is given in [23] in order to cite a result in that paper verbatim (see Proposition 6.1). We remark that in [5], the definition of a radial distance is slightly different. + +⁶More precisely, Cohen-Addad et al. [12] prove that the *branchwidth* of G is at most twice the branchwidth of H. Since the treewidth (plus 1) of a graph is lower bounded by its branchwidth and upper bounded by $\frac{3}{2}$ its branchwidth (see [12]), the proposition follows. \ No newline at end of file diff --git a/samples/texts/1754951/page_60.md b/samples/texts/1754951/page_60.md new file mode 100644 index 0000000000000000000000000000000000000000..0e6ae7fd4c51763e5bc2cf4305160d3787975410 --- /dev/null +++ b/samples/texts/1754951/page_60.md @@ -0,0 +1,5 @@ +*Proof.* Let $u$ and $v$ be the endpoints of $S$. Because $S$ is swollen, $u$ and $v$ are internal vertices of the same maximal degree-2 path $P$ of $R$. Thus, because $W$ is sensible, there exists a segment $S_1$ and a segment $S_2$ that the walk $W \in W$ that has $S$ as a segment traverses immediately before visiting the endpoint $u$ of $S$, and immediately after visiting the endpoint $v$ of $S$, respectively. (Here, we supposed without loss of generality that $W$ is traversed from one endpoint to another such that the endpoint $u$ of $S$ is visited before the endpoint $v$ of $S$.) By Lemma 8.14, $S$ is parallel to the subpath $Q$ of $P$ with endpoints $u$ and $v$, and without loss of generality, we suppose that it uses the positive copies of the edges of $Q$. Then, the application of $T$ replaces each one of these positive copies by a negative copy. Thus, the segments $S_1, S, S_2$ are removed and replaced by one segment $S^*$ that consists of $S_1$, the new negative copies of the edges of $Q$, and $S_2$. Hence, $W'$ has fewer (by 2) segments than $W$. Therefore, because $W$ is sensible, outer-terminal, has no special U-turn and is pushed onto $R$, it is clear that $W'$ also has these properties. Now, we show that $W'$ also has the property that it is extremal. Clearly, because $W$ is extremal, the multiplicity of $W'$ is also upper bounded by $2n$, and for any two parallel edges $e_i, e_j \in E(W')$ that are not parallel to an edge of $Q$, where $i \ge 1$ and $j \le -1$, we have $(i-1) + |j+1| \ge 2n$. Now, consider some two edges $e_i, e_j \in E(W')$ that are parallel to an edge $e_0$ of $Q$, where $i \ge 1$ and $j \le -1$. Let $e_t$ be the edge of $S$ parallel to $e_i$ and $e_j$, and let $e_r$ be the edge with whom $e_t$ is replaced in the application of $t$ (thus, $e_t \in E(W) \setminus E(W')$ and $e_r \in E(W') \setminus E(W)$). Without loss of generality, suppose that $t \ge 1$. Then, by the applicability of $T$, we have that $i \ge t+1$ and $j \le r$. Thus, $(i-1) + |j+1| \ge t + |r+1| = (t-1) + |(r-1)+1|$. Furthermore, by the definition of a move-through tuple, either $e_{r-1} \in E(W)$ or $r = -2n$. In the first case, because $W$ is extremal, we obtain that $(t-1) + |(r-1)+1| \ge 2n$, and in the second case we obtain that $(t-1) + |(r-1)+1| \ge 2n$ as well (because $t \ge 1$). + +It remains to prove that the potential of $\mathcal{W}'$ is upper bounded by the potential of $\mathcal{W}$. For this purpose, we consider several cases as follows. First, suppose that $S_1, S$ and $S_2$ belong to the same segment group in $\mathcal{W}$, then the only effect on the potential that might increase it is the removal of the two crossings at the endpoints of $S$. However, by the definition of a swollen segment, these crossings have opposite labels, and hence their removal does not affect the potential. (Possibly the segment group that contained $S_1, S$ and $S_2$ has shrunk to a singleton group with respect to $\mathcal{W}'$, but this does not increase the potential.) Now, suppose that only one among $S_1$ and $S_2$ is in the same segment group as $S$ in $\mathcal{W}$, and without loss of generality, suppose that it is $S_1$. Then, in $\mathcal{W}'$ the segment $S^*$ belongs to a singleton group (as its endpoints belong to different maximal degree-2 paths of $R$ or it has an endpoint in $V_{=1}(R) \cup V_{\ge 3}(R)$). Moreover, in $\mathcal{W}$ the segment $S_2$ belongs to a singleton segment group. Thus, one singleton segment group has been replaced by another, and the size of existing segment groups might have shrunk. Since the only change in terms of crossing is that the crossing at the endpoints of $S$ were eliminated, and as in the previous case, this does not effect the potential, we conclude that the potential does not increase. + +Lastly, we consider the case where $S_1, S$ and $S_2$ belong to different segment groups in SegGro($\mathcal{W}$). These three groups are singleton groups —$S_1$ and $S_2$ have endpoints that belong to different maximal degree-2 paths of $R$ (or an endpoint in $V_{=1}(R) \cup V_{\ge 3}(R)$), and $S$ lies in between them. In $\mathcal{W}'$, these three groups are eliminates, which results in a decrease of 3 in the potential. If $S^*$ forms a singleton segment group, the overall the potential decreases by 2. Else, $S^*$ joins an existing segment group, and we have several subcases as follows. In the first subcase, suppose that $S^*$ joins only the group that contains the segment $\tilde{S}_1$ of $\mathcal{W}$ consecutive to $S_1$ in $\mathcal{W}$ (that is not $S$). Then, the crossing at the endpoint of $S_2$ is now contributing (1 or -1) to the sum of labels in the potential of the group, and if $\tilde{S}_1$ was in a singleton group in $\mathcal{W}$, then so are its two crossings. Overall, this results in a contribution of at most 3, so in total the potential does not increase. The subcase where $S^*$ joins only the group that contains the segment $\tilde{S}_2$ of $\mathcal{W}$ consecutive to $S_2$ is symmetric. Now, consider the subcase where $S^*$ joins both of these groups and hence we merge them with respect to $\mathcal{W}'$. In this subcase, we have four new \ No newline at end of file diff --git a/samples/texts/1754951/page_61.md b/samples/texts/1754951/page_61.md new file mode 100644 index 0000000000000000000000000000000000000000..08ce38eeab50c44b63b18975b6da0c36665c6b10 --- /dev/null +++ b/samples/texts/1754951/page_61.md @@ -0,0 +1,29 @@ +crossings that may contribute to the sum of labels, but we have also merged two groups which +makes the potential decreased by at least 1, so overall the potential does not increase. +☐ + +Lastly, we assert that all swollen segments can be eliminated as follows. + +**Lemma 8.17.** Let $(G, S, T, g, k)$ be a good Yes-instance of Planar Disjoint Paths, and $R$ be a backbone Steiner tree. Then, there exists a sensible, outer-terminal, extremal weak linkage $W$ in $H$ that is pushed onto $R$, has no special U-turns and swollen segments, is discretely homotopic in $H$ to some solution of $(G, S, T, g, k)$ and Potential($W$) $\le \alpha_{\text{potential}}(k)$. + +*Proof.* By Lemma 8.12, there exists a sensible, outer-terminal, extremal weak linkage in $H$ that has no special U-turns, is pushed onto $R$, discretely homotopic in $H$ to some solution of $(G, S, T, g, k)$ and whose potential is upper bounded by $\alpha_{\text{potential}}(k)$. Among all such weak linkages, let $W$ be one with minimum number of segments. To conclude the proof, it suffices to argue that $W$ has no swollen segments. + +Suppose, by way of contradiction, that $W$ has at least one swollen segment. Then, by Lemma 8.14, $W$ has an innermost swollen segment $S$. By Lemma 8.15, the move-through tuple $T$ of $S$ is well-defined, and its application is valid. Furthermore, by Lemma 8.16, the resulting weak linkage $W'$ is sensible, outer-terminal, extremal, pushed onto $R$, has no special U-turns and fewer segments than $W$, and its potential is upper bounded by the potential of $W$. Since discrete homotopy is an equivalence relation, $W'$ is discretely homotopic to some solution of $(G, S, T, g, k)$. However, this is a contradiction to the choice of $W$. $\square$ + +Lastly, we prove that having eliminated all swollen segments indeed implies that the total number of segments is small. + +**Lemma 8.18.** Let $(G, S, T, g, k)$ be a nice instance of Planar Disjoint Paths, and $R$ be a Steiner tree. Let $W$ be a sensible, outer-terminal, extremal weak linkage that is pushed onto $R$ and has no special U-turns and swollen segments. Then, $|\text{Seg}(W)| \le \text{Potential}(W)$. + +*Proof.* To prove that $|\text{Seg}(W)| \le \text{Potential}(W)$, it suffices to show that for every segment group $W \in \text{SegGro}(W)$, we have that $|\text{Seg}(W)| \le \text{Potential}(W)$. For segment groups of size 1, this inequality is immediate. Thus, we now consider a segment group $W \in \text{SegGro}(W)$ of size at least 2. Then, + +$$ \text{Potential}(W) = 1 + \sum_{(e,e') \in E(W) \times E(W)} |\text{label}_P^W(e,e')|, $$ + +where $P$ is the maximal degree-2 path of $R$ such that all of the endpoints of all of the segments in $\text{Seg}(W)$ are its internal vertices. Because there do not exist swollen segments, we have that $\text{label}_P^W$ assigns either only non-negative values (0 or 1) or only non-positive values (0 or -1). Without loss of generality, suppose that it assigns only non-negative values. Now, notice that every pair of edges consecutively visited by $W$ that below to different segments of $W$ is assigned 1 (because it creates a crossing with $P$). However, the number of segments of $W$ is upper bounded by one plus the number of such pairs of edges. Thus, $|\text{Seg}(W)| - 1 \le \sum_{(e,e') \in E(W) \times E(W)} |\text{label}_P^W(e,e')|$. From this, we conclude that $|\text{Seg}(W)| \le \text{Potential}(W)$. $\square$ + +## **8.4 Completion of the Simplification** + +The purpose of this section is to prove the following lemma. + +**Lemma 8.19.** Let $(G, S, T, g, k)$ be a good Yes-instance of Planar Disjoint Paths, and $R$ be a backbone Steiner tree. Then, there exists a simplified weak linkage $W$ in $H$ that is discretely homotopic in $H$ to some solution of $(G, S, T, g, k)$. + +To this end, we first eliminate all remaining (non-special) U-turns based on Lemmas 8.11, 8.17 and 8.18, similarly to the proof of Lemma 8.12. \ No newline at end of file diff --git a/samples/texts/1754951/page_64.md b/samples/texts/1754951/page_64.md new file mode 100644 index 0000000000000000000000000000000000000000..9b0f39fe11f6844cbcd6faf86818c29cb3fa0744 --- /dev/null +++ b/samples/texts/1754951/page_64.md @@ -0,0 +1,29 @@ +Figure 28: Illustration of the outerplanar graph in Lemma 9.1. + +Thus, for every vertex $v \in V^*(R)$, $|pairing_v|$ is bounded by twice the degree of $v$ in $R$. Since +$|V^*(R)| \le 12k$, the sum of the degrees in $R$ of the vertices in $V^*(R)$ is upper bounded by $24k$. +From this, we conclude that $|\bigcup_{v \in V^*(R)} pairing_v| \le \alpha_{npair}(k)$. $\square$ + +Now, we associate a pairing with a pushed weak linkage. + +**Definition 9.3 (Pairing of a Weak Linkage).** Let $(G, S, T, g, k)$ be an instance of Planar Disjoint Paths. Let $R$ be a Steiner tree, and let $W$ be a weak linkage pushed onto $R$. For a vertex $v \in V^*(R)$, the pairing of $W$ at $v$ is the set that contains every pair of edges $(e, e')$ in $R$ that are incident to $v$ and such that there exists at least one walk in $W$ where $e_i$ and $e'_j$ occur consecutively, where $e_i$ and $e'_j$ are parallel copies of $e$ and $e'$, respectively. More generally, the pairing of $W$ is the collection $\{pairing_u\}_{u \in V^*(R)}$, where $pairing_u$ is the pairing of $W$ at $u$ for every vertex $u \in V^*(R)$. + +Apart from a pairing, to be able to reconstruct a simplified weak linkage, we need additional information in the form of an assignment of numbers to the pairs in the pairing. To this end, we have the definition of a template. + +**Definition 9.4 (Template).** Let $(G, S, T, g, k)$ be an instance of Planar Disjoint Paths. Let $R$ be a Steiner tree. Let $pairing_v$ be a pairing at some vertex $v \in V^*(R)$. A template for $pairing_v$ is a function template$_v$: $pairing_v \to \mathbb{N}$. If the maximum integer assigned by template$_v$ is upper bounded by $N$, for some $N \in \mathbb{N}$, then it is also called an $N$-template. More generally, a template (resp. $N$-template) of a pairing $\{pairing_u\}_{u \in V^*(R)}$ is a collection $\{template_u\}_{u \in V^*(R)}$, where $template_u$ is a template (resp. $N$-template) for $pairing_u$ for every vertex $u \in V^*(R)$. + +We proceed to associate a template with a weak linkage. + +**Definition 9.5 (Template of Weak Linkage).** Let $(G, S, T, g, k)$ be an instance of Planar Disjoint Paths. Let $R$ be a Steiner tree, and let $\mathcal{W}$ be a weak linkage pushed onto $R$. Let $\{\text{pairing}_u\}_{u \in V^*(R)}$ be the pairing of $\mathcal{W}$. For a vertex $v \in V^*(R)$, the template of $\mathcal{W}$ at $v$ is the function template$_v$: $\text{pairing}_v \to \mathbb{N}$ such that for every $(e, e') \in \text{pairing}_v$, we have + +$$ \text{template}_v((e, e')) = | \{ \{\hat{e}, \hat{e}'\} \mid \begin{array}{l} \hat{e} \text{ is parallel to } e, \hat{e}' \text{ is parallel to } e', \\ \exists W \in \mathcal{W} \text{ s.t. } W \text{ traverses } \hat{e} \text{ and } \hat{e}' \text{ consecutively} \end{array} \}| $$ + +More generally, the template of $\mathcal{W}$ is the collection $\{\text{template}_u\}_{u \in V^*(R)}$, where $\text{template}_u$ is the +template of $\mathcal{W}$ at $u$ for every vertex $u \in V^*(R)$. + +Now, we claim that the pairing of a pushed weak linkage non-crossing. + +**Lemma 9.2.** Let $(G, S, T, g, k)$ be an instance of Planar Disjoint Paths. Let $R$ be a Steiner tree. +Then, the pairing of any weak linkage $W$ pushed onto $R$ is non-crossing. + +*Proof.* Let $\{\text{pairing}_v\}_{v \in V^*(R)}$ be the pairing of $\mathcal{W}$. Suppose, by way of contradiction, that $\{\text{pairing}_v\}_{v \in V^*(R)}$ is crossing. If $v \in V_{=1}(R) \cup V_2^*(R)$, then $\text{pairing}_v$ is trivially non-crossing. Thus, there exists a vertex $v \in V_{\ge 3}(R)$ such that $\text{pairing}_v$ is crossing. Let $e^1, e^2, \dots, e^r$ be the \ No newline at end of file diff --git a/samples/texts/1754951/page_65.md b/samples/texts/1754951/page_65.md new file mode 100644 index 0000000000000000000000000000000000000000..c0b2bdd70ab5afb6258336d51855a00b5e8c8f88 --- /dev/null +++ b/samples/texts/1754951/page_65.md @@ -0,0 +1,27 @@ +edges in $E(R)$ incident to $v$ in clockwise order. Because $\text{pairing}_v$ is crossing, there exist two pairs $(e^i, e^j)$ and $(e^x, e^y)$ in $\text{pairing}_v$, where $i < j$ and $x < y$, such that $i < x < j < y$ or $x < i < y < j$. By the definition of $\text{pairing}_v$, this means that there exist walks $\overline{W}, \overline{W} \in W$ (possibly $\overline{W} = \overline{W}$) and edges $\hat{e}^i, \hat{e}^j, \bar{e}^x$ and $\bar{e}^y$ that are parallel to $e^i, e^j, e^x$ and $e^y$, respectively, such that $\overline{W}$ traverses $\hat{e}^i$ and $\hat{e}^j$ consecutively, and $\overline{W}$ traverses $\bar{e}^x$ and $\bar{e}^y$ consecutively. However, because $i < x < j < y$ or $x < i < y < j$, and because parallel edges incident to each vertex appear consecutively in its cyclic order, we derive that $(v, \hat{e}^i, \hat{e}^j, \bar{e}^x, \bar{e}^y)$ is a crossing of $\overline{W}$ and $\overline{W}$. $\square$ + +From Lemmas 9.1 and 9.2, we obtain the following corollary. + +**Corollary 9.1.** Let $(G, S, T, g, k)$ be an instance of Planar Disjoint Paths with a backbone Steiner tree $R$. Let $W$ be a weak linkage pushed onto $R$ with a pairing $\{\text{pairing}_v\}_{v \in V^*(R)}$. Then, $$|\bigcup_{v \in V^*(R)} \text{pairing}_v| \le \alpha_{\text{npair}}(k).$$ + +Additionally, we claim that we can focus our attention on pushed flows whose templates are $\alpha_{\text{weight}}(k)$-templates. + +**Lemma 9.3.** Let $(G, S, T, g, k)$ be a good Yes-instance of Planar Disjoint Paths with a backbone Steiner tree $R$. Then, there exists a simplified weak linkage that is discretely homotopic in $H$ to some solution of $(G, S, T, g, k)$, whose template is an $\alpha_{\text{mul}}(k)$-template. + +*Proof.* By Lemma 8.19, there exists a simplified weak linkage $W$ that is discretely homotopic in $H$ to some solution of $(G, S, T, g, k)$. Because $W$ is simplified, its multiplicity is upper bounded by $\alpha_{\text{mul}}(k)$. Let $\{\text{pairing}_v\}_{v \in V^*(R)}$ and $\{\text{template}_v\}_{v \in V^*(R)}$ be the pairing and template of $W$, respectively. Consider some vertex $v \in V^*(R)$ and pair $(e, e') \in \text{pairing}_v$. To complete the proof, we need to show that $\text{template}_v((e, e')) \le \alpha_{\text{mul}}(k)$. By the definition of a template, + +$$ \text{template}_v((e, e')) = | \{ \{\hat{e}, \bar{e}'\} \mid \hat{e} \text{ is parallel to } e, \bar{e}' \text{ is parallel to } e', \\ \exists W \in W \text{ s.t. } W \text{ traverses } \hat{e} \text{ and } \bar{e}' \text{ consecutively} \}| $$ + +Thus, because every the walks in $W$ are edge-disjoint and each walk in $W$ visits distinct edges (by the definition of a weak linkage), $\text{template}_v((e, e'))$ is upper bounded by the number of edges parallel to $e$ that belong to $E(W)$. Thus, by the definition of the multiplicity of a linkage, we conclude that $\text{template}_v((e, e')) \le \alpha_{\text{mul}}(k)$. $\square$ + +In light of Corollary 9.1 and Lemma 9.3, we define the set of all pairings and templates in which we will be interested as follows. + +**Definition 9.6 (The Set ALL).** Let $(G, S, T, g, k)$ be an instance of Planar Disjoint Paths. Let $R$ be a Steiner tree. The set ALL contains every collection $\{(pairing_v, template_v)\}_{v \in V^*(R)}$ where $\{\text{pairing}_v\}_{v \in V^*(R)}$ is a non-crossing pairing that satisfies $|\bigcup_{v \in V^*(R)} \text{pairing}_v| \le \alpha_{\text{npair}}(k)$, and $\{\text{template}_v\}_{v \in V^*(R)}$ is an $\alpha_{\text{mul}}(k)$-template for $\{\text{pairing}_v\}_{v \in V^*(R)}$. + +From Corollary 9.1 and Lemma 9.3, we have the following result. + +**Corollary 9.2.** Let $(G, S, T, g, k)$ be a good Yes-instance of Planar Disjoint Paths. Let $R$ be a backbone Steiner tree. Then, there exists a simplified weak linkage that is discretely homotopic in $H$ to some solution of $(G, S, T, g, k)$ and satisfies the following property: There exists $\{(pairing_v, template_v)\}_{v \in V^*(R)} \in \text{ALL}$ such that $\{\text{pairing}_v\}_{v \in V^*(R)}$ is the pairing of $W$, and $\{\text{template}_v\}_{v \in V^*(R)}$ is the template of $W$. + +Because we only deal with pairings having just $O(k)$ pairs and upper bound the largest integer assigned by templates by $2^{O(k)}$, the set ALL is “small” as asserted by the following lemma. + +**Lemma 9.4.** Let $(G, S, T, g, k)$ be an instance of Planar Disjoint Paths. Let $R$ be a Steiner tree. Then, $|\text{ALL}| = 2^{O(k^2)}$. Moreover, ALL can be computed in time $2^{O(k^2)}$. \ No newline at end of file diff --git a/samples/texts/1754951/page_66.md b/samples/texts/1754951/page_66.md new file mode 100644 index 0000000000000000000000000000000000000000..25d359702d5493b684373d44cbc27e87ec1cc204 --- /dev/null +++ b/samples/texts/1754951/page_66.md @@ -0,0 +1,17 @@ +*Proof.* First, we upper bound the number of pairings $\{\text{pairing}_v\}_{v \in V^*(R)}$ that satisfy $|\bigcup_{v \in V^*(R)} \text{pairing}_v| \le \alpha_{npair}(k)$. By Observation 6.1, the number of edges in $E^*(R)$ is at most $24k$, and hence the number of pairs of edges in $E^*(R)$ is at most $(24k)^2$. Thus, the number of choices for $\bigcup_{v \in V^*(R)} \text{pairing}_v$ is at most $\binom{24k}{\alpha_{npair}(k)}^2$. Note that each pair of edges in this union belongs to $\text{pairing}_v$ for at most one vertex $v \in V^*(R)$. From this, we conclude that the number of pairings $\{\text{pairing}_v\}_{v \in V^*(R)}$ that satisfy $|\bigcup_{v \in V^*(R)} \text{pairing}_v| \le \alpha_{npair}(k)$ is at most $\binom{24k}{\alpha_{npair}(k)}^2 \cdot 2^{\alpha_{npair}(k)} = 2^{\mathcal{O}(k \log k)}$ (as $\alpha_{npair}(k) = \mathcal{O}(k)$). + +Now, fix some pairing $\{\text{pairing}_v\}_{v \in V^*(R)}$ that satisfies $|\bigcup_{v \in V^*(R)} \text{pairing}_v| \le \alpha_{npair}(k)$. We test if this pairing is a non-crossing pairing, at each vertex $v \in V^*(R)$, by testing all possible 4-tuples of edges in $E_R(v)$. This takes $k^{\mathcal{O}(1)}$ time in total. Then, the number of $\alpha_{mul}(k)$-templates for $\{\text{pairing}_v\}_{v \in V^*(R)}$ is upper bounded by $(\alpha_{mul}(k))^{\alpha_{npair}(k)} = 2^{\mathcal{O}(k^2)}$ (as $\alpha_{mul}(k) = 2^{\mathcal{O}(k)}$ and $\alpha_{npair}(k) = \mathcal{O}(k)$). Thus, we have that $|\text{ALL}| = 2^{\mathcal{O}(k^2)}$. It should also be clear that these arguments, by simple enumeration, imply that ALL can be computed in time $2^{\mathcal{O}(k^2)}$. $\square$ + +**Extension of Pairings and Templates.** To describe the reconstruction of simplified weak linkages from their pairings and templates, we must extend them from $V^*(R)$ to all of $V(R)$. Intuitively, this extension is based on the observation that $\mathcal{W}$ is U-turn free and sensible. Therefore, if a walk in $\mathcal{W}$ visits a maximal degree-2 path in $R$, then it must traverse the entirety of this path. Hence, the pairings and templates at any internal vertex of a degree-2 path can be directly obtained from the endpoint vertices of the path. We begin by identifying which collections of pairings and templates can be extended. + +**Definition 9.7.** Let $(G, S, T, g, k)$ be an instance of Planar Disjoint Paths, and let $R$ be a backbone Steiner tree. And let $\mathcal{A} = \{\langle \text{pairing}_v, \text{template}_v \rangle\}_{v \in V^*(R)}$. Then $\mathcal{A}$ is extensible to all of $V(R)$ if the following conditions are true for every maximal degree-2 path in $R$. + +• Let $u, v \in V_2^*(R)$ such that they lie on the same maximal degree-2 path of $R$. Consider the subpath $\text{path}_R(u, v)$ with endpoints $u$ and $v$, and let $e^u$ and $e^v$ be the edges in $E(\text{path}_R(u, v))$ incident on $u$ and $v$, respectively. Suppose that $e^u$ and $e^v$ are distinct edges. Then $(e, e^u) \in \text{pairing}_u$ if and only if $(e', e^v) \in \text{pairing}_v$, where $E_R(u) = \{e, e^u\}$ and $E_R(v) = \{e', e^v\}$. + +• Assuming that the above condition is true, furthermore $\text{template}_u(e, e^u) = \text{template}_v(e', e^v)$. + +The following lemma shows that pairings and templates of simplified weak linkages are extensible to all of $V(R)$. + +**Lemma 9.5.** Let $(G, S, T, g, k)$ be a good instance of Planar Disjoint Paths. Let $R$ be a backbone Steiner tree. Let $\mathcal{W}$ be a simplified weak linkage with pairing and template $\mathcal{A} = \{\langle \text{pairing}_v, \text{template}_v \rangle\}_{v \in V^*(R)}$. Then, $\mathcal{A}$ is extensible to all of $V(R)$. + +*Proof.* Let $u, v \in V_2^*(R)$ such that they lie on the same maximal degree-2 path of $R$. Consider the subpath $\text{path}_R(u, v)$ with endpoints $u$ and $v$, and let $e^u$ and $e^v$ be the edges in $E(\text{path}_R(u, v))$ incident on $u$ and $v$, respectively. Suppose that $e^u$ and $e^v$ are distinct edges. Then $\text{path}_R(u, v)$ must have an internal vertex. Let $V(\text{path}_R(u, v)) = \{u = w_0, w_1, w_2, \dots, w_p, w_{p+1} = v\}$ and let $E(\text{path}_R(u, v)) = \{e^i = \{w_i, w_{i+1}\} | 0 \le i \le p\}$ where $e^0 = e^u$ and $e^p = e^v$. Consider an internal vertex $w_i$ of $\text{path}_R(u, v)$ and note that $E_R(w_i) = \{e^{i-1}, e^i\}$. Observe that, as $\mathcal{W}$ is U-turn free, $w^i \notin N_R(V_{=1}(R))$ and the endpoints of every walk in $\mathcal{W}$ lies in $V_{=1}(R)$, there is no walk in $\mathcal{W}$ that visits two parallel copies of an edge in $E_R(w_i)$ consecutively, as that would constitute a U-turn. Therefore, any walk in $\mathcal{W}$ that visits $e^{i-1}_{j_{i-1}}$ must also visit $e^i_{j_i}$, where $e^{i-1}_{j_{i-1}}$ and $e^i_{j_i}$ are parallel copies of $e^{i-1}$ and $e^i$ respectively. This holds for all vertices $w_1, w_2, \dots, w_p$. Let $E^*(\text{path}_R(u, v))$ be the collection of all the parallel copies of every edge in $E(\text{path}_R(u, v))$. Then, for a walk $W \in W$, let $\tilde{W}_1, \tilde{W}_2, \dots, \tilde{W}_t$ be the maximal subwalks \ No newline at end of file diff --git a/samples/texts/1754951/page_67.md b/samples/texts/1754951/page_67.md new file mode 100644 index 0000000000000000000000000000000000000000..98e4a81071ed9c5280718ef72ce0d2c1f8c68bc7 --- /dev/null +++ b/samples/texts/1754951/page_67.md @@ -0,0 +1,17 @@ +Figure 4: Segments arising from the crossings of a walk with the Steiner tree. + +Figure 5: Separators and flows for a long maximal degree-2 path *P* in *R*. + +the proof where we analyze a (hypothetical) solution towards pushing it onto R. Our most basic notion in this analysis is that of a *segment*, defined as follows (see Fig. 4). + +**Definition 2.2.** For a walk W in the radial completion of G that is edge-disjoint from R, a segment is a maximal subwalk of W that does not “cross” R. + +Let $\text{Seg}(W)$ denote the set of segments of $W$. Clearly, $\text{Seg}(W)$ is a partition of $W$. Ideally, we would like to upper bound the number of segments of (all the paths of) a solution by $2^{\mathcal{O}(k^2)}$. However, this will not be possible because, while $R$ is easily seen to have only $\mathcal{O}(k)$ vertices of degree 1 or at least 3, it can have “long” maximal degree-2 paths which can give rise to numerous segments (see Fig. 4). To be more concrete, we say that a maximal degree-2 path of $R$ is *long* if it has more than $2^{ck}$ vertices (for some constant $c$), and it is *short* otherwise. Then, as the paths of a solution are vertex disjoint, the following observation is immediate. + +**Observation 2.1.** Let $\mathcal{P}$ be a solution. Then, its number of segments that have at least one endpoint on a short path, or a vertex of degree other than 2, of $R$, is upper bounded by $2^{\mathcal{O}(k)}$. + +To deal with segments crossing only long paths, several new ideas are required. In what follows, we first explain how to handle segments going across different long paths, whose number can be bounded (unlike some of the other types of segments we will encounter). + +**Segments Between Different Long Paths.** To deal with such segments, we modify $R$ (in the second step of its construction). For each long path $P$ with endpoints $u$ and $v$, we will compute two minimum-size vertex sets, $S_u$ and $S_v$, such that $S_u$ separates (i.e., intersects all paths with one endpoint in each of the two specified subgraphs) the following subgraphs in the radial completion of $G$: (i) the subtree of $R$ that contains $u$ after the removal of a vertex $u_1$ of $P$ that is “very close” to $u$, and (ii) the subtree of $R$ that contains $v$ after the removal of a vertex $u_2$ that is “close” to $u$. The condition satisfied by $S_v$ is symmetric (i.e. $u$ and $v$ switch their roles; see Fig. 5). Here, “very close” refers to distance $2^{c_1k}$ and “close” refers to distance $2^{c_2k}$ on the path, for some constants $c_1 < c_2$. Let $u'$ and $v'$ be the vertices of $P$ in the intersection with the separators $S_u$ and $S_v$ respectively. (The selection of $u'$ not to be $u$ itself is of use in the third modification of $R$.) + +To utilize these separators, we need their sizes to be upper bounded by $2^{\mathcal{O}(k)}$. For our initial $R$, such small separators may not exist. However, the modification we present now will \ No newline at end of file diff --git a/samples/texts/1754951/page_68.md b/samples/texts/1754951/page_68.md new file mode 100644 index 0000000000000000000000000000000000000000..362b9fa6904e43de35854f4c34cf6887ff447121 --- /dev/null +++ b/samples/texts/1754951/page_68.md @@ -0,0 +1,25 @@ +of W restricted to $E^*(\mathbf{path}_R(u, v))$. Then each $\widehat{W}_i$ is a path from $u$ to $v$ along the parallel copies of $e^0, e^1, \dots, e_p$. Hence, if $(e, e^u) = (e, e_0) \in \text{pairing}_u$ if and only if $(e', e^v) = (e', e_p) \in \text{pairing}_v$ where $e$ and $e'$ are the edges in $E(R) \setminus E(\mathbf{path}_R(u, v))$ that are incident on $u$ and $v$, respectively. Further, observe that, each time a walk $W \in W$ visits a parallel copy of $e^u = e^0$ (immediately after visiting a parallel copy of $e$) it must traverse a parallel copy of $e^1, e^2, \dots, e^p = e^v$ consecutively (and then immediately visit a parallel copy of $e'$), and vice-versa. Therefore by definition, $\text{template}_u(e, e^u) = \text{template}_{w_1}(e^u, e^1) = \text{template}_{w_2}(e^1, e^2) = \dots = \text{template}_v(e_{p-1}, e^v) = \text{template}_v(e', e^v)$. $\square$ + +We now have the following definition. + +**Definition 9.8 (Extension of Pairings and Templates).** Let $(G, S, T, g, k)$ be a good instance of Planar Disjoint Paths. Let $R$ be a backbone Steiner tree. Let $\mathcal{A} = \{(\text{pairing}_v, \text{template}_v)\}|_{v \in V^*(R)\setminus R}\}$. + +Then, the extension of $\mathcal{A}$ to $V(R)$ is the collection $\tilde{\mathcal{A}} = \{(\widehat{\text{pairing}}_v, \widehat{\text{template}}_v)\}|_{v \in V(R)}$ such that: + +* If $\mathcal{A}$ is not extensible (Definition 9.7), then $\tilde{\mathcal{A}}$ is invalid. + +* Otherwise, $\mathcal{A}$ is extensible and we have two cases. + +- If $v \in V_{=1}(R) \cup V_{\ge 3}(R)$, then $\widehat{\text{pairing}}_v = \text{pairing}_v$ and $\widehat{\text{template}}_v = \text{template}_v$. + +- Otherwise, $v \in V_{=2}(R)$ and let $u, w \in V_{=1}(R) \cup V_{\ge 3}(R)$ such that $v \in V(\mathbf{path}_R(u, v))$. Let $e_u$ and $e_w$ be the two edges in $\mathbf{path}_R(u, v)$ incident on $u$ and $w$, respectively. If $(e, e_u) \notin \text{pairing}_u$ for any $e \in E_R(u)$, then $\text{pairing}_v = \emptyset$. Otherwise, there is some $e \in E_R(u)$ such that $(e, e_u) \in \text{pairing}_u$. Then $\text{pairing}_v = \{(e', e'')\}$ where $e'$ and $e''$ are the two edges in $E(R)$ incident on $v$. Further, $\text{template}_v(e', e'') = \text{template}_u(e, e_u)$. + +Let $\widehat{\mathcal{ALL}}$ denote the collection of extensions of all the pairings and templates in ALL. Then we have the following corollary of Lemma 9.4. + +**Lemma 9.6.** Let $(G, S, T, g, k)$ be an instance of Planar Disjoint Paths. Let $R$ be a Steiner tree. Then, $\lvert\widehat{\mathcal{ALL}}\rvert = 2^{\mathcal{O}(k^2)}$. Moreover, $\widehat{\mathcal{ALL}}$ can be computed in time $2^{\mathcal{O}(k^2)n}$. + +*Proof.* Given $\mathcal{A} = \{(\text{pairing}_v, \text{template}_v)\}|_{v \in V^*(R)} \in \mathsf{ALL}$, we apply Definition 9.8 to obtain the extension $\tilde{\mathcal{A}}$. Note that, for every vertex $v \in V^*(R)$, we have that $\lvert\text{pairing}_v\rvert = \mathcal{O}(k)$ and the numbers assigned by $\text{template}_v$ are bounded by $2^{\mathcal{O}(k)}$. Further, since $\lvert V^*(R) \rvert \le 12k$, we can test the conditions in Definition 9.8 in $k^{\mathcal{O}(1)}$ time. Finally we can construct $\tilde{\mathcal{A}} = \{(\widehat{\text{pairing}}_v, \widehat{\text{template}}_v)\}|_{v \in V(R)}$ in total time $2^{\mathcal{O}(k)n}$, as $\lvert V(R) \rvert \le n$. Since $\lvert\mathsf{ALL}\rvert = 2^{\mathcal{O}(k^2)}$ and it can be enumerated in time $2^{\mathcal{O}(k^2)}$, it follows that $\lvert\widehat{\mathsf{ALL}}\rvert = 2^{\mathcal{O}(k^2)}$ and it can be enumerated in $2^{\mathcal{O}(k^2)n}$ time. $\square$ + +In the rest of this section, we only require pairings and templates that are extended to all of $V(R)$. For convenience, we abuse the notation to denote the extension of a collection of pairings and templates, $\mathcal{A} \in \mathsf{ALL}$ to all of $V(R)$, by $\{\text{pairing}_v\}|_{v \in V(R)}$ and $\{\text{template}_v\}|_{v \in V(R)}$, respectively. The following corollary follows from the definition of $\widehat{\mathsf{ALL}}$ and Corollary 9.2. + +**Corollary 9.3.** Let $(G, S, T, g, k)$ be a good Yes-instance of Planar Disjoint Paths. Let $R$ be a backbone Steiner tree. Then, there exists a simplified weak linkage that is discretely homotopic in $H$ to some solution of $(G, S, T, g, k)$ and satisfies the following property: There exists $\{( (\text{pairing}_v, \text{template}_v) ) |_{v \in V(R)} \in \widehat{\mathsf{ALL}}$ such that $\{\text{pairing}_v\} |_{v \in V(R)}$ is the pairing of $W$, $\{\text{template}_v\} |_{v \in V(R)}$ is the template of $W$. \ No newline at end of file diff --git a/samples/texts/1754951/page_69.md b/samples/texts/1754951/page_69.md new file mode 100644 index 0000000000000000000000000000000000000000..8cd1c8c937344f1cb3b380f813105c2f9528bef7 --- /dev/null +++ b/samples/texts/1754951/page_69.md @@ -0,0 +1,35 @@ +**Stitching of Weak Linkages.** Let us now introduce the notion of a stitching, which gives a localized view of a weak linkage pushed onto $R$ at each vertex in $V(R)$. Intuitively, the stitching at a vertex $v \in V(R)$ is function on the set of edges incident on $v$, that maps each edge to the next (or the previous) edge in a weak linkage. Note that, only a subset of the edges incident on $v$ may participate in a weak linkage. Therefore, we introduce the following notation: an edge is mapped to $\perp$ to indicate that it is not part of weak linkage. Also recall that, for a vertex $v \in V(R)$, order$_v$ is an enumeration of the edges in $\hat{E}_R(v)$ in either clockwise or anticlockwise order, where $\hat{E}_R(v) = \{e \in E_H(v) | e \text{ is parallel to an edge } e' \in E(R)\}$. + +**Definition 9.9 (Stitching at a Vertex).** Let $(G, S, T, g, k)$ be a good Yes-instance of Planar Disjoint Paths, and let $R$ be a backbone Steiner tree. For a vertex $v \in V(R)$, a function $f_v$: $E_H(v) \to E_H(v) \cup \perp$ is a stitching at $v$ if it satisfies the following conditions. + +* For any edge $e \in E_H(v) \setminus \hat{E}_R(v)$, $f_v(e) = \perp$. + +* For a pair of (possibly non-distinct) edges $e, e' \in \hat{E}_R(v)$, $f_v(e) = e'$ if and only if $f_v(e') = e$. + +* If $v \in S \cup T$, then there is exactly one edge such that $f_v(e) = e$. Otherwise, there is no such edge. + +* If $e_1, e_2, e_3, e_4 \in \hat{E}_R(v)$ such that $f_v(e_1) = e_2$ and $f_v(e_3) = e_4$. Then $\{e_1, e_2\}$ and $\{e_3, e_4\}$ are disjoint and non-crossing in order$_v$.²¹ + +Let $\{f_v\}_{v \in V(R)}$ be a collection of functions such that $f_v$ is a stitching at $v$ for each $v \in V(R)$. Then, this collection is called a stitching if for every edge $e = \{u, v\} \in E_H(R)$, $f_u(e) = \perp$ if and only if $f_v(e) = \perp$. + +Let us now describe the stitching of a weak linkage that is pushed onto $R$. + +**Definition 9.10 (Stitching of a Weak Linkage Pushed onto R).** Let $(G, S, T, g, k)$ be a good Yes-instance of Planar Disjoint Paths, and let $R$ be a backbone Steiner tree. Let $\mathcal{W}$ be a weak linkage pushed onto $R$. Then we define the stitching of $\mathcal{W}$ as the collection of functions $\{\text{stitch}_v\}_{v \in V(R)}$, where $\text{stitch}_v: E_H(v) \to E_H(v) \cup \perp$ satisfies the following. + +* If there is $W \in \mathcal{W}$ where $e = \{v, w\}$ is the first edge of $W$, then $v \in S \cup T$ and $\text{stitch}_v(e) = e$. + +* If there is $W \in \mathcal{W}$ where $e = \{u, v\}$ is the last edge of $W$, then $v \in S \cup T$ and $\text{stitch}_v(e) = e$. + +* If there is a walk $W \in \mathcal{W}$ such that $e, e' \in E_H(v)$ are consecutive edges with a common endpoint $v \in V(R)$, then $\text{stitch}_v(e) = e'$ and $\text{stitch}_v(e') = e$. + +* If $e \in E_H(v)$ is not part of any walk in $\mathcal{W}$, then $\text{stitch}_v(e) = \perp$. + +It is easy to verify that $\{\text{stitch}_v\}_{v \in V(R)}$ is indeed a stitching. Let us make a few more observations on the properties of this stitching. + +**Observation 9.1.** Let $(G, S, T, g, k)$ be a good Yes-instance of Planar Disjoint Paths, and let $R$ be a backbone Steiner tree. Let $\mathcal{W}$ be a weak linkage pushed onto $R$ and let $\{\text{stitch}_v\}_{v \in V(R)}$ be the stitching of $\mathcal{W}$. Let $\{\text{pairing}_v\}_{v \in V(R)}$ and $\{\text{template}_v\}_{v \in V(R)}$ be the pairing and template of $\mathcal{W}$, respectively. Then the following holds. + +* Let $e_i, e'_j \in E_H(v)$, then $\text{stitch}_v(e_i) = e'_j$ if and only if $\text{stitch}_v(e'_j) = e_i$. + +* Let $e, e' \in E_R(v)$. Then $(e, e') \in \text{pairing}_v$ if and only if there is a pair $e_i, e'_j$ of edges in $E_H(R)$, where $e_i$ is parallel to $e$ and $e'_j$ is parallel to $e'$, such that $\text{stitch}_v(e_i) = e'_j$ and $\text{stitch}_v(e'_j) = e_i$. Further, the number of pairs of parallel edges is equal to $\text{template}_v(e, e')$. + +²¹That is, in a clockwise (or anticlockwise) enumeration of $\hat{E}_R(v)$ starting from $e_1$, these edges occur as either $e_1, e_2, e_3, e_4$ or $e_1, e_3, e_4, e_2$, where without loss of generality we assume that $e_3$ occurs before $e_4$ in this ordering. \ No newline at end of file diff --git a/samples/texts/1754951/page_7.md b/samples/texts/1754951/page_7.md new file mode 100644 index 0000000000000000000000000000000000000000..706e86d1367ca2df0dafb43cc5fd14371a39bcbf --- /dev/null +++ b/samples/texts/1754951/page_7.md @@ -0,0 +1,27 @@ +word (the unique word of length 0) is denoted by 1. We say that a word $w = a_1a_2 \cdots a_t$ over $\Sigma \cup \Sigma^{-1}$ is reduced if there does not exist $i \in \{1, 2, \dots, t-1\}$ such that $a_i = a_{i+1}^{-1}$. We denote the (infinite) set of reduced words over $\Sigma \cup \Sigma^{-1}$ by $\text{RW}(\Sigma)$. The concatenation $w \circ \hat{w}$ of two words $w = a_1a_2 \cdots a_t$ and $\hat{w} = b_1b_2 \cdots b_\ell$ is the word $w^* = a_1a_2 \cdots a_t b_1 b_2 \cdots b_\ell$. The product $w \cdot \hat{w}$ of two words $w = a_1a_2 \cdots a_t$ and $\hat{w} = b_1b_2 \cdots b_\ell$ in $\text{RW}(\Sigma)$ is a word $w^*$ defined as follows: + +$$w^* = a_1a_2 \cdots a_{t-r}b_{r+1}b_{r+2} \cdots b_\ell$$ + +where $r$ is the largest integer in $\{1, 2, \dots, \min(t, \ell)\}$ such that, for every $i \in \{1, 2, \dots, r\}$, $b_i = a_{t+1-i}^{-1}$. Note that $w^*$ is a reduced word, and the product operation is associative. The reduction of a word $w = a_1a_2 \cdots a_t$ over $\Sigma \cup \Sigma^{-1}$ is the (reduced) word $w^* = a_1 \cdot a_2 \cdots a_t$. + +**Definition 3.2 (Homology).** Let $D$ be a directed plane graph with outer face $f$, and denote the set of faces of $D$ by $\mathcal{F}$. Let $\Sigma$ be an alphabet. Two functions $\phi, \psi : A(D) \to \text{RW}(\Sigma)$ are homologous if there exists a function $h : \mathcal{F} \to \text{RW}(\Sigma)$ such that $h(f) = 1$, and for every arc $e \in A(D)$, we have $h(f_1)^{-1} \cdot \phi(e) \cdot h(f_2) = \psi(e)$ where $f_1$ and $f_2$ are the faces at the left-hand side and the right-hand side of $e$, respectively. + +The following observation will be useful in later results. + +**Observation 3.1.** Let $\alpha, \beta, \gamma : A(D) \to \text{RW}(\Sigma)$ be three functions such that $\alpha, \beta$ and $\beta, \gamma$ are pairs of homologous functions. Then, $\alpha, \gamma$ is also a pair of homologous functions. + +*Proof.* Let $f$ and $g$ be the functions witnessing the homology of $\alpha, \beta$ and $\beta, \gamma$, respectively. Then, it is easy to check that the function $h = g \circ f$ (i.e. the composition of $g$ and $f$) witnesses the homology of $\alpha, \gamma$. $\square$ + +Towards the definition of flow, we denote an instance of Directed Planar Disjoint Paths by a tuple $(D, S, T, g, k)$ where $D$ is a directed plane graph, $S, T \subseteq V(D)$, $k = |S|$ and $g : S \to T$. We assume that $g$ is bijective because otherwise the given instance is a No-instance. A solution of an instance $(D, S, T, g, k)$ of Directed Planar Disjoint Paths is a set $\mathcal{P}$ of pairwise vertex-disjoint directed paths in $D$ that contains, for every vertex $s \in S$, a path directed from $s$ to $g(s)$. When we write $\text{RW}(T)$, we treat $T$ as an alphabet—that is, every vertex in $T$ is treated as a symbol. + +**Definition 3.3 (Flow).** Let $(D, S, T, g, k)$ be an instance of Directed Planar Disjoint Paths. Let $\phi : A(D) \to \text{RW}(T)$ be a function. For any vertex $v \in V(D)$, denote the concatenation $\phi(e_1)^{\epsilon_1} \circ \phi(e_2)^{\epsilon_2} \cdots \phi(e_r)^{\epsilon_r}$ by $\text{conc}(v)$, where $e_1, e_2, \dots, e_r$ are the arcs incident to $v$ in clockwise order where the first arc $e_1$ is chosen arbitrarily, and for each $i \in \{1, 2, \dots, r\}$, $\epsilon_i = 1$ if $v$ is the head of $e_i$ and $\epsilon_i = -1$ if $v$ is the tail of $e_i$. Then, the function $\phi$ is a flow if:⁷ + +1. For every vertex $v \in V(D) \setminus (S \cup T)$, the reduction of $\text{conc}(v)$ is 1. + +2. For every vertex $v \in S$, (i) $\text{conc}(v) = a_0a_1 \dots a_{\ell-1}$ is a word of length $\ell \ge 1$, and (ii) there exists $i \in \{0, 1, \dots, \ell-1\}$ such that the reduction of $a_i a_{(i+1)} \bmod{\ell} \cdots a_{(i+\ell-1)} \bmod{\ell}$ equals $a_i$, where $a_i = g(v)$ if the arc $e$ associated with $a_i$ has $v$ as its tail, $a_i = g(v)^{-1}$ otherwise. + +3. For every vertex $v \in T$, (i) $\text{conc}(v) = a_0a_1\dots a_{\ell-1}$ is a word of length $\ell \ge 1$, and (ii) there exists $i \in \{0, 1, \dots, \ell-1\}$ such that the reduction of $a_i a_{(i+1)} \bmod{\ell} \cdots a_{(i+\ell-1)} \bmod{\ell}$ equals $a_i$, where $a_i = v$ if the arc $e$ associated with $a_i$ has $v$ as its head, $a_i = v^{-1}$ otherwise. + +In the above definition, the conditions on the reduction of $\text{conc}(v)$ for each vertex $v \in V(D)$ are called *flow conservation* constraints. Informally speaking, these constraints resemble “usual” flow conservation constraints and ensure that every two walks that carry two different alphabets do not cross. The association between solutions to $(D, S, T, g, k)$ and flows is defined as follows. + +⁷We note that there is slight technical difference between our definition and the definition in Section 3.1 in [42]. There, a flow must put only a single alphabet ($v$ or $v^{-1}$ in $T \cup T^{-1}$) on the arcs incident on vertices in $S \cup T$. \ No newline at end of file diff --git a/samples/texts/1754951/page_70.md b/samples/texts/1754951/page_70.md new file mode 100644 index 0000000000000000000000000000000000000000..10dd7593beb6791236007827f6fc4603681dd91b --- /dev/null +++ b/samples/texts/1754951/page_70.md @@ -0,0 +1,29 @@ +* If $e_i, e'_j$ and $e_p^*, \hat{e}_q$ are pairs of edges in $E_H(v)$ such that $\text{stitch}_v(e_i) = e'_j$ and $\text{stitch}_v(e_p^*) = \hat{e}_q$, then the pairs $e_i, e'_j$ and $e_p^*, \hat{e}_q$ are non-crossing in order $v$. + +* If the multiplicity of $\mathcal{W}$ is upperbounded by $\ell$, then for each edge $e = \{u, v\} \in E(R)$, $\left|\{e' \in E(H) \mid e' \text{ is parallel to } e \text{ and } \text{stitch}_v(e') \neq \perp\}\right| \le k \cdot \ell$. + +## 9.2 Translating a Template Into a Stitching + +Given (i) an instance $I = (G, S, T, g, k)$ of Planar Disjoint Paths, (ii) a backbone Steiner tree $R$, and (iii) a collection $\{(pairing_v, template_v)\}|_{v \in V(R)} \in \overline{\text{ALL}}$, our current objective is to either determine that $\{(pairing_v, template_v)\}|_{v \in V(R)}$ is invalid or construct a multiplicity function $\ell$ and a stitching $\{f_v\}|_{v \in V(R)}$ to reconstruct the weak linkage. The cases where we determine that $\{(pairing_v, template_v)\}|_{v \in V(R)}$ is invalid will be (some of the) cases where there exists no simplified weak linkage whose pairing and template are $\{pairing_v\}|_{v \in V(R)}$ and $\{template_v\}|_{v \in V(R)}$, respectively. Let us begin with the notion of multiplicity function $\ell$ of a collection of pairings and templates, as follows. + +**Definition 9.11 (Multiplicity Function).** Let $(G, S, T, g, k)$ be an instance of Planar Disjoint Paths, and $R$ be a Steiner tree. Let $\mathcal{A} = \{(pairing_v, template_v)\}|_{v \in V(R)} \in \overline{\text{ALL}}$. For every vertex $v \in V^*(R)$, let $\ell_v$ be the function that assigns $\sum_{e':(e,e')\in pairing_v} \text{template}_v((e,e'))$ to every edge $e \in E(R)$ incident to $v$. If one of the following conditions is satisfied, then the multiplicity function extracted from $\mathcal{A}$ is invalid. + +1. There exists an edge $e = \{u,v\}$ such that $u,v \in V^*(R)$ and $\ell_u(e) \neq \ell_v(e)$. + +2. There exists a terminal $v \in S \cup T$ such that $\text{pairing}_v = \emptyset$. + +Otherwise, the multiplicity function extracted from $\mathcal{A}$ is valid and it is the function $\ell: E_{1,3+}(R) \to \mathbb{N}_0$ such that for each $e \in E_{1,3+}(R)$, $\ell(e) = \ell_v(e)$ where $v$ is an endpoint of $e$ in $V^*(R)^{.22}$ + +Let $\mathcal{W}$ be a weak linkage pushed onto $R$. The multiplicity function of a $\mathcal{W}$ is defined as the multiplicity function $\ell$ extracted from $\mathcal{A}$, the pairings and templates of $\mathcal{W}$. It is clear that the multiplicity of $\mathcal{W}$ is $\max_{e \in E(R)} \ell(e)$. + +**Observation 9.2.** Let $(G, S, T, g, k)$ be an instance of Planar Disjoint Paths with a simplified weak linkage $\mathcal{W}$, and let $\ell$ is the multiplicity function of $\mathcal{W}$. For any $e \in E(R)$, $\ell(e) \le \alpha_{\text{mul}}(k)$. + +Having extracted a multiplicity function, we turn to extract a stitching. Towards this, recall the embedding of $H$ with respect to $R$, and the resulting enumeration of edges around vertices in $V(R)$ (see Section 6.6). Let us now describe the stitching extraction at a terminal vertex. + +**Definition 9.12 (Stitching Extraction at Terminals).** Let $(G, S, T, g, k)$ be a nice instance of Planar Disjoint Paths. Let $R$ be a Steiner tree. Consider a collection $\mathcal{A} = \{(pairing_v, template_v)\}|_{v \in V(R)} \in \overline{\text{ALL}}$. Let $\ell$ be the multiplicity function extracted from $\mathcal{A}$, and suppose that $\ell$ is valid. Let $v \in S \cup T$, and let $e^*$ be the unique edge in $E(R)$ incident to $v$. If $\ell(e^*)$ is an even number, then the local stitching extracted from $\mathcal{A}$ at $v$ is invalid. Otherwise, the stitching extracted from $\mathcal{A}$ at $v$ is valid and it is the involution $f_v : E_H(v) \to E_H(v) \cup \perp$ defined as follows. + +$$ f_v(e) = \begin{cases} e_{\ell(e)+1-i}^* & \text{if } e = e_i^* \text{ and } 1 \le i \le \ell(e) \\ \perp & \text{otherwise.} \end{cases} $$ + +Next, we describe how to extract a stitching at a vertex $v \in V_{=2}(R) \cup V_{\ge 3}(R)$. + +22The choice of the endpoint when both belong to $V^*(R)$ is immaterial by the definition of invalidity. \ No newline at end of file diff --git a/samples/texts/1754951/page_71.md b/samples/texts/1754951/page_71.md new file mode 100644 index 0000000000000000000000000000000000000000..fbd0c1c67f5ebeba1850f7de6f000510d117d462 --- /dev/null +++ b/samples/texts/1754951/page_71.md @@ -0,0 +1,31 @@ +**Definition 9.13 (Stitching Extraction at Non-Terminals).** Let $(G, S, T, g, k)$ be a nice instance of Planar Disjoint Paths. Let $R$ be a backbone Steiner tree. Consider a collection $\mathcal{A} = \{(\text{pairing}_v, \text{template}_v)\}_{v \in V(R)} \subset \widehat{\text{ALL}}$. Let $\ell$ be the multiplicity function extracted from $\mathcal{A}$, and suppose that $\ell$ is valid. Let $v \in V_2(R) \cup V_{\ge 3}(R)$, and suppose that let $e^1, e^2, \dots, e^r$ denote the arcs in $E(R)$ incident to $v$ enumerated as per $\text{order}_v$ starting from $e^1$. Then the define a function $f_v : E_H(v) \to E_H(v) \cup \perp$ as follows. + +* For each $(e, e') \in \text{pairing}_v$ such that $\text{template}_v(e, e') > 0$, where $e$ occurs before $e'$ in $\text{order}_v$, let $\text{inner}(e, e') = \{e^* \in E_R(v) \mid e^*$ occurs between $e$ and $e'$ in $\text{order}_v\}$, and $\text{outer}(e, e') = \{e^* \in E_R(v) \mid \text{either } e^* \text{ occurs before } e \text{ or occurs after } e' \text{ in } \text{order}_v\}$. + +* Then, for each $i \in \{1, \dots, \text{template}_v(e, e')\}$, let $f_v(e_{i+x}) = e'_{y-i}$ and $f_v(e'_{y-i}) = e_{i+x}$ where + +$$x = \sum_{e^* \in \text{outer}(e,e')} \text{template}_v(e, e^*),$$ + +and + +$$y = 1 + \text{template}_v(e, e') + \sum_{e^* \in \text{inner}(e,e')} \text{template}_v(e, e^*)$$ + +* For all other edges in $E_H(v)$, define $f_v(e) = \perp$. + +If the assignment $f_v$ is fixed point free, then $f_v$ is the stitching extracted from $\mathcal{A}$ at $v$, which is said to be valid. Otherwise, it is invalid + +Lastly, based on Definitions 9.12 and 9.13, we extract the stitching as follows, + +**Definition 9.14 (Stitching Extraction).** Let $(G, S, T, g, k)$ be a nice instance of Planar Disjoint Paths. Let $R$ be a Steiner tree. Consider a collection $\mathcal{A} = \{(\text{pairing}_v, \text{template}_v)\}_{v \in V(R)} \subset \widehat{\text{ALL}}$. For each $v \in V(R)$, let $f_v$ be the stitching extracted from $\mathcal{A}$ at $v$. Then the stitching extracted from $\mathcal{A}$ is invalid if it satisfies one of the following conditions. + +* There is a vertex $v \in V_1(R)$ such that the stitching extracted from $\mathcal{A}$ at $v$ is invalid. + +* There is an edge $e = \{u, v\} \in E(H)$ parallel to an edge in $E(R)$ such that $f_u(e) = \perp$ and $f_v(e) \neq \perp$. + +Otherwise, the stitching extracted from $\mathcal{A}$ is valid and defined as the collection $\{f_v\}_{v \in V(R)}$ where $f_v$ is the stitching extracted from $\mathcal{A}$ at $v$ for every $v \in V(R)$. + +Less obviously, we also show that in case we are given a collection in $\widehat{\text{ALL}}$ that corresponds to weak linkage, not only is the stitching extracted from that collection valid, but also, most crucially, it is the stitching we were originally given (under the assumption that the pair of flow and stitching we deal with is simplified). In other words, we are able to faithfully reconstruct a stitching from the template of weak linkage. The implicit assumption in this lemma that $\{(\text{pairing}_v, \text{template}_v)\}_{v \in V(R)}$ belongs to $\widehat{\text{ALL}}$ is supported by Corollary 9.3. + +**Lemma 9.7.** Let $(G, S, T, g, k)$ be a good Yes-instance of Planar Disjoint Paths, and let $R$ be a backbone Steiner tree. Let $\mathcal{W}$ be a simplified weak linkage in $H$ and let $\ell$ be the multiplicity function of $\mathcal{W}$. Consider the collection $\mathcal{A} = \{(\text{pairing}_v, \text{template}_v)\}_{v \in V(R)} \subset \widehat{\text{ALL}}$, such that it is the collection of pairings and templates of $\mathcal{W}$. Let $\{f_v\}_{v \in V(R)}$ be the stitching extracted from $\mathcal{A}$. Then for every vertex $v \in V_1(R)$, $\textit{stitch}_v(e) = f_v(e)$ for every edge $e \in E_H(v)$. + +*Proof.* Let $E^*(v) = \{e_i^* | 1 \le i \le l(e^*)\}$ where $e_i^*$ denotes the *i*-th parallel copy of $e^*$, in the enumeration in $\text{order}_v$. Observe that, since $\mathcal{W}$ is simplified, $E^*(v)$ is exactly the set of edges from $E_H(v)$ that appear in $\mathcal{W}$. Hence, for any edge $e \in E_H(v)$, if $e \notin E^*(v)$ then $\textit{stitch}_v(e) = \perp$. \ No newline at end of file diff --git a/samples/texts/1754951/page_73.md b/samples/texts/1754951/page_73.md new file mode 100644 index 0000000000000000000000000000000000000000..b63cab3f694af4f92e8a65701788fe26005042e4 --- /dev/null +++ b/samples/texts/1754951/page_73.md @@ -0,0 +1,19 @@ +$E_R(v)$ and $\hat{e}_j$ is the $j$-th parallel copy of $\hat{e}$. Let us choose $e'$ (with respect to $e$) so that the $x_{e,e'}$ is minimized, and then choose $e_i$ such that $i$ is minimized. Here, note that $i > x_{e,e'}$. Let us argue that $\hat{e} = e'$. Suppose not, and note that $(e, \hat{e}) \in \text{pairing}_v$ and $\text{template}_v(e, \hat{e}) > 0$. Then we have three cases depending on the position of these edges in $\text{order}_v$, either $e < \hat{e} < e'$, or $\hat{e} < e < e'$, or $e < e' < \hat{e}$. Consider the first case, and note that every parallel copy of $\hat{e}$ occurs before all parallel copies of $e'$ and after all parallel copies of $e$ in $\text{order}_v$. We claim that for any $e_p \in \{e_{i+1}, \dots, e_{\ell(e)}\}$, $\text{stitch}_v(e_p) \notin \{e'_1, \dots, e'_{\ell(e')} \}$. If this claim were false, then observe that, as $e < \hat{e} < e'$, we have $e_i < e_p < \hat{e}_j < \text{stitch}_v(e_p)$ in $\text{order}_v$. Hence $e_i, \hat{e}_j$ and $e_p, \text{stitch}_v(e_p)$ are crossing pairs at $v$, in the weak linkage $\mathcal{W}$, which is a contradiction. On the other hand, if $\text{stitch}_v(e_p) \notin \{e'_1, \dots, e'_{\ell(e')} \}$ for any $e_p \in \{e_{i+1}, \dots, e_{\ell(e)}\}$, then we claim that $\text{stitch}_v$ maps strictly fewer than $\text{template}_v(e, e')$ edges from $\{e_1, \dots, e_{\ell(e)}\}$ to $\{e'_1, \dots, e'_{\ell(e')} \}$. Indeed, we choose $e'$ such that $x_{e,e'}$ is minimized, and hence the edges in $\{e_1, \dots, e_{x_{e,e'}}\}$ are not mapped to any edge in $\{e'_1, \dots, e'_{\ell(e')} \}$. And since, no edge in $\{e_i, e_{i+1}, \dots, e_{\ell(e)}\}$ maps to $\{e'_1, \dots, e'_{\ell(e')} \}$, only the edges in $\{e(x_{e,e'}+1), \dots, e_{i-1}\}$ remain, which is strictly fewer than $\text{template}_v(e, e')$. But this contradicts the definition of $\text{template}_v(e, e')$. Hence, it cannot be the case that $e < \hat{e} < e'$ in $\text{order}_v$. Next, consider the case when $e < e' < \hat{e}$. Note that $\hat{e} \in \text{outer}(e, e')$, and by definition $x_{e,\hat{e}} < x_{e,e'}$. Since we choose $e'$ to minimize $x_{e,e'}$, and we didn't choose $e' = \hat{e}$, $\text{stitch}_v$ maps the edges in $\{e(x_{e,\hat{e}}+1), \dots, e(x_{e,\hat{e}}+\text{template}_v(e,\hat{e}))\}$ to $\text{template}_v(e, \hat{e})$ edges in $\{\hat{e}_1, \dots, \hat{e}_{\ell(\hat{e})}\}$. Therefore, if $\text{stitch}_v(e_i) = \hat{e}_j$, then there are $\text{template}_v(e, \hat{e}) + 1$ parallel copies of $e$ that are mapped to parallel copies $\hat{e}$, which is a contradiction. Hence it is not possible that $e < e' < \hat{e}$. The last case, $\hat{e} < e < e'$ is similar to the previous case, since $\hat{e} \in \text{outer}(e, e')$ in this case as well. Hence, we conclude that if $\text{stitch}_v(e_i) = \hat{e}_j$ then $\hat{e} = e'$. Therefore, when $e$ occurs before $e'$ in $\text{order}_v$, $\text{stitch}_v$ maps each edge in $\{e(x_{e,e'}+1), \dots, e(x_{e,e'}+\text{template}_v(e,e'))\}$ to some edge in $\{e'_1, e'_2, \dots, e'_{\ell(e')} \}$. + +By a symmetric argument, we obtain that for any edge in $\{e'(y_{e,e'}-\text{template}_v(e,e'))$, $\dots$, $e'(y_{e,e'}-1)\}$ maps to an edge in $\{e_1, e_2, \dots, e_{\ell(e)}\}$.²³ + +We now proceed to further restrain the mapping of edges to the ranges determined by $x_{e,e'}, y_{e,e'}$ and $\text{template}_v(e, e')$. + +**Claim 9.2.** *Consider a pair $(e, e') \in \text{pairing}_v$ such that $\text{template}_v(e, e') > 0$ Then, stitch$_v$ maps $\{e(x_{e,e'}+1), \dots, e(x_{e,e'}+\text{template}_v(e,e'))\}$ to $\{e'(y_{e,e'}-\text{template}_v(e,e')), \dots, e'(y_{e,e'}-1)\}$ and vice-versa²⁴.* + +*Proof.* Suppose not, and without loss of generality assume that $e$ occurs before $e'$ in $\text{order}_v$. Then consider the case when there is an edge $e_i \in \{e(x_{e,e'}+1), \dots, e(x_{e,e'}+\text{template}_v(e,e'))\}$ such that $\text{stitch}_v(e_i) = e'_j$, where either $j > y_{e,e'} - 1$ or $j < y_{e,e'} - \text{template}_v(e, e')$. Then consider the collection $\{e'_j\} \cup \{e'(y_{e,e'}-\text{template}_v(e,e')), \dots, e'(y_{e,e'}-1)\}$, and observe that each edge in this collection is mapped to a distinct edge in $\{e_1, e_2, \dots, e_{\ell(e)}\}$. But then there are $\text{template}_v(e, e') + 1$ edges in $\{e'_1, \dots, e'_{\ell(e')} \}$ that map to an edge in $\{e_1, \dots, e_{\ell(e)}\}$ under stitch$_v$. This is a contradiction to the definition of $\text{template}_v(e, e')$. ◇ + +Finally, we show that stitch$_v$ is equal to $f_v$. + +**Claim 9.3.** *Consider a pair $(e, e') \in \textit{pairing}_v$ such that $\textit{template}_v(e, e') > 0$. Then for each $i \in \{1, 2, \dots, \textit{template}_v(e, e')\}$, stitch$_v(e_{x_{e,e'}+i}) = e'_j$ and stitch$_v(e'_{y_{e,e'}-i}) = e_{x_{e,e'}+i}$.* + +*Proof.* Suppose not and consider the case when $\textit{stitch}_v(e_{x_{e,e'}+i}) = e'_j$ where $j \neq y_{e,e'} - i$. Note that $e'_j \in \{e'(y_{e,e'}-\textit{template}_v(e,e'))$, $\dots$, $e'(y_{e,e'}-1)\}$ by previous arguments. Consider the case when $j > y_{e,e'} - i$. We claim that, the edges in $\{e'(y_{e,e'}-\textit{template}_v(e,e'))$, $\dots$, $e'_{(j-1)}\}$ must map + +²³Note that, this is equivalent to the case when $e'$ occurs before $e$ in $\textit{order}_v$. Here, we obtain a contradiction by choosing $e$ (with respect to $e'$) that maximizes $y_{e,e'}$, and then choosing the maximum $i$ such that $y_{e,e'} - \textit{template}_v(e, e') \le i \le y_{e,e'}$ and $\textit{stitch}_v(e'_i) \notin \{e_1, \dots, e_\ell(e)\}$. + +²⁴Note that, by definition of stitch$_v$, this immediately implies the other direction. \ No newline at end of file diff --git a/samples/texts/1754951/page_75.md b/samples/texts/1754951/page_75.md new file mode 100644 index 0000000000000000000000000000000000000000..1fe3058fcc411c3a84e20599e12f1f38be968c4d --- /dev/null +++ b/samples/texts/1754951/page_75.md @@ -0,0 +1,26 @@ +* For each $v \in S \cup T$, Let $e_v \in E_H(v)$ be the unique edge such that $f_v(e_v) = e_v$. + +* Then the walk $W_v$ is defined as the sequence of edges $e_0, e_1, e_2, \dots, e_{p_v}$, where $e_{v_0} = e_v$, and for each $i \in \{0, \dots, p_v - 1\}$, the edge $e_i = \{v_i, v_{i+1}\}$ satisfies (i) $f_{v_{i+1}}(e_i) = e_{i+1}$ where $v_0 = v$; and (ii) $f_{v_{p_v}}(e_{p_v}) = p_v$. + +* We iteratively construct a sequence of walks $W_{v_1}, W_{v_2}, \dots$, where the walk $W_{v_i}$ starts from a vertex $v_i$ that is not the endpoint of any of the previous walks. Finally, we output $\mathcal{W}$ as the collection of these walks. + +It is clear that running time for the construction of a weak linkage from a stitching $\{f_v\}_{v \in V(R)}$ is upperbounded by the number of pairs of edges in $E_H(V(R))$ such that they are images of each other in the stitching. The following observation is follows directly from Definition 9.15 and Definition 9.10. and the fact that + +**Observation 9.3.** Let $(G, S, T, g, k)$ be a good instance of Planar Disjoint Paths, and let $R$ be a backbone Steiner tree. Let $\mathcal{W}$ be a weak linkage that is pushed onto $R$, and let $\{\text{stitch}_v\}_{v \in V(R)}$ be the stitching of $\mathcal{W}$. Then the weak linkage constructed from this stitching is equal to $\mathcal{W}$. + +Let $\mathcal{W}_{\text{ALL}}$ denote the collection of weak linkages extracted from $\widetilde{\text{ALL}}$. The following Lemma is the main result of this section. + +**Lemma 9.11.** Let $(G, S, T, g, k)$ be a good Yes-instance of Planar Disjoint Paths, and let $R$ be a backbone steiner tree. Then, there exists a collection of $2^{\mathcal{O}(k^2)}$ simplified linkages $\mathcal{W}_{\text{ALL}}$ such that there is a weak linkage $\mathcal{W} \in \mathcal{W}_{\text{ALL}}$ that is discretely homotopic in $H$ to some solution of $(G, S, T, g, k)$. Further, collection can be enumerated in $2^{\mathcal{O}(k^2)n}$ time. + +*Proof.* By Lemma 8.19, the given instance has a solution that is discretely homotopic to some simplified weak linkage $\mathcal{W}$, and by Corollary 9.3, the pairing and template of $\mathcal{W}$, denoted by $\mathcal{A}$ lies in the collection $\widetilde{\text{ALL}}$. Then, by Lemma 9.9, the stitching $\{f_v\}_{v \in V(R)}$ is equal to $\{\text{stitch}_v\}_{v \in V(R)}$, the stitching of $\mathcal{W}$, and it can be computed in $2^{\mathcal{O}(k)n}$ time by Lemma 9.10. Finally, we can construct a weak linkage $\mathcal{W}'$ from the stitching $\{f_v\}_{v \in V(R)}$, and by Observation 9.3, $\mathcal{W}' = \mathcal{W}$. Note that, as $\mathcal{W}$ is a simplified weak linkage, its multiplicity is upperbounded by $\alpha_{\text{mul}}(k) = 2^{\mathcal{O}(k)}$. Hence by Observation 9.1 and 9.2, the number of pairs in $(e, e') \in E_H(v) \times E_H(v)$ such that $f_v(e) = e'$ is upperbounded by $k \cdot \alpha_{\text{mul}}(k) = 2^{\mathcal{O}(k)}$. Hence, it is clear that we reconstruct $\mathcal{W}$ from the stitching $\{f_v\}_{v \in V(R)}$ in time $2^{\mathcal{O}(k)n}$ time. + +To enumerate the collection $\mathcal{W}_{\text{ALL}}$, we iterate over $\widetilde{\text{ALL}}$. For each $\mathcal{A} \in \widetilde{\text{ALL}}$, we attempt to construct a stitching and if it returns an invalid stitching, we move on to next iteration. Otherwise, we construct a weak linkage from this stitching and output it. Observe that for each $\mathcal{A} \in \widetilde{\text{ALL}}$ we can compute the corresponding weak linkage $\mathcal{W}$, if it exists, in time $2^{\mathcal{O}(k)n}$ time. Since $|\widetilde{\text{ALL}}| = 2^{\mathcal{O}(k^2)}$, clearly $|\mathcal{W}_{\text{ALL}}| = 2^{\mathcal{O}(k^2)}$ and it can be enumerated in $2^{\mathcal{O}(k^2)n}$ time, we can enumerate $\mathcal{W}_{\text{ALL}}$ in $2^{\mathcal{O}(k^2)n}$ time. $\square$ + +# 10 The Algorithm + +Having set up all required definitions and notions, we are ready to describe our algorithm. +Afterwards, we will analyze its running time and prove its correctness. + +## 10.1 Execution of the Algorithm + +We refer to this algorithm as PDPAlg. It takes as input an instance $(\tilde{G}, \tilde{S}, \tilde{T}, \tilde{g}, k)$ of Planar Disjoint Paths, and its output is the decision whether this instance is a Yes-instance. The specification of the algorithm is as follows. \ No newline at end of file diff --git a/samples/texts/1754951/page_76.md b/samples/texts/1754951/page_76.md new file mode 100644 index 0000000000000000000000000000000000000000..03a7499c4c93f043a857ec0f0ce8ca9eb984d1ca --- /dev/null +++ b/samples/texts/1754951/page_76.md @@ -0,0 +1,31 @@ +**Step I: Preprocessing.** First, PDPAlg invokes Corollary 4.1 to transform $(\tilde{G}, \tilde{S}, \tilde{T}, \tilde{g}, k)$ into an equivalent good instance $(G, S, T, g, k)$ of Planar Disjoint Paths where $|V(G)| = \mathcal{O}(|V(\tilde{G})|)$. + +**Step II:** Computing a Backbone Steiner Tree. Second, PDPAlg invokes Lemma 6.8 with respect to $(G, S, T, g, k)$ to compute a backbone Steiner tree, denoted by $R$. Then, PDPAlg computes the embedding of $H$ with respect to $R$ (Section 6.6). + +**Step III:** Looping on $W_{\text{ALL}}$. Now, PDPAlg invokes Lemma 9.11 to enumerate $W_{\text{ALL}}$. For each weak linkage $\mathcal{W} \in W_{\text{ALL}}$, the algorithm applies the algorithm of Corollary 5.1 to $(G, S, T, g, k)$ and $\mathcal{W}$; if the algorithm finds a solution, then PDPAlg determines that $(\tilde{G}, \tilde{S}, \tilde{T}, \tilde{g}, k)$ is Yes-instance and terminates. + +**Step IV: Reject.** PDPAlg determines that $(\tilde{G}, \tilde{S}, \tilde{T}, \tilde{g}, k)$ is a No-instance and terminates. + +## 10.2 Running Time and Correctness + +Let us first analyze the running time of PDPAlg. + +**Lemma 10.1.** *PDPAlg runs in time $2^{\mathcal{O}(k^2)} n^{\mathcal{O}(1)}$.* + +*Proof.* By Corollary 4.1, the computation of $(G, S, T, g, k)$ in Step I is performed in time $2^{\mathcal{O}(k)} n^2$. By Lemma 6.8, the computation of $R$ in Step II is performed in time $2^{\mathcal{O}(k)} n^{3/2} \log^3 n$. Moreover, the computation of $R$ can clearly be done in time linear in $n$. Let $H = H_G$ be the radial completion of $G$ enriched with $4|V(G)| + 1$ parallel copies of each edge, and note that $|V(H)| = \mathcal{O}(|V(G)|)$. By Observation 6.9, we can compute the embedding of $H$ with respect to $R$ in time $\mathcal{O}(n^2)$. By Lemma 9.11, $|W_{\text{ALL}}| = 2^{\mathcal{O}(k^2)}$, and it can be enumerated in $2^{\mathcal{O}(k^2)} n$ time. Finally, for each $W \in W_{\text{ALL}}$, Corollary 5.1 takes $n^{\mathcal{O}(1)}$ time to test if there is a solution that is discretely homotopic to $\mathcal{W}$. Thus, PDPAlg runs in time $2^{\mathcal{O}(k^2)} n^{\mathcal{O}(1)}$. $\square$ + +The reverse direction of the correctness of PDPAlg is trivially true. + +**Lemma 10.2.** *Let $(\tilde{G}, \tilde{S}, \tilde{T}, \tilde{g}, k)$ be an instance of Planar Disjoint Paths. If $(\tilde{G}, \tilde{S}, \tilde{T}, \tilde{g}, k)$ is accepted by PDPAlg, then $(\tilde{G}, \tilde{S}, \tilde{T}, \tilde{g}, k)$ is a Yes-instance.* + +Now, we handle the forward direction of the correctness of PDPAlg. + +**Lemma 10.3.** *Let $(\tilde{G}, \tilde{S}, \tilde{T}, \tilde{g}, k)$ be an instance of Planar Disjoint Paths. If $(\tilde{G}, \tilde{S}, \tilde{T}, \tilde{g}, k)$ is a Yes-instance, then $(\tilde{G}, \tilde{S}, \tilde{T}, \tilde{g}, k)$ is accepted by PDPAlg.* + +*Proof.* Suppose that $(\tilde{G}, \tilde{S}, \tilde{T}, \tilde{g}, k)$ is a Yes-instance of Planar Disjoint Paths. Then, by Corollary 4.1, $(G, S, T, g, k)$ is a Yes-instance of Planar Disjoint Paths. Then by Lemma 8.19 and Lemma 9.11, there is a collection $W_{\text{ALL}}$ of $2^{\mathcal{O}(k^2)}$ simplified weak linkages containing at least one simplified weak linkage $W^*$, that is discretely homotopic in $H$ to some solution of $(G, S, T, g, k)$. Here $H$ is the radial completion of $G$ enriched with $4|V(G)| + 1$ parallel copies of each edge. Then in Step III, Corollary 5.1 ensures that we obtain a solution to $(G, S, T, g, k)$ in the iteration we consider $W^*$. Hence PDPAlg accepts the instance $(\tilde{G}, \tilde{S}, \tilde{T}, \tilde{g}, k)$. $\square$ + +Lastly, we remark that PDPAlg can be easily modified not only to reject or accept the instance $(\tilde{G}, \tilde{S}, \tilde{T}, \tilde{g}, k)$, but to return a solution in case of acceptance, within time $2^{\mathcal{O}(k^2)} n^{\mathcal{O}(1)}$. + +## References + +[1] I. ADLER, 6th workshop on graph classes, optimization, and width parameters 2013, list of open problems. http://www.cs.upc.edu/~sedthilk/grow/Open_Problems_GROW_2013.pdf, 2013. 1, 3 \ No newline at end of file diff --git a/samples/texts/1754951/page_77.md b/samples/texts/1754951/page_77.md new file mode 100644 index 0000000000000000000000000000000000000000..aae6e54b1efe40b96f512b5d27d2ef432ba95162 --- /dev/null +++ b/samples/texts/1754951/page_77.md @@ -0,0 +1,27 @@ +[2] I. ADLER, S. G. KOLLIPOULOS, P. K. KRAUSE, D. LOKSHTANOV, S. SAURABH, AND D. M. THILIKOS, *Irrelevant vertices for the planar disjoint paths problem*, J. Comb. Theory, Ser. B, 122 (2017), pp. 815-843. 1, 2, 3, 7, 10, 15, 16 + +[3] I. ADLER AND P. K. KRAUSE, *A lower bound for the tree-width of planar graphs with vital linkages*, CoRR, abs/1011.2136 (2010). 3 + +[4] J. BASTE AND I. SAU, *The role of planarity in connectivity problems parameterized by treewidth*, Theor. Comput. Sci., 570 (2015), pp. 1-14. 2 + +[5] H. L. BODLAENDER, F. V. FOMIN, D. LOKSHTANOV, E. PENNINKX, S. SAURABH, AND D. M. THILIKOS, (*Meta*) kernelization, J. ACM, 63 (2016), pp. 44:1-44:69. 7, 13, 25 + +[6] C. CHEKURI AND J. CHUZHOY, *Polynomial bounds for the grid-minor theorem*, J. ACM, 63 (2016), pp. 40:1-40:65. 2 + +[7] J. CHUZHOY AND D. H. K. KIM, *On approximating node-disjoint paths in grids*, in Approximation, Randomization, and Combinatorial Optimization. Algorithms and Techniques, APPROX/RANDOM 2015, August 24-26, 2015, Princeton, NJ, USA, vol. 40 of LIPIcs, Schloss Dagstuhl - Leibniz-Zentrum fuer Informatik, 2015, pp. 187-211. 1 + +[8] J. CHUZHOY, D. H. K. KIM, AND S. LI, *Improved approximation for node-disjoint paths in planar graphs*, in Proceedings of the 48th Annual ACM SIGACT Symposium on Theory of Computing, STOC 2016, Cambridge, MA, USA, June 18-21, 2016, ACM, 2016, pp. 556-569. 1 + +[9] J. CHUZHOY, D. H. K. KIM, AND R. NIMAVAT, *New hardness results for routing on disjoint paths*, in Proceedings of the 49th Annual ACM SIGACT Symposium on Theory of Computing, STOC 2017, Montreal, QC, Canada, June 19-23, 2017, ACM, 2017, pp. 86-99. 1 + +[10] ——, *Almost polynomial hardness of node-disjoint paths in grids*, in Proceedings of the 50th Annual ACM SIGACT Symposium on Theory of Computing, STOC 2018, Los Angeles, CA, USA, June 25-29, 2018, ACM, 2018, pp. 1220-1233. 1 + +[11] ——, *Improved approximation for node-disjoint paths in grids with sources on the boundary*, in 45th International Colloquium on Automata, Languages, and Programming, ICALP 2018, July 9-13, 2018, Prague, Czech Republic, vol. 107 of LIPIcs, 2018, pp. 38:1-38:14. 1 + +[12] V. COHEN-ADDAD, É. C. DE VERDIÈRE, P. N. KLEIN, C. MATHIEU, AND D. MEIER-FRANKENFELD, *Approximating connectivity domination in weighted bounded-genus graphs*, in Proceedings of the 48th Annual ACM SIGACT Symposium on Theory of Computing, STOC 2016, Cambridge, MA, USA, June 18-21, 2016, 2016, pp. 584-597. 13 + +[13] S. CORNELSEN AND A. KARRENBAUER, *Accelerated bend minimization*, J. Graph Algorithms Appl., 16 (2012), pp. 635-650. 29 + +[14] M. CYGAN, D. MARX, M. PILIPCZUK, AND M. PILIPCZUK, *The planar directed k-vertex-disjoint paths problem is fixed-parameter tractable*, in 54th Annual IEEE Symposium on Foundations of Computer Science, FOCS 2013, 26-29 October, 2013, Berkeley, CA, USA, 2013, pp. 197-206. 1, 3, 9, 37, 38, 39 + +[15] E. D. DEMAINE, F. V. FOMIN, M. T. HAJIAGHAYI, AND D. M. THILIKOS, *Subexponential parameterized algorithms on bounded-genus graphs and H-minor-free graphs*, J. ACM, 52 (2005), pp. 866-893. 1, 2 \ No newline at end of file diff --git a/samples/texts/1754951/page_79.md b/samples/texts/1754951/page_79.md new file mode 100644 index 0000000000000000000000000000000000000000..a2e3516e8623f578fabfc6a757bb489fe046b5f9 --- /dev/null +++ b/samples/texts/1754951/page_79.md @@ -0,0 +1,27 @@ +[16] E. D. DEMAINE AND M. HAJIAGHAYI, *Linearity of grid minors in treewidth with applications through bidimensionality*, Combinatorica, 28 (2008), pp. 19–36. + +[17] G. DING, A. SCHRIJVER, AND P. D. SEYMOUR, *Disjoint paths in a planar graph—a general theorem*, SIAM Journal on Discrete Mathematics, 5 (1992), pp. 112–116. + +[18] F. V. FOMIN, D. LOKSHTANOV, D. MARX, M. PILIPCZUK, M. PILIPCZUK, AND S. SAURABH, *Subexponential parameterized algorithms for planar and apex-minor-free graphs via low treewidth pattern covering*, in IEEE 57th Annual Symposium on Foundations of Computer Science, FOCS 2016, New Brunswick, New Jersey, USA, IEEE Computer Society, 2016, pp. 515–524. + +[19] F. V. FOMIN, D. LOKSHTANOV, AND S. SAURABH, *Excluded grid minors and efficient polynomial-time approximation schemes*, J. ACM, 65 (2018), pp. 10:1–10:44. + +[20] S. FORTUNE, J. E. HOPCROFT, AND J. WYLLIE, *The directed subgraph homeomorphism problem*, Theor. Comput. Sci., 10 (1980), pp. 111–121. + +[21] A. FRANK, *Packing paths, cuts, and circuits-a survey*, Paths, Flows and VLSI-Layout, (1990), pp. 49–100. + +[22] M. GROHE, K. KAWARABAYASHI, AND B. A. REED, *A simple algorithm for the graph minor decomposition - logic meets structural graph theory*, in Proceedings of the Twenty-Fourth Annual ACM-SIAM Symposium on Discrete Algorithms, SODA 2013, New Orleans, Louisiana, USA, January 6-8, 2013, SIAM, 2013, pp. 414–431. + +[23] B. M. P. JANSEN, D. LOKSHTANOV, AND S. SAURABH, *A near-optimal planarization algorithm*, in Proceedings of the Twenty-Fifth Annual ACM-SIAM Symposium on Discrete Algorithms, SODA 2014, Portland, Oregon, USA, January 5-7, 2014, 2014, pp. 1802–1811. + +[24] R. M. KARP, *Reducibility among combinatorial problems*, in Proceedings of a symposium on the Complexity of Computer Computations, held March 20-22, 1972, at the IBM Thomas J. Watson Research Center, Yorktown Heights, New York, USA, 1972, pp. 85–103. + +[25] R. M. KARP, *On the computational complexity of combinatorial problems*, Networks, 5 (1975), pp. 45–68. + +[26] K. KAWARABAYASHI, Y. KOBAYASHI, AND B. A. REED, *The disjoint paths problem in quadratic time*, J. Comb. Theory, Ser. B, 102 (2012), pp. 424–435. + +[27] K. KAWARABAYASHI AND P. WOLLAN, *A shorter proof of the graph minor algorithm: the unique linkage theorem*, in Proceedings of the 42nd ACM Symposium on Theory of Computing, STOC 2010, Cambridge, Massachusetts, USA, 5-8 June 2010, ACM, 2010, pp. 687–694. + +[28] ——, *A simpler algorithm and shorter proof for the graph minor decomposition*, in Proceedings of the 43rd ACM Symposium on Theory of Computing, STOC 2011, San Jose, CA, USA, 6-8 June 2011, ACM, 2011, pp. 451–458. + +[29] P. N. KLEIN AND D. MARX, *A subexponential parameterized algorithm for subset TSP on planar graphs*, in Proceedings of the Twenty-Fifth Annual ACM-SIAM Symposium on Discrete Algorithms, SODA 2014, Portland, Oregon, USA, January 5-7, 2014, SIAM, 2014, pp. 1812–1830. \ No newline at end of file diff --git a/samples/texts/1754951/page_8.md b/samples/texts/1754951/page_8.md new file mode 100644 index 0000000000000000000000000000000000000000..ccacaefd3410df6f85f858186c7240a20692bd6e --- /dev/null +++ b/samples/texts/1754951/page_8.md @@ -0,0 +1,25 @@ +**Definition 3.4.** Let $(D, S, T, g, k)$ be an instance of Directed Planar Disjoint Paths. Let $\mathcal{P}$ be a solution of $(D, S, T, g, k)$. The flow $\phi: A(D) \to \mathrm{RW}(T)$ associated with $\mathcal{P}$ is defined as follows. For every arc $e \in A(D)$, define $\phi(e) = 1$ if there is no path in $\mathcal{P}$ that traverses $e$, and $\phi(e) = t$ otherwise where $t \in T$ is the end-vertex of the (unique) path in $\mathcal{P}$ that traverses $e$. + +The following proposition, due to Schrijver [42], also holds for the above definition of flow. + +**Proposition 3.3** (Proposition 5 in [42]). There exists a polynomial-time algorithm that, given an instance $(D, S, T, g, k)$ of Directed Planar Disjoint Paths and a flow $\phi$, either finds a solution of $(D, S, T, g, k)$ or determines that there is no solution of $(D, S, T, g, k)$ such that the flow associated with it and $\phi$ are homologous. + +We need a slightly more general version of this proposition because we will work with an +instance of Directed Planar Disjoint Paths where $D$ contains some “fake” edges that emerge when +we consider the radial completion of the input graph—the edges added to the graph in order to +attain its radial completion are considered to be fake. + +**Corollary 3.1.** *There exists a polynomial-time algorithm that, given an instance (*$D*, S*, T*, g*, k$*) of Directed Planar Disjoint Paths, a flow $\phi$ and a subset $X \subseteq A(D)$, either finds a solution of ($D - X, S*, T*, g*, k$) or determines that there is no solution of ($D - X, S*, T*, g*, k$) such that the flow associated with it and $\phi$ are homologous.*⁸ + +*Proof.* Given $(D,S,T,g,k)$, $\phi$ and $X$, the algorithm constructs an equivalent instance $(D',S,T,g,k)$ of DIRECTED PLANAR DISJOINT PATHS and a flow $\phi'$ as follows. Each arc $(u,v) \in X$ is replaced by a new vertex $w$ and two new arcs (whose drawing coincides with the former drawing of $(u,v)$), $(u,w)$ and $(v,w)$, and we define $\phi'(u,w) = \phi(u,v)$ and $\phi'(v,w) = \phi(u,v)^{-1}$. For all other arcs $a \in A(D) \cap A(D')$, $\phi'(a) = \phi(a)$. It is immediate to verify that $\phi'$ is a flow in $D'$, and that $(D',S,T,g,k)$ admits a solution that is homologous to $\phi'$ if and only if $(D,S,T,g,k)$ admits a solution that is disjoint from $X$ and homologous to $\phi$. Indeed, any solution of one of these instances is also a solution of the other one. Now, we apply Proposition 3.3 to either obtain a solution to $(D',S,T,g,k)$ homologous to $\phi'$ or conclude that no such solution exists. □ + +# 4 Preprocessing to Obtain a Good Instance + +We denote an instance of **Planar Disjoint Paths** similarly to an instance of **Directed Planar Disjoint Paths** (in Section 3) except that now the graph is denoted by **G** rather than **D** to stress the fact that it is undirected. Formally, an instance of **Planar Disjoint Paths** is a tuple (**G**, *S*, *T*, *g*, *k*) where **G** is a plane graph, *S*, *T* ⊆ *V*(*G*), *k* = |*S*| and *g*: *S* → *T* is bijective. Moreover, we say that (*G*, *S*, *T*, *g*, *k*) is *nice* if every vertex in *S* ∪ *T* has degree 1 and *S* ∩ *T* = ∅. The vertices in *S* ∪ *T* are called *terminals*. Let *H**G* be the radial completion of *G*. We choose a plane embedding of *H**G* so that one of the terminals, *t** ∈ *T*, will lie on the outer face.9 A *solution* of an instance (G, S, T, g, k) of **Planar Disjoint Paths** is a set P of pairwise vertex-disjoint paths in G that contains, for every vertex s ∈ S, a path with endpoints s and g(s). + +The following proposition eliminates all long sequences of $S \cup T$-free concentric cycles. + +**Proposition 4.1** ([2]). There exists a 2O(k)n2-time algorithm that, given an instance (G, S, T, g, k) of Planar Disjoint Paths, outputs an equivalent instance (G', S, T, g, k) of Planar Disjoint Paths where G' is a subgraph of G that has no sequence of S ∪ T-free concentric cycles whose length is larger than 2ck for some fixed constant c ≥ 1. + +⁸Note that $\phi$ and homology concern $D$ rather than $D-X$. +⁹This can be ensured by starting with an embedding of $H_G$ on a sphere, picking some face where $t^*$ lies as the outer face, and then projecting the embedding onto the plane. \ No newline at end of file diff --git a/samples/texts/1754951/page_80.md b/samples/texts/1754951/page_80.md new file mode 100644 index 0000000000000000000000000000000000000000..77cb08b327896d52013b9a2702efaa1ebae2655a --- /dev/null +++ b/samples/texts/1754951/page_80.md @@ -0,0 +1,31 @@ +[30] M. R. KRAMER AND J. VAN LEEUWEN, *The complexity of wirerouting and finding minimum area layouts for arbitrary VLSI circuits*, Advances in computing research, 2 (1984), pp. 129–146. + +[31] R. J. LIPTON AND K. W. REGAN, *David Johnson: Galactic Algorithms*, Springer Berlin Heidelberg, Berlin, Heidelberg, 2013, pp. 109–112. + +[32] D. LOKSHTANOV, D. MARX, AND S. SAURABH, *Slightly superexponential parameterized problems*, SIAM J. Comput., 47 (2018), pp. 675–702. + +[33] J. F. LYNCH, *The equivalence of theorem proving and the interconnection problem*, ACM SIGDA Newsletter, 5 (1975), pp. 31–36. + +[34] B. MOHAR AND C. THOMASSEN, *Graphs on Surfaces*, Johns Hopkins series in the mathematical sciences, Johns Hopkins University Press, 2001. + +[35] Y. PERL AND Y. SHILOACH, *Finding two disjoint paths between two pairs of vertices in a graph*, J. ACM, 25 (1978), pp. 1–9. + +[36] M. PILIPCZUK, M. PILIPCZUK, P. SANKOWSKI, AND E. J. VAN LEEUWEN, *Network sparsification for steiner problems on planar and bounded-genus graphs*, ACM Trans. Algorithms, 14 (2018), pp. 53:1–53:73. + +[37] N. ROBERTSON AND P. D. SEYMOUR, *Graph minors*.xiii. the disjoint paths problem, J. Comb. Theory, Ser. B, 63 (1995), pp. 65–110. + +[38] ——, *Graph minors*. XVI. excluding a non-planar graph, J. Comb. Theory, Ser. B, 89 (2003), pp. 43–76. + +[39] ——, *Graph minors*. XXII. irrelevant vertices in linkage problems, J. Comb. Theory, Ser. B, 102 (2012), pp. 530–563. + +[40] N. ROBERTSON, P. D. SEYMOUR, AND R. THOMAS, *Quickly excluding a planar graph*, J. Comb. Theory, Ser. B, 62 (1994), pp. 323–348. + +[41] P. SCHEFFLER, *A practical linear time algorithm for disjoint paths in graphs with bounded tree-width*, TU, Fachbereich 3, 1994. + +[42] A. SCHRIJVER, *Finding k disjoint paths in a directed planar graph*, SIAM J. Comput., 23 (1994), pp. 780–788. + +[43] A. SCHRIJVER, *Combinatorial optimization: polyhedra and efficiency*, vol. 24, Springer Science & Business Media, 2003. + +# A Properties of Winding Number + +In this section we sketch a proof of Proposition 7.1 using homotopy. Towards this, we introduce some notation that are extensions of the terms introduced Section 7 in the continuous setting. Recall that we have a plane graph Ring($I_{in}$, $I_{out}$) and we are interested in the winding number of paths in this graph, where $I_{in}$ and $I_{out}$ are two cycles such that $I_{out}$ is the outer-face, and there are no vertices or edges in the interior of $I_{in}$. Let us denote the (closed) curves defined by these two curves by $\rho_{out}$ and $\rho_{in}$, respectively. Then consider the collection of all the points in the plane that lie in the exterior of $\rho_{in}$ and interior of $\rho_{out}$. The closure of this set of points defines a surface called a *ring*, which we denote by Ring($\rho_{in}, \rho_{out}$) by abusing notation. Observe that the graph Ring($I_{in}, I_{out}$) is embedded in this ring, where the vertices of $I_{in}$ and $I_{out}$ lie on $\rho_{in}$ and $\rho_{out}$ respectively. \ No newline at end of file diff --git a/samples/texts/1754951/page_81.md b/samples/texts/1754951/page_81.md new file mode 100644 index 0000000000000000000000000000000000000000..4efd87b2d57fb4997bcc77b8e57ac0df827cacfe --- /dev/null +++ b/samples/texts/1754951/page_81.md @@ -0,0 +1,17 @@ +A curve $\alpha$ in the $\text{Ring}(\rho_{\text{in}}, \rho_{\text{out}})$ traverses it if it has one endpoint in $\rho_{\text{in}}$ and the other in $\rho_{\text{out}}$. We then orient this curve from its endpoint in $\rho_{\text{in}}$ to its endpoint $\rho_{\text{out}}$. A curve $\beta$ visits $\text{Ring}(\rho_{\text{in}}, \rho_{\text{out}})$ if both its endpoints line on either $\rho_{\text{in}}$ or $\rho_{\text{out}}$. In this case we orient this curve as follows. We first fix an arbitrary ordering of all points in the curve $\rho_{\text{in}}$ and another one for all the points in the curve $\rho_{\text{out}}$. We then orient $\beta$ from the smaller endpoint to the greater one. + +Consider two curves $\alpha, \alpha'$ that are either traversing or visiting $\text{Ring}(\rho_{\text{in}}, \rho_{\text{out}})$ are *homotopic* if there exists a homotopy of the ring that fixes $\rho_{\text{in}}$ and $\rho_{\text{out}}$ and transforms $\alpha$ to $\alpha'$. Note that homotopic curves have same endpoints. Two curves $\beta, \beta'$ are *transversally intersecting* if $\beta \cap \beta'$ is a finite collection of points. Let us remark that the above orientation is preserved under homotopy, since any two homotopic curves have the same endpoints. Furthermore, when we speak of oriented curves in a ring, it is implicit that such curves are either visitors or traversing the ring. Now we are ready to define the winding number of oriented curves in a ring. + +**Definition A.1 (Winding Number of Transversally Intersecting Curves).** Two curves $\alpha, \beta$ in $\text{Ring}(\rho_{\text{in}}, \rho_{\text{out}})$ intersect transversally if $\alpha \cap \beta$ is a finite set of points. For two curves $\alpha$ and $\beta$ in $\text{Ring}(\rho_{\text{in}}, \rho_{\text{out}})$ that intersect transversally, we define the winding number $\overline{\text{WindNum}}(\alpha, \beta)$ as the signed number of traversings of $\beta$ along $\alpha$. That is, for every intersection point of $\alpha$ and $\beta$ we record +1 if $\beta$ crosses $\alpha$ from left to right, and -1 if it crosses from right to left (of course, with respect to the chosen direction of traversing $\beta$) and 0 if it does not cross at that point. The winding number $\overline{\text{WindNum}}(\alpha, \beta)$ is the sum of the recorded numbers. + +It can be easily observed that if $\alpha$ and $\alpha'$ are homotopic curves traversing $\text{Ring}(\rho_{\text{in}}, \rho_{\text{out}})$ and both intersect $\beta$ transversally, then $\overline{\text{WindNum}}(\alpha, \beta) = \overline{\text{WindNum}}(\alpha', \beta')$. Here we rely on the fact that two homotopic curves have the same end-points. Observe that every intersection point of the curves $\alpha'$ and $\beta'$ is a traversing point, i.e. the point assigned either +1 or -1. Therefore, we can extend the notion of the winding number to pairs of curves not necessarily intersecting transversally as follows. + +**Definition A.2 (Winding Number).** If $\alpha, \beta$ are two curves in $\text{Ring}(\rho_{\text{in}}, \rho_{\text{out}})$, then we define $\overline{\text{WindNum}}(\alpha, \beta)$ to be the winding number $\overline{\text{WindNum}}(\alpha', \beta')$ for any $\alpha', \beta'$ such that $\alpha$ and $\alpha'$ are homotopic, $\beta$ and $\beta'$ are homotopic, and $\alpha', \beta'$ intersect transversally, and each common point of $\alpha'$ and $\beta'$ is a traversing point. + +Note that such $\alpha', \beta'$ always exist and the definition of $\overline{\text{WindNum}}(\alpha, \beta)$ does not depend on the particular curves. Let us now proceed towards a proof of Proposition 7.1. + +**Lemma A.1.** Suppose $\alpha, \beta, \gamma$ are curves traversing a ring $\text{Ring}(\rho_{\text{in}}, \rho_{\text{out}})$ that pairwise intersect transversally. Further, letting $a, b, c$ and $a', b', c'$ be the endpoints of $\alpha, \beta, \gamma$ on $\rho_{\text{in}}$ and $\rho_{\text{out}}$ respectively, suppose that $a, b, c$ are different and appear in the clockwise order on $\rho_{\text{in}}$, and that $a', b', c'$ are different and appear in the clockwise order on $\rho_{\text{out}}$. Then + +$$ \overline{\text{WindNum}}(\alpha, \beta) + \overline{\text{WindNum}}(\beta, \gamma) = \overline{\text{WindNum}}(\alpha, \gamma). $$ + +*Proof.* First, we argue that we may assume that $\overline{\text{WindNum}}(\alpha, \gamma) = 0$. This can be done as follows. Let $k = \overline{\text{WindNum}}(\alpha, \gamma)$. Glue a ring $\text{Ring}(\rho_{\text{out}}, \rho'_{\text{out}})$ to $\text{Ring}(\rho_{\text{in}}, \rho_{\text{out}})$ along $\rho_{\text{out}}$, for some non-self-traversing closed curve $\rho'_{\text{out}}$ that encloses $\rho_{\text{out}}$, thus obtaining ring $\text{Ring}(\rho_{\text{in}}, \rho'_{\text{out}})$. Pick $a'', b'', c''$ in the clockwise order on $\rho'_{\text{out}}$. Extend $\alpha$ to a curve $\alpha'$ traversing $\text{Ring}(\rho_{\text{in}}, \rho'_{\text{out}})$ using any curve within $\text{Ring}(\rho_{\text{out}}, \rho'_{\text{out}})$ connecting $a'$ with $a''$. Next, extend $\beta$ to $\beta'$ in the same way, but choose the extending segment so that it does not cross $\alpha'$. Finally, extend $\gamma$ to $\gamma'$ in the same way, but choose the extending segment so that it crosses $\alpha'$ (and thus also $\beta'$) exactly $-k$ times (where we count signed traversings). Thus we have $\overline{\text{WindNum}}(\alpha', \gamma') = 0$ and if we replace $\alpha, \beta, \gamma$ with $\alpha', \beta', \gamma'$, both sides of the postulated equality are decremented by $k$. Hence, it suffices to prove this equality for $\alpha', \beta', \gamma'$, for which we known that $\overline{\text{WindNum}}(\alpha', \gamma') = 0$. \ No newline at end of file diff --git a/samples/texts/1754951/page_82.md b/samples/texts/1754951/page_82.md new file mode 100644 index 0000000000000000000000000000000000000000..aefcb81386718fe9810017959b3f1f1607b896f4 --- /dev/null +++ b/samples/texts/1754951/page_82.md @@ -0,0 +1,17 @@ +Having assumed that $\overline{\text{WindNum}}(\alpha, \gamma) = 0$, it remains to prove that $\overline{\text{WindNum}}(\alpha, \beta) + \overline{\text{WindNum}}(\beta, \gamma) = 0$, or equivalently + +$$ \overline{\text{WindNum}}(\alpha, \beta) + \overline{\text{WindNum}}(\gamma^{-1}, \beta) = 0. \qquad (1) $$ + +Since $\overline{\text{WindNum}}(\alpha, \gamma) = 0$, we may further replace $\alpha$ and $\gamma$ with homotopic curves that do not cross at all. Note that this does not change the winding numbers in the postulated equality. Hence, from now on we assume that $\alpha$ and $\gamma$ are disjoint. + +Let us connect $c$ with $a$ using an arbitrary curve $\epsilon$ through the interior of the disk enclosed by $\rho_{\text{in}}$, and let us connect $a'$ with $c'$ using an arbitrary curve $\epsilon'$ outside of the disk enclosed by $\rho_{\text{out}}$. Thus, the concatenation of $\alpha, \epsilon', \gamma^{-1}, \epsilon$ is a closed curve in the plane without self-traversings; call it $\delta$. Then $\delta$ separates the plane into two regions $R_1, R_2$. Since $a, b, c$ appear in the same order on $\rho_{\text{in}}$ as $a', b', c'$ on $\rho_{\text{out}}$, it follows that $b$ and $b'$ are in the same region, say $R_1$. + +Now consider travelling along $\beta$ from the endpoint $b$ to the endpoint $b'$. Every traversing of $\alpha$ or $\gamma$ along $\beta$ is actually a traversing of $\delta$ that contributes to the left hand side of (1) with +1 if on the traversing $\beta$ passes from $R_1$ to $R_2$, and with -1 if it passes from $R_2$ to $R_1$. Since $\beta$ starts and ends in $R_1$, the total sum of those contributions has to be equal to 0, which proves (1). $\square$ + +**Lemma A.2.** For any curves $\alpha, \beta, \gamma$ traversing a ring $\text{Ring}(\rho_{\text{in}}, \rho_{\text{out}})$, it holds that + +$$ |(\overline{\text{WindNum}}(\alpha, \beta) + \overline{\text{WindNum}}(\beta, \gamma)) - \overline{\text{WindNum}}(\alpha, \gamma)| \le 1. $$ + +*Proof.* By slightly perturbing the curves using homotopies, we may assume that they pair-wise intersect transversally. Further, we modify the curves in the close neighborhoods of $\rho_{\text{in}}$ and $\rho_{\text{out}}$ so that we may assume that the endpoints of $\alpha, \beta, \gamma$ on $\rho_{\text{in}}$ and $\rho_{\text{out}}$ are pair-wise different and appear in the same clockwise order on both cycles; for the latter property, we may add one traversing between two of the curves, thus modifying one of the numbers $\overline{\text{WindNum}}(\alpha, \beta), \overline{\text{WindNum}}(\beta, \gamma), \overline{\text{WindNum}}(\alpha, \gamma)$ by one. It now remains to use Lemma A.1. $\square$ + +**Proof of Proposition 7.1.** Recall that the graph $\text{Ring}(I_{\text{in}}, I_{\text{out}})$ is embedded in the ring. The first property of follows directly from the definition of winding numbers. For the second property, we apply Lemma A.2 to the curves defined by the paths $\alpha, \beta$ and $\gamma$ in $\text{Ring}(\rho_{\text{in}}, \rho_{\text{out}})$. $\square$ \ No newline at end of file diff --git a/samples/texts/1754951/page_9.md b/samples/texts/1754951/page_9.md new file mode 100644 index 0000000000000000000000000000000000000000..5126fb6bf0074ba93d44fbd8b8797f8aa0dd4412 --- /dev/null +++ b/samples/texts/1754951/page_9.md @@ -0,0 +1,15 @@ +Additionally, the following proposition reduces the treewidth of $G$. In fact, Proposition 4.1 was given by Adler et al. [2] as a step towards the proof of the following proposition. However, while the absence of a long sequence of concentric cycles implies that the treewidth is small, the reversed statement is not correct (i.e. small treewidth does not imply the absence of a long sequence of concentric cycles). Having small treewidth is required but not sufficient for our arguments, thus we cite both propositions. + +**Proposition 4.2** (Lemma 10 in [2]). There exists a $2^{\mathcal{O}(k)}n^2$-time algorithm that, given an instance $(G, S, T, g, k)$ of Planar Disjoint Paths, outputs an equivalent instance $(G', S, T, g, k)$ of Planar Disjoint Paths where $G'$ is a subgraph of $G$ whose treewidth is upper bounded by $2^{ck}$ for some fixed constant $c \ge 1$.¹⁰ + +The purpose of this section is to transform an arbitrary instance of Planar Disjoint Paths into a so called “good” instance, defined as follows. + +**Definition 4.1.** An instance $(G, S, T, g, k)$ of Planar Disjoint Paths is good if it is nice, at least one terminal $t^* \in T$ belongs to the outer faces of both $G$ and its radial completion $H_G$, the treewidth of $G$ is upper bounded by $2^{ck}$, and $G$ has no $S \cup T$-free sequence of concentric cycles whose length is larger than $2^{ck}$. Here, $c \ge 1$ is the fixed constant equal to the maximum among the fixed constants in Propositions 4.1 and 4.2. + +Towards this transformation, note that given an instance $(G, S, T, g, k)$ of Planar Disjoint Paths and a terminal $v \in S$, we can add to $G$ a degree-1 vertex $u$ adjacent to $v$, and replace $v$ by $u$ in $S$ and in the domain of $g$. This operation results in an equivalent instance of Planar Disjoint Paths. Furthermore, it does not increase the treewidth of $G$ (unless the treewidth of $G$ is 0, in which it increases to be 1). The symmetric operation can be done for any terminal $v \in T$. By repeatedly applying these operations, we can easily transform $(G, S, T, g, k)$ to an equivalent nice instance. Moreover, the requirement that at least one terminal $t^* \in T$ belongs to the outer faces of both $G$ and its radial completion can be made without loss of generality by drawing $G$ appropriately in the first place. Thus, we obtain the following corollary of Propositions 4.1 and 4.2. Note that $k$ remains unchanged. + +**Corollary 4.1.** There exists a $2^{\mathcal{O}(k)}n^2$-time algorithm that, given an instance $(G, S, T, g, k)$ of Planar Disjoint Paths, outputs an equivalent good instance $(G', S', T', g', k)$ of Planar Disjoint Paths where $|V(G')| = \mathcal{O}(|V(G)|)$. + +We remark that our algorithm, presented in Section 10, will begin by applying Corollary 4.1. To simplify arguments ahead, it will be convenient to suppose that every edge in $H_G$ has $4|V(G)| + 1 = 4n + 1$ parallel copies. Thus, we slightly abuse the notation $H_G$ and use it to refer to $H_G$ enriched with such a number of parallel copies of each edge. For a pair of adjacent vertices $u, v \in V(H_G)$, we will denote the $4n + 1$ parallel copies of edges between them by $e_{-2n}, e_{-2n+1}, \dots, e_{-1}, e_0, e_1, e_2, \dots, e_{2n}$ where $e = \{u, v\}$, such that when the edges incident to $u$ (or $v$) are enumerated in cyclic order, the occurrences of $e_i$ and $e_{i+1}$ are consecutive for every $i \in \{-2n, -2n+1, \dots, 2n-1\}$, and $e_{-2n}$ and $e_{2n}$ are the outermost copies of $e$. Thus, for every $i \in \{-2n+1, -2n+2, \dots, 2n-1\}$, $e_i$ lies on the boundary of exactly two faces: the face bounded by $e_{i-1}$ and $e_i$, and the face bounded by $e_i$ and $e_{i+1}$. When the specification of the precise copy under consideration is immaterial, we sometimes simply use the notation $e$. + +¹⁰While the running time in Lemma 10 in [2] is stated to be $2^{2^{\mathcal{O}(k)}} n^2$, the proof is easily seen to imply the bound $2^{\mathcal{O}(k)} n^2$ in our statement. The reason why Adler et al. [2] obtain a double-exponential dependence on $k$ when they solve *Planar Disjoint Paths* is *not* due to the computation that attains a tree decomposition of width $\text{tw} = 2^{\mathcal{O}(k)}$, but it is because that upon having such a tree decomposition, they solve the problem in time $2^{\mathcal{O}(\text{tw})} n = 2^{2^{\mathcal{O}(k)}} n$. \ No newline at end of file diff --git a/samples/texts/2274792/page_1.md b/samples/texts/2274792/page_1.md new file mode 100644 index 0000000000000000000000000000000000000000..90d7e5c98aca2322a0c336dbf1759beeff5f36d4 --- /dev/null +++ b/samples/texts/2274792/page_1.md @@ -0,0 +1,18 @@ +Supplemental material for: + +Mothers teach daughters because daughters teach +granddaughters: the evolution of sex-biased transmission + +Matthew R. Zefferman + +# Appendix A Stability of the monomorphic equilibria in the uniparental teaching model + +There are six monomorphic equilibria in the uniparental teaching model in the absence of innovation. These can be found by fixing each allele in turn and determining if cultural transmission loses or fixes the cultural in both sexes. The monomorphic equilibria are as follows: + +When $b > 2t$ there is a monomorphic equilibrium where the $A_1$ allele is fixed and the cultural trait is fixed in both sexes. + +When $b > t$ there is a monomorphic equilibrium where the $A_2$ allele is fixed and the cultural trait is fixed in females but lost in males. + +When any of the alleles ($A_1$, $A_2$, $A_3$, or $A_4$) are fixed, there is a monomorphic equilibrium where the cultural trait is lost in both sexes. + +To determine the stability of each equilibrium, I calculated the characteristic equation for the system given by Equations 2 and 3 evaluated at the equilibrium. I then evaluated the eigenvalues numerically, where required, by substituting the values for $b$, and $t$ over the ranges explored in the main text. The results of this analysis are described below and shown in Figure A.1. Figure A.1 shows a close correspondence between the results of the stability analysis and the numerical simulation from the main text. \ No newline at end of file diff --git a/samples/texts/2274792/page_10.md b/samples/texts/2274792/page_10.md new file mode 100644 index 0000000000000000000000000000000000000000..e2b92517d97cbd49c7e28a7163098b993b5a4816 --- /dev/null +++ b/samples/texts/2274792/page_10.md @@ -0,0 +1 @@ +Figure C.3: With a transmission error rate of $\epsilon = 0.1$ the disparate benefits model has similar results to the version in the main text. There is a range where the reproductive cost of teaching is about half the reproductive benefit of the trait where the trait persists at a low frequency (A), females are taught at a high rate (B), and males are taught at a much lower rate (C). However, the trait frequency is low over a greater range than in the model in the main text. \ No newline at end of file diff --git a/samples/texts/2274792/page_11.md b/samples/texts/2274792/page_11.md new file mode 100644 index 0000000000000000000000000000000000000000..545576ea5e2e99096c26af5197172d8618720e46 --- /dev/null +++ b/samples/texts/2274792/page_11.md @@ -0,0 +1,5 @@ +# Appendix D Power Analysis of Sponging Success + +There is some debate over whether sponging provides a reproductive benefit to spongers. The evidence is mixed. Mann et al. 2008 observed that sponging bottlenose dolphins had 18% greater calving success than nonspongers, though due to small sample size this finding was not statistically significant (at *p* = 0.05). To examine this issue further, I conducted a power analysis to determine how small of a reproductive benefit could be reliably detected by Mann et al. 2008. The power analysis, shown in the R code below, indicates that the smallest difference in calving success that could be reliably detected (at 1 – *β* = 0.8) was a 43% (one-sided) or 48% (two-sided) greater calving success for spongers over nonspongers. Importantly, this minimally detectable reproductive advantage is much larger than is required by my models. In fact, it is well above even the largest reproductive advantage (25%) that I considered in my simulations and it is also much larger than the reproductive advantage (18%) found in the empirical data. + +From Mann et al. 2008, the number of non-spongers in the power analysis is 116 with a mean calving success of 0.132 and a standard error of 0.008. The number of spongers is 16 with a mean calving success of 0.15 with a standard error of 0.018. \ No newline at end of file diff --git a/samples/texts/2274792/page_12.md b/samples/texts/2274792/page_12.md new file mode 100644 index 0000000000000000000000000000000000000000..b0efaeab0d5f54b78145d7bf43b9f9080db417e4 --- /dev/null +++ b/samples/texts/2274792/page_12.md @@ -0,0 +1 @@ +Figure A.1: There is close correspondence between the stability analysis and the numeric simulations from the main text. There is one stable monomorphic equilibrium, which occurs over the range shown in A. In this equilibrium the $A_1$ allele is fixed and the cultural trait is fixed in both sexes. This range corresponds to the results of the numeric simulation where the $A_1$ allele and the cultural trait goes to fixation as shown in B and C. \ No newline at end of file diff --git a/samples/texts/2274792/page_13.md b/samples/texts/2274792/page_13.md new file mode 100644 index 0000000000000000000000000000000000000000..0cd65c953d4859b5ec2f878ff7d95c65f10075f2 --- /dev/null +++ b/samples/texts/2274792/page_13.md @@ -0,0 +1,47 @@ +`install.packages("pwr")` + +`library(pwr)` + +#### INPUT THE SAMPLE SIZES, MEAN CALVING SUCCESS AND STANDARD ERRORS FROM MANN, ET AL. 2008. + +``` +# Non-spongers: + +n_nonspongers=116 +mean_CS_nonspongers=0.132 +SE_CS_nonspongers=0.008 + +# Spongers: + +n_spongers=16 +mean_CS_spongers=0.156 +SE_CS_spongers=0.018 + +#### CALCULATE DIFFERENCE IN MEASURED CALVING SUCCESS BETWEEN SPONGERS AND NONSPONGERS + +percent_increase=(mean_CS_spongers - mean_CS_nonspongers)/mean_CS_nonspongers + +print(paste("Percent Increase in Sponger Calving Success: ", 100*round(percent_increase, digits=2), "%")) +``` + +#### CONVERT STANDARD ERRORS TO STANDARD DEVIATIONS + +``` +SD_CS_nonspongers=SE_CS_nonspongers*sqrt(n_nonspongers) +SD_CS_spongers=SE_CS_spongers*sqrt(n_spongers) +``` + +#### CALCULATE TERMS FOR POWER ANALYSIS + +``` +SS_within=(n_nonspongers-1)*SD_CS_nonspongers^2 + (n_spongers-1)*SD_CS_spongers^2 +df_within=n_nonspongers + n_spongers - 2 +SD_within=sqrt(SS_within/df_within) +``` + +#### PERFORM POWER ANALYSIS FOR A TWO-SIDED TEST + +``` +## RUN POWER ANALYSIS +power_test_results=pwr.t2n.test(n1=n_nonspongers, n2=n_spongers, d=NULL, sig.level = 0.05, power = 0.8, +``` \ No newline at end of file diff --git a/samples/texts/2274792/page_14.md b/samples/texts/2274792/page_14.md new file mode 100644 index 0000000000000000000000000000000000000000..cbe05fe420d5e69178aa5c9be3d68b89e6bd2f7f --- /dev/null +++ b/samples/texts/2274792/page_14.md @@ -0,0 +1,30 @@ +alternative = "two.sided") + +## CALCULATE DETECTABLE MEAN DIFFERENCE IN CALVING SUCCESS +**mean_diff** = SD_within * power_test_results$d + +## COMPARE DIFFERENCE IN MEANS TO NONSPONGER CALVING SUCCESS +percent_increase=mean_diff/mean_CS_nonspongers + +## RETURN RESULTS +print("TWO-SIDED_TEST:\") + +print(paste("Difference_inmeans: ", mean_diff, " _Percent_increase: ", 100*round(percent_increase, digits=2,"%))) + +### PERFORM POWER ANALYSIS FOR A ONE-SIDED TEST + +## RUN POWER ANALYSIS + +power_test_results=pwr.t2n.test(nl=n_nonspongers, n2=n_spongers, d=NULL, sig.level = 0.05, power = 0.8, + alternative = "greater") + +## CALCULATE DETECTABLE MEAN DIFFERENCE IN CALVING SUCCESS +**mean_diff** = SD_within * power_test_results$d + +## COMPARE DIFFERENCE IN MEANS TO NONSPONGER CALVING SUCCESS +percent_increase=mean_diff/mean_CS_nonspongers + +## RETURN RESULTS +print("ONE-SIDED_TEST:\") + +print(paste("Difference_inmeans: ", mean_diff, " _Percent_increase: ", 100*round(percent_increase, digits=2),"%")) \ No newline at end of file diff --git a/samples/texts/2274792/page_15.md b/samples/texts/2274792/page_15.md new file mode 100644 index 0000000000000000000000000000000000000000..bd75ef069db2578662ac1281c45a1bb634c53e40 --- /dev/null +++ b/samples/texts/2274792/page_15.md @@ -0,0 +1,26 @@ +## A.1 Stability of the equilibrium where the $A_1$ allele and cultural trait are fixed in both sexes + +To determine the stability of this equilibrium, I found the characteristic equation for the system given by Equations 2 and 3 evaluated where $x_{f11} = x_{m11} = 1$. + +$$ +\begin{aligned} +& \frac{\lambda^6}{128(1+b-2t)^6(1+b)^2} (2(1+b-2t)\lambda - 1) ((1+b-2t)\lambda - 1) \\ +& \quad (2(1+b-2t)\lambda - (1+b-2t)-1) (4(1+b)(1+b-2t)\lambda^2 - 2(1+b)(2+2b-3t)\lambda + b(1+b-t)) \\ +& \quad (8(1+b)(1+b-2t)\lambda^3 + 4(1+b)(1+b-2t)(2+b-2t)\lambda^2 - 2(1+b)(b-t)(1+b-2t)\lambda + \\ +& \qquad \qquad \qquad \qquad b(1+b-t)) = 0 +\end{aligned} +$$ + +I then evaluated the resulting eigenvalues numerically by substituting the values for $b$ and $t$ over the range explored in the main text. Figure A.1A shows the range of parameters where this equilibrium is stable. There is a close correspondence between this region and where the $A_1$ allele (Figure A.1B) and the cultural trait (Figure A.1C) go to fixation in the numerical simulations described in the main text. + +## A.2 Stability of $A_2$ equilibrium with culture in females + +To determine the stability of this equilibrium, I calculated the characteristic equation for the system given by Equations 2 and 3 evaluated where $x_{f21} = x_{m20} = 1$. + +$$ +\begin{aligned} +& \frac{\lambda^5}{256(1+b-t)^5} \Biggl(4(1+bt)\lambda^2 - 2(2+2b-3t)\lambda - b(1+b-t)\Biggr) \\ +& \quad \Biggl((4(1+b-t)\lambda^2 - 2(2+b-t)\lambda - b)\Biggr) \Biggl(2(1+b-t)\lambda - 1\Biggr) \Biggl((1+b-t)\lambda - 1\Biggr) \\ +& \quad \Biggl(8(1+b-t)\lambda^3 - 4(2+b-t)\lambda^2 - 2(b^2 - 2bt + 4b - 2t)\lambda + b\Biggr) = 0 +\end{aligned} +$$ \ No newline at end of file diff --git a/samples/texts/2274792/page_16.md b/samples/texts/2274792/page_16.md new file mode 100644 index 0000000000000000000000000000000000000000..b40d1cc5a773c0ad210d3bf8972e1b2043ece243 --- /dev/null +++ b/samples/texts/2274792/page_16.md @@ -0,0 +1,17 @@ +I then evaluated the resulting eigenvalues numerically by substituting the values for $b$ and $t$ over the range explored in the main text where $b > t$ (the conditions for the existence of the equilibrium). This equilibrium is not stable over this range, which corresponds to the results of the numerical simulations from the main text where the system never evolved to this equilibrium. + +## A.3 Stability of $A_1$ equilibrium without culture + +To determine the stability of this equilibrium, I calculated the characteristic equation for the system given by Equations 2 and 3 evaluated where $x_{f10} = x_{m10} = 1$. + +$$ \frac{\lambda^9(1-\lambda)^3(\lambda - (1+b-2t)) (2\lambda - (1+b-t))}{2} = 0 $$ + +Since three of the resulting eigenvalues are equal to one, the equilibrium is unstable or neutrally stable for all parameters. This makes intuitive sense since the fitness of all four alleles in both sexes are equivalent in the absence of the cultural trait, allowing for drift between the alleles. + +## A.4 Stability of $A_2$ equilibrium without culture + +To determine the stability of this equilibrium, I calculated the characteristic equation for the system given by Equations 2 and 3 evaluated where $x_{f20} = x_{m20} = 1$. + +$$ \frac{\lambda^9(1-\lambda)^3(\lambda - (1+b-t)) (2\lambda - (1+b-2t))}{2} = 0 $$ + +Since three of the resulting eigenvalues are equal to one, the equilibrium is unstable or neutrally stable for all parameters. This makes intuitive sense since the fitnesses of all four alleles in both sexes are equivalent in the absence of the cultural trait. \ No newline at end of file diff --git a/samples/texts/2274792/page_17.md b/samples/texts/2274792/page_17.md new file mode 100644 index 0000000000000000000000000000000000000000..ea2510aaee83201dd5d4664f7ad7e9a28b33efde --- /dev/null +++ b/samples/texts/2274792/page_17.md @@ -0,0 +1,15 @@ +## A.5 Stability of $A_3$ equilibrium with culture lost + +To determine the stability of this equilibrium, I calculated the characteristic equation for the system given by Equations 2 and 3 evaluated where $x_{f30} = x_{m30} = 1$. + +$$ \frac{\lambda^9(1-\lambda)^3(2\lambda-(1+b-t))(2\lambda-(1+b-2t))}{2} = 0 $$ + +Since three of the resulting eigenvalues are equal to one, the equilibrium is unstable or neutrally stable for all parameters. This makes intuitive sense since the fitnesses of all four alleles in both sexes are equivalent in the absence of the cultural trait. + +## A.6 Stability of $A_4$ equilibrium with culture lost + +To determine the stability of this equilibrium, I calculated the characteristic equation for the system given by Equations 2 and 3 evaluated where $x_{f40} = x_{m40} = 1$. + +$$ \frac{\lambda^9(1-\lambda)^3(2\lambda-(1+b-t))(2\lambda-(1+b-2t))}{2} = 0 $$ + +Since three of the resulting eigenvalues are equal to one, the equilibrium is unstable or neutrally stable for all parameters. This makes intuitive sense since the fitnesses of all four alleles in both sexes are equivalent in the absence of the cultural trait. \ No newline at end of file diff --git a/samples/texts/2274792/page_18.md b/samples/texts/2274792/page_18.md new file mode 100644 index 0000000000000000000000000000000000000000..d0991d8d357593505d45f79b1eb20fa6fdc08a4c --- /dev/null +++ b/samples/texts/2274792/page_18.md @@ -0,0 +1,13 @@ +# Appendix B Stability of the monomorphic equilibria in the disparate benefits model + +There are six monomorphic equilibria in the disparate benefits model in the absence of innovation. These can be found by fixing each allele in turn and determining if cultural transmission loses or fixes the cultural in both sexes. The monomorphic equilibria are as follows: + +When $b_f > \mu$ and $b_m > \mu$, there is an equilibrium where the $B_1$ allele is fixed and cultural trait is fixed in both sexes. + +When $b_f > \mu$, there is a monomorphic equilibrium where the $B_2$ allele is fixed and the cultural trait is fixed in females but lost in males. + +When any of the alleles ($B_1, B_2, B_3$, or $B_4$) are fixed, there is a monomorphic equilibrium where the cultural trait is lost in both sexes. + +To determine the stability of each equilibrium, I calculated the characteristic equation for the system given by Equations 7 and 8 evaluated at the equilibrium. I then evaluated the eigenvalues numerically, where required, by substituting the values for $b_f, b_m$ and $\mu$ over the range explored in the main text. The results of this analysis are described below and shown in Figure B.1. Figure B.1 shows a close correspondence between the results of the stability analysis and the numerical simulation from the main text. + +There is one stable monomorphic equilibrium, which occurs over the range shown in A. In this equilibrium the $A_1$ allele is fixed and the cultural trait is fixed in both sexes. This range corresponds to the results of the numeric simulation where the $A_1$ allele and the cultural trait goes to fixation as shown in B and C. \ No newline at end of file diff --git a/samples/texts/2274792/page_19.md b/samples/texts/2274792/page_19.md new file mode 100644 index 0000000000000000000000000000000000000000..787dff1874bd87234726a727c81f14be503d0fdf --- /dev/null +++ b/samples/texts/2274792/page_19.md @@ -0,0 +1 @@ +Figure B.1: There is close correspondence between the stability analysis and the numeric simulations from the main text. There are two stable monomorphic equilibria which occur over the ranges shown in A. In one of these equilibria the $B_2$ is fixed and the cultural trait is fixed in females, but absent in males. This corresponds to the results of the numeric simulation as shown in B and C. In the other equilibrium, the $B_1$ allele and cultural trait are fixed in both sexes. This also corresponds to the results of the numeric simulations as shown in D and E. \ No newline at end of file diff --git a/samples/texts/2274792/page_2.md b/samples/texts/2274792/page_2.md new file mode 100644 index 0000000000000000000000000000000000000000..18f03c4ea5c42a614c36ba18825d9f4022cf4498 --- /dev/null +++ b/samples/texts/2274792/page_2.md @@ -0,0 +1,11 @@ +$$ \frac{\lambda^9 (\lambda - 1)^3 (2\lambda - (1 + b_f - \mu))^2}{4} = 0 $$ + +Since three of these eigenvalues are equal to one, the equilibrium is unstable or neutrally stable for all parameters. This makes intuitive sense since the fitness of all four alleles in both sexes is equivalent in the absence of culture. + +## B.6 Stability of the $B_4$ equilibrium with culture lost + +To determine the stability of this equilibrium, I calculated the characteristic equation for the system given by Equations 7 and 8 evaluated where $y_{f40} = y_{m40} = 1$. + +$$ \frac{\lambda^9 (\lambda - 1)^3 (2\lambda - (1 + b_f - \mu))^2}{4} = 0 $$ + +Since three of these eigenvalues are equal to one, the equilibrium is unstable or neutrally stable for all parameters. This makes intuitive sense since the fitness of all four alleles in both sexes is equivalent in the absence of culture. \ No newline at end of file diff --git a/samples/texts/2274792/page_20.md b/samples/texts/2274792/page_20.md new file mode 100644 index 0000000000000000000000000000000000000000..3128fb340b1b9b1bf6dd7f805c8aa1730a4ac7d0 --- /dev/null +++ b/samples/texts/2274792/page_20.md @@ -0,0 +1,23 @@ +## B.1 Stability of the $B_1$ equilibrium with culture + +To determine the stability of this equilibrium, I calculated the characteristic equation for the system given by Equations 7 and 8 evaluated where $y_{f11} = y_{m11} = 1$. + +$$ \frac{\lambda^8}{32(1+b_m-\mu)^3(1+b_f-\mu)^4} \left(2(1+b_m-\mu)\lambda-1\right) \left((1+b_f-\mu)\lambda-1\right) $$ + +$$ \left(2(1+b_f-\mu)\lambda-1\right) \left(2(1+b_m-\mu)(1+b_f-\mu)\lambda - (2+b_m+b_f-2\mu)\right) $$ + +$$ \left(4(1+b_m-\mu)(1+b_f-\mu)\lambda^2 - 2(1+b_m-\mu)(2+b_f-\mu)\lambda + (b_m-\mu)\right) = 0 $$ + +Explicit expressions for the eigenvalues can be found, though the eigenvalues resulting from the cubic factor of the characteristic polynomial are difficult to evaluate symbolically. Therefore, I evaluated the eigenvalues numerically by substituting the values for $b_f - \mu$, and $b_m - \mu$ explored in the main text. Figure B.1 shows a close correspondence between the region where this equilibrium is stable and where $B_1$ and the cultural trait go to fixation in the numerical simulations in the main text. + +## B.2 Stability of the $B_2$ equilibrium with culture in females + +To determine the stability of this equilibrium, I calculated the characteristic equation for the system given by Equations 7 and 8 evaluated where $y_{f21} = y_{m20} = 1$. + +$$ \frac{\lambda^8}{32(1+b_f-\mu)^4} \left(2\lambda - (2+b_m-\mu)\right) \left((1+b_f-\mu)\lambda - 1\right) $$ + +$$ \left(-4(1+b_f-\mu)\lambda^2 + 2(1+(1+b_m-\mu)(1+b_f-\mu))\lambda - (b_m-\mu)\right) $$ + +$$ \left(2(1+b_f-\mu)\lambda - (2+b_f-\mu)\right) \left(2(1+b_f-\mu)\lambda - 1\right) = 0 $$ + +I evaluated the resulting eigenvalues numerically by substituting the values of $b_f - \mu$ and \ No newline at end of file diff --git a/samples/texts/2274792/page_21.md b/samples/texts/2274792/page_21.md new file mode 100644 index 0000000000000000000000000000000000000000..bc3ac5c703b621b557156140ceaa4b21ffa55310 --- /dev/null +++ b/samples/texts/2274792/page_21.md @@ -0,0 +1,21 @@ +$b_m - \mu$ explored in the main text. Figure B.1 shows a close correspondence between the region where this equilibrium is stable and where $B_2$ and the cultural trait go to fixation in the numerical simulations in the main text. + +### B.3 Stability of the $B_1$ equilibrium with culture lost + +To determine the stability of this equilibrium, I calculated the characteristic equation for the system given by Equations 7 and 8 evaluated where $y_{f10} = y_{m10} = 1$. + +$$ \frac{\lambda^9 (\lambda - 1)^3 (2\lambda - (1 + b_f - \mu)) (\lambda - (1 + b_f - \mu))}{2} = 0 $$ + +Since three of these eigenvalues are equal to one, the equilibrium is unstable or neutrally stable for all parameters. This makes intuitive sense since the fitness of all four alleles in both sexes is equivalent in the absence of culture. + +### B.4 Stability of the $B_2$ equilibrium with culture lost + +To determine the stability of this equilibrium, I calculated the characteristic equation for the system given by Equations 7 and 8 evaluated where $y_{f20} = y_{m20} = 1$. + +$$ \frac{\lambda^9 (\lambda - 1)^3 (2\lambda - (1 + b_f - \mu)) (\lambda - (1 + b_f - \mu))}{2} = 0 $$ + +Since three of these eigenvalues are equal to one, the equilibrium is unstable or neutrally stable for all parameters. This makes intuitive sense since the fitness of all four alleles in both sexes is equivalent in the absence of culture. + +### B.5 Stability of the $B_3$ equilibrium with culture lost + +To determine the stability of this equilibrium, I calculated the characteristic equation for the system given by Equations 7 and 8 evaluated where $y_{f30} = y_{m30} = 1$. \ No newline at end of file diff --git a/samples/texts/2274792/page_3.md b/samples/texts/2274792/page_3.md new file mode 100644 index 0000000000000000000000000000000000000000..d1b817643de25d44807b0149ad4fb4ca99b39564 --- /dev/null +++ b/samples/texts/2274792/page_3.md @@ -0,0 +1,13 @@ +# Appendix C Model Variations + +This appendix presents variations of the uniparental teaching and disparate benefits models. In the first variation, teaching rules for the uniparental teaching model are given by two loci (one for teaching sons and one for teaching daughters) each with two alleles instead of one locus with four alleles as in the main text. In the second variation of the uniparental teaching model, teaching has an error rate of 10%. Similarly, in a variation of the disparate benefits model, learning has an error rate of 10%. + +## C.1 Two Locus Model + +In the uniparental teaching model in the main text, individuals have one locus with four alleles. To check that my results are not dependent on the one locus assumption, here I describe an alternate model where individuals have two loci with two alleles each where one locus controls the teaching of sons and the other the teaching of daughters. Individuals are haploid with one of two sexes, male and female, and two genetic loci A and B. $A_1$ females teach daughters and $A_2$ females do not. $B_1$ females teach sons and $B_2$ females do not. Males do not teach. Cultural transmission rules for each combination of alleles are given in Table C.1. + +
AllelesDaughter RuleSon Rule
$A_1B_1$$\tau_{f11} = 1$$\tau_{m11} = 1$
$A_1B_2$$\tau_{f12} = 1$$\tau_{m22} = 0$
$A_2B_1$$\tau_{f21} = 0$$\tau_{m21} = 1$
$A_2B_2$$\tau_{f22} = 0$$\tau_{m22} = 0$
+ +Table C.1: Teaching rules for each allele where $\tau_{fjk} = 1$ indicates that $A_j B_k$ mothers teach daughters and $\tau_{mjk} = 1$ indicates that mothers teach sons. When $\tau_{fjk} = 0$ or $\tau_{mjk} = 0$, mothers with allele $A_j B_k$ do not teach that sex. These rules are used in Equations C.2 and C.3 + +An individual's fitness depends on whether it has learned the cultural trait and whether it teaches offspring. All individuals have a baseline fitness of 1. Teaching is costly, limiting \ No newline at end of file diff --git a/samples/texts/2274792/page_4.md b/samples/texts/2274792/page_4.md new file mode 100644 index 0000000000000000000000000000000000000000..ac497709aced2e74dd8e44f4dc510818400ae6f4 --- /dev/null +++ b/samples/texts/2274792/page_4.md @@ -0,0 +1,11 @@ +
AllelesFemale fitnessesMale fitnesses
with traitwithout traitwith traitwithout trait
A1B1wf111 = 1 + b - 2κwf110 = 1wm111 = 1 + bwm110 = 1
A1B2wf121 = 1 + b - κwf120 = 1wm121 = 1 + bwm120 = 1
A2B1wf211 = 1 + b - κwf210 = 1wm211 = 1 + bwm210 = 1
A2B2wf221 = 1 + bwf220 = 1wm221 = 1 + bwm220 = 1
+ +Table C.2: Expected fitnesses for encultured and unencultured individuals of each allele and sex. $w_{sjk1}$ is the expected fitness for an encultured individual of sex s and allele $A_j B_k$. $w_{sjk0}$ is the expected fitness for an unencultured individual. b is the benefit of the cultural trait and $\kappa$ is the cost of teaching offspring of one sex. + +a mother's future reproductive potential after her first offspring. In the model, females who teach offspring of one sex pay a reproductive cost of $\kappa$ and females who teach offspring of both sexes pay a reproductive cost of $2\kappa$. Individual of both sexes who learn the trait gain a reproductive benefit of $b$. Table C.2 gives the expected fitnesses for individuals where $w_{sjk1}$ is the fitness for an individual of sex s and allele $A_j B_k$ that has learned cultural trait and $w_{sk0}$ is the fitness of an individual who has not learned the cultural trait. + +As in the main text, reproductive competition occurs in both males and females with each individual's contribution to the next generation determined by its relative fitness with its sex, as described in Equation C.1. The frequency of individuals of type $A_j B_k$ in sex s is $x_{sjkc}$, where $c=1$ if the individual knows the cultural trait and $c=0$ if the individual does not. $x'_{sjkc}$ is the type frequency after genetic selection and $x''_{sjkc}$ as the frequency after cultural transmission and innovation. + +$$x'_{sjkc} = \frac{x_{sjkc}w_{sjkc}}{\sum_g \sum_h \sum_i x_{sghi}w_{sghi}} \quad (C.1)$$ + +An individual only acquires a cultural trait through teaching if its mother has both an allele for teaching offspring of the individual's sex and the cultural trait. This is reflected in Equation C.2, which gives the frequency of individuals of sex s who have allele $A_j B_k$ and the cultural trait after cultural transmission. \ No newline at end of file diff --git a/samples/texts/2274792/page_5.md b/samples/texts/2274792/page_5.md new file mode 100644 index 0000000000000000000000000000000000000000..a8d6b14ea65579263d9a3004565f0b368a853071 --- /dev/null +++ b/samples/texts/2274792/page_5.md @@ -0,0 +1,13 @@ +$$x''_{sjk1} = \frac{1}{2} \left( \tau_{sjk} x'_{fjk1} + \sum_a \sum_b x'_{makb} \sum_d \tau_{sjd} x'_{fjd1} + \sum_g \sum_h x'_{mjgh} \sum_m \tau_{smk} x'_{fmk1} \right. \\ \left. + \sum_n x'_{njkn} \sum_p \sum_q \tau_{spq} x'_{fpq1} \right) \quad (C.2)$$ + +$$\begin{aligned} x''_{sjk0} = \frac{1}{2} & (x'_{fjk0} + (1-\tau_{sjk})x'_{fjk1} + \sum_a \sum_b x'_{makb} \sum_d ((1-\tau_{sjd})x'_{fjd1} + x'_{fjd0}) \\ & + \sum_g \sum_h x'_{mjgh} \sum_m ((1-\tau_{smk})x'_{fmk1} + x'_{fmk0}) \\ & + \sum_n x'_{njk} \sum_p \sum_q ((1-\tau_{spq})x'_{fpq1} + x'_{fpq0})) \end{aligned} \quad (C.3)$$ + +Individuals innovate a trait with a probability of $\nu$, increasing the fraction of encultured individuals, Equation C.4, and decreasing the fraction of unencultured individuals, Equation C.5. + +$$x'''_{sk1} = x''_{sjk1} + \nu x''_{sjk0} \quad (C.4)$$ + +$$x'''_{sjk0} = x''_{sjk0}(1 - \nu) \quad (C.5)$$ + +As in the main text, I ran numeric simulations of this system for various combinations of teaching costs, $t$, and reproductive benefits of the cultural trait, $b$ and an innovation rate of $r = 0.005$. As in the main text, for each parameter combination, I ran the simulation starting with four initial allele frequencies. In each frequency one allele started at 5% of the population and the rest were evenly distributed among the rest of the population. I started the frequency of the cultural trait at zero in all four initial conditions and ran the simulations until they either converged to a shared equilibrium or ran for $10^7$ generations. + +The results of the simulation are shown in Figure C.1 \ No newline at end of file diff --git a/samples/texts/2274792/page_6.md b/samples/texts/2274792/page_6.md new file mode 100644 index 0000000000000000000000000000000000000000..c9e58577072ba4a2eb6d572c8bff105eafb24b96 --- /dev/null +++ b/samples/texts/2274792/page_6.md @@ -0,0 +1 @@ +Figure C.1: The Two Loci version of the uniparental transmission model has similar results as the One Loci version. There is a range where the reproductive cost of teaching is about half the reproductive benefit of the trait where the trait persists at a low frequency (A), females are taught at a high rate (B), and males are taught at a much lower rate (C). \ No newline at end of file diff --git a/samples/texts/2274792/page_7.md b/samples/texts/2274792/page_7.md new file mode 100644 index 0000000000000000000000000000000000000000..93c381eae9af0a2616bf6546db6268e87047cbd1 --- /dev/null +++ b/samples/texts/2274792/page_7.md @@ -0,0 +1,12 @@ +## C.2 Uniparental Teaching with Transmission Errors + +This model is identical to the uniparental teaching model in the main text except that +teaching has an error rate, $\epsilon$, which is the frequency with which a mother pays a teaching +cost, but her offspring fails to learn the behavior. Equation 2 is replaced by Equation C.6 +and Equation 3 is replaced by Equation C.7. + +$$x''_{sk1} = \frac{1}{2} \left( \underbrace{(τ_{sk} - ϵ)x'_{fk1}}_{\text{Allele from mother}} + \underbrace{(x'_{mk0} + x'_{mk1}) \sum_h (τ_{sh} - ϵ)x'_{fh1}}_{\text{Allele from father}} \right) \quad (C.6)$$ + +$$x''_{sk0} = \frac{1}{2} \left( \underbrace{x'_{fk0} + (1 - τ_{sk} + ϵ)x'_{fk1}}_{\text{Allele from mother}} + \underbrace{(x'_{mk0} + x'_{mk1}) \sum_h (x'_{fh0} + (1 - τ_{sh} + ϵ)x'_{fh1})}_{\text{Allele from father}} \right) \quad (C.7)$$ + +As in the main text, I ran numeric simulations of this system for various combinations of teaching costs, *t*, and reproductive benefits of the cultural trait, *b*, and an innovation rate of *r* = 0.005. Since more than 90% of females learn the trait from their mother I conservatively set *ϵ* to 0.1 as the error rate for the simulation. As in the main text, for each parameter combination, I ran the simulation starting with four initial allele frequencies. In each frequency one allele started at 5% of the population and the rest were evenly distributed among the rest of the population. I started the frequency of the cultural trait at zero in all four initial conditions and ran the simulations until they converged to a shared equilibrium or ran for 10⁷ generations. \ No newline at end of file diff --git a/samples/texts/2274792/page_8.md b/samples/texts/2274792/page_8.md new file mode 100644 index 0000000000000000000000000000000000000000..49be2a8ae4eb9f37b28fcde73183e7b9c563199c --- /dev/null +++ b/samples/texts/2274792/page_8.md @@ -0,0 +1 @@ +Figure C.2: With a transmission error rate of $\epsilon = 0.1$ the uniparental transmission model has similar results to the version in the main text. There is a range where the reproductive cost of teaching is about half the reproductive benefit of the trait where the trait persists at a low frequency (A), females are taught at a high rate (B), and males are taught at a much lower rate (C). However, the trait frequency is low over a greater range than in the model in the main text. \ No newline at end of file diff --git a/samples/texts/2274792/page_9.md b/samples/texts/2274792/page_9.md new file mode 100644 index 0000000000000000000000000000000000000000..6e024eca01fdb560e68240ab9f93a0e0075c761e --- /dev/null +++ b/samples/texts/2274792/page_9.md @@ -0,0 +1,9 @@ +### C.3 Disparate Benefits with Transmission Errors + +This model is identical to the disparate benefits model in the main text except that teaching has an error rate, $\epsilon$, which is the frequency with which an individual who pays a learning cost fails to learn the behavior. Equation 7 is replaced by Equation C.8 and Equation 8 is replaced by Equation C.9. + +$$ y''_{sk1} = \frac{1}{2}(r_{sk} - \epsilon) \underbrace{\left( y'_{fk1} + (y'_{mk0} + y'_{mk1}) \sum_h y'_{fh1} \right)}_{\substack{\text{Allele} \\ \text{from mother}}} \quad (C.8) $$ + +$$ y''_{sk0} = \frac{1}{2} \underbrace{\left( y'_{fk0} + (1-r_{sk}+\epsilon)y'_{fk1} + (y'_{mk0}+y'_{mk1}) \sum_h (y'_{fh0} + (1-r_{sk}+\epsilon)y'_{fh1}) \right)}_{\substack{\text{Allele from mother} \\ \text{Allele from father}}} \quad (C.9) $$ + +As in the main text, I ran numeric simulations of this system for various combinations of female net benefits, $b_f - \mu$ and male net benefits $b_m - \mu$, and an innovation rate of $r = 0.005$. Since more than 90% of females learn the trait from their mother I conservatively set $\epsilon$ to 0.1 as the error rate for the simulation. As in the main text, for each parameter combination, I ran the simulation starting with four initial allele frequencies. In each frequency one allele started at 5% of the population and the rest were evenly distributed among the rest of the population. I started the frequency of the cultural trait at zero in all four initial conditions and ran the simulations until they converged to a shared equilibrium or ran for $10^7$ generations. \ No newline at end of file diff --git a/samples/texts/2926839/page_2.md b/samples/texts/2926839/page_2.md new file mode 100644 index 0000000000000000000000000000000000000000..0aae0b6387d56e98cb58ba842125a58627083a08 --- /dev/null +++ b/samples/texts/2926839/page_2.md @@ -0,0 +1,36 @@ +Journal of Mechanical Science and Technology 00 (2013) 0000–0000 +www.springerlink.com/content/1738-494x + +# Robust multiobjective optimization method using satisfying trade-off method + +Masahiro Toyoda¹ and Nozomu Kogiso¹,* + +¹Department of Aerospace Engineering, Osaka Prefecture University, Sakai, Osaka 599-8531, Japan + +(Manuscript Received 000 0, 2013; Revised 000 0, 2013; Accepted 000 0, 2013) + +## Abstract + +This study proposes a robust multiobjective optimization approach using the satisfying trade-off method (STOM). STOM is a multi-objective optimization method that obtains a highly accurate single Pareto solution. Conventionally, a robust design is formulated as a single-objective optimization problem, where the objective function is defined as the weighted sum of the mean and standard deviation of the performance index. In this study, the mean and standard deviation are formulated as individual objective function. The effect of uncertainty can be investigated through Pareto surface. In STOM, the multiobjective optimization problem is transformed into the equivalent single objective problem by introducing an aspiration level. As the obtained single Pareto solution corresponds to the aspiration level that implies the ratio of the designer's desired objective function, the designer can investigate only the desired space in detail through setting the aspiration without obtaining full Pareto surfaces. The validity of using STOM for a robust multiobjective optimization problem is discussed using numerical examples. + +Keywords: Multiobjective optimization, Robust design, Satisficing trade-off method, Uncertainty + +## 1. Introduction + +Robust optimal design is widely applied to engineering design problems that consider the uncertainties of design variables and parameters such as the material constants and applied load conditions [1, 2]. Robust design optimization is conventionally defined as a single-objective optimization method, where the objective function is defined as the weighted sum of the expected value and the standard deviation of the performance index. Such a weighted sum approach is one of the methods for transforming a multiobjective optimization problem into the equivalent single-objective optimization problem. It is also known that this approach sometimes fails to obtain the desired optimum solution, especially when the Pareto set has a concave shape. + +This paper proposes a method for directly formulating a robust design problem as a multiobjective optimization problem [3], where both the expected value and standard deviation of the performance index are adopted as objective functions. Then, the robust design problem is solved using the satisfying trade-off method (STOM) [4]. STOM is known to be an interactive optimization method and has been applied to many different kinds of multiobjective optimization problems [5, 6]. STOM converts a multiobjective optimization problem into + +the equivalent single-objective optimization problem by introducing an aspiration level that corresponds to the user's preference for each objective function value. In addition, the user can interactively repeat the optimization until the desired Pareto solution is found by updating the aspiration level. The automatic trade-off analysis method [7] is one of the methods used to reasonably update the aspiration level. + +The validity of using STOM for a robust multiobjective optimization problem is discussed through numerical examples. In particular, the accuracy of the Pareto set obtained by parametrically changing the aspiration level is demonstrated. Then, the effect of the random variables on the Pareto set is investigated in the small region of the Pareto set. + +## 2. Robust Multiobjective Optimal Design + +The robust design optimization is formulated to obtain the design with the smallest deterioration in performance under a variety of uncertain design parameters, as well as a reasonable higher performance, as illustrated in Fig. 1, where $z_0$ and $\Delta z$ denote the mean and variation of the random variables, respectively. The deterministic optimal solution $x_{opt}$ may typically show larger variation in the objective function $\Delta f_{opt}$ under variation in the random variables $z$. In contrast, the robust optimal solution will yield a smaller corresponding variation $\Delta f_{robust}$ even if the objective function of the robust design $f(x_{robust})$ is worse than that of the deterministic optimal design $f(x_{opt})$. + +*Corresponding author. Tel.: +81-72-254-9245, Fax.: +81-72-254-9906 +E-mail address: kogiso@aero.osakafu-u.ac.jp + +¹Recommended by Editor 000 000 +© KSME & Springer 2013 \ No newline at end of file diff --git a/samples/texts/2926839/page_3.md b/samples/texts/2926839/page_3.md new file mode 100644 index 0000000000000000000000000000000000000000..72fb93ddfc7d6adc0eba88f0c1f32acb82cc1b7a --- /dev/null +++ b/samples/texts/2926839/page_3.md @@ -0,0 +1,69 @@ +Fig. 1. Concept of robust optimization + +The objective function of the robust design optimization is conventionally formulated as the weighted sum of the mean value and the standard deviation of the objective function, as follows: + +$$ +\text{Minimize} \quad f_r(x) = E[f(x,z)] + \alpha \sqrt{\mathrm{Var}[f(x,z)]} \quad (1) +$$ + +where x and z are the design and random variables, respective- +ly, and α is a weighting coefficient. As the linear approxima- +tion, the mean value and variance of the objective function f +can be evaluated according to the following equation [8]: + +$$ +E[f(\mathbf{x},z)] \approx f(E[\mathbf{x}]) E[z] \quad (2) +$$ + +$$ +\operatorname{Var}[f(\mathbf{x}, z)] \approx \sum_{i=1}^{n_x} \left(\frac{\partial f}{\partial x_i}\right)^2 \operatorname{Var}[x_i] + \sum_{i=1}^{n_z} \left(\frac{\partial f}{\partial z_i}\right)^2 \operatorname{Var}[z_i] \quad (3) +$$ + +where $n_x$ and $n_z$ are the numbers of the design and random variables, respectively. + +The single-objective optimization using Eq. (1) sometimes fails to obtain the desired solution, especially when the Pareto set has a non-convex shape, as shown in Fig. 2. In this study, the robust design optimization is formulated as the following multiobjective optimization problem: + +$$ +\begin{align} +\text{Minimize: } & f(x,z) = (f_1, f_2) \tag{4} \\ +& f_1 = E[f(x,z)] \nonumber \\ +& f_2 = \sqrt{\text{Var}[f(x,z)]} \nonumber \\ +\text{subject to: } & g_j(x,z) \le 0 \quad (j=1,\dots,m) \nonumber \\ +& x_i^L \le x_i \le x_i^U \quad (i=1,\dots,n_x) \nonumber +\end{align} +$$ + +where $g_j(x, z)$ and $(j = 1, ..., m)$ are constraint conditions, and $x_i^U$ and $x_i^L$ are the upper and lower limits of the design variables, respectively. + +**3. Satisficing Trade-off Method (STOM)** + +STOM is known to be an interactive optimization method and converts a multiobjective optimization problem into the equivalent single-objective optimization problem by introducing an aspiration level that corresponds to the user's preference for each objective function value. The flow of the STOM is + +Fig. 2. Case of non-convex Pareto set + +Fig. 3. Computational flow of STOM + +summarized in Fig. 3 and briefly described as follows. + +Step 1: Set the ideal point $f_i^I$, ($i$ = 1, ..., $k$) of each objective function. The ideal point is usually determined by solving a single-objective optimization problem considering only the corresponding objective function, $f_i(\mathbf{x}, \mathbf{z})$. The ideal point for the mean performance is obtained by solving the deterministic design problem. + +Step 2: Set the aspiration level $f_i^A$, ($i$ = 1, ..., $k$) of each objec- +tive function and evaluate the weight coefficient as fol- +lows: + +$$ +w_i = \frac{1}{f_i^A - f_i^I}, \quad (i = 1, \dots, k) \tag{5} +$$ + +Step 3: Formulate the multiobjective optimization problem in Eq. (4) into the weighted Tchebyshev norm problem as follows: + +$$ +\begin{align*} +\text{Minimize:} \quad & \max_{i=1,\cdots,k} w_i (f_i(\mathbf{x}) - f_i^I) && (6) \\ +\text{subject to:} \quad & g_j(\mathbf{x},\mathbf{z}) < 0, && (j=1,\cdots,m) \\ +& x_i^L \le x_i \le x_i^U && (i=1,\cdots,n_x) +\end{align*} +$$ + +Step 4: The min-max problem in Eq. (6) is transformed into +the equivalent single-objective problem by introducing a \ No newline at end of file diff --git a/samples/texts/2926839/page_4.md b/samples/texts/2926839/page_4.md new file mode 100644 index 0000000000000000000000000000000000000000..b0c1bc0cb2d0f8e40912fb3f6ef92ddf02258ef4 --- /dev/null +++ b/samples/texts/2926839/page_4.md @@ -0,0 +1,57 @@ +Fig. 4. Pareto optimal solution when searching by STOM + +new design variable y as follows. + +$$ +\begin{align*} +\text{Minimize: } & y \tag{7} \\ +\text{subject to: } & w_i(f_i(x,z) - f_i^L) \le y, \quad (j=1,\dots,k) \\ +& g_j(d,z) < 0, \quad (j=1,\dots,m) \\ +& x_i^L \le x_i \le x_i^U, \quad (i=1,\dots,n_x) +\end{align*} +$$ + +Step 5: If the objective function values are satisfactory, the search is finished. Otherwise, update the aspiration level $f_i^A$ and return to Step 2. + +The weight coefficient $w_i$ plays an important role in obtaining the Pareto solution in the direction of the aspiration level, which is directly related to the designer's preference. As shown in Fig. 4, the Pareto optimal solution is usually located on the line connecting the ideal point and the aspiration level in the objective function space, regardless of whether or not the aspiration level lies in the feasible region. When Eq. (7) is solved using a nonlinear programming method, an accurate Pareto optimal solution is obtained. + +An accurate Pareto set is obtained by parametrically changing the aspiration level. On the other hand, the designers can investigate only the desired region in detail by arranging the aspiration level properly without obtaining the full Pareto set. + +**4. Numerical Examples** + +**4.1 Example 1: two-bar truss structure** + +Consider the following problem involving the two-bar truss structure shown in Fig. 5 [9]. The original design problem is to find the nominal diameter for member $x_1$ and the height of structure $x_2$ to minimize the element stress under the constraints of the total volume $g_1(x)$ and buckling stress $g_2(x)$. The external force $F$ is set at 150 kN, the thickness of the circular tube $T$ is set at 2.5 mm, the width of structure $B$ is set at 750 mm, and the elastic modulus $E$ is set at 210 GPa. The original optimization problem is formulated as follows: + +$$ +\text{Minimize:} \quad f(x) = \frac{F\sqrt{B^2 + x_2^2}}{2\pi T x_1 x_2} \qquad (8) +$$ + +Fig. 5. Two-bar truss design problem + +$$ +\begin{array}{ll} +\text{subject to:} & g_1(\mathbf{x}) = \frac{2\pi T x_1 \sqrt{B^2 + x_2^2}}{70000} \le 1 \\ +& g_2(\mathbf{x}) = \frac{4F_1(B^2 + x_2^2)^{3/2}}{E\pi^3 T x_1 x_2(T^2 + x_1^2)} \le 1 \\ +& 1 \le x_1 \le 100 \\ +& 1 \le x_2 \le 1000 +\end{array} +$$ + +The robust design problem is considered when the member diameter and height have uncertainties, with ($Δx_1$, $Δx_2$) used to denote random variables, where the mean value of $x_1$ and $x_2$ are treated as design variables. In this example, the standard deviation of a random variable is assumed to be $Δx_i/3$ [9]. The robust multiobjective design problem is formulated as follows: + +$$ +\begin{align} +\text{Minimize:} \quad & f_1(\mathbf{x}) = f(E[\mathbf{x}]) = \frac{F\sqrt{B^2 + x_2^2}}{2\pi T x_1 x_2} \tag{9} \\ +& f_2(\mathbf{x}, \Delta\mathbf{x}) = \sqrt{\mathrm{Var}[f(\mathbf{x}, \Delta\mathbf{x})]} \nonumber \\ +& = \frac{1}{3}\sqrt{\left(\frac{\partial f}{\partial x_1}\right)^2 \Delta x_1^2 + \left(\frac{\partial f}{\partial x_2}\right)^2 \Delta x_2^2} \nonumber \\ +\text{subject to:} \quad & \hat{g}_1(\mathbf{x}) = g_1(E[\mathbf{x}]) + \sqrt{\mathrm{Var}[g_1(\mathbf{x})]} \le 1 \nonumber \\ +& \hat{g}_2(\mathbf{x}) = g_2(E[\mathbf{x}]) + \sqrt{\mathrm{Var}[g_2(\mathbf{x})]} \le 1 \nonumber \\ +& 1 + \Delta x_1 \le x_1 \le 10 - \Delta x_1 \nonumber \\ +& 10 + \Delta x_2 \le x_2 \le 100 - \Delta x_2 \nonumber +\end{align} +$$ + +where $\hat{g}_j(x)$ are the constraints for the considered variations of the design variables evaluated as the first-order approximation [8]. The constraints mean the volume and buckling load limits should be satisfied under variations. + +The Pareto front under several values of $\Delta x$ is obtained by parametrically changing the aspiration level, as shown in Fig. 6. The ideal points are set by solving each single-objective function problem, as listed in Table 1. Fig. 6 (a) shows the Pareto set when changing $\Delta x_1$ to 1.0, 1.5, and 2.0, where $\Delta x_2$ is constant at $\Delta x_2 = 5.0$. On the other hand, Fig. 6 (b) corresponds to the Pareto set when changing $\Delta x_2$ to 5.0, 7.0, and 10.0, where $\Delta x_1$ is constant at $\Delta x_1 = 1.0$. Note that the blue \ No newline at end of file diff --git a/samples/texts/2926839/page_5.md b/samples/texts/2926839/page_5.md new file mode 100644 index 0000000000000000000000000000000000000000..b866d9acdadb6a03b6097c6054ad812d79f22b42 --- /dev/null +++ b/samples/texts/2926839/page_5.md @@ -0,0 +1,38 @@ +Fig. 6. Pareto front in example 1 + +curves in both figures indicate the same Pareto set under ($\Delta x_1$, $\Delta x_2$) = (1.0, 5.0). It is found that smooth Pareto fronts are obtained in both cases, because each Pareto design is obtained by using a sequential quadratic programming method. + +As $\Delta x_1$ becomes larger, the Pareto front moves in the upper right direction, as shown in Fig. 6 (a). On the other hand, the Pareto front moves in the upper left direction as $\Delta x_2$ becomes larger, as shown in Fig. 6 (b). + +A comparison of these figures shows that the variation of $x_1$ has a larger effect on the Pareto set, especially on $f_2$, which is the variation of the normal stresses. + +## 4.2 Example 2: ten-bar truss structure + +Consider the ten-bar truss problem shown in Fig. 7, where the representative length L is set at 360, and the applied load P is 100 [10]. The original deterministic design problem is formulated to minimize the structural volume in terms of the member's cross-sectional area subject to the member's stress constraints. + +This problem is modified to give the following robust multiobjective optimization problem. Assume that the applied load P and member strength have variations and are treated as random variables. The robust design problem is formulated as three objective function problems, which are defined as the structural volume and the mean and standard deviation of the + +Fig. 7. Ten-bar truss design problem + +Table 1. Ideal points for each condition in example 1. + +
Δx1Δx2f1'f2'
1.05.0324.022.46
1.55.0325.313.65
2.05.0326.614.87
1.07.0324.052.53
1.010.0324.092.62
+ +tip displacement $d_2$ in terms of the member's cross-sectional area $x_i$, where $x_i$ is assumed to be deterministic. The robust multiobjective optimization problem is formulated as follows: + +$$ +\begin{array}{ll} +\text{Minimize:} & f_1(x) = \sum_{i=1}^{10} l_i x_i \\ +& f_2(x,z) = E[d_2(x,z)] \\ +& f_3(x,z) = \sqrt{\text{Var}[d_2(x,z)]} \\ +\text{subject to:} & -s + \Delta s \le \sigma_i(x,z) \le s - \Delta s \quad (i=1,\dots,10) \\ +& 0.1 \le x_i \le 10.0 \quad (i=1,\dots,10) +\end{array} +\qquad (10) +$$ + +where $l_i$ is the member length, $\sigma_i$ is the member stress, and $s$ and $\Delta s$ are the mean value and standard deviation of the member strength, respectively. The mean value and standard deviation of the applied load are set at $E[P] = 100.0$ and $\Delta P = 10.0$, respectively. The mean value and standard deviation of the member strength are set at $E[s] = 25.0$ and $\Delta s = 1.0$, respectively. + +The obtained Pareto set in the objective function space is shown in Fig. 8 with different viewing angles, where each point is obtained by changing the aspiration level parametrically under the ideal point $f^1 = (1.659 \times 10^4, 1.956, 1.345 \times 10^{-2})^\mathrm{T}$, as determined from each single-objective optimization problem. It is found that a smooth Pareto surface is obtained, because each Pareto design is obtained by using a sequential quadratic programming method. The Pareto set indicates that a trade-off exists between the structural volume, mean of the tip displacement, and variance of the tip displacement. + +Next, the correspondences of the selected Pareto designs are investigated in detail. First, designs A, B, C, and D on the Pareto set shown in Fig. 8 are selected, because the Pareto solutions have almost the same volume and mean of the tip displacement, but have different standard deviations of the tip displacement, as shown in Fig. 9. Design A is the most robust \ No newline at end of file diff --git a/samples/texts/2926839/page_6.md b/samples/texts/2926839/page_6.md new file mode 100644 index 0000000000000000000000000000000000000000..b82e0bdf4ab6fb78518bd366f8ce03abe6c0824f --- /dev/null +++ b/samples/texts/2926839/page_6.md @@ -0,0 +1,22 @@ +Fig. 8. Pareto front in example 2 + +Fig. 9. Comparison of Pareto designs + +among these four designs, because the variation of the tip displacement is the smallest. The aspiration levels (Asp.) and obtained objective function values (Pareto) are listed in Table 2. This investigation is possible without obtaining full Pareto set actually, if the aspiration levels are adequately selected for obtaining these four designs. + +Fig. 10 shows the changes in the member cross-sectional areas of these designs, where the horizontal axis indicates the standard deviation of the tip displacement $f_3(x)$. The leftmost + +Table 2. Objective function values of selected Pareto designs + +
DesignVolume ×104E[d3]√Var[d2] ×10-2
AAsp.1.2501.4001.943
Pareto2.0232.7927.808
BAsp.1.2021.4262.166
Pareto1.9952.7908.430
CAsp.1.2101.4502.450
Pareto1.9822.7709.108
DAsp.1.2001.4002.600
Pareto1.9922.7419.622
+ +Fig. 10. Transitions of member cross-sectional areas for designs A, B, C, and D +Table 3. Comparison of cross-sectional areas + +
ABCDchange
x18.9119.6119.89710.0
x20.10.10.10.1
x38.3418.3348.3168.299
x46.9426.8796.0005.672
x50.10.10.10.1
x60.10.10.10.1
x75.9045.8945.8685.841
x86.4566.5037.1478.016
x99.9518.9148.4768.017
x100.10.10.10.1
+ +value corresponds to design A, the most robust design, and the rightmost value corresponds to design D. The cross-sectional areas of members 2, 5, 6, and 10 are not shown in this figure, because these areas converge to the lower bound (0.1) in all the designs. The details of the cross-sectional area values are listed in Table 3. As the variation of the tip displacement becomes larger, the cross-sectional areas of members 1 and 8 increase, and those of 4 and 9 decrease. The cross-sectional area distributions of designs A, B, C, and D are illustrated in Fig. 11. Note that members 1 and 8 are located to the left fixed side, but 4 and 9 are to the right free side. In other words, the variation of the tip displacement decreases when the cross-sectional area of the tip-side member increases. + +## 5. Conclusions + +This paper proposed a robust multiobjective design method that uses the STOM [4]. Through numerical examples, the \ No newline at end of file diff --git a/samples/texts/2952501/page_1.md b/samples/texts/2952501/page_1.md new file mode 100644 index 0000000000000000000000000000000000000000..30806260a18c297f21d6eb36d2dd5cabc95597dc --- /dev/null +++ b/samples/texts/2952501/page_1.md @@ -0,0 +1,26 @@ +Electron. J. Probab. 26 (2021), article no. 18, 1–48. +ISSN: 1083-6489 https://doi.org/10.1214/21-EJP579 + +Convergence of Eulerian triangulations + +Ariane Carrance* + +Abstract + +We prove that properly rescaled large planar Eulerian triangulations converge to the Brownian map. This result requires more than a standard application of the methods that have been used to obtain the convergence of other families of planar maps to the Brownian map, as the natural distance for Eulerian triangulations is a canonical oriented pseudo-distance. To circumvent this difficulty, we adapt the layer decomposition method, as formalized by Curien and Le Gall in [13], which yields asymptotic proportionality between three natural distances on planar Eulerian triangulations: the usual graph distance, the canonical oriented pseudo-distance, and the Riemannian metric. This notably gives the first mathematical proof of a convergence to the Brownian map for maps endowed with their Riemannian metric. Along the way, we also construct new models of infinite random maps, as local limits of large planar Eulerian triangulations. + +**Keywords:** random maps; scaling limits of maps; local limits of maps; branching processes. +**MSC2020 subject classifications:** 05C80; 60B05; 60J80; 05A16. +Submitted to EJP on January 14, 2020, final version accepted on January 10, 2021. +Supersedes arXiv:1912.13434. + +# 1 Introduction + +## 1.1 Context + +Eulerian triangulations are face-bicolored triangulations. They can be encountered in several contexts. As their definition is quite straightforward, they are already an object of interest in themselves in enumerative combinatorics (see [27, 10, 8, 4]). Moreover, they are in bijection with combinatorial objects such as **constellations** and **bipartite maps**, and geometrical objects such as **Belyi surfaces** (see [19]). They also correspond to the two-dimensional case of **colored tensor models**, an approach to quantum gravity that generalizes matrix models to any dimension (see Part I of [12] for an introduction to this topic). + +The main aim of this paper is to show that large planar rooted Eulerian triangulations converge to the Brownian map (see Theorem 3.1 for a more precise statement). Along the + +*Université Paris-Saclay, France. Supported by the ERC Advanced Grant 740943 GeoBrown. +E-mail: ariane.carrance@math.cnrs.fr \ No newline at end of file diff --git a/samples/texts/2952501/page_10.md b/samples/texts/2952501/page_10.md new file mode 100644 index 0000000000000000000000000000000000000000..9677c8e6afcc33df473a3c19e2ebb77e369e41c5 --- /dev/null +++ b/samples/texts/2952501/page_10.md @@ -0,0 +1,66 @@ +get that, if $|s-s'| \vee |t-t'| \le \eta$, then, for $n$ large enough, the right-hand side of (3.3) is smaller than $4\omega(Z, \eta) + \varepsilon$, where we denote by $\omega(Z, \eta)$ the supremum $\sup_{|I|\le \eta} \omega(Z, I)$, and $\omega(f, I)$ is the modulus of continuity of $f$ on the interval $I$. + +Thus, along a subsequence, we have the joint convergence: + +$$ +(C_{(n)}, L_{(n)}, \vec{D}_{(n)}) \xrightarrow[n\to\infty]{(d)} (\mathbb{E}, Z, D), \qquad (3.4) +$$ + +for some random continuous process D on $[0, 1]^2$. In the rest of this proof, we fix a +subsequence so that (3.4) holds, and work along this subsequence. + +Note that, from Theorem 1.2, we also have the joint convergence of $D_{(n)}$ to $c_0D$. This already implies that $D$ is symmetric, and thus is a pseudo-metric. We now want to show that $D = D^*$ a.s., which will conclude the proof, since this will imply the uniqueness of the limit $D$. + +First, it is straightforward to get from (2.2) and (3.4) that, for any $s \in [0, 1]$: + +$$ +D(s_*, s) = Z_s - \inf Z. \tag{3.5} +$$ + +We will now show that a.s., for every $s, t \in [0, 1]$, + +$$ +D(s,t) \le D^\circ(s,t). \tag{3.6} +$$ + +To prove this claim, let us get back to $\bar{T}_n$ and $\mathcal{T}_n$. From (3.1), for any $\varepsilon > 0$ and any $\delta \in (0, 1)$, for any $n$ large enough, the event + +$$ +|d_n(u,v) - c_0 d_n(v,u)| \le \varepsilon n^{1/4} \quad \forall u,v \in V(\bar{T}_n) +$$ + +holds with probability at least $1 - \delta$. + +On that event, we have, for any $u, v, w \in V(\bar{T}_n)$, + +$$ +\begin{align*} +\vec{d}_n(u, v) &\le \vec{d}_n(u, w) + \vec{d}_n(w, v) \le \frac{1}{c_0} d_n(u, w) + \vec{d}_n(w, v) + \varepsilon n^{1/4} \\ +&\le \vec{d}_n(w, u) + \vec{d}_n(w, v) + 2\varepsilon n^{1/4}. \tag{3.7} +\end{align*} +$$ + +Thus, going back to the proof of Proposition 2.11, when estimating oriented distances +from the length of the concatenation of two predecessor geodesics, rather than having +to multiply this length by 2, we just need to add $2\varepsilon n^{1/4}$. + +Therefore, for any $0 \le i < j \le 2n$, + +$$ +\vec{D}_{(n)}\left(\frac{i}{n}, \frac{j}{n}\right) \le L_{(n)}\left(\frac{i}{n}\right) + L_{(n)}\left(\frac{j}{n}\right) \\ +- 2\max\left(\min_{k\in\{i,\dots,j\}} L_{(n)}\left(\frac{k}{n}\right), \min_{k\in\{j,\dots,2n\}\cup\{0,\dots,i\}} L_{(n)}\left(\frac{k}{n}\right)\right) + \frac{2}{n^{1/4}} + 2\varepsilon. +$$ + +Thus, letting $n \to \infty$ (along our subsequence), for any $\varepsilon > 0$, we have $D \le D^\circ + 2\varepsilon$ +a.s., so that we get the desired inequality (3.6). + +Moreover, as *D* satisfies the triangle inequality, we have + +$$ +D(s,t) \le D^*(s,t) \quad \forall s,t \in [0,1] \text{ a.s.} \tag{3.8} +$$ + +To replace this inequality by an equality, it now suffices to show that, for $U,V$ +chosen uniformly and independently at random in $[0,1]$, and independently from the rest, +$D(U,V) \stackrel{(d)}{=} D^*(U,V)$. Indeed, this would imply $D = D^*$ a.e., and thus $D = D^*$ since both +are continuous. To prove this, from (3.5), it is enough to show that $D(U,V) \stackrel{(d)}{=} D(s_*,U). \ No newline at end of file diff --git a/samples/texts/2952501/page_11.md b/samples/texts/2952501/page_11.md new file mode 100644 index 0000000000000000000000000000000000000000..46f0a4c6e31f547b8bd2a63c4191bdf0988ad852 --- /dev/null +++ b/samples/texts/2952501/page_11.md @@ -0,0 +1,43 @@ +To prove this, let us get back to the discrete level for a moment. Let $u_n, v_n$ be two vertices of $\mathcal{T}_n$ chosen independently and uniformly at random. As $\mathcal{T}_n$ re-pointed at $u_n$ has the same law as $\bar{\mathcal{T}}_n$, we have + +$$ \vec{d}_n(u_n, v_n) \stackrel{(d)}{=} \vec{d}_n(o_n, v_n). \quad (3.9) $$ + +Similarly to the case of usual triangulations in [20], this implies the desired equality in distribution $D(U, V) \stackrel{(d)}{=} D(s_*, U)$. Indeed, set $U_n = \lceil(2n-1)U\rceil$ and $V_n = \lceil(2n-1)V\rceil$, which are both uniformly distributed over $\{1, 2, \dots, 2n-1\}$, so that + +$$ \frac{U_n}{n} \xrightarrow{n \to \infty} U, \quad \frac{V_n}{n} \xrightarrow{n \to \infty} V. $$ + +Then, from (3.4), we have + +$$ \vec{D}_{(n)} \left( \frac{U_n}{n}, \frac{V_n}{n} \right) \xrightarrow{n \to \infty} \vec{D}(U, V). $$ + +Now, from (3.9), we have that the distribution of $\vec{D}(U,V)$ is also the limiting distribution of + +$$ L_{(n)} \left( \frac{U_n}{n} \right) - \min L_{(n)} + 1, $$ + +so that $\vec{D}(U,V)$ has the same distribution as $Z_U - \inf Z$, which is also the distribution of $D(s_*, U)$, from (3.5). This concludes the proof. $\square$ + +# 4 Technical preliminaries + +## 4.1 Consequences of the convergence of the rescaled labels + +We now prove a few technical properties of $\vec{d}$ that stem from the convergence given in Theorem 2.12. + +For any integer $n \ge 1$, let $\rho_n$ be the root vertex of the random triangulation $\mathcal{T}_n$, uniform over the rooted planar Eulerian triangulations with $n$ black faces. We denote by $\bar{\mathcal{T}}_n$, the triangulation $\mathcal{T}_n$ together with a distinguished vertex $o_n$, picked uniformly at random in $\mathcal{T}_n$. We then have the following result: + +**Proposition 4.1.** *The following convergence holds:* + +$$ n^{-1/4} \vec{d}(o_n, \rho_n) \xrightarrow[n\to\infty]{(d)} \sup Z. $$ + +Consequently, the sequence $(n^{-1/4}\vec{d}(o_n, \rho_n))_{n\ge 1}$ is bounded in probability and bounded away from zero in probability, as well as the sequence $(n^{-1/4}\vec{d}(\rho_n, o_n))_{n\ge 1}$. + +*Proof.* Recall that $\bar{\mathcal{T}}_n$ is in correspondence with a random tree $\mathcal{T}_n$, uniform over the well-labeled plane trees with $n$ edges, whose labelling we denote by $l_n$. We have, from (2.2), that + +$$ \vec{d}(o_n, \rho_n) = - \min_{v \in V(\mathcal{T}_n)} l(v) + 1. $$ + +Then, using the convergence of Theorem 2.12, we get that the quantity + +$$ n^{-1/4} \left( - \min_{v \in V(\mathcal{T}_n)} l(v) + 1 \right) $$ + +converges in distribution to $(-\inf Z) \stackrel{(d)}{=} \sup Z$. + +This directly implies the bounds in probability for the sequence $(n^{-1/4}\vec{d}(o_n, \rho_n))_{n\ge 1}$. For those pertaining to $(n^{-1/4}\vec{d}(\rho_n, o_n))_{n\ge 1}$, recall that for any two vertices $u, v$ of an Eulerian triangulation, $\frac{1}{2}\vec{d}(u, v) \le \vec{d}(v, u) \le 2\vec{d}(u, v).$ $\square$ \ No newline at end of file diff --git a/samples/texts/2952501/page_12.md b/samples/texts/2952501/page_12.md new file mode 100644 index 0000000000000000000000000000000000000000..fb72b303db9b31ef18a4c4faf4392d19abc913c3 --- /dev/null +++ b/samples/texts/2952501/page_12.md @@ -0,0 +1,17 @@ +way, we explore uncharted properties of planar Eulerian triangulations. This allows us to construct, in the case of Eulerian triangulations, many random objects and structures whose equivalents already exist for other families of planar maps. + +Let us now briefly sketch how this exploration ties in together with proving Theorem 3.1. + +If one wants to prove that a family of planar maps converges to the Brownian map, the classical method is to use a **bijection** between this family, and a family of labeled trees, whose labels keep track of the distances in the map. Obtaining a joint scaling limit for the trees and their label functions is a classical procedure, however, it then remains to deduce from this limit, a scaling limit for the metric space induced by the maps. This was first done independently by Le Gall for triangulations and $2p$-angulations [20], and Miermont for quadrangulations [25], using different technical tools. The list of families amenable to this method has been expanded since then to general maps, general bipartite maps, simple triangulations and odd $p$-angulations [7, 1, 3, 2]¹. + +A more recent method applies to local modifications of distances, in families that are already known to converge to the Brownian map. This method, established by Curien and Le Gall in [13] for usual triangulations, uses a **layer decomposition** of the maps, rather than a bijection with trees. This makes it possible to use an ergodic subadditivity argument, to obtain that the modified and original distances are asymptotically proportional. This method has recently been extended by Lehéricy to planar quadrangulations (and general maps, via Tutte's bijection) in [23], using the layer decomposition of quadrangulations established by Le Gall and Lehéricy in [21]. Note that a first notion of layer decomposition was already introduced by Krikun for usual triangulations (without self-loops) [18] and for quadrangulations [17]. + +In the case of Eulerian triangulations, there exists a bijection with a family of labeled trees, but, as we will explain in the sequel, these labels do not correspond to the usual graph distance from the root, but to an oriented pseudo-distance. This implies that we cannot a priori recover the distances from the labels, so that, while it is still easy to get a scaling limit at the level of labeled trees, we are stuck there without any additional ingredient. This ingredient turns out to be the layer decomposition. Indeed, the usual graph distance can be seen as a local modification of the oriented pseudo-distance, so that the layer decomposition method applies to Eulerian triangulations equipped with these two distances. This method then yields that the oriented pseudo-distance is asymptotically proportional to the usual graph distance, so that the labels do keep track of it up to a small error. This proves to be enough to obtain convergence to the Brownian map. + +This is the first time that a combination of these two methods is needed to show such a convergence. It would be interesting to apply this to other families of maps, such as Eulerian quadrangulations. + +Our layer decomposition of Eulerian triangulations also allows us to prove their convergence to the Brownian map when endowed with the **Riemannian metric**, which is inherited from the Euclidean geometric realization obtained by gluing equilateral triangles according to the combinatorics of the map. This result is the first of its kind to be proven mathematically, and as such it reinforces the link between random maps and models of 2D quantum gravity in theoretical physics, such as Causal Dynamical Triangulations (see for instance [5]), in which it is the geometric realization itself that is studied. + +Note that one could want to prove the convergence of planar Eulerian triangulations to the Brownian map using their bijection with bipartite maps, as the convergence for + +¹ We stay purposefully vague here, as some of these results rely on bijections with other types of decorated trees. \ No newline at end of file diff --git a/samples/texts/2952501/page_13.md b/samples/texts/2952501/page_13.md new file mode 100644 index 0000000000000000000000000000000000000000..27faead7454c4d219adede8027de5397ad01ce8a --- /dev/null +++ b/samples/texts/2952501/page_13.md @@ -0,0 +1,45 @@ +Convergence of Eulerian triangulations + +For a rooted Eulerian triangulation (possibly with a boundary) $\Delta$, let $N(\Delta)$ be the number of black triangles of $\Delta$. Then: + +**Proposition 4.2.** Let $\alpha > 0$, and let us denote by $B_r(\bar{T}_n, o_n)$ the ball (for the oriented distance) of radius $r$ in $\bar{T}_n$, centered at $o_n$. For any $\varepsilon \in (0, 1)$, there exists some $b \in (0, 1)$ such that + +$$ \liminf_{n \to \infty} \mathbb{P}(N(B_{\alpha n^{1/4}}(\bar{T}_n, o_n)) > bn) \geq 1 - \varepsilon. $$ + +*Proof.* Let us roughly sketch the idea of the proof. Recall that $\bar{T}_n$ is in correspondence with a random tree $\mathcal{T}_n$, uniform over the well-labeled plane trees with $n$ edges. Moreover, the black triangles of $\bar{T}_n$ correspond to the edges of $\mathcal{T}_n$, so that, for any type-$m$ black triangle $t$ of $\bar{T}_n$: + +$$ \mathrm{Leb}\{s \in [0, 2n) \mid u_{|s|}^{(n)} \text{ is the vertex of type } m \text{ in } t\} = 2. $$ + +Thus, using (2.2), we have + +$$ \frac{1}{n} \cdot N(\mathcal{B}_{\alpha n^{1/4}}(\bar{T}_n, o_n)) \geq \int_0^1 \mathbb{1}_{\{l_n(\lfloor 2ns \rfloor) - \min l_n \leq \alpha n^{1/4} - 1\}} ds. $$ + +Note that we have, for any $s \in [0, 1]$: + +$$ |l_n(\lfloor 2ns \rfloor) - n^{1/4}L_{(n)}(s)| \leq 1, $$ + +so that: + +$$ \int_0^1 \mathbb{1}_{\{l_n(\lfloor 2ns \rfloor) - \min l_n \leq \alpha n^{1/4} - 1\}} ds \geq \int_0^1 \mathbb{1}_{\{L_{(n)}(s) - \min L_{(n)} \leq \alpha - 2/n^{1/4}\}} ds. $$ + +Therefore: + +$$ \mathbb{P}(N(\mathcal{B}_{\alpha n^{1/4}}(\bar{T}_n, o_n)) > bn) \geq \mathbb{P}\left(\int_0^1 \mathbb{1}_{\{L_{(n)}(s) - \min L_{(n)} \leq \alpha - 2/n^{1/4}\}} ds > b\right). $$ + +Now, the liminf of the probability on the right-hand side of the previous inequality can be bounded below by + +$$ \mathbb{P}\left(\int_{0}^{1} \mathbb{1}_{\{Z_s - \inf Z \leq \frac{\alpha}{2}\}} ds > b\right), $$ + +which tends to 1 as $b$ tends to 0, as $Z$ is continuous. + +This concludes the proof. $\square$ + +**Proposition 4.3.** For any $\varepsilon > 0$ and any $\delta \in (0, 1)$ there exists an integer $k \ge 1$ such that, for any sufficiently large $n$, if $o_n^1, \dots, o_n^k$ are chosen uniformly and independently in $V(T_n)$, we have + +$$ \mathbb{P}\left(\sup_{x \in V(T_n)} \left(\inf_{1 \le j \le k} \vec{d}(x, o_n^j)\right) > \varepsilon n^{1/4}\right) \le \delta. $$ + +*Proof.* Let us fix an integer $K \ge 1$. Recall that we write $(u_i^{(n)})_{0\le i\le 2n-1}$ for the vertices of $\mathcal{T}_n$ along its contour exploration. Then, for $k$ large enough, for any sufficiently large $n$, + +$$ \mathbb{P}\left(\forall i \in \{0, \dots, 2K-1\} \exists j \in \{1, \dots, k\} \exists m \in \{\lfloor \frac{in}{k} \rfloor, \dots, \lfloor \frac{(i+1)n}{k} \rfloor\}, o_m^j = u_m^{(n)}\right) \geq 1 - \frac{\delta}{2}. \quad (4.1) $$ + +We will now argue on the event in (4.1). \ No newline at end of file diff --git a/samples/texts/2952501/page_14.md b/samples/texts/2952501/page_14.md new file mode 100644 index 0000000000000000000000000000000000000000..c52e5061fea905c91e5772420ebbbf9f76fbaa53 --- /dev/null +++ b/samples/texts/2952501/page_14.md @@ -0,0 +1,79 @@ +Using (2.2), we have, for any $i,j \in \{0,1,\dots,2n\}$, + +$$ +\vec{d}_n(u_i^{(n)}, u_j^{(n)}) \le 2(l_n(u_i^{(n)}) + l_n(u_j^{(n)})) - 2\tilde{l}_n(i,j) + 2, +$$ + +so that, for any $n$ sufficiently large: + +$$ +\sup_{x \in V(\tau_n)} \left( \inf_{1 \le j \le k} \vec{d}(x, o_n^j) \right) \le 4 \max_{0 \le i \le 2K-1} \omega \left( l_n \left[ \lfloor \frac{in}{K} \rfloor, \lfloor \frac{(i+1)n}{K} \rfloor \right] \right) + 4 +$$ + +where $\omega(f, I)$ is the modulus of continuity of the function $f$ on the interval $I$. +Therefore, we have + +$$ +\liminf_n \mathbb{P} \left( \sup_{x \in V(\tau_n)} \left( \inf_{1 \le j \le k} \vec{d}(x, o_n^j) \right) < \varepsilon n^{1/4} \right) \geq \mathbb{P} \left( \omega(Z, \frac{1}{K}) < \frac{\varepsilon}{5} \right), +$$ + +by using once again the convergence of Theorem 2.12. (We denote by $\omega(Z, \eta)$ the +supremum $\sup_{|I| \le \eta} \omega(Z, I)$.) + +Now, as Z is a.s. continuous on $[0, 1]$, it is uniformly continuous, so that, for any $\epsilon > 0$, +for $K$ large enough, + +$$ +\mathbb{P}\left(\omega(Z, \frac{1}{K}) < \frac{\varepsilon}{5}\right) \geq 1 - \frac{\delta}{2}, +$$ + +which concludes the proof. + +## 4.2 Enumeration results + +We will need some asymptotic results on the generating series $B(t, z)$ of Eulerian triangulations with a semi-simple alternating boundary, as defined in Section 2.4: + +$$ +B(t, z) = \sum_{n,p \ge 0} B_{n,p} t^n z^p, +$$ + +where $B_{n,p}$ is the number of Eulerian triangulations with semi-simple alternating boundary of length $2p$ and with $n$ black triangles. Jérémie Bouttier and the author obtain in [9] a rational parametrization of $B(t, z)$ in, which yields the following asymptotic result: + +**Theorem 4.4.** We have + +$$ +[t^n] B(t, z) = \sum_{p \ge 0} B_{n,p} z^p \underset{n \to \infty}{\sim} \frac{3}{2} \frac{z}{\sqrt{\pi(z-1)(4z-1)^3}} 8^n n^{-5/2} \quad \forall z \in [0, \frac{1}{4}). \quad (4.2) +$$ + +This implies that: + +$$ +\begin{equation} +\left\{ +\begin{array}{@{}l@{}} +B_{n,p} \underset{n\to\infty}{\sim} C(p) 8^n n^{-5/2} \quad \forall p \\ +C(p) \underset{p\to\infty}{\sim} \frac{\sqrt{3}}{2\pi} 4^p \sqrt{p} \quad \text{and} \quad \sum_{p\ge 1} C(p) z^p = \frac{3}{2} \frac{z}{\sqrt{\pi(z-1)(4z-1)^3}} \quad \forall z \in [0, \frac{1}{4}) +\end{array} +\right. +\tag{4.3} +\end{equation} +$$ + +Note that (4.2) is much stronger than (4.3). Indeed, it states that, for any $\epsilon > 0$, for +any $n$ large enough, we have, for all $z \in [0, 1/4)$, + +$$ +(1 - \varepsilon)8^n n^{-5/2} f(z) \le g_n(z) \le (1 + \varepsilon)8^n n^{-5/2} f(z), \quad (4.4) +$$ + +where + +$$ +f(z) = \frac{3}{2} \frac{z}{\sqrt{\pi(z-1)(4z-1)^3}} +$$ + +and + +$$ +g_n(z) = \sum_{p \ge 0} B_{n,p} z^p . +$$ \ No newline at end of file diff --git a/samples/texts/2952501/page_15.md b/samples/texts/2952501/page_15.md new file mode 100644 index 0000000000000000000000000000000000000000..c86989d853121e2112295cfaaf22517822d8c64e --- /dev/null +++ b/samples/texts/2952501/page_15.md @@ -0,0 +1,29 @@ +Thus, as both $f$ and $g_n$ are analytic functions on $[0, 1/4)$, by taking the successive derivatives of the terms in (4.4), we obtain equivalent bounds for the successive coefficients of $f$ and $g_n$ seen as power series: + +$$ (1 - \varepsilon)8^n n^{-5/2}([z^p]f(z)) \le B_{n,p} \le (1 + \varepsilon)8^n n^{-5/2}([z^p]f(z)), $$ + +for any $n$ large enough and for any $p$. + +This yields that, for all $n, p \ge 1$, + +$$ cC(p)8^n n^{-5/2} \le B_{n,p} \le c' C(p)8^n n^{-5/2}, \quad (4.5) $$ + +for some constants $0 < c < c'$ independent of $n$ and $p$. + +Let us now focus on the coefficients + +$$ Z(p) := \sum_{n \ge 0} \left(\frac{1}{8}\right)^n B_{n,p}. \quad (4.6) $$ + +The calculations of [9] also yield the following exact formula: + +$$ \sum_{p \ge 0} Z(p) z^p = \sum_{\substack{p \ge 0, \\ n \ge 0}} \left(\frac{1}{8}\right)^n B_{n,p} z^p = \frac{1 + 7z - 8z^2 + \sqrt{(z-1)(4z-1)^3}}{2(1-z)} \quad \forall z \in [0, \frac{1}{4}), \quad (4.7) $$ + +which gives the asymptotic behavior: + +$$ Z(p) \underset{p \to \infty}{\sim} \frac{1}{4} \sqrt{\frac{3}{\pi}} 4^p p^{-5/2} \quad \text{and} \quad Z(0) = 1. \quad (4.8) $$ + +In particular, for any $p \ge 1$, the sum $Z(p) = \sum_n B_{n,p} 8^{-n}$ is finite, which makes it possible to define the **Boltzmann distribution** on Eulerian triangulations of the $2p$-gon (with a semi-simple alternating boundary), that assigns a weight $8^{-n}/Z(p)$ to each such triangulation having $n$ black triangles. A random triangulation sampled according to this measure will be called a **Boltzmann Eulerian triangulation of perimeter $2p$**. + +Note that there is a natural bijection between Eulerian triangulations of the $2$-gon (with an alternating boundary), and rooted planar Eulerian triangulations, which simply consists in “zipping” or “unzipping” the root edge (see Figure 10). This simple observation will be useful in the sequel. + +Figure 10: The bijection between Eulerian triangulations of the 2-gon with an alternating boundary, and rooted planar Eulerian triangulations. \ No newline at end of file diff --git a/samples/texts/2952501/page_16.md b/samples/texts/2952501/page_16.md new file mode 100644 index 0000000000000000000000000000000000000000..b32e723772a994bb5188ffde27ff0fcf1351ff50 --- /dev/null +++ b/samples/texts/2952501/page_16.md @@ -0,0 +1,7 @@ +# 5 Skeleton decomposition + +We previously considered the balls of planar rooted Eulerian triangulations, and the associated hulls, defined with the oriented distance from the root. + +To obtain the layer decomposition of finite planar Eulerian triangulations that will be crucial to the rest of this paper, we want to generalize the notions of hulls, to the layers of a planar Eulerian triangulation $A$, that lie between the boundaries of two different hulls of $A$ (see Figure 11). More precisely, we want to have a good definition of oriented distance from the boundary of a rooted planar Eulerian triangulation with an alternating boundary (or, as will be the case for our layers, two disjoint boundaries with one distinguished as the “bottom” one): such an oriented distance will once again induce a structure of sets of simple closed curves $\{C_n\}$, going through type-$n$ modules, and, if we look at different layers making up some triangulation $A$, we want the union of these sets of curves to be $\{C_n(A)\}$ (see Figure 11). For that purpose, if $A$ is a planar Eulerian triangulation with a (distinguished) alternating boundary $\partial_0 A$, rooted on this boundary, we define on $V(A)$, the **oriented distance from the boundary** $\partial_0 A$ as the pullback of the oriented distance on $A'$, where $A'$ is the planar Eulerian triangulation obtained from $A$ by gluing into pairs the edges of $\partial_0 A$, as shown in Figure 12. Thus, this distance alternates between 0 and 1 on $\partial_0 A$, and, for an inner vertex $v$ of $A$, is the shortest oriented distance from a 0-labeled outer vertex, to $v$ (see Figure 12). + +Figure 11: The sets of simple closed curves $\{C_n(A)\}$, cutting up the planar Eulerian triangulation $A$ into layers of increasing distance from the origin $\rho$, also separate the different layers of a subtriangulation $\Delta$ lying between two such curves: this guides us for the good notion of oriented distance from the bottom boundary of $\Delta$. \ No newline at end of file diff --git a/samples/texts/2952501/page_17.md b/samples/texts/2952501/page_17.md new file mode 100644 index 0000000000000000000000000000000000000000..0a3e544357dac162dd1e907a0ec42ff1dbed12b6 --- /dev/null +++ b/samples/texts/2952501/page_17.md @@ -0,0 +1,23 @@ +Figure 12: The geodesic oriented distances in an Eulerian triangulation $A$ with a (distinguished) alternating boundary $\partial_0 A$ (left) are given by the distances from the root in the triangulation $A'$, obtained from $A$ by gluing into pairs the edges of $\partial_0 A$ (right). The triangles adjacent to the boundary are depicted in black and white, the rest of triangulation is sketched in gray, and the boundary is in red. + +Now that we have a satisfactory notion of distance, as before, we will be interested +in the union of faces of $A$ incident to vertices at (oriented) distance less than $n$ from +the boundary. We will denote this union $B_{n+1}(A)$. As was the case for usual balls, the +faces of $B_{n+1}(A)$ adjacent to its boundary parts, other than the original boundary $\partial A$, +will correspond to modules of type $n+1$, for the oriented distance from $\partial_0 A$. Once again, +we will have the convention that these boundary parts are simple, and we will glue them +to semi-simple boundaries. If $A$ is pointed at a vertex $v$ at oriented distance at least +$n+2$ from the boundary, we can also define a notion of hull for $B_{n+1}(A)$, which will be +an Eulerian triangulation with two boundaries of specific types. In this section, we will +first develop the description of such triangulations, before dealing with random Eulerian +triangulations with one boundary, and their hulls. + +Note that the chain of arguments and notation of this section follow closely those of [13, Section 5]: in order to be both concise and precise, we detail in the proofs of this section only the additional subtleties and difficulties arising in our case compared to the equivalent results of [13]. + +**5.1 Cylinder triangulations** + +**Definition 5.1.** We call *Eulerian cylinder triangulation of height* $r \ge 1$, an *Eulerian triangulation with two boundaries, one (the bottom of the cylinder) being alternating and semi-simple, the other one (the top) being a succession of modules (see Figure 13)*, and such that any module adjacent to the top boundary is of distance type *r* with respect to the bottom. + +We denote by $\partial\Delta$ its bottom boundary, and by $\partial^*\Delta$ its top boundary. The root is an edge on $\partial\Delta$ oriented such that the bottom face sits on its right. + +Let $\Delta$ be an Eulerian cylinder triangulation of height $r$. Let $2p$ be the bottom boundary length, and $2q$ the top boundary length. For $1 \le j \le r$, the ball $B_j(\Delta)$ is defined as the union of all edges and faces of $\Delta$ incident to at least a vertex at distance $ r+1$. If $\bar{d}(\partial\Delta, v) \le r+1$, we can set $B_r^\bullet(\Delta) = \Delta$. + +Let $\mathcal{T}_n^{(p)}$ be a uniform random triangulation over the set of Eulerian triangulations with a semi-simple alternating boundary of length $2p$ and with $n$ black triangles. We denote by $\overline{\mathcal{T}}_n^{(p)}$ the pointed triangulation obtained by choosing a uniform random inner vertex of $\mathcal{T}_n^{(p)}$. Let $\Delta$ be a cylinder triangulation of height $r$, of respective bottom and top boundary lengths $2p$ and $2q$, with $N$ black triangles, with $n \ge N$. Using the skeleton decomposition, we associate to $\Delta$ a $(p, q, r)$-admissible forest $\mathcal{F}$, together with triangulations $(M_v)_{v \in \mathcal{F}}$ filling in the "slots" between the modules of $\mathcal{M}(\Delta)$. We write $N(M_v)$ for the number of black triangles of $M_v$, for every $v \in \mathcal{F}^*$. + +**Lemma 5.3.** We have + +$$ \lim_{n \to \infty} \mathbb{P} \left( B_r^\bullet \left( \overline{\mathcal{T}}_n^{(p)} \right) = \Delta \right) = \frac{4^{-q} C(q)}{4^{-p} C(p)} \prod_{v \in \mathcal{F}^*} \theta(c_v) \frac{8^{-N(M_v)}}{Z(c_v + 1)}, \quad (5.1) $$ + +where + +$$ \theta(k) = \frac{1}{8} 4^{-k+1} Z(k+1), \quad (5.2) $$ + +with $Z(k)$ defined as in (4.6). + +*Proof.* First note that this result is the equivalent of [13, Lemma 2]. It is obtained very similarly, though in our case we start from a slightly less explicit expression, as shown in (5.3): this stems from the fact that our triangulations do not necessarily have simple boundaries. \ No newline at end of file diff --git a/samples/texts/2952501/page_2.md b/samples/texts/2952501/page_2.md new file mode 100644 index 0000000000000000000000000000000000000000..e84f23a1d90ca5ced249b86cc8a88efe7afb663e --- /dev/null +++ b/samples/texts/2952501/page_2.md @@ -0,0 +1,37 @@ +As will be clearer in the proof of Theorem 3.1, this factor of 2 is really the stumbling block that prevents us from reaching the convergence to the Brownian map using only the bijective approach. + +## 2.3 Convergence of the labeled trees + +From what precedes, starting from a uniform random rooted, pointed planar Eulerian triangulation with $n$ black faces, we get a uniform random well-labeled tree $\mathcal{T}_n$ with $n$ edges. Let us now explain how we can make sense of taking a continuum scaling limit of the latter. We first define the **contour process** of $\mathcal{T}_n$: let $e_0, e_1, \cdots, e_{2n-1}$ be the sequence of oriented edges bounding the unique face of $\mathcal{T}_n$, starting with the root edge, and ordered counterclockwise around this face. Then let $u_i = e_i^-$ be the $i$-th visited vertex in this contour exploration, and set the **contour process** of $\mathcal{T}_n$ at time $i$: + +$$C_n(i) := d_{\mathcal{T}_n}(u_0, u_i), \quad 0 \le i \le 2n - 1,$$ + +with the convention that $u_{2n} = u_0$ and $C_n(2n) = 0$. We also extend $C_n$ by linear interpolation between integer times: for $0 \le s \le 2n$ + +$$C_n(s) = (1 - \{s\})C_n(\lfloor s \rfloor) + \{s\}C_n(\lfloor s \rfloor + 1),$$ + +where $\{s\} = s - \lfloor s \rfloor$ is the fractional part of $s$. Thus, the contour process $C_n$ is a non-negative path of length $2n$, starting and ending at 0, with increments of 1 between integer times. We will use the **rescaled contour process** of $\mathcal{T}_n$: + +$$C_{(n)}(t) = \frac{C_n(2nt)}{\sqrt{2n}}, \quad 0 \le t \le 1.$$ + +We define similarly the **rescaled label function** of $\mathcal{T}_n$: + +$$L_{(n)}(t) = \frac{L_n(2nt)}{n^{1/4}}, \quad 0 \le t \le 1,$$ + +where, similarly, we start by defining $L_n(i)$ as the label of $u_i$ for $i \in \{0, 1, \dots, 2n\}$, then interpolate between integer times. + +Finally, for a continuous, non-negative function $f : [0, 1] \to \mathbb{R}_+$ such that $f(0) = f(1) = 0$, for any $s, t \in [0, 1]$, we set + +$$\check{f}(s,t) = \inf\{f(u)|s \wedge t \le u \le s \vee t\}.$$ + +Then we have the following result: + +**Theorem 2.12.** [16] It holds that + +$$ (C_{(n)}, L_{(n)}) \xrightarrow[n]{(d)} (\wp, Z), \qquad (2.3) $$ + +*in distribution in $C([0, 1], \mathbb{R})^2$, where $\wp$ is a standard Brownian excursion, and, conditionally on $\wp$, $Z$ is a continuous, centered Gaussian process with covariance* + +$$\operatorname{Cov}(Z_s, Z_t) = \check{e}_{s,t}, s,t \in [0,1].$$ + +As this convergence will be crucial to ultimately prove the convergence of Eulerian triangulations to the Brownian map, to describe and analyse these triangulations, we will need to use their *oriented distances*, instead of the usual graph distance. \ No newline at end of file diff --git a/samples/texts/2952501/page_20.md b/samples/texts/2952501/page_20.md new file mode 100644 index 0000000000000000000000000000000000000000..5bae46a8f1091071a1b27cc852fa56f21ee9bf7a --- /dev/null +++ b/samples/texts/2952501/page_20.md @@ -0,0 +1,37 @@ +To simplify notation, let us note in this proof $\rho = 8$ and $\alpha = 4$. The property $B_r^\bullet(\bar{T}_n^{(p)}) = \Delta$ holds if and only if $\mathcal{T}_n^{(p)}$ is obtained from $\Delta$ by gluing to the top boundary³ an arbitrary triangulation with a semi-simple alternating boundary of length $2q$, and with $n-N$ black triangles, and if the distinguished vertex is chosen among the inner vertices of the glued triangulation. Thus: + +$$ \mathbb{P} (B_r^\bullet (\bar{T}_n^{(p)}) = \Delta) = \frac{B_{n-N,q}}{B_{n,p}} \cdot \frac{\#\text{inner vertices in glued triangulation}}{\#\text{inner vertices in total triangulation}}. \quad (5.3) $$ + +Therefore: + +$$ \lim_{n \to \infty} \mathbb{P} (B_r^\bullet (\bar{T}_n^{(p)}) = \Delta) = \frac{C(q)}{C(p)} \rho^{-N}. \quad (5.4) $$ + +As we have: + +$$ N = |\mathcal{M}(\Delta)| - p + \sum_{v \in \mathcal{F}^*} N(M_v) = \sum_{1 \le i \le q} |\tau_i| - p + \sum_{v \in \mathcal{F}^*} N(M_v) = q + \sum_{v \in \mathcal{F}^*} (c_v + N(M_v)) - p, $$ + +we get: + +$$ \lim_{n \to \infty} \mathbb{P} (B_r^\bullet (\bar{T}_n^{(p)}) = \Delta) = \frac{\rho^{-q} C(q)}{\rho^{-p} C(p)} \prod_{v \in \mathcal{F}^*} \rho^{-c_v} \rho^{-N(M_v)}. $$ + +Now, since $\sum_{v \in \mathcal{F}^*} (c_v - 1) = p - q$, we can multiply the right-hand side by $(\alpha\rho)^{p-q-\sum_{v \in \mathcal{F}^*} (c_v-1)}$, which yields: + +$$ \lim_{n \to \infty} \mathbb{P} (B_r^\bullet (\bar{T}_n^{(p)}) = \Delta) = \frac{\alpha^{-q} C(q)}{\alpha^{-p} C(p)} \prod_{v \in \mathcal{F}^*} \rho^{-1} \alpha^{-c_v+1} \rho^{-N(M_v)}, $$ + +that is: + +$$ \lim_{n \to \infty} \mathbb{P} (B_r^\bullet (\bar{T}_n^{(p)}) = \Delta) = \frac{\alpha^{-q} C(q)}{\alpha^{-p} C(p)} \prod_{v \in \mathcal{F}^*} \theta(c_v) \frac{\rho^{-N(M_v)}}{Z(c_v + 1)}, $$ + +for $\theta(k) = \rho^{-1}\alpha^{-k+1}Z(k+1)$. $\square$ + +Let us give a few properties of $\theta$ that will be useful in the sequel. These properties are obtained from the analytic combinatorial work in [9], rather than explicit enumeration as was the case for usual triangulations in [13]. + +First, the asymptotics of Z give: + +$$ \theta(k) \sim_k \frac{1}{2} \sqrt{\frac{3}{\pi}} k^{-5/2}. \quad (5.5) $$ + +Moreover, $\theta$ has the following generating function $g_\theta$: + +$$ g_\theta(x) = \sum_{k=0}^{\infty} \theta(k)x^k = 1 - \frac{3}{\left(\sqrt{\frac{4-x}{1-x}} + 1\right)^2 - 1} \quad \forall x \in [0, 1]. \quad (5.6) $$ + +³Note that, as $\Delta$ is rooted, we can fix an arbitrary rule to determine where to glue the root of the other triangulation. \ No newline at end of file diff --git a/samples/texts/2952501/page_21.md b/samples/texts/2952501/page_21.md new file mode 100644 index 0000000000000000000000000000000000000000..008f2996defb341942e8a9f0092e955977c93495 --- /dev/null +++ b/samples/texts/2952501/page_21.md @@ -0,0 +1,53 @@ +Convergence of Eulerian triangulations + +Indeed, the generating function of $\theta$ may be written, for $0 \le x < 1$: + +$$ +\begin{align*} +\sum_{k \ge 0} \theta(k) x^k &= \sum_{k \ge 0} \rho^{-1} \alpha^{-k+1} Z(k+1) x^k = \frac{\alpha}{\rho} \sum_{k \ge 0} \left(\frac{x}{\alpha}\right)^k Z(k+1) \\ +&= \frac{\alpha^2}{x\rho} \sum_{k \ge 1} \left(\frac{x}{\alpha}\right)^k Z(k) = \frac{\alpha^2}{x\rho} \left( \sum_{k \ge 0} \left(\frac{x}{\alpha}\right)^k Z(k) - Z(0) \right) \\ +&= \frac{2}{x} \left( \frac{1}{2} \left( 1 + \frac{7}{4}x - \frac{x^2}{2} + \sqrt{\left(\frac{x}{4}-1\right)(x-1)^3} \right) - 1 \right) \\ +&= \frac{-4 + 9x - 2x^2 + 2\sqrt{(x-4)(x-1)^3}}{x(4-x)} \\ +&= 1 - \frac{3}{\left(\sqrt{\frac{4-x}{1-x}} + 1\right)^2 - 1} = g_\theta(x). +\end{align*} +$$ + +It is straightforward to obtain from this that $\theta$ is a probability distribution with mean 1, so that, considered as the offspring distribution of a branching process, it is critical. + +Let $Y = (Y_r)_{r \ge 0}$ be a Galton-Watson process with offspring distribution $\theta$, and let us write $\mathcal{P}_k(\cdot)$ for the law of $Y$ given $Y_0 = k$, and $\mathcal{E}_k[\cdot]$ for the corresponding expectation. Then, for every $r \ge 1$, the generating function of $Y_r$ under $\mathcal{P}_1$ is the $r$-th iterate $g_\theta^{(r)}$ of $g_\theta$. It is easy to show that this iterate has a very nice expression for any positive integer $r$: + +$$ +\mathcal{E}_1[x^{Y_r}] = g_\theta^{(r)}(x) = 1 - \frac{3}{\left(\sqrt{\frac{4-x}{1-x}} + r\right)^2 - 1} \quad \forall x \in [0, 1]. \tag{5.7} +$$ + +Note that a similarly convenient expression for the *r*-th iterate of the generating func- +tion also exists for the offspring distributions associated to the skeleton decompositions +of usual triangulations [18, 13] and of quadrangulations [17, 21]. + +Using the transfer theorem (see Theorem VI.3 in [15]), we deduce from (5.7) that + +$$ +\mathcal{P}_1 (Y_r = k) \underset{k \to \infty}{\sim} \sqrt{\frac{3}{\pi}} \frac{r}{2} k^{-5/2}. \qquad (5.8) +$$ + +Let us denote by $\mathbb{F}_{p,q,r}$ the set of $(p,q,r)$-admissible forests. We also define the set $\mathbb{F}'_{p,q,r}$ of pointed forests satisfying the same conditions as $(p,q,r)$-admissible forests, except that the tree with a distinguished vertex is not necessarily $\tau_1$, and the set $\mathbb{F}''_{p,q,r}$ of forests which satisfy the same conditions but do not have a distinguished vertex. + +We now prove that the “skeleton part” of (5.1) defines a probability measure on $\mathbb{F}_{p,r} = \cup_{q \ge 1} \mathbb{F}_{p,q,r}$, similarly to [13, Lemma 3]: + +**Lemma 5.4.** For every $p \ge 1$ and $r \ge 1$, + +$$ +\sum_{q=1}^{\infty} \sum_{\mathcal{F} \in \mathbb{F}_{p,q,r}} \frac{4^{-q} C(q)}{4^{-p} C(p)} \prod_{v \in \mathcal{F}^*} \theta(c_v) = 1. \quad (5.9) +$$ + +Proof. Like in the proof of [13, Lemma 3], (5.9) amounts to + +$$ +\sum_{q=1}^{\infty} \frac{h(q)}{h(p)} P_q (Y_r = p) = 1, \qquad (5.10) +$$ + +with + +$$ +h(k) = 2\sqrt{\pi} \frac{4^{-k} C(k)}{k}. \tag{5.11} +$$ \ No newline at end of file diff --git a/samples/texts/2952501/page_22.md b/samples/texts/2952501/page_22.md new file mode 100644 index 0000000000000000000000000000000000000000..c2da588ca9688347af80fcebdad70314303e2e2d --- /dev/null +++ b/samples/texts/2952501/page_22.md @@ -0,0 +1,43 @@ +Now, (5.10) is equivalent to + +$$ +\sum_{q=1}^{\infty} h(q) P_q (Y_r = p) = h(p) \tag{5.12} +$$ + +or, in other words, to the fact that *h* is an infinite stationary measure for *Y*. + +Let $\Pi$ be the generating function of the sequence $(h(k))_{k \ge 1}$: + +$$ +\Pi(x) := \sum_{k=1}^{\infty} h(k) x^k = \sum_{k=1}^{\infty} \frac{1}{k} C(k) \left(\frac{x}{4}\right)^k . +$$ + +Contrary to the case of usual triangulations, we do not have an explicit expression for $h$, but, by integrating (4.3), we obtain one for $\Pi$: + +$$ +\Pi(x) = \sqrt{\frac{4-x}{1-x}} - 2 \quad \forall 0 < x < 1 +$$ + +To prove that $h$ is an infinite stationary measure for $Y$, it is enough to check that $\Pi(g_\theta(x)) - \Pi(g_\theta(0)) = \Pi(x)$ for every $x \in [0, 1]$, which follows from the explicit formulas for $g_\theta$ and $\Pi$. $\square$ + +With Lemma 5.4, we can define a probability measure $\mathbf{P}_{p,r}$ on $\mathbb{F}_{p,r}$ by setting, for any $\mathcal{F} \in \mathbb{F}_{p,q,r}$, + +$$ +\mathbf{P}_{p,r}(\mathcal{F}) := \frac{4^{-q}C(q)}{4^{-p}C(p)} \prod_{v \in \mathcal{F}^*} \theta(c_v). \quad (5.13) +$$ + +Let us note $\mathbb{C}_{p,r}$ the set of Eulerian triangulations of the cylinder of height $r$ and bottom boundary length $2p$. We can define a probability measure $\mathbb{P}_{p,r}$ on $\mathbb{C}_{p,r}$, by first setting the skeleton to be distributed according to $\mathbb{P}_{p,r}$, then, conditionally on the skeleton, filling the slots by independent Boltzmann triangulations (whose boundary lengths are prescribed by the skeleton). Thus, Lemma 5.3 amounts to stating that, if $\Delta \in \mathbb{C}_{p,r}$, + +$$ +\lim_{n \to \infty} \mathbb{P} \left( B_r^\bullet \left( \bar{T}_n^{(p)} \right) = \Delta \right) = \mathbb{P}_{p,r}(\Delta). \quad (5.14) +$$ + +In other words, the law of $B_r^\bullet(\bar{T}_n^{(p)})$ converges weakly to $\mathbb{P}_{p,r}$ as $n \to \infty$. + +Note that the expression (5.13) implies that, if a random cylinder triangulation $A$ is distributed as $\mathbb{P}_{p,r}$, then, for all $1 \le s \le r$, its hull $B_s^\bullet(A)$ will be distributed as $\mathbb{P}_{p,s}$, or, in other words, the laws ($\mathbb{P}_{p,r}$)$_{r \ge 1}$ are consistent. This implies that the sequence of random maps $(\mathcal{T}_n^{(p)})_n$ has a local distributional limit. To express this result more precisely, we need to generalize the notion of hulls to some infinite maps. First, for any infinite planar Eulerian triangulation $A$ with a boundary, we can define its ball $B_r(A)$ like in the finite case. Then, if $A$ has a unique end, only one connected component of $A \setminus B_r(A)$ is infinite, so that we can fill all the finite holes, to get the hull $B_r^\bullet(A)$. + +We then have the following result: + +**Proposition 5.5.** For any integer $p \ge 1$, the sequence of random maps $(\mathcal{T}_n^{(p)})_n$ converges in distribution, in the sense of local limits of rooted maps, to an infinite map that we call the **uniform infinite Eulerian triangulation of the 2p-gon**, and that we denote by $\mathcal{T}_\infty^{(p)}$. It is a random infinite Eulerian triangulation of the plane, with an alternating, semi-simple boundary of length $2p$, that has a unique end almost surely, and such that $B_r^\bullet(\mathcal{T}_\infty^{(p)})$ has law $\mathbb{P}_{p,r}$, for every integer $r \ge 1$. + +For $p = 1$, we can perform the transformation described in Figure 10, which yields a random infinite planar Eulerian triangulation, which we denote by $\mathcal{T}_\infty$. This random \ No newline at end of file diff --git a/samples/texts/2952501/page_23.md b/samples/texts/2952501/page_23.md new file mode 100644 index 0000000000000000000000000000000000000000..1f689f54138aaac6be86f04a901f35ccca34aca6 --- /dev/null +++ b/samples/texts/2952501/page_23.md @@ -0,0 +1,53 @@ +these has already been proven in [1]. However, this would necessitate to treat the +distances on an Eulerian triangulation as a local modification of the distances on the +corresponding bipartite map, and thus use a layer decomposition of bipartite maps. As +this has not been achieved yet, this route is a priori not easier than the one undertaken +here, which has the advantage of uncovering a lot of properties of Eulerian triangulations. +However, achieving a layer decomposition of bipartite maps would be interesting in +itself. + +**1.2 Outline** + +In the whole paper, $\mathbf{c}_0$ refers to the constant $\mathbf{c}_0 \in [2/3, 1]$ appearing in Proposition 8.1 below. The main result of this paper is Theorem 3.1. As the full statement of this theorem necessitates a bit of notation, we postpone it to Section 3. We can however already give a much weaker version of it: + +**Theorem 1.1.** Let $\mathcal{T}_n$ be a uniform random rooted Eulerian planar triangulation with $n$ black faces, equipped with its usual graph distance $d_n$, and let $V(\mathcal{T}_n)$ denote its vertex set. Let $(\mathbf{m}_\infty, D^*)$ be the Brownian map. The following convergence holds + +$$ +n^{-1/4} \cdot (V(\mathcal{T}_n), d_n) \xrightarrow[n \to \infty]{(d)} \mathbf{c}_0 \cdot (\mathbf{m}_\infty, D^*), +$$ + +for the Gromov-Hausdorff distance on the space of isometry classes of compact metric +spaces. + +We give a detailed definition of the Brownian map in Section 3, and we refer to [11] +for a precise definition of the Gromov-Hausdorff distance. + +We will see how Theorem 1.1 can be obtained from the following result: + +**Theorem 1.2.** Let $\mathcal{T}_n$ be a uniform random rooted Eulerian planar triangulation with $n$ black faces, and let $V(\mathcal{T}_n)$ be its vertex set. We denote by $d_n$ its usual graph distance, and by $\vec{d}_n$ its canonical oriented pseudo-distance. For every $\varepsilon > 0$, we have + +$$ +\mathrm{P}\left(\sup_{(x,y) \in V(\mathcal{T}_n)} |d_n(x,y) - \mathbf{c}_0 \vec{d}_n(x,y)| > \varepsilon n^{1/4}\right) \xrightarrow{n \to \infty} 0. +$$ + +After giving a precise description of the structure of Eulerian triangulations endowed +with their oriented pseudo-distance in Section 2, in Section 3 we will give the complete +statement of Theorem 3.1, and explain how to prove it using Theorem 1.2. Sections 4 +to 8 are then devoted to proving Theorem 1.2. + +Let us sketch the different steps of this proof. After some technical statements +in Section 4, pertaining either to asymptotic estimates of $\vec{d}$, or to asymptotics of the +enumeration of Eulerian triangulations with a boundary, we detail in Section 5 the +decomposition of finite rooted planar Eulerian triangulations (possibly with a boundary) +into *layers*, determined by the oriented distance from the root. This decomposition +makes it possible to describe the random triangulation $\mathcal{T}_n$, defined like in Theorem 1.1, in +terms of a branching process whose generations are associated to the layers of $\mathcal{T}_n$. This +nice description of $\mathcal{T}_n$ allows us to take the limit $n \to \infty$, to define the Uniform Infinite +Planar Eulerian triangulation, $\mathcal{T}_\infty$, that is naturally endowed with a decomposition into +an infinite number of layers. Now, in Section 6, we take a local limit of $\mathcal{T}_\infty$ where we view +these layers "from infinity", which yields the Lower Half-Planar Eulerian Triangulation $\mathcal{L}$. +In Section 8, we explain how the construction of this half-plane model makes it possible +to obtain Theorem 1.2. First, the layers of $\mathcal{L}$ are i.i.d., which makes it straightforward +to apply an ergodic subbadditivity argument to the graph distance $d$ between the root +of $\mathcal{L}$ and the $n$-th layer of $\mathcal{L}$. Then, we detail how this result can carry over to finite +Eulerian triangulations, first for the graph distance between the root and a random \ No newline at end of file diff --git a/samples/texts/2952501/page_24.md b/samples/texts/2952501/page_24.md new file mode 100644 index 0000000000000000000000000000000000000000..73451027c4b70a535ac2bbec047c9cc7362eb09a --- /dev/null +++ b/samples/texts/2952501/page_24.md @@ -0,0 +1,41 @@ +infinite map is the local limit of uniform rooted planar Eulerian triangulations with $n$ black faces when $n \to \infty$, therefore we call it the **Uniform Infinite Planar Eulerian Triangulation** (UIPET). + +The UIPET is the equivalent of well-known models of random infinite planar maps such as the UIPT or the UIPQ (see [6, 17]), in the case of Eulerian triangulations. Note that this present work gives the first construction of the UIPET. + +Let $L_r^{(p)}$ be the length of the top cycle of $B_r^\bullet(\mathcal{T}_\infty^{(p)})$. When $p=1$, we write $L_r$ for $L_r^{(1)}$ for simplicity. + +Let us first note that $\mathcal{T}_\infty^{(p)}$ exhibits a spatial Markov property. Let $r, s$ be integers with $1 \le r < s$, and $\Delta \in \mathbb{C}_{p,s}$. Let $2q$ be the length of the boundary $\partial_r\Delta$. We can obtain $\Delta$ by gluing a triangulation $\Delta'' \in \mathbb{C}_{q,s-r}$ on top of a triangulation $\Delta' \in \mathbb{C}_{p,r}$, whose top boundary has length $q$. From the explicit formula of (5.13), we get + +$$ \mathbb{P}_{p,s}(\Delta) = \mathbb{P}_{p,r}(\Delta') \cdot \mathbb{P}_{q,s-r}(\Delta''). \quad (5.15) $$ + +Therefore, conditionally on $\{L_r^{(p)} = q\}$, $B_s^\bullet(\mathcal{T}_\infty^{(p)}) \setminus B_r^\bullet(\mathcal{T}_\infty^{(p)})$ follows $\mathbb{P}_{q,s-r}$, and is independent of $B_r^\bullet(\mathcal{T}_\infty^{(p)})$. By letting $s \to \infty$, we obtain that, conditionally on $\{L_r^{(p)} = q\}$, the triangulation $\mathcal{T}_\infty^{(p)} \setminus B_r^\bullet(\mathcal{T}_\infty^{(p)})$ is distributed as $\mathcal{T}_\infty^{(q)}$ and is independent of $\mathcal{T}_\infty^{(p)}$. + +We now give a technical but useful result on the law of $L_r$, which is the equivalent in our case of [13, Lemma 4]. + +**Lemma 5.6.** There exists a constant $C_0 > 0$ such that for any $\alpha \ge 0$, and for any integers $r, p \ge 1$, + +$$ \mathbb{P}(L_r = p) \le \frac{C_0}{r^2} \quad (5.16) $$ + +and + +$$ \mathbb{P}\left(L_r \ge \alpha r^2\right) \le C_0 e^{-\alpha/4}. \quad (5.17) $$ + +Let us fix some notation before getting to the proof of Lemma 5.6. For $1 \le r < s$, let $\mathcal{F}_{r,s}^{(1)}$ be the skeleton of $B_s^\bullet(\mathcal{T}_\infty^{(1)}) \setminus B_r^\bullet(\mathcal{T}_\infty^{(1)})$. We let $\tilde{\mathcal{F}}_{r,s}^{(1)}$ by the non-pointed forest obtained by a uniform cyclic permutation of $\mathcal{F}_{r,s}^{(1)}$, and by forgetting the distinguished vertex. Thus, on the event $\{L_r = p\} \cap \{L_s = q\}$, $\tilde{\mathcal{F}}_{r,s}^{(1)}$ is a random element of $\mathbb{F}_{p,q,s-r}^{(2)}$. + +*Proof.* These bounds are obtained very similarly to those of Lemma 4 in [13], with the (small) additional difficulty that in our case we do not have an explicit expression for $h$, but only asymptotics. We give the full argument here as it consists in a short but rather involved computation. + +First observe that + +$$ \mathbb{P}(L_r = p) = \sum_{\mathcal{F} \in \mathbb{F}_{1,p,r}''} \mathbb{P}(\tilde{\mathcal{F}}_{0,r}^{(1)} = \mathcal{F}) = \sum_{\mathcal{F} \in \mathbb{F}_{1,p,r}''} \frac{h(p)}{h(1)} \prod_{v \in \mathcal{F}^*} \theta(c_v). $$ + +Thus + +$$ \mathbb{P}(L_r = p) = \frac{h(p)}{h(1)} \mathcal{P}_p(Y_r = 1). $$ + +From the definition of $h$ and the asymptotics of $\mathcal{C}(p)$, there exists a constant $C_1$ such that, for every $p \ge 1$, + +$$ h(p) \le \frac{C_1}{\sqrt{p}}. $$ + +Moreover, from (5.7), we have + +$$ \mathcal{P}_1(Y_r = 0) = 1 - \frac{3}{(r+2)^2 - 1}, \quad (5.18) $$ \ No newline at end of file diff --git a/samples/texts/2952501/page_25.md b/samples/texts/2952501/page_25.md new file mode 100644 index 0000000000000000000000000000000000000000..cd2f31fcd397ec00c20b601181eb2b53c5683433 --- /dev/null +++ b/samples/texts/2952501/page_25.md @@ -0,0 +1,46 @@ +hence + +$$ +\begin{align*} +\mathcal{P}_p (Y_r = 1) &= \lim_{x \downarrow 0} x^{-1} (\mathcal{E}_p [x^{Y_r}] - \mathcal{P}_p (Y_r = 0)) \\ +&= \lim_{x \downarrow 0} x^{-1} \left( \left( 1 - \frac{3}{\left(\sqrt{\frac{4-x}{1-x}} + r\right)^2 - 1} \right)^p - \left( 1 - \frac{3}{(r+2)^2 - 1} \right)^p \right) \\ +&= \frac{9p(r+2)}{2((r+2)^2-1)((r+2)^2-4)} \left(1 - \frac{3}{(r+2)^2-1}\right)^p. +\end{align*} +$$ + +Therefore, for some constant $C_3 > 0$, + +$$ +\mathbb{P}(L_r = p) \leq \frac{C_1}{h(1)} \sqrt{p} \frac{9(r+2)}{((r+2)^2-1)((r+2)^2-4)} \left(1 - \frac{3}{(r+2)^2-1}\right)^{p-1} \leq \frac{C_3}{r^2} \sqrt{\frac{p}{r^2}} e^{-3p/r^2}. +$$ + +The bound (5.16) immediately follows. As for (5.17), since the function $x \mapsto \sqrt{x}e^{-3x}$ is decreasing for $x \ge 1/6$, we have, for $\alpha \ge 1/6$, for some constant $C_4 > 0$, + +$$ +\mathbb{P}(L_r > \alpha r^2) \leq \sum_{p=\alpha r^2+1}^{\infty} \frac{C_3}{r^2} \sqrt{\frac{p}{r^2}} e^{-3p/r^2} \leq \frac{C_3}{r^2} \int_{\alpha r^2}^{\infty} \sqrt{\frac{x}{r^2}} e^{-3x/r^2} dx \leq C_4 e^{-\alpha/4}. \quad \square +$$ + +We now fix a positive constant $a \in (0, 1)$. For every integer $r \ge 1$, let $N_r^{(a)}$ be uniformly random in $\{|ar^2|+1, \dots, [a^{-1}r^2]\}$. We also consider a sequence $\tau_1, \tau_2, \dots$ of independent Galton-Watson trees with offspring distribution $\theta$, independent of $N_r^{(a)}$. For every integer $j \ge 0$, we write $[\tau_i]_j$ for the tree $\tau_i$ truncated at generation $j$. + +Using the same arguments that yield Proposition 5 from Lemma 4 in [13], the above lemma implies the following bound: + +**Proposition 5.7.** There exists a constant $C_1$, which only depends on $a$, such that, for every sufficiently large integer $r$, for every choice of $s \in \{r+1, r+2, \dots\}$, for every choice of integers $p$ and $q$ with $ar^2 < p, q \le a^{-1}r^2$, for every forest $\mathcal{F} \in F''_{p,q,s-r}$, + +$$ +\mathbb{P}(\tilde{\mathcal{F}}_{r,s}^{(1)} = \mathcal{F}) \le C_1 \mathbb{P}(([\tau_1]_{s-r}, \dots, [\tau_{N_r^{(a)}}]_{s-r}) = \mathcal{F}). \quad (5.19) +$$ + +**5.3 Leftmost mirror geodesics** + +We now define a type of paths in Eulerian cylinder triangulations that will be useful +in the sequel. + +Let $\Delta$ be an Eulerian cylinder triangulation of height $r \ge 1$. Let $x$ be a type-$j$ vertex of $\partial_j \Delta$, with $1 \le j \le r$. We define the **leftmost mirror geodesic** from $x$ to the bottom cycle in the following way. Enumerate in clockwise order around $x$ all the half-edges incident to it, starting from the half-edge of $\partial_j \Delta$ that is to the right of $x$. The first edge on the leftmost mirror geodesic starting from $x$ is the last edge connecting $x$ to $\partial_{j-1} \Delta$ arising in this order. The path is then continued by induction (see Figure 14). Note that, taken in the reverse order, this path is an oriented geodesic, hence the name *mirror geodesic*. (Such a precision is not necessary in [13], that deals with proper, symmetric distances.) + +The coalescence of leftmost geodesics from distinct vertices can be characterized +by the skeleton of $\Delta$. Indeed, let $u, v$ be two distinct type-$r$ vertices of $\partial^*\Delta$. Let $\mathcal{F}$ be +the skeleton of $\Delta$, $\mathcal{F}'$ the subforest of $\mathcal{F}$ consisting of the trees rooted between $u$ and $v$ +left-to-right in $\partial^*\Delta$, and $\mathcal{F}''$ be the rest of the trees in $\mathcal{F}$. Then, for any $k \in \{1, 2, \dots, r\}$, +the leftmost mirror geodesics from $u$ and $v$ merge before step $k$ (possibly exactly at step +$k$) if and only if at least one of the two forests $\mathcal{F}'$ and $\mathcal{F}''$ have height strictly smaller +than $k$. \ No newline at end of file diff --git a/samples/texts/2952501/page_26.md b/samples/texts/2952501/page_26.md new file mode 100644 index 0000000000000000000000000000000000000000..5f553d7d67f107245414a93b1f2dddaf0caa1038 --- /dev/null +++ b/samples/texts/2952501/page_26.md @@ -0,0 +1,13 @@ +Figure 14: Some leftmost mirror geodesics (depicted in blue) in a portion of an Eulerian cylinder triangulation. + +# 6 The Lower Half-Plane Eulerian Triangulation + +We now construct a triangulation of the lower half-plane $\mathbb{R} \times \mathbb{R}_{-}$ that will be crucial to prove Theorem 1.2, and that also is an object of interest in itself. Note that this construction is very similar that of the LHPT in [13, Section 3.2]. + +We start with a doubly infinite sequence $(\mathcal{T}_i)_{i \in \mathbb{Z}}$ of independent Galton-Watson trees with offspring distribution $\theta$. They are embedded in the lower half-plane so that, for every $i \in \mathbb{Z}$, the root of $\mathcal{T}_i$ is $(\frac{1}{2} + i, 0)$, and such that the collection of all vertices of all the $\mathcal{T}_i$ is $(\frac{1}{2} + \mathbb{Z}) \times \mathbb{Z}_{\le 0}$, with vertices at height $k$ being of the form $(\frac{1}{2} + i, -k)$. We also assume that the embedding is such that the collection of vertices of the $\mathcal{T}_i$, for $i \ge 0$, is $(\frac{1}{2} + \mathbb{Z}_{\ge 0}) \times \mathbb{Z}_{\le 0}$ (see Figure 15). + +We can now build the triangulation itself. We start with the “distinguished” modules, which will play the role of skeleton modules for our infinite triangulation. They are naturally associated with the vertices of the infinite collection of trees in the following way. To each vertex $(\frac{1}{2} + i, j)$ in one of the trees, we associate a module whose type $n+1$ vertices are $(i,j)$ and $(i+1,j)$. The type $n$ vertex is $(k, j-1)$, where $k$ is the minimal integer such that $(\frac{1}{2} + k, j-1)$ is the child of $(\frac{1}{2} + i', j)$, for some $i' > i$. The last vertex, of type $n+2$, is set to be $(\frac{1}{2} + i, j + \varepsilon)$, for an arbitrary $0 < \varepsilon < 1$. As for the (outer) edges of these skeleton modules, we draw them such that they are all distinct, and do not cross. Having completely determined the configuration of the skeleton edges from the infinite collection of trees, we fill in the slots bounded by these modules, with independent Boltzmann Eulerian triangulation of appropriate perimeters. (Note that each point of the form $(i,j)$, with $j \ge 1$, is at the top of a slot of perimeter $2(c_{i,j} + 1)$, where $c_{i,j}$ is the number of children of $(\frac{1}{2} + i, j)$ in the infinite collection of trees.) + +We obtain an Eulerian triangulation of the lower-half plane, which we will note $\mathcal{L}$ and call the **Lower Half-Plane Eulerian Triangulation** (LHPET). It is rooted at the edge from $(0,0)$ to $(\frac{1}{2}, \varepsilon)$. + +We will denote by $\mathcal{L}_{[0,r]}$ the infinite rooted planar map obtained by keeping only the first $r$ layers of $\mathcal{L}$ (having the skeleton modules at level $r$ as ghost modules), and denote by $\mathcal{L}_r$ the lower boundary of $\mathcal{L}_{[0,r]}$. For integers $0 \le m < n$, we also define $\mathcal{L}_{[m,n]}$, to be \ No newline at end of file diff --git a/samples/texts/2952501/page_27.md b/samples/texts/2952501/page_27.md new file mode 100644 index 0000000000000000000000000000000000000000..5f6c7839ca0143d1786813553243b99bf6687e9e --- /dev/null +++ b/samples/texts/2952501/page_27.md @@ -0,0 +1,28 @@ +the map obtained by keeping only the layers of $\mathcal{L}$ that lie between the levels $m$ and $n$ (the skeleton modules at level $m$ making up the top boundary of $\mathcal{L}_{[m,n]}$, and the ones at level $n$ being its bottom ghost modules). + +Figure 15: Construction of the LHPET. + +While we will not use this result in the sequel, note that $\mathcal{L}$ is the local limit of $\mathcal{T}_\infty^{(p)}$ +"seen from infinity". This statement is made more precise in the following proposition: + +**Proposition 6.1.** Set $p \ge 1$, and for every $r \ge 1$, define $\tilde{B}_s^\bullet(\mathcal{T}_\infty^{(p)})$ as the hull $B_s^\bullet(\mathcal{T}_\infty^{(p)})$ re-rooted at an edge uniform on those of $\partial_s(\mathcal{T}_\infty^{(p)})$ that are oriented so that the top face is lying on their left. Then + +$$ +\tilde{B}_s^\bullet (\mathcal{T}_\infty^{(p)}) \xrightarrow[s \to \infty]{(d)} \mathcal{L} +$$ + +in the sense of local limits of rooted planar maps. + +An equivalent result was stated for usual triangulations in [13, Proposition 7], but its +proof was not detailed, since it is similar to the proof of the equivalent convergence to +the *Upper Half-Plane Triangulation*. We give a proof of our result both for the sake of +completeness, and because it involves nonetheless a few arguments that are different +from the ones for the upper half-plane models. + +*Proof.* Recall that, for an Eulerian triangulation $A$ (possibly with a boundary) and an integer $r \ge 1$, we denote by $B_r(A)$ the ball of radius $r$ of $A$, that is, the union of all edges and faces of $A$ incident to a vertex at (oriented) distance strictly less than $r$ from the root. Proving the proposition amounts to showing that, for every $r \ge 1$, for every rooted planar map $A$, + +$$ +\mathbb{P}\left(B_r\left(\tilde{B}_s^{\bullet}\left(\mathcal{T}_{\infty}^{(p)}\right)\right) = A\right) \xrightarrow{s \to \infty} \mathbb{P}(B_r(\mathcal{L}) = A). \quad (6.1) +$$ + +To obtain this convergence, we will need a bit of additional notation. We fix $r \ge 1$, and note $[\mathcal{T}]_r$ for the tree $\mathcal{T}$ truncated at height $r$, and similarly for a forest. For any $s \ge 1$, we write $\mathcal{F}_{0,s}^{(p)} = (\mathcal{T}_0^{(p)}, \mathcal{T}_1^{(p)}, \dots, \mathcal{T}_{L_s^{(p)}}^{(p)})$ for the skeleton of $\tilde{B}_s^{\bullet}(\mathcal{T}_{\infty}^{(p)})$. Let us fix $k \ge 1$. For any $q \ge 1$, for any forest $\mathcal{F} = (\sigma_0, \dots, \sigma_{l-1}) \in \mathbb{F}_{q,l,r}$ with $l \ge 2k+1$, we write $\Phi_k(\mathcal{F}) = (\sigma_{i-k}, \dots, \sigma_{i-1}, \sigma_i, \dots, \sigma_{i+k})$, where $i$ is a uniform index on $0, \dots, l-1$, and the indices for the $\sigma_j$ are extended to $\mathbb{Z}$ by periodicity. \ No newline at end of file diff --git a/samples/texts/2952501/page_28.md b/samples/texts/2952501/page_28.md new file mode 100644 index 0000000000000000000000000000000000000000..dc76c72d00fda82134542deb2aba6e9345e215a8 --- /dev/null +++ b/samples/texts/2952501/page_28.md @@ -0,0 +1,42 @@ +We will prove that, for every collection $\mathcal{F}_k = (\tau_{-k}, \dots, \tau_0, \dots, \tau_k)$ of $2k+1$ plane trees of maximal height $r$, + +$$ P(\{\Phi_k([\mathcal{F}_{0,s}]_r) = \mathcal{F}_k\} \cap \{L_s^{(p)} \ge 2k+1\}) \xrightarrow{s \to \infty} P([\mathcal{T}_{-k}]_r, \dots, [\mathcal{T}_0]_r, \dots, [\mathcal{T}_k]_r) = \mathcal{F}_k). \quad (6.2) $$ + +If $k$ is large enough, we can find a set $\mathbf{F}_k$ of forests such that the probability of the event + +$$ ([\mathcal{T}_{-k}]_r, \dots, [\mathcal{T}_0]_r, \dots, [\mathcal{T}_k]_r) \in \mathbf{F}_k $$ + +is close to 1, and such that, on that event, the ball $B_r(\mathcal{L})$ is a deterministic function of the truncated trees $[\mathcal{T}_{-k}]_r, \dots, [\mathcal{T}_0]_r, \dots, [\mathcal{T}_k]_r$ and of the triangulations with a boundary filling in the slots associated with the vertices of these trees. (Note that we need $k$ to be large, so that the $(2k+1)$ central trees of the skeleton of $\mathcal{L}$ and the associated slots are enough to cover the ball $B_r(\mathcal{L})$, not only vertically, which is a given, but also horizontally.) Likewise, on the event $\{\Phi_k([\mathcal{F}_{0,s}]_r) \in \mathbf{F}_k\} \cap \{L_s^{(p)} \ge 2k+1\}$, the ball $B_r(\tilde{\mathcal{B}}_s^* (\mathcal{T}_\infty^{(p)}))$ is given by the same deterministic function of the trees in $\Phi_k([\mathcal{F}_{0,s}]_r)$ and of the associated triangulations with a boundary. + +Moreover, we claim that, for every fixed $p \ge 1$ and $j \ge 1$, + +$$ P(L_s^{(p)} = j) \xrightarrow{s \to \infty} 0. \qquad (6.3) $$ + +Indeed, we can write + +$$ P(L_s^{(p)} = j) = \frac{h(j)}{h(p)} P_j(Y_s = p), $$ + +and, from $E_j[x^{Y_s}] = (g_\theta^{(s)}(x))^j$, we get that $P_j(Y_s = p) \xrightarrow{s \to \infty} 0$. + +Thus, the desired convergence of (6.1) will follow from (6.2) and (6.3). + +It remains to prove (6.2). Let us fix $\mathcal{F}_k$ as above. From the definition of the $\mathcal{T}_i$, we have + +$$ P([\mathcal{T}_{-k}]_r, \dots, [\mathcal{T}_0]_r, \dots, [\mathcal{T}_k]_r) = \mathcal{F}_k) = \prod_{v \in (\tau_{-k}, \dots, \tau_k)^*} \theta(c_v), \quad (6.4) $$ + +where, as before, for a forest $\mathcal{F}$, $\mathcal{F}^*$ denotes the set of vertices in $\mathcal{F}$ that are not at the maximal height, and, for such a vertex $v$, $c_v$ is its number of children. + +Now, using the definition of the law $P_{p,s}$ of $B_s^* (\mathcal{T}_\infty^{(p)})$, the left-hand side of (6.2) is equal to + +$$ +\begin{aligned} +& \sum_{l=2k+1}^{\infty} \sum_{\mathcal{F} \in \mathcal{F}_{p,l,s}, \Phi_k(\mathcal{F}) = \mathcal{F}_k} \frac{4^{-l} C(l)}{4^{-p} C(p)} \prod_{v \in \mathcal{F}^*} \theta(c_v) \\ +& = \left( \prod_{v \in (\tau_{-k}, \dots, \tau_k)^*} \theta(c_v) \right) \cdot \left( \sum_{l=2k+1}^{\infty} \frac{4^{-l} C(l)}{4^{-p} C(p)} \prod_{\substack{v \in (\sigma_0, \dots, \sigma_{l-2k-1})^* \\ \#(\sigma_0(s)+\cdots+\#(\sigma_{l-2k-1}(s))+\cdots+\#(\tilde{\sigma}_1(s-r))+\cdots+\#(\tilde{\sigma}_{m_k}(s-r))=p}}} \theta(c_v) \right), +\end{aligned} +$$ + +where $m_k$ is the number of vertices at generation $r$ in $\mathcal{F}_k$, while $\sigma_0, \dots, \sigma_{l-2k-1}$ stand for the trees (of maximal height $s$) not selected in $\mathcal{F}_k$, and $\tilde{\sigma}_1, \dots, \tilde{\sigma}_{m_k}$ stand for the trees (of maximal height $s-r$) obtained after truncation of the selected trees. + +Let us denote by $A_s$ the second term of the second line of the previous equation. To conclude the proof, it suffices to show that + +$$ \liminf_s A_s \ge 1. \qquad (6.5) $$ \ No newline at end of file diff --git a/samples/texts/2952501/page_29.md b/samples/texts/2952501/page_29.md new file mode 100644 index 0000000000000000000000000000000000000000..c84936cf7984fc6d80af791e0dc32a7e3d8abf12 --- /dev/null +++ b/samples/texts/2952501/page_29.md @@ -0,0 +1,55 @@ +Indeed, in that case the liminf of the quantities in the left-hand side of (6.2) are greater than or equal to the right-hand side, for any choice of the forest $\mathcal{F}_k$. As the sum of the quantities on the right-hand side of (6.2) over these choices is equal to 1, necessarily the desired convergence holds. + +Let us thus show (6.5). Set $\varphi(l) := 4^{-l}C(l)$. We have + +$$A_s = \sum_{l=2k+1}^{\infty} \frac{\varphi(l)}{\varphi(p)} \sum_{q=0}^{p} P_{l-(2k+1)}(Y_s=q) P_{m_k}(Y_{s-r}=p-q).$$ + +First, as $\theta$ is a critical offspring distribution, we get from [26] that, for any $q \ge 0$, + +$$\frac{P_{m_k}(Y_{s-r} = p-q)}{P_{m_k}(Y_s = p-q)} \xrightarrow{s \to \infty} 1.$$ + +Thus, for any $l \ge 2k+1$, for every $\varepsilon > 0$, for any sufficiently large $s$, + +$$ +\begin{align*} +\sum_{q=0}^{p} P_{l-(2k+1)}(Y_s=q) P_{m_k}(Y_{s-r}=p-q) &\ge (1-\varepsilon) \sum_{q=0}^{p} P_{l-(2k+1)}(Y_s=q) P_{m_k}(Y_s=p-q) \\ +&\ge (1-\varepsilon) P_{l-(2k+1)+m_k}(Y_s=p). +\end{align*} +$$ + +This implies that: + +$$A_s \ge (1 - \varepsilon) \sum_{l=2k+1}^{\infty} \frac{\varphi(l)}{\varphi(p)} P_{l-(2k+1)+m_k} (Y_s = p).$$ + +Now, from the asymptotics of $C(l)$, we have that, for some $l_0 \ge 0$, for any $l \ge l_0$, we have + +$$\varphi(l) \ge (1 - \varepsilon)\varphi(l - (2k + 1) + m_k),$$ + +so that, + +$$A_s \ge (1 - \varepsilon)^2 \sum_{l=m_k \lor l_0}^{\infty} \frac{\varphi(l)}{\varphi(p)} P_l(Y_s = p).$$ + +Recall that $\varphi(l) = lh(l)$, which yields: + +$$ +\begin{align*} +A_s &\ge (1-\epsilon)^2 \sum_{l=m_k \lor l_0 \lor p}^{\infty} \frac{h(l)}{h(p)} P_l(Y_s=p) \\ +&= (1-\epsilon)^2 \left( 1 - \sum_{l=0}^{m_k \lor l_0 \lor p-1} \frac{h(l)}{h(p)} P_l(Y_s=p) \right), +\end{align*} +$$ + +the last equality stemming from (5.12). + +Finally, we use once again the fact that, for any fixed $l$, + +$$P_l (Y_s = p) \xrightarrow{s \to \infty} 0,$$ + +to get that, for any $\epsilon > 0$, + +$$\liminf_s A_s \ge (1 - \epsilon)^2.$$ + +As $\epsilon$ was completely arbitrary in the above chain of arguments, we get that + +$$\liminf_s A_s \ge 1.$$ + +This completes the proof of the proposition. $\square$ \ No newline at end of file diff --git a/samples/texts/2952501/page_3.md b/samples/texts/2952501/page_3.md new file mode 100644 index 0000000000000000000000000000000000000000..7e9e62df46638179a05b579f1edc4b45b83baa0d --- /dev/null +++ b/samples/texts/2952501/page_3.md @@ -0,0 +1,9 @@ +## 2.4 Structure for oriented distance + +Let us consider a rooted Eulerian triangulation $A$, equipped with its canonical orientation and oriented geodesic distance $\bar{d}$. For each type-$n$ black face $f$ of $A$, there is exactly one white face $f'$ that shares its $n+1 \to n-1$ edge. We call the union of $f$ and $f'$ a **type-$n$ module**. Now, imagine that for each type-$n$ module of $A$, we trace the “diagonal” linking its two type-$n$ vertices, and direct it from the black triangle to the white one (see Figure 5). We will call this, directing the module **left-to-right**. + +Figure 5: The union of type-$n$ module diagonals can be decomposed into a set of simple closed curves by pairing, at each vertex of type $n$, each ingoing diagonal with the next one clockwise, which is necessarily outgoing. + +We now explain how to describe the union of the diagonals of type-$n$ modules as a set of simple closed curves. First note that, by construction, this union of diagonals only goes through vertices of type $n$. Moreover, around each vertex $u$ of type $n$, these directed diagonals alternate between ingoing and outgoing. Indeed, around $u$, after each black type-$n$ triangle, there is necessarily a white type-$n$ triangle before the next black type-$n$ triangle. This stems from the fact that the triangles around $u$ can only be of type $n-1$, $n$ or $n+1$, and that along each edge, the oriented geodesic distance can only change by 1 or 2 (see Figure 5). Now, to resolve the intersections at type-$n$ vertices, we can take the convention that if a curve arrives at a vertex $u$ by an ingoing diagonal $\delta$, it will immediately leave $u$ by the first outgoing diagonal that we encounter going clockwise around $u$, starting at $\delta$ (see Figure 5). + +This yields a set of closed curves that we denote by $C_n(A)$. By construction, the curves in $C_n(A)$ separate vertices at oriented distance $n+1$ or higher from the origin, and they go counter-clockwise around these vertices (i.e., the origin lies on their right). \ No newline at end of file diff --git a/samples/texts/2952501/page_33.md b/samples/texts/2952501/page_33.md new file mode 100644 index 0000000000000000000000000000000000000000..bf32d29e82989a39bd50fef3c7d12bcabac29c25 --- /dev/null +++ b/samples/texts/2952501/page_33.md @@ -0,0 +1,33 @@ +Now, taking $K$ even larger if necessary, we can also have $cK/4 \ge 1$, and $1/2^{K/4} \le \varepsilon/2$, which does give that, with probability at least $1 - \varepsilon$, for all $j \ge Kr^2$, $\tilde{d}_{\mathcal{L}}((0,0), (j,0)) \ge r$. The case of negative $j$ can be treated in the same way. + +Let us now turn to the second assertion. Assume that there exist $j \ge 2K'r^2$ and $i \in \{-K'r^2, \dots, K'r^2\}$, such that $\tilde{d}_{\mathcal{L}}((i,0), (j,0)) < r$. Then, any geodesic from $(i,0)$ to $(j,0)$ must stay in the layer $\mathcal{L}_{[0,r]}$, and therefore must intersect the leftmost mirror geodesic from $(K'r^2, 0)$ to the line $\mathcal{L}_r$, so that + +$$ \tilde{d}_{\mathcal{L}}((K'r^2, 0), (j, 0)) < 3r. $$ + +But then, by the first assertion of the proposition, the probability of such an event is bounded above by $\varepsilon$. Considering also the case $j < -K'r^2$, we obtain the desired result. $\square$ + +An alternative proof of this result, adapting to Eulerian triangulations the method used in [13] for usual triangulations, can be found in Chapter 8 of [12]. Note that [23], that adapts the results of [13] to planar quadrangulations, and that was written simultaneously to the present work, also uses a block decomposition similar to [14]. + +## 7.2 Upper bounds + +After having proved in the previous subsection lower bounds for the distances along the boundary of $\mathcal{L}$, we now prove upper bounds for these quantities, that will carry to the UIPT of the digon, $\mathcal{T}_{\infty}^{(1)}$, thanks to Proposition 5.7. For that purpose, we follow closely the chain of arguments leading to Proposition 17 in [13, Section 4.3]. These bounds are expressed in terms of coalescence of leftmost mirror geodesics: + +**Proposition 7.6.** Let $\delta > 0$ and $\gamma > 0$. We can choose an integer $A \ge 1$ such that, for every sufficiently large $n$, with probability at least $1 - \delta$: + +> $\forall i \in \{-n+1, -n+2, \dots, n\}$, the leftmost mirror geodesic starting from $(i,0)$ coalesces with the one starting from $(-n + \lfloor 2ln/A \rfloor, 0)$, for some $0 \le l \le A$, before hitting $\mathcal{L}_{[\gamma\sqrt{n}]}$. + +The proof of this proposition can be adapted straightforwardly from that of [13, Proposition 16]. + +We now derive a similar result for $\mathcal{T}_{\infty}^{(1)}$. Recall the notation $L_r$ for the number of skeleton modules on $\partial^* B_r^\bullet(\mathcal{T}_{\infty}^{(1)})$. + +For any integer $n \ge 1$, we write $u_0(n)$ for a vertex chosen uniformly at random in the vertices of type $n$ of $\partial^* B_n^\bullet(\mathcal{T}_{\infty}^{(1)})$, and $u_1(n), \dots, u_{L_n-1}(n)$ for the other type-$n$ vertices of $\partial^* B_n^\bullet(\mathcal{T}_{\infty}^{(1)})$, enumerated clockwise, starting from $u_0(n)$. We extend the definition of $u_i(n)$ to $i \in \mathbb{Z}$ by periodicity. + +**Proposition 7.7.** Let $\gamma \in (0, 1/2)$ and $\delta > 0$. For every integer $A \ge 1$, let $H_{n,A}$ be the event where any leftmost mirror geodesic to the root starting from a type-$n$ vertex of $\partial^* B_n^\bullet(\mathcal{T}_{\infty}^{(1)})$ coalesces before time $\lfloor \gamma n \rfloor$ with the leftmost mirror geodesic to the root starting from $u_{\lfloor kn^2/A \rfloor}(n)$, for some $0 \le k \le \lfloor n^{-2} L_n A \rfloor$. Then, we can choose $A$ large enough that, for every sufficiently large $n$, + +$$ P(H_{n,A}) \le 1 - \delta. $$ + +*Proof.* The idea of the proof is to carry the result of Proposition 7.6 over to the case of $\mathcal{T}_{\infty}^{(1)}$, using the comparison principle of Proposition 5.7. To apply it, one needs to consider the intersection of $H_{n,A}$ with an event of the form + +$$ \{\lfloor an^2 \rfloor < L_n \le \lfloor a^{-1}n^2 \rfloor\} \cap \{\lfloor an^2 \rfloor < L_{n-\lfloor \gamma n \rfloor} \le \lfloor a^{-1}n^2 \rfloor\}. \quad (7.2) $$ + +Lemma 5.6 ensures that we can choose an $a > 0$ such that this latter event holds with probability at least $1 - \delta/2$. \ No newline at end of file diff --git a/samples/texts/2952501/page_34.md b/samples/texts/2952501/page_34.md new file mode 100644 index 0000000000000000000000000000000000000000..358af9c98d998c76bb083ea83e54ad6848338136 --- /dev/null +++ b/samples/texts/2952501/page_34.md @@ -0,0 +1,37 @@ +uniform vertex, then between any two vertices, as stated in Theorem 1.2. The transfer of the results from $\mathcal{L}$ to finite triangulations necessitates estimates on the distances in $\mathcal{L}$ that are derived in Section 7. + +Finally, Section 9 tackles the case of the Riemannian metric, using the same arguments as for the usual graph distance, to show that it is asymptotically proportional to the oriented pseudo-distance, and that, endowed with it, planar Eulerian triangulations still converge to the Brownian map. + +As our use of the layer decomposition to get the asymptotic proportionality of the oriented and usual distances follows closely the chain of arguments of [13] (albeit with additional difficulties), we purposefully use similar notation, and will omit some details of proofs when they are very similar and do not present any additional subtleties in our case. This is especially the case in Section 5, Section 7.2 and Section 8. + +# 2 Structure of Eulerian triangulations and bijection with trees + +## 2.1 Basic definitions + +We start by giving basic definitions related to graphs and maps, that will be needed in the sequel. + +**Definition 2.1.** Let $G$ be a finite connected graph. A **map** with underlying graph $G$ is an embedding $f$ of $G$ into an (orientable) surface $S$ such that + +* the images of the (open) edges of $G$ are homeomorphic to (open) segments + +* the images of different edges do not intersect, except at their extremities if they correspond to the same vertex + +* the connected components of $S \setminus f(G)$ are homeomorphic to the open disk; these components are called the **faces** of the map. + +A **planar** map is a map embedded into the sphere. + +A **rooted** map is a map equipped with a distinguished oriented edge, called its **root edge**. The starting vertex of the root edge is called the **root vertex**. + +A **pointed** map is a map equipped with a distinguished vertex. + +Maps are usually considered up to orientation-preserving homeomorphisms of the surface $S$. The only automorphism of a map that fixes an oriented edge is the trivial one, so that rooted maps do not have any non-trivial automorphisms. In the sequel, we will consider maps up to isomorphism, unless specified. + +Another way to define a map up to isomorphism is to equip its underlying graph with a cyclic ordering of the edges around each vertex. + +**Definition 2.2.** A **corner** in a map is an angular sector between two consecutive edges in the cyclic order around a vertex. + +Two notions that will be useful in the sequel are those of maps with boundaries, and submaps: + +**Definition 2.3.** A **map with boundaries** is a map $m$ with a certain number of distinguished faces, that are called its **external faces**. The other faces of $m$ are naturally called its **internal faces**. Likewise, the vertices of $m$ that are not incident to any external face are called its **inner vertices**. We allow two external faces to share vertices, but not edges. We will usually denote by $\partial m$ the boundary cycle of a map $m$ with one boundary. + +**Definition 2.4.** Let $m$ be a rooted map, and let $m'$ be a rooted map with simple boundaries. We say that $m'$ is a **submap** of $m$, and write $m' \subset m$, if $m$ can be obtained from $m'$, by gluing to each boundary $f_i$ of $m'$ some finite map $u_i$ with a (possibly non-simple) boundary. \ No newline at end of file diff --git a/samples/texts/2952501/page_36.md b/samples/texts/2952501/page_36.md new file mode 100644 index 0000000000000000000000000000000000000000..f51d1905bc1735028bd8022a3ac9cf816d411434 --- /dev/null +++ b/samples/texts/2952501/page_36.md @@ -0,0 +1,31 @@ +*Proof.* Let us give a sketch of the proof, as it is very similar to the proof of Proposition 19 in [13]. Recall the notation $u_j^{(n)}$ for the type-$n$ vertices of $\partial^*\tilde{B}_n^\bullet$. The first key step is to use Proposition 7.5 to get that a non-oriented shortest path from some $u_j^{(n)}$ to $\partial^*\tilde{B}_{n-|η_n|}^\bullet$ that stays in $\tilde{B}_n^\bullet$ cannot meander too much in the layer $\tilde{B}_n^\bullet \setminus \tilde{B}_{n-|η_n|}^\bullet$, and, more precisely, that it must stay in the region bounded by the leftmost mirror geodesics starting at $u_{j-|cn^2|}^{(n)}$ and $u_{j+|cn^2|}^{(n)}$ respectively, for some $c > 0$. Then, to bound probabilities of events on that sector of $\tilde{B}_n^\bullet \setminus \tilde{B}_{n-|η_n|}^\bullet$, Proposition 5.7 together with Lemma 5.6 allows us to replace the skeleton of $\tilde{B}_n^\bullet \setminus \tilde{B}_{n-|η_n|}^\bullet$ by independent Galton-Watson trees. We can therefore transfer the property of Proposition 8.1 from $\mathcal{L}$ to $\tilde{B}_n^\bullet \setminus \tilde{B}_{n-|η_n|}^\bullet$. Finally, to consider all vertices of $\partial^*\tilde{B}_n^\bullet$, we use the coalescence property obtained in Proposition 7.7, which amounts to saying that it suffices to consider for the values of $j$ a fixed number $C$, large but independent of $n$. + +The details of the proof can be adapted verbatim from the proof of Proposition 20 in [13] (replacing $d_{gr}$ by $\vec{d}$, and $d_{fpp}$ by $d_>$, with a small caveat. + +Indeed, when using the coalescence property of Proposition 7.7 (which corresponds to (57) in [13]), one must pay attention to two things. + +First, Proposition 7.7 only gives an upper bound on the distances between vertices of $\partial^* \tilde{B}_n^\bullet$ of type $n$, and the $C$ chosen $u_j^{(n)}$. To also include the vertices of type $n+1$, one must add an additional margin of 1 to the bounds, which, for any fixed $\epsilon$, can be smaller than $\epsilon c_0 \eta n/2$, for $n$ large enough. + +A second restriction of the application of Proposition 7.7 is that it ensures that *oriented* geodesics from the root to a type-$n$ vertex $v$ of $\partial^* \tilde{B}_n^\bullet$ and to one of the chosen $u_j^{(n)}$, are merged up to a level $\gamma n$. Thus, the upper bound on the oriented distance between $v$ and $u_j^{(n)}$ is not $2\gamma n$ but $3\gamma n$. + +Thus, rather than $\gamma = \varepsilon c_0 \eta / 2$, we take $\gamma = \varepsilon c_0 \eta / 6$, to obtain the equivalent of (57) in [13] for all vertices of $\partial^* \tilde{B}_n^\bullet$. $\square$ + +We then derive a more global result from the one of Proposition 8.2: + +**Proposition 8.3.** For every $\epsilon \in (0,1)$, + +$$ P((c_0 - \epsilon)n \le d(\rho, v) \le (c_0 + \epsilon)n \text{ for every vertex } v \in \partial^* B_n^\bullet) \xrightarrow{n \to \infty} 1. $$ + +The proof of this result is straightforwardly adapted from the proof Proposition 20 of [13], replacing $d_{gr}$ by $\vec{d}$, and $d_{fpp}$ by $d$. + +## 8.2 Asymptotic proportionality of distances in finite triangulations + +We now turn to finite triangulations. More precisely, we consider $\mathcal{T}_n^{(1)}$, uniform on the Eulerian triangulations of the digon with $n$ black triangles. Recall that such triangulations are in bijection with (rooted) Eulerian triangulations with $n$ black faces, from Figure 10. We write $\rho_n$ for the root of $\mathcal{T}_n^{(1)}$, and $d$ for the non-oriented distance on $\mathcal{T}_n^{(1)}$. + +**Proposition 8.4.** Let $o_n$ be uniform over the inner vertices of $\mathcal{T}_n^{(1)}$. Then, for every $\epsilon > 0$, + +$$ P(|d(\rho_n, o_n) - c_0 \vec{d}(\rho_n, o_n)| > \epsilon n^{1/4}) \xrightarrow{n \to \infty} 0. $$ + +To derive this from the previous results on $\mathcal{T}_\infty^{(1)}$, we will first establish an absolute continuity relation between finite triangulations and this infinite model. + +Recall that $C_{1,r}$ is the set of Eulerian triangulations of the cylinder of height $r$ and bottom boundary length 2. For $\Delta \in C_{1,r}$, we denote by $N(\Delta)$ the number of black triangles in $\Delta$. Finally, we write $\bar{\mathcal{T}}_n^{(1)}$ for the triangulation $\mathcal{T}_n^{(1)}$ together with \ No newline at end of file diff --git a/samples/texts/2952501/page_40.md b/samples/texts/2952501/page_40.md new file mode 100644 index 0000000000000000000000000000000000000000..83f651e5d5d051d59cfb77b4fce5374c0febbd0b --- /dev/null +++ b/samples/texts/2952501/page_40.md @@ -0,0 +1,21 @@ +Consequently, + +$$d^R \geq \left( \frac{\sqrt{3}}{8} \right) \vec{d}.$$ + +*Proof.* Let us prove the first bound, as the second is a direct consequence of it, together with (2.1). + +Let $A$ be a triangulation, and $S(A)$ its Euclidean geometric realization. For any continuous path $\gamma: [0, 1] \to S(A)$, we will construct an edge path $\gamma_E$ in $A$, such that, if $\gamma$ is a geodesic, then its length $l(\gamma)$ can be bounded from below by $\sqrt{3}/4$ times the number of edges in $\gamma_E$, which gives the desired inequality. + +Let thus $\gamma$ be a continuous path in $S(A)$. We construct a sequence $(u_0, u_1, \dots, u_k)$ of vertices of $A$, in the following way. We start by setting $u_0$ to be the vertex closest to $\gamma(0)$ (if there is an ambiguity, we just pick one of the possible vertices in an arbitrary way). Let $f_0$ be the first triangle that $\gamma$ crosses (i.e., gets out of after having spent a positive time in it). Let then $u_1$ be the vertex closest to the point where $\gamma$ leaves for the last time any of the triangles incident to $u_0$. We then define $u_2, \dots, u_k$, etc. similarly. This yields a *finite* sequence of vertices $u_0, u_1, \dots, u_k$: indeed, $\gamma$ cannot get close to an infinite number of distinct vertices of $S(A)$. Note also that $u_k$ is necessarily the closest vertex to $\gamma(1)$. + +By construction, for any $0 \le i \le k-1$, $u_i$ and $u_{i+1}$ are neighbors in $A$, so that the sequence does induce a path $\gamma_E$ of $k$ edges. + +Figure 17: The shortest distance that $\gamma$ can cross between the vicinity $u_i$ and the vicinity of $u_{i+1}$ corresponds to the altitude of an equilateral triangle of side length $1/2$, which is equal to $\sqrt{3}/4$ (the boundaries of the Voronoi cells associated to the vertices of the triangle are dashed, and $\gamma$ is in purple). + +Consider now a geodesic $\gamma$ in $S(A)$ (with respect to $d^R$), going from a vertex $v$ to a different vertex $w$. Then, the restriction of $\gamma$ to any triangle it crosses is necessarily a straight line (since this is a geodesic on a Euclidean triangle). Therefore, the portion of $\gamma$ going from the first moment that $\gamma$ is closest to $u_i$, to the last moment it is closest to $u_{i+1}$, crosses at least one triangle, from one edge to another, and also crossing the Voronoi cell of one of the vertices. This implies that this portion of $\gamma$ has a length of at least $\sqrt{3}/4$ (see Figure 17). Thus, + +$$l(\gamma) \geq \frac{\sqrt{3}}{4} \cdot k = \frac{\sqrt{3}}{4} \cdot l(\gamma_e).$$ + +This concludes the proof. $\square$ + +Recall that we write $\rho$ for the root vertex $(0,0)$ of the LHPET $\mathcal{L}$, and that $\mathcal{L}_r$ is the lower boundary of the layer $\mathcal{L}_{[0,r]}$. We have the following result: \ No newline at end of file diff --git a/samples/texts/2952501/page_43.md b/samples/texts/2952501/page_43.md new file mode 100644 index 0000000000000000000000000000000000000000..96fd8f64c18b3a357764b639e80809f784733321 --- /dev/null +++ b/samples/texts/2952501/page_43.md @@ -0,0 +1,31 @@ +Convergence of Eulerian triangulations + +[14] Nicolas Curien, Tom Hutchcroft, and Asaf Nachmias, *Geometric and spectral properties of causal maps*, J. Eur. Math. Soc. (JEMS) **22** (2020), no. 12, 3997–4024. MR-4176785 + +[15] P. Flajolet and R. Sedgewick, *Analytic combinatorics*, Cambridge University Press, Cambridge, 2009. MR-2483235 + +[16] S. Janson and J.-F. Marckert, *Convergence of discrete snakes*, J. Theoret. Probab. **18** (2005), no. 3, 615–647. MR-2167644 + +[17] K. Krikun, *Local structure of random quadrangulations*, arXiv:0512304 + +[18] M. Krikun, *Uniform infinite planar triangulation and related time-reversed critical branching process*, Journal of Mathematical Sciences **131** (2005), no. 2, 5520–5537. MR-2050691 + +[19] S. Lando and A. Zvonkin, *Graphs on surfaces and their applications*, Encyclopaedia of Mathematical Sciences, vol. 141, Springer-Verlag, Berlin, 2004, With an appendix by Don B. Zagier, Low-Dimensional Topology, II. MR-2036721 + +[20] J.-F. Le Gall, *Uniqueness and universality of the brownian map*, Ann. Probab. **41** (2013), 2880–2960. MR-3112934 + +[21] J.-F. Le Gall and T. Lehéricy, *Separating cycles and isoperimetric inequalities in the uniform infinite planar quadrangulation*, Ann. Probab. **47** (2019), no. 3, 1498–1540. MR-3945752 + +[22] J.-F. Le Gall and M. Weill, *Conditioned brownian trees*, Ann. Inst. H. Poincaré Probab. Statist **42** (2006), no. 4, 455–489. MR-2242956 + +[23] T. Lehéricy, *First-passage percolation in random planar maps and tutte's bijection*, arXiv:1906.10079 + +[24] T.M. Liggett, *An improved subadditive ergodic theorem*, Ann. Probab. **13** (1985), 1279–1285. MR-0806224 + +[25] G. Miermont, *The brownian map is the scaling limit of uniform random plane quadrangulations*, Acta Math. **210** (2013), 319–401. MR-3070569 + +[26] F. Papangelou, *A lemma on the Galton-Watson process and some of its consequences*, Proc. Amer. Math. Soc. **19** (1968), 1469–1479. MR-0232457 + +[27] W. T. Tutte, *A census of slicings*, Canadian J. Math. **14** (1962), 708–722. MR-0142470 + +**Acknowledgments.** I warmly thank Grégory Miermont for his crucial help and advice, as well as Nicolas Curien and Christina Goldschmidt for their insightful remarks. I also thank the referees for their very helpful feedback. \ No newline at end of file diff --git a/samples/texts/2952501/page_45.md b/samples/texts/2952501/page_45.md new file mode 100644 index 0000000000000000000000000000000000000000..0ff52536600de4be0add98d71941784ef4e3ca93 --- /dev/null +++ b/samples/texts/2952501/page_45.md @@ -0,0 +1,27 @@ +We also define the **oriented geodesic distance** to any vertex of A, as the oriented +distance from the origin (that is, the root vertex $\rho$) to that vertex. This gives a labeling of +the vertices of A, such that the sequence of labels around any triangle, starting from the +minimal label, is of the form $n \to n+1 \to n+2$. + +Let us now introduce a bit of notation that will be of use in the sequel. + +**Definition 2.7.** In a rooted Eulerian triangulation *A*, a vertex of type *n* is a vertex whose canonical labeling by the oriented geodesic distance is *n*. An edge of type *n* → *m* is an oriented edge that starts at a vertex of type *n* and ends at a vertex of type *m*. A triangle of type *n* is a triangle adjacent to a vertex of type *n* − 1, one of type *n* and one of type *n* + 1. + +By keeping only the edge of type $n + 1 \rightarrow n + 2$ in each black face of type $n + 1$, we construct a graph $T$ whose vertices, that correspond to $V(A) \setminus \{\rho\}$, are labeled by integers. Moreover, it is **well-labeled** in the sense that the labels of adjacent vertices differ by exactly 1, and that the root vertex has label 1. By construction, those labels are all positive, but we do not include it in the definition of being well-labeled, for reasons that will be clear soon. + +**Lemma 2.8.** [10] For any planar rooted Eulerian triangulations $A$, the corresponding labeled graph $T$ is a plane tree. + +This tree (which is a spanning tree of the subgraph of $A$ induced by $V(A) \setminus \{\rho\}$) is naturally rooted at the corner of a vertex of type 1, that corresponds to the root edge of $A$ (see Figure 2). + +Figure 2: The construction of the well-labeled tree associated to the triangulation of Figure 1. + +**Theorem 2.9.** [10] The above mapping $\varphi_n$ from the set of rooted planar Eulerian triangulations with $n$ faces, to the set of well-labeled trees with $n$ edges with positive labels, is a bijection. + +Bouttier-Di Francesco-Guitter have also detailed the inverse construction from trees +to triangulations, that we recall now. + +The inverse construction consists in building iteratively the black triangles of A. +Starting from a well-labeled rooted plane tree with positive integers, the first step +consists in adding an origin (labeled 0). We then create a black triangle of type 1 to +the right of each edge of type 1 → 2, by adding edges between the origin and the two +vertices of the edge. The creation of these black triangles splits the original external \ No newline at end of file diff --git a/samples/texts/2952501/page_5.md b/samples/texts/2952501/page_5.md new file mode 100644 index 0000000000000000000000000000000000000000..baf3a8d3932b5f574f91ae86b64df63bb96f56d3 --- /dev/null +++ b/samples/texts/2952501/page_5.md @@ -0,0 +1,9 @@ +Figure 7: From the rule we have chosen to resolve intersections (see Figure 5), two curves in $C_n(A)$ going through the same vertex cannot encircle one other (left), and one curve cannot go through the same vertex twice (right), as it would imply that all oriented paths from the origin to at least one vertex of type $n-1$ or less, would have at least length $n+1$. + +Thus, $B_n(A)$ is an Eulerian triangulation with simple boundaries², as many as curves in $C_n(A)$. Moreover, the faces of $B_n(A)$ adjacent to these boundaries compose the type-$n$ modules of $A$, so that each part of $\partial B_n(A)$ is **alternating**, that is, the adjacent faces alternate between black and white. + +Let us also formalize the definition of the complement of $B_n(A)$. It is naturally obtained from $A$ by removing the faces and edges of $A$ that are incident to at least a vertex at distance $n-1$ or less from the origin. Note that it is made of as many connected components as there are curves in $C_n(A)$, as it is also the number of boundaries of $B_n(A)$. Consider some $\mathscr{C} \in C_n(A)$, and write $M(A, \mathscr{C})$ for the corresponding connected component of $A \setminus B_n(A)$. $M(A, \mathscr{C})$ is a planar Eulerian triangulation with a boundary, which has the same length as the corresponding one of $B_n(A)$, and is also alternating. However, the boundary of $M(A, \mathscr{C})$ is not necessarily simple. More precisely, a type-(n+1) vertex v of A that sits on the boundary of M(A, $\mathscr{C}$) is attached to the type-(n → n + 1) edges and type-(n + 1) faces that are adjacent to v in A, as they were excluded from $B_n(A)$, so that v may be a separating vertex in the external face of M(A, $\mathscr{C}$). This is not the case for type-n vertices, as $B_n(A)$ contains all the type-(n - 1 → n) edges and type-n faces of A. Thus, the boundary of $M(A, \mathscr{C})$ can have separating vertices, but only on boundary vertices that have, in clockwise order, a black triangle before them and a white one after (as it corresponds to the type-(n + 1) vertices of A). We call such boundary conditions **semi-simple**. + +Let v be a distinguished vertex of A at oriented distance at least $n+2$ from the root. We can now define the **hull** $\mathcal{B}_n^\bullet(A)$ of $B_n(A)$, as the union of $B_n(A)$ and all the connected components of its complement that do not contain v. More precisely, for each curve $\mathscr{C} \in C_n(A)$ that does not separate v from the origin, we glue the boundary of $M(A, \mathscr{C})$ to the corresponding boundary of $B_n(A)$. This operation is well-defined, as the latter is simple, and they both have the same length. The resulting map $\mathcal{B}_n^\bullet(A)$ has only one boundary, that corresponds to $\mathscr{C}^*$, the unique curve of $C_n(A)$ that separates v from the + +²Note that these boundaries may share a vertex, but not an edge. \ No newline at end of file diff --git a/samples/texts/2952501/page_6.md b/samples/texts/2952501/page_6.md new file mode 100644 index 0000000000000000000000000000000000000000..843c32948cfa3d0755164e104f258ecafd2fbe8e --- /dev/null +++ b/samples/texts/2952501/page_6.md @@ -0,0 +1,11 @@ +Figure 8: In an Eulerian triangulation (top), we cut along the edges of type $n \to n+1$ to separate the ball of radius $n$ from the components of its complement (bottom). This possibly induces the duplication of edges and vertices in the ball. + +origin. + +In the sequel, we will use the notion of **local distance** between rooted maps. Let $\mathcal{M}$ be the set of finite rooted maps, for $m, m' \in \mathcal{M}$, we define the **local distance** between $m$ and $m'$ as + +$$d_{\text{loc}}(m, m') = \frac{1}{1 + \sup\{R \ge 1 | \mathcal{B}_R^d(m) = \mathcal{B}_R^d(m')\}},$$ + +where $\mathcal{B}_R^d(m)$ is defined similarly as before, replacing $\vec{d}$ by the usual graph distance. It is clearly a distance on $\mathcal{M}$, and the completion $(\overline{\mathcal{M}}, d_{\text{loc}})$ of the space $(\mathcal{M}, d_{\text{loc}})$ is a Polish space. The notion of convergence in this space will be called **local limit**. The elements of $\overline{\mathcal{M}} \setminus \mathcal{M}$ are thus **infinite** maps that can be defined as the local limit of finite rooted maps. + +Note that, from (2.1), if $A_n$ is a sequence of rooted Eulerian triangulations (possibly \ No newline at end of file diff --git a/samples/texts/2952501/page_7.md b/samples/texts/2952501/page_7.md new file mode 100644 index 0000000000000000000000000000000000000000..f2d3c14c4547dd274274d2ab1135f1613fe5b2b0 --- /dev/null +++ b/samples/texts/2952501/page_7.md @@ -0,0 +1,5 @@ +Figure 9: A rooted, pointed Eulerian triangulation (bottom right) and its balls (left) and corresponding hulls (right). The module "diagonals" are in dashed purple, and the boundaries of the balls and hulls in solid gray. We can see an example of duplication of vertices in the ball of radius 2, marked in red. + +with a boundary), and *A* a rooted planar map, the property that all oriented balls of *A**n* converge to those of *A*, as *n* tends to infinity, is equivalent to the same property for non-oriented balls, which is precisely the definition of the convergence of *A**n* to *A* in the sense of local limits of rooted planar maps. + +As the topology induced by the local distance is what will really matter in the sequel, rather than the actual value of the local distance between two maps, we can forget the general definition of the local distance, and just compare the oriented balls of Eulerian triangulations. \ No newline at end of file diff --git a/samples/texts/2952501/page_8.md b/samples/texts/2952501/page_8.md new file mode 100644 index 0000000000000000000000000000000000000000..162643469b63cc4a882de0b44b982eea9ab043ef --- /dev/null +++ b/samples/texts/2952501/page_8.md @@ -0,0 +1,39 @@ +### 3 Convergence to the Brownian map + +We will now state and give the proof of the main result of this paper. + +Before doing so, let us recall the construction of the Brownian map, and introduce some notation. As in Section 2.3, we write $\mathfrak{e}$ for a standard Brownian excursion, and $Z$ for the "head" of the Brownian snake driven by $\mathfrak{e}$, i.e., conditionally on $\mathfrak{e}$, $Z$ a continuous, centered Gaussian process on $[0, 1]$ with covariance + +$$ \mathrm{Cov}(Z_s, Z_t) = \mathfrak{e}_{s,t}, \quad s, t \in [0, 1]. $$ + +The Brownian excursion $\mathfrak{e}$ encodes the **Continuum Random Tree** $(\mathcal{T}_{\mathfrak{e}}, d_{\mathfrak{e}})$, defined by: + +$$ \begin{gathered} d_{\mathfrak{e}}(s,t) = \mathfrak{e}(s) + \mathfrak{e}(t) - 2\mathfrak{e}_{s,t} \\ \mathcal{T}_{\mathfrak{e}} = [0, 1]/\{d_{\mathfrak{e}} = 0\}. \end{gathered} $$ + +The function $d_{\mathfrak{e}}$, which is a pseudo-distance on $[0, 1]$, induces a true distance on $\mathcal{T}_{\mathfrak{e}}$ via the canonical projection $p_{\mathfrak{e}}: [0, 1] \to \mathcal{T}_{\mathfrak{e}}$, to $\mathcal{T}_{\mathfrak{e}}$. + +Almost surely, there is a unique $s \in [0, 1]$ such that $Z_s = \inf Z$ [22]. We then denote this point by $s_*$, and $x_* = p_{\mathfrak{e}}(s_*)$ its projection on $\mathcal{T}_{\mathfrak{e}}$. + +We define, for $s \le t \in [0, 1]$, + +$$ D^{\circ}(s,t) = D^{\circ}(t,s) := Z_s + Z_t - 2 \max( \min_{r \in [s,t]} Z_r, \min_{r \in [t,1] \cap [0,s]} Z_r ). $$ + +This function does not satisfy the triangle inequality, which leads us to introduce + +$$ D^*(s,t) := \inf \left\{ \sum_{i=1}^k D^{\circ}(s_i, t_i) \mid k \ge 1, s_1 = s, t_k = t, d_{\mathfrak{e}}(t_i, s_{i+1}) = 0 \quad \forall i \in \{1, 2, \dots, k\} \right\}. $$ + +We can now define the **Brownian map**, by setting $m_{\infty} = [0, 1]/\{D^* = 0\}$, and equipping this space with the distance induced by $D^*$, which we still denote by $D^*$. + +Let $\mathcal{T}_n$ be a uniform random rooted Eulerian planar triangulation with $n$ black faces, equipped with its usual graph distance $d_n$, and its oriented pseudo-distance $\vec{d}_n$. Let $\overline{\mathcal{T}}_n$ be the triangulation $\mathcal{T}_n$ together with a distinguished vertex $o_n$ picked uniformly at random. Recall from Section 2.2 that $\overline{\mathcal{T}}_n$ is the image, by the BDG bijection, of a random labeled tree $\mathcal{T}_n$, uniformly distributed over the set of well-labeled rooted plane trees with $n$ edges. We denote by $l_n$ the labels of the vertices of $\mathcal{T}_n$, and enumerate as in Section 2.2 the vertices (or rather, the corners) of $\mathcal{T}_n$, by setting $u_i^{(n)}$ to be the $i$-th vertex visited by the contour process of $\mathcal{T}_n$, for $0 \le i \le 2n$. As before, we denote by $L_{(n)}$ the rescaled labels of the vertices of $\overline{\mathcal{T}}_n$. + +We define the symmetrization $\overleftrightarrow{d_n}$ of $\vec{d}_n$, by + +$$ \overleftrightarrow{d_n}(u,v) = \frac{\vec{d}_n(u,v) + \vec{d}_n(v,u)}{2}. $$ + +We also define a rescaled oriented distance $\tilde{D}_{(n)}$ on $[0, 1]^2$, by first setting, for $i, j \in \{0, 1, \dots, 2n\}$: + +$$ \tilde{D}_{(n)} \left( \frac{i}{n}, \frac{j}{n} \right) = \frac{\vec{d}_n(u_i^{(n)}, u_j^{(n)})}{n^{1/4}}, $$ + +then linearly interpolating to extend $\tilde{D}_{(n)}$ to $[0, 1]^2$. + +We define similarly $D_{(n)}$ from $d_n$, as well as $\overleftrightarrow{D}_{(n)}$ from $\overleftrightarrow{d}_n$. \ No newline at end of file diff --git a/samples/texts/2952501/page_9.md b/samples/texts/2952501/page_9.md new file mode 100644 index 0000000000000000000000000000000000000000..e813df42c72606caf0ab69369e5580fcedd79bec --- /dev/null +++ b/samples/texts/2952501/page_9.md @@ -0,0 +1,48 @@ +**Theorem 3.1.** Let $(\mathbf{m}_{\infty}, D^*)$ be the Brownian map. There exists some constant $c_0 \in [2/3, 1]$, such that the following convergence in distribution holds: + +$$ (C_{(n)}, L_{(n)}, \vec{D}_{(n)}, \overleftrightarrow{D}_{(n)}, D_{(n)}) \xrightarrow[n \to \infty]{(d)} (\mathbb{e}, Z, D^*, D^*, \mathbf{c}_0 D^*). $$ + +Consequently, we have the following joint convergences + +$$ n^{-1/4} \cdot (V(\mathcal{T}_n), \overleftrightarrow{d}_n) \xrightarrow[n \to \infty]{(d)} (\mathbf{m}_{\infty}, D^*) $$ + +$$ n^{-1/4} \cdot (V(\mathcal{T}_n), d_n) \xrightarrow[n \to \infty]{(d)} \mathbf{c}_0 \cdot (\mathbf{m}_{\infty}, D^*), $$ + +for the Gromov-Hausdorff distance on the space of isometry classes of compact metric spaces. + +Note that we would like to have a statement similar to the one on $\overleftrightarrow{d}_n$ for $\vec{d}_n$. However, as $\vec{d}_n$ is not a proper distance, it does not induce a metric space structure on $V(T_n)$. Thus, we would need to generalize the Gromov-Hausdorff topology to spaces equipped with a non-symmetric pseudo-distance, to be able to write such a statement. + +*Proof.* We admit here Theorem 1.2, that will be proven later in the paper: for every $\varepsilon > 0$, we have + +$$ P\left( \sup_{(x,y) \in V(\mathcal{T}_n)} |d_n(x,y) - \mathbf{c}_0 \vec{d}_n(x,y)| > \varepsilon n^{1/4} \right) \xrightarrow{n \to \infty} 0. \quad (3.1) $$ + +We proceed similarly to the case of usual triangulations in [20]. + +For this whole proof, we work with the pointed triangulation $\bar{\mathcal{T}}_n$, but, as all Eulerian triangulations with $n$ black faces have the same number of vertices, this does not introduce any bias for the underlying, non-pointed triangulation, so that the final statement also holds for $\mathcal{T}_n$. + +We have, from Proposition 2.11, for any $0 \le i < j \le 2n$: + +$$ +\begin{aligned} +& \vec{D}_n\left(\frac{i}{n}, \frac{j}{n}\right) \\ +& \le \frac{2}{n^{1/4}} \left( l_n(u_i^{(n)}) + l_n(u_j^{(n)}) - 2 \max\left(\min_{k \in \{i, \dots, j\}} l_n(u_k^{(n)}), \min_{k \in \{j, \dots, 2n\} \cup \{0, \dots, i\}} l_n(u_k^{(n)})\right) + 2 \right). +\end{aligned} +\quad (3.2) +$$ + +As noted before, if we did not have the global multiplicative factor of 2 in (3.2), we could then proceed as for usual triangulations and other well-known families of planar maps. Thus, the rest of this proof will consist in proving that Theorem 1.2 makes it possible to “get rid” of this cumbersome factor. + +We claim that the sequence of the rescaled distances $(\vec{D}_{(n)}(s,t))_{s,t \in [0,1]}$ is tight. First note that, for any $s, s', t, t' \in [0,1]$, we have + +$$ |\vec{D}_{(n)}(s,t) - \vec{D}_{(n)}(s',t')| \le 2 (\vec{D}_{(n)}(s,s') + \vec{D}_{(n)}(t',t)). \quad (3.3) $$ + +Indeed, $\vec{D}_n$, like $\vec{d}_n$, satisfies the triangle inequality, so that + +$$ +\begin{align*} +\vec{D}_{(n)}(s,t) - \vec{D}_{(n)}(s',t') &\leq \vec{D}_{(n)}(s,s') + \vec{D}_{(n)}(t',t) \\ +\vec{D}_{(n)}(s',t') - \vec{D}_{(n)}(s,t) &\leq \vec{D}_{(n)}(s',s) + \vec{D}_{(n)}(t,t'), +\end{align*} +$$ + +which gives (3.3) when taking into account that, while $\vec{D}_{(n)}$ is not symmetric, we do have, for any $s,t \in [0,1]$, $\vec{D}_{(n)}(s,t) \le 2\vec{D}_{(n)}(t,s)$. Then, using (3.2) and Theorem 2.12, we \ No newline at end of file diff --git a/samples/texts/3220451/page_2.md b/samples/texts/3220451/page_2.md new file mode 100644 index 0000000000000000000000000000000000000000..1abbd1efffecece063acb55fa9336f409b38ea49 --- /dev/null +++ b/samples/texts/3220451/page_2.md @@ -0,0 +1,57 @@ +WM and then the effectiveness of such an algorithm is validated in Section 4. Finally, in Section 5, the conclusion is summarized and the future work is discussed. + +**Notations.** $\mathbb{R}$ and $\mathbb{N}^+$ represent the space of all real numbers and positive integers, respectively. $\mathbb{R}^n$ and $\mathbb{R}^{m \times n}$ stand for the $n$-dimensional Euclidean space and the set of all $m \times n$ real matrices, respectively. For any vector $x \in \mathbb{R}^n$, $\|x\|$ denotes the Euclidean norm of $x$. Additionally, for any nonnegative matrix $P \in \mathbb{R}^{n \times n}$, $\|x\|_P$ means the weighted norm of $x$ (i.e., $\|x\|_P = \sqrt{x^T P x}$). + +## 2. PROBLEM FORMULATION + +Consider the following nonlinear system with IFs + +$$ +\begin{cases} +x(k+1) = g(x(k)) + w(k) + Ff(k) \\ +y(k) = h(x(k)) + v(k) + Gf(k), +\end{cases} +\quad (1) +$$ + +where $x(k) \in \mathbb{R}^{n_x}$, $y(k) \in \mathbb{R}^{n_y}$ and $f(k) \in \mathbb{R}^{n_f}$ are the state vector, measurement output and IF signal, respectively. $w(k) \in \mathbb{R}^{n_x}$ and $v(k) \in \mathbb{R}^{n_y}$ are mutually uncorrelated zero-mean Gaussian white noises with respective covariance matrices $R_w$ and $R_v$. $F$ and $G$ are known matrices with appropriate dimensions. $g(\cdot)$ and $h(\cdot)$ are known nonlinear functions. + +The intermittent fault $f(k)$ is assumed to satisfy the following form + +$$ f(k) = \sum_{s=1}^{\infty} (\Theta(k - k_{s,1}) - \Theta(k - k_{s,2}))m_s(k), \quad s \in \mathbb{N}^{+}, \quad (2) $$ + +where $k_{s,1}$ and $k_{s,2}$ are the sth unknown appearing time and disappearing time of IF $f(k)$, respectively. $\Theta(\cdot)$ is a function satisfying $\Theta(i) = 1$ ($i \ge 0$) and $\Theta(i) = 0$ ($i < 0$). $m_s(k)$ is the sth unknown fault magnitude. Define $d_{s,1} = k_{s,2} - k_{s,1}$ and $d_{s,2} = k_{s+1,1} - k_{s,2}$ as the sth active duration time and inactive duration time of $f(k)$. In this paper, we suppose that there exist two known constants $\bar{d}_1 > 0$ and $\bar{d}_2 > 0$ satisfying $d_{s,1} \le \bar{d}_1$ and $d_{s,2} \le \bar{d}_2$ ($s \in \mathbb{N}^+$), where $\bar{d}_1$ and $\bar{d}_2$ are respectively called the lower bounds of fault active duration and fault inactive duration. + +If a residual $r(k)$ satisfies the following two conditions: + +(1) there exists a constant $0 \le \tau_1 < \bar{d}_1$ such that $r(k) \ge J_{\text{th},1}$ holds for all $k \in [k_{s,1} + \tau_1, k_{s,2})$ ($s \in \mathbb{N}^+$), where $J_{\text{th},1}$ is the detection threshold for the appearing time and $k_{s,1} + \tau_1$ is the sth appearing time detected by the residual $r(k)$; + +(2) there exists a constant $0 \le \tau_2 < \bar{d}_2$ such that $r(k) < J_{\text{th},2}$ holds for all $k \in [k_{s,2} + \tau_2, k_{s+1,1})$ ($s \in \mathbb{N}^+$), where $J_{\text{th},2}$ is the detection threshold for the disappearing time and $k_{s,2} + \tau_2$ is the sth disappearing time detected by the residual $r(k)$, + +it is said that IF $f(k)$ is detectable by the residual $r(k)$. + +**Remark 1.** The core task for IF detection is to detect all appearing and disappearing times of IFs. If condition (1) is fulfilled, the designed residual $r(k)$ must be larger than the threshold $J_{\text{th},1}$ before fault $f(k)$ disappears, which means that there must exist a period of alarm time during $[k_{s,1}, k_{s,2})$ ($s \in \mathbb{N}^+$). Condition (2) shows that $r(k)$ can decrease below the threshold $J_{\text{th},2}$ before the next fault + +Fig. 1. IF and the residual of EKF in the case of $f_a = 1$ + +Fig. 2. IF and the residual of EKF in the case of $f_a = 2$ + +$f(k)$ appears, which ensures that the sth disappearing time and the s+1th appearing time of IF $f(k)$ can be clearly distinguished. Combining the two conditions, it is easy to deduce that all appearing and disappearing times of IF $f(k)$ can be detected by the residual $r(k)$. + +**Example 1:** Consider the nonlinear system with the following parameters + +$$ +\begin{align*} +x(k) &= [x_1(k), x_2(k)]^T, & g(x(k)) &= [g_1(x(k)), g_2(x(k))]^T, \\ +g_1(x(k)) &= 0.89x_1(k) + 0.1x_2(k) - 0.11 \sin(x_1(k)x_2(k)), \\ +g_2(x(k)) &= 0.9x_2(k) - 0.2x_1(k) + 0.01 \cos(x_2^2(k)), \\ +h(x(k)) &= 0.5x_1(k) + x_2(k), \\ +F &= [2, 0]^T, & G &= 0, & R_w &= 0.05^2 I, & R_v &= 0.05^2. +\end{align*} +$$ + +The IF $f(k)$ is chosen as + +$$ f(k) = \begin{cases} f_a, & k \in [50, 70] \cup [85, 105] \cup [120, 150] \\ & \cup [165, 203] \cup [215, 243] \cup [255, 270], \\ 0, & \text{otherwise}. \end{cases} $$ + +By means of EKF, the estimate $\hat{x}^*(k)$ can be derived. Then the residual is defined as $r(k) = y(k) - h(\hat{x}^*(k))$. The trajectories of $r(k)$ in the case of $f_a = 1$ and $f_a = 2$ are respectively depicted in Figs. 1 and 2. \ No newline at end of file diff --git a/samples/texts/3220451/page_4.md b/samples/texts/3220451/page_4.md new file mode 100644 index 0000000000000000000000000000000000000000..094e41aa29ed5f7c240a5df45c13f49ee4ddf0c5 --- /dev/null +++ b/samples/texts/3220451/page_4.md @@ -0,0 +1,50 @@ +the prior estimate. Therefore, the key point for detecting +the disappearing times of IFs is to appropriately discard +the prior estimate. In this paper, according to the index +$\rho(k)$, the unreliability degree of prior estimate can be +clearly reflected. Then the weight matrices $P(k)$ and +$Q(k)$ are dynamically regulated, which can guarantee the +performance of estimator, in the meantime suppress the +smearing effects. + +Based on the above analysis, the following IF detection +algorithm is obtained. + +IF Detection Algorithm for Nonlinear Systems + +(1) Set $\bar{\rho}$ and $\underline{\rho}$ according to $R_w$ and $R_v$. Select $N$ and $L$ depending on $\bar{d}_1$ and $\bar{d}_2$. The threshold $J_{\text{th}}$ is obtained by 100 Monte Carlo simulations without faults. + +(2) If $k > N+L$, calculate the index $\rho(k)$ and the weight matrices $P(k)$ and $Q(k)$. Otherwise, jump to Step (6). + +(3) Compute the QCF (7) by means of $P(k)$, $Q(k)$, $\bar{x}(k-N|k)$ and $y(k-i)$ ($i=0, \dots, N$). + +(4) By utilizing the PSO algorithm, solve the suboptimal estimate $\hat{x}^o(k-N)$ of OP (4). + +(5) Calculate the residual $r(k)$ and evaluation function $J(k)$. If $J(k) \ge J_{\text{th}}$, faults occur. Otherwise, no fault occurs. + +(6) Let $k = k + 1$. Return to Step (2). + +4. A NUMERICAL EXAMPLE + +Consider Example 1 in Section 2 and the parameters of +MHEDWM are selected as follows + +$$N = 4, L = 3, \bar{\rho} = 0.1, \underline{\rho} = 0.001, x(0) = [-1, 1]^T.$$ + +The simulation results are shown in Figs. 3-8. Figs. 3 and 4 depict the trajectories of system state and state estimates in the absence of faults, where $x_i(k)$, $\hat{x}_i^o(k)$, $\hat{x}_i^*(k)$ and $\hat{x}_i^*(k)$ ($i=1,2$) respectively represent the system state, the estimate of MHEDWM, the estimate of moving horizon estimation with constant weight matrices (MHECWM) $P(k) = 0.5I$, $Q(k) = 0.5I$, and the estimate of EKF. Figs. 5 and 7 respectively describe the trajectories of IF, the evaluation functions of MHEDWM and MHECWM, and the corresponding thresholds in the case of $f_a = 1$ and $f_a = 2$. The alarm times detected by MHEDWM and MHECWM in the case of $f_a = 1$ and $f_a = 2$ are respectively shown in Figs. 6 and 8. + +Define $\delta = 1/Z \sum_{z=1}^{Z} 1/k_f \sum_{k=1}^{k_f} e_z^T(k)e_z(k)$ as the mean square estimate error (MSEE), where Z is the number of simulation tests, $k_f$ is the step number of each simulation test, and $e_z(k)$ is the estimate error in the zth simulation test. In the case of no fault and after 100 simulations tests, the corresponding MSEEs of MHEDWM, MHECWM and EKF are derived as follows + +$$\delta_{\text{MHEDWM}} = 0.0152, \delta_{\text{MHECWM}} = 0.0148, \delta_{\text{EKF}} = 0.0196.$$ + +From the simulation results, it can be seen that 1) MHECWM is a hysteretic estimation algorithm with the highest estimation accuracy and the smearing effects; 2) EKF is a real-time estimation algorithm with the worst smearing effects; 3) MHEDWM is a hysteretic estimation algorithm with the second-highest estimation accuracy and without the smearing effects, which can detect all appearing and disappearing times of IF $f(k)$ accurately + +Fig. 3. System state $x_1(k)$ and its estimates + +Fig. 4. System state $x_2(k)$ and its estimates + +Fig. 5. IF and residuals in the case of $f_a = 1$ + +and timely by properly selecting $N$ and $L$. Therefore, +MHEDWM is superior to MHECWM and EKF in the +problem of IF detection for nonlinear systems. \ No newline at end of file diff --git a/samples/texts/3826450/page_1.md b/samples/texts/3826450/page_1.md new file mode 100644 index 0000000000000000000000000000000000000000..86e7c7514b4bcbbed633b52c25e82d9b2c3057e3 --- /dev/null +++ b/samples/texts/3826450/page_1.md @@ -0,0 +1,51 @@ +Recursive Construction of Stable Assemblies +of Recurrent Neural Networks + +Michaela Ennis¹,* + +Leo Kozachkov²,* + +Jean-Jacques Slotine³ + +¹Harvard Medical School, Harvard University + +²Department of Brain and Cognitive Sciences, Massachusetts Institute of Technology + +³ Nonlinear Systems Laboratory, Massachusetts Institute of Technology + +* equal contribution + +Abstract + +Advanced applications of modern machine learning will likely involve combinations of trained net- +works, as are already used in spectacular systems such as DeepMind's AlphaGo. Recursively building +such combinations in an effective and stable fashion while also allowing for continual refinement of the +individual networks - as nature does for biological networks - will require new analysis tools. This paper +takes a step in this direction by establishing contraction properties of broad classes of nonlinear recur- +rent networks and neural ODEs, and showing how these quantified properties allow in turn to recursively +construct stable networks of networks in a systematic fashion. The results can also be used to stably +combine recurrent networks and physical systems with quantified contraction properties. Similarly, they +may be applied to modular computational models of cognition. + +# 1 Introduction + +Neuro-inspired machine learning has profoundly altered many fields such as computer vision, natural language processing, and computational neuroscience [2, 10]. While models trained with e.g. deep learning are remarkably powerful, they are for the most part ‘black boxes’. This opaqueness can be dangerous in safety-critical applications, such as autonomous driving or human-centered robotics, and it limits conceptual progress. In the case of recurrent models, one difficulty is that providing a certificate of stability is currently impossible or computationally impractical. Given that stability is a fundamental property of dynamical systems – and is intimately linked to concepts of control, generalization, data-efficiency, and robustness – being able to guarantee the stability of a recurrent model is an important step towards making sure deep models behave as we expect them to. Furthermore, overall stability and convergence guarantees will become central as ever larger feedback combinations of trained networks are developed, in a fashion reminiscent of the modular functional architecture of the brain [33, 10, 40] + +In this spirit, there has been a recent flux of work focusing on applications of contraction analysis [16] to recurrent models. Loosely speaking, a dynamical system is said to be contracting if any two of its trajectories converge to each other exponentially, regardless of initial conditions. It can be shown that the non-autonomous system + +$$ +\dot{\mathbf{x}} = \mathbf{f}(\mathbf{x}, t) +$$ + +is contracting if there exists a metric $\mathbf{M}(\mathbf{x}, t) = \mathbf{\Theta}(\mathbf{x}, t)^T \mathbf{\Theta}(\mathbf{x}, t) > 0$ such that uniformly + +$$ +\dot{\mathbf{M}} + \mathbf{M}\mathbf{J} + \mathbf{J}^T\mathbf{M} \leq -\beta\mathbf{M} +$$ + +where **J** = $\frac{\partial f}{\partial x}$ and $\beta > 0$. For more details see the main reference [16]. Similarly, a non-autonomous +discrete-time system + +$$ +\mathbf{x}_{t+1} = \mathbf{f}(\mathbf{x}_t, t) +$$ \ No newline at end of file diff --git a/samples/texts/3826450/page_18.md b/samples/texts/3826450/page_18.md new file mode 100644 index 0000000000000000000000000000000000000000..1dc90fce34f33fc742c92d0271573e81d0926cef --- /dev/null +++ b/samples/texts/3826450/page_18.md @@ -0,0 +1,29 @@ +where the variable $\mathbf{y}$ is interpreted as a vector of firing rates, rather than membrane potentials. The two models are related by the transformation $\mathbf{x} = \mathbf{W}\mathbf{y} + \mathbf{b}$, which yields + +$$\tau \dot{\mathbf{x}} = \mathbf{W}(-\mathbf{y} + \phi(\mathbf{W}\mathbf{y} + \mathbf{b})) + \tau \dot{\mathbf{b}} = -\mathbf{x} + \mathbf{W}\phi(\mathbf{x}) + \mathbf{v}$$ + +where $\mathbf{v} \equiv \mathbf{b} + \tau \dot{\mathbf{b}}$. Thus $\mathbf{b}$ is a low-pass filtered version of $\mathbf{v}$ (or conversely, $\mathbf{v}$ may be viewed as a first order prediction of $\mathbf{b}$) and the contraction properties of the system are unaffected by the affine transformation. Note that the above equivalence holds even in the case where $\mathbf{W}$ is not invertible. In this case, the two models are proven to be equivalent, provided that $\mathbf{b}(0)$ and $\mathbf{y}(0)$ satisfy certain conditions—which are always possible to satisfy [19]. Therefore, any contraction condition derived for the $x$ (or $y$) system automatically implies contraction of the other system. We exploit this freedom later on in the paper. + +Our first theorem is motivated by the observation that if the y-system is to be interpreted as a vector of firing rates, it must stay positive for all time. For a linear, time-invariant system with positive states, diagonal stability is equivalent to stability. Therefore a natural question is if diagonal stability of a linearized y-system implies anything about stability of the nonlinear system. More formally, given an excitatory neural network (i.e $\forall ij, W_{ij} \ge 0$), if the linear system + +$$\dot{\mathbf{x}} = -\mathbf{x} + g\mathbf{W}\mathbf{x}$$ + +is stable, then there exists a positive diagonal matrix $\mathbf{P}$ such that: + +$$\mathbf{P}(g\mathbf{W} - \mathbf{I}) + (g\mathbf{W} - \mathbf{I})^T \mathbf{P} < 0$$ + +The following theorem shows that the nonlinear system (1) is indeed contracting in metric $\mathbf{P}$, and extends this result to a more general $\mathbf{W}$ by considering only the magnitudes of the weights. + +**Theorem 1.** Let $|\mathbf{W}|$ denote the matrix formed by taking the element-wise absolute value of $\mathbf{W}$. If there exists a positive, diagonal $\mathbf{P}$ such that: + +$$\mathbf{P}(g|\mathbf{W}| - \mathbf{I}) + (g|\mathbf{W}| - \mathbf{I})^T \mathbf{P} < 0$$ + +then (1) is contracting in metric $\mathbf{P}$. Moreover, if $W_{ii} \le 0$, then $|\mathbf{W}|_{ii}$ may be set to zero to reduce conservation. + +This condition is particularly straightforward in the common special case where the network does not have any self weights, with the leak term driving stability. While it can be applied to a more general $\mathbf{W}$, the condition will of course not be met if the network was relying on highly negative values on the diagonal of $\mathbf{W}$ for linear stability. As demonstrated by counterexample in the proof of Theorem 1, it can be impossible to use the same metric $\mathbf{P}$ for the nonlinear RNN in such cases. + +Theorem 1 allows many weight matrices with low magnitudes or a generally sparse structure to be verified as contracting in the nonlinear system (1), by simply checking a linear stability condition (as linear stability is equivalent to diagonal stability for Metzler matrices too [20]). This is particularly useful because sparsity is a recurring theme in neuroscience, both enabling stability in theory [14], and occurring in practice, with only ~0.00001% of potential connections forming synapses in the human brain [32]. + +Beyond verifying contraction, Theorem 1 actually provides a metric, with little need for additional computation. Not only is it of inherent interest that the same metric can be shared across systems in this case, it is also of use in machine learning applications, where stability certificates are becoming increasingly necessary. Critically, it is feasible to enforce the condition during training via L2 regularization on $\mathbf{W}$. More generally, there are a variety of systems of interest that meet this condition but do not meet the well-known maximum singular value condition, including those with a hierarchical structure. + +While regularization may push networks towards satisfying Theorem 1, strictly enforcing the condition during optimization is not straightforward. This motivates our next three theorems, which derive contraction results for specially structured weight matrices. Unlike Theorem 1, these results have direct parameterizations which can easily be plugged into modern optimization libraries. \ No newline at end of file diff --git a/samples/texts/3826450/page_19.md b/samples/texts/3826450/page_19.md new file mode 100644 index 0000000000000000000000000000000000000000..3dfdc68510742b8d213b905e7e7700de3d7da6d8 --- /dev/null +++ b/samples/texts/3826450/page_19.md @@ -0,0 +1,15 @@ +**Theorem 2.** If $W = W^T$ and $gW \prec I$, then (1) is contracting. + +Our proof works by constructing a constant metric for the system from the eigendecomposition of the weight matrix alone, and can be computed efficiently. The application of Theorem 2 is depicted with a representative simulation result in Figure 2. Moreover, as shown in [39], any symmetric matrix **W** that satisfies Theorem 2 may be written: + +$$g\mathbf{W} = \mathbf{I} - (\mathbf{V}^T\mathbf{V} + \epsilon\mathbf{I})$$ + +for some square **V** and $\epsilon > 0$. This direct parameterization can be used to enforce the contraction condition during optimization. + +**Remark 1** (Almost Necessary and Sufficient). For symmetric **W**, it was shown in [17] that $g\mathbf{W} - \mathbf{I} > 0$ implies that there exists a constant input **v** such that (1) has at least one unstable fixed point. Since the inputs enter affinely into the system, this implies that $g\mathbf{W} - \mathbf{I} < 0$ is in fact necessary as well as sufficient for global contraction in a constant metric. + +**Remark 2** (Connection to Transformer Networks). When **W** is symmetric, (1) may be seen as a continuous-time Hopfield network. Continuous-time Hopfield networks with symmetric weights were recently shown to be closely related to Transformer architectures [15, 24]. Specifically, the dot-product attention rule may be seen as a discretization of the continuous-time Hopfield network with softmax activation function [15]. Our results here provide a simple sufficient (and nearly necessary, see above remark) condition for global exponential stability of a given trajectory for the Hopfield network. In the case where the input into the network is constant, this trajectory is a fixed point. Moreover, each trajectory associated with a unique input is guaranteed to be unique. Finally, we note that our results are flexible with respect to activation functions so long as they satisfy the slope-restriction condition. This flexibility may be useful when, for example, considering recent work showing that standard activation functions may be advantageously replaced by attention mechanisms [7]. + +Figure 2: Top) The activation of a randomly chosen neuron, taken from a symmetric, contracting network with several different initial conditions. As expected, initial conditions are forgotten exponentially. Bottom) The maximum eigenvalue of the Jacobian taken from a single run, in the identity metric and in a valid contraction metric as constructed in Theorem 2. In the identity metric, the maximum eigenvalue exceeds zero, meaning that contraction cannot be concluded. In the valid metric, the maximum eigenvalue is always less than zero. + +Imposing symmetry on **W** may result in too strong of a stability condition for many purposes. The following two theorems provide contraction conditions for asymmetric weight matrices, which do not overtly constrain the magnitude of the weights as in Theorem 1. \ No newline at end of file diff --git a/samples/texts/3826450/page_20.md b/samples/texts/3826450/page_20.md new file mode 100644 index 0000000000000000000000000000000000000000..a287d004e07137c6a38466b7be8248bd0e97a3d6 --- /dev/null +++ b/samples/texts/3826450/page_20.md @@ -0,0 +1,29 @@ +**Theorem 3.** If there exists positive diagonal matrices $P_1$ and $P_2$, as well as $Q = Q^T > 0$ such that + +$$W = -P_1QP_2$$ + +then (1) is contracting in metric $M = (P_1QP_1)^{-1}$. + +As in the symmetric case, this condition is easy to enforce by simply directly parameterizing **W** as the product of the three matrices above, using the fact that any **Q** = **Q***T* > 0 may be written as **Q** = **V***T*V with **V** nonsingular. + +**Theorem 4.** If $g\mathbf{W} - \mathbf{I}$ is triangular and Hurwitz, then (1) is contracting in a diagonal metric. + +Note that in the case of a triangular weight matrix, the system (1) may be seen as a feedforward (i.e hierarchical) network. Therefore, this result follows from the combination properties of contracting systems. However, our proof provides a means of explicitly finding a metric for this system. + +The next theorem provides another continuous-time contraction condition for asymmetric $\mathbf{W}$, and also provides a direct parameterization in the context of learning/training, as well as application for Neural ODEs—which we discuss in section 2.2. + +**Theorem 5.** If there exists a positive diagonal matrix $\mathbf{P}$ such that: + +$$g^2 \mathbf{W}^T \mathbf{P} \mathbf{W} - \mathbf{P} < 0$$ + +then (1) is contracting in metric $\mathbf{P}$. + +Note that this is equivalent to the discrete-time diagonal stability condition developed in [25], for a constant metric. Note also that when $\mathbf{M} = \mathbf{I}$, Theorem 5 is identical to checking the maximum singular value of $\mathbf{W}$, a previously established condition for stability of (1). In Figure 3, we demonstrate how a much larger set of weight matrices are found via the condition when $\mathbf{M} = \mathbf{P}$ instead. + +Similar to Theorem 4 above, this contraction condition also enjoys a direct parameterization which may be used in training. Namely, the above condition on $\mathbf{W}$ is satisfied if $\mathbf{W}$ can be written as: + +$$\mathbf{W} = g^{-2}\mathbf{P}^{-1/2}\mathbf{U}\mathbf{S}\mathbf{V}^T\mathbf{P}^{1/2}$$ + +where $\mathbf{U}, \mathbf{V}$ are orthogonal matrices and $\mathbf{S}$ is a non-negative diagonal matrix with $0 \le S_{ii} < 1$. Note that the matrix $e^{\mathbf{N}}$ with $\mathbf{N} = -\mathbf{N}^T$ is an orthogonal matrix. This can be used to parameterize $\mathbf{U}$ and $\mathbf{V}$. + +Figure 3: Linear projection (using principal components analysis) of randomly generated matrices satisfying the stability condition in Theorem 5 with $\mathbf{M} = \mathbf{I}$ and $\mathbf{M} = \mathbf{P}$. \ No newline at end of file diff --git a/samples/texts/3826450/page_7.md b/samples/texts/3826450/page_7.md new file mode 100644 index 0000000000000000000000000000000000000000..dd0de9e92bf7b9896fe67bb5085b0e9278a69b5d --- /dev/null +++ b/samples/texts/3826450/page_7.md @@ -0,0 +1,59 @@ +**A.5 Proof of Theorem 5** + +**Theorem.** If there exists a positive diagonal matrix **P** such that: + +$$g^2 \mathbf{W}^T \mathbf{P} \mathbf{W} - \mathbf{P} < 0$$ + +then (1) is contracting in metric **P**. + +*Proof.* Consider the generalized Jacobian: + +$$\mathbf{F} = \mathbf{P}^{1/2} \mathbf{J} \mathbf{P}^{-1/2} = -\mathbf{I} + \mathbf{P}^{1/2} \mathbf{W} \mathbf{P}^{-1/2} \mathbf{D}$$ + +where $\mathbf{D}$ is a diagonal matrix with $D_{ii} = \frac{d\phi_i}{dx_i}$. Using the subadditivity of the matrix measure $\mu_2$ of the +generalized Jacobian we get: + +$$\mu_2(\mathbf{F}) \leq -1 + \mu_2(\mathbf{P}^{1/2}\mathbf{W}\mathbf{P}^{-1/2}\mathbf{D})$$ + +Now using the fact that $\mu_2(\cdot) \le || \cdot ||_2$ we have: + +$$\mu_2(\mathbf{F}) \le -1 + ||\mathbf{P}^{1/2}\mathbf{W}\mathbf{P}^{-1/2}\mathbf{D}||_2 \le -1 + g ||\mathbf{P}^{1/2}\mathbf{W}\mathbf{P}^{-1/2}||_2$$ + +Using the definition of the 2-norm, imposing the condition $\mu_2(\mathbf{F}) \le 0$ may be written: + +$$g^2 \mathbf{W}^T \mathbf{P} \mathbf{W} - \mathbf{P} < 0$$ + +which completes the proof. +□ + +**A.6 Proof of Theorem 6** + +**Theorem.** Let $\mathbf{D}$ be a positive, diagonal matrix with $D_{ii} = \frac{d\phi_i}{dx_i}$, and let $\mathbf{P}$ be an arbitrary, positive diagonal matrix. If: + +$$(g\mathbf{W} - \mathbf{I})\mathbf{P} + \mathbf{P}(g\mathbf{W}^T - \mathbf{I}) \preceq -c\mathbf{P}$$ + +and + +$$\dot{\mathbf{D}} - c g^{-1} \mathbf{D} \preceq -\beta \mathbf{D}$$ + +for $c, \beta > 0$, then (1) is contracting in metric $\mathbf{D}$ with rate $\beta$. + +*Proof.* Consider the differential, quadratic Lyapunov function: + +$$V = \delta\mathbf{x}^T \mathbf{P} \mathbf{D} \delta\mathbf{x}$$ + +where $\mathbf{D} > 0$ is as defined above. The time derivative of $V$ is: + +$$\dot{V} = \delta\mathbf{x}^T \mathbf{P} \dot{\mathbf{D}} \delta\mathbf{x} + \delta\mathbf{x}^T (-2\mathbf{P}\mathbf{D} + \mathbf{P}\mathbf{D}\mathbf{W}\mathbf{D} + \mathbf{D}\mathbf{W}^T\mathbf{D}\mathbf{P}) \delta\mathbf{x}$$ + +The second term on the right can be factored as: + +$$ +\begin{align*} +\delta\mathbf{x}^T(-2\mathbf{P}\mathbf{D} + \mathbf{P}\mathbf{D}\mathbf{W}\mathbf{D} + \mathbf{D}\mathbf{W}^T\mathbf{D}\mathbf{P})\delta\mathbf{x} &= \\ +&= \delta\mathbf{x}^T\mathbf{D}(-2\mathbf{P}\mathbf{D}^{-1} + \mathbf{P}\mathbf{W} + \mathbf{W}^T\mathbf{P})\mathbf{D}\delta\mathbf{x} &\le \\ +&= \delta\mathbf{x}^T\mathbf{D}(-2\mathbf{P}\mathbf{g}^{-1} + \mathbf{P}\mathbf{W} + \mathbf{W}^T\mathbf{P})\mathbf{D}\delta\mathbf{x} = \\ +&= \delta\mathbf{x}^T\mathbf{D}[\mathbf{P}(\mathbf{W}-\mathbf{g}^{-1}\mathbf{I})] + (\mathbf{W}^T-\mathbf{g}^{-1}\mathbf{I})\mathbf{P}]\mathbf{D}\delta\mathbf{x} &\le \\ +&= -c\mathbf{g}^{-1}\delta\mathbf{x}^T\mathbf{P}\mathbf{D}^2\delta\mathbf{x} +\end{align*} +$$ \ No newline at end of file diff --git a/samples/texts/3826450/page_8.md b/samples/texts/3826450/page_8.md new file mode 100644 index 0000000000000000000000000000000000000000..77328fb407624904f3beb6d8cc095fc88fe5a5a4 --- /dev/null +++ b/samples/texts/3826450/page_8.md @@ -0,0 +1,53 @@ +where the last inequality was obtained by substituting in the first assumption above. Combining this with +the expression for $\dot{V}$, we have: + +$$ +\dot{V} \leq \delta \mathbf{x}^T \mathbf{P} \dot{\mathbf{D}} \delta \mathbf{x} - c g^{-1} \delta \mathbf{x}^T \mathbf{P} \mathbf{D}^2 \delta \mathbf{x} +$$ + +Substituting in the second assumption, we have: + +$$ +\dot{V} \leq \delta \mathbf{x}^T \mathbf{P} (\dot{\mathbf{D}} - c g^{-1} \mathbf{D}^2) \delta \mathbf{x} \leq -\beta \delta \mathbf{x}^T \mathbf{P} \mathbf{D} \delta \mathbf{x} = -\beta V +$$ + +and thus $V$ converges exponentially to 0 with rate $\beta$. $\square$ + +**A.7 Proof of Theorem 7** + +**Theorem.** $g\mathbf{W}_{sym} - \mathbf{I} = g\frac{\mathbf{W}+\mathbf{W}^T}{2} - \mathbf{I} < 0$, i.e. contraction of the linear system in the identity metric, is not a sufficient condition for the general nonlinear system (1) to be contracting in a constant metric. High levels of antisymmetry in $\mathbf{W}$ can make it impossible to find such a metric, which we demonstrate via a 2 × 2 counterexample of the form + +$$ +\mathbf{W} = \begin{bmatrix} 0 & -c \\ c & 0 \end{bmatrix} +$$ + +with $c \ge 2$. + +*Proof.* The nonlinear system is globally contracting in a *constant* metric if there exists a symmetric, positive definite **M** such that the symmetric part of the Jacobian for the system, (**MWD**)$_{sym}$ – **M** is negative definite uniformly. Therefore (**MWD**)$_{sym}$ – **M** ≺ 0 must hold for all possible **D** if **M** is a constant metric the system *globally* contracts in with any allowed activation function, as some combination of settings to obtain a particular **D** can always be found. + +Thus to prove the main claim, we present here a simple 2-neuron system that is contracting in the iden- +tity metric with linear activation function, but can be shown to have no M that simultaneously satisfies the +(MWD)sym – M ≺ 0 condition for two different possible D matrices. + +To begin, take + +$$ +\mathbf{W} = \begin{bmatrix} 0 & -2 \\ 2 & 0 \end{bmatrix} +$$ + +Note that any off-diagonal magnitude ≥ 2 would work, as this is the point at which 1/2 of one of the weights (found in $\mathbf{W}_{sym}$ when the other is zeroed) will have magnitude too large for $(\mathbf{WD})_{sym} - \mathbf{I}$ to be stable. + +Looking at the linear system, we can see it is contracting in the identity because + +$$ +\mathbf{W}_{sym} - \mathbf{I} = \begin{bmatrix} -1 & 0 \\ 0 & -1 \end{bmatrix} < 0 +$$ + +Now consider $(\mathbf{MWD})_{sym} - \mathbf{M}$ with $\mathbf{D}$ taking two possible values of + +$$ +\mathbf{D}_1 = \begin{bmatrix} 1 & 0 \\ 0 & 0 \end{bmatrix} \quad \text{and} \quad \mathbf{D}_2 = \begin{bmatrix} 0 & 0 \\ 0 & 1 \end{bmatrix} +$$ + +We want to find some symmetric, positive definite **M** = [a m] such that (MWD₁)sym - **M** and +(MWD₂)sym - **M** are both negative definite. \ No newline at end of file diff --git a/samples/texts/3826450/page_9.md b/samples/texts/3826450/page_9.md new file mode 100644 index 0000000000000000000000000000000000000000..77021b94b0afff0f09fcbfa9e77cde30d1993237 --- /dev/null +++ b/samples/texts/3826450/page_9.md @@ -0,0 +1,57 @@ +Working out the matrix multiplication, we get + +$$ +(\mathbf{MWD}_1)_{sym} - \mathbf{M} = \begin{bmatrix} 2m-a & b-m \\ b-m & -b \end{bmatrix} +$$ + +and + +$$ +(\mathbf{MWD}_2)_{\mathrm{sym}} - \mathbf{M} = \begin{bmatrix} -a & -(a+m) \\ -(a+m) & -2m-b \end{bmatrix} +$$ + +We can now check necessary conditions for negative definiteness on these two matrices, as well as for positive definiteness on **M**, to try to find an **M** that will satisfy all these conditions simultaneously. In this process we will reach a contradiction, showing that no such **M** can exist. + +A necessary condition for positive definiteness in a real, symmetric $n \times n$ matrix $\mathbf{X}$ is $x_{ii} > 0$, and for negative definiteness $x_{ii} < 0$. Another well known necessary condition for definiteness of a real symmetric matrix is $|x_{ii} + x_{jj}| > |x_{ij} + x_{ji}| = 2|x_{ij}| \quad \forall i \neq j$. See [38] for more info on these conditions. + +Thus we will require *a* and *b* to be positive, and can identify the following conditions as necessary for our 3 +matrices to all meet the requisite definiteness conditions: + +$$ +2m < a \tag{5} +$$ + +$$ +-2m < b +\quad (6) +$$ + +$$ +|2m - (a+b)| > 2|b-m| \tag{7} +$$ + +$$ +|-2m-(a+b)| > 2|a+m| \qquad (8) +$$ + +Note that the necessary condition for $\mathbf{M}$ to be PD, $a + b > 2|m|$, is not listed, as it is automatically +satisfied if (5) and (6) are. + +It is easy to see that if $m = 0$, conditions (7) and (8) will result in the contradictory conditions $a > b$ and $b > a$ respectively, so we will require a metric with off-diagonal elements. To make the absolute values easier to deal with, we will check $m > 0$ and $m < 0$ cases independently. + +First we take $m > 0$. By condition (5) we must have $a > 2m$, so between that and knowing the signs of all unknowns are positive, we can reduce many of the absolute values. Condition (7) becomes $a+b-2m > |2b-2m|$, and condition (8) becomes $a+b+2m > 2a+2m$, which is equivalent to $b > a$. If $b > a$ we must also have $b > m$, so condition (7) further reduces to $a+b-2m > 2b-2m$, which is equivalent to $a > b$. Therefore we have again reached contradictory conditions. + +A very similar approach can be applied when $m < 0$. Using condition (6) and the known signs we reduce condition (7) to $2|m| + a + b > 2b + 2|m|$, i.e. $a > b$. Meanwhile condition (8) works out to $a + b - 2|m| > 2a - 2|m|$, i.e. $b > a$. + +Therefore it is impossible for a single constant **M** to accommodate both **D**₁ and **D**₂, so that no constant +metric can exist for **W** to be contracting in when a nonlinearity is introduced that can possibly have deriva- +tive reaching both of these configurations. One real world example of such a nonlinearity is ReLU. Given +a sufficiently high negative input to one of the units and a sufficiently high positive input to the other, **D** +can reach one of these configurations. The targeted inputs could then flip at any time to reach the other +configuration. + +An additional condition we could impose on the activation function is to require it to be a strictly increasing function, so that the activation function derivative can never actually reach 0. We will now show that a very similar counterexample applies in this case, by taking + +$$ +\mathbf{D}_{1*} = \begin{bmatrix} 1 & 0 \\ 0 & \epsilon \end{bmatrix} \quad \text{and} \quad \mathbf{D}_{2*} = \begin{bmatrix} \epsilon & 0 \\ 0 & 1 \end{bmatrix} +$$ \ No newline at end of file diff --git a/samples/texts/3953766/page_13.md b/samples/texts/3953766/page_13.md new file mode 100644 index 0000000000000000000000000000000000000000..bd1812ea7c68f0095556052d9bf4e4cd8cd149fd --- /dev/null +++ b/samples/texts/3953766/page_13.md @@ -0,0 +1,49 @@ +Assuming that, on part of its boundary, $\Gamma_N \subset \partial\Omega$, the body is subject to a surface force (traction) $\mathbf{f}_N$ that is also derivable from a potential function, we are interested in the minimisation of *total potential energy* + +$$E_t = E - \int_{\Gamma_N} \mathbf{f}_N \cdot \mathbf{u} \, dA, \qquad (B.2)$$ + +subject to $\det \mathbf{F} = \det \mathbf{G}$, over all displacement fields $\mathbf{u} = \mathbf{u}(\mathbf{X}) = \mathbf{x} - \mathbf{X}$, with gradient tensor $\nabla \mathbf{u} = \mathbf{F} - \mathbf{I}$, such that the kinematic constraint (2) holds. Following a standard procedure of computing variations $\delta E$ in $E$ for infinitesimal variations $\delta \mathbf{u}$ in $\mathbf{u}$, the following first variation in $E_t$ is obtained (see [66, p. 310-312]), + +$$ +\begin{align*} +\delta E_t &= \delta E - \int_{\Gamma_N} \mathbf{f}_N \cdot \delta \mathbf{u} \, dA \\ +&= \int_{\Omega} \mathbf{P}^{(nc)} : \delta \mathbf{F} \, dV - \int_{\Gamma_N} \mathbf{f}_N \cdot \delta \mathbf{u} \, dA \tag{B.3} \\ +&= - \int_{\Omega} \operatorname{Div} \mathbf{P}^{(nc)} : \delta \mathbf{u} \, dV, +\end{align*} +$$ + +where $\mathbf{P}^{(nc)}$ is the first Piola-Kirchhoff stress tensor (given explicitly by (A.7) in Appendix A). + +The *principle of stationary potential energy* then states that, of all admissible displacements fields that the body can assume, those for which the total potential energy assumes a stationary value, such that $\delta E_t = 0$, correspond to static equilibrium state described by + +$$-\operatorname{Div} \mathbf{P}^{(nc)}(\mathbf{X}) = 0, \qquad (B.4)$$ + +together with the traction boundary condition causing the deformation, $\mathbf{P}^{(nc)}\mathbf{N} = \mathbf{f}_N$, where $\mathbf{N}$ is the outward unit normal vector to the external surface $\Gamma_N$. + +In order to simplify the mechanical analysis, it is convenient to rewrite the above equations equivalently in terms of the elastic stresses for the base polymeric network. As the elastic deformation is applied directly to the reference state, using the relations between the stress tensors for the LCE, we obtain [63] + +$$ +\begin{align*} +\delta E &= \int_{\Omega} \mathbf{P}^{(nc)} : \delta \mathbf{F} \, dV \\ +&= \int_{\Omega} \mathbf{G}^{-1} \mathbf{P} : \delta (\mathbf{G}\mathbf{A}) \, dV \\ +&= \int_{\Omega} \mathbf{G}^{-1} \mathbf{A} \mathbf{S} : \delta (\mathbf{G}\mathbf{A}) \, dV \\ +&= \int_{\Omega} \mathbf{S} : [\ (\mathbf{G}^{-1}\mathbf{A})^T \delta (\mathbf{G}\mathbf{A}) ] \, dV, +\end{align*} +\qquad (B.5) +$$ + +where **P** and **S** are the Piola-Kirchhoff stress tensors given by (A.2) and (A.3), respectively. Hence, + +$$\delta E = \int_{\Omega} \mathbf{S} : \delta \mathbf{E} \, dV, \qquad (B.6)$$ + +where **E** is the Green-Lagrange strain tensor defined by (8). The equilibrium equation (B.4) is then equivalent to + +$$-\operatorname{Div} \mathbf{S}(\mathbf{X}) = 0, \qquad (B.7)$$ + +with the corresponding boundary condition $\mathbf{SN} = \mathbf{A}^{-1}\mathbf{G}\mathbf{f}_N$. + +The potential energy (B.1) can be expressed equivalently as [63] + +$$E = \int_{\Omega} [W(\mathbf{A}) - p(\det \mathbf{A} - 1)] dV, \qquad (B.8)$$ + +where $W(\mathbf{A})$ is the elastic strain-energy function on which the LCE strain-energy function $W^{(nc)}(\mathbf{F}, \mathbf{n})$ is based, and $p(\det \mathbf{A} - 1)$ enforces the incompressibility condition $\det \mathbf{A} = 1$. \ No newline at end of file diff --git a/samples/texts/3953766/page_14.md b/samples/texts/3953766/page_14.md new file mode 100644 index 0000000000000000000000000000000000000000..c6e129160d93d21e333875991b2aa3cc6c92d1bc --- /dev/null +++ b/samples/texts/3953766/page_14.md @@ -0,0 +1,27 @@ +# C Analysis at $q \to \infty$ + +The special limiting case where the wave number goes to infinity is interesting as it corresponds to the Biot instability and arises when the stiffness ratio $\beta = \mu_f/\mu_s$ is sufficiently small. In practice, this instability is not observed because creasing, folding, or a subcritical wrinkling instability occur before that limit. Nevertheless, it is an important reference point to organise the bifurcation diagram, and we include this here for completeness. + +## C.1 Hyperelastic film on LCE substrate + +When the bilayer system is formed by a hyperelastic film and a liquid crystal elastomer substrate, the entries of the 6 × 6 matrix **M** in (50) are given in Table 2. As $q \to \infty$, the coefficient of the leading term, containing $e^{qh(\lambda a^{1/6}+1)}$, must tend to zero. Hence, the determinant of the following 4 × 4 matrix obtained by dividing rows 3 and 4 by $\mu_s$, and removing columns 1 and 2, and rows 5 and 6 from the 6 × 6 matrix **M**, must vanish, + +$$ \overline{\mathbf{M}} = \begin{bmatrix} 1 & 1 & -1 & -1 \\ -1 & -a^{1/6}\lambda & -1 & -\frac{a^{1/6}\lambda}{g_1^2 g_2} \\ \frac{2\mu_f}{a^{1/3}} & \mu_f\left(\lambda^2 + \frac{1}{a^{1/3}}\right) & -\frac{2\mu_s g_1^2 g_2^2}{a^{1/3}} & -\mu_s\left(\frac{\lambda^2}{g_1^2} + \frac{g_1^2 g_2^2}{a^{1/3}}\right) \\ \mu_f\left(\lambda^2 + \frac{1}{a^{1/3}}\right) & \frac{2\mu_f\lambda}{a^{1/6}} & \mu_s\left(\frac{\lambda^2}{g_1^2} + \frac{g_1^2 g_2^2}{a^{1/3}}\right) & \frac{2\mu_s\lambda g_2}{a^{1/6}} \end{bmatrix}. \quad (\text{C.1}) $$ + +**Case 1.** First, we assume that the director is uniformly aligned in the second direction, such that $g_1 = a^{-1/6}$ and $g_2 = a^{1/3}$. For $q \to \infty$, the real solutions of equation det $\overline{\mathbf{M}} = 0$ are $\lambda = a^{-1/6}$ and $\lambda = \lambda_*$ satisfying equation + +$$ \lambda^3 \left( a^{1/2}\beta^2 + 2a^{5/6}\beta + a^{7/6} \right) + \lambda^2 \left( a^{1/3}\beta^2 + 6a^{2/3}\beta + a \right) \\ + \lambda \left( 3a^{1/6}\beta^2 - 2a^{1/2}\beta + 3a^{5/6} \right) - \beta^2 + 2a^{1/3}\beta - a^{2/3} = 0, \quad (\text{C.2}) $$ + +for all $0 < \beta \neq a^{1/3}$. + +**Case 2.** Second, we assume that the director is uniformly aligned in the first direction, such that $g_1 = a^{1/3}$ and $g_2 = a^{-1/6}$. For $q \to \infty$, the real solutions of equation det $\overline{\mathbf{M}} = 0$ are $\lambda = a^{-1/6}$, $\lambda = a^{1/3}$, and $\lambda = \lambda_*$ satisfying equation + +$$ \lambda^3 \left[ a^{5/6}\beta^2 + \beta \left( a^{2/3} + a^{1/6} \right) + 1 \right] + \lambda^2 \left[ a^{2/3}\beta^2 + \beta \left( a + 4a^{1/2} + 1 \right) + a^{1/3} \right] \\ + \lambda \left[ 3a^{1/2}\beta^2 - \beta \left( a^{1/3} + a^{5/6} \right) + 3a^{2/3} \right] - a^{1/3}\beta^2 + 2a^{2/3}\beta - a = 0, \quad (\text{C.3}) $$ + +for all $0 < \beta \neq a^{1/3}$. + +## C.2 LCE film on hyperelastic substrate + +For a bilayer system consisting of a liquid crystal elastomer film on a hyperelastic substrate, the entries of the 6 × 6 matrix **M** in (50) are given in Table 3. As $q \to \infty$, the coefficient of the leading term, containing $e^{qh(\lambda a^{1/6}+1)}$, must tend to zero. Hence, the determinant of the following 4 × 4 matrix obtained by dividing rows 3 and 4 by $\mu_s$, and removing columns 1 and 2, and rows 5 and 6 from the 6 × 6 matrix **M**, must be approximately zero, + +$$ \overline{\mathbf{M}} = \begin{bmatrix} 1 & 1 & -1 & -1 \\ -1 & -\frac{a^{1/6}\lambda}{g_1^2 g_2} & -1 & -a^{1/6}\lambda \\ \frac{2\mu_f g_1^2 g_2^2}{a^{1/3}} & \mu_f\left(\frac{\lambda^2}{g_1^2} + \frac{g_1^2 g_2^2}{a^{1/3}}\right) & -\frac{2\mu_s}{a^{1/3}} & -\mu_s\left(\lambda^2 + \frac{1}{a^{1/3}}\right) \\ \mu_f\left(\frac{\lambda^2}{g_1^2} + \frac{g_1^2 g_2^2}{a^{1/3}}\right) & \frac{2\mu_f\lambda g_2}{a^{1/6}} & \mu_s\left(\lambda^2 + \frac{1}{a^{1/3}}\right) & \frac{2\mu_s\lambda g_2}{a^{1/6}} \end{bmatrix}. \quad (\text{C.4}) $$ \ No newline at end of file diff --git a/samples/texts/3953766/page_15.md b/samples/texts/3953766/page_15.md new file mode 100644 index 0000000000000000000000000000000000000000..71467d7af1cde697b417264dbe140a7a047bef07 --- /dev/null +++ b/samples/texts/3953766/page_15.md @@ -0,0 +1,43 @@ +**Case 3.** First, the director is uniformly aligned in the second direction, such that $g_1 = a^{-1/6}$ and $g_2 = a^{1/3}$. For $q \to \infty$, the real solutions of equation det $\overline{\mathbf{M}} = 0$ are $\lambda = a^{-1/6}$ and $\lambda = \lambda_*^*$ satisfying equation + +$$ +\begin{aligned} +& \lambda^3 (a^{7/6}\beta^2 + 2a^{5/6}\beta + a^{1/2}) + \lambda^2 (a\beta^2 + 6a^{2/3}\beta + a^{1/3}) \\ +& \quad + \lambda (3a^{5/6}\beta^2 - 2a^{1/2}\beta + 3a^{1/6}) - a^{2/3}\beta^2 + 2a^{1/3}\beta - 1 = 0, +\end{aligned} +\quad (\text{C.5}) $$ + +for all $0 < \beta \neq a^{-1/3}$. + +**Case 4.** Second, the director is uniformly aligned in the first direction, such that $g_1 = a^{1/3}$ and $g_2 = a^{-1/6}$. For $q \to \infty$, the real solutions of equation det $\overline{\mathbf{M}} = 0$ are $\lambda = a^{-1/6}$, $\lambda = a^{1/3}$, and $\lambda = \lambda_*^*$ satisfying equation + +$$ +\begin{aligned} +& \lambda^3 [\beta^2 + \beta (a^{2/3} + a^{1/6}) + a^{5/6}] + \lambda^2 [a^{1/3}\beta^2 + \beta (a + 4a^{1/2} + 1) + a^{2/3}] \\ +& \quad + \lambda [3a^{2/3}\beta^2 - \beta (a^{1/3} + a^{5/6}) + 3a^{1/2}] - a\beta^2 + 2a^{2/3}\beta - a^{1/3} = 0, +\end{aligned} +\quad (\text{C.6}) $$ + +for all $0 < \beta \neq a^{-1/3}$. + +**Acknowledgement.** We gratefully acknowledge the support by the Engineering and Physical Sciences Research Council of Great Britain under research grants EP/R020205/1 to Alain Goriely and EP/S028870/1 to L. Angela Mihai. + +## References + +[1] Agrawal A, Luchette P, Palffy-Muhoray P, Biswal SL, Chapman WG, Verduzco R. 2012. Surface wrinkling in liquid crystal elastomers, Soft Matter 8(27), 7138-7142 (doi: 10.1039/c2sm25734c). + +[2] Agrawal A, Yun TH, Pesek SL, Chapman WG, Verduzco R. 2014. Shape-responsive liquid crystal elastomer bilayers, Soft Matter 10(9), 1411-1415 (doi: 10.1039/c3sm51654g). + +[3] Alawiye H, Farrell PE, Goriely A. 2020. Revisiting the wrinkling of elastic bilayers II: Post-bifurcation analysis, Journal of the Mechanics and Physics of Solids 143, 104053 (doi: 10.1016/j.jmps.2020.104053). + +[4] Alawiye H, Kuhl E, Goriely A. 2019. Revisiting the wrinkling of elastic bilayers I: Linear analysis, Philosophical Transactions of the Royal Society A 377, 20180076 (doi: 10.1098/rsta.2018.0076). + +[5] Anderson DR, Carlson DE, Fried E. 1999. A continuum-mechanical theory for nematic elastomers, Journal of Elasticity 56, 33-58 (doi: 10.1023/A:1007647913363). + +[6] Audoly B, Boudaoud A. 2008a. Buckling of a stiff film bound to a compliant substrate - Part I: Formulation, linear stability of cylindrical patterns, secondary bifurcations, Journal of the Mechanics and Physics of Solids 56, 2401-2421 (doi: 10.1016/j.jmps.2008.03.003). + +[7] Audoly B, Boudaoud A. 2008b. Buckling of a thin film bound to a compliant substrate - Part II: A global scenario for the formation of herringbone pattern, Journal of the Mechanics and Physics of Solids 56, 2422-2443 (doi: 10.1016/j.jmps.2008.03.002). + +[8] Audoly B, Boudaoud A. 2008c. Buckling of a thin film bound to a compliant substrate - Part III: Herringbone solutions at large buckling parameter, Journal of the Mechanics and Physics of Solids 56, 2444-2458 (doi: 10.1016/j.jmps.2008.03.001). + +[9] Bella P, Kohn RV. 2014. Metric-induced wrinkling of a thin elastic sheet, Journal of Nonlinear Science 24(6), 1147-1176 (doi: 10.1007/s00332-014-9214-9). \ No newline at end of file diff --git a/samples/texts/3953766/page_16.md b/samples/texts/3953766/page_16.md new file mode 100644 index 0000000000000000000000000000000000000000..92e7633da5b3187e796758e6a9894f455590ba39 --- /dev/null +++ b/samples/texts/3953766/page_16.md @@ -0,0 +1,35 @@ +[10] Ben Amar M, Bordner A. 2017. Mimicking cortex convolutions through the wrinkling of growing soft bilayers, Journal of Elasticity 129(1-2), 213-238 (doi: 10.1007/s10659-017-9622-9). + +[11] Ben Amar M, Ciarletta P. 2010. Swelling instability of surface-attached gels as a model of tissue growth under geometric constraints, Journal of the Mechanics and Physics of Solids 58(7), 935-954 (doi: 10.1016/j.jmps.2010.05.002). + +[12] Biot MA. 1937. Bending of an infinite beam on an elastic foundation, Journal of Applied Mechanics 203 (1936-37), A1-A7. + +[13] Biot MA. 1963. Surface instability of rubber in compression, Applied Scientific Research, Section A 12, 168182 (doi: 10.1007/BF03184638). + +[14] Bladon P, Terentjev EM, Warner M. 1994. Deformation-induced orientational transitions in liquid crystal elastomers, Journal de Physique II 4, 75-91 (doi: 10.1051/jp2:1994100). + +[15] Budday S, Kuhl E, Hutchinson JW. 2015. Period-doubling and period-tripling in growing bilayered systems, Philosophical Magazine 95(28-30), 3208-3224 (doi: 0.1080/14786435.2015.1014443). + +[16] Cai S, Breid D, Crosby AI, Suo Z, Hutchinson J W. 2011. Periodic patterns and energy states of buckled films on compliant substrates, Journal of the Mechanics and Physics of Solids 59, 1094-1114 (doi:10.1016/j.jmps.2011.02.001). + +[17] Cai ZX, Fu YB. 1999. On the imperfection sensitivity of a coated elastic half-space, Proceedings of the Royal Society A 455, 3285-3309 (doi: 10.1098/rspa.1999.0451). + +[18] Cai ZX, Fu YB. 2019. Effects of pre-stretch, compressibility and material constitution on the period-doubling secondary bifurcation of a film/substrate bilayer, International Journal of Non-Linear Mechanics 115, 11-19 (doi: 10.1016/j.ijnonlinmec.2019.05.001). + +[19] Carroll M. 2004. A representation theorem for volume-preserving transformations, International Journal of Non-Linear Mechanics 39(2), 219224 (doi: 10.1016/S0020-7462(02)00167-1). + +[20] Cao YP, Hutchinson JW. 2012. From wrinkles to creases in elastomers: the instability and imperfection-sensitivity of wrinkling, Proceedings of the Royal Society A 468, 94-115 (doi: 10.1098/rspa.2011.0384). + +[21] Cao YP, Hutchinson JW. 2012. Wrinkling phenomena in neo-Hookean film/substrate bilayers, Journal of Applied Mechanics 79(3), 031019 (doi: 10.1115/1.4005960). + +[22] Cesana P, Plucinsky P, Bhattacharya K. 2015. Effective behavior of nematic elastomer membranes, Archive for Rational Mechanics and Analysis 218(2), 863-905 (doi: 10.1007/s00205-015-0871-0). + +[23] Chung JY, Nolte AJ, Stafford CM. 2011. Surface wrinkling: A versatile platform for measuring think-film properties, Advanced Materials 23, 349-368 (doi: 10.1002/adma.201001759). + +[24] Chen X, Hutchinson JW. 2004. Herringbone buckling patterns of compressed thin films on compliant substrates, Journal of Applied Mechanics 71, 597-603 (doi: 10.1115/1.1756141). + +[25] Ciarletta P. 2011. Generating functions for volume-preserving transformations, International Journal of Non-Linear Mechanics 46(9), 1275-1279 (doi: 10.1016/j.ijnonlinmec.2011.07.001). + +[26] Ciarletta P. 2014. Wrinkle-to-fold transition in soft layers under equibiaxial strain: a weakly nonlinear analysis, Journal of the Mechanics and Physics of Solids 73, 118-133. + +[27] Ciarletta P, Fu YB. 2015. A semi-analytical approach to Biot instability in a growing layer: strain gradient correction, weakly non-linear analysis and imperfection sensitivity, International Journal of Non-Linear Mechanics 75, 3845 (doi: 10.1016/j.ijnonlinmec.2015.03.002). \ No newline at end of file diff --git a/samples/texts/3953766/page_18.md b/samples/texts/3953766/page_18.md new file mode 100644 index 0000000000000000000000000000000000000000..ae2d84600cf88e22f116e590ca9c00a840c6103e --- /dev/null +++ b/samples/texts/3953766/page_18.md @@ -0,0 +1,35 @@ +[45] Goriely A. 2017. The Mathematics and Mechanics of Biological Growth, Springer-Verlag, New York. + +[46] Holland MA, Li B, Feng XQ, Kuhl E. 2017. Instabilities of soft films on compliant substrates, Journal of the Mechanics and Physics of Solids 98, 350-365 (doi: 10.1016/j.jmps.2016.09.012). + +[47] Huang ZY, Hong W, Suo Z. 2005. Nonlinear analyses of wrinkles in a film bonded to a compliant substrate, Journal of the Mechanics and Physics of Solids 53, 2101-2118 (doi: 10.1016/j.jmps.2005.03.007). + +[48] Hutchinson JW. 2013. The role of nonlinear substrate elasticity in the wrinkling of thin films, Philosophical Transactions of the Royal Society A 371, 20120422 (doi: 10.1098/rsta.2012.0422). + +[49] Jiang ZC, Xiao YY, Zhao Y. 2019. Shining light on liquid crystal polymer networks: preparing, reconfiguring, and driving soft actuators, Advanced Optical Materials 7, 1900262 (doi: 10.1002/adom.201900262). + +[50] Kang SH, Na JH, Moon SN, Lee WI, Yoo PJ, Lee SD. 2012. Self-organized anisotropic wrinkling of molecularly aligned liquid crystalline polymer, Lagmuir 28, 3576-3582 (doi: 10.1021/la203799h). + +[51] Kohn RV, O'Brien E. 2018. The wrinkling of a twisted ribbon, Journal of Nonlinear Science 28, 1221-1249 (doi: 10.1007/s00332-018-9447-0). + +[52] Kowalski BA, Mostajeran C, Godman NP, Warner M, White TJ. 2017. Curvature by design and on demand in liquid crystal elastomers, Physical Review E 97, 012504. + +[53] Krieger MS, Dias MA. 2019. Tunable wrinkling of thin nematic liquid crystal elastomer sheets, Physical Review E 100, 022701 (doi: 10.1103/PhysRevE.100.022701). + +[54] Kuenstler AS. Hayward RC. 2019. Light-induced shape morphing of thin films, Current Opinion in Colloid & Interface Science 40, 70-86 (doi: 10.1016/j.cocis.2019.01.009). + +[55] Kundler I, Finkelmann H. 1995. Strain-induced director reorientation in nematic liquid single crystal elastomers, Macromolecular Rapid Communications 16, 679-686 (doi: 10.1002/marc.1995.030160908). + +[56] Lejeune E, Javili A, Linder C. 2016. Understanding geometric instabilities in thin films via a multi-layer model, Soft Matter 12, 806-816 (doi: 10.1039/c5sm02082d). + +[57] Lejeune E, Javili A, Weickenmeier J, Kuhl E, Linder C. 2016. Tri-layer wrinkling as a mechanism for anchoring center initiation in the developing cerebellum, Soft Matter 12, 5613-5620 (doi: 10.1039/C6SM00526H). + +[58] Li B, Cao Y-P, Feng X-Q, Gao H. 2012. Mechanics of morphological instabilities and surface wrinkling in soft materials: a review, Soft Matter 8, 5728-5745 (doi:10.1039/c2sm00011c). + +[59] Lubarda VA. 2004. Constitutive theories based on the multiplicative decomposition of deformation gradient: thermoelasticity, elastoplasticity and biomechanics, Applied Mechanics Reviews 57(2), 95-108 (doi: 10.1115/1.1591000). + +[60] McCracken JM, Donovan BR, White TJ. 2020. Materials as Machines, Advanced Materials 32, 1906564 (doi: 10.1002/adma.201906564). + +[61] Mihai LA, Goriely A. 2017. How to characterize a nonlinear elastic material? A review on nonlinear constitutive parameters in isotropic finite elasticity, Proceedings of the Royal Society A 473, 20170607 (doi: 10.1098/rspa.2017.0607). + +[62] Mihai LA, Goriely A. 2020. Likely striping in stochastic nematic elastomers, Mathematics and Mechanics of Solids 25(10), 1851-1872 (doi: 0.1177/1081286520914958). \ No newline at end of file diff --git a/samples/texts/3953766/page_19.md b/samples/texts/3953766/page_19.md new file mode 100644 index 0000000000000000000000000000000000000000..d2fd15faaf1553d7d958989204a699f5682a216e --- /dev/null +++ b/samples/texts/3953766/page_19.md @@ -0,0 +1,37 @@ +[63] Mihai LA, Goriely A. 2020. A plate theory for nematic liquid crystalline solids, Journal of the Mechanics and Physics of Solids 144, 104101 (doi: 10.1016/j.jmps.2020.104101). + +[64] Mihai LA, Goriely A. 2021. Instabilities in liquid crystal elastomers, Material Research Society (MRS) Bulletin (doi: 10.1557/s43577-021-00115-2). + +[65] Nayyar V, Ravi-Chandar K, Huang R. 2011. Stretch-induced stress patterns and wrinkles in hyperelastic thin sheets, International Journal of Solids and Structures 48(25-26), 3471-3483 (doi: 10.1016/j.ijsolstr.2011.09.004). + +[66] Ogden RW. 1997. Non-Linear Elastic Deformations, 2nd ed, Dover, New York. + +[67] Pang X, Lv J-a., Zhu C, Qin L, Yu Y. 2019. Photodeformable azobenzene-containing liquid crystal polymers and soft actuators, Advanced Materials, 1904224 (doi: 10.1002/adma.201904224). + +[68] Pipkin AC. 1986. The relaxed energy density for isotropic elastic membranes, IMA journal of Applied Mathematics 36(1), 8599 (doi: 10.1093/imamat/36.1.85). + +[69] Plucinsky P, Bhattacharya K. 2017. Microstructure-enabled control of wrinkling in nematic elastomer sheets, Journal of the Mechanics and Physics of Solids 102, 125-150 (doi: 10.1016/j.jmps.2017.02.009). + +[70] Rivlin RS, Saunders DW. 1951. Large elastic deformations of isotropic materials. VII. Experiments on the deformation of rubber, Philosophical Transactions of the Royal Society of London A 243(865), 251-288. + +[71] Šilhavý M. 2007. Ideally soft nematic elastomers, Networks Heterogeneous Media 2, 279-311. + +[72] Soni H, Pelcovits RA, Powers TR. 2016. Wrinkling of a thin film on a nematic liquid-crystal elastomer, Physical Review E 94, 012701 (doi: 10.1103/PhysRevE.94.012701). + +[73] Steigmann, DJ. 1990. Tension-field theory, Proceedings of the Royal Society A 29, 141-173 (doi: 10.1098/rspa.1990.0055). + +[74] Steigmann DJ, Pipkin AC. 1989. Finite deformations of wrinkled membranes, The Quarterly Journal of Mechanics and Applied Mathematics 42(3), 427-440 (doi: 10.1093/qjmam/42.3.427). + +[75] Sun JY, Xia S, Moon MW, Oh KH, Kim KS. 2012. Folding wrinkles of a thin stiff layer on a soft substrate, Proceedings of the Royal Society A 468, 932-953 (doi: 10.1098/rspa.2011.0567). + +[76] Talroze RV, Zubarev ER, Kuptsov SA, Merekalov AS, Yuranova TI, Plate NA, Finkelmann H. 1999. Liquid crystal acrylate-based networks: polymer backbone-LC order interaction, Reactive and Functional Polymers 41, 1-11 (doi: 10.1016/S1381-5148(99)00032-2). + +[77] Taylor M, Bertoldi K, Steigmann DJ. 2014. Spatial resolution of wrinkle patterns in thin elastic sheets at finite strain, Journal of the Mechanics and Physics of Solids 62, 163-180 (doi: 10.1016/j.jmps.2013.09.024). + +[78] Treloar LRG. 1944. Stress-strain data for vulcanized rubber under various types of deformation, Transactions of the Faraday Society 40, 59-70. + +[79] Treloar LRG. 2005. The Physics of Rubber Elasticity, 3rd ed, Oxford University Press, Oxford, UK. + +[80] Truesdell C, Noll W. 2004. The Non-Linear Field Theories of Mechanics, 3rd ed, Springer-Verlag, New York. + +[81] Ula SW, Traugutt NA, Volpe RH, Patel RP, Yu K, Yakacki CM. 2018. Liquid crystal elastomers: an introduction and review of emerging technologies, Liquid Crystals Review 6, 78-107 (doi: 10.1080/21680396.2018.1530155). \ No newline at end of file diff --git a/samples/texts/3953766/page_2.md b/samples/texts/3953766/page_2.md new file mode 100644 index 0000000000000000000000000000000000000000..57e08f10893fa234bcf5d4e0bcaba756b646fad6 --- /dev/null +++ b/samples/texts/3953766/page_2.md @@ -0,0 +1,39 @@ +following homogeneous system of eight linear equations in the eight unknown coefficients: + +$$ \zeta_f(0) = \zeta_s(0), \quad (42) $$ + +$$ \zeta'_{f}(0) = \zeta'_{s}(0), \quad (43) $$ + +$$ \frac{\mu_f}{\lambda_1^2 \lambda_2^2} \left( \frac{1}{q} \zeta''_{f}(0) + q \zeta_{f}(0) \right) = \frac{\mu_s}{\alpha_1^2 \alpha_2^2} \left( \frac{1}{q} \zeta''_{s}(0) + q \zeta_{s}(0) \right), \quad (44) $$ + +$$ \frac{\mu f}{q^2 \lambda_1^2 \lambda_2^2} \zeta'''_f(0) - \mu f \left( \lambda_1^2 + \frac{2}{\lambda_1^2 \lambda_2^2} \right) \zeta'_f(0) = \frac{\mu s}{q^2 \alpha_1^2 \alpha_2^2} \zeta'''_s(0) - \mu s \left( \alpha_1^2 + \frac{2}{\alpha_1^2 \alpha_2^2} \right) \zeta'_s(0), \quad (45) $$ + +$$ \frac{1}{q} \zeta''_{f}(h) + q \zeta_{f}(h) = 0, \quad (46) $$ + +$$ \frac{1}{q^2 \lambda_1^2 \lambda_2^2} \zeta'''_f(h) - \left( \lambda_1^2 + \frac{2}{\lambda_1^2 \lambda_2^2} \right) \zeta'_f(h) = 0, \quad (47) $$ + +$$ \frac{1}{q} \zeta''_{s}(-H) + q\zeta_{s}(-H) = 0, \quad (48) $$ + +$$ \frac{1}{q^2\alpha_1^2\alpha_2^2}\zeta'''_s(-H) - \left(\alpha_1^2 + \frac{2}{\alpha_1^2\alpha_2^2}\right)\zeta'_s(-H) = 0. \quad (49) $$ + +When the substrate is infinitely thick, the external conditions on the bottom surface $\Gamma_{3s}$ are replaced by the assumption that $u_i^{(s)} \to 0$, $i = 1, 2, 3$, as $x_3 \to -\infty$. In this case, $c_3^{(s)} = c_4^{(s)} = 0$, and this homogeneous algebraic system reduces to + +$$ M c = 0, \quad (50) $$ + +where $c = [c_1^{(f)}, c_2^{(f)}, c_3^{(f)}, c_4^{(f)}, c_1^{(s)}, c_2^{(s)}]^T$ and the entries of the 6×6 coefficient matrix $M = (M_{ij})_{i,j=1,\dots,6}$ are explicitly given in Table 1. + +Table 1: The coefficients $\{M_{ij}\}_{i,j=1,\dots,6}$ of the homogeneous system of linear equations (50). + +
Mijj = 1j = 2j = 3j = 4j = 5j = 6
i = 11111-1-1
i = 21λ12λ2-112λ2-112α2
i = 3(2μf12λ22)μf12 + 1/λ12λ22)(2μf12λ22)μf12 + 1/λ12λ22)(-2μs12α22)s12 + 1/α12α22)
i = 4-(μf12 + 1/λ12λ22))- (2μf2)μf12 + 1/λ12λ22)(2μf2)μs12 + 1/α12α22)(2μs2)
i = 5(eqhλ12λ212λ22) eqheqhλ12λ212 + 1/λ12λ22)(e-qhλ12λ212λ22) e-qhe-qhλ12λ212 + 1/λ12λ22)(e-qhλ12λ212λ22) e-qhλ12λ2(e-qhλ12λ212λ22) e-qhλ12λ2
i = 6-eqhλ1212+ 1/λ12λ22)- (e-qhλ12λ212λ22) e-qhλ12λ2(e-qhλ12λ212λ22) e-qhλ12λ2(e-qhλ12λ212λ22) e-qhλ12λ2(e-qhλ12λ21                                                                                                                      (50)
+ +For a system formed from a liquid crystal elastomer film and an elastic substrate, the stretch ratios $\{\lambda_i\}_{i=1,2,3}$ and $\{\alpha_i\}_{i=1,2,3}$ are swapped in (50). + +# 5 Wrinkling under biaxial stretch + +The general equations for the linear stability analysis in Section 4 are valid for all bilayer systems and deformations described in Section 3. As a particular application, we now examine the wrinkling of our bilayer system with an infinitely thick substrate, subject to biaxial stretch, such that [62] + +$$ \lambda_1 = \lambda, \quad \lambda_2 = \frac{a^{1/6}}{\lambda}, \quad \lambda_3 = \frac{1}{a^{1/6}} \quad (51) $$ + +and + +$$ \alpha_1 = \frac{\lambda_1}{g_1} = \frac{\lambda}{g_1}, \quad \alpha_2 = \frac{\lambda_2}{g_2} = \frac{a^{1/6}}{\lambda g_2}, \quad \alpha_3 = \frac{\lambda_3}{g_3} = 1. \quad (52) $$ \ No newline at end of file diff --git a/samples/texts/3953766/page_20.md b/samples/texts/3953766/page_20.md new file mode 100644 index 0000000000000000000000000000000000000000..44cebf7e02d2918d514ccb86682c2f4e6f62d0fd --- /dev/null +++ b/samples/texts/3953766/page_20.md @@ -0,0 +1,37 @@ +[82] Verwey GC, Warner M. 1995. Soft rubber elasticity, Macromolecules 28, 4303-4306. + +[83] Verwey GC, Warner M, Terentjev EM. 1996. Elastic instability and stripe domains in liquid crystalline elastomers, Journal de Physique II France 6(9), 1273-1290 (doi: 10.1051/jp2:1996130). + +[84] Wan G, Jin C, Trase I, Zhao S, Chen Z. 2018. Helical structures mimicking chiral seedpod opening and tendril coiling, Sensors 18(9), 2973 (doi: 10.3390/s18092973). + +[85] Wang M, Hu XB, Zuo B, Huang S, Chen XM, Yang H. 2020. Liquid crystal elastomer actuator with serpentine locomotion, Chemical Communications 56, 7597 (doi: 10.1039/d0cc02823a). + +[86] Warner M. 2020. Topographic mechanics and applications of liquid crystalline solids, Annual Review of Condensed Matter Physics 11, 125-145 (doi: 10.1146/annurev-conmatphys-031119-050738). + +[87] Warner M, Bladon P, Terentjev E. 1994. "Soft elasticity" - deformation without resistance in liquid crystal elastomers, Journal de Physique II France 4, 93-102 (doi: 10.1051/jp2:1994116). + +[88] Warner M, Gelling KP, Vilgis TA. 1988. Theory of nematic networks, The Journal of Chemical Physics 88, 4008-4013 (doi: 10.1063/1.453852). + +[89] Warner M, Terentjev EM. 1996. Nematic elastomers - a new state of matter?, Progress in Polymer Science 21, 853-891. + +[90] Warner M, Terentjev EM. 2007. Liquid Crystal Elastomers, paper back, Oxford University Press, Oxford, UK. + +[91] Warner M, Wang, XJ. 1991. Elasticity and phase behavior of nematic elastomers, Macromolecules 24, 4932-4941 (doi: 10.1021/ma00017a033). + +[92] Wen Z, Yang K, Raquez JM. 2020. A review on liquid crystal polymers in free-standing reversible shape memory materials, Molecules 25, 1241 (doi:10.3390/molecules25051241). + +[93] White TJ, Broer DJ. 2015. Programmable and adaptive mechanics with liquid crystal polymer networks and elastomers, Nature Materials 14, 1087-1098 (doi: 10.1038/nmat4433). + +[94] Xia Y, Honglawan A, Yang S. 2019. Tailoring surface patterns to direct the assembly of liquid crystalline materials, Liquid Crystals Reviews 7(1) 30-59 (doi: 10.1080/21680396.2019.1598295). + +[95] Xu F, Zhao S. 2020. Thermal wrinkling of liquid crystal polymer shell/core spheres, Extreme Mechanics Letters 40, 100860 (doi: 10.1016/j.eml.2020.100860). + +[96] Yang D, He LH. 2014. Photo-triggered wrinkling of glassy nematic films, Smart Materials and Structures 23, 045012 (doi: 10.1088/0964-1726/23/4/045012). + +[97] Yang D, He LH. 2015. Nonlinear analysis of photo-induced wrinkling of glassy twist nematic films on compliant substrates, Acta Mechanica Sinica 31(5), 672-678 (doi: 10.1007/s10409-015-0463-0). + +[98] Zhang Y, Xuan C, Jiang Y, Huo Y. 2019. Continuum mechanical modeling of liquid crystal elastomers as dissipative ordered solids, Journal of the Mechanics and Physics of Solids 126, 285-303 (doi: 10.1016/j.jmps.2019.02.018). + +[99] Zhang Y, Zhang Z, Huo Y. 2020. Nucleation and critical conditions for stripe domains in monodomain nematic elastomer sheets under uniaxial loading, Journal of the Mechanics and Physics of Solids 144, 104110 (doi: 10.1016/j.jmps.2020.104110). + +[100] Zhao S, Xu F, Fu C, Huo Y. 2019. Controllable wrinkling patterns on liquid crystal polymer film/substrate systems by laser illumination, Extreme Mechanics Letters 30, 100502 (doi: 10.1016/j.eml.2019.100502). \ No newline at end of file diff --git a/samples/texts/3953766/page_21.md b/samples/texts/3953766/page_21.md new file mode 100644 index 0000000000000000000000000000000000000000..b4a99e5ca37fcc7b1048c1b4611e8da5d23e923a --- /dev/null +++ b/samples/texts/3953766/page_21.md @@ -0,0 +1 @@ +[101] Zubarev ER, Kuptsov SA, Yuranova TI, Talroze RV, Finkelmann H. 1999. Monodomain liquid crystalline networks: reorientation mechanism from uniform to stripe domains, Liquid Crystals 26, 1531-1540 (doi: 10.1080/026782999203869). \ No newline at end of file diff --git a/samples/texts/3953766/page_22.md b/samples/texts/3953766/page_22.md new file mode 100644 index 0000000000000000000000000000000000000000..4de2850ebec8ed9e7af76eca7d7ba24533c62fd5 --- /dev/null +++ b/samples/texts/3953766/page_22.md @@ -0,0 +1,9 @@ +linear stability and a weekly nonlinear post-bifurcation analysis of a growing neo-Hookean film on a stiffer or softer neo-Hookean substrate, subjected to lateral compressive forces acting on the whole system were presented in [4] and [3], respectively. Notably, [3, 4, 10, 27] employed a stream function [11, 19, 25] operating on a mixture of the reference and deformed coordinate systems to replace the use of Lagrange multipliers associated with the incompressibility constraint. + +(III.2) *Models based on the three-dimensional finite elasticity.* For a growing soft layer subjected to equi-biaxial compression, in [26], a weakly nonlinear analysis of the wrinkling instability was performed using the multiple-scale perturbation method applied to the incremental theory in finite elasticity. + +Figure 1: Wrinkling parallel to the nematic director **n** when $a > 1$ and the system is at high temperature then cooled, so it extends along the director and contracts in the perpendicular direction [1, 2]. + +At the constitutive level, for ideal monodomain liquid crystal elastomers, where the mesogens are uniformly aligned throughout the material, a general formulation is provided by the phenomenological neoclassical strain-energy function proposed in [14, 88, 91]. This model is based on the molecular network theory of rubber elasticity [79], and its parameters are directly measurable experimentally or derived from macroscopic shape changes [89, 90]. Here, we adopt the stress-free state of the isotropic phase at high temperature as the reference configuration [28, 33–36, 62–64], rather than the nematic phase in which the cross-linking was produced [5, 14, 83, 87, 88, 91, 98]. Within this theoretical framework, the material deformation can be expressed as a composite deformation from a reference configuration to the current configuration, via an elastic distortion followed by a natural (stress free) shape change. The multiplicative decomposition of the associated gradient tensor is similar to those found in the constitutive theories of thermoelasticity, elastoplasticity, and growth [45, 59], but it is fundamentally different since, for nematic materials, the stress free geometrical change is superposed on the elastic deformation, which is applied directly to the reference state. This difference is important since, although the elastic configuration obtained by this deformation may not be observed in practice, it may still be possible for the nematic body to assume such a configuration under suitable external stimuli. The resulting elastic stresses can then be used to analyse the final deformation where the particular geometry also plays a role [63, 64]. + +In this study, we examine nonlinear elastic wrinkling of three-dimensional (3D) bilayer systems containing a monodomain liquid crystal elastomer layer and a homogeneous isotropic incompressible hyperelastic layer. First, we present a general linear instability analysis, which is valid for a range of bilayer systems and deformations. We then apply this analysis to a biaxial stretch combining elastic and natural deformations. For rubber-like material, experimental testing under biaxial (also known as “pure shear”) stretch has been carried out since the original work by Rivlin & Saunders (1951) [70]. When a nematic LCE sample is subjected to biaxial stretch, depending on the stretch ratios, the deformation induces phases of shear striping or wrinkling [22, 30, 34, 36, 62]. We consider the following two cases: a film of hyperlastic material attached to a liquid crystal elastomer substrate, and a liquid \ No newline at end of file diff --git a/samples/texts/3953766/page_23.md b/samples/texts/3953766/page_23.md new file mode 100644 index 0000000000000000000000000000000000000000..37329b3cec1aea1a1e37c836f3e60b54168462fa --- /dev/null +++ b/samples/texts/3953766/page_23.md @@ -0,0 +1,21 @@ +Figure 2: Wrinkling perpendicular to the nematic director **n** when $a < 1$ and the system is at low temperature then heated, so it contracts along the director and extends in the perpendicular direction [1, 2]. + +crystal elastomer film bonded to a hyperelastic substrate. In each case, assuming that the nematic director is uniformly aligned parallel to the film plane, and that biaxial forces act either parallel or perpendicular to the director field, we determine the critical wave number and stretch ratio for the onset of wrinkling. We find that: + +(i) When the liquid crystal elastomer is at high temperature then cooled, it extends along the director and contracts in the perpendicular direction causing wrinkling parallel to the director, as sketched in Figure 1; + +(ii) When the liquid crystal elastomer is at low temperature then heated, it contracts along the director and extends in the perpendicular direction causing wrinkling perpendicular to the director, as sketched in Figure 2; + +(iii) At constant temperature, if the nematic layer is compressed along the director, it can lead to reorientation of the director until it aligns perpendicular to the compressive force, after which, further compression produces wrinkling parallel to the rotated director. + +In Section 2, we present the constitutive description for LCEs. In Section 3, we introduce the bilayer system and the minimisation problem for its equilibrium state. Section 4 is devoted to the infinitesimal incremental perturbation of the displacement field with respect to a homogeneously deformed state. Critical values for the formation of wrinkles in specific bilayer systems under biaxial stretch are obtained in Section 5. In Section 6, we briefly consider the case when the film deformation is approximated using a plate theory, and demonstrate the validity of this approximation when the film is much stiffer than the substrate. In the final section, we draw concluding remarks and outline some potential future developments. + +## 2 The neoclassical model + +The neoclassical strain-energy density function describes an ideal liquid crystal elastomer and takes the form + +$$W^{(nc)}(\mathbf{F}, \mathbf{n}) = W(\mathbf{A}), \quad (1)$$ + +where **F** denotes the deformation gradient from the isotropic state, **n** is a unit vector field, known as the *director*, and $W(\mathbf{A})$ represents the strain-energy density of the isotropic polymer network, depending only on the (local) elastic deformation tensor **A**. The tensors **F** and **A** satisfy the relation (see Figure 3) + +$$\mathbf{F} = \mathbf{G}\mathbf{A}, \quad (2)$$ \ No newline at end of file diff --git a/samples/texts/3953766/page_24.md b/samples/texts/3953766/page_24.md new file mode 100644 index 0000000000000000000000000000000000000000..1a48fc457d595ae453715d046d398c467a969955 --- /dev/null +++ b/samples/texts/3953766/page_24.md @@ -0,0 +1,25 @@ +where + +$$ +\mathbf{G} = a^{1/3}\mathbf{n} \otimes \mathbf{n} + a^{-1/6}(\mathbf{I} - \mathbf{n} \otimes \mathbf{n}) = a^{-1/6}\mathbf{I} + (a^{1/3} - a^{-1/6})\mathbf{n} \otimes \mathbf{n}, \quad (3) +$$ + +is the spontaneous deformation tensor defining a change of frame of reference from the isotropic phase to a nematic phase. Note that, the spontaneous tensor **G** given by (3) is symmetric, i.e., **G** = **G***T* (where the superscript “*T*” denotes the transpose operation), but the elastic tensor **A** may not be symmetric. In (3), *a* > 0 is the temperature-dependent stretch parameter, ⊗ is the usual tensor product of two vectors, and **I** = diag(1, 1, 1) is the identity tensor. We assume that *a* is spatially-independent (i.e., no differential swelling). This parameter can be estimated by examining the thermal or light-induced response of the nematic elastomer with uniform planar alignment [52]. The ratio *r* = *a*1/3/(*a*-1/6) = *a*1/2 represents the anisotropy parameter, which, in an ideal nematic solid, is the same in all directions. In the nematic phase, both the cases with *r* > 1 (prolate molecules) and *r* < 1 (oblate molecules) are possible, while when *r* = 1, the energy function reduces to that of an isotropic hyperelastic material [32]. Monodomains can be synthesised with the parameter *a* taking values from 1.05 (glasses) to 60 (nematic rubber), which would correspond to changes in natural length between 7% and 400% (spontaneous extension ratio *a*1/3 between 1.02 and 4) [29]. + +Figure 3: Schematic of the composite deformation of a nematic solid. + +In (1), the elastic strain-energy function W is minimised by any deformation satisfying $\mathbf{A}\mathbf{A}^T = \mathbf{I}$ [66, 80], whereas the nematic strain-energy $W^{(nc)}$ is minimised by any deformation satisfying $\mathbf{F}\mathbf{F}^T = \mathbf{G}^2$. Hence, every pair $(\mathbf{G}\mathbf{R}, \mathbf{n})$, with $\mathbf{R}$ an arbitrary rigid-body rotation (i.e., $\mathbf{R}^{-1} = \mathbf{R}^T$ and $\det \mathbf{R} = 1$), is a natural (i.e., stress free) state for this material model. By (3), the following identity holds + +$$ +\mathbf{R}^T \mathbf{G} \mathbf{R} = a^{-1/6} \mathbf{I} + (a^{1/3} - a^{-1/6}) (\mathbf{R}^T \mathbf{n}) \otimes (\mathbf{R}^T \mathbf{n}). \quad (4) +$$ + +The director **n** is an observable (spatial) quantity. Denoting by **n**₀ the reference orientation of the local director corresponding to the cross-linking state, **n** may differ from **n**₀ both by a rotation and a change in *r*. + +We confine our attention to the case when the hyperelastic strain-energy function in (1) describes an incompressible neo-Hookean material [78], i.e., + +$$ +W(\mathbf{A}) = \frac{\mu}{2} [\operatorname{tr}(\mathbf{A}\mathbf{A}^T) - 3], \quad (5) +$$ + +where “tr” denotes the trace operator, and $\mu > 0$ represents the constant shear modulus at small strain. \ No newline at end of file diff --git a/samples/texts/3953766/page_25.md b/samples/texts/3953766/page_25.md new file mode 100644 index 0000000000000000000000000000000000000000..bdd9ace0275b458dc3b5f848a0f415e530dec5bd --- /dev/null +++ b/samples/texts/3953766/page_25.md @@ -0,0 +1,65 @@ +For a finite elastic deformation with gradient tensor **A**, the left and right Cauchy-Green tensors +are defined, respectively, by + +$$ +\mathbf{B} = \mathbf{A}\mathbf{A}^T \quad \text{and} \quad \mathbf{C} = \mathbf{A}^T\mathbf{A}. \tag{6} +$$ + +Using these deformation tensors, the elastic Almansi strain tensor is equal to [66, pp. 90-91] + +$$ +\mathbf{e} = \frac{1}{2} (\mathbf{I} - \mathbf{B}^{-1}), \qquad (7) +$$ + +and the elastic Green-Lagrange strain tensor is [66, pp. 89-90] + +$$ +\mathbf{E} = \frac{1}{2} (\mathbf{C} - \mathbf{I}) . \tag{8} +$$ + +Based on the hyperelastic model defined by (5), we construct the following strain-energy function of the form (1), + +$$ +W^{(nc)}(\mathbf{F}, \mathbf{n}) = \frac{\mu}{2} \left\{ a^{1/3} \left[ \operatorname{tr}(\mathbf{F}\mathbf{F}^T) - (1-a^{-1})\mathbf{n} \cdot \mathbf{F}\mathbf{F}^T\mathbf{n} \right] - 3 \right\}, \quad (9) +$$ + +where we assume that the nematic director is ‘free’ to rotate relative to the elastic matrix, i.e., **F** and +**n** are independent variables [28, 34, 36]. For this case, the relations between the stress tensors for the +LCE described by (9) and the base polymeric network modelled by (5) are presented in Appendix A +(see also [63]). In particular, the following additional condition related to the general lack of symmetry +of the Cauchy stress tensor in LCEs holds [5, 98], + +$$ +\frac{\partial W^{(nc)}}{\partial \mathbf{n}} = \mathbf{0}. \tag{10} +$$ + +We note that many macroscopic deformations of LCEs induce a director re-orientation, such that the +director becomes parallel to the direction of the largest principal stretch. + +**3 The nonlinear bilayer system** + +We now consider a system composed of two solid layers, namely a thin film and a thick substrate, +attached continuously to each other across a contact interface. In the undeformed reference configu- +ration, each body occupies a compact domain $\bar{\Omega}_k \subset \mathbb{R}^3$, $k \in \{f, s\}$, with the index “f” for the film +and “s” for the substrate, such that its interior is an open, bounded, connected set $\Omega_k \subset \mathbb{R}^3$ and its +boundary $\partial\Omega_k = \bar{\Omega}_k \setminus \Omega_k$ is Lipschitz continuous (in particular, we assume that a unit normal vector +$\mathbf{N}$ exists almost everywhere on $\partial\Omega_k$). We denote by $\Omega = \Omega_f \cup \Omega_s$ the domain occupied by this system, +such that $\Omega_f \cap \Omega_s = \emptyset$, with total boundary $\partial\Omega = \partial\Omega_f \cup \partial\Omega_s$, and by $\Gamma_C = \partial\Omega_f \cap \partial\Omega_s \subset \partial\Omega$ the +open subset representing the interface between the two layers. The external boundary $\Gamma = \partial\Omega \setminus \Gamma_C$ +is partitioned as $\Gamma = \Gamma_D \cup \Gamma_N$, such that on $\Gamma_D$, essential (displacement) boundary conditions are +prescribed, while on $\Gamma_N$, natural (traction) boundary conditions are imposed. + +We denote by **X** = (X₁, X₂, X₃) and **x** = (x₁, x₂, x₃) the Cartesian coordinates for the reference +and the deformed state, respectively. We then designate the plane formed by the first two directions +as the film's plane and the third direction as the film's thickness direction, and take Ωf = (-L1, L1) × +(-L2, L2) × (0, h), Ωs = (-L1, L1) × (-L2, L2) × (-H, 0) and ΓC = (-L1, L1) × (-L2, L2) × {0}, such +that the thickness of the film is much smaller than that of the substrate, i.e., h ≪ H. + +We assume that each layer is made of either a purely elastic material characterised by a neo- +Hookean strain-energy function of the form (5), or a liquid crystal elastomer described by a neoclassical +strain-energy function of the form (9). Therefore, in the reference configuration, the material in each +layer is stress-free and isotropic. We denote by μf and μs the respective constitutive coefficients, and +their ratio by β = μfs. + +For the two-layer system, the kinematically admissible displacement fields u = u(X) = x − X have +gradient tensor ∇u = F(u) − I, and satisfy the incompressibility condition det F(u) = 1, the essential \ No newline at end of file diff --git a/samples/texts/3953766/page_26.md b/samples/texts/3953766/page_26.md new file mode 100644 index 0000000000000000000000000000000000000000..9125c4ce6a900f5500c31f36488af69a3831428f --- /dev/null +++ b/samples/texts/3953766/page_26.md @@ -0,0 +1,48 @@ +boundary conditions on $\Gamma_D$, and the continuity condition $[\mathbf{u}] = \mathbf{u}_+ - \mathbf{u}_- = 0$ across the interface $\Gamma_C$. +In addition, for the LCE layer, the kinematic constraint (2) holds. + +For the hyperelastic material, we define the function + +$$ \Psi_k(\mathbf{F}(\mathbf{u}), p_k) = W_k(\mathbf{F}(\mathbf{u})) - p_k(\det \mathbf{F}(\mathbf{u}) - 1), \quad (11) $$ + +where $W_k(\mathbf{F}(\mathbf{u}))$ takes the form (5) with $\mu = \mu_k$ and $\mathbf{A} = \mathbf{F}(\mathbf{u}) = \mathbf{I} + \nabla \mathbf{u}$, and $p_k$ is the Lagrange multiplier for the incompressibility constraint $\det \mathbf{F}(\mathbf{u}) = 1$. + +Similarly, for the LCE, we define + +$$ \Psi_k (\mathbf{F}(\mathbf{u}), \mathbf{n}, p_k) = W_k^{(nc)} (\mathbf{F}(\mathbf{u}), \mathbf{n}) - p_k (\det \mathbf{F}(\mathbf{u}) - 1), \quad (12) $$ + +where $W_k^{(nc)}(\mathbf{F}, \mathbf{n})$ is given by (9) with $\mu = \mu_k$ and $\mathbf{F} = \mathbf{F}(\mathbf{u}) = \mathbf{I} + \nabla\mathbf{u}$, and $p_k$ is the Lagrange multiplier for the incompressibility constraint $\det \mathbf{F}(\mathbf{u}) = \det \mathbf{G} = 1$. + +The elastic energy stored by each individual body is equal to (see Appendix B for details) + +$$ E_k = \int_{\Omega_k} \Psi_k dV, \quad k \in \{f, s\}, \qquad (13) $$ + +with $\Psi_k$ described by either (11) or (12). Thus, the energy functional is $E_k = E_k(\mathbf{u}, p_k)$ for the hyperelastic material, and $E_k = E_k(\mathbf{u}, \mathbf{n}, p_k)$ for the liquid crystal elastomer. + +Assuming that the two layers are subjected to dead-load traction $f_N$ on $\Gamma_N \subset \partial\Omega \setminus \Gamma_C$, we are interested in the minimisation of total potential energy + +$$ E_t = E_f + E_s - \int_{\Gamma_N} \mathbf{f}_N \cdot \mathbf{u} \, dA \quad (14) $$ + +over all kinematically admissible displacement fields $\mathbf{u} = [u_1, u_2, u_3]^T$. + +Recalling that (10) holds, the eight Euler-Lagrange equations for the minimisation problem in $\Omega_k$, $k \in \{f, s\}$ are: + +$$ \frac{\partial}{\partial X_1} \left( \frac{\partial \Psi_k}{\partial u_{1,1}} \right) + \frac{\partial}{\partial X_2} \left( \frac{\partial \Psi_k}{\partial u_{1,2}} \right) + \frac{\partial}{\partial X_3} \left( \frac{\partial \Psi_k}{\partial u_{1,3}} \right) = 0, \quad (15) $$ + +$$ \frac{\partial}{\partial X_1} \left( \frac{\partial \Psi_k}{\partial u_{2,1}} \right) + \frac{\partial}{\partial X_2} \left( \frac{\partial \Psi_k}{\partial u_{2,2}} \right) + \frac{\partial}{\partial X_3} \left( \frac{\partial \Psi_k}{\partial u_{2,3}} \right) = 0, \quad (16) $$ + +$$ \frac{\partial}{\partial X_1} \left( \frac{\partial \Psi_k}{\partial u_{3,1}} \right) + \frac{\partial}{\partial X_2} \left( \frac{\partial \Psi_k}{\partial u_{3,2}} \right) + \frac{\partial}{\partial X_3} \left( \frac{\partial \Psi_k}{\partial u_{3,3}} \right) = 0, \quad (17) $$ + +$$ \det (\mathbf{I} + \nabla \mathbf{u}) - 1 = 0, \quad (18) $$ + +where $u_{i,j} = \partial u_i / \partial X_j$, $i, j = 1, 2, 3$. These equations are complemented by the following conditions across the interface: + +* On $\Gamma_C = (-L_1, L_1) \times (-L_2, L_2) \times \{0\}$, the jump in displacement is equal to zero, i.e., + +$$ [\mathbf{u}] = 0, \quad (19) $$ + +and the surface tractions are equal in magnitude and opposite in orientation, i.e., + +$$ \frac{\partial\Psi_f}{\partial u_{i,3}} = \frac{\partial\Psi_s}{\partial u_{i,3}}, \quad i=1,2,3. \quad (20) $$ + +Equations (15)-(18) and contact conditions (19)-(20) are completed by the following external boundary conditions: \ No newline at end of file diff --git a/samples/texts/3953766/page_27.md b/samples/texts/3953766/page_27.md new file mode 100644 index 0000000000000000000000000000000000000000..3fcf82e80b273d20afac2c7d28d02ef2f3546d32 --- /dev/null +++ b/samples/texts/3953766/page_27.md @@ -0,0 +1,41 @@ +* On the side surfaces $\Gamma_{1f} = \{-L_1, L_1\} \times (-L_2, L_2) \times (0, h)$ and $\Gamma_{1s} = \{-L_1, L_1\} \times (-L_2, L_2) \times (-H, 0)$: + +$$u_1 = (\lambda_1 - 1) X_1, \quad \frac{\partial \Psi_k}{\partial u_{i,1}} = 0, \quad i = 2,3, \quad k \in \{f,s\}, \qquad (21)$$ + +where $\lambda_1 > 0$ is fixed. + +* On the front and back surfaces $\Gamma_{2f} = (-L_1, L_1) \times \{-L_2, L_2\} \times (0, h)$ and $\Gamma_{2s} = (-L_1, L_1) \times \{-L_2, L_2\} \times (-H, 0)$: + +$$u_2 = (\lambda_2 - 1) X_2, \quad \frac{\partial \Psi_k}{\partial u_{i,2}} = 0, \quad i = 1,3, \quad k \in \{f,s\}, \qquad (22)$$ + +where $\lambda_2 > 0$ is fixed. + +* The top and bottom surfaces $\Gamma_{3f} = (-L_1, L_1) \times (-L_2, L_2) \times \{h\}$ and $\Gamma_{3s} = (-L_1, L_1) \times (-L_2, L_2) \times \{-H\}$ are traction free, i.e., + +$$\frac{\partial \Psi_k}{\partial u_{i,3}} = 0, \quad i = 1,2,3, \quad k \in \{f,s\}. \qquad (23)$$ + +In addition to the specified contact and boundary conditions, as a general rule, at a corner where one of the adjacent edges is subject to essential conditions and the other to natural conditions, the essential conditions take priority, and when both edges meeting at a corner are subject to natural conditions, these conditions are imposed simultaneously at the corner. Also, in order to avoid rigid-body rotations of the entire system, the lower left-hand corner is clamped, i.e., **u** = **0** at that corner. + +Assuming that, for the LCE body, the director aligns uniformly in the first or second direction, we denote by **G** = diag($g_1$, $g_2$, $1/(g_1g_2)$) the spontaneous deformation tensor. It can be verified that the homogeneous displacement + +$$\mathbf{u}^{(0)} = \left( \lambda_1 - 1, \lambda_2 - 1, \frac{1}{\lambda_1 \lambda_2} - 1 \right), \qquad (24)$$ + +together with the following Lagrange multipliers, + +$$p_k^{(0)} = \mu_k \lambda_1^{-2} \lambda_2^{-2} \quad \text{for the elastic layer,} \qquad (25)$$ + +$$p_k^{(0)} = \mu_k \alpha_1^{-2} \alpha_2^{-2} \quad \text{for the LCE layer,} \qquad (26)$$ + +where $\alpha_1 = \lambda_1/g_1$ and $\alpha_2 = \lambda_2/g_2$, satisfies equations (15)-(18), contact conditions (19)-(20) and boundary conditions (21)-(23). + +# 4 Incremental perturbation + +Next, we assume an infinitesimal incremental perturbation of the displacement field with respect to the homogeneously deformed state described by (24)-(26). The perturbed displacement field takes the form + +$$\mathbf{u} = \mathbf{u}^{(0)} + \epsilon\mathbf{u}^{(k)} \quad \text{in } \Omega_k, \quad k \in \{f,s\}, \qquad (27)$$ + +where $\mathbf{u}^{(0)}$ is the displacement corresponding to the homogeneously deformed state, $\mathbf{u}^{(k)}$ is the infinitesimal incremental displacement, and $0 < \epsilon \ll 1$ is a small parameter. The associated Lagrange multipliers are equal to + +$$\mathbf{p} = \mathbf{p}_k^{(0)} + \epsilon\mathbf{p}^{(k)} \quad \text{in } \Omega_k, \quad k \in \{f,s\}. \qquad (28)$$ + +For the incremental perturbation, the continuous contact and external boundary conditions are as follows: \ No newline at end of file diff --git a/samples/texts/3953766/page_28.md b/samples/texts/3953766/page_28.md new file mode 100644 index 0000000000000000000000000000000000000000..ee33862ccf1395c0c7493bf21fd6ce023472e04b --- /dev/null +++ b/samples/texts/3953766/page_28.md @@ -0,0 +1,45 @@ +• On $\Gamma_C = (-L_1, L_1) \times (-L_2, L_2) \times \{0\}$: + +$$ u_1^{(f)} = u_1^{(s)}, \quad u_2^{(f)} = u_2^{(s)}, \quad u_3^{(f)} = u_3^{(s)}, \qquad (29) $$ + +$$ \frac{\mu_f}{\lambda_1^2 \lambda_2^2} (u_{1,3}^{(f)} + u_{3,1}^{(f)}) = \frac{\mu_s}{\alpha_1^2 \alpha_2^2} (u_{1,3}^{(s)} + u_{3,1}^{(s)}), \qquad (30) $$ + +$$ \frac{\mu_f}{\lambda_1^2 \lambda_2^2} (u_{2,3}^{(f)} + u_{3,2}^{(f)}) = \frac{\mu_s}{\alpha_1^2 \alpha_2^2} (u_{2,3}^{(s)} + u_{3,2}^{(s)}), \qquad (31) $$ + +$$ \frac{2\mu_f}{\lambda_1^2 \lambda_2^2} (u_{1,1}^{(f)} + u_{2,2}^{(f)}) + p^{(f)} = \frac{2\mu_s}{\alpha_1^2 \alpha_2^2} (u_{1,1}^{(s)} + u_{2,2}^{(s)}) + p^{(s)}. \qquad (32) $$ + +• The top surface $\Gamma_{3f} = (-L_1, L_1) \times (-L_2, L_2) \times \{h\}$ is free, i.e., + +$$ u_{1,3}^{(f)} + u_{3,1}^{(f)} = 0, \quad u_{2,3}^{(f)} + u_{3,2}^{(f)} = 0, \quad \frac{2\mu_f}{\lambda_1^2 \lambda_2^2} (u_{1,1}^{(f)} + u_{2,2}^{(f)}) + p^{(f)} = 0. \qquad (33) $$ + +• The bottom surface $\Gamma_{3s} = (-L_1, L_1) \times (-L_2, L_2) \times \{-H\}$ is free, i.e., + +$$ u_{1,3}^{(s)} + u_{3,1}^{(s)} = 0, \quad u_{2,3}^{(s)} + u_{3,2}^{(s)} = 0, \quad \frac{2\mu_s}{\alpha_1^2 \alpha_2^2} (u_{1,1}^{(s)} + u_{2,2}^{(s)}) + p^{(s)} = 0. \qquad (34) $$ + +We search for solutions of the form + +$$ u_1^{(k)} = -\frac{1}{q}\zeta_k'(x_3)\sin(qx_1), \quad u_2^{(k)} = 0, \quad u_3^{(k)} = \zeta_k(x_3)\cos(qx_1), \quad k \in \{f, s\}, \qquad (35) $$ + +where the prime denotes differentiation and $q > 0$ is the wave number. + +We first take a system consisting of a hyperelastic film and a liquid crystal elastomer substrate [2]. Then the opposite configuration of a liquid crystal elastomer film atop a hyperelastic substrate can be directly obtained by swapping variables, as will be explained later. + +From the Euler-Lagrange equations, we obtain the following fourth-order ordinary differential equations, + +$$ \zeta_f^{iv} - \zeta_f'' q^2 (\lambda_1^4 \lambda_2^2 + 1) + \zeta_f q^4 \lambda_1^4 \lambda_2^2 = 0, \qquad (36) $$ + +$$ \zeta_s^{iv} - \zeta_s'' q^2 (\alpha_1^4 \alpha_2^2 + 1) + \zeta_s q^4 \alpha_1^4 \alpha_2^2 = 0, \qquad (37) $$ + +where $\zeta'_k$, $\zeta''_k$, $\zeta'''_k$, and $\zeta^{iv}_k$ are the derivatives of order 1, 2, 3 and 4 of $\zeta_k$ with respect to $x_3$, respectively, and $k \in \{f,s\}$. The corresponding Lagrange multipliers take the form + +$$ p^{(f)} = \left( \frac{\mu_f}{q^2 \lambda_1^2 \lambda_2^2} \zeta_f''' - \mu_f \lambda_1^2 \zeta_f' \right) \cos(qx_1), \qquad (38) $$ + +$$ p^{(s)} = \left( \frac{\mu_s}{q^2 \alpha_1^2 \alpha_2^2} \zeta_s''' - \mu_s \alpha_1^2 \zeta_s' \right) \cos(qx_1). \qquad (39) $$ + +When $\lambda_1^2 \lambda_2 \neq 1$, the following general solutions satisfy equations (36) and (37), respectively, + +$$ \zeta_f = c_1^{(f)} e^{qx_3} + c_2^{(f)} e^{q\lambda_1^2\lambda_2 x_3} + c_3^{(f)} e^{-qx_3} + c_4^{(f)} e^{-q\lambda_1^2\lambda_2 x_3}, \qquad (40) $$ + +$$ \zeta_s = c_1^{(s)} e^{qx_3} + c_2^{(s)} e^{q\alpha_1^2\alpha_2 x_3} + c_3^{(s)} e^{-qx_3} + c_4^{(s)} e^{-q\alpha_1^2\alpha_2 x_3}. \qquad (41) $$ + +To determine the eight unknown coefficients, {$c_i^{(f)}, c_i^{(s)}$}_{i=1,...,4}, in the above equations, we substitute expressions (40) and (41) in the contact and external boundary conditions (29)-(34). We obtain the \ No newline at end of file diff --git a/samples/texts/3953766/page_6.md b/samples/texts/3953766/page_6.md new file mode 100644 index 0000000000000000000000000000000000000000..b3e120fb5b5e1a5891f928cf601908e95b979504 --- /dev/null +++ b/samples/texts/3953766/page_6.md @@ -0,0 +1,6 @@ +Figure 5: Case 2 (elastic film on LCE substrate, with director aligned in the first direction): The critical stretch ratio $0 < \lambda_{cr} < \min\{a^{1/12}, a^{1/3}\}$ and critical wave number $q_{cr}$ as functions of relative stiffness ratio $\beta = \mu_f/\mu_s$ when $g_1 = a^{1/3}$ and $g_2 = a^{-1/6}$. For different magnitudes of the nematic parameter $a$, the solid lines represent critical values, while the dashed lines show the value $\min\{a^{1/12}, a^{1/3}\}$, respectively. + +Table 3: The coefficients $\{M_{ij}\}_{i,j=1,\dots,6}$ of the homogeneous system of linear equations (50), for a liquid crystal elastomer film on a hyperelastic substrate, with $\{\lambda_i\}_{i=1,2,3}$ given by (51) and $\{\alpha_i\}_{i=1,2,3}$ given by (52). + +
Mijj = 1j = 2j = 3j = 4j = 5j = 6
i = 11111-1-1
i = 21a1/6λ
g12g2
-1-a1/6λ
g12g2
-1-a1/6λ
i = 3fg12g22/a1/3μfi2 + g12g22/a1/3)fg12g22/a1/3μfi2 + g12g22/a1/3)-2μs/a1/3s2 + 1/a1/3)
i = 4fi2 + g12g22/a1/3)-2μfλg2/a1/6μfi2 + g12g22/a1/3)fλg2/a1/6μs2 + 1/a1/3)sλ/a1/6
i = 5(2g12g22/a1/3)eqi2 + g12g22/a1/3)eqλa1/6/(g12g2)(2g12g22/a1/3)e-qi2 + g12g22/a1/3)e-qλa1/6/(g12g2)00
i = 6-(λi2 + g12g22/a1/3)e-q-2e-qλa1/6/(g12g2)i2 + g12g22/a1/3)e-q(2e-qλa1/6/(g12g2)) e-qλa1/6(g12+g22) / (g12+g22) e-qλa-6/30/7/30/7/30/7/30/7/30/7/30/7/30/7/30/7/30/7/30/7/30/7/30/7/30/7/30/7/30/7/30/7/30/7/30/7/30/7/30/7/30/7/30/7/30/7/30/7/30/7/30/7/30/7/30/7/30/7/30/7/30/7/30/7/30/7/30/7/30/7/30/7/30/7/30/7/30/7/30/7/30/7/30/7/30/7/30/7/30/7/30/7/30/7/30/7/30/7/30/7/30/7/30/7/30/7/30/7/30/7/30/7/30/7/30/7/30/7/30/7/30/7/30/7/30/7/30/7/30/7/30/7/30/7/30/7/30/7/30/7/30/7/30/7/30/7/30/7/30/7/30/7/30/7/30/7/30/7/30/7/30/7/30/7/30//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//7//8888888888888888888888888888888888888888888888888888888888888888888888888888888888888888889999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999999\\end{table} + diff --git a/samples/texts/3953766/page_7.md b/samples/texts/3953766/page_7.md new file mode 100644 index 0000000000000000000000000000000000000000..ea9b176b1a2c800fc378dab1523aa44b602d9901 --- /dev/null +++ b/samples/texts/3953766/page_7.md @@ -0,0 +1,21 @@ +**Case 3.** First, the director is initially aligned in the second direction, such that $g_1 = a^{-1/6}$ and $g_2 = a^{1/3}$. The asymptotic limit of large stiffness ratio $\beta = \mu_f/\mu_s$ gives + +$$ \lambda_{\text{cr}} = a^{-1/6} - \frac{a^{-7/18}}{2} \left(\frac{3}{\beta}\right)^{2/3} + \mathcal{O}\left(\beta^{-4/3}\right), \quad (63) $$ + +$$ q_{\text{cr}} = a^{-1/9} \left( \frac{3}{\beta} \right)^{1/3} + \mathcal{O} \left( \beta^{-1} \right). \qquad (64) $$ + +Figure 6: Case 3 (LCE film, with director aligned in the second direction, on elastic substrate): The critical stretch ratio $0 < \lambda_{\text{cr}} < \min\{a^{1/12}, a^{1/3}\}$ and critical wave number $q_{\text{cr}}$ as functions of relative stiffness ratio $\beta = \mu_f/\mu_s$ when $g_1 = a^{-1/6}$ and $g_2 = a^{1/3}$. For different magnitudes of the nematic parameter $a$, the solid lines represent critical values, while the dashed lines show the value $\min\{a^{1/12}, a^{1/3}\}$, respectively. + +For finite values of $0 < q < \infty$ (see Appendix C for $q \to \infty$), Figure 6 displays the critical stretch ratios $\lambda_{\text{cr}}$ and critical wave numbers $q_{\text{cr}}$ as functions of the stiffness ratio $\beta$: + +* When $a < 1$, wrinkles parallel to the director form at the critical stretch ratio $0 < \lambda_{\text{cr}} < a^{1/3}$; + +* When $a > 1$, wrinkles parallel to the director are obtained at the critical stretch ratio $0 < \lambda_{\text{cr}} < a^{1/12}$. + +At constant temperature, the discussion is similar to that of Case 1. Namely, if the prolate LCE film is in its natural state prior to being attached to the elastic substrate, then decreasing the value of $\lambda < a^{-1/6}$ will not cause shear striping in the film. + +**Case 4.** Second, the director is initially aligned in the first direction, such that $g_1 = a^{1/3}$ and $g_2 = a^{-1/6}$. In the asymptotic limit of large stiffness ratio $\beta = \mu_f/\mu_s$, we find + +$$ \lambda_{\text{cr}} = a^{1/3} - \frac{a^{4/9} (a^{1/2} + 1)}{2^{5/3}} \left(\frac{3}{\beta}\right)^{2/3} + O\left(\beta^{-4/3}\right), \quad (65) $$ + +$$ q_{\text{cr}} = \left( \frac{a^{2/3} + a^{1/6}}{2} \right)^{1/3} \left( \frac{3}{\beta} \right)^{1/3} + O\left( \beta^{-1} \right). \quad (66) $$ \ No newline at end of file diff --git a/samples/texts/3953766/page_9.md b/samples/texts/3953766/page_9.md new file mode 100644 index 0000000000000000000000000000000000000000..2c52251f0d4e29928a26bc69fa3cf705e0c98568 --- /dev/null +++ b/samples/texts/3953766/page_9.md @@ -0,0 +1,39 @@ +where $\xi''$ and $\xi^{iv}$ are the derivatives of order 2 and 4, respectively, of the normal deflection $\xi$ with +respect to $x_1$, $S$ is the plane stress component in the first direction of the uniformly deformed film, +and $f_s$ is the normal pressure exerted in the third direction by the (infinitely deep) substrate. + +Taking the wrinkling solution [12,24] + +$$ \xi = \xi_0 \cos(qx_1), \quad f_s = -2\mu_s q\xi = -2\mu_s q\xi_0 \cos(qx_1), \qquad (68) $$ + +yields + +$$ S = -\frac{\mu_f}{3}q^2h^2 - \frac{2\mu_s}{qh}. \qquad (69) $$ + +The critical wave number, which minimises the above stress, and the corresponding critical stress then take the form (see also [1]) + +$$ q_{cr} = \frac{1}{h} \left( \frac{3}{\beta} \right)^{1/3}, \quad S_{cr} = -\mu_f \left( \frac{3}{\beta} \right)^{2/3}, \qquad (70) $$ + +where $\beta = \mu_f/\mu_s$, as before. We recover here the scaling with respect to $\beta$ given in [53] based on a LCE plate theory. Equation (67) and the critical values given by (70) are valid for both elastic and LCE films. The difference between these two types of films lies in the material constitutive law, and hence, in the formula for the stress $S_{cr}$ in terms of the deformation (see [63] for details). Specifically, in each of the four cases described in the previous section, the critical plane stress $S_{cr}$ and the critical stretch ratio $\lambda_{cr}$ are related as follows: + +**Cases 1 and 2.** The film is made of an isotropic elastic material, hence, + +$$ S_{cr} = 2\mu_f (\lambda_{cr} - a^{-1/6}) \qquad (71) $$ + +and, by (70), + +$$ \lambda_{cr} = a^{-1/6} - \frac{1}{2} \left(\frac{3}{\beta}\right)^{2/3}. \qquad (72) $$ + +**Case 3 and 4.** The film is made of a monodomain liquid crystal elastomer, hence, + +$$ S_{cr} = 2\mu_f (\lambda_{cr} - g_1) \qquad (73) $$ + +and, by (70), + +$$ \lambda_{cr} = g_1 - \frac{1}{2} \left( \frac{3}{\beta} \right)^{2/3}. \qquad (74) $$ + +In each case, the critical stretch value for the plate model approximates well the corresponding value for the 3D models in the asymptotic limit of large stiffness ratio, $\beta \to \infty$. We conclude that the plate model is applicable when the film is much stiffer than the substrate. + +# 7 Conclusion + +We conducted a linear stability analysis for the onset of wrinkling in nonlinear two-layer systems formed from a hyperelastic film on a liquid crystal elastomer substrate, or a liquid crystal elastomer film on a hyperelastic substrate. We assumed that the hyperelastic material is described by a neo-Hookean strain-energy function, while the LCE is characterised by the neoclassical strain density. We also assumed that the substrate is infinitely deep and that the nematic director is uniformly aligned either parallel with or orthogonal to the biaxial forces causing wrinkling. We note that bilayer systems where the substrate is of arbitrary thickness can also be analysed by the same method if the original 8 × 8 algebraic system, which we derived, is used instead of its reduced 6 × 6 form corresponding to the \ No newline at end of file diff --git a/samples/texts/5029325/page_10.md b/samples/texts/5029325/page_10.md new file mode 100644 index 0000000000000000000000000000000000000000..4d808b76c6a408a168ff2ecf0d87f54dd0f131a2 --- /dev/null +++ b/samples/texts/5029325/page_10.md @@ -0,0 +1,19 @@ +## 1.4 Open Problems + +In this work we prove that there exist unsatisfiable k-CNF formulas in $n$ variables that require $\delta$-regular Resolution refutations of size at least $2^{(1-\epsilon)n}$, where $k = \tilde{O}(\epsilon^{-4})$ and where $\delta = \tilde{O}(\epsilon^{-4})$. Hence a natural question is whether it is possible to improve the dependency of $\delta$ and $k$ on $\epsilon$. + +More generally, we have some proof systems stronger than $\delta$-regular Resolution, such as Resolution itself, Polynomial Calculus + Resolution, RES($k$), Cutting Planes, for which we know that there are some unsatisfiable CNFs in $n$ variables which require refutations of size $2^{\Omega(n)}$. Are those proof systems consistent with SETH? + +## 2 Preliminaries + +A *literal* is either a variable $x$ or its negation $\neg x$. A *clause* $C$ is a disjunction of literals and by its *width* we mean the number of literals appearing in $C$ and we denote this by $|C|$. A *Conjunctive Normal Form (CNF)* formula is a conjunction of a set of clauses. + +Given a boolean function $f$ on a set of variables $X$, a *partial assignment* is a function $\rho : X \to \{0, 1, *\}$. We call *domain* of $\rho$, dom($\rho$) the set $\rho^{-1}(\{0, 1\})$. The restriction of $f$ to $\rho$ denoted by $f|_{\rho}$ is a function on $\rho^{-1}(*)$ obtained from $f$ by fixing the value of all variables in $\rho^{-1}(0) \cup \rho^{-1}(1)$ according to $\rho$. We write $\rho \subseteq \sigma$ if for all $x \in X$, $\rho(x) \neq *$ implies $\sigma(x) = \rho(x)$. For a partial assignment $\rho$ for which $\rho(x) = *,$ by $\rho \cup \{(x, b)\}$ we denote a partial assignment $\rho'$ such that for all $y \neq x$, $\rho'(y) = \rho(y)$ and $\rho'(x) = b$. Given a (partial) assignment $\rho$ and a subset $B \subseteq X$, $\rho|_B$ is a partial assignment defined only on the variables in $B$ such that for all $x \in B$, $\rho|_B(x) = \rho(x)$. + +Resolution [7,24] is a proof system for refuting unsatisfiable CNF formulas. The only inference rule in Resolution is given as follows + +$$ \frac{C \lor x, D \lor \neg x}{C \lor D}, $$ + +where $C$ and $D$ are clauses and $x$ is a variable. We say that $x$ is *resolved* and $C \lor D$ is called the *resolvant* of $C \lor x$ and $D \lor \neg x$. A *Resolution derivation* of a clause *D* from a CNF $\varphi$ is a sequence $\Pi = (C_1, ..., C_\tau)$ of clauses such that $C_\tau = D$ and each $C_i$ is either an *axiom*, that is a clause from $\varphi$, or it is derived by applying the Resolution rule on some clause $C_j$ and $C_{j'}$ such that $j, j' < i$. We will denote this by $\Pi : \varphi \vdash D$. When defining subsystems of Resolution we consider hardcoded in the sequence of clauses $\Pi$ also a function providing from which previous clauses a clause in $\Pi$ is inferred or if it is a clause from $\varphi$. Having at hand such function then a Resolution derivation $\Pi$ is given a structure of a DAG and hence we can talk of *paths* in the derivation intending paths in the DAG associated to the derivation. If $\varphi$ is an unsatisfiable formula, a *Resolution refutation* of $\varphi$ is a derivation of $\bot$, the empty clause, from $\varphi$. Resolution is *sound and complete*, that is we can derive $\bot$ from a CNF formula if and only if it is unsatisfiable. + +A $\delta$-regular *Resolution derivation* of a clause *D* from a formula $\varphi$ in $n$ variables is a Resolution derivation in which along any derivation path at most a fraction of $\delta$ variables are resolved multiple times. Hence a 0-regular Resolution refutation is just \ No newline at end of file diff --git a/samples/texts/5029325/page_11.md b/samples/texts/5029325/page_11.md new file mode 100644 index 0000000000000000000000000000000000000000..6ee320341949dc9d2804e9d8f3aafaa2d27d6aad --- /dev/null +++ b/samples/texts/5029325/page_11.md @@ -0,0 +1,33 @@ +a standard regular refutation and a 1-regular Resolution refutation is one without any constraint. + +The *size* of a Resolution derivation is the number of clauses appearing in it. We denote the minimum size of a derivation of *D* from $\varphi$ by **size**($\varphi \vdash D$). We also denote the minimum size of a $\delta$-regular derivation of *D* from $\varphi$ by **size**$_{\delta}$ ($\varphi \vdash D$). Similarly we define the *width* of a derivation to be the width of the largest clause appearing in it. We denote the minimum width of a derivation of *D* from $\varphi$ by **width**($\varphi \vdash D$). + +### 3 A Game View of Resolution + +In this section we present a common framework, Definition 3.1, for the games described by Atserias and Dalmau [1] and Pudlák [22] and then we recall the characterizations of width and size in Resolution. + +**Definition 3.1 (Game**(*φ*, *R*)**).** Given an unsatisfiable CNF formula *φ* in *n* variables and a set of partial assignments *R* containing the empty assignment, we define a game, **Game**(*φ*, *R*), between two players **Prover** (he) and **Delayer** (she). + +At each step *i* of the game a partial assignment αᵢ ∈ R is maintained (α₀ is the empty partial assignment), then at step *i* + 1 the following moves take place: + +1. **Prover** picks some variable *x* ∉ dom(αᵢ). + +2. **Delayer** then has to answer *x* = *b* for some bit *b* ∈ {0, 1}. + +3. Prover set αᵢ₊₁ ∈ R such that αᵢ₊₁ ⊆ αᵢ ∪ {⟨x, b⟩}. + +If at any point in the game αᵢ falsifies φ then Prover wins; otherwise we say that Delayer wins. As customary, we say that Prover has a winning strategy for the game if for any strategy of Delayer, he can play so that he wins the game. Otherwise we say that Delayer has a winning strategy. + +If in each run of the game Prover can query at most a fraction of δ variables, we call the corresponding game **Game**δ(φ, R). + +For a suitable choice of R the Game(φ, R) is exactly the one used by Atserias and Dalmau [1] to characterise the minimal width of Resolution refutations of φ. In particular in [1] the following result is shown (rephrased here with the notations we just set up). + +**Theorem 3.2 (Atserias and Dalmau [1]).** Let φ be an unsatisfiable CNF formula and let R be the set of all possible partial assignments with a domain of size strictly less than w. The following are equivalent + +1. Prover has a winning strategy for Game(φ, R); + +2. $width(\varphi \vdash \bot) < w.$ + +Due to this equivalence, for this particular choice of R, we will denote Game(φ, R) by width-Game(φ, w). + +The next result is essentially due to Pudlák [22]: he shows that we can also characterize the minimal size of Resolution refutations of φ in terms of these games. From a Resolution refutation Π we can construct a winning strategy for Prover with a set R \ No newline at end of file diff --git a/samples/texts/5029325/page_12.md b/samples/texts/5029325/page_12.md new file mode 100644 index 0000000000000000000000000000000000000000..f211338898dabfcaa5ca900b630e6aaca48d6900 --- /dev/null +++ b/samples/texts/5029325/page_12.md @@ -0,0 +1,43 @@ +of the same size of Π and vice versa. Moreover a play of the $\mathbf{\Gamma}_{\delta}(\varphi, \mathcal{R})$ corresponds to a path in Π and, if Π is $\delta$-regular, in each run the set of variables **Prover** is going to query many times is at most a $\delta$ fraction of the total number of variables. + +**Theorem 3.3** Let $\varphi$ be an unsatisfiable CNF and let $\delta$ be any real in the interval $[0, 1]$. The following are equivalent + +1. there exists a set of partial assignments $\mathcal{R}$ such that $|\mathcal{R}| \le s$ for which **Prover** has a winning strategy for $\mathbf{\Gamma}_{\delta}(\varphi, \mathcal{R})$; + +2. $\text{size}_{\delta}(\varphi \vdash \bot) \le s$. + +Notice that this Theorem states an equivalence but in what follows we will only +use the fact that (2) implies (1). + +4 Games and Xorifications + +Given a CNF formula $\varphi$ on the variables $x_1, \dots, x_n$, we define the $\ell$-xorification of $\varphi$ +as follows: it is a formula on the new variables $y_i^j$, where $1 \le i \le n$ and $1 \le j \le \ell$ +and it is obtained by replacing each $x_i$ with $y_i^1 \oplus \dots \oplus y_i^\ell$ and expanding the formula +as a CNF formula. We denote the obtained CNF formula by $\varphi[\oplus^\ell]$ and note that if $\varphi$ is +a $k$-CNF in $n$ variables, then $\varphi[\oplus^\ell]$ is a $k\ell$-CNF in $n\ell$ variables. Due to this notation +we will refer to the variables of $\varphi$ as the $x$-variables and to the variables of $\varphi[\oplus^\ell]$ as +the $y$-variables. Moreover we say that all the $y$-variables $y_i^1, \dots, y_i^\ell$ form a block of +variables corresponding to the $x$-variable $x_i$. We say that a partial assignment over the +$y$-variables fixes a value for a $x$-variable $x_i$ if it assigns all the $y$-variables in the block +corresponding to $x_i$. + +**Restated Theorem 1.2** Let $\varphi$ be an unsatisfiable CNF formula in $n$ variables and $w$, $\delta$ and $\ell$ be parameters. If the width to refute $\varphi$ is Resolution is at least $w$ then the size to refute $\varphi[\oplus^\ell]$ in $\delta$-regular Resolution is at least $2^{(1-\epsilon)w\ell}$, where $\epsilon = \frac{1}{\ell}\log(\frac{e^3\ell n}{w}) + \frac{\delta n}{w}\log\frac{e^3\ell}{\delta}$. + +*Proof* For each partial assignment $\alpha$ over the $y$-variables there is naturally associated +a partial assignment $\alpha'$ over the $x$-variables, defined as follows + +$$ +\alpha'(x_i) = \begin{cases} +\alpha(y_i^1) \oplus \cdots \oplus \alpha_r(y_i^\ell) & \text{if } \forall j = 1, \ldots, \ell, y_i^j \in \operatorname{dom}(\alpha), \\ +* & \text{otherwise.} +\end{cases} +$$ + +By Theorem 3.3, it is enough to show that if **Prover** wins **Game**$_{\delta}(\varphi[\oplus^\ell], \mathcal{R})$ then + +$$ +|\mathcal{R}| \geq 2^{w(\ell - \log(\frac{e^3\ell n}{w}) - \frac{\delta\ell n}{w} \log \frac{e^3\ell}{\delta})}. +$$ + +So suppose Prover wins Game$_{\delta}(\varphi[\oplus^{\ell}], R)$ for some set of partial assignments $R$. Since width$(\varphi \vdash \bot) \ge w$, by Theorem 3.2, there is a winning strategy $\sigma$ for Delayer in the game width-Game$(\varphi, w)$. \ No newline at end of file diff --git a/samples/texts/5029325/page_13.md b/samples/texts/5029325/page_13.md new file mode 100644 index 0000000000000000000000000000000000000000..3ff30f16704788835dd5d77ce571a7e1a2188fce --- /dev/null +++ b/samples/texts/5029325/page_13.md @@ -0,0 +1,25 @@ +For each total assignment $\beta$ on the y-variables, consider a strategy $\sigma_\beta$ for **Delayer** in the game $\mathbf{Game}_\delta(\varphi[\oplus^\ell], \mathcal{R})$ as follows. Let $\alpha_r$ be the partial assignment on y-variables at stage $r$ of the game $\mathbf{Game}_\delta(\varphi[\oplus^\ell], \mathcal{R})$ and $y_i^j$ the variable queried by **Prover** at stage $r+1$. Then the strategy $\sigma_\beta$ for **Delayer** goes as follows: + +1. if there exists $j' \neq j$ such that $y_i^{j'} \notin \text{dom}(\alpha_r)$, set $y_i^j$ to $\beta(y_j^i)$; + +2. otherwise, if for all $j' \neq j$, $y_i^{j'} \in \text{dom}(\alpha_r)$, then look at the value $b \in \{0, 1\}$ the strategy $\sigma$ sets the variable $x_i$ when given the partial assignment $\alpha_r'$. Then set $y_i^j$ to $q \in \{0, 1\}$ such that + +$$q \oplus \bigoplus_{j' \neq j} \alpha_r (y_i^{j'}) = b.$$ + +This can be done since $x_i \equiv y_i^1 \oplus \cdots \oplus y_i^\ell$ and the value of $x_i$ can be set freely to 0 or 1 appropriately even after all but one of $y_i^1, \dots, y_i^\ell$ have been set. Moreover, by induction on $r$, it is easy to see that the strategy $\sigma$ must provide an answer when challenged by **Prover** by any variable $x_i$ not on the record $\alpha_r$, hence $\sigma_\beta$ is well-defined. + +Moreover, it is easy to see that for each total assignment $\beta$ over the y-variables, $\sigma_\beta$ is a winning strategy for **Delayer** in the game width-Game$(\varphi[\oplus^\ell], w\ell)$. Since we are assuming that **Prover** has a winning strategy for $\mathbf{Game}_\delta(\varphi[\oplus^\ell], \mathcal{R})$, in particular, this means that for any $\beta$ he wins against the **Delayer**'s strategy $\sigma_\beta$. This means that for each total assignment $\beta$ over the y-variables, $\mathcal{R}$ must contain some partial assignment, denoted by $\rho_\beta$, with domain of size at least $w\ell$ and such that at least $w$ blocks of y-variables are completely fixed by $\rho_\beta$. Without loss of generality we assume that each $\rho_\beta$ fixes exactly $w$ blocks of y-variables, that is if $\rho_\beta$ is setting more y-variables we simply ignore some of the variables and only consider $w$ blocks. Our goal is to show that we have ‘many distinct’ such partial assignments $\rho_\beta$. + +Let $B \subseteq [n]$ denote a generic set of size $w$ and consider for each possible such $B$ the set $S_B$ of the total assignments $\beta$s such that $\rho_\beta$ is fixing all the $y_i^1, \dots, y_i^\ell$ corresponding to some $i$ in $B$. There are $2^{n\ell}$ possible total assignments $\beta$ and $\binom{n}{w}$ possible sets $B$, hence by the pigeonhole principle, there is a set $B^* \subseteq [n]$ of size $w$ such that + +$$|S_{B^*}| \geq \frac{2^{n\ell}}{\binom{n}{w}}. \quad (1)$$ + +Let $S_{B^*}'$ be the set of partial assignments $\beta|_{B^*}$ where $\beta \in S_{B^*}$. We clearly have that + +$$|S_{B^*}| \le |S_{B^*}'| \cdot 2^{n\ell - |\overline{B^*}|} = |S_{B^*}'| \cdot 2^{n\ell - w\ell}.$$ + +By Eq. (1), we get + +$$|S_{B^*}'| \geq \frac{2^{w\ell}}{\binom{n}{w}}. \quad (2)$$ + +We have now that both $S_{B^*}'$ and $\{\rho_\beta \in \mathcal{R} : \beta \in S_{B^*}\}$ consist of assignments with domain the y-variables $y_i^j$ such that $i \in B^*$ and $1 \le j \le l$. We show that $|\{\rho_\beta \in \mathcal{R}\}: \ No newline at end of file diff --git a/samples/texts/5029325/page_2.md b/samples/texts/5029325/page_2.md new file mode 100644 index 0000000000000000000000000000000000000000..c46d053be98aa3f99240f5f9ce1f19c23667ab64 --- /dev/null +++ b/samples/texts/5029325/page_2.md @@ -0,0 +1,27 @@ +$β \in S_{B^*} |$ cannot be too small compared to $|S'_{B^*}|$, this will be, intuitively, due to the fact that the $β$s we start with are very different. + +Let $Z^\beta$ be the set of variables that Prover re-queried when playing against $\sigma_\beta$ and for any $i = 1, \dots, n$ let $Z_i^\beta = Z^\beta \cap \{y_i^1, \dots, y_i^\ell\}$. By hypothesis, Prover is allowed to re-query in each game at most a $\delta$ fraction of variables, hence $|Z^\beta| \le \delta\ln n$. + +When Delayer follows the strategy $\sigma_\beta$ and fixes all y-variables in a block corresponding to $x_i$, the assignment produced $\rho_\beta$ is within Hamming distance $|Z_i^\beta| + 1$ from $\beta$ in that block. This means that for each $\beta \in S_{B^*}$ and for each $i$, $\rho_\beta|_{\{y_1^{i_1}, \dots, y_\ell^{i_\ell}\}}$ has Hamming distance at most $|Z_i^\beta| + 1$ from some partial assignment in $S'_{B^*}$ restricted to $\{y_1^{i_1}, \dots, y_\ell^{i_\ell}\}$. Let $\mathcal{Z}$ be the set of all possible sets Z that are subsets of the y-variables of size $\delta\ell n$ and such that there exists $\beta \in S_{B^*}$ with $Z^\beta \subseteq Z$. For any $i=1, \dots, n$ let $Z_i = Z \cap \{y_i^1, \dots, y_i^\ell\}$. Then, by counting the variables where $\rho_\beta$ and an assignment in $S'_{B^*}$ could differ, we have that + +$$|S'_{B^*}| \le |\{\rho_\beta \in \mathcal{R} : \beta \in S_{B^*}\}| \cdot \sum_{Z \in \mathcal{Z}} \prod_{i \in B^*} 2^{|Z_i|+1} \binom{\ell}{|Z_i|+1}. \quad (3)$$ + +Hence we have the following chain of inequalities + +$$|S'_{B^*}|^{\text{eq.}(3)} \le |\{\rho_\beta \in \mathcal{R} : \beta \in S_{B^*}\}| \cdot \sum_{Z \in \mathcal{Z}} \prod_{i \in B^*} 2^{|Z_i|+1} \binom{\ell}{|Z_i|+1} \quad (4)$$ + +$$\le |\{\rho_\beta \in \mathcal{R} : \beta \in S_{B^*}\}| \cdot \sum_{Z \in \mathcal{Z}} \prod_{i \in B^*} \left( \frac{e^{2\ell}}{|Z_i|+1} \right)^{|Z_i|+1} \quad (5)$$ + +$$\le |\{\rho_\beta \in \mathcal{R} : \beta \in S_{B^*}\}| \cdot \sum_{Z \in \mathcal{Z}} \left( \frac{\sum_{i \in B^*} e^{2\ell}}{\sum_{i \in B^*} (|Z_i|+1)} \right)^{\sum_{i \in B^*} (|Z_i|+1)} \quad (6)$$ + +$$\le |\{\rho_\beta \in \mathcal{R} : \beta \in S_{B^*}\}| \cdot \binom{\ell n}{\delta \ell n} \cdot \left(\frac{\sum_{i \in B^*} e^{2\ell}}{w}\right)^{\delta \ell n + w} \quad (7)$$ + +$$= |\{\rho_\beta \in \mathcal{R} : \beta \in S_{B^*}\}| \cdot \binom{\ell n}{\delta \ell n} \cdot (e^{2\ell})^{\delta \ell n + w} \quad (8)$$ + +The inequality (6) follows from the weighted AM-GM inequality¹ and the inequality (7) follows from the fact that $w \le \sum_{i \in B^*} (|Z_i| + 1) \le \delta\ell n + w$. Putting all together we have that + +¹ The weighted Arithmetic Mean - Geometric Mean inequality says that given non-negative numbers $a_1, \dots, a_n$ and non-negative weights $w_1, \dots, w_n$ then + +$$\prod_i a_i^{w_i} \le \left( \frac{\sum_i w_i a_i}{w} \right)^w,$$ + +where $w = \sum_i w_i$. We applied this inequality with $a_i = \frac{e^{2\ell}}{|Z_i|+1}$ and $w_i = |Z_i| + 1$. \ No newline at end of file diff --git a/samples/texts/5029325/page_7.md b/samples/texts/5029325/page_7.md new file mode 100644 index 0000000000000000000000000000000000000000..10ce12a6ab7ca4608642064c4bf937bfed6133f3 --- /dev/null +++ b/samples/texts/5029325/page_7.md @@ -0,0 +1,13 @@ +## 1.1 Previous Work + +Exponential lower bounds supporting ETH have long been known for natural proof systems such as Resolution since mid 1980s, see e.g. [28]. These are $2^{\Omega(n)}$ lower bounds for k-CNF formulas on n variables and hence not strong enough to support SETH. Some thirteen years passed until the first lower bounds supporting SETH were shown. Pudlák and Impagliazzo [23] proved such lower bounds for tree-like Resolution via Prover-Delayer games. Another thirteen years later, Beck and Impagliazzo [5] obtained a very strong width lower bound which simplified and improved the result of [23] for tree-like Resolution and they were able to prove lower bounds supporting SETH for regular Resolution, a sub-system of Resolution. Beck and Impagliazzo in [5] showed that there are unsatisfiable k-CNF formulas in n variables requiring refutations of size at least $2^{n(1-\epsilon_k)}$ in regular Resolution. Their proof is an adaptation of a probabilistic technique from [4] and, from an high level, it can be seen as a variation of the bottleneck counting of Haken in [16]. In their argument a rule is given which maps assignments to particular clauses of the proof, at which a significant amount of 'work' is done. + +Prior to this work, the strongest proof system with lower bounds supporting SETH was regular Resolution [5]. This work proves lower bounds supporting SETH in a subsystem of Resolution stronger than regular Resolution and moreover it gives a different, conceptually simpler, game-theoretic proof of the fact that SETH is consistent with regular Resolution. + +## 1.2 Results + +In this work we consider proof systems that are intermediate between regular Resolution and Resolution. The *Resolution* proof system [7, 24] is a proof system for refuting unsatisfiable CNF formulas. A Resolution refutation of a CNF formula $\varphi$ is a sequence of clauses ending with the empty clause such that each clause is either a clause from $\varphi$ or it is derived from previous clauses in the sequence according to the following inference rule: + +$$ \frac{C \lor x, \quad D \lor \neg x}{C \lor D}, $$ + +where C and D are clauses and x is a variable. Clearly a Resolution refutation can be annotated with directed edges keeping track of the applications of the inference rule, in particular each clause in the sequence will have either 0 or 2 predecessors according to if it is a clause from $\varphi$ or an inferred clause. The resulting directed graph is a DAG and the sequence of clauses in the Resolution refutation is a topological ordering of it. Notice that given a resolution Refutation the DAG we can associate is not uniquely determined. Anyway, when defining subsystems of Resolution based on restrictions on the DAG structures allowed for a proof, it is customary to consider a Resolution refutation directly as a DAG with vertices labeled with clauses. This is the case for *tree-like* Resolution where are allowed as valid Resolution refutations only the ones with a tree-like structure. \ No newline at end of file diff --git a/samples/texts/5029325/page_8.md b/samples/texts/5029325/page_8.md new file mode 100644 index 0000000000000000000000000000000000000000..433fef45aae46b0f122cedd371c883701f477966 --- /dev/null +++ b/samples/texts/5029325/page_8.md @@ -0,0 +1,17 @@ +A $\delta$-regular Resolution refutation of a formula $\varphi$ is a Resolution derivation in which along any path of the refutation DAG at most a fraction $\delta$ of the variables of $\varphi$ are resolved multiple times. Hence a 0-regular Resolution refutation is just a standard regular Resolution refutation, that is a Resolution refutation where no variable is resolved multiple times along any path. A 1-regular Resolution refutation is just one without any constraint. A more formal definition of Resolution and $\delta$-regular Resolution is given in Sect. 2 together with all the other necessary preliminaries. + +The main result of this work is the following. + +**Theorem 1.1 (Main theorem)** For any large $n$ and $k$, there exists an unsatisfiable $k$-CNF formula $\psi$ on $n' \ge n$ variables such that any $\delta$-regular Resolution refutation of $\psi$ requires size at least $2^{(1-\epsilon_k)n}$ where both $\epsilon_k$ and $\delta$ are $\tilde{O}(k^{-1/4})$. + +We recall that the width of a Resolution refutation is the number of literals in the largest clause appearing in the refutation. The way we prove this result is via a strong width lower bound, that is a lower bound of the form $(1-\epsilon_k)n$, with $\epsilon_k \to 0$, relative to Resolution refutations of some particular $k$-CNFs in $n$ variables. Width lower bounds of this form were proved in [5] and improved in the asymptotic in [8] (see Theorem 4.1 for the precise statement we are going to use). + +The reader could be tempted to think that once we are given a strong Resolution width of the form above then a size lower bound as in Theorem 1.1 will follow by the standard relation between width and size by Ben-Sasson and Wigderson [6]. In [6] the authors showed that if a formula requires refutations of large width, it also requires refutations with many clauses. More precisely they showed that if a $k$-CNF formula can only have Resolution refutations of width at least $W$, then it requires Resolution size at least $2^{(W-k)^2/16n}$, where $n$ is the number of variables. The constant loss in the exponent is the reason why we do not immediately get $2^{(1-\epsilon_k)n}$ size lower bounds from strong width lower bounds. However, the result in [6] holds for any $k$-CNF without any particular assumption. If the formula is structured in some sense, for instance if it is a *xorification*, we show we can avoid this loss (in a subsystem of Resolution). + +The $\ell$-xorification of a CNF formula $\varphi$ in $n$ variables is a new CNF formula in $n\ell$ variables obtained substituting each variable $x_i$ in $\varphi$ with $\ell$ new variables $y_i^1 \oplus \cdots \oplus y_i^\ell$ and then expanding again as a CNF. We denote the CNF resulting from such operation $\varphi[\oplus^\ell]$. Our main technical result, Theorem 1.2 informally states that if a $k$-CNF $\varphi$ requires width $w$ to be refuted in Resolution, then any $\delta$-regular Resolution refutation of $\varphi[\oplus^\ell]$ requires size $2^{(1-\epsilon)nw}$, where $\epsilon$ is a function of $k$, $\ell$, $\delta$ and $w$. More precisely we prove the following: + +**Theorem 1.2** Let $\varphi$ be an unsatisfiable CNF formula in $n$ variables and $w$, $\delta$ and $\ell$ be parameters. If the width to refute $\varphi$ is Resolution is at least $w$ then the size to refute $\varphi[\oplus^\ell]$ in $\delta$-regular Resolution is at least $2^{(1-\epsilon)w\ell}$, where $\epsilon = \frac{1}{\ell} \log(\frac{e^3\ell n}{w}) + \frac{\delta n}{w} \log \frac{e^3\ell}{\delta}$. + +Once we have proved this result then Theorem 1.1 follows just by carefully tuning the parameters. + +The way we prove Theorem 1.2 relies on two known games characterizations: the *Pudlák game* characterizing Resolution size [22] and the *Atserias-Dalmau game* characterizing Resolution width [1]. In Sect. 3 we give a precise account for both games in a common setting and terminology, see Definition 3.1. We conclude this \ No newline at end of file diff --git a/samples/texts/5029325/page_9.md b/samples/texts/5029325/page_9.md new file mode 100644 index 0000000000000000000000000000000000000000..3d6661ff137832ce9bf6887d51129e914cfa3c62 --- /dev/null +++ b/samples/texts/5029325/page_9.md @@ -0,0 +1,26 @@ +introductory part giving some intuition behind those games and the proof of Theorem 1.2. + +In the *Pudlák game*, informally, we have two players, *Prover* and *Delayer*, that play on some formula $\varphi$. *Prover* has the objective of showing that the formula $\varphi$ is unsatisfiable by querying variables. *Delayer* on the other hand wants to play as long as possible before the formula is falsified while answering to the queries *Prover* asks her. The size of Resolution proofs of $\varphi$ is then characterized as the minimal number of records, i.e. partial assignments, *Prover* has to consider in a winning strategy [22]. If we force the *Prover* in each game to re-query a fraction of at most $\delta$ variables from $\varphi$ then the minimal number of records such *Prover* has to consider in a winning strategy characterize the size in $\delta$-regular Resolution. This is the content of Theorem 3.3 which we leave without proof since it is a trivial observation over the result in [22]. + +Hence to prove a Resolution (or a $\delta$-regular Resolution) size lower bound we show that, in order to win, **Prover** must keep a large number of records and we can do that by producing a lot of sufficiently different strategies for **Delayer**. **Prover** must win against each of them, hence in his winning strategy he must have a lot of distinct records, since the strategies of **Delayer** are sufficiently different. In the literature this is done essentially by making **Prover** play against a **Delayer** that plays accordingly to a random strategy [12,22]. Then the size lower bound, that is a lower bound on the number of records that **Prover** must have in a winning strategy, is obtained by probabilistic arguments. This may very likely lead to some loss in the constants that we need to avoid to prove a SETH lower bound for Resolution size so we choose another way: we play the Pudlák game over a xorified formula $\varphi[\oplus^\ell]$. + +The construction of multiple strategies for Delayer relies on the characterization +of Resolution width as a game [1] where again we have a Prover and a Delayer, +the goal of the prover is to falsify the formula φ but he can use assignments with a +bounded number of variables. At a very high level, a winning strategy for Delayer in +the width game on φ gives rise to a multitude of strategies for Delayer on the Pudlák +game on φ[⊕^l]. The new strategies act differently from each other on φ[⊕^l], but in +a sense they all act the same as the original strategy for Delayer in the width game +on the original formula φ. The size lower bound then follows by a counting argument +exploiting the combinatorial properties of the xorified formula in such a way that the +number of Delayer strategies, for the Pudlák game played on φ[⊕^l], does indeed +hugely amplify. + +Notice that Theorem 1.2 does not depend on the particular formula we choose to +apply it but, to get the result in Theorem 1.1, we need to apply it to a particular formula +φ for which we have strong width lower bounds, such formulas are provided by [5] +and, with better asymptotic, by [8]. + +## 1.3 Outline of Paper + +In the next Section we give some preliminaries and notations about Resolution and δ-regular Resolution. Section 3 contains the common framework for *Pudlák games* [22] characterizing *size* in Resolution and the *Atserias and Dalmau games* [1] characterizing *width* in Resolution. Section 4 contains the core results of this work (Theorem 1.1 and Theorem 1.2). \ No newline at end of file diff --git a/samples/texts/7104097/page_1.md b/samples/texts/7104097/page_1.md new file mode 100644 index 0000000000000000000000000000000000000000..9d0b603bada784edc405c6151196536e2944144d --- /dev/null +++ b/samples/texts/7104097/page_1.md @@ -0,0 +1,26 @@ +# The Range 1 Query (R1Q) Problem* + +Michael A. Bender¹², Rezaul Chowdhury¹, Pramod Ganapathi¹, +Samuel McCauley¹, and Yuan Tang³ + +¹ Department of Computer Science, Stony Brook University, Stony Brook, NY, USA +{bender, rezaul, pganapathi, smccauley}cs.stonybrook.edu + +² Tokutek, Inc. + +³ Software School, Fudan University, Shanghai, China +yuantang@csail.mit.edu + +**Abstract.** We define the *range 1 query* (R1Q) problem as follows. Given a *d*-dimensional ($d \ge 1$) input bit matrix $A$, preprocess $A$ so that for any given region $\mathcal{R}$ of $A$, one can efficiently answer queries asking if $\mathcal{R}$ contains a 1 or not. We consider both orthogonal and non-orthogonal shapes for $\mathcal{R}$ including rectangles, axis-parallel right-triangles, certain types of polygons, and spheres. We provide space-efficient deterministic and randomized algorithms with constant query times (in constant dimensions) for solving the problem in the word RAM model. The space usage in bits is sublinear, linear, or near linear in the size of $A$, depending on the algorithm. + +**Keywords:** R1Q, range query, range emptiness, randomized, rectangular, orthogonal, non-orthogonal, triangular, polygonal, circular, spherical. + +## 1 Introduction + +Range searching is one of the fundamental problems in computational geometry [1,12]. It arises in application areas including geographical information systems, computer graphics, computer aided design, spatial databases, and time series databases. Range searching encompasses different types of problems, such as range counting, range reporting, emptiness queries, and optimization queries. + +The *range 1 query* (R1Q) problem is defined as follows. Given a *d*-dimensional ($d \ge 1$) input bit matrix $A$ (consisting of 0's and 1's), preprocess $A$ so that one can efficiently answer queries asking if any given range $\mathcal{R}$ of $A$ is empty (does not contain a 1) or not, denoted by $R1Q_A(\mathcal{R})$ or simply $R1Q(\mathcal{R})$. In 2-D, the range $\mathcal{R}$ can be a rectangle, a right triangle, a polygon or a circle. + +In this paper, we investigate solutions in the word RAM model sharing the following characteristics. First of all, we want queries to run in constant time, even for $d \ge 2$ dimensions. Second, we are interested in solutions that have space linear or sublinear in the number of bits in the input grid. Note that while our sublinear bounds are parameterized by the number of 1s in the grid, this is still larger + +* Rezaul Chowdhury & Pramod Ganapathi are supported in part by NSF grant CCF-1162196. Michael A. Bender & Samuel McCauley are supported in part by NSF grants IIS-1247726, CCF-1217708, CCF-1114809, and CCF-0937822. \ No newline at end of file diff --git a/samples/texts/7104097/page_11.md b/samples/texts/7104097/page_11.md new file mode 100644 index 0000000000000000000000000000000000000000..c298d55a32810757f643df549a28fdaf8b13873d --- /dev/null +++ b/samples/texts/7104097/page_11.md @@ -0,0 +1,9 @@ +We first show that for any given $p \in [0, \log N]$ at most $2\gamma$ fraction of the queries of size $2^p$ can return incorrect answers. Consider any two consecutive blocks of size $2^p$, say, blocks $i \in [0, \frac{N}{2^p} - 1)$ and $i+1$. Exactly $2^p$ different queries of size $2^p$ will cross the boundary between these two blocks. The answer to each of these queries will depend on the estimates of $R_p[i]$ and $L_p[i+1]$ obtained from the CM sketches. Under our construction the estimates are $\hat{R}_p[i] \le (1+\gamma)R_p[i] \le R_p[i] + \gamma \cdot 2^p$ and $\hat{L}_p[i+1] \le (1+\gamma)L_p[i+1] \le L_p[i+1] + \gamma \cdot 2^p$. Hence, at most $\gamma \cdot 2^p$ of those $2^p$ queries will produce incorrect results due to the error in estimating $R_p[i]$, and at most $\gamma \cdot 2^p$ more because of the error in estimating $L_p[i+1]$. Thus with probability at least $(1-\delta)^2$, at most $2\gamma$ fraction of those $2^p$ queries will return wrong results. Recall from Section 2.1 that we answer given queries by decomposing the query range into two overlapping query ranges. Hence, with probability at least $(1-\delta)^4 \ge 1-4\delta$, at most $2\gamma+2\gamma = 4\gamma$ fraction of all queries can produce wrong answers. + +**Theorem 4.** Given a 1-D bit array of length *N* containing *N*₁ nonzero entries, and two parameters γ ∈ (0, 1/4) and δ ∈ (0, 1/4), and an integer constant *c* > 1, one can construct a data structure occupying O(N₁ log³ N log₁+γ (1/δ) + N¹/c log log N) bits (and discard the input array) to answer each R1Q in O(log c/δ) worst-case time such that with probability at least 1 − 4δ at most 4γ fraction of all query results will be wrong. + +**Sampling Based Algorithm.** Suppose we are allowed to use only $\mathcal{O}(s)$ bits of space (in addition to the input array $A$), and $s = \Omega(\log_2 N)$. We are also given two constants $\varepsilon \in (0, 1)$ and $\delta \in (0, 1)$. We build $L_p$ and $R_p$ arrays for each $p \in [\log \frac{N}{s} + \log \log N, \log N]$, and an MSB lookup table to support constant time MSB queries for integers in $[1, s/\log N]$. Consider the query $R_1 Q_A(i, j)$. If $j-i+1 \le w$, we answer the query correctly in constant time by reading at most 2 words from $A$ and using bit shifts. If $j-i+1 \ge 2^{\log \frac{N}{s} + \log \log N} = \frac{N \log N}{s}$, we use the $L_p$ and $R_p$ arrays to correctly answer the query in constant time. If $w < j-i+1 < \frac{N \log N}{s}$, we sample $\lceil \frac{1}{\epsilon} \ln (\frac{1}{\delta}) \rceil$ entries uniformly at random from $A[i...j]$, and return their bitwise OR. It is easy to show that the $L_p$ and $R_p$ tables use $\mathcal{O}(s)$ bits in total, and the MSB table uses $o(s)$ bits of space. The query time is clearly $\mathcal{O}(\frac{1}{\epsilon} \ln (\frac{1}{\delta}))$. + +**Error Bound.** If at least an $\epsilon$ fraction of the entries in $A[i...j]$ are nonzero then the probability that a sample of size $\lceil \frac{1}{\epsilon} \ln(\frac{1}{\delta}) \rceil$ chosen uniformly at random from the range will pick at least one nonzero entry is $\ge 1-(1-\epsilon)^{\frac{1}{\epsilon} \ln(\frac{1}{\delta})} \approx 1-\delta$. + +**Theorem 5.** Given a 1-D bit array of length *N*, a space bound *s* = *Ω*(log *N*), and two parameters *ε* ∈ (0,1) and *δ* ∈ (0,1), one can construct a data structure occupying only O(*s*) bits of space (in addition to the input array) that in O(1/ε ln(1/δ)) time can answer each R1QA(i,j) correctly with probability at least 1 − δ provided at least an ε fraction of the entries in A[i...j] are nonzero. If j−i+1 ≤ *w* or j−i+1 ≥ N log N/s, the query result is always correct. \ No newline at end of file diff --git a/samples/texts/7104097/page_2.md b/samples/texts/7104097/page_2.md new file mode 100644 index 0000000000000000000000000000000000000000..84220c3292690c8cbccc994aadb6799577805da3 --- /dev/null +++ b/samples/texts/7104097/page_2.md @@ -0,0 +1,19 @@ +**Fig. 4.** Black grid points contain 1's and white grid points contain 0's. Polygon in (a) satisfies Prop. 1. Polygons in (b) and (c) do not satisfy Prop. 1. Still, R1Q can be answered for (c). + +in constant time). Observe that we can answer R1Q(*ABC*) if we can answer R1Q for triangles *ADG* and *CEH*, and the rectangle *BDFE*. R1Q for the rectangle can be answered using our deterministic algorithm described in Section 2.2. R1Q for a right triangle of a particular orientation with height or base length equal to a power of two can be answered in constant time. This is done by checking whether the point stored (from preprocessing) with the appropriate endpoint of the hypotenuse for that specific orientation is inside the triangle or not. + +**Theorem 6.** Given a 2-D bit matrix of size $N = \sqrt{N} \times \sqrt{N}$ containing $N_0$ zero bits, one can construct a data structure occupying $O(N \log N + N_0 \log^2 N)$ bits in $O(N^{1.5})$ time (and discard the input matrix) to answer each axis-aligned right triangular R1Q with the three vertices on the grid points in $O(1)$ time. + +## 3.2 Polygonal R1Q + +Consider a simple polygon with its vertices on grid points satisfying the following. + +*Property 1.* For every two adjacent vertices $(a, b)$ and $(c, d)$, one of the two right triangles with the third vertex being either $(a, d)$ or $(c, b)$ is completely inside the polygon. + +It can be shown that such a polygon can be decomposed into a set of possibly overlapping right triangles and rectangles with only grid points as vertices that completely covers the polygon (see Fig. 4(a)). Examples of polygons that do not satisfy the constraint are given in Fig. 4(b,c), but we can still answer R1Q for the polygon in (c). A simple polygon with $k$ vertices satisfying property 1 can be decomposed into $O(k)$ right triangles and rectangles and hence can be answered in $O(k)$ time. + +## 3.3 Spherical R1Q + +The spherical R1Q problem is defined as follows. Given a $d$-dimensional ($d \ge 2$) input bit matrix $A$, preprocess $A$ such that given any grid point $p$ in $A$ and a radius $r \in \mathbb{R}^{+}$, find efficiently if there exists a 1 in the $d$-sphere centered at $p$ of radius $r$. Here, we present the algorithm for 2-D. The approach can be extended to higher dimensions. + +A nearest 1-bit of a grid point $p$ is called a *nearest neighbor* (NN) of $p$. We preprocess $A$ by computing and compressing the NNs of all grid points in $A$ \ No newline at end of file diff --git a/samples/texts/7104097/page_3.md b/samples/texts/7104097/page_3.md new file mode 100644 index 0000000000000000000000000000000000000000..c2a1d430446c7fc7ca6bbdd2dca394051a55966d --- /dev/null +++ b/samples/texts/7104097/page_3.md @@ -0,0 +1,30 @@ +(only one NN per grid point). We then answer a spherical R1Q by checking +whether a NN of the given center point is inside the circle of given radius. + +**Preprocessing.** We store the locations of the NNs of the grid points of $A$ in a temporary NN matrix that occupies $\mathcal{O}(N \log N)$ bits, but can be compressed to occupy $\mathcal{O}(N\sqrt{\log N})$ bits as follows. + +We divide the grid into $\sqrt{\frac{\log N}{6}} \times \sqrt{\frac{\log N}{6}}$ blocks. We store the NN position for all points on the boundary of each block. The interior points will be replaced with arrows ($\rightarrow, \leftarrow, \uparrow, \downarrow$) and bullets ($\bullet$) as follows. If a grid point $p$ contains a 1 then $p$ is replaced with a $\bullet$. An arrow at a grid point gives the direction of its NN. If we follow the arrows from any interior point, we end up in either a boundary point or an interior point containing a 1. For any given block the matrix created as above will be called a *symbol matrix* representing the block. + +Two blocks are of the same type if they have the same symbol matrix. Each +symbol can be represented using only three bits. Since each block has $(\log N)/6$ +symbols, there are $2^{\frac{3(\log N)}{6}} = \sqrt{N}$ possible block types. For each block type we +create a *position matrix* that stores, for each grid point within a block, the pointer +to its NN if the NN is an interior point, or a pointer to a boundary point if the +NN is an exterior or boundary point. The boundary point will have stored its +own NN position in the input array. + +We can now discard the original input matrix, and replace it with the follow- +ing compressed representation. For each block in the input matrix we store its +block type (i.e., a pointer to the corresponding block type) followed by the NN +positions of its boundary points. For each block type we retain its position matrix. + +**Query Execution.** We can answer a spherical R1Q by checking whether the NN position of the center point is inside the query sphere. The approach of finding the NN position is as follows. We find the block to which the given point belongs and follow the pointer to its block type. We check the position stored at the given point in the position matrix. If it points to an internal point, then that point is the correct NN. If it points to a boundary point, we again follow the pointer stored at the boundary point to get the correct NN. + +**Theorem 7.** Given a 2-D bit array of size *N*, one can construct a data structure occupying $\mathcal{O}(N\sqrt{\log N})$ bits (and discard the input array) to answer each spherical R1Q in $\mathcal{O}(1)$ time. + +**Acknowledgments.** We like to thank Michael Biro, Dhruv Matani, Joseph S. B. Mitchell, and anonymous referees for insightful comments and suggestions. + +References + +1. Agarwal, P.K., Erickson, J.: Geometric range searching and its relatives. Contemporary Mathematics 223, 1–56 (1999) +2. Bender, M.A., Farach-Colton, M., Pemmasani, G., Skiena, S., Sumazin, P.: Lowest common ancestors in trees and directed acyclic graphs. Journal of Algorithms 57(2), 75–94 (2005) \ No newline at end of file diff --git a/samples/texts/7104097/page_4.md b/samples/texts/7104097/page_4.md new file mode 100644 index 0000000000000000000000000000000000000000..0b8189ffaea43ca90a5092a22c9311292d4b3c7e --- /dev/null +++ b/samples/texts/7104097/page_4.md @@ -0,0 +1,25 @@ +3. Chazelle, B., Rosenberg, B.: Computing partial sums in multidimensional arrays. In: SoCG. pp. 131–139. ACM (1989) + +4. Cormode, G., Muthukrishnan, S.: An improved data stream summary: the count-min sketch and its applications. Journal of Algorithms 55(1), 58–75 (2005) + +5. De Berg, M., Cheong, O., van Kreveld, M., Overmars, M.: Computational geometry. Springer (2008) + +6. Fischer, J.: Optimal succinctness for range minimum queries. In: LATIN, pp. 158–169. Springer (2010) + +7. Fischer, J., Heun, V., Stiihler, H.: Practical entropy-bounded schemes for o (1)-range minimum queries. In: Data Compression Conference. pp. 272–281. IEEE (2008) + +8. Golynski, A.: Optimal lower bounds for rank and select indexes. TCS 387(3), 348–359 (2007) + +9. González, R., Grabowski, S., Mäkinen, V., Navarro, G.: Practical implementation of rank and select queries. In: Poster Proc. WEA. pp. 27–38 (2005) + +10. Navarro, G., Nekrich, Y., Russo, L.: Space-efficient data-analysis queries on grids. TCS (2012) + +11. Overmars, M.H.: Efficient data structures for range searching on a grid. Journal of Algorithms 9(2), 254–275 (1988) + +12. Sharir, M., Shaul, H.: Semialgebraic range reporting and emptiness searching with applications. SIAM Journal on Computing 40(4), 1045–1074 (2011) + +13. Tang, Y., Chowdhury, R., Kuszmaul, B.C., Luk, C.K., Leiserson, C.E.: The Pochoir stencil compiler. In: SPAA. pp. 117–128. ACM (2011) + +14. Yao, A.C.: Space-time tradeoff for answering range queries. In: STOC. pp. 128–136. ACM (1982) + +15. Yuan, H., Atallah, M.J.: Data structures for range minimum queries in multidimensional arrays. In: SODA. pp. 150–160 (2010) \ No newline at end of file diff --git a/samples/texts/7104097/page_5.md b/samples/texts/7104097/page_5.md new file mode 100644 index 0000000000000000000000000000000000000000..8d77111aa7519dd159b235d462946a8133609071 --- /dev/null +++ b/samples/texts/7104097/page_5.md @@ -0,0 +1,13 @@ +than the information-theoretic lower bounds. For our motivating applications, information-theoretically optimal space is less important than constant query times. Third, we are interested in grid inputs [10, 11], viewing the problem in terms of pixels/voxels rather than a set of spatial points. This grid perspective enables constant-time operations such as table lookup and hashing. Finally, we are interested in both orthogonal and nonorthogonal queries, and we require solutions that are concise enough to be implementable. + +**Previous Results.** The R1Q problem can be solved using data structures such as balanced binary search trees, kd-trees, quad trees, range trees, partition trees, and cutting trees (see [5]), which take the positions of the 1-bits as input. It can also be solved using a data structure of Overmars [11], which uses priority search trees, y-fast tries, and q-fast tries and takes the entire grid as input. However, in $d$-D ($d \ge 2$), in the worst case these data structures have a query time at least polylogarithmic and occupy a near-linear number of bits. + +The R1Q problem can also be solved via range partial sum [3, 14] and the range minimum query (RMQ) [2, 6, 7, 15] problems. Though several efficient algorithms have been developed to solve the problem in 1-D and 2-D, their generalizations to 3-D and higher dimensions occupying a linear number of bits are not known yet. Also, there is little work on space-efficient constant-time RMQ solutions for non-orthogonal ranges. + +The R1Q problem can also be solved using rank queries [8, 9]. Again, its generalization to 2-D and higher dimensions has not yet been studied. + +**Motivation.** We encountered the R1Q and R0Q (whether a range contains a 0) problems while trying to optimize stencil computations in the *Pochoir* stencil compiler [13], where we had to answer octagonal R1Q and octagonal R0Q on a static 2-D property grid. Stencil computations have applications in physics, computational biology, computational finance, mechanical engineering, adaptive statistical design, weather forecasting, clinical medicine, image processing, quantum dynamics, oceanic circulation modeling, electromagnetics, multigrid solvers, and many other areas (see the references in [13]). + +In Fig. 1, we provide a simplified exposition of the problem encountered in Pochoir. There are two grids of the same size: a static property grid and a dynamic value grid. Each property grid cell is set to 1 if it satisfies property $\mathcal{P}$ and 0 otherwise. When Pochoir needs to update a range $\mathcal{R}$ in the value grid (see Alg. 1), its runtime system checks whether all or none of the points in $\mathcal{R}$ satisfy $\mathcal{P}$ in the property grid, and based on the query result it uses an appropriate precompiled optimized version of the original code (see Algs. 3, 4) to update the range in the value grid. To check if all points in $\mathcal{R}$ satisfy $\mathcal{P}$, Pochoir uses $R_0\mathcal{Q}(\mathcal{R})$, and to check if no points in $\mathcal{R}$ satisfy $\mathcal{P}$, it uses $R_1\mathcal{Q}(\mathcal{R})$. + +Pochoir needs time-, space-, and cache-efficient data structures to answer R1Q. It can also tolerate some false-positive errors. The solutions should achieve constant query time and work in all dimensions. Although it is worth trading off space to achieve constant query times, space is still a scarce resource. \ No newline at end of file diff --git a/samples/texts/7104097/page_6.md b/samples/texts/7104097/page_6.md new file mode 100644 index 0000000000000000000000000000000000000000..858088ffb7d730b7d43e511464a9519d88f5d210 --- /dev/null +++ b/samples/texts/7104097/page_6.md @@ -0,0 +1,22 @@ +**Fig. 1.** Examples of the procedures in Pochoir that make use of R1Q and R0Q. + +**Our Contributions.** We solve the R1Q problem for orthogonal and non-orthogonal ranges. Our major contributions as shown in Table 1 are as follows: + +1. [Orthogonal Deterministic.] We present a deterministic data structure to answer R1Q for orthogonal ranges in all dimensions and for any data distribution. It occupies linear space in bits and answers queries in constant time for any constant dimension. + +2. [Orthogonal Randomized.] We present randomized data structures to answer R1Q for orthogonal ranges. The structures occupy sublinear space in bits and provide a tradeoff between query time and error probability. + +3. [Non-Orthogonal Deterministic.] We present deterministic data structures to answer R1Q for non-orthogonal shapes such as axis-parallel right-triangles (for 2-D) and spheres (for all dimensions). The structures occupy near-linear space in bits and answer queries in constant time. + +We use techniques such as power hyperrectangles, power right-triangles, sketches, sampling, the four Russians trick, and compression in our data struc- +tures. A careful combination of these techniques allows us to solve a large class of +R1Q problems. Techniques such as power hyperrectangles, table lookup, and the +four Russians trick are already common in RMQ-style operations, while sketches, +power right-triangles, and compression are not. + +**Organization of the Paper.** Section 2 presents deterministic and randomized algorithms to answer orthogonal R1Qs on a grid in constant time for constant dimensions. Section 3 presents deterministic algorithms to answer non-orthogonal R1Qs on a grid, for axis-parallel right triangles, some polygons, and spheres. + +## 2 Orthogonal Range 1 Queries (R1Q) + +In this section, we present deterministic and randomized algorithms for answering +orthogonal R1Qs in constant time and up to linear space. \ No newline at end of file diff --git a/samples/texts/7104097/page_8.md b/samples/texts/7104097/page_8.md new file mode 100644 index 0000000000000000000000000000000000000000..203fdb46f2119df24ac069e5a80b08c157731f0f --- /dev/null +++ b/samples/texts/7104097/page_8.md @@ -0,0 +1,19 @@ +**Query Execution.** To answer R1Q$_A(i, j)$, we consider two cases: (1) *Intra-word queries*: If $(i, j)$ lies inside one word, we answer R1Q using bit shifts. (2) *Inter-word queries*: If $(i, j)$ spans multiple words, then the query gets split into three subqueries: (a) R1Q from $i$ to the end of its word, (b) R1Q of the words between $i$'s and $j$'s word (both exclusive), and (c) R1Q from the start of $j$'s word to $j$. + +The answer to an inter-word query is 1 if and only if the R1Q for at least one of the three subqueries is 1. The first and third subqueries are intra-word queries and can be answered using bit shifts. Let the words containing indices $i$ and $j$ be $I$ and $J$, respectively. Then, the second subquery, denoted by $R1Q_{L_0}(I+1, J-1)$, is answered as follows. Using the MSB of $J-I-1$, we find the largest integer $p$ such that $2^p \le J - I - 1$. The query $R1Q_{L_0}(I+1, J-1)$ is then decomposed into the following two overlapping queries of size $2^p$ each: $R1Q_{L_0}(I+1, I+2^p)$ and $R1Q_{L_0}(J-2^p, J-1)$. If either of those two ranges contains a 1 then the answer to the original query will be 1, and 0 otherwise. We show below how to answer $R1Q_{L_0}(I+1, I+2^p)$. Query $R1Q_{L_0}(J-2^p, J-1)$ is answered similarly. + +Split $L_0$ into blocks of size $2^p$. Then, the range $R1Q_{L_0}(I+1, I+2^p)$ can be covered by one or two consecutive blocks. Let $I+1$ be in the $k$th block. If the range lies in one block, we find whether a 1 exists in that block by checking whether $L_p[k] < 2^p$ is true. If the range is split across two consecutive blocks, we find whether a 1 exists in at least one of the two blocks by checking whether at least one of $R_p[k] \le (k+1)2^p - I$ or $L_p[k+1] \le I + 2^p - (k+1)2^p$ is true. + +## 2.2 Deterministic d-D Algorithm + +For d-D ($d \ge 2$) R1Q, the input is a bit matrix $A$ of size $N = n \times n$, but the algorithm extends an algorithm for a 2-D matrix of size $N = n \times n$, but the algorithm extends to higher dimensions. For simplicity, we assume $n$ is a power of 2. The query $R1Q([i_1, j_1][i_2, j_2])$ asks if there exists a 1 in the submatrix $A[i_1 \dots j_1][i_2 \dots j_2]$. + +**Preprocessing.** For each $p, q \in [0, \log n]$, we partition $A$ into $\frac{n}{2^p} \times \frac{n}{2^q}$ blocks, each of size $2^p \times 2^q$ called a $(p, q)$-block. For each $(p, q)$ pair, we construct four tables of size $\frac{N}{2^{p+q}} \times \min(2^p, 2^q)$ each: + +(i) *TL**p,q*: if $p \le q$, *TL**p,q*[*i*,*j*][$k$] indicates that any rectangle of height $k \in [0, 2^p)$ starting from the top-left corner of the current block must have width at least *TL**p,q*[*i*,*j*][$k$] in order to include at least one 1-bit. + +(ii) *BL*, *TR*, *BR*: similar to *TL* but starts from the bottom-left, top-right and bottom-right corners, respectively. + +In all cases, a stored value of $\max(2^p, 2^q)$ indicates that the block has no 1. + +**Query Execution.** Given a query [*i*₁, *j*₁][*i*₂, *j*₂], we find the largest integers *p* and *q* such that $2^p \le j_1 - i_1 + 1$ and $2^q \le j_2 - i_2 + 1$. The original query range can then be decomposed into four overlapping (*p*, *q*)-blocks, which we call *power rectangles*, each with a corner at one of the four corners of the original rectangle, as in Fig. 2(a). If any of these four rectangles contains a 1, the answer to the original query will be 1, and 0 otherwise. We show below how to answer an R1Q for a power rectangle. \ No newline at end of file diff --git a/samples/texts/7104097/page_9.md b/samples/texts/7104097/page_9.md new file mode 100644 index 0000000000000000000000000000000000000000..754eea36ff4dc90483783c71e3b820a89d345fe8 --- /dev/null +++ b/samples/texts/7104097/page_9.md @@ -0,0 +1,13 @@ +**Fig. 2.** Rectangles: (a) Query rectangle split into four possibly overlapping power rectangles. (b) Power rectangle divided into four regions by four split rectangles. + +We consider the partition of A into preprocessed $(p, q)$-blocks. It is easy to see that each of the four power rectangles of size $2^p \times 2^q$ will intersect at most four preprocessed $(p, q)$-blocks. We call each rectangle contained in both the power rectangle and a $(p, q)$-block a *split rectangle* (see Fig. 2(b)). The R1Q for a split rectangle can be answered using a table lookup, checking if the table values of the appropriate $(p, q)$-blocks are inside the power rectangle boundary, as shown in Fig. 2(b). The proof of the following theorem will be given in the full paper. + +**Theorem 2.** Given a $d-D$ input grid of size $N = n^d$, each orthogonal R1Q on the grid can be answered deterministically in $O(4^d d)$ time after preprocessing the grid in $\Theta(N)$ time using $O((d+1)!(2/\ln 2)^d N)$ bits of space. In 1-D, the space can be reduced to $O(N/\log N)$ bits. + +## 2.3 Randomized Algorithms + +In this section, we present randomized algorithms that build on the deterministic algorithms given in Sections 2.1 and 2.2. We describe the algorithms for one dimension only. Extensions to higher dimensions are straightforward. + +**Sketch Based Algorithms.** Our algorithms provide probabilistic guarantees based on the Count-Min (CM) sketch data structure proposed in [4]. Let $N_1$ be the number of 1-bits in the input bit array $A[0...N-1]$ for any data distribution. Then, the (preprocessing) time and space complexities depend on $N_1$ while the query time remains constant. + +A CM sketch with parameters $\epsilon \in (0, 1]$ and $\delta \in (0, 1)$ can store a summary of any given vector $\mathbf{a} = \langle a_0, a_1, \ldots, a_{n-1} \rangle$ with $a_i \ge 0$ in only $\lceil \frac{\epsilon}{\delta} \rceil [\ln \frac{1}{\delta}] \log ||\mathbf{a}||_1$ bits of space, where $||\mathbf{a}||_1$ (or $||\mathbf{a}||$) = $\sum_{i=0}^{n-1} a_i$, and can provide an estimate $\hat{a}_i$ of any $a_i$ with the following guarantees: $a_i \le \hat{a}_i$, and with probability at least $1-\delta$, $\hat{a}_i \le a_i + \epsilon ||\mathbf{a}||_1$. It uses $t = \lceil \ln \frac{1}{\delta} \rceil$ hash functions $h_1 \ldots h_t : \{0 \ldots n-1\} \rightarrow \{1 \ldots b\}$ chosen uniformly at random from a pairwise-independent family, where bucket size $b = \lceil \frac{\epsilon}{\delta} \rceil$. These hash functions are used to update a 2-D matrix $c[1:t][1:b]$ of bt counters initialized to 0. For each $i \in [0, n-1]$ and each $j \in [1,t]$ one then updates $c[j][h_j(i)]$ to $c[j][h_j(i)] + a_i$. After the updates, an estimate $\hat{a}_i$ for any given query point $a_i$ is obtained as $\min_{1 \le j \le t} c[j][h_j(i)]$. \ No newline at end of file diff --git a/samples/texts/7675318/page_1.md b/samples/texts/7675318/page_1.md new file mode 100644 index 0000000000000000000000000000000000000000..4efa2f42e6632a98c3df7403d4c25ca9f09c2144 --- /dev/null +++ b/samples/texts/7675318/page_1.md @@ -0,0 +1,35 @@ +# MORAVA K-THEOIRES OF CLASSIFYING SPACES OF COMPACT LIE GROUPS + +MASAKI KAMEKO + +**ABSTRACT.** This is the abstract for the talk I will give in the 54th Topology Symposium in August 6, 2007 at the University of Aizu. I will talk about the computational aspect of cohomology and generalized cohomology theories of classifying spaces of connected compact Lie groups. + +## 1. INTRODUCTION + +Let $p$ be a prime number. Let $G$ be a connected compact Lie group and let us denote by $BG$ its classifying space. We say $G$ has $p$-torsion if and only if $H_*(G; \mathbb{Z})$ has $p$-torsion. In the case $G$ has no $p$-torsion, the mod $p$ cohomology of $BG$ is well-known. However, in the case $G$ has $p$-torsion, the computation of the mod $p$ cohomology is not an easy task. I refer the reader for the book of Mimura and Toda [9] for the detailed account on the cohomology of classifying spaces of compact Lie groups. + +I will give the current state of computation of the mod $p$ cohomology theory, Brown-Peterson cohomology and Morava $K$-theories of classifying spaces of some connected compact Lie groups. The coefficient ring of Brown-Peterson cohomology is + +$$BP^* = \mathbb{Z}_{(p)}[v_1, v_2, \dots, v_n, \dots],$$ + +where $\deg v_n = -2(p^n - 1)$ and the coefficient ring of the Morava $K$-theory $K(n)$ is + +$$K(n)^* = \mathbb{Z}/p[v_n, v_n^{-1}].$$ + +I will describe the results of the joint work with Yagita [2] and some other results obtained after writing of [2]. One of the explicit computational results is as follows: + +**Theorem 1.1.** For $p = 2$, $G = G_2$, the Morava $K$-theory $K(n)^*(BG_2)$ of the classifying space of the exceptional Lie group $G_2$ is given by + +$$K(n)^* \otimes_{BP^*} (BG_2).$$ + +As an $\text{gr } BP^*$-module, we have that + +$$\text{gr } BP^*(BG_2)$$ + +is isomorphic to + +$$\text{gr } BP^*[[y_8, y_{12}, y_{14}]]/(y_{14}\rho_0, y_{14}\rho_{-2}, y_{14}\rho_{-6}) \bigoplus \text{gr } BP^*[[y_8, y_{12}, y_{14}]]\{w_4\},$$ + +where the index indicates the degree. + +The author was partially supported by Japan Society for the Promotion of Science, Grant-in-Aid for Scientific Research (C) 19540105. \ No newline at end of file diff --git a/samples/texts/7675318/page_5.md b/samples/texts/7675318/page_5.md new file mode 100644 index 0000000000000000000000000000000000000000..0eacd517f986ceb4434752790ffa112507b4fb1b --- /dev/null +++ b/samples/texts/7675318/page_5.md @@ -0,0 +1,102 @@ +The computation of the Brown-Peterson cohomology above is due to Kono and +Yagita in [6]. The computation of Morava K-theories is a new result. Kono +and Yagita compute the Brown-Peterson cohomology of BSpin(n) for n ≤ 10 and +BPU(4n+2), BSp(4n+2), too. Wilson compute the Brown-Peterson cohomology +of BO(n) in [12] and Kono and Yagita compute their Morava K-theories in [6]. +Recently, Inoue and Yagita compute the Brown-Peterson cohomology and Morava +K-theories of BSO(n) for n ≥ 2 in [1]. In the case p = 3, we have the following +results for simply connected simple Lie groups. + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + +
(p = 3)GBPMorava K
F4
E6??
E7
E8????
+ +The computation of $p = 3$, $G = F_4$, $PU(3)$ is done in [6]. In the case $p = 3$, $G = E_6$, in [2], we computed the Brown-Peterson cohomology of $BE_6$, however, we could not conclude + +$$ +K(n)^*(BE_6) = K(n)^* \otimes_{BP} BP^*(BE_6). +$$ + +It remains to be an open problem. In the case *p* = 5, we have the following result +for simply connected simple Lie groups. + +$$ +\begin{array}{c|c|c|c} +(p=5) & G & BP & \text{Morava } K \\ +\hline +& E_8 & \bigcirc & \bigcirc \\ +\end{array} +$$ + +4. ADAMS SPECTRAL SEQUENCE + +In [6], they use the Atiyah-Hirzebruch spectral sequence in order to compute +the Brown-Peterson cohomology. The Atiyah-Hirzebruch spectral sequence in [6] +does not collapse at the $E_2$-level. In [2], we use the Adams spectral sequence in +order to compute $P(n)$-cohomology and show that the assumption in Theorem 3.3 +holds. The Adams spectral sequence is an spectral sequence converging to $P(n)$ +cohomology whose $E_2$-term is given by + +$$ +\mathrm{Ext}_{\mathcal{A}}(H^*(P(n); \mathbb{Z}/p), H^*(BG; \mathbb{Z}/p)) = \mathrm{Ext}_{\mathcal{E}_n}(\mathbb{Z}/p, H^*(BG; \mathbb{Z}/p)), +$$ + +where $\mathcal{A}$ is the mod $p$ Steenrod algebra and $\mathcal{E}_n$ is the subalgebra generated by Milnor operations $Q_n$, $Q_{n+1}$, .... The $E_2$-term could be computed by taking the (co)homology of the (co)chain complex + +$$ +d_1 : grP(n)^*\hat{\otimes} H^*(BG; \mathbb{Z}/p) \rightarrow grP(n)^*\hat{\otimes} H^*(BG; \mathbb{Z}/p) +$$ + +where + +$$ +\operatorname{gr} P(n)^* = \mathbb{Z}/p[v_n, v_{n+1}, \dots, ] +$$ + +and + +$$ +d_1(v \otimes x) = \sum_{k=n}^{\infty} vv_k \otimes Q_k x. +$$ + +We compute this (co)chain complex. In this talk, we deal with the case $p = 2$, +$G = G_2$, $E_6$ and give an outline of the computation. + +First, we deal with the case $p = 2$, $G = G_2$. Let us recall the mod 2 cohomology of the classifying space of the exceptional Lie group $G_2$. There is an non-toral elementary abelian 2-subgroup of rank 3, say $A$, in $G_2$. The induced homomorphism + +$$ +H^*(BG_2;\mathbb{Z}/2) \to H^*(BA;\mathbb{Z}/2) +$$ \ No newline at end of file diff --git a/samples/texts/7675318/page_6.md b/samples/texts/7675318/page_6.md new file mode 100644 index 0000000000000000000000000000000000000000..5c9cbe645e34587840e64bd68669fe6d39a74129 --- /dev/null +++ b/samples/texts/7675318/page_6.md @@ -0,0 +1,52 @@ +is a monomorphism and the image of this induced homomorphism is the ring +of invariants of general linear group $GL_3(\mathbb{F}_2)$ acting on the polynomial algebra +$H^*(BA;\mathbb{Z}/2)$ in the usual manner. So, the mod 2 cohomology of $BG_2$ is the Dick- +son invariants + +$$D = \mathbb{Z}/2[y_0, y_1, y_2]$$ + +where $\deg y_k = 2^3 - 2^k$ and through this inclusion, we have the action of the Steenrod algebra, in particular, the action of Milnor operations on $H^*(BG_2; \mathbb{Z}/2)$. Let $R$ be the subalgebra of $D$ generated by $y_0^2, y_1^2$ and $y_2^2$. Then $D$ is a free $R$-module with the basis + +$$\{1, y_0, y_0y_1, y_0y_2, y_1, y_2, y_1y_2, y_0y_1y_2\}.$$ + +Let us write $\beta_0 = y_0^2$, $\beta_1 = y_0y_1$, $\beta_2 = y_0y_2$ and $\beta_3 = y_0$. Then, we have + +$$D = R\{1, \beta_1, \beta_2, \beta_3, \frac{\beta_1\beta_2}{\beta_0}, \frac{\beta_1\beta_3}{\beta_0}, \frac{\beta_2\beta_3}{\beta_0}, \frac{\beta_1\beta_2\beta_3}{\beta_0}\}.$$ + +**Proposition 4.1.** There exit $f_1, f_2, f_3 \in grP(n)^*\hat{\otimes}R$ such that + +$$d_1(\beta_1) = f_1\beta_0, \quad d_1(\beta_2) = f_2\beta_0, \quad d_1(\beta_3) = f_3\beta_0,$$ + +where $\deg f_1 = 0$, $\deg f_2 = -2$, $\deg f_3 = -6$. Moreover, $f_1, f_2, f_3$ is a regular sequence in $grP(n)^*\hat{\otimes}R$. + +Since $R$ has no odd degree elements and since + +$$\deg d_1(\beta_1\beta_2\beta_3/\beta_0^2) = 4,$$ + +the following proposition completes the proof of Theorem 1.1. The element + +$$d_{1}(\beta_{1}\beta_{2}\beta_{3}/\beta_{0}^{2})$$ + +corresponds to $w_4$ in Theorem 1.1. + +**Proposition 4.2.** Let $R$ be a graded algebra over $\mathbb{Z}/2$ and suppose that $R$ is an integral domain. Let $f_1, \dots, f_n$ be elements in $R$. Let $z_0$ be an element in $R$ and consider $R$-submodules $C_k$ of + +$$R[z_0^{-1}] \otimes \Delta(\beta_1, \dots, \beta_n)$$ + +generated by + +$$\frac{\beta_{i_1} \cdots \beta_{i_k}}{z_0^{k}},$$ + +where $\Delta(\beta_1, \dots, \beta_n)$ is a simple system of generators, $1 \le i_1 < i_2 < \cdots < i_k \le n$, $k > 0$ and $C_0 = R$. Consider the differential $d$ given by $d(\beta_k) = z_0 f_k$ for $k = 1, \dots, n$ and by $d(xy) = d(x)y + xd(y)$. If $f_1, \dots, f_n$ is a regular sequence, then the homology of $(C, d)$ is given by + +$$H_{0}(C, d) = R/(z_{0}f_{1}, \dots, z_{0}f_{n})$$ + +and + +$$H_k(C, d) = \{0\}$$ + +for $k > 0$. + +Next, we deal with the case $p = 2$, $G = E_6$. The mod 2 cohomology $H^*(BE_6; \mathbb{Z}/2)$ is generated by $y_4, y_6, y_7, y_{10}, y_{18}, y_{32}, y_{48}$ and $y_{34}$ with the relations + +$$y_{7}y_{10} = 0, \quad y_{7}y_{18} = 0, \quad y_{7}y_{34} = 0, \quad y_{34}^{2} = y_{10}^{2}y_{48} + y_{18}^{2}y_{32} + \text{lower terms},$$ \ No newline at end of file diff --git a/samples/texts/7856234/page_6.md b/samples/texts/7856234/page_6.md new file mode 100644 index 0000000000000000000000000000000000000000..26a87fe2e5c6a7571f441bf7474280e912fbb605 --- /dev/null +++ b/samples/texts/7856234/page_6.md @@ -0,0 +1,27 @@ +REFERENCES + +ANSYS Inc. (2014). ANSYS Fluent Version 15.0 User's Guide. + +AFSHIN E., MITRA J. (2012). Comparison of Mixture and VOF Models for Numerical Simulation of Air-entrainment in Skimming Flow over Stepped Spillways Journal of Science Direct. Procedia engineering Vol.28, pp 657 – 66. + +BENTALHA C., HABI M. (2015). Numerical Simulation of Air Entrainment for Flat-sloped Stepped Spillway, The Journal of Computational Multiphase Flows, Vol. 7, n°1, pp. 33-41. + +BENTALHA C., HABI M. (2017). Numerical simulation of water flow along stepped spillways with non uniform step heights, Larhyss Journal, n°31, pp. 115-129. + +BENTALHA C., HABI M. (2019). Numerical study of the flow field for moderate-slope stepped spillways. Larhyss Journal, n°37, pp.39-51. + +BOMBARDELLI FA, MEIRELES I, MATOS J. (2010). Laboratory measurements and multi-block numerical simulations of the mean flow and turbulence in the non-aerated skimming flow region of steep stepped spillways. Environmental Fluid Mechanics. Vol. 11. issue .3 pp.263-288. + +CHANSON H. (1997). Air Bubble Entrainment in Free-Surface Turbulent Shear Flows. Academic Press, London, UK, 401 pages ISBN 0-12-168110-6. + +CHANSON H. (2001) .The Hydraulics of Stepped Chutes and Spillways. Balkema Publisher,. Rotterdam, The Netherlands. ISBN 90 5809 352 2, 384 p. + +CHARLES E. RICE., KEM C.KADAVY. (1996). Model of A roller compacted concrete stepped spillway. Journal of Hydraulic Engineering, ASCE, Vol 122, n°6, pp.292-297. + +CHENG XIANGJU., CHEN YONGCAN., LUO LIN. (2006). Numerical simulation of air-water two-phase flow over stepped spillways. Science in China Series E, Technological Sciences, Vol 49, n°6 , pp.674-684 + +CHEN Q., G.Q. DAI, H.W. LIU., (2002). Volume of Fluid Model for Turbulence Numerical Simulation of Stepped Spillway Over Flow. Journal of Hydraulic Engineering, ASCE Vol 128, n°7,pp. 683-688. + +CHINNARASRI C., KOSITGITTIWONG K., JULIEN PY. (2012). Model of flow over spillways by computational fluid dynamics. Proceedings of the Institution of Civil Engineers, Vol. 167, n° 3, pp. 164-175 + +CHRISTODOULOU G.C. (1993). Energy dissipation on stepped spillways. Journal of Hydraulic Engineering, ASCE Vol 119, n°5 ,pp. 644-650. \ No newline at end of file