arXiv++ Combinatorics

We consider a very simple dynamical system on weighted graphs which we call Iterative Graph Normalization (IGN) and a variant in which we apply a non-linear activation function to the weights after each normalization. We show that the indicator vectors of the Maximal Independent Sets of the graph are the only binary fixed points of IGN, that they are attractive under simple conditions on the activation function and we characterize their basins of attraction. We enumerate a number of other fixed points and we prove repulsivity for some classes. Based on extensive experiments and different theoretical arguments we conjecture that IGN always converges and converges to a binary solution for non-linear activations. If our conjectures are correct, IGN would thus be a differentiable approximation algorithm for the Maximum Weight Independent Set problem (MWIS), a central NP-hard optimization problem with numerous applications. IGN is closely related to a greedy approximation algorithm of MWIS by Kako et al. which has a proven approximation ratio. Experimental results show that IGN provides solutions of very similar quality. In the context of the Assignment Problem, IGN corresponds to an iterative matrix normalization scheme which is closely related to the Sinkhorn-Knopp algorithm except that it projects to a permutation matrix instead of a doubly stochastic matrix. We relate our scheme to the Softassign algorithm and provide comparative results. As Graph Normalization is differentiable, its iterations can be embedded into a machine learning framework and used to train end-to-end any model which includes a graphical optimization step which can be cast as a maximum weight independent set problem. This includes problems such as graph and hypergraph matching, sequence alignment, clustering, ranking, etc. with applications in multiple domains.

A $0$-$1$ matrix $M$ is saturating for a $0$-$1$ matrix $P$ if $M$ does not contain a submatrix that can be turned into $P$ by changing some $1$ entries to $0$ entries, and changing an arbitrary $0$ to $1$ in $M$ introduces such a submatrix in $M$. In saturation problems for $0$-$1$ matrices we are interested in estimating the minimum number of $1$ entries in an $m \times n$ matrix that is saturating for $P$, in terms of $m$ and $n$. In other words, we wish to give good estimates for the saturation function of $P$. Recently, Brualdi and Cao initiated the study of saturation problems in the context of $0$-$1$ matrices. We extend their work in several directions. We prove that every $0$-$1$ forbidden matrix has its saturation function either in $Θ(1)$ or $Θ(n)$ in the case when we restrict ourselves to square saturating matrices. Then we give a partial answer to a question posed by Brualdi and Cao about the saturation function of $J_k$, which is obtained from the identity matrix $I_k$ by putting the first row after the last row. Furthermore, we exhibit a $5\times 5$ permutation matrix with the saturation function bounded from the above by a fixed constant. We complement this result by identifying large classes of $0$-$1$ matrices with linear saturation function. Finally, we completely resolve the related semisaturation problem as far as the constant vs. linear dichotomy is concerned.

Let $p(Y_1, \dots, Y_d, Z_1, \dots, Z_e)$ be a self-adjoint noncommutative polynomial, with coefficients from $\mathbb{C}^{r \times r}$, in the indeterminates $Y_1, \dots, Y_d$ (considered to be self-adjoint), the indeterminates $Z_1, \dots, Z_e$, and their adjoints $Z_1^*, \dots, Z_e^*$. Suppose $Y_1, \dots, Y_d$ are replaced by independent random $n \times n$ matching matrices, and $Z_1, \dots, Z_e$ are replaced by independent random $n \times n$ permutation matrices. Assuming for simplicity that $p$'s coefficients are $0$-$1$ matrices, the result can be thought of as a kind of random $rn$-vertex graph $G$. As $n \to \infty$, there will be a natural limiting infinite graph $X$ that covers any finite outcome for $G$. A recent landmark result of Bordenave and Collins shows that for any $\varepsilon > 0$, with high probability the spectrum of a random $G$ will be $\varepsilon$-close in Hausdorff distance to the spectrum of $X$ (once the suitably defined "trivial" eigenvalues are excluded). We say that $G$ is "$\varepsilon$-near fully $X$-Ramanujan". Our work has two contributions: First we study and clarify the class of infinite graphs $X$ that can arise in this way. Second, we derandomize the Bordenave-Collins result: for any $X$, we provide explicit, arbitrarily large graphs $G$ that are covered by $X$ and that have (nontrivial) spectrum at Hausdorff distance at most $\varepsilon$ from that of $X$. This significantly generalizes the recent work of Mohanty et al., which provided explicit near-Ramanujan graphs for every degree $d$ (meaning $d$-regular graphs with all nontrivial eigenvalues bounded in magnitude by $2\sqrt{d-1} + \varepsilon$). As an application of our main technical theorem, we are also able to determine the "eigenvalue relaxation value" for a wide class of average-case degree-$2$ constraint satisfaction problems.

We identify the dimension of the centralizer of the symmetric group $\mathfrak{S}_d$ in the partition algebra $\mathcal{A}_d(δ)$ and in the Brauer algebra $\mathcal{B}_d(δ)$ with the number of multidigraphs with $d$ arrows and the number of disjoint union of directed cycles with $d$ arrows, respectively. Using Schur-Weyl duality as a fundamental theory, we conclude that each centralizer is related with the $G$-invariant space $P^d(M_n(\mathbf{k}))^G$ of degree $d$ homogeneous polynomials on $n \times n$ matrices, where $G$ is the orthogonal group and the group of permutation matrices, respectively. Our approach gives a uniform way to show that the dimensions of $P^d(M_n(\mathbf{k}))^G$ are stable for sufficiently large $n$.

A $d$-dimensional matrix is called \emph{$1$-polystochastic} if it is non-negative and the sum over each line equals~$1$. Such a matrix that has a single $1$ in each line and zeros elsewhere is called a \emph{$1$-permutation} matrix. A \emph{diagonal} of a $d$-dimensional matrix of order $n$ is a choice of $n$ elements, no two in the same hyperplane. The \emph{permanent} of a $d$-dimensional matrix is the sum over the diagonals of the product of the elements within the diagonal. For a given order $n$ and dimension $d$, the set of $1$-polystochastic matrices forms a convex polytope that includes the $1$-permutation matrices within its set of vertices. For even $n$ and odd $d$, we give a construction for a class of $1$-permutation matrices with zero permanent. Consequently, we show that the set of $1$-polystochastic matrices with zero permanent contains at least $n^{n^{3/2}(1/2-o(1))}$ $1$-permutation matrices and contains a polytope of dimension at least $cn^{3/2}$ for fixed $c,d$ and even $n\to\infty$. We also provide counterexamples to a conjecture by Taranenko about the location of local extrema of the permanent. For odd $d$, we give a construction of $1$-permutation matrices that decompose into a convex linear sum of positive diagonals. These combine with a theorem of Taranenko to provide counterexamples to a conjecture by Dow and Gibson generalising van der Waerden's conjecture to higher dimensions.

In this paper we extend the study of three state lively quantum walks on cycles by considering the coin operator as a linear sum of permutation matrices, which is a generalization of the Grover matrix. First we provide a complete characterization of orthogonal matrices of order $3\times 3$ which are linear sum of permutation matrices. Consequently, we determine several groups of complex, real and rational orthogonal matrices. We establish that an orthogonal matrix of order $3\times 3$ is a linear sum of permutation matrices if and only if it is permutative. Finally we determine period of lively quantum walk on cycles when the coin operator belongs to the group of orthogonal (real) linear sum of permutation matrices.

Finite metric spaces are characterized by a polyhedral cone defined in terms of the positivity of the distance functions and the triangle inequalities. Their classification is based on the decomposition of an associated polyhedral cone, called the "metric fan". The complete classification of $n$-point metric spaces is available only for $n\le 6$. As the number of classes increases rapidly with the number of elements, it is desirable to have coarser equivalence class decompositions based on certain invariants of finite metric spaces. If $(X,d)$ is a finite metric space with elements $P_i$ and with distance functions $d_{ij}$, the Gromov product at $P_i$ is defined as $Δ_{ijk}=1/2(d_{ij}+d_{ik}-d_{jk})$. Assuming that the set of Gromov product at $P_i$ has a unique smallest element $Δ_{ijk}$, the association of the edge $P_jP_k$ to $P_i$ defines the "Gromov product structure". The "pendant-free" reduction of the finite metric space is the graph obtained by removing the edges $P_jP_k$ corresponding to the minimal Gromov products $Δ_{ijk}$ at $P_i$. In the present work, we define a matrix representation for a Gromov product structure $S$ on an $n$-point metric space, by $n\times n$ matrix $G_S$. We prove that if two metric spaces have Gromov product structures that can be mapped to each other by a permutation of the indices, then their matrices are similar via the corresponding permutation matrix. Matrix invariants of $G_S$ are used to define subclasses of Gromov product structures and their application to $n=5$ and $n=6$-point spaces are given.

To any $n \times n$ Latin square $L$, we may associate a unique sequence of mutually orthogonal permutation matrices $P = P_1, P_2, ..., P_n$ such that $L = L(P) = \sum kP_k$. Brualdi and Dahl (2018) described a generalisation of a Latin square, called an alternating sign hypermatrix Latin-like square (ASHL), by replacing $P$ with an alternating sign hypermatrix (ASHM). An ASHM is an $n \times n \times n$ (0,1,-1)-hypermatrix in which the non-zero elements in each row, column, and vertical line alternate in sign, beginning and ending with $1$. Since every sequence of $n$ mutually orthogonal permutation matrices forms the planes of a unique $n \times n \times n$ ASHM, this generalisation of Latin squares follows very naturally, with an ASHM $A$ having corresponding ASHL $L = L(A) =\sum kA_k$, where $A_k$ is the $k^{\text{th}}$ plane of $A$. This paper addresses some open problems posed in Brualdi and Dahl's article, firstly by characterising how pairs of ASHMs with the same corresponding ASHL relate to one another and providing a tight lower bound on $n$ for which two $n \times n \times n$ ASHMs can correspond to the same ASHL, and secondly by exploring the maximum number of times a particular integer may occur as an entry of an $n \times n$ ASHL. A general construction is given for an $n \times n$ ASHL with the same entry occurring $\lfloor\frac{n^2 + 4n -19}{2}\rfloor$ times, improving considerably on the previous best construction, which achieved the same entry occuring $2n$ times.

Character polynomials are used to study the restriction of a polynomial representation of a general linear group to its subgroup of permutation matrices. A simple formula is obtained for computing inner products of class functions given by character polynomials. Character polynomials for symmetric and alternating tensors are computed using generating functions with Eulerian factorizations. These are used to compute character polynomials for Weyl modules, which exhibit a duality. By taking inner products of character polynomials for Weyl modules and character polynomials for Specht modules, stable restriction coefficients are easily computed. Generating functions of dimensions of symmetric group invariants in Weyl modules are obtained. Partitions with two rows, two columns, and hook partitions whose Weyl modules have non-zero vectors invariant under the symmetric group are characterized. A reformulation of the restriction problem in terms of a restriction functor from the category of strict polynomial functors to the category of finitely generated FI-modules is obtained.

For a permutation $π$, and the corresponding permutation matrix, we introduce the notion of {\em discrete derivative}, obtained by taking differences of successive entries in $π$. We characterize the possible derivatives of permutations, and consider questions for permutations with certain properties satisfied by the derivative. For instance, we consider permutations with distinct derivatives, and the relationship to so-called Costas arrays.

We consider the three finite free convolutions for polynomials studied in a recent paper by Marcus, Spielman, and Srivastava. Each can be described either by direct explicit formulae or in terms of operations on randomly rotated matrices. We present an alternate approach to the equivalence between these descriptions, based on combinatorial Weingarten methods for integration over the unitary and orthogonal groups. A key aspect of our approach is to identify a certain \emph{quadrature property}, which is satisfied by some important series of subgroups of the unitary groups (including the groups of unitary, orthogonal, and signed permutation matrices), and which yields the desired convolution formulae.

We provide several constructions of special unextendible entangled bases with fixed Schmidt number $k$ (SUEB$k$) in $\mathbb{C}^{d}\otimes \mathbb{C}^{d'}$ for $2\leq k\leq d\leq d'$. We generalize the space decomposition method in Guo [Phys. Rev. A 94, 052302 (2016)], by proposing a systematic way of constructing new SUEB$k$s in $\mathbb{C}^{d}\otimes \mathbb{C}^{d'}$ for $2\leq k < d \leq d'$ or $2\leq k=d< d'$. In addition, we give a construction of a $(pqdd'-p(dd'-N))$-number SUEB$pk$ in $\mathbb{C}^{pd}\otimes \mathbb{C}^{qd'}$ from an $N$-number SUEB$k$ in $\mathbb{C}^{d}\otimes \mathbb{C}^{d'}$ for $p\leq q$ by using permutation matrices. We also connect a $(d(d'-1)+m)$-number UMEB in $\mathbb{C}^{d}\otimes \mathbb{C}^{d'}$ with an unextendible partial Hadamard matrix $H_{m\times d}$ with $m<d$, which extends the result in [Quantum Inf. Process. 16(3), 84 (2017)].

The Birkhoff polytope, defined to be the convex hull of $n\times n$ permutation matrices, is a well studied polytope in the context of the Ehrhart theory. This polytope is known to have many desirable properties, such as the Gorenstein property and existence of regular, unimodular triangulations. In this paper, we study analogues of the Birkhoff polytope for finite irreducible Coxeter groups of other types. We focus on a type-$B$ Birkhoff polytope $\mathcal{BB}(n)$ arising from signed permutation matrices and prove that it and its dual polytope are reflexive, and hence Gorenstein, and also possess regular, unimodular triangulations. Noting that our triangulation proofs do not rely on the combinatorial structure of $\mathcal{BB}(n)$, we define the notion of an orthant-lattice property polytope and use this to prove more general results for the existence of regular, unimodular triangulations and unimodular covers for a significant family of reflexive polytopes. We conclude by remarking on some connections to Gale-duality, Birkhoff polytopes of other types, and possible applications of orthant-lattice property.

This paper introduces a new graph construction, the permutational power of a graph, whose adjacency matrix is obtained by the composition of a permutation matrix with the adjacency matrix of the graph. It is shown that this construction recovers the classical zig-zag product of graphs when the permutation is an involution, and it is in fact more general. We start by discussing necessary and sufficient conditions on the permutation and on the adjacency matrix of a graph to guarantee their composition to represent an adjacency matrix of a graph, then we focus our attention on the cases in which the permutational power does not reduce to a zig-zag product. We show that the cases of interest are those in which the adjacency matrix is singular. This leads us to frame our problem in the context of equitable partitions, obtained by identifying vertices having the same neighborhood. The families of cyclic and complete bipartite graphs are treated in details.

It is known that the automorphism group of the elementary abelian $2$-group $Z_2^n$ is isomorphic to the general linear group $GL(n,F_2)$ of degree $n$ over $F_2$. Let $W$ be the collection of permutation matrices of order $n$. It is clear that $W\le GL(n,F_2)$. In virtue of this, we consider the Cayley graph $Cay(Z_2^n,S)$, where $S$ is the union of some orbits under the action of $W$. We call such graphs the orbit Cayley graphs over $Z_2^n$. In this paper, we give eight infinite families of strongly regular graphs among orbit Cayley graphs over $Z_2^n$, in which six families are new as we know. By the way, we formulate the spectra of orbit Cayley graphs as well.

We present a bijection between permutation matrices and descending plane partitions without special parts, which respects the quadruple of statistics considered by Behrend, Di Francesco and Zinn--Justin. This bijection involves the inversion words of permutations and the "usual" representation of descending plane partitions as families of non--intersec\-ting lattice paths.

Motivated by the study of polytopes formed as the convex hull of permutation matrices and alternating sign matrices, we define several new families of polytopes as convex hulls of sign matrices, which are certain {0,1,-1}-matrices in bijection with semistandard Young tableaux. We investigate various properties of these polytopes, including their inequality descriptions, vertices, facets, and face lattices, as well as connections to alternating sign matrix polytopes and transportation polytopes.

Given two $k\times n$ matrices $A$ and $B$, we describe a couple of methods to solve the matrix equation $XA=BY$, where $X$ is an invertible $k\times k$ matrix, and $Y$ is an $n\times n$ permutation matrix, both of which we want to determine. We are interested in pursuing those techniques that have algebraic geometric flavor. An application to solving such a matrix equation comes from the cryptanalysis of McEliece cryptosystem. By using codewords of minimum weight of a linear code, in concordance with these methods of solving $XA=BY$, we present an efficient way to determine the entire encryption keys for the McEliece cryptosystems built on Reed-Solomon codes.

Keszegh (2009) proved that the extremal function $ex(n, P)$ of any forbidden light $2$-dimensional 0-1 matrix $P$ is at most quasilinear in $n$, using a reduction to generalized Davenport-Schinzel sequences. We extend this result to multidimensional matrices by proving that any light $d$-dimensional 0-1 matrix $P$ has extremal function $ex(n, P,d) = O(n^{d-1}2^{α(n)^{t}})$ for some constant $t$ that depends on $P$. To prove this result, we introduce a new family of patterns called $(P, s)$-formations, which are a generalization of $(r, s)$-formations, and we prove upper bounds on their extremal functions. In many cases, including permutation matrices $P$ with at least two ones, we are able to show that our $(P, s)$-formation upper bounds are tight.

The independence number of a square matrix $A$, denoted by $α(A)$, is the maximum order of its principal zero submatrices. Let $S_n^{+}$ be the set of $n\times n$ nonnegative symmetric matrices with zero trace. Denote by $J_n$ the $n\times n$ matrix with all entries equal to one. Given any integer $n$, we prove that a linear map $φ: S_n^+\rightarrow S_n^+$ satisfies $$α(φ(X))= α(X) {\quad\rm for~ all\quad}X\in S_n^+$$ if and only if there is a permutation matrix $P$ such that $$φ(X)=H\circ(P^TXP)\quad { \rm for~ all\quad}X\in S_n^+,$$ where $H=φ(J_n-I_n)$ with all off-diagonal entries positive.