Richard A. Brualdi • Ángeles Carmona P. van den Driessche • Stephen Kirkland Dragan Stevanović
Combinatorial Matrix Theory
Editors for this volume: Andrés M. Encinas, Universitat Politècnica de Catalunya Margarida Mitjana, Universitat Politècnica de Catalunya Richard A. Brualdi Ángeles Carmona Mathematics Department Departament de Matemàtiques University of Wisconsin Universitat Politècnica de Catalunya Madison, WI, USA Barcelona, Spain
P. van den Driessche Stephen Kirkland Department of Mathematics and Statistics Department of Mathematics University of Victoria University of Manitoba Victoria, BC, Canada Winnipeg, MB, Canada
Dragan Stevanović Mathematical Institute Serbian Academy of Sciences and Arts Belgrade, Serbia
ISSN 2297-0304 ISSN2297-0312 (electronic) Advanced Courses in Mathematics - CRM Barcelona ISBN 978-3-319-70952-9 ISBN 978-3-319-70953-6 (eBook) https://doi.org/10.1007/978-3-319-70953-6
Library of Congress Control Number: 2018935110
© Springer International Publishing AG, part of Springer Nature 2018 Foreword
This book contains the notes of lectures delivered by Richard A. Brualdi, Angeles´ Carmona, Stephen Kirkland, Dragan Stevanovi´cand Pauline van den Driessche at the Centre de Recerca Matem`atica(CRM) in Bellaterra, Barcelona from June 29th to July 3rd, 2015. The Advanced Course on Combinatorial Matrix Theory was mainly addressed to PhD students and post-docs in related areas, and also to more established researchers who wanted to get involved in these subjects. Combinatorial matrix theory is a rich branch of matrix theory; it is both an active field of research and a widespread toolbox for many scientists. Combina- torial properties of matrices are studied on the basis of qualitative rather than quantitative information, so that the ideas developed can provide consistent infor- mation about a model even when the data is incomplete or inaccurate. The theory behind qualitative methods can also contribute to the development of effective quantitative matrix methods. The topics covered in the Advanced Course included permutation matrices, alternating sign matrices, tournaments, sign pattern matrices, minimum rank and its distribution, boundary value problems on finite networks, the group inverse for the Laplacian matrix of a graph, and bounds on the spectral radius of the Laplacian matrix. The activity consisted of five series of four lectures each. Accordingly, the material is divided into five chapters in the book. There were also two short sessions devoted to informal presentations of the current work of most of the participants. The first chapter corresponds to the lectures delivered by Richard A. Brualdi on some combinatorially defined matrix classes, specifically on three families, namely permutation, alternating sign, and tournament matrices. Permutation ma- trices are the matrices associated with permutations, and it is shown that the Bruhat order on this symmetric group is related to Gaussian elimination and leads to the Bruhat decomposition of a nonsingular matrix. Alternating sign ma- trices are generalizations of permutation matrices, and the extension of Bruhat order is a lattice, specifically its McNielle completion. A tournament matrix is the adjacency matrix corresponding to a tournament; that is, an orientation of the complete graph. The generation problem is presented and analyzed on loopy, Hankel, and skew-Hankel tournaments. The second chapter is devoted to the series of lectures given by Pauline van den Driessche on sign pattern matrices. It discusses the study of spectral properties of matrices with a given sign pattern. Several classes of sign patterns are reviewed, including sign patterns that allow all possible spectra; those allowing all possible inertias; those allowing stability; and those that may give rise to Hopf bifurcation in associated dynamical systems. Some classes of these matrices are explored in more detail using techniques from matrix theory, graph theory, and analysis. Moreover, some open problems are suggested to encourage further work. Dragan Stevanovi´cdelivered lectures on the spectral radius of graphs, col- lected in Chapter 3. The spectral radius of a graph is a tool to provide bounds of parameters related to the properties of a graph. More precisely, eigenvalues and eigenvectors of graph matrices have become standard mathematical tools nowa- days due to their wide applicability in network analysis and computer science, with the most prominent graph matrices being the adjacency and the Laplacian matrix. In this chapter, lower and upper limits of the spectral radius of adjacency and Laplacian matrices are addressed, with special attention to testing techniques and common properties of shapes of the boundaries. In addition, approximate formulas for the spectral radius of the adjacency matrix are also discussed. The fourth chapter contains the lectures delivered by Stephen Kirkland on the group inverse for the Laplacian matrix of a graph. In recent times, Laplacian matrices for undirected graphs have received a good deal of attention, in part because the spectral properties of the Laplacian matrix are related to a number of features of interest of the underlying graph. It turns out that a certain generalized inverse, the group inverse, of a Laplacian matrix also carries information about the graph in question. This chapter explores the group inverse of the Laplacian matrix and its relationship to graph structure. Connections with the algebraic connectivity and the resistance distance are made, and the computation of the group inverse of a Laplacian matrix is also considered from a numerical viewpoint. The last chapter, authored by Angeles´ Carmona, is devoted to the study of boundary value problems on finite networks. The starting point is the description of the basic difference operators: the derivative, gradient, divergence, curl and Laplacian, or more generally, Schr¨odingeroperators. The next step is to define the discrete analogue of a manifold with a boundary, which includes the concept of outer normal field and proves the Green identities. At that point, the focus is on some aspects of discrete potential theory. The discrete analog of the Green and Poisson kernels are defined and their relationship with the so-called Dirichlet- to-Neuman map established. Finally, some applications to matrix theory and to organic chemistry, such as the M-inverse problem and the Kirchhoff index compu- tation, are considered. We would like to express our gratitude to the director, Prof. Joaquim Bruna, and staff of the Centre de Recerca Matem`aticanot only for their excellent job in organizing this course but also for their kindness and support during the event. We are also in debt to Elsevier, the Societat Catalana de Matem`atiques,and the Real Sociedad Matem´aticaEspa˜nolafor their financial support. Finally, we thank all the participants for their active involvement and spe- cially to the five lecturers for their accurate preparation of these notes. We hope that their publication will contribute to increasing knowledge of combinatorial matrix theory. Andr´esM. Encinas and Margarida Mitjana Contents
Foreword v
1 Some Combinatorially Defined Matrix Classes 1 By Richard A. Brualdi
1.1 Permutations and Permutation Matrices ...... 1 1.1.1 Basic Properties ...... 1 1.1.2 Generation ...... 3 1.1.3 Bruhat Order ...... 3 1.1.4 Matrix Bruhat Decomposition ...... 5 1.1.5 Flags ...... 8 1.1.6 Involutions and Symmetric Integral Matrices ...... 11 1.2 Alternating Sign Matrices ...... 15 1.2.1 Basic Properties ...... 15 1.2.2 Other Views of ASMs ...... 16 1.2.3 The λ-determinant ...... 19 1.2.4 Maximal ASMs ...... 20 1.2.5 Generation ...... 21 1.2.6 MacNeille Completion and the Bruhat Order ...... 22 1.2.7 Bruhat Order Revisited ...... 25 1.2.8 Spectral Radius of ASMs ...... 31 1.3 Tournaments and Tournament Matrices ...... 32 1.3.1 The Inverse Problem ...... 33 1.3.2 Generation ...... 34 1.3.3 Loopy Tournaments and Their Generation ...... 36 1.3.4 Hankel Tournaments and Their Generation ...... 38 1.3.5 Combinatorially Skew-Hankel Tournaments and Their Generation ...... 40 Bibliography ...... 45 2 Sign Pattern Matrices 47 By P. van den Driessche
2.1 Introduction to Sign Pattern Matrices ...... 47 2.1.1 Notation and Definitions ...... 47 2.2 Potential Stability of Sign Patterns ...... 48 2.2.1 Stability Definitions ...... 48 2.2.2 Stability of a Dynamical System ...... 49 2.2.3 Characterization of Sign Stability ...... 50 2.2.4 Basic Facts for Potential Stability ...... 51 2.2.5 Known Results on Potential Stability for Small Orders . . . 52 2.2.6 Sufficient Condition for Potential Stability ...... 52 2.2.7 Construction of Higher-Order Potentially Stable Sign Patterns ...... 54 2.2.8 Number of Nonzero Entries ...... 56 2.2.9 Open Problems Related to Potential Stability ...... 57 2.3 Spectrally Arbitrary Sign Patterns ...... 57 2.3.1 Some Definitions Relating to Spectra of Sign Patterns . . . 57 2.3.2 A Family of Spectrally Arbitrary Sign Patterns ...... 59 2.3.3 Minimal Spectrally Arbitrary Patterns and Number of Nonzero Entries ...... 61 2.3.4 Reducible Spectrally Arbitrary Sign Patterns ...... 62 2.3.5 Some Results on Potentially Nilpotent Sign Patterns . . . . 63 2.3.6 Some Open Problems Concerning SAPs ...... 64 2.4 Refined Inertia of Sign Patterns ...... 64 2.4.1 Definition and Maximum Number of Refined Inertias . . . . 64 2.4.2 The Set of Refined Inertias Hn ...... 65 2.4.3 Sign Patterns of Order 3 and H3 ...... 66 2.4.4 Sign Patterns of Order 4 and H4 ...... 66 2.4.5 Sign Patterns with All Diagonal Entries Negative ...... 67 2.4.6 Detecting Periodic Solutions in Dynamical Systems . . . . . 68 2.4.7 Some Open Problems Concerning Hn ...... 72 2.5 Inertially Arbitrary Sign Patterns ...... 73 2.5.1 Definition and Relation to Other Properties ...... 73 2.5.2 Generalization of the Nilpotent-Jacobian Method ...... 74 2.5.3 Reducible IAPs ...... 76 2.5.4 A Glimpse at Zero-Nonzero Patterns ...... 76 2.5.5 A Taste of More General Patterns ...... 77 2.5.6 Some Open Problems Concerning IAPs ...... 78 Bibliography ...... 79 3 Spectral Radius of Graphs 83 By Dragan Stevanovi´c 3.1 Graph-Theoretical Definitions ...... 83 3.2 The Adjacency Matrix and Its Spectral Properties ...... 85 3.3 The Big Gun Approach ...... 89 3.4 The Eigenvector Approach ...... 95 3.5 The Characteristic Polynomial Approach ...... 105 3.6 Walk Counting ...... 110 Bibliography ...... 127
4 The Group Inverse of the Laplacian Matrix of a Graph 131 By Stephen Kirkland 4.1 Introduction ...... 131 4.2 The Laplacian Matrix ...... 132 4.3 The Group Inverse ...... 138 4.4 L# and the Bottleneck Matrix ...... 141 4.5 L# for Weighted Trees ...... 144 4.6 Algebraic Connectivity ...... 148 4.7 Joins ...... 152 4.8 Resistance Distance ...... 154 4.9 Computational Considerations ...... 160 4.10 Closing Remarks ...... 168 Bibliography ...... 171
5 Boundary Value Problems on Finite Networks 173 By Angeles´ Carmona 5.1 Introduction ...... 173 5.2 The M-Matrix Inverse Problem ...... 174 5.3 Difference Operators on Networks ...... 176 5.3.1 Schr¨odinger Operators ...... 181 5.4 Glossary ...... 184 5.5 Networks with Boundaries ...... 185 5.6 Self-Adjoint Boundary Value Problems ...... 188 5.7 Monotonicity and the Minimum Principle ...... 193 5.8 Green and Poisson Kernels ...... 195 5.9 The Dirichlet-to-Robin Map ...... 199 5.10 Characterization of Symmetric M-Matrices as Resistive Inverses . 201 5.10.1 The Kirchhoff Index and Effective Resistances ...... 202 5.10.2 Characterization ...... 204 5.11 Distance-regular Graphs with the M-Property ...... 207 5.11.1 Strongly Regular Graphs ...... 209 5.11.2 Distance-regular Graphs with Diameter 3 ...... 210 Bibliography ...... 215 Chapter 1
Some Combinatorially Defined Matrix Classes by Richard A. Brualdi
1.1 Permutations and Permutation Matrices
In this section we consider the symmetric group of permutations of a finite set and their partial order known as the Bruhat order. Regarding a permutation as a permutation matrix, this partial order is related to Gaussian elimination and leads to the matrix Bruhat decomposition of a nonsingular matrix, and then to a characterization of flags in a vector space. We also describe a correspondence between permutations that are involutions (symmetric permutation matrices) and a certain class of nonnegative integral matrices.
1.1.1 Basic Properties
One of the most basic concepts in mathematics is that of a permutation of a finite set. Let σ be a permutation of {1, 2, . . . , n}, where we write σ = (i1, i2, . . . , in) to denote that σ(k) = ik for k = 1, 2, . . . , n. Here, i1, i2, . . . , in are distinct and {i1, i2, . . . , in} = {1, 2, . . . , n}. We denote the set of all permutations of {1, 2, . . . , n} by Sn. Each permutation σ of {1, 2, . . . , n} can be identified with an n × n permu- tation matrix P = Pσ = [pij], where p1i1 = p2i2 = ··· = pnin = 1 and pij = 0 otherwise. 2 Chapter 1. Some Combinatorially Defined Matrix Classes
Example 1.1.1. With n = 5 and σ = (3, 5, 4, 2, 1) we have
0 0 1 0 0 1 0 0 0 0 1 1 P = 0 0 0 1 0 or, as we often write, P = 1 , 0 1 0 0 0 1 1 0 0 0 0 1 blocking the rows and columns and suppressing the zeros.
Let σ = (i1, i2, . . . , in) be a permutation. An ascent of σ is a pair k, k+1 with ik < ik+1. We also say that an ascent occurs at position k. Similarly, a descent of σ is a pair k, k +1 with ik > ik+1, and we also say that a descent occurs at position k. The permutation σ in Example 1.1.1 has an ascent occuring at position 1 and descents occuring at positions 2, 3, and 4. In general the number of ascents plus the number of descents in a permutation in Sn equals n − 1. An inversion in a permutation σ = (i1, i2, . . . , in) is less restrictive than a descent; it is a pair k, l of positions with k < l such that ik > il. Thus an inversion corresponds to a pair of integers in σ which are out of the natural order 1, 2, . . . , n. In terms of the permutation matrix P , an inversion corresponds to a 2 × 2 submatrix of P ,
0 1 P [k, l | i , i ] = = Q , l k 1 0 2 in rows k and l, and columns il and ik. The transposition that interchanges ik and il in σ, and thus replaces the 2×2 submatrix P [k, l | il, ik] in the permutation matrix P with the 2 × 2 identity matrix I2, is a new permutation with fewer inversions, perhaps a lot fewer. But if each of the integers il+1, il+2, . . . , ik−1 is either smaller than il or larger than ik, then the transposition decreases the number of inversions by exactly one. In terms of the permutation matrix P , if the (l −k −1)×(l −k −1) submatrix of P determined by rows k+1, . . . , l−1 and columns il+1, il+2, . . . , ik−1 is a zero matrix, then the replacement of P [k, l | ik, il] = Q2 with I2 decreases the number of inversions by exactly one. Example 1.1.2. The permutation σ in Example 1.1.1 has eight inversions. The identity matrix In (corresponding to the permutation σ = (1, 2, . . . , n)) has no inversions. The anti-identity matrix Qn (corresponding to the permutation σ = n 2 (n, n−1,..., 1)) has the maximum number 2 of inversions and satisfies Qn = In. For example, 1 1 Q5 = 1 1 1 1.1. Permutations and Permutation Matrices 3