Totally Nonnegative Matrices
4.5/5
()
About this ebook
Totally nonnegative matrices arise in a remarkable variety of mathematical applications. This book is a comprehensive and self-contained study of the essential theory of totally nonnegative matrices, defined by the nonnegativity of all subdeterminants. It explores methodological background, historical highlights of key ideas, and specialized topics.
The book uses classical and ad hoc tools, but a unifying theme is the elementary bidiagonal factorization, which has emerged as the single most important tool for this particular class of matrices. Recent work has shown that bidiagonal factorizations may be viewed in a succinct combinatorial way, leading to many deep insights. Despite slow development, bidiagonal factorizations, along with determinants, now provide the dominant methodology for understanding total nonnegativity. The remainder of the book treats important topics, such as recognition of totally nonnegative or totally positive matrices, variation diminution, spectral properties, determinantal inequalities, Hadamard products, and completion problems associated with totally nonnegative or totally positive matrices. The book also contains sample applications, an up-to-date bibliography, a glossary of all symbols used, an index, and related references.
Read more from Shaun M. Fallat
Princeton Series in Applied Mathematics
Related to Totally Nonnegative Matrices
Titles in the series (34)
A Dynamical Systems Theory of Thermodynamics Rating: 0 out of 5 stars0 ratingsImpulsive and Hybrid Dynamical Systems: Stability, Dissipativity, and Control Rating: 5 out of 5 stars5/5Self-Regularity: A New Paradigm for Primal-Dual Interior-Point Algorithms Rating: 5 out of 5 stars5/5Entropy Rating: 5 out of 5 stars5/5Selfsimilar Processes Rating: 4 out of 5 stars4/5The Traveling Salesman Problem: A Computational Study Rating: 5 out of 5 stars5/5Analytic Theory of Global Bifurcation: An Introduction Rating: 0 out of 5 stars0 ratingsThermodynamics: A Dynamical Systems Approach Rating: 5 out of 5 stars5/5Auxiliary Signal Design for Failure Detection Rating: 0 out of 5 stars0 ratingsGraph Theoretic Methods in Multiagent Networks Rating: 5 out of 5 stars5/5Distributed Control of Robotic Networks: A Mathematical Approach to Motion Coordination Algorithms Rating: 5 out of 5 stars5/5Genomic Signal Processing Rating: 5 out of 5 stars5/5Positive Definite Matrices Rating: 5 out of 5 stars5/5Algebraic Curves over a Finite Field Rating: 5 out of 5 stars5/5Wave Scattering by Time-Dependent Perturbations: An Introduction Rating: 5 out of 5 stars5/5Control Theoretic Splines: Optimal Control, Statistics, and Path Planning Rating: 5 out of 5 stars5/5Matrices, Moments and Quadrature with Applications Rating: 5 out of 5 stars5/5Robust Optimization Rating: 5 out of 5 stars5/5Totally Nonnegative Matrices Rating: 5 out of 5 stars5/5Topics in Quaternion Linear Algebra Rating: 5 out of 5 stars5/5Modern Anti-windup Synthesis: Control Augmentation for Actuator Saturation Rating: 5 out of 5 stars5/5Hidden Markov Processes: Theory and Applications to Biology Rating: 5 out of 5 stars5/5Mathematical Analysis of Deterministic and Stochastic Problems in Complex Media Electromagnetics Rating: 5 out of 5 stars5/5Delay-Adaptive Linear Control Rating: 0 out of 5 stars0 ratingsMatrix Completions, Moments, and Sums of Hermitian Squares Rating: 5 out of 5 stars5/5Stability and Control of Large-Scale Dynamical Systems: A Vector Dissipative Systems Approach Rating: 3 out of 5 stars3/5Mathematical Methods in Elasticity Imaging Rating: 0 out of 5 stars0 ratingsPDE Control of String-Actuated Motion Rating: 0 out of 5 stars0 ratings
Related ebooks
Self-Regularity: A New Paradigm for Primal-Dual Interior-Point Algorithms Rating: 5 out of 5 stars5/5Spectral Radius of Graphs Rating: 0 out of 5 stars0 ratingsFundamentals of Advanced Mathematics V3 Rating: 0 out of 5 stars0 ratingsAbstract Methods in Partial Differential Equations Rating: 0 out of 5 stars0 ratingsDifferential Forms in Algebraic Topology Rating: 5 out of 5 stars5/5Real Analysis: A Historical Approach Rating: 0 out of 5 stars0 ratingsNonlinear Filtering and Smoothing: An Introduction to Martingales, Stochastic Integrals and Estimation Rating: 0 out of 5 stars0 ratingsBoundary Value Problems and Fourier Expansions Rating: 0 out of 5 stars0 ratingsInvitation to Dynamical Systems Rating: 5 out of 5 stars5/5Stochastic Control by Functional Analysis Methods Rating: 0 out of 5 stars0 ratingsQuadratic Form Theory and Differential Equations Rating: 0 out of 5 stars0 ratingsStochastic Analysis of Mixed Fractional Gaussian Processes Rating: 0 out of 5 stars0 ratingsEffective Dynamics of Stochastic Partial Differential Equations Rating: 0 out of 5 stars0 ratingsGraphs and Tables of the Mathieu Functions and Their First Derivatives Rating: 0 out of 5 stars0 ratingsCalculus of Variations Rating: 0 out of 5 stars0 ratingsNumerical Methods for Two-Point Boundary-Value Problems Rating: 0 out of 5 stars0 ratingsApplications of Finite Groups Rating: 0 out of 5 stars0 ratingsDynamic Programming and Stochastic Control Rating: 3 out of 5 stars3/5Bayesian Analysis of Stochastic Process Models Rating: 0 out of 5 stars0 ratingsConvexity and Graph Theory Rating: 0 out of 5 stars0 ratingsRandom Signal Analysis in Engineering Systems Rating: 0 out of 5 stars0 ratingsPeriodic Differential Equations: An Introduction to Mathieu, Lamé, and Allied Functions Rating: 0 out of 5 stars0 ratingsErgodic Theory and Topological Dynamics Rating: 0 out of 5 stars0 ratingsFourier Analysis in Probability Theory Rating: 0 out of 5 stars0 ratingsTheory of Lie Groups Rating: 0 out of 5 stars0 ratingsStochastic Stability and Control Rating: 0 out of 5 stars0 ratingsInequalities for Differential and Integral Equations Rating: 0 out of 5 stars0 ratingsSingular Optimal Control Problems Rating: 0 out of 5 stars0 ratingsFoundations of Stochastic Analysis Rating: 0 out of 5 stars0 ratingsPartial Differential Equations of Parabolic Type Rating: 0 out of 5 stars0 ratings
Mathematics For You
My Best Mathematical and Logic Puzzles Rating: 5 out of 5 stars5/5The Thirteen Books of the Elements, Vol. 1 Rating: 0 out of 5 stars0 ratingsThe Little Book of Mathematical Principles, Theories & Things Rating: 3 out of 5 stars3/5Standard Deviations: Flawed Assumptions, Tortured Data, and Other Ways to Lie with Statistics Rating: 4 out of 5 stars4/5Quantum Physics for Beginners Rating: 4 out of 5 stars4/5Basic Math & Pre-Algebra For Dummies Rating: 4 out of 5 stars4/5Algebra - The Very Basics Rating: 5 out of 5 stars5/5The Everything Everyday Math Book: From Tipping to Taxes, All the Real-World, Everyday Math Skills You Need Rating: 5 out of 5 stars5/5The Math Book: From Pythagoras to the 57th Dimension, 250 Milestones in the History of Mathematics Rating: 3 out of 5 stars3/5Calculus Made Easy Rating: 4 out of 5 stars4/5The Math of Life and Death: 7 Mathematical Principles That Shape Our Lives Rating: 4 out of 5 stars4/5The Everything Guide to Algebra: A Step-by-Step Guide to the Basics of Algebra - in Plain English! Rating: 4 out of 5 stars4/5Flatland Rating: 4 out of 5 stars4/5Mental Math Secrets - How To Be a Human Calculator Rating: 5 out of 5 stars5/5Game Theory: A Simple Introduction Rating: 4 out of 5 stars4/5ACT Math & Science Prep: Includes 500+ Practice Questions Rating: 3 out of 5 stars3/5Geometry For Dummies Rating: 5 out of 5 stars5/5Practice Makes Perfect Algebra II Review and Workbook, Second Edition Rating: 4 out of 5 stars4/5Real Estate by the Numbers: A Complete Reference Guide to Deal Analysis Rating: 0 out of 5 stars0 ratingsAlgebra I For Dummies Rating: 4 out of 5 stars4/5Alan Turing: The Enigma: The Book That Inspired the Film The Imitation Game - Updated Edition Rating: 4 out of 5 stars4/5The Elements of Euclid for the Use of Schools and Colleges (Illustrated) Rating: 0 out of 5 stars0 ratingsSummary of The Black Swan: by Nassim Nicholas Taleb | Includes Analysis Rating: 5 out of 5 stars5/5Algebra I Workbook For Dummies Rating: 3 out of 5 stars3/5
Reviews for Totally Nonnegative Matrices
2 ratings0 reviews
Book preview
Totally Nonnegative Matrices - Shaun M. Fallat
consecutively
Preface
A matrix is called totally nonnegative (resp. totally positive) if the determinant of every square submatrix is a nonnegative (resp. positive) number. This seemingly unlikely occurrence arises in a remarkable variety of ways (see Section 0.2 for a sample) and carries with it an array of very attractive mathematical structure. Indeed, it is the most useful and aesthetic matricial topic not covered in the broad references [HJ85] and [HJ91]. It is our purpose here to give a largely self-contained development of the most fundamental parts of that structure from a modern theoretical perspective. Applications of totally nonnegative matrices and related numerical issues are not the main focus of this work, but are recognized as being integral to this subject. We also mention a number of more specialized facts with references and give a substantial collection of references covering the subject and some related ideas.
Historically, the books [GK60, GK02, Kar68] and [GM96] and the survey paper [And87] have been the most common references in this area. However, each has a somewhat special perspective and, by now, each is missing some modern material. The most recent of these, [GM96], is more than fifteen years old and is a collection of useful, but noncomprehensive papers broadly in the area. In [Kar68] and [And87] the perspective taken leads to a notation and organization that can present difficulty to the reader needing particular facts. Perhaps the most useful and fundamental reference over a long period of time [GK60], especially now that it is readily available in English, is from the perspective of one motivating and important application and is now about sixty years old.
The present book takes a core, matrix theoretic perspective common to all sources of interest in the subject and emphasizes the utility of the elementary bidiagonal factorizations (see Chapter 2), which, though its roots are rather old, has only recently emerged as the primary tool (among very few available at present) to understand the beautiful and elaborate structure of totally nonnegative matrices. This tool is largely unused in the prior references. Along with the elementary bidiagonal factorization, planar diagrams, a recently appearing concept, are introduced as a conceptual combinatorial tool for analyzing totally nonnegative matrices. In addition to the seemingly unlimited use of these combinatorial objects, they lend themselves to obtaining prior results in an elegant manner, whereas in the past many results in this area were hard fought. As we completed this volume, the book [Pin10] appeared. It takes a different and more classical view of the subject, with many fewer references. Of course, its appearance is testimony to interest in and importance of this subject.
Our primary intent is that the present volume be a useful reference for all those who encounter totally nonnegative matrices. It could also be readily used to guide a seminar on the subject for either those needing to learn the ideas fundamental to the subject or those interested in an attractive segment of matrix analysis. For brevity, we have decided not to include textbook-type problems to solidify the ideas for students, but for purposes for which these might be useful, a knowledgeable instructor could add appropriate ones.
Our organization is as follows. After an initial introduction that sets major notation, gives examples, and describes several motivating areas, Chapter 1 provides much of the needed special background and preliminary results used throughout. (Any other needed background can be found in [HJ85] or a standard elementary linear algebra text.) Chapter 2 develops the ubiquitously important elementary bidiagonal factorization and related issues, such as LU factorization and planar diagrams that allow a combinatorial analysis of much about totally nonnegative matrices. Though an m-by-n matrix has many minors, total nonnegativity/positivity may be checked remarkably efficiently. This and related ideas, such as sufficient collections of minors, are the subject of Chapter 3.
Viewed as a linear transformation, it has long been known that a totally positive matrix cannot increase the sequential variation in the signs of a vector (an important property in several applications). The ideas surrounding this fact are developed in Chapter 4 and several converses are given. The eigenvalues of a totally positive matrix are positive and distinct and the eigenvectors are highly structural as well. A broad development of spectral structure and principal submatrices etc. are given in Chapter 5. Like positive semidefinite and M-matrices, totally nonnegative matrices enjoy a number of determinantal inequalities involving submatrices. These are explored in Chapter 6.
The remaining four chapters contain introductions to a variety of more specialized and emerging topics, including the distribution of rank in a TN matrix (Chapter 7); Hadamard (or entry-wise) products of totally nonnegative matrices (Chapter 8); various aspects of matrix completion problems associated with totally nonnegative matrices (Chapter 9); and a smorgesboard of further interesting topics are surveyed in the final chapter (Chapter 10).
An index and directory of terminology and notation is given at the end, as well as a lengthy list of relevant references, both ones referred to in the text and others of potential interest to readers. It is inevitable, unfortunately, that we may have missed some relevant work in the literature.
The reader should be alerted to an unfortunate schism in the terminology used among authors on this subject. We consistently use totally nonnegative
and totally positive
(among a hierarchy of refinements). Some authors use, instead, terms such as totally positive
(for totally nonnegative) and strictly totally positive
(for totally positive). This is at odds with conventional terminology for real numbers. We hope that there can be a convergence to the more natural and unambiguous terms over time.
We would like to thank B. Turner for a useful reading of a late draft of our manuscript, and several colleagues and friends for helpful suggestions. Finally, we note that the so-called sign-regular matrices (all minors of a given size share a common sign that may vary from size to size) share many properties with totally nonnegative and totally positive matrices. It could be argued that they present the proper level of generality in which to study certain parts of the structure. We have chosen not to focus upon them (they are mentioned occasionally) in order not to deflect attention from the most important sign-regular classes: namely the totally nonnegative and totally positive matrices.
SMF
CRJ
Chapter
Introduction
The central theme in this book is to investigate and explore various properties of the class of totally nonnegative matrices.
At first it may appear that the notion of total positivity is artificial; however, this class of matrices arises in a variety of important applications. For example, totally positive (or nonnegative) matrices arise in statistics (see [GM96, Goo86, Hei94]), mathematical biology (see [GM96]), combinatorics (see [BFZ96, Bre95, Gar96, Pen98]), dynamics (see [GK02]), approximation theory (see [GM96, Pri68, CP94B]), operator theory (see [Sob75]), and geometry (see [Stu88]). Historically, the theory of totally positive matrices originated from the pioneering work of Gantmacher and Krein ([GK60]) and was elegantly brought together in a revised version of their original monograph [GK02]). Also, under the influence of I. Schoenberg (see, for example, [Sch30]), Karlin published an influential treatise, Total Positivity ([Kar68]), which mostly concerns totally positive kernels but also deals with the discrete version, totally positive matrices. Since then there has been a considerable amount of work accomplished on total positivity, some of which is contained in the exceptional survey paper by T. Ando ([And87]). Two more recent survey papers have appeared ([Fal01] [Hog07, Chap. 21]) and both take the point of view of bidiagonal factorizations of totally nonnegative matrices.
Before we continue, as mentioned in the Preface, we issue here a word of warning. In the existing literature, the terminology used is not always consistent with what is presented above. Elsewhere in the literature the term totally positive (see below) corresponds to strictly totally positive
(see, for example, the text [Kar68]) and the term totally nonnegative defined here sometimes corresponds to totally positive
(as in some of the papers in [GM96]).
0.0 DEFINITIONS AND NOTATION
The set of all m-by-n , the set of all real numbers, then we may shorten this notation to Mm,n. Our matrices will, for the most part, be real entried. In the case m=nor Mm,n or Mnwill indicate that the entries of A . The transpose of a matrix A will be denoted by AT.
, of A is the submatrix obtained from A by deleting the rows indexed by α and columns indexed by β. Throughout the book, we let αc denote the complement , then the submatrix A[α, β] is referred to as a nonprincipal submatrix of A, we refer to the principal submatrix A[α] as a leading principal submatrix. In the same manner, for an n, x[α] denotes the entries of x denote the n-by-n diagonal matrix with main diagonal entries xiby M and N, respectively.
A minor in a given matrix A , the minor detA[α, β] may also be denoted by Aα, β, and the principal minor (if A is square) detA[α] may be abbreviated to Aα.
with all positive (nonnegative). The former are called the totally positive (totally nonnegative) matrices and are denoted by TP (TN); we denote the latter as TPk (TNk). TP (TN) is used both as an abbreviation and as a name of a set. The reader should consult [HJ85, HJ91] or any other standard reference on matrix theory for standard notation and terminology not defined herein. In terms of compounds, TPk, for example, simply means that the first through the kth compounds are entry-wise positive.
Several related classes of matrices are both of independent interest and useful in developing the theory of TP (TN) matrices. We give our notation for those classes here; we continue to use the subscript k to denote a requirement made on all minors of no more than k rows. No subscript indicates that the requirement is made on all minors. Each class is defined irrespective of the m and n in Mm,n, except that a subscript k .
By InTN (InTNk), we denote the square and invertible TN matrices (TNk matrices), and by IrTN (IrTNk), we mean the square and irreducible TN matrices (TNk matrices). The intersection of these two classes is denoted by IITN (IITNk); this class is also referred to as the oscillatory matrices [GK60], which are defined as those TN matrices with a TP integral power. The equivalence will be demonstrated in Chapter 2. We may also use the symbol OSC to denote oscillatory matrices.
A TN (TNk) matrix may have a zero line (row or column), which often leads to an exception to convenient statements. So we denote TN (TNk) matrices with no zero lines as TN’ (TNk′).
A square, triangular matrix cannot be TP, but one that is TN and has all its minors positive unless they are identically 0 because only of the triangular structure is called triangular TP) are the intersections of upper ΔTP and lower ΔTP matrices, and thus are InTN.
For convenience of reference we present these classes in tabular form in Table 1. Note: Omission of subscript k means all minors.
Table 1: Various Classes of Matrices of Interest
The ith standard basis vector is the n-vector whose only nonzero entry occurs in the ith component and that entry is a one, and is denoted by ei; the (i,j)th standard basis matrix, the m-by-n matrix whose only nonzero entry is in the (i,j)th position and this entry is a one, will be denoted by Eij.
Observe that if m=n. We also let e denote the n-vector consisting of all ones (the size of e will be determined from the context). Finally, we let Jm,n ) and In denote the m-by-n matrix of all ones and the n-by-n identity matrix, respectively. The subscript is dropped when the sizes of these matrices is clear from the context.
0.1 JACOBI MATRICES AND OTHER EXAMPLES OF TN MATRICES
An n-by-n is called a diagonal . For example,
is a 4-by-4 diagonal matrix. We may also write A .
An n-by-n is referred to as a tridiagonal . For example,
are referred to as the superdiagonal, and the entries of A are called subdiagonal.
Hence a matrix is tridiagonal if the only nonzero entries of A are contained on its sub-, main, and superdiagonal.
Tridiagonal matrices are perhaps one of the most studied classes of matrices. Much of the reason for this is that many algorithms in linear algebra require significantly less computational labor when they are applied to tridiagonal matrices. Some elementary examples include
(1) eigenvalues,
(2) solving linear systems,
(3) LU factorization,
(4) evaluating determinants.
can be evaluated by solving the recursive equation
Equation (1) is a simple consequence of the Laplace expansion of det A are (smaller sized) tridiagonal matrices.
The recursive relation (1) also ties the eigenvalues of a (symmetric) tridiagonal matrix to certain orthogonal polynomials (Chebyshev), as the characteristic polynomials of A and its principal submatrices also satisfy a relation like (1). In addition, (1) arises in the Runge-Kutta methods for solving partial differential equations.
Suppose the tridiagonal matrix A has the form
Observe that a tridiagonal matrix A .
An n-by-n matrix is called a P0-matrix if all its principal minors are nonnegative. We let P0 denote the class of all P0-matrices. Furthermore, an n-by-n matrix is called a P-matrix if all its principal minors are positive. We let P denote the class of all P-matrices.
Suppose A , then A is a P0-matrix. To verify this claim we need the following fact. If A can be written as
, where ⊕ denotes direct sum. In particular,
.
Consider the case n = 3. Then by (1),
it follows from (1) that
Consider the general case and let A’ be the n-by-n matrix obtained from A by using an elementary row operation to eliminate c,
.
Furthermore, we note that by (1) we have
So, if a nonnegative, irreducible tridiagonal matrix satisfies the simple dominance relation it is a P0-matrix. In fact, even more is true.
it is a simple consequence that
.
is any submatrix of A is a product of principal minors and nonprincipal minors, where the principal minors are based on contiguous index sets and the nonprincipal minors are simply entries from the super- and/or subdiagonal.
For example, suppose A is a 10-by-10 tridiagonal matrix of the form in (2).
Then
has all its minors nonnegative.
We remark here that the irreducibility assumption was needed for convenience only. If the tridiagonal matrix is reducible, then we simply concentrate on the direct summands and apply the same arguments as above.
The upshot of the previous discussion is that if A , then A is TN.
In the pivotal monograph [GK60], the chapter on oscillatory matrices began by introducing a class of tridiagonal matrices the authors called Jacobi matrices. An n-by-n is a tridiagonal matrix of the form
is called a normal Jacobi matrix . Jacobi matrices were of interest to Gantmacher and Krein for at least two reasons. First, many of the fundamental properties of general oscillatory matrices can be obtained by an examination of analogous properties of Jacobi matrices. Second Jacobi matrices are among the basic types of matrices that arose from studying the oscillatory properties of an elastic segmental continuum (no supports between the endpoints) under small transverse oscillations.
One of the first things that we observe about Jacobi matrices is that they are not TN or even entrywise nonnegative. However, they are not that far off, as we shall see in the next result.
The following result on the eigenvalues of a tridiagonal matrix is well known, although we present a proof here for completeness.
Lemma 0.1.1 Let T be an n-by-n tridiagonal matrix in the form
, then the eigenvalues of T are real and have algebraic multiplicity one (i.e., are simple). Moreover, T is similar (via a positive diagonal matrix) to a symmetric nonnegative tridiagonal matrix.
Proof. be the n-by-n , and for k . Then it is readily verified that
is symmetric and T have the same eigenvalues, this implies that the eigenvalues of T are real. Suppose λ is an eigenvalue of Tare deleted, then the resulting matrix is an (n-1)-by-(n. Hence this submatrix has rank nhas rank at least nhas rank at most n-1 since λ is an eigenvalue of Tis also a nonnegative matrix.
A diagonal matrix is called a signature matrix if all its diagonal entries are ± 1. Furthermore, two n-by-n matrices A and B are said to be signature similar , where S is a signature matrix. It is not difficult to verify that a normal Jacobi matrix is signature similar to a nonnegative tridiagonal matrix. Furthermore, the eigenvalues of a normal Jacobi matrix are real and distinct.
), then it is similar to a symmetric TN tridiagonal matrix. In particular (since a symmetric TN matrix is an example of a positive semidefinite matrix [HJ85]), the eigenvalues are nonnegative and distinct by Lemma 0.1.1.
Moreover, if the above normal Jacobi matrix is also invertible, then the eigenvalues are positive and distinct.
Since the eigenvalues are distinct, each eigenvalue has exactly one eigenvector (up to scalar multiple), and these (unique) eigenvectors satisfy many more interesting sign
properties.
Let
above, define
Recall that the functions Dk satisfy the recurrence relation
. Evidently,
and so on.
A key observation that needs to be made is the following:
differ from zero and have opposite sign."
Suppose that, for some k,
Go back to the recurrence relation (5), and notice that if the above inequalities hold, then
. But then observe
. Now we arrive at a contradiction since
must reverse sign from + to −. A consequence of this is that the roots of successive Dk(that is, the roots interlace).
Observation: The sequence
has j.
.
. Assume for the moment that for each i(for simplicity’s sake).
Suppose
yields the system of equations
; hence the first n-1 equations are linearly independent, and the nth equation is a linear combination of the previous n-1.
Looking more closely at the first n - 1 equations we arrive at the relation
Hence the coordinates of x .
has exactly j - 1 sign changes. The interlacing of sign changes in the eigenvectors follows as well.
We close with a remark that offers a nice connection between tridiagonal TN matrices and a