ETNA Kent State University

Similar documents
A note on polar decomposition based Geršgorin-type sets

H-matrix theory vs. eigenvalue localization

Three Bounds on the Minimal Singular Value: A Comparison

Some inequalities for sum and product of positive semide nite matrices

Gershgorin s Circle Theorem for Estimating the Eigenvalues of a Matrix with Known Error Bounds

b jσ(j), Keywords: Decomposable numerical range, principal character AMS Subject Classification: 15A60

Journal of Inequalities in Pure and Applied Mathematics

Gershgorin type sets for eigenvalues of matrix polynomials

A SIMPLIFIED FORM FOR NEARLY REDUCIBLE AND NEARLY DECOMPOSABLE MATRICES D. J. HARTFIEL

Completions of P-Matrix Patterns

THE REAL AND THE SYMMETRIC NONNEGATIVE INVERSE EIGENVALUE PROBLEMS ARE DIFFERENT

On the adjacency matrix of a block graph

A note on 5 5 Completely positive matrices

First, we review some important facts on the location of eigenvalues of matrices.

1 Linear Regression and Correlation

c 1995 Society for Industrial and Applied Mathematics Vol. 37, No. 1, pp , March

Kernels of Directed Graph Laplacians. J. S. Caughman and J.J.P. Veerman

R, 1 i 1,i 2,...,i m n.

Square Roots of Real 3 3 Matrices vs. Quartic Polynomials with Real Zeros

For δa E, this motivates the definition of the Bauer-Skeel condition number ([2], [3], [14], [15])

A lower bound for the Laplacian eigenvalues of a graph proof of a conjecture by Guo

RESEARCH ARTICLE. An extension of the polytope of doubly stochastic matrices

VARIETIES OF ABELIAN TOPOLOGICAL GROUPS AND SCATTERED SPACES

Interlacing Inequalities for Totally Nonnegative Matrices

Springer Series in 36 Computational Mathematics

Notes on Linear Algebra and Matrix Theory

Math 322. Spring 2015 Review Problems for Midterm 2

Spectra of Semidirect Products of Cyclic Groups

MATH STUDENT BOOK. 11th Grade Unit 10

RITZ VALUE BOUNDS THAT EXPLOIT QUASI-SPARSITY

CONSTRUCTING PIECEWISE LINEAR 2-KNOT COMPLEMENTS

Copositive Plus Matrices

arxiv: v1 [math.na] 6 Aug 2010

Spectral inequalities and equalities involving products of matrices

Polynomials of small degree evaluated on matrices

A 3 3 DILATION COUNTEREXAMPLE

REFLEXIVITY OF THE SPACE OF MODULE HOMOMORPHISMS

Cospectral graphs and the generalized adjacency matrix

Completely positive matrices of order 5 with ĈP -graph

TOPOLOGICAL GROUPS MATH 519

Topological properties

Multiple Integrals and Probability Notes for Math 2605

The minimum rank of matrices and the equivalence class graph

MATH 304 Linear Algebra Lecture 8: Vector spaces. Subspaces.

THE RELATION BETWEEN THE QR AND LR ALGORITHMS

On the closures of orbits of fourth order matrix pencils

A NOTE ON A BASIS PROBLEM

On the distance and distance signless Laplacian eigenvalues of graphs and the smallest Gersgorin disc

Methods for Constructing Distance Matrices and the Inverse Eigenvalue Problem

Fast Matrix Multiplication*

QUALITATIVE CONTROLLABILITY AND UNCONTROLLABILITY BY A SINGLE ENTRY

Improved Upper Bounds for the Laplacian Spectral Radius of a Graph

AN EXTREMUM PROPERTY OF SUMS OF EIGENVALUES' HELMUT WIELANDT

COMPOSITIONS WITH A FIXED NUMBER OF INVERSIONS

Linear Algebra (Math-324) Lecture Notes

A Combinatorial Interpretation of the Numbers 6 (2n)! /n! (n + 2)!

The eigenvalues of an n x n complex matrix A are the roots of the n th degree polynomial det(a-λi) = 0. Further the positive square roots of the

Matrix Inequalities by Means of Block Matrices 1

To appear in Monatsh. Math. WHEN IS THE UNION OF TWO UNIT INTERVALS A SELF-SIMILAR SET SATISFYING THE OPEN SET CONDITION? 1.

Banach Journal of Mathematical Analysis ISSN: (electronic)

CAAM 454/554: Stationary Iterative Methods

RATIONAL REALIZATION OF MAXIMUM EIGENVALUE MULTIPLICITY OF SYMMETRIC TREE SIGN PATTERNS. February 6, 2006

INVESTIGATING THE NUMERICAL RANGE AND Q-NUMERICAL RANGE OF NON SQUARE MATRICES. Aikaterini Aretaki, John Maroulas

Math 4377/6308 Advanced Linear Algebra

In 1912, G. Frobenius [6] provided the following upper bound for the spectral radius of nonnegative matrices.

CANONICAL LOSSLESS STATE-SPACE SYSTEMS: STAIRCASE FORMS AND THE SCHUR ALGORITHM

ISOPERIMETRIC INEQUALITY FOR FLAT SURFACES

Strictly Positive Definite Functions on a Real Inner Product Space

A NEW EFFECTIVE PRECONDITIONED METHOD FOR L-MATRICES

Strictly positive definite functions on a real inner product space

Diagonalizing Hermitian Matrices of Continuous Functions

Math 203A - Solution Set 1

Spectral radius of bipartite graphs

x by so in other words ( )

Generalized Numerical Radius Inequalities for Operator Matrices

A Characterization of (3+1)-Free Posets

SUMSETS MOD p ØYSTEIN J. RØDSETH

General Properties for Determining Power Loss and Efficiency of Passive Multi-Port Microwave Networks

MATH 1553-C MIDTERM EXAMINATION 3

Two applications of the theory of primary matrix functions

NONNEGATIVE MATRICES EACH OF WHOSE POSITIVE DIAGONALS HAS THE SAME SUM

Circulant Matrices. Ashley Lorenz

A ROLE FOR DOUBLY STOCHASTIC MATRICES IN GRAPH THEORY

A Note on Eigenvalues of Perturbed Hermitian Matrices

ELA

On marker set distance Laplacian eigenvalues in graphs.

APPLICATIONS OF FD APPROXIMATIONS FOR SOLVING ORDINARY DIFFERENTIAL EQUATIONS

A Perron-type theorem on the principal eigenvalue of nonsymmetric elliptic operators

Centralizers of Finite Subgroups in Simple Locally Finite Groups

UNIQUENESS OF SOLUTIONS TO MATRIX EQUATIONS ON TIME SCALES

Rose-Hulman Undergraduate Mathematics Journal

Intrinsic products and factorizations of matrices

ELA

Implicit schemes of Lax-Friedrichs type for systems with source terms

LAYERED NETWORKS, THE DISCRETE LAPLACIAN, AND A CONTINUED FRACTION IDENTITY

On the Schur Complement of Diagonally Dominant Matrices

The Infinity Norm of a Certain Type of Symmetric Circulant Matrix

TWELVE LIMIT CYCLES IN A CUBIC ORDER PLANAR SYSTEM WITH Z 2 -SYMMETRY. P. Yu 1,2 and M. Han 1

The eigenvalue distribution of block diagonally dominant matrices and block H-matrices

Graphs with prescribed star complement for the. eigenvalue.

Transcription:

F N Electronic Transactions on Numerical Analysis Volume 8, 999, pp 5-20 Copyright 999, ISSN 068-963 ON GERŠGORIN-TYPE PROBLEMS AND OVALS OF CASSINI RICHARD S VARGA AND ALAN RAUTSTENGL Abstract Recently, two Geršgorin-type matrix questions were raised These are answered here, using ovals of Cassini ey words Geršgorin circle theorem, ovals of Cassini by AMS subject classification 5A8 Introduction Given a matrix, its usual Geršgorin disks are given ) and, if 54 "! $# %'&) %+* &-, denotes the spectrum of, ie, 3 54 A@BC D B 0/ 23 54 76 23 54829 #;:<# = 4 is an eigenvalue of EGF the famous Geršgorin circle theorem [3] or Horn and Johnson [4, p 344] and [7, p 6]) is that 2) 3 54H 6 23 54) M6 543N At this moment, only = = @O P the diagonal @ entries 6 2 Q4 F = @ / 23 54 and the row sums F of ) determine the Geršgorin disks F 6 54 and the inclusion set, of 2), in the complex plane Then, with the notation R A@S92T$0U0UVUW = these T = pieces of information are used to define the following set of matrices: 3) XY It is evident that notation Z[ \ J]2 C ]2P ^ _ P / 23ZG4) / Q4 : R and for all Z`4 Z_ X, for each, must also satisfy the inclusion of 2), and, with the 4) acbmd Z`4 e Received March 30, 998 Accepted for publication August 3, 998 Recommended by V Mehrmann Institute for Computational Mathematics, Department of Mathematics and Computer Science,, ent, Ohio 44242 USA) varga@msckentedu) This research supported by National Science Foundation Grant DMS-9707359 Department of Mathematics, Case Western Reserve University 0900 Euclid Ave, Cleveland, OH 4406 USA) axk66@pocwruedu) 5

p F F N 6 On a Geršgorin Problem and Ovals of Cassini we have 5) H 6 3 54N As we shall see, the inclusion in 5) is not always one of equality Recently, the following questions were asked of one of us: Question Can a precise description of be given Question 2 When is it the case that is a single closed disk with a nonempty interior) in the complex plane The point of this note is to answer the above questions We remark that the set of 4) =gf T is, for, in general larger than the minimal Geršgorin set cf [8]) for any matrix, simply _ h J P5D because the minimal Geršgorin set uses more specific information about the matrix in determining its minimal Geršgorin set ; A J PiC Specifically, the associated set of matrices, for the minimal Geršgorin set of a given matrix, is Xkj`l Z[ \ J] $$ ] _ ] : mg R 6) and for all ie, = in = = @ P addition to knowing the diagonal entries F, one is also given the moduli of all! 9O4 @ P nondiagonal entries F,on%'&),o* &$% =p, for a total of pieces of information Though it is not essential for this paper, we remark that it is evident from 3) and 6) that 7) X j`l H X so that X j`l 4qH rc for any =sf T$ where t it can be = f shown T that the equalities = rt of containment in 7) cannot in general hold for all, but in the case, there holds 8) X j`l X so that X j`l 4- uc p p N for any 2 On Question We need some additional notations and results Given the T vwt matrix 2) Zt ]MW x]0ip ] p x] p p p p its associated oval of Cassini is defined as 22) Then, for a matrix y 3ZG4 A@ C z k! ]0i {4+! ] p p 4 $#u ]MWp U] p C =;} T, we denote its ovals of Cassini by y M3 54 \@ k! P 24+! W 4 #;/ Q4^U / Q4 :~ tm R N 23) for all in J ƒ There are p y 7 Q4 ovals of Cassini associated with the matrix, and from a well-known result of A Brauer [] see also [4, p 380]), we have that if

Richard S Varga and Alan rautstengl 7 24) y Q4, n% &^,o* &$% y P 3 543 then 25) Q4 H y Q4N But, as the ovals of Cassini in 23) use only the information given by @ P F it is also true from 3) that and @ / 23 54 F Ic, H y Q4N 26) y 7 543 :ˆ~ tm We remark from 23) that each oval of Cassini, is a bounded and closed hence, compact) y Q4 set in the complex plane, as is their finite union in 24) Next, if is a set in, then denote its closure and Š <Œ denotes its complement The boundary of is defined, as usual, by O Š y 3 54 Ž A J $ = } T THEOREM 2 Given any, then Our first result, which sharpens the inclusion in 26), shows exactly how fills out 27) ) y 54- y Wp 543 when = T$ and 28) ) y 3 54 when =;}Ž N = rt Z X Proof For, each matrix in is, from 3), necessarily of the form Z[ W / 03 543 ' / p Q4 ' O p p with p N 29) arbitrary real numbers B Z Z If is any eigenvalue of, then det! BO 4^ u, so that from 29), i! B^4+3 p p! B^4) / 03 54^U / p 543 + N Hence, 20)! B U p p! B / M Q4^U / p 3 54N Bš As 20) corresponds to the case of equality in 23), we see that y ip Q4 Since this is true = for T B Z X y i any eigenvalue of any in and since, from 24), p 3 54c y 54 in this case, then ch y Wp 3 54œ y 54 Conversely, it is easily seen that each point of y ip 3 54 is, for suitable choices of real and p Z, an eigenvalue of some in 29), so ^ y ip Q4) y Q4 that, the desired result of 27)

@ ž 8 On a Geršgorin Problem and Ovals of Cassini =Y}_ Z h J] To establish 28), first assume that, and consider a matrix in, which has the partitioned form 2) where 22) with Ẑ W i ' Ÿ7 ' p p Z[ Z i Z ip Z p p with # Ÿ #s/ 03 543œ # # / p Q4 and Ÿ p arbitrary Ÿ s real ]2Wc A W numbers, / V Q4 9 # m #;= and with s / for all Now, for any choices of and with and p Q4 Ẑ i, the entries of the block p can be chosen so that the row =}[ / 0Z`4 / sums, and p 3ZG4 Z, in the first two rows of, equal those of Similarly, because, the row sums of the Z matrix p p of 2) can be chosen to be Z the same as those in the remaining row sums of Thus, by our construction, the matrix of 2) is /M an 54^ u X = element of We remark that this construction fails to work in the case, unless ) But from the partitioned form in 2), it is evident that 23) 3ZG4) Ẑ W M4 3Z p p 4N ŸM i However, from the parameters p Ẑ i in in 22), it can be seen from 23), that, for Ẑ each W y Wp 54, there are choices for these parameters such that is an eigenvalue of # In # / Ẑ W Ÿ # Ÿ # / M 54 other words, the eigenvalues of, on varying and with and p 54 y W, fill out p 54 Z, where we note that the remaining eigenvalues of namely, y Z those M3 54 of p p y 3 54 ) must still lie, from 26), in As this applies to all ovals of Cassini :D~ m Z, for, upon suitable permutations of the rows and columns of of 2), then ) y 54, = =s}d for all Z X For the remaining case of 28), any matrix in can be expressed as W ŸM ' / 03 54! Ÿ4 ' Z[ p p / p 3 54! 4 ' 24) / 3 54!š 4 where 25) # Ÿ #s/ 03 543œ # # / p Q4 Fi are arbitrary real numbers and # #;/ Q4 Now, choose any complex number in the oval of Cassini y i p Q4 satisfy 26) k! i U! p p #;/ 54U / p Q4N and, ie, cf 23)), let The above inequality implies that either Q! i œ#r/ M 54 or ˆ! p p ª#r/ p Q4 Assuming, without loss of generality, that ˆ! i ª#[/ 03 54 such that "! i AŸ and "! i œ Ÿ7 ' is valid Ÿ, choose the parameters and, and similarly, choose the parameters and

X X Richard S Varga and Alan rautstengl 9 such that k! p p and! p p Ÿ<U #Ž/ Q4 U /, where, from 26), p 3 54 '9M9+ Then with the vector «Z u in, it can be verified that «, where œ A i ±ŽŸM ' p ' ±; p p 27) ' ±z / 3 54!š 4 N Then, on setting / 3 543²VT and ³ ±;, we have h Z Thus, ««, and Z is an eigenvalue of Hence, each y W in p 54 Z X is an eigenvalue of some in Consequently, y Q4 = as this :~ construction tm for can be applied to any point of any oval of Cassini with, then ) y 3 54, which X completes the proof of 28) We remark that there is a way to enlarge the set, in the spirit of what has been done in the treatment =r} T of minimal Geršgorin sets cf [8]), which gives = \T a unified =u}r inclusion result for any, which does not distinguish between the cases and, as in Theorem 2 To this end, set XY µ 28) X H X µ so that 29) Zt A J]2 ]2P ^, and this of course implies H But from the definitions in 23), we also have 220) 22) THEOREM 22 Given any and #;/ Z`4 #s/ 3 54 for all : R µ N X µ 4H y Q4N Ž A J $ =;} T with, then µ ) y Q4N Proof From Theorem = 2 AT and the inclusions = T of 29) and 220), it is only µ necessary to establish 22) in the case, ie, for, it suffices to show that q y 3 54œ y ip 3 54 The Z proof µ = T of this is similar to the proof in the first part of Theorem 2 For, each matrix in is, from 28), of the form Z[ W Ÿ7 ' ' O with p p # Ÿ #\/ 3 54k # # / p Q4, and 222) B Z Z p arbitrary real numbers If is any eigenvalue of, then det! BO 4^ u implies that 3 W! B^4+ p p! B)4) Ÿ so that 3 i! B)42 p p! B^4 _Ÿ i for all real p N Ÿ / V Q4 / But as and run, respectively, through the intervals and p Q4, and as p Z µ run through all real y numbers, i it is evident that the eigenvalues of the matrices in out cf 23)) the set p Q4^ y 3 54 = gt µ, ie, in this case ^ y ip Q4 y 54 and fill

20 On a Geršgorin Problem and Ovals of Cassini t 3 On Question 2 The answer to Question 2 is straight-forward, based on the =} results T of Section 2 and the following observations Given a matrix, then, with the definitions of 6 +3 54 of ) and y 54 of 23), it is easy to verify that 3) y Q4¹H 6 3 54 6 3 54 : mº R :»~ [m for all with with equality ) _ W / 3 54) / 3 54) u $ holding/ W Q4- only if / Q4 f N or if and The set inclusions in 3) imply, with the definitions of 6 Q4 of 2) and y Q4 of 24), that one always has 32) y 3 54qH 6 3 54N ¼ J Pˆ So, to answer Question 2, it can be seen from Theorems 2 and 22 that, if $ = } T, = then ut a) for = ½T$ is never a single closed disk with a nonempty interior µ b) for W œ is a single closed disk with a nonempty interior only if p p / 0 54U / and p Q4 f ) X µ 4 is a simple closed disk with nonempty interior only = f T c) for Ÿ7 R Ÿ ~ if there are in with such y P3 54 ¾0¾c + / ¾ Q4U / i Q4 f that y ¾0 Q4 and, and all remaining ovals of Cassini lie in 4 Final Remarks It was a surprise for us to see that the determination of, in Theorem 2, was completely in terms of ovals of Cassini, a topic rarely seen in the current research literature in linear algebra In fact, on consulting many 6) books on linear algebra, we could find only three books where ovals of Cassini are even mentioned: in an exercise in Varga [7, p 22]), in a lengthier discussion in Horn and Johnson [4, pp 380-38], and in Marcus and Minc [5, p 49] But, in none of these books was it stated that the ovals of Cassini are Q4 in general at least as good cf 32)) as the Geršgorin disks, in estimating the spectrum $ of a given matrix in This appears in Brauer s paper [] of 947 which curiously does not mention Geršgorin s earlier paper [3] of 93) What apparently was of more interest in the literature is the fact that Brauer s use of two rows at a time, to determine an inclusion set for the eigenvalues of a given matrix, cannot in general be extended in the same manner to À rows at a time, for À }g, and explicit counterexamples to this are given in [4, p 382] and [5, p 49] However, a correct generalization of Brauer s results, using graph theory, is nicely given in Brualdi [2] REFERENCES [] A BRAUER, Limits for the characteristic roots of a matrix II, Duke Math J, 4 947), pp 2 26 [2] R BRUALDI, Matrices, eigenvalues, and directed graphs, Lin Multilin Alg 982), pp 43 65 [3] S GERSCHGORIN, Über die Abgrenzung der Eigenwerte einer Matrix, Izv Akad Nauk SSSR Ser Mat, 7 93), pp 749 754 [4] R A HORN AND C R JOHNSON, Matrix Analysis, Cambridge University Press, Cambridge, 985 [5] M MARCUS AND M MINC, A Survey of Matrix Theory and Matrix Inequalities, Allyn and Bacon, Boston, 964 [6] O TAUSSY, Bounds for characteristic roots of matrices, Duke Math J 5 948), pp 043-044 [7] R S VARGA, Matrix Iterative Analysis, Prentice-Hall, Englewood Cliffs, NJ, 962 [8] R S VARGA, Minimal Gerschgorin Sets, Pacific J Math, 5 965), pp 79 729 [9] X ZHANG AND D GU, A note on Brauer s Theorem, Lin Algebra Appl 96 994), pp 63-74