An iterative method to solve a nonlinear matrix equation

Similar documents
A new error bound for linear complementarity problems for B-matrices

Common Coupled Fixed Point of Mappings Satisfying Rational Inequalities in Ordered Complex Valued Generalized Metric Spaces

Eigenvalue localization for complex matrices

On the Variations of Some Well Known Fixed Point Theorem in Metric Spaces

PAijpam.eu ON TENSOR PRODUCT DECOMPOSITION

Introduction to Optimization Techniques. How to Solve Equations

The log-behavior of n p(n) and n p(n)/n

Estimation of Backward Perturbation Bounds For Linear Least Squares Problem

Research Article A New Second-Order Iteration Method for Solving Nonlinear Equations

A General Iterative Scheme for Variational Inequality Problems and Fixed Point Problems

Computation of Error Bounds for P-matrix Linear Complementarity Problems

ITERATIVE SOLUTION OF TWO MATRIX EQUATIONS

APPROXIMATE FUNCTIONAL INEQUALITIES BY ADDITIVE MAPPINGS

Some New Iterative Methods for Solving Nonlinear Equations

Differentiable Convex Functions

Modified Decomposition Method by Adomian and. Rach for Solving Nonlinear Volterra Integro- Differential Equations

A collocation method for singular integral equations with cosecant kernel via Semi-trigonometric interpolation

Disjoint unions of complete graphs characterized by their Laplacian spectrum

TRACES OF HADAMARD AND KRONECKER PRODUCTS OF MATRICES. 1. Introduction

Introduction to Optimization Techniques

Higher-order iterative methods by using Householder's method for solving certain nonlinear equations

Properties of Fuzzy Length on Fuzzy Set

II. EXPANSION MAPPINGS WITH FIXED POINTS

IJITE Vol.2 Issue-11, (November 2014) ISSN: Impact Factor

Sequences and Series of Functions

1+x 1 + α+x. x = 2(α x2 ) 1+x

Some Variants of Newton's Method with Fifth-Order and Fourth-Order Convergence for Solving Nonlinear Equations

Find quadratic function which pass through the following points (0,1),(1,1),(2, 3)... 11

LOWER BOUNDS FOR THE BLOW-UP TIME OF NONLINEAR PARABOLIC PROBLEMS WITH ROBIN BOUNDARY CONDITIONS

Some Common Fixed Point Theorems in Cone Rectangular Metric Space under T Kannan and T Reich Contractive Conditions

ON WELLPOSEDNESS QUADRATIC FUNCTION MINIMIZATION PROBLEM ON INTERSECTION OF TWO ELLIPSOIDS * M. JA]IMOVI], I. KRNI] 1.

NEW FAST CONVERGENT SEQUENCES OF EULER-MASCHERONI TYPE

Sequences and Limits

On Nonsingularity of Saddle Point Matrices. with Vectors of Ones

6.3 Testing Series With Positive Terms

Solving a Nonlinear Equation Using a New Two-Step Derivative Free Iterative Methods

On Weak and Strong Convergence Theorems for a Finite Family of Nonself I-asymptotically Nonexpansive Mappings

On forward improvement iteration for stopping problems

Oscillation and Property B for Third Order Difference Equations with Advanced Arguments

NUMERICAL METHOD FOR SINGULARLY PERTURBED DELAY PARABOLIC PARTIAL DIFFERENTIAL EQUATIONS

CS537. Numerical Analysis and Computing

Product measures, Tonelli s and Fubini s theorems For use in MAT3400/4400, autumn 2014 Nadia S. Larsen. Version of 13 October 2014.

DECOMPOSITION METHOD FOR SOLVING A SYSTEM OF THIRD-ORDER BOUNDARY VALUE PROBLEMS. Park Road, Islamabad, Pakistan

Numerical Solution of the Two Point Boundary Value Problems By Using Wavelet Bases of Hermite Cubic Spline Wavelets

CS321. Numerical Analysis and Computing

A Note on the Kolmogorov-Feller Weak Law of Large Numbers

Research Article Quasiconvex Semidefinite Minimization Problem

arxiv: v2 [math.fa] 21 Feb 2018

Exact Solutions for a Class of Nonlinear Singular Two-Point Boundary Value Problems: The Decomposition Method

A note on the modified Hermitian and skew-hermitian splitting methods for non-hermitian positive definite linear systems

A Genetic Algorithm for Solving General System of Equations

Solution of Differential Equation from the Transform Technique

Math Solutions to homework 6

ON POINTWISE BINOMIAL APPROXIMATION

Chapter 3 Inner Product Spaces. Hilbert Spaces

Research Article A Note on Ergodicity of Systems with the Asymptotic Average Shadowing Property

Brief Review of Functions of Several Variables

A 2nTH ORDER LINEAR DIFFERENCE EQUATION

Approximation theorems for localized szász Mirakjan operators

THE SOLUTION OF NONLINEAR EQUATIONS f( x ) = 0.

Bounds for the Extreme Eigenvalues Using the Trace and Determinant

Numerical Method for Blasius Equation on an infinite Interval

62. Power series Definition 16. (Power series) Given a sequence {c n }, the series. c n x n = c 0 + c 1 x + c 2 x 2 + c 3 x 3 +

Iterative Techniques for Solving Ax b -(3.8). Assume that the system has a unique solution. Let x be the solution. Then x A 1 b.

On the Determinants and Inverses of Skew Circulant and Skew Left Circulant Matrices with Fibonacci and Lucas Numbers

Math 341 Lecture #31 6.5: Power Series

-ORDER CONVERGENCE FOR FINDING SIMPLE ROOT OF A POLYNOMIAL EQUATION

Recurrence Relations

A New Solution Method for the Finite-Horizon Discrete-Time EOQ Problem

Xhevat Z. Krasniqi and Naim L. Braha

Comparison Study of Series Approximation. and Convergence between Chebyshev. and Legendre Series

Self-normalized deviation inequalities with application to t-statistic

MATH 10550, EXAM 3 SOLUTIONS

arxiv: v1 [cs.sc] 2 Jan 2018

Math 210A Homework 1

Applied Mathematical Sciences, Vol. 9, 2015, no. 3, HIKARI Ltd,

A NEW CLASS OF 2-STEP RATIONAL MULTISTEP METHODS

Discrete-Time Systems, LTI Systems, and Discrete-Time Convolution

6. Kalman filter implementation for linear algebraic equations. Karhunen-Loeve decomposition

An Analysis of a Certain Linear First Order. Partial Differential Equation + f ( x, by Means of Topology

Chapter 6 Infinite Series

Uniform Strict Practical Stability Criteria for Impulsive Functional Differential Equations

Stability of fractional positive nonlinear systems

Lecture 19: Convergence

Generalization of Contraction Principle on G-Metric Spaces

Newton Homotopy Solution for Nonlinear Equations Using Maple14. Abstract

Strong Convergence Theorems According. to a New Iterative Scheme with Errors for. Mapping Nonself I-Asymptotically. Quasi-Nonexpansive Types

CHAPTER 5. Theory and Solution Using Matrix Techniques

PRELIM PROBLEM SOLUTIONS

AN OPEN-PLUS-CLOSED-LOOP APPROACH TO SYNCHRONIZATION OF CHAOTIC AND HYPERCHAOTIC MAPS

Solutions to HW Assignment 1

SYMMETRIC POSITIVE SEMI-DEFINITE SOLUTIONS OF AX = B AND XC = D

ITERATIVE SOLUTION OF TWO MATRIX EQUATIONS. 1. Introduction In this paper, we are concerned with the iterative solution of the matrix equations (1.

TMA4205 Numerical Linear Algebra. The Poisson problem in R 2 : diagonalization methods

Lyapunov Stability Analysis for Feedback Control Design

ON THE LEHMER CONSTANT OF FINITE CYCLIC GROUPS

Asymptotic distribution of products of sums of independent random variables

A Characterization of Compact Operators by Orthogonality

Periodic solutions for a class of second-order Hamiltonian systems of prescribed energy

IN many scientific and engineering applications, one often

Transcription:

Electroic Joural of Liear Algebra Volume 3 Volume 3: (206) Article 44 206 A iterative method to solve a oliear matrix equatio Peg Jigjig Hua Uiversity, Chagsha 40082, PR Chia, jjpeg202@63.com Liao Apig Hua Uiversity, Chagsha 40082, PR Chia Peg Zheyu Guili Uiversity of Electroic Techology, Guili, yuzhep@63.com Follow this ad additioal works at: http://repository.uwyo.edu/ela Part of the Numerical Aalysis ad Computatio Commos Recommeded Citatio Jigjig, Peg; Apig, Liao; ad Zheyu, Peg. (206), "A iterative method to solve a oliear matrix equatio", Electroic Joural of Liear Algebra, Volume 3, pp. 620-632. DOI: https://doi.org/0.300/08-380.295 This Article is brought to you for free ad ope access by Wyomig Scholars Repository. It has bee accepted for iclusio i Electroic Joural of Liear Algebra by a authorized editor of Wyomig Scholars Repository. For more iformatio, please cotact scholcom@uwyo.edu.

AN ITERATIVE METHOD TO SOLVE A NONLINEAR MATRIX EQUATION JINGJING PENG, ANPING LIAO, AND ZHENYUN PENG Abstract. I this paper, a iterative method to solve oe kid of oliear matrix equatio is discussed. For each iitial matrix with some coditios, the matrix sequeces geerated by the iterative method are show to lie i a fixed ope ball. The matrix sequeces geerated by the iterative method are show to coverge to the oly solutio of the oliear matrix equatio i the fixed closed ball. I additio, the error estimate of the approximate solutio i the fixed closed ball, ad a umerical example to illustrate the covergece results are give. Key words. Noliear matrix equatio, Iterative method, Newto s iterative method, Covergece theorem. AMS subject classificatios. 5A24, 5A39, 65F30.. Itroductio. We cosider the oliear matrix equatio (.) X +A X A = Q, where Q is a m m positive defiite matrix, A is a arbitrary real m m matrix, ad is a positive iteger greater tha. The solutio of the matrix equatio (.) is a problem of practical importace. I may physical applicatios, oe must solvethe system of the liear equatio Mx = f, where the positive defiite matrix M arises from a fiite differece approximatio to a elliptic partial differetial operator (e.g. see [2]). As a example, let M = The M = M +diag(i X,0), where ( I A A Q M = ( X A A Q ). ). Received by the editors o March 24, 205. Accepted for publicatio o September 3, 206. Hadlig Editor: Brya L. Shader. Research was supported by Natioal Natural Sciece Foudatio of Chia (o. 2604, o. 3007). Hua Uiversity, Chagsha 40082, PR Chia (jjpeg202@63.com, liaoap@hu.c). Guili Uiversity of Electroic Techology, Guili 54004, PR Chia (yuzhep@63.com). 620

Iterative Methods to Solve a Noliear Matrix Equatio 62 If X is a positive solutio to (.), the the LU decompositio of M is ( ) ( )( ) X A I 0 X A A = Q A X, I 0 X the system Mx = f is trasformed to [ ( )( I 0 X A A X I 0 X ) ( I X 0 + 0 0 ) ] x = f, ad the Sherma-Morriso-Woodbury formula [25] ca be applied to compute the solutio of Mx = f directly, or based o the iterative formula ( )( ) ( ) I 0 X A I X 0 A X x k+ = f x k, k = 0,,2,... I 0 X 0 0 to solve the system Mx = f. Aother example is i cotrol theory, see the refereces give i [9]. Several authors have studied the existece, the rate of covergece as well as the ecessary ad sufficiet coditios of the positive defiite solutios of (.) or similar matrix equatios. For example, the fixed poit iteratio methods to solve (.) have bee ivestigated i [6, 7, 4, 26]. The properties of the positive defiite solutios of (.) ad the perturbatio aalysis of the solutios have bee itroduced i [23, 27]. Some special cases of (.) have bee ivestigated. Fixed poit iteratio methods to solve (.) i cases = 2 ad Q = I have bee studied i [4, 28, 29, 36]. For the case = or cases = ad Q = I, fixed poit iteratio methods are cosidered i [, 4, 5,, 2, 5, 7, 9, 20, 22, 24, 32]. The iversio free variat of the basic fixed poit iteratio methods are cosidered i [0, 6, 38]. Cyclic reductio methods are cosidered i [8, 9, 32]. Properties of the positive defiite solutios of (.) ad perturbatio aalysis of the solutios have bee discussed i [20, 22, 24]. More geeral cases have also bee aalyzed [3, 8, 2, 30, 33, 34, 35, 37]. I this paper, we cosider Newto s iterative method to solve the oliear matrix equatio(.). Weshow,forthegivetheiitialmatrixX 0, thatthematrixsequeces {X k } geerated by the iterative method areicluded i a fixed ope ball B(X 0,). We also show that the {X k } geerated by the iterative method coverges to the oly solutio of the oliear matrix equatio (.) i the fixed closed ball B(X 0,). I additio, the error estimate of the approximate solutio i the fixed closed ball B(X 0,) is preseted, ad a umerical example to illustrate the covergece results is give. We use A F ad A to deote the Frobeius orm ad the spectral orm of the matrix A, respectively. The otatio R m ad H deotes, respectively, the set of all complex m matrices ad all Hermitia matrices. A stads for

622 J. Peg, A. Liao, ad Z. Peg the cojugate traspose of the matrix A. A B stads for the Kroecker product of the matrices A ad B. 2. Covergece of the iterative methods. I this sectio, we discuss Newto s iteratio method to solve the oliear matrix equatio (.), ad show that the iterates coverge to the oly solutio of the oliear matrix equatio (.) i a fixed closed ball. Let F(X) = X +A X A Q, the matrix fuctio F(X) is Frechet-differetiable at the osigular matrix X ad the Frechet-derivate is give by (2.) We have the followig lemmas. F X (E) = E A X i EX +i A. Lemma 2.. Let X be a osigular with X + A 2 <. The the liear operator F X is osigular, ad (F X ) X + A 2. Proof. The iequality F X(E) = E A X i EX +i A E A X i EX +i A E X + A 2 E ad the assumptio X + A 2 < imply that F X (E) = 0 if ad oly if E = 0, that is, the operatorf X is a ijectio. Sice F X is a operatoro the fiite dimesio liear space C, F X is a surjectio. Therefore, F X is osigular, ad (F X) = mi{ F X (H) / H : H 0} X + A 2. Lemma 2.2. For osigular matrices X ad Y, we have (2.2) + F X F Y A 2 ( X +2 i Y i ) X Y.

Iterative Methods to Solve a Noliear Matrix Equatio 623 Proof. For a arbitrary matrix E, we have (F X F Y )(E) = A X i EX +i A = + = = + = + Therefore, (2.2) holds. A X i EX +i A A X i EY +i A A Y i EY +i A A X i EY +i A A Y i EY +i A A X i E(X +i Y +i )A+ A (X i Y i )EY +i A i A X i E[ X ( i k) (X Y )Y k ]A i A [ X k (X Y )Y (i k) ]EY +i A i A X i E[ X ( i k+) (Y X)Y k ]A i A [ X k (Y X)Y (i k) ]EY +i A i A 2 ( X k+ Y k+ ) X Y E + A 2 X k+ Y k+ ) X Y E i ( i = A 2 ( X k+ Y k+ ) X Y E + A 2 ( k= i+ X k+ Y k+ ) X Y E = A 2 ( X k+ Y k+ ) X Y E + = A 2 ( X +2 i Y i ) X Y E.

624 J. Peg, A. Liao, ad Z. Peg The followig lemma from [3, 3] will be useful to prove the ext theoretical results. Lemma 2.3. Let A,B R ad assume that A is ivertible with A α. If A B β ad αβ <, the B is also ivertible, ad B α/( αβ). Applyig Newto s method to solve the oliear matrix equatio (.), we have (2.3) or equivaletly, (2.4) (2.5) X k+ = X k (F X k ) (F(X k )), k = 0,,2,... E k A X i k E kx +i k A = F(X k ), X k+ = X k +E k, k = 0,,2,... Lettig H(X) = (F X 0 ) (F(X)), we see that the Newto iterates for the matrix fuctio H(X) coicide with those of the matrix fuctio F(X) sice (2.6) X k+ X k = (H X k ) (H(X k )) = (F X k ) (F(X k )). (2.7) Lemma 2.4. Let X 0 be a osigular matrix such that holds, ad let The 0 < = (+)( X 0 A 2 + Q X 0 ) X (i) H(X 0 ) + ; 0 + A 2 < ( X B(X 0,) = {X X X 0 < }. (ii) H X H Y X Y for all X,Y B(X 0,); (iii) (H X ) X X 0 / for all X B(X 0,); 0 2 2 ) /(+2) X0 (iv) H(X) H(Y) H Y (X Y) 2 X Y 2 for all X,Y B(X 0,). Proof. (i) By the defiitio of ad the estimate obtaied by Lemmas 2. ad 2.2, we have H(X 0 ) = (F X 0 ) (F(X 0 ))

Iterative Methods to Solve a Noliear Matrix Equatio 625 (F X 0 ) F(X 0 ) X0 + A 2.( X 0 A 2 + Q X 0 ) = X 0 A 2 + Q X 0 X0 = + A 2 +. (ii) Usig the estimates obtaied by Lemma 2. ad 2.2, we have H X H Y = (F X 0 ) F X (F X 0 ) F Y (F X 0 ) F X F Y) + X0 ( X +2 i Y i ) X Y. + A 2. A 2 Sice X X 0 <, Y X 0 < ad X0 < ( X 0 2 2 ) /+2 <, we have by Lemma 2.3 that X ad Y are both osigular, ad X X 0 Y X X. Therefore, 0 H X H Y X 0 0 +2 A 2 0 + A 2.(+) X ( X0. X Y )+2 X0 + A 2.(+) X =. X0 2 2 ( X Y )+2. X 0 X Y. X, 0 0 2 ( X0 A 2 + Q X 0 ) ( X 0 )+2. X Y (iii) Accordig to the defiitio of the matrix fuctio H(X), it is easy to get (H X 0 ) =. The estimate (ii) implies that H X H X 0 X X 0 < for all X B(X 0,). Thus, we have by Lemma 2.3 that (H X ) X X 0 / for all X B(X 0,). (iv) Usig the Newto-Leibiz formula ad (iii), we have H(X) H(Y) H Y (X Y) = (H( t)y +tx H Y )(X Y)dt 0 X Y 0 H ( t)y+tx H Y dt X Y 2 tdt = 2 X Y 2. Lemma 2.5. Assume that the iitial matrix X 0 satisfies (2.7). The the Newto s iterates X k, k 0, for the matrix fuctio H(X), ad hece for the matrix fuctio 0

626 J. Peg, A. Liao, ad Z. Peg F(X), belog to the ope ball B(X 0,), ad furthermore, X k X k hold for all k. 2 k (+), X k X 0 ( 2 k (+) ), (H X k ) 2 k (+), H(X k ) 2 2k (+) 2. Proof. First, let us check that the above estimates hold for k =. Clearly, the poit X = X 0 (H X 0 ) (H(X 0 )) = X 0 H(X 0 ) is well defied sice H X 0 is ivertible. Also ad by (iii) of Lemma 2.4, X X 0 = H(X 0 ) + < ( + ), (H X ) X X 0 / +. By defiitio of X, ad by (iv) of Lemma 2.4 agai, H(X ) = H(X ) H(X 0 ) H X 0 (X X 0 ) 2 X X 0 2 2(+) 2. Assume that the estimates hold for k =,2,...,m for some iteger m. The poit X m+ = X m (H X m ) (H(X m )) is thus well defied sice H X m is ivertible. Moreover, by the iductio hypothesis ad by estimates of (iii) of Lemma 2.4 (for the third ad fourth estimates), X m+ X m (H X m ) H(X m ) 2 m (+), X m+ X 0 X m X 0 + X m+ X m ( 2 m (+) )+ 2 m (+) = ( 2 m (+) ), (H X m+ ) X m+ X 0 / 2m (+), H(X m+ ) = H(X m+ ) H(X m ) H X m (X m+ X m ) 2 X m+ X m 2 2 2m+ (+) 2.

Iterative Methods to Solve a Noliear Matrix Equatio 627 Hece, the estimates also hold for k = m+. Lemma 2.6. Assume that the iitial matrix X 0 satisfies (2.7), the the Newto s iterates X k, k 0, coverges to a zero X of H(X), ad hece of F(X), which belogs to the closed ball B(X 0,). Moreover, for all k 0. X k X 2 k (+) Proof. The estimates X k X k, k, established i Lemma 2.5 2 k (+) clearly imply that {X k } k= is a Cauchy sequece. Sice X k B(X 0,) B(X 0,) ad B(X 0,) is a complete metric space (as a closed subset of the Baach space), there exists X B(X 0,) such that X = lim k X k. Sice H(X k ) fuctio, 2 2k (+) 2, k, by Lemma 2.5, ad H(X) is a cotiuous H( X) = lim k H(X k) = 0. Hece, the poit X is a zero of F(X). Give itegers k ad l, we have, agai by Lemma 2.5, X k X k+l so that, for each k, l+p j=k X j+ X j j=k 2 j (+) = 2 k (+), X k X = lim X k X k+l k 2 k (+). Lemma 2.7. Assume that the iitial matrix X 0 satisfies (2.7), the the zero X of H(X), ad hece of F(X), i the closed ball B(X 0,) uique. Proof. We first show that, if Z B(X 0,) ad H( Z) = 0, the X k Z 2 k (+) for all k 0. Clearly, this is true if k = 0; so assume that this iequality holds for k =,2,...,m, for some iteger m 0. Notig that we ca write X m+ Z = X m (H X m ) (H(X m )) Z = (H X m ) (H( Z) H(X m ) H X m ( Z X m )),

628 J. Peg, A. Liao, ad Z. Peg we ifer from Lemma 2.4 ad 2.5, ad from iductio hypothesis that X m+ Z (H Xm ) 2 Z X m 2 Hece, the iequality X k Z which shows that Z = X. 2 k (+) lim X m Z = X Z = 0, m Lemmas 2.5-2.7 ow imply the followig. 2 m (+). holds for all k. Cosequetly, Theorem 2.8. Assume that the iitial matrix X 0 satisfies (2.7). The the sequece {X k } defied by (2.3) is such that X k B(X 0,) for all k 0 ad coverges to a solutio X B(X 0,) of the oliear matrix equatio (.). Moreover, for all k 0, X k X 2 k (+), ad the poit X is the oly solutio of the oliear matrix equatio (.) i B(X 0,). 3. Numerical examples. I this sectio, we preset a umerical example to illustrate the covergece results of the Newto s method to solve the equatio (.). Our computatioal experimets were performed o a IBM ThikPad of mode T40 with 2.5 GHz ad 3.0 RAM. All tests were performed i the MATLAB 7. with a 32-bit Widows XP operatig system. The followig example we cosider the matrix equatio (.) with = 2, ad use LSQR M algorithm from [33] to solve the subproblem (2.4) of the Newto s method. For coveiece, we let A = A X k,a 2 = A X 2 k,b = X 2 k A,B 2 = X k A,C = F(X k). The the LSQR M algorithm to solve the subproblem (2.4), that is, ca be described as follows. E +A EB +A 2 EB 2 = C LSQR M:(Algorithm for solvig the matrix equatio (2.4) with = 2) () Iitializatio. Set iitial matrix E 0 H. Compute β = C E 0 A E 0 B A 2 E 0 B 2 F, U = (C E 0 A E 0 B A 2 E 0 B 2 )/β, α = U +A T U B T +AT 2 U B T 2 F, V = (U +A T U B T +A T 2U B T 2 )/α, W = V, φ = β, ρ = α. (2) Iteratio. For i =,2,..., util the stoppig criteria have bee met

Iterative Methods to Solve a Noliear Matrix Equatio 629 (a) β i+ = V i +A V i B +A 2 V i B 2 α i U i F,U i+ = (V i +A V i B +A 2 V i B 2 α i U i )/β i+, (b) α i+ = U i+ + A T U i+ B T + A T 2U i+ B T 2 β i+ V i F,V i+ = (U i+ + A T U i+b T +AT 2 U i+b T 2 β i+v i )/α i+. (c) ρ i = ( ρ 2 i + β2 i+ )/2, c i = ρ i /ρ i, s i = β i+ /ρ i, θ i+ = s i α i+, ρ i+ = c i α i+, φ i = c i φi, φ i+ = s i φi. (d) E i = E i +(φ i /ρ i )W i, (e) W i+ = V i+ (θ i+ /ρ i )W i. The stoppig criteria is used as φ k+ α k+ c k 0 0 here. Other stoppig criteria ca also be used, readers ca see [33] for details. Example 3.. Let = 2, ad give matrices A ad Q be as follows. A =.3963.988 0.0292 0.394 0.592.655 2.0658 0.693 0.7079.6776 0.5023.6029 0.687 0.964 0.76.9080 0.4926.3365 0.322 0.005 0.2489 0.6592 0.2735.594 0.6207 0.3987 0.6705.885.7459.328 0.830 0.844 0.7252 0.4953 0.5459.455.5887 0.873.764.0907,.02.255 0.6380.76 0.056.7247 0.7847 0.474 2.087.4742.4575 0.777 0.457 0.4660 0.2668.529.8423 0.9436 0.7286 0.9480 0.533 0.3008 0.889 0.0295 Q =.5272 3.5007.8948 0.5634 0.066.8747 3.6932 0.5252 3.5007 8.543 3.5379.9406 2.788 5.8077 2.0738 2.2306.8948 3.5379 0.409 0.8632 0.7259.3282 0.856 0.0928 0.5634.9406 0.8632 9.453 0.8946 0.4670.9463.0822 0.066 2.788 0.7259 0.8946.5623 3.9067 2.6642.8856..8747 5.8077.3282 0.4670 3.9067 24.522.6249 3.9570 3.6932 2.0738 0.856.9463 2.6642.6249 20.0556 2.3762 0.5252 2.2306 0.0928.0822.8856 3.9570 2.3762 4.896 By direct compute, we kow that Q is a positive defiite matrix, ad the followig estimates hold: = (+) Q A 2 Q + A 2 =.7778< ( Q 2 2 ) /(+2) Q = 3.0523. Hece, give matrices A ad Q satisfy the coditio of Theorem 2.8. Usig Newto s method ad iterate 4 steps, we have X 4 =

630 J. Peg, A. Liao, ad Z. Peg.389 3.5530.8764 0.5452 0.0729.8990 3.6976 0.5792 3.5530 8.4520 3.5457.9983 2.6767 5.7772 2.0723 2.229.8764 3.5457 0.3887 0.8389 0.6987.3395 0.83 0.0960 0.5452.9983 0.8389 9.2778 0.7866 0.5225.9495.49 0.0729 2.6767 0.6987 0.7866.4663 3.8653 2.6763.8484,.8990 5.7772.3395 0.5225 3.8653 24.4734.605 3.9995 3.6976 2.0723 0.83.9495 2.6763.605 9.9757 2.3480 0.5792 2.229 0.0960.49.8484 3.9995 2.3480 4.808 with X A X 2 A Q F = 3.9450 0 2, X 4 Q = 0.342 < =.7778. 4. Coclusios. I this paper, Newto s iterative method to solve the oliear matrix equatio X +A X A = Q is discussed. For the give iitial matrix X 0, the results that the matrix sequeces {X k } geerated by the iterative method are icluded i the fixed ope ball B(X 0,)(Lemma 2.5) ad that the matrix sequeces {X k } geerated by the iterative method coverges to the oly solutio of the matrix equatio X + A X A = Q i the fixed closed ball B(X 0,) (Lemmas 2.6 ad 2.7) are proved. I additio, the error estimate of the approximate solutio i the fixed closed ball B(X 0,) (Lemma 2.6), ad a umerical example to illustrate the covergece results are preseted. The advatage of Newto s iterative method to solve the oliear matrix equatio X + A X A = Q are that the fixed closed ball i which the uique solutio of the matrix equatio icluded ca be determied, the uique solutio i the fixed closed ball ca be obtaied, ad the expressio of the error estimate of the approximate solutio i the fixed closed ball ca be give. May tests show that, if the iitial matrix X 0 = Q, the sequece {X k } defied by (2.3) coverges to the maximal positive defiite solutio X L of the oliear matrixequatio(.). IftheiitialmatrixX 0 = (AQ A ) /, thesequece{x k } defiedby(2.3)covergestothemiimalpositivedefiite solutiox S oftheoliear matrix equatio (.). Here the maximal (miimal) solutio X L (X S ) meas that for everypositivedefiite solutioy ofthe oliearmatrix equatio(.) satisfiesx L Y (Y X S ). That is, X L Y (Y X S ) is a positive defiite matrix. Ufortuately these results ca ot be proved here. The disadvatage is that the rate of covergece is relatively slower tha some existig fixed poit iteratio methods. This is because the ier iteratio, that is, LSQR M algorithm may eed to iterate may times to achievethe required accuracy. Ackowledgemet. The authors thak the aoymous referee for valuable suggestios which helped them to improve this article.

Iterative Methods to Solve a Noliear Matrix Equatio 63 REFERENCES [] W.N. Aderso, T.D. Morley, ad G.E. Trapp. Positive solutios to X = A BX B. Liear Algebra Appl., 34:53-62, 990. [2] B.L. Buzbee, G.H. Golub, ad C.W. Nielso. O direct methods for solvig Poissos equatios. SIAM J. Numer. Aal., 7:627 656, 970. [3] S.H. Du ad J. Hou. Positive defiite solutios of operator equatios X m + A X A = I. Liear Multiliear Algebra, 5: 63-73, 2003. [4] S.M. El-Sayed. A twosided iterative method for computig positive defiite solutios of a oliear matrix equatio. ANZIAM J., 44: 8, 2003. [5] S.M. El-Sayed ad A.M. Al-Dbiba. A ew iversio free iteratio for solvig the equatio X +A X A = Q. J. Comput. Appl. Math., 8:48 56, 2005. [6] S.M. El-Sayed ad A.M. Al-Dbiba. O positive defiite solutios of the oliear matrix equatio X +A X A = I. Appl. Math. Comput., 5:533 54, 2004. [7] S.M. El-Sayed ad M. EI-Alem. Some properties for the existece of a positive defiite solutio of matrix equatio X +A X 2m A = I. Appl. Math. Comput., 28:99-08, 2002. [8] S.M. El-Sayed ad M.G. Petkov. Iterative methods for oliear matrix equatios X + A X α A = I. Liear Algebra Appl., 403:45 52, 2005. [9] J.C. Egwerda. O the existece of a positive defiite solutio of the matrix equatio X + A T X A = I. Liear Algebra Appl., 94:9 08, 993. [0] J.C. Egwerda. O the existece of a positive def iite solutio of the matrix equatio X + A T X A = I. Liear Algebra Appl., 94:9-08, 993. [] A. Ferrate ad B.C. Levy. Hermitia solutio of the matrix X = Q NX N. Liear Algebra Appl., 247:359 373, 996. [2] S. Fital ad C.H. Guo. A ote o the fixed-poit iteratio for the matrix equatios X ± A X A = I. Liear Algebra Appl., 429:2098 22, 2008. [3] Y.H. Gao. Newto s method for the quadratic matrix equatio. Appl. Math. Comput., 82:772 779, 2006. [4] X.X. Guo. O Hermitia positive defiite solutio of oliear matrix equatio. J. Comp. Math., 23:53 526, 2005. [5] G.H. Guo. Covergece rate of a iterative method for a oliear matrix equatio. SIAM J. Matrix Aal. Appl., 23:295 302, 200. [6] G.H. Guo ad P. Lacaster. Iterative solutio of two matrix equatios. Math. Comput., 228:589 603, 999. [7] C.H. Guo, Y.C. Kuo, ad W.W. Li. Complex symmetric stabilizig solutio of the matrix equatio X +A T X A = Q. Liear Algebra Appl., 435:87 92, 20. [8] C.H. Guo ad W.W. Li. Solvig a structured quadratic eigevalure problem by a structuredpreservig doublig algorithm. SIAM J. Matrix Aal. Appl., 32:2784 280, 200. [9] C.H. Guo ad W.W. Li. The matrix equatio X+A T X A = Q ad its applicatio i ao research. SIAM J. Sci. Comput., 32:3020 3038, 200. [20] V.I. Hasaov. Notes o two perturbatio estimates of the extreme solutios to the equatios X ±A X A = Q. Appl. Math. Comput., 26:355 362, 200. [2] V.I. Hasaov. Positive defiite solutios of the matrix equatios X ± A X q A = Q. Liear Algebra Appl., 404:66 82, 2005. [22] V.I. Hasaov ad I.G. Ivaov. O two perturbatio estimates of the extreme solutios to the equatios X ±A X A = Q. Liear Algebra Appl., 43:8 92, 2006. [23] V. Hasaov ad I.G. Ivaov. Solutios ad Perturbatio estimates for the matrix equatios X ±A X A = Q. Appl. Math. Comput., 56:53 525, 2004. [24] V.I. Hasaov, I.G. Ivaov, ad F. Uhlig. Improved perturbatio estimates for the matrix equatios X ±A X A = Q. Liear Algebra Appl., 379:3 35, 2004. [25] A.S. Householder. The Theory of Matrices i Numerical Aalysis. Blaisdell, New York, 964.

632 J. Peg, A. Liao, ad Z. Peg [26] I.G. Ivaov. O positive defiite solutios of the family of matrix equatios X+A X A = Q. J. Comput. Appl. Math., 93:277 30, 2006. [27] I.G. Ivaov. Perturbatio aalysis for solutios of X ± A X A = Q. Liear Algebra Appl., 395:33 33, 2005. [28] I.G. Ivaov ad S.M. El-Sayed. Properties of positive defiite solutios of the equatio X + A X 2 A = I. Liear Algebra Appl., 279:303 36, 998. [29] I.G. Ivaov, V. Hasaov, ad B.V. Michev. O matrix equatios X ±A X 2 A = I. Liear Algebra Appl., 326:27 44, 200. [30] X.G. Liu ad H. Gao. O the positive defiite solutios of the matrix equatios X s ± A T X t A = I. Liear Algebra Appl., 368:83 97, 2003. [3] P.P. Liu ad S.G. Zhag. Newto s method for solvig a class of oliear matrix equatios. J. Comput. Appl. Math., 256:254 267, 204. [32] B. Meii. Efficet computatio of the extreme solutios of X + A X A = Q ad X A X A = Q. Math. Comput., 239:89 204, 200. [33] Z.Y. Peg. A matrix LSQR iterative method to solve matrix equatio AXB = C. It. J. Comput. Math., 87:820 830, 200. [34] Z.Y. Peg ad S.M. El-Sayed. O positive defiite solutio of a oliear matrix equatio. Numer. Liear Algebra Appl., 4:99 3, 2007. [35] Z.Y. Peg, S.M. El-Sayed, ad X.L. Zhag. Iterative methods for the extremal positive defiite solutio of the matrix equatio X+A X α A = Q. J. Comput. Appl. Math., 200:520 527, 2007. [36] M.A. Ramada, T.S. El-Daaf, ad N.M. El-Shazly. Iterative positive defiite solutios of the two oliear matrix equatios X ± A T X 2 A = I. Appl. Math. Comput., 64:89 200, 2005. [37] G.Z. Yao. Theories ad Algorithms for Several Noliear Matrix Equatios. PhD dissertatio, Hua Uiversity, 2004. [38] X. Zha. Computig the extremal positive defiite solutios of a matrix equatio. SIAM J. Sci. Comput., 7:67 74, 996.