Inverse Iteration on Defective Matrices

نویسنده

  • Nai-fu Chen
چکیده

Very often, inverse iteration is used with shifts to accelerate convergence to an eigenvector. In this paper, it is shown that, if the eigenvalue belongs to a nonlinear elementary divisor, the vector sequences may diverge even when the shift sequences converge to the eigenvalue. The local behavior is discussed through a 2 X 2 example, and a sufficient condition for the convergence of the vector sequence is given. Introduction. If an accurate approximation a to an eigenvalue X of a matrix B is available, then inverse iteration is an attractive technique for computing the associated eigenvector. We choose an arbitrary unit vector vQ and a fixed shift a. Then for / = 1, 2, ... we solve (,) (BofyWj = Vj_v Vj = wy/Uwyll, where || • || is the users' preferred vector norm. Semisimple case. If X is a simple eigenvalue with unit eigenvector x, if a is close enough to X, and if v0 is not an unfortunate choice, then the vector sequence {u} converges linearly to x and the convergence factor is very favorable. This well-known result holds also for multiple eigenvalue X provided that: (i) the dimension of X's eigenspace is equal to X's algebraic multiplicity (i.e. linear elementary divisors), (ii) the spectral projection of v0 into X's eigenspace is not zero (i.e. the starting vector is not deficient in x). If a is known to equal X to within working precision of the computer, then only one or two steps of the iteration are necessary. Defective Case. Wilkinson [3] and Varah [2] pointed out that the situation is not so nice if X has generalized eigenvectors of grade higher than one, i.e., when X belongs to a nonlinear elementary divisor. In exact arithmetic the iteration converges not linearly, but harmonically like 1 // as /' —► °°. Even worse is the fact that except for very special choices of u0, the vectors v2 and v3 will be poorer approximations than Uj ! Variable Shifts. Inverse iteration can also be used with variable shifts. If the sequence of shifts {a} converges to X as / —► °°, then the vector sequence generated Received January 20, 1976; revised June 21, 1976. AMS (MOS) subject classifications (1970). Primary 65F15; Secondary 15A18.

برای دانلود متن کامل این مقاله و بیش از 32 میلیون مقاله دیگر ابتدا ثبت نام کنید

ثبت نام

اگر عضو سایت هستید لطفا وارد حساب کاربری خود شوید

منابع مشابه

A STABLE COUPLED NEWTON'S ITERATION FOR THE MATRIX INVERSE $P$-TH ROOT

The computation of the inverse roots of matrices arises in evaluating non-symmetriceigenvalue problems, solving nonlinear matrix equations, computing some matrixfunctions, control theory and several other areas of applications. It is possible toapproximate the matrix inverse pth roots by exploiting a specialized version of New-ton's method, but previous researchers have mentioned that some iter...

متن کامل

A Parallel Unsymmetric Inverse Iteration Solver

We describe a matrix multiply based block unsymmetric inverse iteration solver for upper Hessenberg matrices. Our kernel is robust in that it prevents overrow by scaling. It uses new techniques to ensure performance is not sacriiced when scaling is not necessary. Finally, we give results on a parallel implementation on an Intel Paragon TM supercomputer.

متن کامل

A Generalized Inverse Iteration for Computing Simple Eigenvalues of Nonsymmetric Matrices

Technische Universit at Dresden Herausgeber: Der Rektor A Generalized Inverse Iteration for Computing Simple Eigenvalues of Nonsymmetric Matrices Hubert Schwetlick and Ralf L osche IOKOMO-07-97 December 1997 Preprint-Reihe IOKOMO der DFG-Forschergruppe Identi kation und Optimierung komplexer Modelle auf der Basis analytischer Sensitivitatsberechnungen an der Technischen Universitat Dresden ...

متن کامل

Stair Matrices and Their Generalizations with Applications to Iterative Methods Ii: Iteration Arithmetic and Preconditionings

Iteration arithmetic is formally introduced based on iteration multiplication and αaddition which is a special multisplitting. This part focuses on construction of convergent splittings and approximate inverses for Hermitian positive definite matrices by applying stair matrices, their generalizations and iteration arithmetic. Analysis of the splittings and the approximate inverses is also prese...

متن کامل

An iterative method for the Hermitian-generalized Hamiltonian solutions to the inverse problem AX=B with a submatrix constraint

In this paper, an iterative method is proposed for solving the matrix inverse problem $AX=B$ for Hermitian-generalized Hamiltonian matrices with a submatrix constraint. By this iterative method, for any initial matrix $A_0$, a solution $A^*$ can be obtained in finite iteration steps in the absence of roundoff errors, and the solution with least norm can be obtained by choosing a special kind of...

متن کامل

ذخیره در منابع من


  با ذخیره ی این منبع در منابع من، دسترسی به آن را برای استفاده های بعدی آسان تر کنید

برای دانلود متن کامل این مقاله و بیش از 32 میلیون مقاله دیگر ابتدا ثبت نام کنید

ثبت نام

اگر عضو سایت هستید لطفا وارد حساب کاربری خود شوید

عنوان ژورنال:

دوره   شماره 

صفحات  -

تاریخ انتشار 2010