A method of congruent type for linear systems with conjugate-normal coefficient matrices |
| |
Authors: | M. Ghasemi Kamalvand Kh. D. Ikramov |
| |
Affiliation: | (1) University of Lorestan, Khorramabad, Islamic Republic of Iran;(2) Faculty of Computational Mathematics and Cybernetics, Moscow State University, Moscow, 119992, Russia |
| |
Abstract: | Minimal residual methods, such as MINRES and GMRES, are well-known iterative versions of direct procedures for reducing a matrix to special condensed forms. The method of reduction used in these procedures is a sequence of unitary similarity transformations, while the condensed form is a tridiagonal matrix (MINRES) or a Hessenberg matrix (GMRES). The algorithm CSYM proposed in the 1990s for solving systems with complex symmetric matrices was based on the tridiagonal reduction performed via unitary congruences rather than similarities. In this paper, we construct an extension of this algorithm to the entire class of conjugate-normal matrices. (Complex symmetric matrices are a part of this class.) Numerical results are presented. They show that, on many occasions, the proposed algorithm has a superior convergence rate compared to GMRES. |
| |
Keywords: | conjugate-normal matrices unitary similarity transformations generalized Lanczos process GMRES CSYM |
本文献已被 SpringerLink 等数据库收录! |
|