Inexact Matrix-Vector Products in Krylov Methods for Solving Linear Systems: A Relaxation Strategy

Authors:
Amina Bouras;Valérie Frayssé
Affiliations:
-;-
Venue:
SIAM Journal on Matrix Analysis and Applications
Year:
2005

Citing 0
Cited 7

A FETI-like domain decomposition method for coupling finite elements and boundary elements in large-size problems of acoustic scattering

Computers and Structures
Linear multifrequency-grey acceleration recast for preconditioned Krylov iterations

Journal of Computational Physics
Algorithm 881: A Set of Flexible GMRES Routines for Real and Complex Arithmetics on High-Performance Computers

ACM Transactions on Mathematical Software (TOMS)
Constraint Schur complement preconditioners for nonsymmetric saddle point problems

Applied Numerical Mathematics
An Efficient Newton-Krylov Implementation of the Constrained Runs Scheme for Initializing on a Slow Manifold

Journal of Scientific Computing
Block triangular Schur complement preconditioners for saddle point problems and application to the Oseen equations

Applied Numerical Mathematics
Evaluation of the performance of inexact GMRES

Journal of Computational and Applied Mathematics

Quantified Score

Hi-index	0.01

Visualization

Abstract

Embedded iterative linear solvers are being used more and more often in linear algebra. An important issue is how to tune the level of accuracy of the inner solver to guarantee the convergence of the outer solver at the best global cost. As a first step towards the challenging goal of controlling embedded linear solvers, inexact Krylov methods are used as a model of inner-outer iterations with external Krylov scheme. This paper experimentally shows that Krylov methods for solving linear systems can still perform very well in the presence of carefully monitored inexact matrix-vector products. This surprising behavior of inexact Krylov methods, as opposed to Newton-like methods, is investigated in detail, and potentially important applications are mentioned. A new relaxation strategy for the inner accuracy is proposed for Krylov methods with inexact matrix-vector products; its efficiency is supported by a wide range of numerical experiments on different algorithms and contrasted against other potential approaches.