On choice of preconditioner for minimum residual methods for nonsymmetric matrices

Report

Existing convergence bounds for Krylov subspace methods such as GMRES for nonsymmetric linear systems give little mathematical guidance for the choice of preconditioner. Here, we establish a desirable mathematical property of a preconditioner which guarantees that convergence of a minimum residual method will essentially depend only on the eigenvalues of the preconditioned system, as is true in the symmetric case. Our theory covers only a subset of nonsymmetric coefficient matrices but computations indicate that it might be more generally applicable.

Authors: Pestana, J; Wathen, A

• Publisher: SIMAX
• Publication date: 2010-08-01