The superlinear convergence behaviour of GMRES

From MaRDI portal





The authors analyse the convergence behaviour of the generalized minimal residual (GMRES) algorithm [cf. \textit{Y. Saad} and \textit{M. H. Schultz}, SIAM J. Sci. Stat. Comput. 7, 856-869 (1986; Zbl 0599.65018)]. Their result shows that a reduction in the speed of convergence in a certain phase of the iteration process can be expected as soon as certain eigenvalues of the given matrix are sufficiently well approximated by Ritz values. The main results of this paper are concentrated in three theorems, where the norm of the residual vector of the original and by the authors modified GMRES are compared. The conclusion, according to the authors, is that as soon as eigenvalues of the original operator are sufficiently well approximated by Ritz values, GMRES from then on converges at least as fast as for a related system in which these eigenvalues (and their eigenvector components) are missing.




Cited in
(73)








This page was built for publication: The superlinear convergence behaviour of GMRES

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q1318439)