Differences in the effects of rounding errors in Krylov solvers for symmetric indefinite linear systems (Q2706285)

The authors discuss in what a way and to what extent the well known MINRES (minimal residual), GMRES (generalized minimal residual) and SYMMLQ (symmetric LQ) methods differ in their sensitivity to rounding errors. They show that the method of solution may lead, under certain circumstances, to large additional errors, which are not corrected by continuing the iteration process. A way of analyzing the above mentioned methods is proposed which does not attempt to derive sharper upper bounds but to derive upper bounds for relevant differences between these processes in finite precision arithmetic. The authors' approach allows to answer some practical questions: NEWLINENEWLINENEWLINE(i) When and why is MINRES less accurate than SYMMLQ?NEWLINENEWLINENEWLINE(ii) Is MINRES suspect for ill-conditioned systems?NEWLINENEWLINENEWLINE(iii) Why and when does SYMMLQ converge slower than MINRES and GMRES?NEWLINENEWLINENEWLINE(iv) Why does MINRES sometimes lead to rather large residuals, whereas the error in the approximation is significantly smaller?