<i>s</i>-Step Iterative Methods for (Non)Symmetric (In)Definite Linear Systems (Q3984475)

This paper reviews the following conjugate gradient-like methods, in order to solve linear systems, with a non necessarily symmetric positive definite matrix: conjugate residuals (CR), generalized CR (GRC), GRC(k), Orthomin(k), \(s\)-step minimal residual (\(s\)-MR), \(s\)-GRC, \(s\)- Orthomin(k), \(s\)-step minimal residual (\(s\)-MR), \(s\)-Orthomin(k) and GMRES\((s)\). The \(s\)-versions deal in the \(i\)-th iteration with the Krylov affine subspaces spanned by the first \(s\) terms in the Krylov sequence generated by the \(i\)-th residual; thus 1-MR is the classical steepest descent method; \(s\)-MR is algebraically equivalent to GMRES\((s)\) as defined by \textit{Y. Saad} and \textit{M. Schultz} [SIAM J. Sci. Stat. Comput. 7, 856-869 (1986; Zbl 0599.65018)], when considered as a direct method, i.e. in error free computations. The author proves several new convergence results for some of the methods already mentioned and also gives a detailed description of their operational cost. He also points out that the \(s\)-versions are expected to have better performance than the standard ones on parallel computers, which is to be examined in a forthcoming paper.