Chebyshev acceleration for SOR-like method

From MaRDI portal





The authors present a new iteration method for systems of linear algebraic equations with a symmetric matrix of order \(m+n\) (\(m\geq n\)) consisting of one positive-definite block of order \(m\) and one zero block of order \(n\) in the main diagonal and of two mutually transposed off diagonal blocks of rank \(n\). This new method, called a generalized successive overrelaxation (SOR)-like method using Chebyshev semi-iteration (GSOR SI), is motivated as an improvement of the SOR-like method from \textit{G. H. Golub, X. Wu} and \textit{J.-Y. Yuan} [BIT 41, 71--85 (2001; Zbl 0991.65036)]. Convergence of the GSOR SI is proved and the asymptotic rate of convergence of the GSOR SI is shown to be greater than that of the SOR-like method under some assumptions concerning the preconditioner and the value of relaxation parameter. Two concrete examples illustrate that indeed the GSOR SI is more effective than the SOR-like method.











This page was built for publication: Chebyshev acceleration for SOR-like method

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q5460572)