Optimum parameter for the SOR-like method for augmented systems

From MaRDI portal





Saddle point problems lead to linear equations \[ \begin{pmatrix} A & B \\ B^T & 0 \end{pmatrix} \begin{pmatrix} x \\ y\end{pmatrix} = \begin{pmatrix} b\\ c \end{pmatrix} \] Recently, \textit{G. H. Golub, X. Wu} and \textit{J.-Y. Yuan} [BIT 41, No.~1, 71--85 (2001; Zbl 0991.65036)] introduced the successive overrelaxation (SOR)-like methods with the inverse of the matrix \[ \begin{pmatrix} A & 0 \\ \omega B^T & Q \end{pmatrix} \] as an approximative inverse. Here, \(-Q\) acts as an approximation of the Schur complement \(B^TA^{-1}B\) and \(\omega\) is a relaxation factor. The optimal \(\omega\) is given in terms of the maximal and minimal eigenvalues of \(-QB^TA^{-1}B\). An acceleration of the iteration via Chebyshev polynomials is discussed.











This page was built for publication: Optimum parameter for the SOR-like method for augmented systems

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q3154808)