An iterative solution method for solving sparse nonsymmetric linear systems (Q1079918)

\textit{O. Widlund} [SIAM J. Numer. Anal. 15, 801-812 (1978; Zbl 0398.65030)] proposed a Lanczos method for nonsymmetric sparse linear systems which is based on the algorithm of \textit{P. Concus} and \textit{G. Golub} [Lect. Notes Econ. math. Syst. 134, 56-65 (1976; Zbl 0344.65020)]. \textit{D. Young} and \textit{K. Jea} [Linear Algebra Appl. 34, 159-194 (1980; Zbl 0463.65025)] have studied the acceleration of Widlunds method, which yields a very effective method for the nonsymmetrizable cases, but it works only if the coefficient matrix A is positive real. The author generalizes Widlund's method without this assumption. This generalized Widlund (GW) method converges, if A is nearly symmetric and has workable splits. This property is a generalization of ''positive real''. Such matrices arise in elliptic boundary value problems, where the first order partial derivative terms have different signs. Some numerical examples are given comparing the GW method with usual relaxation algorithms.