On hybrid acceleration of a linear stationary iterative process (Q917218)

A hybrid iterative process for solving large unstructured linear systems with symmetric positive semidefinite matrix is proposed. The method consists in a combination of Gauss-Seidel iterations with SOR iterations having the relaxation parameter equal to two. The convergence is proved assuming the consistency of the system. There is a conjecture (supported by numerical experiments) that the method accelerates substantially the ordinary Gauss-Seidel iterations. Applications to linear systems with square matrices are also considered.