The semi-convergence properties of MHSS method for a class of complex nonsymmetric singular linear systems (Q403082)

The paper is focused on solving singular linear systems \(Ax=b\) with a large sparse non-Hermitian matrix with positive definite Hermitian part \(\frac12(A+A^*)\). The authors analyze the semi-convergence of the modified Hermitian skew-Hermitian splitting (MHSS) method, which is based on splitting of \(A\) to a real and an imaginary part, i.e., \(A=W+iT\). The spectrum of the iteration matrix that is involved in the iterative process is analyzed. The authors then derive the quasi-optimal iteration parameter that minimizes the upper bound obtained for the semi-convergence. The MHSS is then used as a preconditioner for the generalized minimal residual (GMRES) method. Numerical experiments compare the computational costs and the accuracy of the standard Hermitian skew-Hermitian splitting (HSS) and MHSS. Further, the computational cost and accuracy of different variants of the GMRES method with incomplete LU (ILU), HSS, and MHSS precoditioners are compared.