Practical convergent splittings and acceleration methods for non-Hermitian positive definite linear systems (Q360463)

Let \(A\) be a (large and sparse) non-Hermitian matrix, then for solving the linear system \(Ax=b\) by a Chebyshev semi-iterative scheme, this matrix is split as \(A=M-N\) with \(M\) or \(N\) (positive definite) Hermitian. This leads to iterations of the form \(x_{k+1}=H(x_k;\alpha_k)\) with \(H(x,\alpha)=\alpha(M^{-1}Nx+M^{-1}b-x)+x\). In this paper, the analysis is performed for three such splittings. Necessary and sufficient conditions for convergence are given depending on the properties of the matrices \(M\) and \(N\) as well as lower bounds for the rate of convergence under some additional conditions involving also the choice of the scalar parameters \(\alpha_k\). This \(\alpha_k\) controls the stepsize by minimizing some weighted residual norm.