Minimum residual Hermitian and skew-Hermitian splitting iteration method for non-Hermitian positive definite linear systems (Q1731615): Difference between revisions

The classical Hermitian-skew-Hermitian Splitting (HSS) iteration method solves a system $Ax=b$ with $A\in\mathbb{C}^{n\times n}$ a positive definite matrix by a two-step iteration: $x^{(k+1/2)}=x^{(k)}+\beta_k\delta^{(k)}$ and $x^{(k+1)}=x^{(k+1/2)}+\gamma_k \delta^{(k+1/2)}$ with $\beta_k=\delta_k=1$. The $\delta^{(k)}$ and $\delta^{(k+1/2)}$ are solutions of systems involving both the Hermitian ($H=(A+A^*)/2$) and skew-Hermitian matrices ($S=(A-A^*)/2$) parts of the matrix $A$ and a positive parameter $\alpha$. In this paper, the authors add parameters $\beta_k$ and $\gamma_k$ to minimize the residuals $r^{(k+1/2)}=b-Ax^{(k+1/2)}$ and $r^{(k+1)}=b-Ax^{(k+1)}$ and call the procedure ``minimal residual HSS method''. They show that their parameters give an overall minimum for $\|r^{(k+1)}\|$ and that if the method is used as a preconditioner, it is not a classical Krylov subspace method. An estimate, although not trivial to compute, is given for the (linear) rate of convergence. Several numerical examples show the performance and how it compares with other iterative methods.