The superiority of a new type (2,2)-step iterative method over the related Chebyshev method (Q1765806)

\((k,l)\)-step iterative methods for solving linear systems of equations have been proposed in the eighties but they can be found only rarely in the literature since then, in particular not in papers related to applications. Since the access to the original literature is difficult, it would have been very helpful for the reader if the authors would have explained what are these methods and what are the basic ideas behind them. Let \(\Omega\) be an ellipse which contains the eigenvalues of the iteration matrix. The paper studies the question if there is a \((2,2)\)-step method which converges faster than the Chebyshev method which is optimal for \(\Omega\). The question is answered positively by constructing an appropriate class of methods which depends on two parameters. Numerical tests on two model problems confirm the analytical results. For a banded Toeplitz matrix, the (2,2)-step method needs even less iterations than popular iterative schemes like GMRES, QMR and BICGSTAB.