The application of Leja points to Richardson iteration and polynomial preconditioning (Q808157)

Let \(Ax=b\) be a system of algebraic equations with a non-Hermitian matrix A, and let K be a compact set in \({\mathbb{C}}\) which contains the spectrum of A. The paper deals with the choice of parameters \(\delta_ j\), such that the (nonstationary) Richardson iteration \(x_{j+1}=x_ j-\delta_ j(Ax_ j-b),\) \(j=0,1,2,..\). converges and is stable. It is shown that the inverses of the Leja points are good choices. The Leja points \(z_ 1,z_ 2,...\in K\) are iteratively defined. Let \(\omega\) (z) be a positive weight function. If \(z_ 0,z_ 1,...,z_{j- 1}\) are already given, then \(z_ j=z\) is chosen such that the expression \(\prod_{k<j}| z-z_ k| \cdot \omega (z)\) becomes maximal. By using properties of the Greens function it is shown that asymptotic optimal convergence and stability is achieved. The numerical results are depicted in 7 figures.