On the use of rational iterative methods for solving large sparse linear systems (Q1195363)

One is mainly interested in linear systems arising from the solution of elliptic partial differential equations by finite difference methods. The object of the work is to develop iterative procedures which are not only rapidly convergent but which are also suited for use on vector and parallel supercomputers. These techniques should be applicable to general problems with convergence rate substantially less than \(O(h^{-3})\). To this end, for solving the linear system \(Au=b\), the iteration method is selected in the form \(u(\text{new})=u(\text{old})+R(A)(b-Au(\text{old}))\) which provides a connection with the time-dependent problem \(du(t)/dt=-Au(t)+b\). The sections deal respectively with time-dependent problems, rational iteration, multilevel implementation of the rational iteration and some numerical experiments (in addition to the usual conclusion).