Dynamic control of a BICGSTAB algorithm (Q1964389)

The aim of this paper is to present a dynamic strategy which allows to detect breakdowns and near-breakdowns and to choose between a look ahead technique and a restart in the BICGSTAB algorithm for solving non-Hermitian sparse linear systems. First, it is recalled how the algorithm can be introduced by means of the formal orthogonal polynomial theory. This allows to develop a simple way to cure breakdowns and near-breakdowns and to locate the right parameters on which the restart strategy will be based. The dynamic strategy proposed is based on the implementation of the discrete stochastic arithmetic. Several examples are given and some conclusions are made.