On some new approximate factorization methods for block tridiagonal matrices suitable for vector and parallel processors (Q1013149)

Stair matrices are used to construct parallel polynomial approximate inverse preconditioners which lead to an efficient parallel solver for sparse block-tridiagonal linear systems. These preconditioners are suitable to maximize parallelism. In adition, some theoretical results concerning these preconditioners are presented and it is further shown how to effectively construct preconditioners for any nonsingular block tridiagonal H-matrices. Besides, a reasonably good performance of these preconditioners is illustrated with numerical experiments driven from second order elliptic partial differential equations and from problems brought in from oil reservoir simulations.