Refining an approximate inverse (Q1591188)

The authors propose algorithmic techniques for improving the performance of stabilized sparse approximate inverse (SAINV) preconditioners for general regular matrices \(A\). The SAINV preconditioners are then used to accelerate some Krylov space method for solving linear systems of the form \(Ax=b\). The algorithmic techniques (symbolic factorization, reordering etc.) are basically borrowed from the direct factorization algorithms. The authors also extend this approach to the block case.