Sparse approximate inverse and multilevel block ILU preconditioning techniques for general sparse matrices (Q1590821)

A preconditioning technique for general sparse matrices, which combines a multilevel block ILU (BILUM) with the sparse approximate inverse techniques, is presented. The two basic methods are introduced and the resulting preconditioner is designed to have the ability of controlling sparsity and increased parallelism compared to the standard BILUM preconditioner. Extensive numerical experiments with a discretized convection-diffusion problem and five additional, well-known matrices from different sparse matrix collections illustrate the behaviour of the proposed preconditioner, but also the realization that it is unlikely that there exists a general purpose preconditioner being superior for all types of problems.