Robust dropping criteria for F-norm minimization based sparse approximate inverse preconditioning (Q2434936)

The authors study how drop tolerances affect the quality and effectiveness of a preconditioner. They focus on the mathematical theory on robust selection criteria for drop tolerances. To develop a robust preconditioning procedure, the authors analyze the effects of drop tolerances on the non-singularity, quality and effectiveness of preconditioners. They establish some relationships between these factors. They propose adaptive robust selection criteria for drop tolerances that can make the preconditioner as sparse as possible and of comparable quality to those obtained by currently known preconditioners, so that it is possible to lower the cost of setup and application. Numerical experiments are given to show that their criteria work well.