Analytical bounds for block approximate factorization methods (Q1208299)

The author derives simple analytical upper bounds on the spectral condition number associated with the preconditioning of large sparse linear systems by the so-called modified block incomplete factorizations. By the application of graph notions to block matrices, he succeeds to get the upper bounds for the preconditioning in the case of a Stieltjes matrix. The results prove to handle a large class of discretization of second-order elliptic partial differential equations with mixed boundary conditions, and support that sparse approximate block-matrix factorizations provide powerful preconditioners.