There are several preconditioning methods for large sparse systems of linear equations. One of the most robust parallel approaches is the sparse approximate inverse (SPAI) preconditioner, which is based on Frobenius norm minimization. Our objective is to extend SPAI in order to satisfy certain additional constraints, the so-called probing constraints. The resulting preconditioner should act in an optimal way on these probing subspaces. The resulting method is the modified sparse approximate inverse (MSPAI) preconditioner. Furthermore, it can be regarded as a generalization of both classical probing approaches and the class of modified incomplete factorizations, which are limited to the preservation of very simple probing constraints. MSPAI does not suffer from any restriction with respect to probing information. Moreover, it is still inherently parallel. Another topic is the efficient implementation of MSPAI. Many numerical examples such as the discretization of partial differential equations prove MSPAI's effectiveness.
«
There are several preconditioning methods for large sparse systems of linear equations. One of the most robust parallel approaches is the sparse approximate inverse (SPAI) preconditioner, which is based on Frobenius norm minimization. Our objective is to extend SPAI in order to satisfy certain additional constraints, the so-called probing constraints. The resulting preconditioner should act in an optimal way on these probing subspaces. The resulting method is the modified sparse approximate inve...
»