Gene Golub Symposium
Preconditioning with Selective Sparse Approximate Inversion
Padma Raghavan and Keita Teranishi, Pennsylvania State University
We use the recursive separability of the graphs of sparse matrices to develop a parallel preconditioner that combines incomplete factorization and sparse approximate inversion. We will discuss our parallel algorithm with an analysis of its computational complexity on model problems. We will also present results on observed parallel speedups and the quality of preconditioning.
