Effective preconditioning through minimum degree ordering interleaved with incomplete factorization

Abstract

In this paper, we study a kind of effective preconditioning technique, which interleaves the incomplete Cholesky (IC) factorization with an approximate minimum degree ordering. An IC factorization algorithm derived from IKJ-version Gaussian elimination is proposed and some details on implementation are presented. Then we discuss the ways to compute the degrees of the unnumbered nodes exactly and approximately using the concept of element absorbing. When used in conjunction with conjugate gradient algorithm, the new preconditioners usually lead to fast convergence. The numerical experiments show that the interleaving of symbolic ordering and numerical IC factorization will generate better preconditioners than those generated by the IC factorization without ordering or with purely symbolic ordering ahead of the factorization.