Crout versions of ilu factorization with pivoting for sparse symmetric matrices

The Crout variant of ILU preconditioner (ILUC) developed recently has been shown to be generally advantageous over ILU with Threshold (ILUT), a conventional row-based ILU preconditioner. This paper explores pivoting strategies for sparse symmetric matrices to improve the robustness of ILUC. We integrate two symmetry-preserving pivoting strategies, the diagonal pivoting and the Bunch-Kanfman pivoting, into ILUC without significant overheads. The performances of the pivoting methods are compared with ILUC and ILUTP ([20]) on a set of problems, including a few arising from saddle-point (KKT) problems.