A sparse-sparse iteration for computing a sparse incomplete factorization of the inverse of an SPD matrix (Q1015916)

The authors propose a method for computing a sparse incomplete factorization of the inverse of sparse symmetric positive definite (SPD) matrix. This inverse can be used as a preconditioner for solving symmetric positive definite linear system of equations by conjugate gradient algorithm. The new preconditioner does not need the sparsity pattern of the inverse factor in advance. For preserving sparsity, it is necessary to specify only two parameters. Computational results are given that show that the proposed preconditioner, in conjunction with the symmetric diagonal scaling, is somewhat better than another well known preconditioner.