Abstract
We consider a preconditioner for linear systems whose coefficient matrices have many nonzero elements and complex nonzero structure. We seek robust preconditioning using sparse complete factorization of the matrix LU=A obtained from drop-thresholding of the original coefficient matrix A. Due to less nonzero elements in A than A, we expect fewer nonzero elements in the preconditioner L and U than the matrix factor obtained from sparse LU factorization of A. We report the performance of our preconditioner for the sparse matrices that arise from our application including comparison with variants of level-of-fill Incomplete Cholesky preconditioning for complex numbers.
