An Efficient Solver for Sparse Linear Systems Based on Rank-Structured Cholesky Factorization

Direct factorization methods for the solution of large, sparse linear systems that arise from PDE discretizations are robust, but typically show poor time and memory scalability for large systems. In this paper, we describe an efficient sparse, rank-structured Cholesky algorithm for solution of the positive definite linear system Ax = b when A comes from a discretized partial-differential equation. Our approach combines the efficient memory access patterns of conventional supernodal Cholesky algorithms with the memory efficiency of rank-structured direct solvers. For several test problems arising from PDE discretizations, our method takes less memory than standard sparse Cholesky solvers and less wall-clock time than standard preconditioned iterations.