A Supernodal Approach To Incomplete LU Factorization

In the thesis,we present a new incomplete LU method to construct a preconditioner for solving sparse linear systems with iterative methods.The new algorithm is derived from the sparse direct solver SuperLU.We propose a new adaptive dropping strategy similar to Y.Saad's ILUTP approach,but ours is heuristic and has a better numerical behavior than traditional ILUTP.We also analyze the breakdown caused by zero pivots,and handle them with perturbation to boost the robustness the algorithm. We present experiments demonstrating that our new method is competitive with the other ILU approaches and is well suited for today's high performace architectures.