RIA: An Unsymmetric-Pattern Multifrontal Method for ParallelSparse LU Factorization

本研究的目的是开发一种并行多正面方法，用于稀疏非对称矩阵的LU分解，该方法充分利用密集矩阵核，同时保持纯非对称模式。所得到的算法将在具有非对称模式的大型稀疏矩阵上具有高性能的潜力。该算法将适用于许多需要求解稀疏非对称线性方程组数值解的科学和工程问题。

The objective of the proposed research is to develop a parallel multifrontal method for LU factorization of sparse unsymmetric matrices that takes full advantage of dense matrix kernels while maintaining a purely unsymmetric pattern. The resulting algorithm will have the potential for high performance on large, sparse matrices with unsymmetric pattern. The algorithm will be applicable to the many problems in science and engineering that require the numerical solution of sparse unsymmetric systems of linear equations.