浅析对称M矩阵的不完全LU分解算法

Incomplete LU-decomposition of Symmetric M-matrix

文章提出了一种针对M矩阵(若A非奇异,A-1≥0,且A的非对角元非正,则称A为M矩阵)的正则分解方法。如果矩阵是对称的,那么这种分解方法能够得到很好的分解效果,而且如果将它与共轭梯度法相结合就能得到一种更快的迭代算法。在文章中证明了这种不完全LU分解算法的稳定性和收敛性。最后,将这种方法应用于几种不同的矩阵。数值实验结果表明,对于高阶稀疏矩阵,这种方法收敛的最快,效果最好。

A particular class of regular splitting of not necessarily symmetric M-matrices is proposed. If the matrix is symmetric, this splitting is combined with the conjugate-gradient method to provide a fast iterative solution algorithm. Comparisons have been made with other well-known methods. In all test problems the new combination was faster than the other methods.