成组Broyden修正矩阵的紧凑形式与成组记忆修正算法

1 引言 成组型线性方程组 其中,p是适中的数值,由于其有相当的实际应用背景,人们一直在研究有效的数值方法,特别是近年来,实际问题中归结出来的成组型方程组,其规模越来越大,又具有稀疏结构,因而使用迭代法是一种有效的途径,目前使用比较多的是Krylov子空间方法中的Lanczos方法,CG方法,GMRES方法等等。这种成组型算法的建立,其基本出发点是使算法具有较少的计算量和存储量,具体体现在: 1)成组型算法在应用于问题(1.1)的求解时,也具有有限终止性性质,而其终止步数一般要比单个型算法的步数减少了户倍,由于成组型算法每迭代一步的计算量基本上等同于单个型算法使用户次的计算量,如此,算法的计算量会有明显的改善。 2)当A存储在二级(secondary)内存时,在迭代计算时需要不断地进行存取交换,由于成组型算法的迭代步数减少了户倍,如此,用在这种交换的时间也要减少户倍,相当有效。 3)由于在成组型算法中,出现的多是AX的形式,其中,故成组型算法便于计算并行化。 4)即使用于求解单个方程组,当A的少数几个极端特征值分离甚远时,这种成组型算法也有可能改善其收敛速度,如成组型的CG方法。 目前,这种成组型算法已体现出很大的实用计算价值,然而其进一步的理论分析还有待深入研究。

REPRESENTATION OF MULTIPLE BROYDEN'S UPDATING MATRIX AND MULTIPLE VERSION OF BROYDEN'S METHOD WITH LIMITED STORAGE

Broyden's method for solving systems of nonlinear equations (including linear equations) has achieved a grate deal of computational success. In 1994, O'Leary extended the Broyden's method to the multiple version for solving multiple systems of nonlinear equations. In this paper, we derive a compact representation of the matrices generated by the multiple version of Broyden's update. This representa-tion allow us to efficiently implement limited memory methods for solving the multi-ple nonlinear systems. The method relieves the requirement for the storage counts and has the savings in the operation counts. The numerical experiments of the multi-ple linear systems show that the method works efficiently, and that the more limited information the method has, the less iterative steps for convergence are required. The total operation counts and the storage counts for the above numerical experi-ments are given.