学术报告

题目：Modified Parallel Multisplitting Iterative

Methods for Linear Systems

报告人：温瑞萍

时间：10 月23 日（周四） 上午10:00-11:00

地点：致远楼102 室

温瑞萍教授介绍：

温瑞萍，太原师范学院数学系教授，全国计算数学学会理事。2009 年7 月以来主持和

参与国家级课题1 项、省级课题4 项；在国内外重要学术期刊发表论文9 篇，其中被三大

检索系统SCI、EI、ISTP 收录7 篇，国家级论文1 篇；成为SCI 核心期刊《Applied Mathematics

Letters》的特邀审稿人。

摘要：

In this talk, we not only want to decreases the difficulty of constricting the

multisplitting of the coefficient matrix, but also release the constrains to the

weighting matrices. We present two optimization models to modify parallel

multisplitting iteration methods for solving positive definite (Hermitian or

non-Hermitian) linear systems.

We proved that the multisplitting parallel iteration methods are convergent under

weak conditions:

(1) the weighting matrices are not necessarily nonnegative and be given in advance;

(2) only one of all the splitting is required to be convergent. In view of computation,

though the great time is taken in the solving the optimization models at each step,

the total iteration numbers will be decreased largely. On the other hand, the

weighting matrices are chosen few nonzero variables and many zero entries, so that

the optimization models are solved conveniently. Some numerical experiments show

that these methods are effective.