【作者】 陈高洁；

【导师】 曾金平；

【作者基本信息】 湖南大学 ， 应用数学， 2006， 博士

【摘要】 区域分解法是建立在给定的计算区域被划分为几个重叠或非重叠的子区域的假设上的一种算法。Schwarz交替法无疑是最早的区域分解法之一。随着并行计算机的出现，区域分解法以其缩小计算规模和高度并行的优点成为设计并行算法最重要的一种方式。本文讨论椭圆型变分问题，包括椭圆算子对应的变分不等式与互补问题，以及偏微分方程的区域分解法。 互补问题是一类典型的变分不等式，它广泛用于阐述和研究物理学、力学、经济学、运筹学、最优控制等数学模型以及交通运输中出现的各种平衡模型，其数值解法的研究发展迅速。目前求解互补问题的迭代算法有很多，区域分解法是其中的研究热点之一。对于对称线性互补问题， Ax+6≥0，x≥0，x~T(Ax+b)=0，其中，A是给定的N×N实对称矩阵，b是N×1向量，在已有的研究成果中，大多数要求其中的系数矩阵A对称正定或者为M阵等。本文中讨论了当其中的系数矩阵为对称双正阵时，区域分解法(包括乘性Schwarz算法、非重叠加性Schwarz算法和重叠加性Schwarz算法)的收敛性质。证明了由这些算法产生的迭代序列的聚点是原互补问题的解。数值算例表明，算法的收敛速度快，体现其优越性。 用区域分解法求解偏微分方程于上世纪八十年代蓬勃兴起，并越来越受到人们的重视。它分为重叠型和非重叠型。以Robin条件为界面条件的重叠型区域分解法也被称为广义Schwarz算法，其区别于古典的Schwarz算法的特点是在子区域之间的界面上采用Dirichlet条件和Neumann条件相结合的Robin条件来代替原来的单纯的Dirichlet条件。本文中分析了一种广义加性Schwarz算法求解Dirichlet边值的偏微分方程问题的收敛率。给出了一维和二维问题的算法收敛率的定量分析，并以相应的数值算例说明参数及重叠区域的大小与算法收敛率之间的关系。数值算例表明，适当的Robin参数和减小重叠区域的大小会提高算法的收敛率。这种算法也可以被用于非重叠型的区域分解。非重叠型区域分解方面的研究目前相关结论不是很多。在大多数文献中，讨论的主要是矩形或带状区域。本文中讨论了非规则的区域-L型区域上的Poisson方程的一种加性非重叠区域分解法。而且，在该区域分解法中也采用了Robin型界面传输条件。证明了该算法在连续情形下的收敛性，并讨论了离散后算法的收敛速度与Robin型界面传输条件中的Robin参数之间的关系。数值算例说明，适当的Robin参数的选取会大大加快该算法的收敛速度。更多还原

【Abstract】 Domain decomposition is one of the most significant way for devising parallel algorithms that can benefit strongly from multiprocessor computers. Domain decomposition methods are generally based on the assumption that the given computational domain is partitioned into subdomains, which may or may not overlap. The alternative Schwarz method is undoubtedly the earliest example of domain decomposition methods. In this dissertation, we consider domain decomposition to solve elliptic variational problems, including variational inequalities, complementarity problem and partial differential equations.Complementarity problems are used to interpret and study the mathematical models of physics, mechanics, economics and optimal control, and various of equilibrium models that arise from traffic conveyance. The methods for the numerical solution of the complementarity problems are developed rapidly. At the present time, there are many iterative methods for solving complementarity problems. Domain decomposition method is a kind of pop iterative method studied by many researchers. For symmetric linear complementarity problems, most results are based on the assumption that the coefficient matrices are symmetric and positive definite or M matrices. In this dissertation, domain decomposition methods for the case of the coefficient matrices are symmetric and copositive are proposed. Convergence of the methods is established. And numerical results are present to show the efficiency of the methods.Domain decomposition methods for partial differential equations were developed in 1980s. From then on, the methods are recognized by more and more researchers. The Schwarz algorithms, which use Robin transmission conditions on the inner boundaries of the subdomains, is also called generalized Schwarz algorithms. Compared with the classical Schwarz algorithm, the generalized Schwarz algorithms replace transmission conditions on the interface between subdomains by the Robin conditions with parameters. In this dissertation, the convergence rate of a generalized additive Schwarz algorithm for solving boundary value problems of differential equations is studied. A quantitative analysis of the convergence rate is given for the one and two dimensional model Dirichlet problems. It shows that small overlapping is preferred for the generalized additive Schwarz algorithm. Some numerical tests also show that a greater acceleration of the algorithm can be obtained by choosing the parameter suitably. The alternative Schwarz method更多还原