抛物型偏微分方程地几类新地并行算法与其收敛性分析.docx

山东大学硕士学位论文 tije corresponding projections掣。：V‘一V?[sing tile parallel theorem and sub— space correction method given by J Xu(㈧)we can get an algorithm described as “，”“。=w““+∑20葺 Second．册inn-odutx、the coarse space VⅣand define the projection璐：V_÷ W()n cv(、rj。subspace．州get矗。(?=1 2，?．。，)by solving the Ailti 2^as (teseribed in this paper．Also wP(：a11 gel|the solution of the coarse fini．te element sDace．we wrire this solution as^0 Then we make the average面=击E识．From this way．we carl make up Ollr second algorithm —e also suppose： Condition(．{1 For a11．}’。∈豆there exists ail open domain DT and i∈ {1．2． ·．AT)such that一∈D?and D，n Q岳Q Condition fBl．Fm an?7。∈ 豆there are at mosl-A0 subdomain that z Call belong tO Using the abstract paralh、1 algoril hm tO solve(】uI parabolic con＼’ection—diffusion problem we(。aIl gel the following Corresponding to these 61l ro algorithms．we get}ll’o different convergence re— suits as follou’iIlg： Theorem 3．1 Suppose that the solution of r3．1)is s“fficiently smooth and that T1一“be ttze sotutior?of地e parallel subspace correction algorithm．When At and h are s听ciently small玎At=o(S2)h=。(日)，叫e have 怯?叫怪c hk+i+At+(豢+刍)等)， ㈦ Ⅲ^e7℃C denotes口qeneric constant independent ol At．h and H but dependent on Theorem 3÷1 Suppose that the,solution 0f(3 1)is sufficiently smooth and that II—be the solution of the averaging parallel algorithmj we have |l“n一-一。川。墨c(n‘一l一土t+(-一志)”)，c。， 一4一 山东大学硕士学位论文 whcT((、denote．x。generic constant mdependent of At．^and H but dependent on 。u． Key words：Domain decomposition：Parabolic problems；Subspace correction Collvergellee rate：Error estimates 一5一 山东大学硕士学位论文 第一章 引 言 数学物理及工程问题，如油、气藏的勘探与开发、大型结构工程航天器的设 计、天气预报、反应堆计算等，无不归结于求解大型偏微分方程．计算区域往往是 高维的、大范围的，其形态可能很不规则，给计算带来很大的困难．在实践中，对 这些问题的计算精度要求越来越高，而单机计算的速度已接近极限，囤此不可避 免地促进了并行计算与区域分解算法的发展． 简而言之，区域分解算法是把计算区域分解为若干子域，子域的形状尽可能 的规刚，于是原问题的求解转化为在子域上求解，区域分解算法具有其它方法无以比拟的优越性： l它把大问题转化为若干小问题，缩小计算规模． 2算法是高度并行的， 3子区域形状如果规则，其上或者允许使用熟知的快速算法，或者已经有解 这类规则问题的高效率软件备用 4允许使用局部拟一致网格，无