首页 | 本学科首页   官方微博 | 高级检索  
     检索      

块三对角线性方程组的重叠分割可扩展并行近似求解方法
引用本文:张衡,张武,封卫兵.块三对角线性方程组的重叠分割可扩展并行近似求解方法[J].上海大学学报(自然科学版),2007,13(2):165-171.
作者姓名:张衡  张武  封卫兵
作者单位:1. 上海大学,计算机工程与科学学院,上海,200072;石河子大学,师范学院数学系,石河子,832000
2. 上海大学,计算机工程与科学学院,上海,200072
基金项目:教育部科学技术研究重点项目;上海市自然科学基金
摘    要:基于并行计算的分治思想,对于严格块对角占优的块三对角线性方程组提出一个可扩展 的块重叠分割并行近似求解方法(PBOA方法).在机器精度内,利用块对角占优的条件,只需要相邻处理器间一次通讯,得到与精确解等价的近似解.在算法设计中,充分考虑计算与通信的重叠和处理机间负载平衡.通过精度分析,给出子方程组的阶数与精度的关系,从而得到通过调整子方程组的阶数来控制精度和并行效率,保证可扩展性的方法,得到的并行计 算效率可随着问题规模的增加而增加.该文的方法在上海大学并行计算机“自强3000”上运行,数值实验的结果与理论分析的结果一致,得到的并行计算效率接近67%,加速比几乎是线性的.

关 键 词:块三对角线性方程组  块对角占优  块LU分解  矩阵分割  相对误差
文章编号:1007-2861(2007)02-0165-07
收稿时间:2006-06-29
修稿时间:2006年6月29日

Approximate Parallel Solution to System of Block Tri-diagonal Linear Equations with Overlapped Partition Algorithm
ZHANG Heng,ZHANG Wu,FENG Wei-bing.Approximate Parallel Solution to System of Block Tri-diagonal Linear Equations with Overlapped Partition Algorithm[J].Journal of Shanghai University(Natural Science),2007,13(2):165-171.
Authors:ZHANG Heng  ZHANG Wu  FENG Wei-bing
Institution:1. School of Computer Engineering and Science, Shanghai University, Shanghai 200072, China; 2. Department of Mathematics, Teachers College, Shihezi University, Shihezi 832000, China
Abstract:A high efficiency scalable parallel algorithm,parallel block overlapped partition approximate(PBOA) algorithm,is proposed for solving block tri-diagonal linear systems on multiple computers.The algorithm is based on the divided-and-conquer idea in parallel computation.When the system is strictly block diagonal dominant,the PBOA is highly parallel and provides an solution that equals the exact solution within machine accuracy with finite communications between adjacent processors.By analyzing the accuracy,the relation between the accuracy and the order of the system is obtained.The authors propose the method for improving efficiency and accuracy.The computation efficiency increases with size of the problem.The proposed method has been implemented on 64 nodes of the ZQ3000 parallel computer at Shanghai University.The numerical results agree with the theoretical analysis.With desired accuracy,linear speedup has been obtained,and the parallel efficiency approaches 67%.
Keywords:block tri-diagonal linear systems  block diagonal dominant  block LU decomposition method  matrix partitioning  relative error
本文献已被 CNKI 维普 万方数据 等数据库收录!
点击此处可从《上海大学学报(自然科学版)》浏览原始摘要信息
点击此处可从《上海大学学报(自然科学版)》下载免费的PDF全文
设为首页 | 免责声明 | 关于勤云 | 加入收藏

Copyright©北京勤云科技发展有限公司  京ICP备09084417号