计算机工程与应用 ›› 2009, Vol. 45 ›› Issue (22): 41-43.DOI: 10.3778/j.issn.1002-8331.2009.22.014

• 研究、探讨 • 上一篇    下一篇

改进的求解线性方程组的并行Arnoldi方法

汪 保1,2,吕全义1,樊艳红1,聂玉峰1   

  1. 1.西北工业大学 应用数学系,西安 710072
    2.西北工业大学 航空学院,西安 710072
  • 收稿日期:2008-04-28 修回日期:2008-07-21 出版日期:2009-08-01 发布日期:2009-08-01
  • 通讯作者: 汪 保

Improved parallel Arnoldi method for solving linear equations

WANG Bao 1,2,LV Quan-yi1,FAN Yan-hong1,NIE Yu-feng1   

  1. 1.Department of Applied Mathematics,Northwestern Polytechnical University,Xi’an 710072,China
    2.Academy of Aeronautics,Northwestern Polytechnical University,Xi’an 710072,China
  • Received:2008-04-28 Revised:2008-07-21 Online:2009-08-01 Published:2009-08-01
  • Contact: WANG Bao

摘要: 以Galerkin原理为基础,提出了求解循环块三对角线性方程组的并行算法。根据系数矩阵的稀疏性,选取适当的子空间的基,使算法不但不会发生中断,并从理论上证明了当系数矩阵对称正定时,该并行算法收敛。最后,在HP rx2600集群上进行的数值实验结果表明,该算法的并行效率很高,理论和实际计算相一致。

关键词: 循环块三对角线性方程组, 并行算法, Arnoldi方法

Abstract: A parallel algorithm based on Galerkin method for cycle block-tridiagonal linear equations on distributed-memory multi-computers is presented.A group of vectors spanning subspace chosen properly,the algorithm is no disrupted.In theory,convergence is proved when the coefficient matrix A is a symmetric positive definite matrix.Finally,some numerical results on HP rx2600 cluster show that practice computing is consistent with theory.

Key words: cycle block-tridiagonal linear equations, parallel algorithm, Arnoldi method