计算物理 ›› 2006, Vol. 23 ›› Issue (1): 50-56.

• 论文 • 上一篇    下一篇

无需线搜索的并行非线性共轭梯度法

王建1,2,3, 迟学斌1, 谷同祥4, 冯仰德1   

  1. 1. 中科院计算机网络信息中心超级计算中心, 北京 100080;
    2. 中科院软件所, 北京 100080;
    3. 中国科学院研究生院, 北京 100039;
    4. 北京应用物理与计算数学研究所计算物理实验室, 北京 100088
  • 收稿日期:2004-10-26 修回日期:2005-04-15 出版日期:2006-01-25 发布日期:2006-01-25
  • 作者简介:王建(1976-),男,山东荷泽,博士,从事并行最优化算法和软件的研究,北三环西路45号开放实验室100086.
  • 基金资助:
    国家重点基础研究专项(G1999032805);中国科学院知识创新工程信息化建设专项"超级计算环境建设与应用"(INF105-SCE)和"超级计算网格节点建设"(863,2002AA-104540)联合资助项目

A Parallel Nonlinear Conjugate Gradient Method with No-line-search

WANG Jian1,2,3, CHI Xue-bin1, GU Tong-xiang4, FENG Yang-de1   

  1. 1. Supercomputing Center, Computer Net Work Information Center, Chinese Academy of Sciences, Beijing 100080, China;
    2. Institute of Software, Chinese Academy of Sciences, Beijing 100080, China;
    3. Graduate School, Chinese Academy of Sciences, Beijing 100039, China;
    4. Laboratory of Computational Physics, Institute of Applied Physics and Computational Mathematics, Beijing 100088, China
  • Received:2004-10-26 Revised:2005-04-15 Online:2006-01-25 Published:2006-01-25

摘要: 给出了一类无需线搜索的无约束最优化并行算法--并行非线性共轭梯度法(NLS-PNCG),用一个固定的公式来计算搜索步长,较常用的共轭梯度法计算量小.并且证明了目的的全局收敛性,给出了数值试验,结果显示NLS-PNCG优于线搜索的非线性共轭梯度法(NCG).

关键词: 无约束最优化, 共轭梯度法, 搜索步长, NLS-PNCG

Abstract: A no-line-search parallel nonlinear conjugate gradient method(NLS-PNCG) for unconstrained optimization is proposed. In this method the step length is evaluated by a fixed formula. It is shown that the NLS-PNCG requires less computation and its performance is superior to those with line search on Deep Comp 6800.

Key words: unconstrained optimization, conjugate gradient methods, step length, NLS-PNCG

中图分类号: