无约束最优化问题的BFGS并行算法与实现.pdfVIP

第35 卷 第15 期 计 算 机 工 程 2009 年8 月 V .35 No.15 Computer Engineering August 2009 ol ·软件技术与数据库· 文章编号：1000—3428(2009)15—0058—03 文献标识码：A 中图分类号：TP301.6 无约束最优化问题的BFGS 并行算法与实现 李文敬，王汝凉，廖伟志 (广西师范学院信息技术系，南宁 530001) 摘 要：介绍无约束最优化问题的 BFGS 算法及其收敛性，提出利用行卷帘格式并行 Cholesky 分解法、同步并行 Wolfe-Powell 非线性搜 索和并行处理 BFGS 修正公式来构建 BFGS 的并行算法，并对该算法的时间复杂性、加速比进行分析。在 PC 机群数值实验的结果表明， BFGS 并行算法提高了无约束最优化问题的求解速度，理论分析与实验结果相一致，并行算法具有线性加速比。 关键词：无约束最优化；BFGS 并行算法；Cholesky 分解；加速比 BFGS Parallel Algorithm of Unconstrained Optimization Problems and Its Implementation LI Wen-jing, WANG Ru-liang, LIAO Wei-zhi (Department of Information Technology, Guangxi Teachers Education University, Nanning 530001) 【Abstract 】Based on analysis of both the Broyden-Fletcher-Goldfarb-Shanno(BFGS) algorithm and its convergence properties of unconstrained optimization problems, this paper presents a parallel algorithms of BFGS by using the row interleaved format parallel decomposition of Cholesky, the synchronous parallel Wolfe-Powell non-linear search and the modified formula of BFGS. The paper also analyszs the time complexity and speedup ratio of the algorithm. Experimental results of PC cluster show that the BFGS parallel algorithm improves solution speed of unconstrained optimization problems. The theoretical analysis and the experimental results of BFGS parallel algorithm are consistency with a linear speedup ratio. 【Key words 】unconstrained optimization; Broyden-Fletcher-Goldfarb-Shanno(BFGS) parallel algorithm; Chole