BFGS算法分析与实现选编.docx

《最优化方法》课程设计题 目：BFGS算法分析与实现院 系： 数学与计算科学学院 专 业： 统计学姓名学号： 左想 1200720133 指导教师：李丰兵日期： 2014年01 月22日摘 要在求解无约束最优化问题的众多算法中,拟牛顿法是颇受欢迎的一类算法。尤其是用于求解中小规模问题时该类算法具有较好的数值效果。 BFGS 算法被认为是数值效果最好的拟牛顿法，其收敛理论的研究也取得了很好的成果. 在一定的条件下，BFGS 算法具有全局收敛性和超线性收敛速度。 然而，对于大规模最优化问题来求解，包括 BFGS 算法在内拟牛顿法具有明显的缺陷。有许多的例子表明，一旦处理问题很大时，一些对小规模问题非常成功的算法变得毫无吸引力。究其原因，主要是由于在中小型问题一些不太重要的因素在求解大规模问题时，变得代价很高。随着速度更快及更复杂的计算机的出现，增强了我们的计算处理能力。 同时也为我们设计算法带来了新的课题。并行计算机的发展为求解大规模最优化问题提供了一条新途径。关键词：BFGS 拟牛顿法；无约束最优化问题；大规模问题AbstractQuasi-Newton methods are welcome numerical methods for solving optimization problems. They are particularly effective when applied to solve small or middle size problems. BFGS method is regarded as the most effective quasi-Newton method due toits good numerical performance. It also possesses very good global and superlinear con vergenceproperties. On the other hand, however, when applied to solve largescaleproblems, quasi-Newton methods including BFGS method don ot perform well. Themajor drawback for a quasi-Newton method, when used to solve large scaleoptimization problem, is that the matrix generated by the method does not retain thesparsity of the Hessian matrix of the objective function. There are examples showing that many success methods for solving small-sized optimization become unattractive once the problem to be tackled is large. An important reason for thisfeature is that some process that is important for small problems may become veryexpensive for large scale problems.The fast development of computer has enhanced our ability to solve large scaleproblems. In particular, the parallel computer provides us a new way to solve largescale problems efficiently. In recent years, there has been growing interest in the studyin parallel methods. It has been found that many good methods that are efficient forsolving small and middle size problems can be parallized. Key Words: BFGS quasi-Newton method; unconstrained optimization problem, large scale problem目录1、引言12、BFGS算法及其修正形