复合不可微最优化问题的非单调信赖域方法  

Nonmonotonic Trust Region Method for Composite Nonsmooth Optimization Problems

作　　者：白延琴[1] 孙小玲[1] 

机构地区：[1]上海大学数学系

出　　处：《工程数学学报》1999年第2期15-21,共7页Chinese Journal of Engineering Mathematics

基　　金：上海市教委青年科学基金

摘　　要：对复合不可微最优化问题提出了一种新的非单调信赖域方法。算法在每个迭代点处构造带信赖域约束的二次规划子问题，新的迭代点采用非单调策略产生，在一般的假设条件下证明了算法的全局收敛性。Abstract This paper presents a nonmonotonic trust region method for compostie nonsmooth optimization problems. The method constructs quadratic programming subproblems with trust region consraint at each iteration. To overcome the Maratos effect rising from the nondifferentiability, the method generates new iteration points by using nonmonotonic strategy. Global convergence of the method is proven under some mild assumptions.

关 键 词：不可微最优化 信赖域 全局收敛性 非单调信赖域 

分 类 号：O224[理学—运筹学与控制论]