一种满足夹角性质的超记忆梯度方法  

A modified super-memory gradient method with angle property

作　　者：孙敏[1] 

机构地区：[1]枣庄学院数学与信息科学系,山东枣庄277160

出　　处：《山东大学学报（理学版）》2008年第6期68-70,76,共4页Journal of Shandong University(Natural Science)

摘　　要：提出了一种修正的超记忆梯度方法。该方法的优点是:(1)在无需线性搜索的条件下,迭代方向就是充分下降方向;(2)迭代方向保持夹角性质。在较弱的条件下,分析了方法的全局收敛性。初步的数值试验表明了方法的有效性。A modified super-memory gradient method was proposed. The method has the following two attractive properties, one is that the iterative direction is always a sufficiently descent direction without utilizing the line search, the other is that the iterative direction always satisfies the angle property. Under mild conditions, the global convergence of the method was proved. The preliminary numerical results show the efficiency of this method.

关 键 词：超记忆梯度法 充分下降 全局收敛性 

分 类 号：O221.2[理学—运筹学与控制论]