一类混合CD-LS共轭梯度法的全局收敛性  

Global convergence of a hybrid CD-LS conjugate gradient method

作　　者：张雁[1] 单锐[1] 王换鹏[1] 靳飞[1] 

机构地区：[1]燕山大学理学院,河北秦皇岛066004

出　　处：《辽宁工程技术大学学报（自然科学版）》2013年第3期409-412,共4页Journal of Liaoning Technical University (Natural Science)

基　　金：河北省教育厅科学研究基金资助项目(2009159)

摘　　要：为了寻找同时具有良好的收敛性和数值表现的共轭梯度法.将CD方法和LS方法结合,选用推广的Wolfe线搜索,构造出一类新的混合共轭梯度法.新的混合共轭梯度法不需要限制推广的Wolfe线搜索条件中的参数,但得到的下降性与CD法一致,具有比CD方法更好的收敛性,并具有全局收敛性.对新算法进行数值试验,通过与CD法和LS法的数值结果进行比较,表明新算法是可行的,尤其对大规模无约束优化问题.In order to find a good convergence and numerical expression of conjugate gradient method at the same time,combining the CD method and the LS method,and using general Wolfe line search,this paper has developed a new hybrid conjugate gradient method.The new hybrid conjugate gradient method is not required to restrict parameters in a general Wolfe line search,but gets the same descent property as the CD method and has a much better convergence.The new method achieves a global convergence.The numerical test of new hybrid algorithm and the comparison with numerical results of the CD method and the LS method show that the new algorithm is feasible,especially for a large-scale unconstrained optimization problem.

关 键 词：无约束优化问题 共轭梯度法 混合 WOLFE线搜索 推广 下降性 全局收敛性 数值表现 

分 类 号：O224.2[理学—运筹学与控制论]