摘要
锥模型优化方法是一类非二次模型优化方法,它在每次迭代中比标准的二次模型方法含有更丰富的插值信息.Di和Sun(1996)提出了解无约束优化问题的锥模型信赖域方法.本文根据Fletcher和Leyffer(2002)的过滤集技术的思想,在Di和Sun(1996)工作的基础上,提出了解无约束优化问题的基于锥模型的过滤集信赖域算法.在适当的条件下,我们证明了新算法的收敛性.有限的数值试验结果表明新算法是有效的.
Conic model optimization methods are one class of non-quadratic model optimization methods.It contains richer interpolation information than the standard quadratic model.Di and Sun(1996) proposed a trust-region method based on conic model for unconstrained optimization.Based on Di and Sun(1996),this paper employs the filter technique due to Fletcher and Leyffer(2002) and proposes a filter-trust-region method based on conic model for unconstrained optimization.Under mild conditions,we prove the convergence of new algorithm.Limited numerical results show that the new algorithm is efficient.
出处
《中国科学:数学》
CSCD
北大核心
2012年第5期527-543,共17页
Scientia Sinica:Mathematica
基金
国家自然科学基金(批准号:11071122
11171159)
教育部博士点基金(批准号:20103207110002)资助项目
关键词
无约束优化
锥模型
信赖域方法
过滤集方法
收敛性
unconstrained optimization
conic model
trust region method
filter technique
convergence