摘要
本文讨论极小化由凸泛函和光滑算子复合而成的目标函数的数值方法,给出了旨在求上述问题的一个平稳点的拟牛顿型算法,它将原问题转化为求解一系列约束凸极小化问题的近似解.在适当的条件下算法具有全局收敛性,当目标函数满足增长条件时算法有超线性的敛速.
This paper deals with the minimization of the composite nonsmooth func- tional which is composed of a smooth operator and a convex functional.An quasi-newton-type algorithm is given,which converts the problem to a series of constrained convex minimization problems.Under adequate conditions,the global convergence is proved and superlinear convergernce rate of the algorithm can be obtained.
出处
《西安交通大学学报》
EI
CAS
CSCD
北大核心
1990年第6期105-112,共8页
Journal of Xi'an Jiaotong University
关键词
牛顿型算法
收敛性
极小化
Newton like method
convergence
minimization
