摘要
基于变换X=VV^T,本文将半定规划问题转换为非线性规划问题,提出了解决此问题的增广拉格朗日算法,并证明了算法的线性收敛性.在此算法中,每一次迭代计算的子问题利用最速下降搜索方向和满足wolf条件的线性搜索法求最优解.数值实验表明,此算法是行之有效的,且优于内点算法.
Based on the change of X = VVT, an augmented lagrangian algorithm to solve convex quadratic SDP is proposed. The algorithm＇s distinguishing feature is a factorization, the gradient method and an exact linesearch procedure. The convergence of the algorithm is shown. Numerical experiments show that our methods are efficient and robust.
出处
《计算数学》
CSCD
北大核心
2014年第2期133-142,共10页
Mathematica Numerica Sinica
关键词
二次半定规划
分解变换
增广拉格朗日算法
线性搜索
convex quadratic SDP
change of factorization
Augmented Lagrangian
exact linesearch
