一类大规模二次规划问题的可行下降分解算法

A Feasible Descent Decomposition Algorithm for Large-scale Quadratic Programming

本文对一类大规模二次规划问题,提出了矩阵剖分的概念和方法,并将问题转化为求解一系列容易求解的小规模二次规划子问题.另外,通过施加某些约束机制,使子问题所产生的迭代点均为可行下降点.在通常的假定下,证明算法具有全局收敛性,大量数值实验表明,本文所提出的新算法是有效的.

Abstract：In this paper, a new algorithm for solving large scale quadratic programming is proposed. We decompose a large scale quadratic programming into a serial of small scale ones whose solutions approximate that of the large scale quadratic programming. Furthermore, the algorithm is a feasible descent one. It is proved that the algorithm proposed is of the global convergence under the certain conditions. Numerical tests show that the algorithm has performed more effectively.