摘要
本文提出了一种极小化不可微合成函数的下降算法,该算法通过内部迭代寻找下降方向,每次内部迭代求解一个二次规划.外部迭代点通过不精确线搜索求得.算法在有限步内得到近似平稳点,经过适当修正后,算法全局收敛到平稳点.
In this paper,we give an algorithm for composite nondifferentiable optimization problems. The algorithm finds descent directions by using inner iterations in which quadratic progrom- mings are solved.Each outer iterative point is obtained by inexact line search.The algorithm finds approximate stationary point within finite number of iterations.After appropriate modifi- cation,the outer iterative point sequence generated by the algorithm will converge to stationary point.
