摘要
本文对Ｐｏｌａｋ等人的组合ＮａｓｅⅠ－Ⅱ可行方向法进行改进，使之不仅能自动地将初始化阶段（Ｐｈａｓｅｌ）和最优化阶段（ＰｈａｓｅⅡ）统一起来，而且保证了满足不等式约束的函数个数不断叠累递增，故称改进后的算法为强组合ＰｈａｓｅⅠ－ⅡＰｈａｓｅⅡ次可行方向法．本文算法包含了一种新的目标局数非单词的非精确线搜索，它保证了算法产生的点列的任何聚点都是问题的Ｋ－Ｔ的点．
In this paper, Polak's combined Phase Ⅰ-Phase Ⅱ methods of feasible directions is modified,such it can not only automatically unify the operations of initialization(Phase Ⅰ) and optimization (Phase Ⅱ),but also guarantees the number of the functionssatisfying inequality constraints is monotonically increased, so this modifed algorithm is called to be a strong combined Phase Ⅰ-Phase Ⅱ method of sub feasible directions. Our algorithm contains a new nonexact and nonmonotonic (about the object value) line search,which guarantees the algorithm convergence to the Kuhn-Tucker Point under the nondegeneracy assumption.
出处
《经济数学》
1995年第1期64-70,共7页
Journal of Quantitative Economics
