线性约束最优化问题的一族次可行方向法

本文给出线性约束最优化问题的一族算法．方法具有如下特点：１）初始迭代点可以任意选取；２）一旦有某一个迭代点进入可行域，方法将成为一族可行方向法；３）算法避开不易处理的罚函数和罚参数．文中采用一种最优性控制函数将初始化阶段和最优化阶段有机地结合起来，正是这种技巧保证了算法的全局收敛性

A SORT OF THE METHODS OF SUBFEASIBLE DIRECTIONS FOR LINEAR CONSTRAINED OPTIMIZATION PROBLEMS

In this paper,a class of algorithms for linear constrained optimization problems is presented.It possesses the following special features:1) Arbitrary point may be chosen as a starting iteration point; 2) The method will become a sort of methods of feasible directions whenever some iteration point goes into the feasible region;3) It escapes from the penalty functions and penalty parameters which are not managed easily.A function controlling optimality is used to combine automatically the phases of initialization and optimization.It is such a technique that engages the global convergence of the algorithm.