本文提出了一个新的非单调序列线性方程组(SSLE)算法.在每次迭代过程中只需解三个具有相同系数矩阵的线性方程组,以替代解二次规划子问题,使得新算法的总计算量大大减少.该算法不需要罚函数也无需滤子,从而避免了由罚参数的选取所带来的困难.并且适用于解所有一般约束优化问题,无需初始点可行.该算法具有全局收敛性.数值结果表明该算法是有效的.
In this paper, we propose a new nonmonotone algorithm using the sequential systems of linear equations(SSLE), which is an infeasible QP-free method, and also a penaltyfunction-free method. We do not use the penalty function or a filter. The new algorithm only needs to solve three systems of linear equations having the same nonsingular coefficient matrix instead of solving QP problems. The global convergence is established. At last, the numerical results are shown.