通过使用线搜索技术,提出了一类具有整体收敛性的不精确修正正割算法解非线性约束优化问题.引入Fletcher罚函数作为价值函数克服了产生Maratos效应.在合理条件下证明了该类算法具有二步q阶超线性收敛速率.进而,对于约束进行很小的额外计算改进了此类算法,以使新算法具有一步q阶超线性收敛速率.数值实验的结果证明了该算法的有效性和可行性.
We propose a class of modified global convergent inexact secant methods in association with line search technique for solving nonlinear constrained optimization problems. By introducing Fletcher's penalty function as a merit function, the Maratos effect can be avoided. The resulting algorithms possess global convergence while maintaining two - step q - superlinear local convergence rates under some reasonable conditions. Furthermore, with one extra evaluation of the constraints at each iteration, the improved algorithms have one - step q - superlinear local convergence rates. The results of numerical experiments indicate that the proposed algorithms are efficient for the given test problems.