非线性规划的不使用导数的解析方法——统一探讨

Non-Derivative Analytic Methods for Nonlinear Programming—A Unified Approach

本文从统一的角度给出了解约束优化问题的修正的解析方法,即用梯度的近似值代替梯度的解析方法,并在较弱的条件下证明了其全局收敛性。本文结果扩大了解析方法的应用范围,为优化算法的实现提供了理论依据.

This paper gives a unified approach to the modified analytic methods for constrained optimization, in which the gradient is replaced by the approximate value of gradient. Under mild conditions, the global convergence of the methods here is proved. By empoloying the results of the paper, the analytic methods find a wider range of application, and the results also give a theoretical basis to the implementation of optimization methods.