О применении одного класса параметрических функций в качестве внешних штрафов при решении нелинейных задач с ограничениями

Abstract

Kaplan’s one-parametric class of functions for solving nonlinear convex minimization problems with one-sided constraints is studied. The set defined by constraints is assumed to have a nonempty interior. In the monographs written by A. Fiacco, G. McCormick, E. Polak, and A. Kaplan, the penalty methods theory and classification of penalty functions are presented quite systematically. With these methods and approaches as backbones, in the present article, we establish that the class of functions under study belongs to the class of exterior penalty functions for problems of convex programming. Application of penalty methods to the solution of nonlinear extremal problems with constraints allows using the toolbox of unconstrained nonlinear optimization, including gradient methods. Kaplan’s penalty functions have fine differential properties. Therefore, they are suitable for use in iterative gradient methods of approximate solution of unconstrained extremal problems. After this, we prove the theorem on convergence of the sequence of approximate solutions of penalized unconstrained extremal problems to the exact solution of the original problem with constraints. As well, we establish a bound on the convergence rate for the penalty method with the one-parametric class of functions serving as penalty functions. This bound is derived provided that the exact resolving of the sequence of unconstrained extremal problems is fulfilled. With the help of these results, one may proceed further with a numerical analysis of the class of problems that are under discussion in the article.