On convergence of "divide the best" global optimization algorithms

Abstract

In this paper a new class of multidimensional global optimization algorithms (called "divide the best" algorithms) is proposed. The class unifies and generalizes the classes of the characteristic methods and the adaptive partition algorithms introduced by Grishagin and Pinter respectively. The new scheme includes also some methods which do not fit either the characteristic or the adaptive partition families. A detailed convergence study is presented. A special attention is paid to cases where sufficient conditions of everywhere dense, local and global convergence are fulfilled only over subregions of the search domain.