Sequential Optimization of Multimodal Discrete Function with Bounded Rate of Change

Abstract

In this paper a search scheme for finding the maximum of a discrete objective function of several variables is proposed. The scheme is sequential and optimal in the sense that it determines the maximum with minimum number of measurements of the function. Unlike other sequential search techniques with this property the scheme places no condition on the number of possible maxima of the objective function. Instead it requires that some bounds on the rate of change of the function per each variable be known. The points at which measurements of the value of the objective function are to be taken are determined successively by maximization of an auxiliary function updated after each measurement. The algorithm is described in detail, and illustrated by two simple examples.