A Search Technique for Multimodal Surfaces

Abstract

During the past decade many different computeroriented techniques for locating the extremum of a mathematically defined n-dimensional surface have been developed for use as aids toward optimum system design. The great majority of these techniques locate only the nearest peak if the surface is multimodal. This paper presents a technique for locating the global extremum of a multimodal surface. The search strategy is divided into three phases: a global search, a selection of hypervolume containing the global extremum, and a final unimodal search. The search strategy is discussed along with an illustrative example problem.