An heuristic strategy for multimodal search?

A heuristic search is described which has the aim of finding practically all extrema of a given non-linear functional. A standard unimodal descent algorithm is employed for finding individual extrema. This basic algorithm is applied repeatedly using various computed initial points and starting directions. Through an additional vise of several learning cycles most available extrema can be found. Numerical experiments indicate that the method is certainly very efficient for functionals of dimensions 15– 20 with 20–25 extrema.