Differential evolution methods based on local searches

In this paper we analyze the behavior of a quite standard Differential Evolution (DE) algorithm applied to the objective function transformed by means of local searches. First some surprising results are presented which concern the application of this method to standard test functions. Later we introduce an application to disk- and to sphere-packing problems, two well known and particularly hard global optimization problems. For these problems some more refined variations of the basic method are necessary in order to take at least partially into considerations the many symmetries those problems possess. Coupling these techniques with DE and local optimization resulted in a new method which, when tested on moderately sized packing problems, was capable of confirming known putative optima for the problem of packing disks, and of discovering quite a significant number of new putative optima for the problem of packing spheres.