An Adaptive Random Search Algorithm for Constrained Minimization

The well-known pattern search method has been randomized by Lawrence and Steiglitz [4] in order to augment its ability to adapt to direction. In this note we introduce other refinements in order to make it more adaptive in step size, and the use of penalty terms is incorporated as so to accommodate constraints. This yields an algorithm that appears to be robust and reliable, as verified by numerical experimentation. Moreover, the method has low storage requirements and is insensitive to the initial choice of a search radius.