A Simple Multistart Algorithm for Global Optimization

A generalization of the multistart algorithm is proposed for finding the globalminimizer of a nonlinear function of n variables. Our method concentrates a quasirandomsample by performing a few inexpensive local searches. The sample is thed reduced byreplacing worse points by flew quasirandom points. A complete local search is performedonly on those points with small function values. This method performs favorably incomparison to other global optimization methods.