Global Optimization Using Diffusion Perturbations with Large Noise Intensity

This work develops an algorithm for global optimization.The algorithm is of gradient ascent typeand uses random perturbations.In contrast to the annealing type procedurcs,the perturbation noise intensityis large.We demonstrate that by properly varying the noise intensity,approximations to the global maximumcan be achieved.We also show that the expected time to reach the domain of attraction of the global maximum,which can be approximated by the solution of a boundary value problem,is finite.Discrete-time algorithmsare proposed;recursive algorithms with occasional perturbations involving large noise intensity are developed.Numerical examples are provided for illustration.

Global Optimization Using Diffusion Perturbations with Large Noise Intensity

This work develops an algorithm for global optimization. The algorithm is of gradient ascent type and uses random perturbations. In contrast to the annealing type procedures, the perturbation noise intensity is large. We demonstrate that by properly varying the noise intensity, approximations to the global maximum can be achieved. We also show that the expected time to reach the domain of attraction of the global maximum, which can be approximated by the solution of a boundary value problem, is finite. Discrete-time algorithms are proposed; recursive algorithms with occasional perturbations involving large noise intensity are developed. Numerical examples are provided for illustration.