Multi-scale optimization

Document Type

Article

Date of Original Version

12-1-2004

Abstract

Global optimization problems with so-called 'rough' or rugged objective function landscapes are studied. These problems often have many, many stationary points and show considerable differences between small and large-scale geometry. A novel multi-scale global optimization algorithm for solving 'rough' objective functions, based on the alternate use of terrain methods and new funneling algorithms, is presented. Small-scale information is gathered using a terrain optimization methodology while funneling algorithms are used to guide the overall optimization calculations and to make 'large' moves within the feasible region. A molecular modeling example is used to clearly illustrate that the proposed methodology is capable of finding a global minimum without calculating all stationary points and can lead to significant reductions in computational work. © 2004 Elsevier B.V. All rights reserved.

Publication Title, e.g., Journal

Computer Aided Chemical Engineering

Volume

18

Issue

C

Share

COinS