Asynchronous Parallel Pattern Search with Bound Constraints

Tamara G. Kolda, Sandia National Labs

2000 SIAM Annual Meeting
Puerto Rico
July 10-14, 2000

Abstract

Asynchronous parallel pattern search (APPS) is a practical method for many optimization engineering problems since it does not require gradient or approximate gradient information and can execute function evaluations in parallel over a heterogeneous and even fault-prone cluster computing environment. However, APPS was designed for unconstrained optimization while most engineering problems have bound constraints. We introduce several alternatives for incorporating bound constraints into APPS and present numerical results for various engineering problems.

Slides

postscript or pdf


Maintained by: Tamara Kolda (tgkolda@sandia.gov).
Disclaimer and Acknowledgment.