Performance Evaluation and Selection of Optimization Software

Arizona State University, Scottsdale AZ

Investigators

Abstract

This project will extend the state of the art in evaluating optimization software. Beginning with a number of subclasses of continuous local optimization problems and progressing to global as well as suitable selected discrete optimization problems a careful evaluation and comparison of available implementations will be done. One essential part of this is the systematic collection and generation of test problems. Next, a group of programs will be chosen in each category and the issue of comparability, for example with respect to termination criteria, will be addressed. The results of applying codes to the test sets will be documented and published both on the web and in print. It is expected that this will lead, just as in the previous work, indirectly to an improvement of many of the programs by their authors. In addition, there will be direct improvements through the PI's experiences and suggestions. This work will complement related activities such as NEOS, netlib, etc. A second product of the proposed work is a guide in the form of a collection of webpages through which a prospective user of optimization methods is directly offered a concrete avenue, typically including appropriate software, to solve the problem at hand. Based on the success of the earlier work, the proposed research may be anticipated to have a substantial impact on both the production of high-quality software for modern architectures and the utilization of that software by the user community, both academic and non-academic.