Parallel Quasi-Newton Methods for Unconstrained Optimization

Abstract

This paper describes work done on the 1024 node NCUBE hypercube at the University of South Carolina in developing methods for efficient local solution of unconstrained minimization problems. The paper begins with a mathematical discussion of quasiNewton methods for unconstrained optimization, and specifically Broyden’s Method. Next it presents the paralfel methods, and discusses the parallel implementation of the most common Broyden method. Finally it lists some numerical results to evaluate the performance of the parallel Broyden methods.