Author(s)

D. El Baz

Abstract

Numerical performances of parallel algorithms for a class of optimization problems D. El Baz LAAS du CNRS, 7, avenue du Colonel Roche, 31077 Toulouse Cedex, France ABSTRACT We present parallel relaxation and gradient algorithms for a class of non- linear optimization problems: single commodity convex network flow prob- lems. We discuss the synchronous and asynchronous implementation of the algorithms on the distributed memory multiprocessor T-node 16-32. We present and analyse computational results. INTRODUCTION We consider a particular class of nonlinear optimization problems: single commodity convex network flow problems. These problems occur in many domains: electrical networks, gas or water distribution, financial models. Typically, nonlinear network flow problems require intensive computations (see [29]). As a consequence the introduction of parallelism seems to be very attractive. We concentrate on the dual problem which is unconstrained, dif- ferenti

Keywords