A modularized package of dual algorithms for solving constrained nonlinear programming problems

Abstract

A package of algorithms for numerical experimentation with constrained optimization methods is described. This package is addressed to dual methods and is modularized to permit experiments with a particular module (such as a penalty function module) without changing any of the other modules. Thus reliable comparisons can be made without questioning differences in implementations of the remaining portions of a constrained algorithm. The feasibility of modularization as well as advantages and disadvantages for dual algorithms are discussed. Some numerical results are also presented.