METHODS FOR EVALUATING NONLINEAR PROGRAMMING SOFTWARE11This work was supported by the Applied Mathematical Sciences Research Program (KC-04-02) of the Office of Energy Research of the U.S. Department of Energy under Contract W-31-109-Eng-38.

In evaluating nonlinear programming codes, it has historically been the case that codes are run on a variety of test problems and summary statistics are presented. More recently randomly generated test problems have been used in an attempt to produce a more complete spectrum of problems. In this paper we consider the advantages and disadvantages of these two approaches and introduce a third approach in which one parameter sets of problems are constructed to demonstrate a single numerical feature. This approach, based on performance profiles, involves the measurement of a single set of parameters which can be used as evaluation criteria.