A Methodological Approach to Testing of NLP-Software

Prior to detailed testing of software there should be: (1) knowledge of the software and the underlying algorithms; (2) knowledge of the class of problems to be solved by this software; and (3) goals of testing.We present a methodological setup of testing nonlinear programming software which is based on prior knowledge. This leads to a discussion of the following topics: representativity of test problems, use of test problem generators, performance measures, presentation of test results, and presentation of final conclusions.The concepts will be illustrated with a project which has been performed to evaluate a set of Newton-like methods for solving systems of nonlinear equations.