OQNLP and MSNLP

OQNLP and MSNLP are multistart heuristic algorithms designed to find global optima of smooth constrained nonlinear programs (NLPs). By ”multistart” we mean that the algorithm calls an NLP solver from multiple starting points, keeps track of all feasible solutions found by that solver, and reports back the best of these as its final solution. The starting points are computed by a scatter search implementation called OptQuest (see www.opttek.com and [Laguna and Marti, 2003]) or by a randomized driver, which generates starting points using probability distributions. There are currently two randomized drivers, Pure Random and Smart Random-see the description of the POINT GENERATION keyword in Section 3, and Appendix B. With OQNLP, all three drivers are provided, while OptQuest is not present in MSNLP. When interfaced with the GAMS modeling language, any GAMS NLP solver can be called. When used as a callable system, MSNLP uses the LSGRG2 NLP solver (see www.optimalmethods.com and (Smith and Lasdon, 1992)), and this is also provided (optionally) in the GAMS version.