Additional Perspectives on Simulation for Optimization

Abstract

Professor Fu has provided an excellent survey of the state-of-the-art in the area of simulation for optimization. His discussion offers insights into topics ranging from the evolution of commercial simulation software to future directions for academic research in the area. As pointed out by Professor Fu, the incorporation of optimization routines into simulation software packages is a relatively recent development. We offer here a possible explanation for this history. One of the principal value propositions for simulation has been the insight to the practitioner obtained by studying the dynamics of the system under study. Often, the most powerful insights have been captured via the animation feature that has become a popular component of virtually all packages. For users that rely heavily on animation, the numerical computation aspect of the simulation engine may be of secondary importance. It follows that although simulation packages can be viewed as sampling-based computational tools for computing performance characteristics of importance, this point of view often does not resonate well within the simulation practitioner community. The fact that simulation has value to the user that goes well beyond its value as a numerical computing platform differentiates simulation from numerical optimization, where the user insights are obtained almost exclusively from the numerical results computed by the optimization routines. Consequently, numerical optimization has evolved as an important subdiscipline of the numerical analysis community, and the quality of the numerics has played a central role in the user community’s preferences with regard to the purchasing of optimization packages. On the other hand, the quality of the numerics has often played a secondary role in purchasing decisions for simulation software, and other characteristics like animation, model-building/model re-use features, and associated report writing software have frequently dominated the user’s selection. Thus, one could argue that market forces have historically not imposed upon commercial vendors a requirement to incorporate the potentially sophisticated numerical algorithms necessary to successfully solve complex simulation-related optimization problems. Of course, market forces change over time. A principal use of simulation historically has centered on its ability to answer “what if” type questions. Often, “what if” questions are intended to explore the possibility of obtaining better solutions than the system design currently implemented or under consideration. Optimization routines offer a systematic means of obtaining improved solutions and present the possibility of actually computing solutions that are optimal according to some user-defined criterion. Thus, the recent addition of optimization routines to simulation software represents a natural evolution of such packages. Perhaps more importantly, as features like animation become commonplace, optimization routines become a potentially key product differentiator relative to competing packages in the simulation marketplace. Finally, the last decade or so has seen the emergence of certain applications areas (such as computational finance) in which the speed and quality of the internal numerics often plays a dominant role in user perceptions of value. As argued above, user demand for high-quality numerics can have a positive effect on inducing vendors to incorporate sophisticated simulation-based optimization algorithms.