TESTING STRATEGIES FOR SIMULATION OPTIMIZATION.

Computer simulation models consisting of systems of differential equations, or other mathematical models, can present special problems to numerical optimization methods. Derivatives are often unavailable, function evaluations can be extremely expensive (e. g. , 1 h on an IBM 3090), and the numerical accuracy of each function value may depend on a complicated chain of calculations and so be impractical to prespecify. This last point makes it difficult to calibrate optimization routines that use finite-difference approximations for gradients. A strategy for comparing optimization techniques for these problems is presented, and several interesting findings for quasi-Newton methods, simplex search, and others are reviewed.