Abstract

Within S-O models, one employs substitutes for exact embedded simulators either to improve computational efficiency or when exact simulators are not available. The modeler must decide whether a particular S-O approach is suitable, and how best to employ it. One selects a substitute (surrogate) simulator based on practicality, computational efficiency, and accuracy. Often, increasing accuracy requires additional computational effort. S-O models have convergence criteria conceptually similar to those of S models. However, S-O models must converge both to a suitably accurate simulation prediction and to an optimal solution.