MATHEMATICAL COMPARISON OF COMBAT COMPUTER-MODELS TO EXERCISE DATA

The powerful techniques of modern nonlinear statistical mechanics are used to compare battalion scale combat computer models (including simulations and wargames) to exercise data. This is necessary if large-scale combat computer models are to be extrapolated with confidence to develop battle-management, C3 and procurement decision-aids, and to improve training. This modeling approach to battalion level missions is amenable to reasonable algebraic and/or heuristic approximations to drive higher-echelon computer models. Each data set is fit to several candidate short-time probability distributions, using methods of "very fast simulated re-annealing" with a Lagrangian (time-dependent algebraic cost-function) derived from nonlinear stochastic rate equations. These candidate mathematical models are further tested by using path-integral numerical techniques we have developed to calculate long-time probability distribution spanning the combat scenario. We have demonstrated proofs of principle, that battalion level combat exercises can be well represented by the computer simulation JANUS(T), and that modern methods of nonlinear nonequilibrium statistical mechanics can well model these systems. Since only relatively simple drifts and diffusions were required, in larger systems, e.g., at brigade and division levels, it might be possible to "absorb" other important variables (C3, human factors, logistics, etc.) into more nonlinear mathematical forms. Otherwise, this battalion level model should be supplemented with a "tree" of branches corresponding to estimated values of these variables.