This Note examines the differences in combat outcomes predicted by models of different resolution applied to identical combat situations. First, hypothetical combat situations are posed, then several models of varying degrees of resolution in the spatial representation, aggregation of forces, and time step are used to predict losses and battle winners. Both stochastic and deterministic simulations are used. Comparison of outcomes provides important insights into the problems of aggregation. Observations from this set of experiments are as follows: Intuition regarding outcomes, causes, and effects is frequently wrong, leading to bad approximations in the aggregate. Scaling for different levels of resolution is possible, but a method of predicting the appropriate scaling technique and factors has not been found. The differences in outcomes between stochastic and deterministic models are most pronounced in the "fair-fight" regime, in which the force balance (accounting for situational factors) is almost even. Because defense analysis frequently operates in this regime (getting "just enough" force to a theater or because constrained defense budget allocations may not permit overwhelming odds), this implies that great care should be taken to understand the possible variance in outcomes.

This Note examines the differences in combat outcomes predicted by models of different resolution applied to identical combat situations. First, hypothetical combat situations are posed, then several models of varying degrees of resolution in the spatial representation, aggregation of forces, and time step are used to predict losses and battle winners. Both stochastic and deterministic simulations are used. Comparison of outcomes provides important insights into the problems of aggregation. Observations from this set of experiments are as follows. Intuition regarding outcomes, causes, and effects is frequently wrong, leading to bad approximations in the aggregate. Scaling for different levels of resolution is possible, but a method of predicting the appropriate scaling technique and factors has not been found. The differences in outcomes between stochastic and deterministic models are most pronounced in the 'fair-fight' regime, in which the force balance (accounting for situational factors) is almost even. Because defense analysis frequently operates in this regime (getting just enough force to a theater or because constrained defense budget allocations may not permit overwhelming odds), this implies that great care should be taken to understand the possible variance in outcomes.

Military conflicts, particularly land combat, possess the characteristics of complex adaptive systems: combat forces are composed of a large number of nonlinearly interacting parts and are organized in a dynamic command-and-control network; local action, which often appears disordered, self-organizes into long-range order; military conflicts, by their nature, proceed far from equilibrium; military forces adapt to a changing combat environment; and there is no master “voice” that dictates the actions of every soldier (i.e., battlefield action is decentralized). Nonetheless, most modern “state of the art” military simulations ignore the self-organizing properties of combat.This book summarizes the results of a multiyear research effort aimed at exploring the applicability of complex adaptive systems theory to the study of warfare, and introduces a sophisticated multiagent-based simulation of combat called EINSTein. EINSTein, whose bottom-up, generative approach to modeling combat stands in stark contrast to the top-down or reductionist philosophy that still underlies most conventional military models, is designed to illustrate how many aspects of land combat may be understood as self-organized, emergent phenomena. Used worldwide by the military operations research community, EINSTein has pioneered the simulation of combat on a small to medium scale by using autonomous agents to model individual behaviors and personalities rather than hardware.

Operations Research (OR) emerged in an effort to improve the effectiveness of newly inducted weapons and equipment during World War II. While rapid growth ofOR led to its becoming an important aid to decision making in all sectors including defense, its contribution in defense remained largely confined to classified reports. Very few books dealing with applications of quantitative decision making techniques in military have been published presumably due to limited availability ofrelevant information. The situation changed rapidly during the last few years. The recognition of the subject of Military Operations Research (MOR) gave tremendous boost to its development. Books and journals on MOR started appearing. The number of sessions on MOR at national and international conferences also registered an increase. The volume of teaching, training and research activities in the field of MOR at military schools and non-military schools enhanced considerably. Military executives and commanders started taking increasing interest in getting scientific answers to questions pertaining to weapon acquisition, threat perception and quantification, assessment of damage or casualties, evaluation of chance of winning a battle, force mix, deployment and targeting of weapons against enemy targets, war games and scenario evaluation. Most of these problems were being tackled on the basis of intuition, judgment and experience or analysis under very simple assumptions. In an increasingly sophisticated and complex defense scenario resulting in advances in equipment and communications, the need for supplementing these practices by scientific research in MOR became imperative.

This study analyzes an original hybrid combat simulation for possible use as the underlying support model for the Joint Warfare Systems (JWARS) analytical simulation. The model employs a fixed increment time advance mechanism but represents individual entities vice aggregated units. Results from an otherwise identical model using a next event time advance mechanism provide a baseline for comparison. The hybrid, using a longer time increment, runs faster than the next event model but produces unacceptable results. The hybrid, using a smaller time increment, more closely approximates the next event model but takes longer to run.