Adaptive simulation model configuration

Models of complex systems may be improved in fidelity, efficiency or both by allowing the way they represent component parts to change as the state of the model and its components move through their state-spaces. A simple model demonstrates that representation changes for a population encountering contaminants may perform better than either conventional form. Here, there was a substantial decrease in runtime relative to a purely i-state configuration (individual-based) model, with comparable fidelity, while the purely p-state (population-based) version exhibited error arising from the "blurring" of contaminant contact through the distributed population. This example demonstrates the utility of model representations maintain important state data across representations so that previous representations can be recovered with minimal error. Triggers for changing representations of components is addressed in a paper exploring a possible set of dynamics associated with a simple, hypothetical model of a seven component ecosystem. The components of the system are described as i-state configuration models or as p-state models, and a mechanism for determining when to change representations is outlined. To support the analysis and selection of representations, a metric-space with the properties of a commutative ring is defined. The elements of the metricspace are trees that can encode the structural character of a set of submodels which comprise the model of a system, and to provide a metric in analysis. Finally, a framework is developed with an example model that closely follows the hypothetical example. It was designed as a reference model, and is freely available on https://github.com/snarkypenguin/Mutans.git This implementation demonstrates dynamic configuration management, maintenance of the states of superceded submodel representations, and the support structures needed to implement models of this type.