Lightening the Performance Burden of Individual-Based Models through Dimensional Analysis and Scale Modeling

摘要

The System Dynamics community is increasingly applying individual-based models for insight. Such models offer particular value for investigating the effects of targeted interventions and for studying systems with populations exhibiting high heterogeneity, complex and dynamic network structures. Unfortunately, simulation of such models for large populations is often extremely expensive. While it is desirable to gain insight into the behavior of models using reduced-scale populations, naive construction of such reduced-size models can yield erroneous conclusions. Within this paper, we have described a rigorous, systematic and general technique for building such models.While the described technique requires further development to address the full richness of modern modeling practice, we believe that it has considerable potential.More broadly, we believe that dimensional analysis offers many inviting avenues for future research. Specifically, we believe there are high benefits to be gained by applying dimensional scaling theory to model analysis. We believe these benefits are likely to be particularly substantial for individual-based models which are less tractable to closed-form analysis. We believe that the concepts and associated tools of incomplete similitude and intermediate asymptotics hold particular promise for further simplification of dimensional reasoning for individual-based systems. By building on dimensional approaches directly confronting the issue of multi-scale analysis and approximation, renormalization group theory may also be of great value.