A computational model for complex systems analysis: Causality estimation

Abstract Analysis of complex systems has migrated from theoretical analysis of model physical systems into the domain of real-world applications with far-reaching implications. This shift has been facilitated by the availability of measurement data, the capability to store and manage large-scale datasets, and significant increase in computational power. These developments have been matched by a concerted effort in the development of new methods of complex systems analysis. Each class of analytical methods are applicable only for certain system types and specific problems, such as estimating causality between coupled complex systems, are still unresolved even for simples cases involving only two connected systems. This study summarizes results for estimating causality using a novel mathematical computational model, System Structure, across several different types of coupled systems. System structure models a system as a dynamic network of connected components and provides a distribution-free and systematic quantification of inter-component dynamics as a basis for making several system-wide inferences including the inference of strength and direction of causal relations. System structure is a non-parametric scalable method that requires few non-restrictive a-priori assumptions, making it generally applicable across a diverse set of problems. The comparative study reported on here demonstrates the ability of system structure to correctly infer different types of causality across diverse system types.