Detecting causalities between strongly coupled dynamical systems

Abstract

A new version of the convergent cross mapping is proposed to detect causalities from records for strongly coupled nonlinear dynamical systems, where the mutual entropy is used to measure nonlinear correlations, and the time delay stability is adopted to filter out false identifications. Calculations on various deterministic dynamic systems show that it is applicable not only to strongly coupled systems but also to non-interacting systems influenced by a common environment. Compared with the original version of convergent cross mapping, under strong couplings our proposed method has significantly higher accuracy, and is more robust to coupling strength. As a typical example, it is used to detect the causal effects between arterial blood pressure (ABP) and intracranial pressure (ICP) of patients diagnosed with traumatic brain injury (TBI). A mono-directional causality from ICP to ABP is identified.