Cluster-weighted modeling: estimation of the Lyapunov spectrum in driven systems

Abstract

Cluster-weighted modeling based techniques are shown to be accurate, efficient, and robust in application to the problem of computing the Lyapunov spectrum from time-series data. We develop a method that is appropriate for application to driven nonlinear dynamical systems and show, in particular, that it is possible to estimate both global and local Lyapunov exponents through this technique. For dynamics on strange nonchaotic attractors, the present approach correctly determines a largest Lyapunov exponent that is negative.