Additive Nonparametric Reconstruction of Dynamical Systems from Time Series.pdfVIP

Additive Nonparametric Reconstruction of Dynamical Systems from Time Series Markus Abel,∗ Karsten Ahnert, J¨urgen Kurths, and Simon Mandelj Institute of Physics, University of Potsdam, 14415 Potsdam, Germany (Dated: February 8, 2008) We present a nonparametric way to retrieve an additive system of differential equations in embed- ding space from a single time series. These equations can be treated with dynamical systems theory and allow for long term predictions. We apply our method to a modified chaotic Chua oscillator in order to demonstrate its potential. 4 PACS numbers: 05.45.Tp, 47.52.+j 0 0 2 Casting physical observations into mathematical equa- exists an attractor A ⊂ Rn with the box-counting dimen- v tions is one of the fundamental tasks to understand and sion d ≤ n. In [2] it has been shown that almost every o predict dynamical systems. Basically, there are two com- smooth map Ψ : Rn → Rm , m 2d, is an embedding, i.e. N plementary approaches to accomplish this task: theoret- a smooth diffeomorphism from A onto its image Ψ(A). ically by convenient considerations and empirically by The condition m 2d is sufficient, therefore cases with 0 data analysis. Both approaches are essential for mod- d m 2d can occur. 3 ern modeling strategies. If For many systems, the dy- Due to differentiability, the dynamics of ξ (t)