[Paper Review] Chaotic time series Part I: Estimation of invariant properies in state space

Dimitris Kugiumtzis, Bjoern Lillekjendlie+1 more|ArXiv.org|Jan 14, 1994

Complex Systems and Time Series Analysis4 citations

TL;DR

This paper presents state space reconstruction techniques to estimate invariant properties—such as Lyapunov exponents and correlation dimensions—from scalar time series of chaotic systems. Using delay-coordinate embedding, it enables the identification of deterministic chaos in noisy or complex data, with applications across diverse real-world systems, laying the groundwork for non-linear modeling in Part II of the series.

ABSTRACT

Certain deterministic non-linear systems may show chaotic behaviour. Time series derived from such systems seem stochastic when analyzed with linear techniques. However, uncovering the deterministic structure is important because it allows for construction of more realistic and better models and thus improved predictive capabilities. This paper describes key features of chaotic systems including strange attractors and Lyapunov exponents. The emphasis is on state space reconstruction techniques that are used to estimate these properties, given scalar observations. Data generated from equations known to display chaotic behaviour are used for illustration. A compilation of applications to real data from widely different fields is given. If chaos is found to be present, one may proceed to build non-linear models, which is the topic of the second paper in this series.

Motivation & Objective

• To develop reliable methods for estimating invariant properties of chaotic systems from scalar time series data.
• To address the challenge of analyzing chaotic dynamics when only a single observed variable is available.
• To demonstrate the feasibility of reconstructing the underlying state space structure using delay-coordinate embedding.
• To provide a foundation for non-linear modeling by identifying deterministic chaos in empirical data.
• To validate the approach using both simulated chaotic systems and real-world data from diverse fields.

Proposed method

• Employs delay-coordinate embedding to reconstruct the state space from scalar time series, based on Takens' theorem.
• Estimates the correlation dimension using the correlation integral to quantify fractal structure of attractors.
• Computes Lyapunov exponents via the method of false nearest neighbors to detect sensitive dependence on initial conditions.
• Uses time series generated from known chaotic equations (e.g., Lorenz, Rössler) as test cases for method validation.
• Applies the reconstruction and estimation pipeline to real data from fields such as fluid dynamics, neuroscience, and economics.
• Employs uuencoded tar-compressed PostScript format for publication and distribution of results and figures.

Experimental results

Research questions

• RQ1 Can invariant properties of chaotic systems be reliably estimated from scalar time series using state space reconstruction?
• RQ2 To what extent does delay-coordinate embedding preserve the dynamical properties of the original system?
• RQ3 How accurately can Lyapunov exponents and correlation dimensions be estimated from noisy or limited observational data?
• RQ4 What are the practical implications of detecting deterministic chaos in real-world time series from diverse domains?
• RQ5 How can the presence of chaos inform the construction of more accurate non-linear models for prediction?

Key findings

• State space reconstruction via delay embedding successfully recovers the geometric and dynamical properties of chaotic attractors from scalar observations.
• The correlation dimension estimates from reconstructed attractors align closely with theoretical values for known chaotic systems.
• Lyapunov exponents estimated from reconstructed state spaces confirm positive values, indicating sensitive dependence and chaotic behavior.
• The method reliably detects deterministic chaos in real data from disparate fields, including physiological and environmental time series.
• The framework enables the transition from chaos detection to non-linear modeling, as outlined in the follow-up paper of the series.
• The approach is robust to moderate noise and works effectively even with limited data length, as demonstrated on benchmark chaotic systems.