[Paper Review] Estimation of initial conditions from a scalar time series

Anil Maybhate, R. E. Amritkar|ArXiv.org|Feb 17, 2000

Time Series Analysis and Forecasting1 references3 citations

TL;DR

This paper presents a novel Newton-Raphson-based method to estimate the full initial conditions of a multivariable dynamical system from a single scalar time series, leveraging time evolution in a modified iterative scheme. The method achieves fast, quadratic convergence even when the conditional Lyapunov exponent is positive, enabling instantaneous synchronization of chaotic systems via identical initial states derived from scalar data.

ABSTRACT

We introduce a method to estimate the initial conditions of a mutivariable dynamical system from a scalar signal. The method is based on a modified multidimensional Newton-Raphson method which includes the time evolution of the system. The method can estimate initial conditions of periodic and chaotic systems and the required length of scalar signal is very small. Also, the method works even when the conditional Lyapunov exponent is positive. An important application of our method is that synchronization of two chaotic signals using a scalar signal becomes trivial and instantaneous.

Motivation & Objective

• To address the challenge of reconstructing full initial conditions of a multivariable dynamical system from a single scalar time series of one state variable.
• To develop a method that works reliably even when the conditional Lyapunov exponent is positive, a regime where existing methods often fail.
• To enable trivial and instantaneous synchronization of chaotic systems using only a scalar signal by accurately estimating the initial state.
• To demonstrate that minimal-length time series—shorter than standard embedding delay times—can suffice for accurate initial condition recovery.

Proposed method

• The method uses a modified multidimensional Newton-Raphson approach that incorporates the time evolution of the system via a sequence of time-step updates.
• It defines the difference vector w^n = y^n - x^n between a trial trajectory y^n and the true trajectory x^n, seeking w^n = 0.
• The evolution of the difference vector is modeled using a matrix update: W^n = A^{n-1} A^{n-2} ... A^0 W^0, where A^n = I + Δt J^n and J^n is the Jacobian of the system's vector field at y^n.
• The method iteratively corrects the initial guess y^0 using the gradient of the system's dynamics, with corrections derived from the time-advanced error vector.
• Convergence is assessed via error norms e_i = ||x^0 - y^0||_i, and the convergence rate is quantified using the parameter α in e_{i,n+1} ≈ α e_{i,n}^2.
• The approach is applied to chaotic systems like the Rössler and Chua’s circuits, and to periodic and intermittent regimes, with consistent performance across diverse dynamical behaviors.

Experimental results

Research questions

• RQ1 Can the full initial state vector of a multivariable dynamical system be accurately reconstructed from a scalar time series of a single component?
• RQ2 Does the method remain effective when the conditional Lyapunov exponent is positive, a condition that typically hinders synchronization?
• RQ3 How small can the required time series length be for reliable initial condition estimation, especially compared to standard embedding techniques?
• RQ4 To what extent does the method’s convergence speed and robustness depend on the system’s dynamical regime (periodic, chaotic, intermittent)?

Key findings

• The method achieves quadratic convergence, with convergence parameters α ≈ 1.97 for e₁ and 1.95 for e₃ when estimating initial conditions from x₁ time series in the Rössler system.
• For time series of x₂, the convergence rate remains quadratic (α ≈ 1.97 for e₁, 1.95 for e₃), even though the conditional Lyapunov exponent is positive.
• When using x₃ time series, convergence is slower than quadratic (α ≈ 1.28 for e₁, 1.30 for e₂), but the method still converges reliably.
• In the Chua’s circuit with a limit cycle, convergence is slower than quadratic (α ≈ 1.29 for e₂, 1.25 for e₃), yet the method successfully estimates the initial state.
• The method works across diverse systems, including the Lorenz system, disk dynamo, three-wave plasma coupling, and four-dimensional phase converter circuits, under periodic, chaotic, and intermittent regimes.
• The method enables instantaneous synchronization of identical chaotic systems by setting the replica’s initial state equal to the estimated initial state from the scalar signal, even when the conditional Lyapunov exponent is positive.