Quantifying local predictability in phase space (Q914223)

Traditionally in the analysis of chaotic dynamics, the largest Lyapunov exponent defines the predictability time scale of the considered dynamical system, by quantifying the average rate at which initial uncertainty is propagated over a strange attractor. However the rate at which adjacent trajectories diverge on such an attractor (local divergence rate) depends on time, and on location in the system phase space. Thus, it appears that the use of the classical Lyapunov exponents to measure predictability is inadequate for applications related to short-term prediction. This situation is due to the elimination of phase-spatially dependent predictability information by the averaging over the attractor, which results from the Lyapunov exponent definition. In this paper the local predictability of the three-dimensional Lorenz equation is studied, from both temporal and phase-spatial points of view. It is shown how the local divergence rates might be used to identify four regions of the phase space with: high predictability, predictability, unpredictability, high unpredictability. Standard statistical analysis techniques, Poincaré maps, and phase- spatial averaging are the tools used to quantify local predictability.