Lyapunov exponent

A Lyapunov exponent quantifies the average rate of separation of infinitesimally close trajectories in a dynamical system. It is the mathematical signature of predictability: a negative exponent means nearby trajectories converge, indicating stability; a positive exponent means they diverge exponentially, indicating chaos. The magnitude of the largest Lyapunov exponent determines the predictability horizon — the time beyond which forecast error exceeds a given threshold.

In a system with n degrees of freedom, there are n Lyapunov exponents forming the Lyapunov spectrum, which characterizes the geometry of the system's attractor. Strange attractors in chaotic systems are distinguished by having at least one positive Lyapunov exponent while remaining globally bounded. The Kaplan-Yorke conjecture relates the Lyapunov spectrum to the fractal dimension of the attractor, linking dynamical instability to geometric complexity.

The Predictability Horizon

The largest Lyapunov exponent sets the fundamental limit on prediction in any dynamical system. If the exponent is λ, then two trajectories separated by an initial distance δ₀ diverge as δ(t) ≈ δ₀e^(λt). The predictability horizon — the time after which forecast error exceeds some threshold Δ — is approximately t_horizon ≈ (1/λ) ln(Δ/δ₀). This is not a practical limitation of our models; it is a structural limitation of the system itself. No model, however perfect, can predict beyond this horizon because the required precision in initial conditions grows exponentially with time.

In atmospheric models, the largest Lyapunov exponent corresponds to a predictability horizon of roughly two weeks — the famous limit beyond which weather forecasting becomes statistically indistinguishable from climatology. In neural systems, the Lyapunov spectrum characterizes the trade-off between sensitivity (needed to respond to stimuli) and stability (needed to maintain functional states).

Lyapunov Dimension and Entropy

The Lyapunov spectrum connects dynamical instability to geometric and information-theoretic quantities. The Kaplan-Yorke dimension approximates the fractal dimension of a strange attractor from the Lyapunov exponents, linking the rate of information loss (positive exponents) to the attractor's geometric complexity. A system with k positive exponents loses information at a rate equal to their sum — the Kolmogorov-Sinai entropy — quantifying the rate at which initial uncertainty is amplified into macroscopic unpredictability.

This connection reveals that chaos is not merely disorder but a specific form of information dynamics. A chaotic system does not destroy information; it encrypts it, spreading microscopic details across macroscopic scales faster than any observer can decode them. The Lyapunov exponents are the encryption rate.

The Lyapunov exponent is not just a number. It is a measure of how quickly the universe forgets its own initial conditions — and how quickly we, as observers, lose the ability to say what happens next.