Lyapunov Exponent

3 min read#foundations

The average rate at which nearby trajectories pull apart — the quantitative fingerprint of chaos.

Contents

Lyapunov Exponent

The Lyapunov exponent \lambda measures how fast two trajectories that start infinitesimally close together separate as time passes. If their initial gap is \delta_0, then on average it grows (or shrinks) like

|\delta(t)| \approx |\delta_0|\,e^{\lambda t}, \qquad \lambda = \lim_{t\to\infty}\frac{1}{t}\ln\frac{|\delta(t)|}{|\delta_0|}.
(1)

The sign of \lambda tells you everything about the system's predictability:

Exponential divergence, seen

The gap between two chaotic trajectories doesn't creep — it explodes. The plot shows the same initial error of 10^{-4} growing under a positive exponent. On a log scale it would be a straight line of slope \lambda; on this linear scale it looks like nothing, then suddenly everything.

A tiny error amplified by a positive Lyapunov exponent
An undetectable initial difference becomes order-1 in finite time.

The predictability horizon

A positive exponent imposes a hard limit on forecasting. If you know the state to precision \delta_0 and can tolerate error up to \Delta, your predictions stay useful only until the Lyapunov time t_h \approx \frac{1}{\lambda}\ln\frac{\Delta}{\delta_0}. Because the dependence on your initial precision is logarithmic, buying a thousand times better measurements only buys you a little more forecast time — the reason weather prediction hits a wall a couple of weeks out.

Many trajectories, one fate

Watch a fan of nearby starts under a chaotic flow. They march together convincingly — and then, at no particular moment, they don't.

Twelve trajectories of a chaotic map, started within a pixel of each other, losing all correlation.

See also