Sensitive Dependence on Initial Conditions

5 min read#chaos

The hallmark of chaos — infinitesimally close starting points diverge exponentially fast, making long-term prediction impossible.

Contents

Sensitive Dependence on Initial Conditions

Sensitive dependence on initial conditions is the defining symptom of chaos: take two states as close together as you like, let the same deterministic rule run, and watch them peel apart at an exponential rate. Because no measurement is ever infinitely precise, that tiny initial gap is inevitable — and chaos blows it up until the prediction is worthless. This is the famous butterfly effect.

The name comes from Edward Lorenz, who in 1972 asked: "Does the flap of a butterfly's wings in Brazil set off a tornado in Texas?" The point is not that butterflies cause tornadoes, but that in a chaotic atmosphere a perturbation that small can grow to dominate the weeks-later weather. Lorenz stumbled onto the effect by accident, restarting a weather simulation from a printout rounded to three digits and watching it diverge completely from the original run.

Exponential divergence

Let \delta_0 be the initial separation between two trajectories and \delta(t) their separation at time t. In a chaotic system the gap grows, on average, like

|\delta(t)| \approx |\delta_0|\, e^{\lambda t},

where \lambda > 0 is the leading Lyapunov Exponent. A positive \lambda is the mathematical signature of sensitive dependence — it measures how fast information about the initial condition is destroyed.

\lambda = \lim_{t\to\infty}\ \lim_{|\delta_0|\to 0}\ \frac{1}{t}\,\ln\frac{|\delta(t)|}{|\delta_0|}

The exponential is brutal. If \lambda gives a tenfold growth per unit time, then improving your initial measurement by a factor of a billion buys only nine extra units of forecast horizon. This is why weather forecasts decay past a week or two no matter how good the instruments get.

Watch two trajectories diverge

The sketch below launches two pendulum-like trajectories whose starting angles differ by just one part in a thousand. For a while they track each other perfectly — then sensitive dependence takes over and they go their separate ways.

Two trajectories of a driven nonlinear oscillator starting 0.001 apart. The faint vertical line marks the moment their separation first exceeds 0.5 — the practical 'prediction horizon'.

The two curves are generated by identical equations; the only difference is the fourth decimal place of the starting angle. That is sensitive dependence in one picture.

Predictable in principle, not in practice

Sensitive dependence forces a careful distinction. A chaotic system remains fully deterministic — the future is uniquely fixed by the present. What collapses is predictability, because predicting requires knowing the present exactly, and the real world only ever hands us approximations. Chaos converts that gap between exact and approximate into runaway error.

Why can't we forecast a chaotic system far into the future even though its rule is perfectly deterministic?

The same mechanism is what carves chaotic motion onto a Strange Attractor: state space is repeatedly stretched (so neighbors separate) and folded back (so the motion stays bounded). Stretching is sensitive dependence; folding is what keeps the Double Pendulum from flying off to infinity.

See also