Chaos: the butterfly effect in a deterministic equation
A deterministic system is one where the present fixes the future completely: same start, same outcome, every time. Chaos is what happens when such a system also stretches every small difference until it is as big as the system itself. The future is fixed, and still unpredictable.
Why it matters
Before this was understood, the working assumption in physics was that small errors stay small: measure a bit better and you predict a bit further. Chaos breaks that bargain. In the Lorenz equations — three coupled rates written down in 1963 as a toy model of convecting air — an error in the sixth decimal place grows to fill the whole range of behaviour in a fixed, short time. Measuring ten times better buys you only a little more forecast, not ten times more. This is why weather forecasts have a horizon, and it is not a shortage of computers.
The mechanism
The Lorenz system tracks three numbers, x, y, z. Each has a rule for how fast it changes, and the rule depends on the current values of the others. With the classic settings (σ = 10, ρ = 28, β = 8/3) the state never settles down and never repeats. It loops around one of two lobes, then flips to the other, in a sequence that looks random but is computed exactly.
Two ingredients make this chaotic. First, stretching: near the centre, trajectories that start close are pulled apart, and the gap grows by roughly a constant factor per unit time. That is exponential growth, and it is the whole secret. A gap of 0.000001 that doubles every 0.7 time units is about 1 after 14 time units. Second, folding: the state cannot escape to infinity, so the stretched trajectories are folded back onto the same bounded region. Stretch and fold forever, and neighbours are shuffled without limit.
None of this uses randomness. Two copies of the system started from identical numbers stay identical forever. The unpredictability is about our knowledge of the starting point, which is never infinitely precise, not about the rule.
Interactive Set how close the two runs start, then press run and watch them separate.
In one breath
A deterministic rule can still stretch small differences exponentially. Bounded plus stretched means folded, and folding scrambles neighbours. So any finite measurement error, however small, becomes a total loss of prediction after a time that grows only slowly as the error shrinks.
Where this comes from
- Decomposing the Dynamics of the Lorenz 1963 model using Unstable Periodic Orbits: Averages, Transitions, and Quasi-Invariant Sets reuse permitted with attribution
arxiv.org/abs/2108.04181 - Quantifying the computability of the Lorenz system linked only, not reproduced
arxiv.org/abs/1306.2782 - A 3D Strange Attractor with a Distinctive Silhouette. The Butterfly Effect Revisited linked only, not reproduced
arxiv.org/abs/1311.6128