Two double pendulums start a millionth of a radian apart. The laws are exact and nothing is random, yet within seconds they swing as if they had never met. Watch the gap between them grow tenfold, again and again.
Each tenfold improvement in the starting difference buys less than two extra seconds of agreement. Shrinking it a millionfold barely delays the split.
Each pendulum is two equal one-metre arms hinged together, swinging under real gravity with no friction. Their motion is computed with the same exact equations and the same fourth-order Runge-Kutta steps. The only difference is the starting angle of the upper arm, which differs by the amount on the slider, spread evenly across the pendulums.
The right panel tracks how far apart the first and last pendulum are, counting both angles and both spin rates. The vertical scale is logarithmic, so steady exponential growth shows up as a straight climb.
Nothing here is random. Give a computer the same numbers and it repeats the run exactly. But any error in the starting state, however small, is multiplied by about the same factor every second. Predicting further ahead needs exponentially better measurements, so in practice the future is closed even though the laws are known.
Poincaré saw this in the three-body problem around 1890. Edward Lorenz rediscovered it in a weather model in 1961 and asked in 1972 whether a butterfly in Brazil could set off a tornado in Texas.
Chaos sets hard limits on prediction, but knowing where those limits are, and how to exploit tiny nudges, turns it into a tool.
In 1961 Lorenz restarted a weather simulation from numbers rounded to three decimals and got a completely different forecast after a few simulated weeks. Small errors in today's observations grow the same way in real weather, which caps useful day-to-day forecasts at roughly ten to fourteen days.
In the demo: a rounding-sized difference ends in a different pendulum.Weather centres such as ECMWF run their model around 50 times from slightly perturbed starting states. Where the runs agree, the forecast is confident. Where they fan out, forecasters give probabilities instead, like a 40% chance of rain.
In the demo: 100 fan is an ensemble; its spread is the uncertainty.Near unstable points between the Sun, Earth and Moon, a tiny burn produces a large change in path. In 1983 NASA used a series of lunar flybys to send ISEE-3 out of its orbit near the Sun-Earth L1 point to meet comet Giacobini-Zinner, renamed ICE. Japan's Hiten reached the Moon in 1991 on a similar low-energy route.
In the demo: sensitivity works both ways; a tiny nudge buys a big change.The electrical waves that coordinate a heartbeat can break into spiral waves and then disordered activity, which is what happens in ventricular fibrillation. Researchers study arrhythmias as nonlinear dynamics, and experiments have used small, well-timed stimuli to steer irregular rhythms back toward regular ones.
In the demo: low angles stay regular; push the energy up and order breaks down.Stirring works because chaotic flow stretches and folds fluid, pulling neighbouring drops apart exponentially fast. Engineers design static mixers, microfluidic chips and industrial stirrers to produce chaotic advection, which blends liquids far faster than diffusion alone.
In the demo: the 100 pendulums start as one line and end up spread everywhere.A chaotic system turns a secret starting value into a stream that looks random and changes completely if the key changes slightly. Researchers have used this for image ciphers and for synchronised chaotic lasers that carried hidden data over a commercial fibre network in Athens in 2005.
In the demo: change the key by 1e-9 and the output becomes unrecognisable.