The Lorenz Waterwheel: A Bucket of Chaos You Can Build

A wheel that reproduces a weather model

In 1963 Edward Lorenz reduced a simplified model of atmospheric convection to three coupled ordinary differential equations and discovered, almost by accident, that they behave chaotically. A decade later, Willem Malkus and Lou Howard built something you can hold in your hands that obeys exactly the same mathematics: a ring of cups, each with a small hole in the bottom, mounted around a wheel that is free to rotate and tilted so gravity can drive it. A steady stream of water pours into the cup currently at the top. Depending on how fast the water comes in, the wheel either settles into a fixed spin, oscillates, or does something that looks nothing like clockwork at all.

From buckets to the Lorenz equations

Each cup gains mass from the inflow when it is near the top and loses mass through its leak at a rate proportional to how full it is. Malkus and Howard described the mass distribution around the ring by its Fourier coefficients and showed that, if you keep only the first harmonic, the wheel's angular velocity ω and the two leading coefficients a and b obey three coupled equations that are algebraically identical to Lorenz's, after a change of variables and constants:

Malkus waterwheel                    Lorenz system (1963)

dω/dt = −Kω + (πgr/I)·a         dx/dt = σ(y − x)
da/dt  = ωb − Ka                    dy/dt = x(ρ − z) − y
db/dt  = −ωa − Kb + q            dz/dt = xy − βz

ω ↔ x     a ↔ y     b ↔ z
inflow rate q plays the role of the Rayleigh number ρ

That correspondence is not a loose analogy — it is a genuine isomorphism between a mechanical system and a convection model, which is why the waterwheel is the textbook physical demonstration of the Lorenz attractor: turning the tap is literally turning the Rayleigh number, the same parameter that governs how vigorously a heated fluid convects.

Three regimes, one dial

Turn the inflow up from zero and the wheel passes through three qualitatively different behaviours. At low inflow, friction dominates and the wheel simply stays still or settles at a fixed tilt with no net rotation — the cups fill and drain in a static, symmetric pattern. Past a first critical inflow, the symmetric state becomes unstable and the wheel picks a direction and spins at a constant rate, exactly balancing gravity's torque from the lopsided water load against friction. Past a second, higher inflow, that steady rotation itself loses stability: the wheel accelerates, slows, sometimes stalls and swings the other way, and it never settles into a periodic pattern. That third regime is the chaotic one, and it is what the live demo above traces out in the abstract (x, y, z) space of the Lorenz equations — the same two-lobed, never-repeating "butterfly" shape, called a strange attractor.

Sensitive dependence, not randomness

Nothing about the waterwheel's equations is random — given exact initial conditions, its future is completely determined. The catch is sensitive dependence on initial conditions: in the chaotic regime, two runs started with a difference in cup fill level too small to ever measure diverge exponentially, doubling their separation on a timescale set by the largest Lyapunov exponent of the system. Predicting exactly when the wheel will next reverse direction therefore becomes impossible in practice within a few of those doubling times, even though the underlying dynamics are fully deterministic and the trajectory never leaves the bounded, fractal-structured attractor. This is the mechanical ancestor of Lorenz's own metaphor for the same phenomenon in the atmosphere — the butterfly effect.

Why it matters beyond the toy

The waterwheel earns its place in physics courses because it makes an abstract, seemingly exotic property of a weather model into something you can watch spin on a tabletop. It demonstrates, with a device simple enough to build from a bicycle wheel and paper cups, that determinism and unpredictability are not opposites — a system with no randomness anywhere in its rules can still make its long-term behaviour effectively unforecastable, which is exactly the situation atmospheric scientists face when they try to extend weather forecasts more than a couple of weeks ahead.

Frequently asked questions

Try it live

Everything above runs in your browser — open Lorenz Waterwheel and change the parameters while it is running. Nothing is installed, nothing is uploaded, the whole model lives in one tab.