OpenWorldLab

Loading interactive view…

Chaos

The double pendulum: deterministic, and still unpredictable

Fifty pendulums, released together, each a billionth of a radian further out than the last. For a long while they move as one. Then, within a few swings, they are everywhere. Nothing random happens at any point: every one of them obeys the same exact equations.

What you are looking at

Each pendulum is two point masses on light rigid rods, the second hanging from the first, swinging without friction (or three or four, if you change Arms). All of them start at rest, at the angles in the controls panel, except that each upper arm starts slightly further round than the one before. The front pendulum is the one you can drag; the colours run through the rest.

The simulation steps the exact equations of motion forward 400 times per simulated second, using the fourth-order Runge–Kutta method. Over twenty seconds the double pendulum’s total energy changes by less than one part in a million, and a quadruple pendulum falling from upright stays within one part in 10,000. So the divergence you see comes from the physics, not from rounding errors building up.

Sensitive dependence

The chart at the bottom left plots the gap between the first two pendulums on a logarithmic scale, where each equal step up the axis multiplies the gap by the same factor. From the Raised high start the line climbs straight until the gap is as large as it can get. A straight line on a log scale means exponential growth: the gap multiplies by a fixed factor every second. The slope is the Growth readout, about 1 per second here, so the gap grows roughly e ≈ 2.7 times each second. That rate estimates the largest Lyapunov exponent of the motion, the rate at which nearby starts separate.

That growth is why a billionth of a radian takes about eighteen seconds to show, and a millionth about eleven: making the starting gap a thousand times smaller gains only seven seconds. To predict the motion twice as far ahead, you would need to know the starting position tens of millions of times more precisely. Edward Lorenz ran into the same limit with weather models in the 1960s, and it is why forecasts have a horizon however good the models become.

Phase space

At any instant a double pendulum is fully described by four numbers: the angle of each arm and how fast each is turning. Together they mark a point in its phase space, the space of every possible state, and as the pendulum moves, the point moves. Turn on Phase space in the controls to see a slice of it for one arm: each pendulum is a dot, with that arm’s angle across and its angular velocity (how fast it turns) up.

At release, all the dots sit on one spot and trace the same path. When the pendulums start to part, the spot stretches into a thread. The thread keeps stretching and folds back on itself, and soon it is a cloud spread over the whole panel. Chaos works by this stretching and folding, over and over, like a baker kneading dough. Without friction, a patch of starting states keeps the same total volume in the full space of states (Liouville’s theorem, with the volume measured using the arms’ momenta rather than their speeds), but the motion draws it out ever thinner in some directions while squeezing it in others. The panel shows only one arm’s share of that, so the cloud on it can still spread wide. From a Gentle swing the dots stay together and keep looping round the middle. When an arm goes right over the top, its angle passes ±180° and its dot reappears at the other edge.

Chaos needs energy

Choose Gentle swing and the pendulums stay together indefinitely. With little energy the motion is quasi-periodic, a blend of two steady rhythms that never quite repeats, and small differences stay small. Choose Normal mode and, for small swings, the motion repeats exactly: with equal masses and rods, starting the lower arm √2 times further out than the upper makes both swing at one frequency ω, given by ω² = (2 − √2) g / l, where g is the strength of gravity and l the length of each rod. Most starts with enough energy to throw the lower arm over the top are chaotic.

More arms

A triple or quadruple pendulum obeys the same kind of equations, with more terms. Each arm’s acceleration depends on all the others (through what physicists call the chain’s mass matrix), so at every step the simulation solves a small set of simultaneous linear equations. More arms give energy more ways to slosh between them, and usually faster divergence. Starting from Nearly upright, a billionth of a radian shows after about twenty seconds with two arms, six with three and three with four.

A chain of n arms has n normal modes: patterns in which, for small swings, every arm moves at the same frequency. Normal mode starts the arms in the pattern of the slowest one, worked out from the small-swing equations, so a triple or quadruple pendulum swings back and forth at a single frequency, as the double does.

Things to try

  • Set the gap between neighbours to 10⁻¹² and then to 10⁻⁶. The Growth readout barely changes, because the rate belongs to the motion, not to the starting gap. (Start them much further apart and the gap fills the chart before there is enough of a straight line to fit.)
  • Drag the lower bob of the front pendulum straight up from a hanging start, and compare how long the pendulums stay together with the upper arm low and high.
  • Use one pendulum with trails on and slow motion to see the lower arm flip over the top.
  • Turn on Phase space and watch from the release of Raised high: the single dot draws out into a thread at about the moment the pendulums visibly part.
  • Choose Nearly upright, then switch Arms from Double to Quadruple and watch how much sooner the fan opens.

References