Main menuGallery

Double Pendulum & Chaos

This exhibit needs WebGL, which your browser or device has turned off.
Try a different browser, or switch hardware acceleration on, to watch the double pendulum.

Double pendulum & chaos

Pendulum A Pendulum B
Time0.00 s
Start difference0.010°
Angle gap now0.010°
Lyapunov λ
Energy of A0.000 J
Energy drift0 %
Both pendulums obey exactly the same equations. Only the starting angle of the first rod differs, by the amount above.

What's the maths?

Where's the maths? A pendulum is trigonometry set in motion: the height of a bob is L cos θ and the sideways pull that swings it back is proportional to sin θ, so the whole machine is built out of the sine and cosine of Sec 2. Turning that into a rule for how θ changes from one instant to the next needs calculus — a differential equation, which is simply a statement about a rate of change. A single pendulum obeys its rule tamely and repeats itself forever, which is why it can run a clock. Hang a second rod from the bottom of the first and the same honest trigonometry produces something new: motion that never repeats. The branch of mathematics that studies this is dynamical systems, better known as chaos theory. Nothing on this screen is random — the computer solves identical equations for both pendulums to about fifteen digits — yet a difference of one hundredth of a degree at the start grows into a difference of a whole swing in under twenty seconds. That is exactly why a weather forecast is useful for a week and useless for a month.

📘 On the Sec 1–4 syllabus
  • Sec 2 · G4 Trigonometry — 4.3 trigonometric ratios (resolving the pendulum)
  • Sec 3/4 (A-Math) · G1 — trigonometric functions as models
  • Sec 3/4 (A-Math) · C1 — derivative as rate of change

Chaos and differential equations are beyond the syllabus.

The state of a double pendulum is just four numbers: the two angles θ₁, θ₂ and the two turning speeds ω₁, ω₂. Newton's laws turn into two equations giving the angular accelerations in terms of those four numbers and nothing else — no luck, no dice, no hidden input. The computer takes tiny steps of 1/1200 of a second and uses the Runge–Kutta method to follow them.

Start two copies a hair apart and the gap between them does not creep up — it multiplies:

|Δθ(t)| ≈ |Δθ(0)| · eλt
  • λ is the Lyapunov exponent. Here λ ≈ 0.4 per second, so the gap multiplies by e ≈ 2.7 roughly every 2½ seconds — about ten times bigger every 6 seconds.
  • That is why the graph plots log₁₀ of the gap: taking logarithms turns multiplying into adding, so exponential growth becomes a straight line. The fitted slope, times ln 10, is λ.
  • Because a starting gap of 0.01° ≈ 10−4.8 rad has to climb only about 5 decades to reach a whole radian, a hundredfold better measurement buys you just 2 more decades — about 12 extra seconds. Precision loses to exponentials.
  • The energy readout is the honesty check. Gravity and the rods do no net work, so the total energy must stay fixed; ours drifts by less than a millionth of a percent in a minute, so the wild motion is physics, not numerical rubbish.
  • Switch to Single pendulum and the same two starts stay locked together: one rod is regular, and for small swings its period is T = 2π√(L/g) — no mass, no amplitude, just the length. (Release it from 90° and the readouts show the real period running about 18 % long: "small angle" really does mean small.) One extra joint is the whole difference between predictable and unpredictable.
Try this: leave the difference at 0.01°, press Release and guess when the two pendulums will visibly part company. Now halve it to 0.005° and watch again — does the parting take twice as long, or only about 1½ seconds longer? Then drop the release angle to 25° and see the chaos switch off completely.
Drag to orbit · scroll or pinch to zoom · Space plays and pauses · R resets.

Divergence

log₁₀ of the gap between A and B against time. A straight line means exponential growth; the dashed fit gives λ = —.
0.010° 90° 1.0× View