EXPERIMENT 001 · MECHANICS

At the edge of chaos.

Nearly identical starts. Completely different paths.

Double pendulum
Starting the experiment…0.0 s
Motion of a family of double pendulums with colored trails
● first last ●5 pendulums
Add charts

Selected charts appear below and update as the pendulums move.

01 /

Why do the paths diverge?

The pendulums have identical lengths and masses, but their lower arms start at slightly different angles. In chaotic motion, even this tiny difference grows rapidly, making long-term prediction difficult. Reduce the angles and compare how the motion changes.

Idealized model
No friction · 1 kg masses
Gravitational acceleration: 9.81 m/s²
Theory behind the experimentWhy almost identical pendulums end up moving differentlyPlain explanation, equations and how the model works

In plain words

A double pendulum is two linked arms with weights. The upper arm hangs from a fixed pivot and the lower arm hangs from the end of the upper one. When the upper weight moves, it drags the lower pivot along, and the lower weight pulls back on the upper one. The two motions are coupled, so you can’t treat them as two separate ordinary pendulums.

The screen shows several independent double pendulums drawn on top of each other. They don’t collide or affect each other. All lengths and masses are the same; only the starting angle of the lower arm differs, and only slightly. The colours show what almost identical starting conditions turn into.

At first the pendulums move almost together. In the chaotic regime the small difference keeps growing: one is already flipping over while another swings back the other way. This is called sensitivity to initial conditions. Chaos here doesn’t mean there are no laws. Every next moment follows from the same equations of mechanics; a long-range forecast just needs the starting state known more and more precisely.

A double pendulum isn’t always chaotic. Pull it only slightly away from the bottom and the motion is usually much more regular. A tangled path on its own doesn’t prove chaos either, so compare regimes, not just how pretty the trails look.

Try it yourself

  1. Set the initial angle difference to zero. All coloured pendulums merge into one, because the model adds no random kicks.
  2. Bring back a small difference and compare large starting angles with small ones. Watch how quickly the motions drift apart. The exact time depends on the settings.
  3. Turn on the divergence and phase-path charts. Sometimes two weights are close together while moving in opposite directions. Same position doesn’t mean same state.

Physics and equations

The state of one pendulum is two angles and two angular velocities:

s=(θ1, θ˙1, θ2, θ˙2)\mathbf s=(\theta_1,\ \dot\theta_1,\ \theta_2,\ \dot\theta_2)

Here θ1\theta_1 and θ2\theta_2 are the angles of the upper and lower arm, both measured from the downward vertical (not one from the other). A dot means a time derivative, so θ˙\dot\theta is angular velocity. The equations use radians; the sliders show degrees.

Let the arm lengths be l1l_1 and l2l_2 and each weight have mass mm. With the yy axis pointing down from the pivot, the weights are at

x1=l1sin⁡θ1,y1=l1cos⁡θ1,x2=x1+l2sin⁡θ2,y2=y1+l2cos⁡θ2 \begin{aligned} x_1&=l_1\sin\theta_1, & y_1&=l_1\cos\theta_1,\\ x_2&=x_1+l_2\sin\theta_2, & y_2&=y_1+l_2\cos\theta_2 \end{aligned}

These formulas connect the angles to the picture on screen. From the speeds and heights of the weights we get the kinetic and potential energy:

K=ml12θ˙12+ml22θ˙222+ml1l2 θ˙1θ˙2cos⁡(θ1−θ2) \begin{aligned} K={}&m l_1^2\dot\theta_1^2+\frac{m l_2^2\dot\theta_2^2}{2}\\ &+m l_1 l_2\,\dot\theta_1\dot\theta_2\cos(\theta_1-\theta_2) \end{aligned}
U=−2mgl1cos⁡θ1−mgl2cos⁡θ2U=-2mgl_1\cos\theta_1-mgl_2\cos\theta_2

Lengths are in metres, mm in kilograms, gg is the gravitational acceleration in m/s², and energy is in joules. The zero of potential energy is arbitrary; only changes in it affect the motion. The last term in KK, with the cosine of the angle difference, is what couples the two arms.

There is no friction, so the total energy E=K+UE=K+U is conserved. The equations of motion follow from the Lagrangian L=K−U\mathcal L=K-U:

ddt∂L∂θ˙i−∂L∂θi=0,i=1,2\frac{d}{dt}\frac{\partial\mathcal L}{\partial\dot\theta_i}-\frac{\partial\mathcal L}{\partial\theta_i}=0,\qquad i=1,2

These are Lagrange’s equations. They are Newton’s laws in disguise, written in terms of angles instead of forces and coordinates.

More: the equations the simulation solves

For two equal masses let δ=θ1−θ2\delta=\theta_1-\theta_2. The angular accelerations (two dots) are written with the numerators N1N_1 and N2N_2:

θ¨1=−N1l1 (3−cos⁡2δ),θ¨2=N2l2 (3−cos⁡2δ), \begin{aligned} \ddot\theta_1&=-\frac{N_1}{l_1\,(3-\cos 2\delta)},\\[4pt] \ddot\theta_2&=\frac{N_2}{l_2\,(3-\cos 2\delta)}, \end{aligned}
N1=3gsin⁡θ1+gsin⁡(θ1−2θ2)+2sin⁡δ (l2θ˙22+l1θ˙12cos⁡δ),N2=2sin⁡δ (2l1θ˙12+2gcos⁡θ1+l2θ˙22cos⁡δ) \begin{aligned} N_1={}&3g\sin\theta_1+g\sin(\theta_1-2\theta_2)\\ &+2\sin\delta\,\bigl(l_2\dot\theta_2^2+l_1\dot\theta_1^2\cos\delta\bigr),\\[6pt] N_2={}&2\sin\delta\,\bigl(2l_1\dot\theta_1^2+2g\cos\theta_1\\ &+l_2\dot\theta_2^2\cos\delta\bigr) \end{aligned}

The acceleration of each arm depends on the state of both. You don’t need to memorise these to follow the experiment.

What the divergence chart shows

The chart compares the first and last pendulums by both angles and velocities:

D2=(Δθ1)2+(Δθ2)2+τ2[(Δθ˙1)2+(Δθ˙2)2] \begin{aligned} D^2={}&(\Delta\theta_1)^2+(\Delta\theta_2)^2\\ &+\tau^2\bigl[(\Delta\dot\theta_1)^2+(\Delta\dot\theta_2)^2\bigr] \end{aligned}
τ=l1+l2g\tau=\sqrt{\frac{l_1+l_2}{g}}

Angle differences are wrapped to the range −π-\pi to π\pi, so a full turn counts as no difference. The time scale τ\tau lets angle and angular-velocity differences be added together. So DD is a dimensionless measure of how different the states are. It isn’t the distance between the weights in metres; that one is shown separately.

A phase chart plots an arm’s angle against its angular velocity. It shows how the state changes even when the weights sit in similar places. The line breaks when the angle passes ±180°, which is how a full turn is drawn.

How the model works

Each weight is 1 kg. The rods are massless and rigid, there is no friction, and g=9.81g=9.81 m/s². The pendulums start at rest. The spread setting is the full difference in lower-arm angle between the first and last pendulum, with the rest evenly spaced in between.

The motion is computed with the fourth-order Runge-Kutta method, step 1/240 of a model second. The charts keep the last 30 seconds. Numerical integration has errors, and they build up over long runs.

The D(t)D(t) chart is handy for comparing regimes, but it isn’t the Lyapunov exponent. Growth on it doesn’t prove chaos by itself, and the computed path isn’t an exact forecast for a real pendulum over unlimited time.

Sources and further reading

myPhysicsLab: Double Pendulum ↗: an independent explanation of the geometry and a derivation of the equations of motion.

Experiment video

Double pendulum

Watch the short video, then try changing the model’s parameters yourself.

Video in Russian

Watch on YouTube