Why is the double pendulum chaotic?

Why is the double pendulum chaotic?

A double pendulum is a pendulum hanging from the end of another. Its equations are exact and deterministic, yet the motion is chaotic: two runs that start almost identically drift apart exponentially fast, so long-term prediction is impossible. The chaos comes from the coupling — each arm changes the force on the other through the angle between them — which makes the equations of motion strongly nonlinear.

What is a double pendulum?

Two rigid rods of lengths $L_1$ and $L_2$ carry masses $m_1$ and $m_2$. The upper rod pivots on a fixed point; the lower one pivots on the first mass. Two angles, $\theta_1$ and $\theta_2$, fully describe the state, together with their rates of change. That is four numbers — a point in a four-dimensional phase space.

Where do the equations come from?

Tracking the tension in two coupled rods is awkward, so physicists use the Lagrangian $\mathcal{L} = T - V$, the kinetic energy minus the potential energy, written in terms of the two angles:

The cross term $\dot\theta_1\dot\theta_2\cos(\theta_1-\theta_2)$ is the coupling: the motion of each arm enters the other's equation. Applying the Euler–Lagrange equation to each angle gives two coupled nonlinear differential equations with no general closed-form solution — they have to be integrated numerically.

What does sensitive dependence look like?

We integrated the exact equations for two double pendulums with $L_1 = L_2 = 1$ m and equal masses, released from rest at $\theta_1 = 120°$, $\theta_2 = -10°$. The second run starts with $\theta_1$ larger by just $10^{-6}$ rad ($0.00006°$):

The gap grows roughly exponentially, $\delta(t) \approx \delta_0 e^{\lambda t}$, where $\lambda$ is the Lyapunov exponent. Beyond the Lyapunov time $\sim 1/\lambda$ the two histories are unrelated. Starting instead from small swings ($5°$, $5°$) with the same $10^{-6}$ rad difference, the runs stayed within about $10^{-6}$ rad for a full minute — regular motion, no chaos.

Why can't we just use a better computer?

Because the problem is not the arithmetic: any measurement of the starting angles has some error, and that error is amplified exponentially. Halving the uncertainty buys only a fixed amount of extra prediction time, $\ln 2/\lambda$. Real chaotic systems — the weather, the three-body problem — face the same wall. What survives is the statistics: the region of phase space the motion visits, not the path.

Keep exploring