Two angles, one mess, infinite futures.
Hang one pendulum from another and write down Lagrange's equations. The recipe is high-school physics. The motion is the textbook face of deterministic chaos — beautiful, bounded, and utterly unpredictable past a few seconds.
A swing has two angles, two velocities — four numbers. That is the entire state of the toy. And yet two starts that look identical to the eye will, within a handful of seconds, dance two completely different dances. The same equations, different futures. Determinism without predictability.
Start one pendulum at θ₁ = 90°, θ₂ = 90°. The tip traces a beautiful, intricate curve. Restart with θ₂ = 90.00001° — a hundred-thousandth of a degree of difference. The two curves coincide for a few seconds, then peel apart, and by ten seconds in they have nothing whatsoever to say to each other.
This is the purest example of chaos in a low-dimensional classical system. Just Newton's laws, just Lagrange's equations — no randomness anywhere. Yet sensitive dependence on initial conditions is on full display, in motion you can see, in a system you can build with two paperclips.
Move the sliders below to set initial angles. Hit play and watch the tip trace its short, doomed history.
Slide θ₁, θ₂ to set initial angles. Velocities start at zero.
Write the position of each bob: (x₁, y₁) for the first and (x₂, y₂) for the second, both as functions of θ₁ and θ₂. Differentiate to get velocities. The kinetic energy T is ½ m₁ v₁² + ½ m₂ v₂². The potential energy V is m₁ g y₁ + m₂ g y₂. Form the Lagrangian L = T − V — a single scalar function on the four-dimensional state space — and the entire dynamics is buried in it. Lagrange's deep insight, from 1788, is that you never need to draw a single free-body diagram: nature minimises an action, and the equations follow.
T = ½ m₁ (L₁ θ̇₁)²
+ ½ m₂ [ (L₁ θ̇₁)² + (L₂ θ̇₂)² + 2 L₁ L₂ θ̇₁ θ̇₂ cos(θ₁ − θ₂) ]
V = − (m₁ + m₂) g L₁ cos θ₁
− m₂ g L₂ cos θ₂
L = T − VOne scalar L on the (θ₁, θ₂, θ̇₁, θ̇₂) phase space. Plug into Euler-Lagrange and the equations of motion drop out.
Applying ∂L/∂θᵢ − d/dt(∂L/∂θ̇ᵢ) = 0 for i = 1, 2 yields two second-order ODEs. After solving for the angular accelerations θ̈₁ and θ̈₂, you get expressions full of sines, cosines, products of velocities, and a denominator that involves 2 m₁ + m₂ − m₂ cos(2θ₁ − 2θ₂). The equations are exact, deterministic, and completely free of friction or noise. They also admit no elementary closed-form solution. To watch the system, you integrate — RK4 is plenty.
Let Δ = θ₁ − θ₂ , M = 2 m₁ + m₂ − m₂ cos(2θ₁ − 2θ₂)
θ̈₁ =
( −g(2 m₁ + m₂) sin θ₁
− m₂ g sin(θ₁ − 2θ₂)
− 2 sin Δ · m₂ ( ω₂² L₂ + ω₁² L₁ cos Δ ) )
/ ( L₁ · M )
θ̈₂ =
( 2 sin Δ
· ( ω₁² L₁ (m₁ + m₂)
+ g (m₁ + m₂) cos θ₁
+ ω₂² L₂ m₂ cos Δ ) )
/ ( L₂ · M )Two coupled second-order nonlinear ODEs. No elementary closed form. The sim above and the twin demo below both integrate them with 4th-order Runge-Kutta.
At low total energy the motion lives on a smooth invariant torus in the 4D phase space — quasi-periodic, bounded, almost boring. Crank the energy up and the torus shreds: by the KAM theorem of Kolmogorov, Arnold and Moser, most invariant tori survive small perturbations, but as energy grows they break one by one until what is left is a chaotic sea. In the middle, a critical transition where islands of order coexist with stochastic layers. The double pendulum is a tabletop demonstration of the whole KAM story.
Smooth quasi-periodic swing. The trajectory winds an invariant torus and never settles, never escapes.
Chaos. The torus has shredded; trajectories fill a fuzzy region of phase space, and no two starts agree for long.
In the chaotic regime, two nearby trajectories drift apart roughly like ε · e^(λt), where λ is the largest Lyapunov exponent — positive, finite, and a property of the system, not the observer. Equivalently, the predictability horizon scales as t* ≈ (1/λ) · log(L/ε): every extra factor of ten in initial precision buys you only a constant number of seconds of foresight. No better stopwatch, no faster computer, no cleverer integrator can outrun the exponent.
In the chaotic regime, gaps grow like e^(λt) — and the predictability horizon scales only as log(1/ε).
Two pendulums. Same rods, same masses, same gravity. Initial angles differ by one hundred-thousandth of a degree. After a few seconds they have nothing in common.
The state lives in a 4D space (θ₁, θ₂, ω₁, ω₂). Pick a hyperplane — say, the moments when θ₁ = 0 with ω₁ > 0 — and record where the trajectory crosses it. The collection of crossings is a Poincaré section: at low energy a clean closed curve (the torus sliced once), at high energy a fuzzy two-dimensional cloud (the chaotic sea), and at the critical energy a beautiful mix of curves and clouds — the geometric signature of the KAM transition.
Closed loops at low energy (intact KAM tori), a scattering of points at high energy (chaotic sea), and a mixed-phase mosaic in between.
The same coupled-rotor mathematics governs two-link robotic arms (where control engineers fight to suppress the chaos), human gait (where a swinging leg is essentially a passive double pendulum), and chains of coupled oscillators that model power grids, heart pacemakers, and Kuramoto's universal synchronisation. The KAM mechanism behind the regime change also runs in the Solar System: the planets, modeled carefully enough, are chaotic on Gigayear timescales.
The Explorer integrates the equations live, lets you tune the masses, rod lengths, gravity, and the two initial angles, and overlays a ghost twin whose start differs by a millionth of a radian so you can watch the two trails peel apart. Energy and divergence read out live as you push the settings.
How and where this technique lives in the world today.
Topics in the same vein.