The Horizon of a Pendulum,
& other small infinities

A pendulum is the simplest machine worth thinking about. A weight, a string, a pivot — that is all. Galileo watched a chandelier in a Pisa cathedral and timed it with his pulse¹, and the pattern he noticed — that the period barely depends on how far you pull it — has been the foundation of clocks, metronomes, and a surprising amount of modern physics ever since.

And yet: pull it far enough, and the same pendulum becomes unpredictable. Not statistically — actually. The equation that describes it is deterministic, solvable in principle, and produces behavior you can simulate but cannot forecast. This is the fact I want to sit with.

Here is the equation, in the form most textbooks give it — and which, were this a proper printed volume, would stand at the foot of the page in a slightly smaller size, set apart by a thin rule:

d²θ/dt² + (g/L)·sin θ = 0

$\sqrt{a^2 + b^2}$

The angle θ, the gravitational constant g, the length L. That is the entire state of the world, as far as a pendulum is concerned. The sine is the interesting character in the sentence — it is what keeps the restoring force pointing roughly toward zero, and it is also what makes the equation nonlinear.

For small θ, sin θ ≈ θ, and the equation becomes a spring. Clean, periodic, boring in a good way. This is the small-angle approximation, and it is the reason the pendulum in a grandfather clock keeps better time than most of your friends.

Push θ past about thirty degrees, though, and the approximation starts to lie. The period grows with amplitude. Add a second pendulum — hang one off the first — and the lie becomes a cascade². Two identical double pendulums, started from angles that differ by a thousandth of a degree, will be in entirely different places a minute later. Not noisy. Not drifting. Different.

The rate at which two nearby trajectories pull apart has a name: the Lyapunov exponent. A positive one means small errors grow exponentially. Double the precision of your measurement and you buy a fixed amount of time — maybe a few seconds, maybe a few hours, depending on the system. After that, you are guessing.

This is the horizon I mean. Not a wall, but a fog: the far distance where even a perfect model, run on a perfect computer, stops being able to tell you what the pendulum is doing, because the tiniest rounding error has grown into the whole answer.

I find this comforting, in a way I do not fully understand. There is a limit to what you can know about a string with a weight on it. There is a limit to what you can know about most things. The equation is still there, still clean, still solvable in the way the word is used in a math class. The world is still deterministic. And yet, past a certain horizon, the pendulum and you are both free.