When does a spinning thing stop being predictable?

A rigid body tumbling in deep space obeys equations we have known for two hundred years. And yet there is a precise sense in which, past a certain point, those very equations refuse to tell us what happens next. This is a small essay about the difference between having a law and being able to use it — told through a spacecraft, a wing-nut, and a number called the Lyapunov exponent.

Two kinds of “we know the equations”

There is a comfortable assumption buried in the phrase deterministic physics: that if we know the equations of motion and the present state exactly, the future is ours to read off. Laplace made this explicit in 1814 — an intellect that knew all forces and all positions “would embrace in a single formula the movements of the greatest bodies of the universe and those of the tiniest atom.”

The twentieth century took that assumption apart, gently, without ever contradicting the equations themselves. The equations remain deterministic. What changed is our understanding of how information about the initial state flows forward in time. In some systems it is preserved — a small error in where you start stays a small error forever. In others it is amplified, exponentially, so that any uncertainty in the present — and there is always some — swells to swallow the prediction whole. Both kinds of system are perfectly deterministic. Only one of them is predictable in practice.

The boundary between them is not philosophical hand-waving. It is measurable, with a single number. We have spent the last little while measuring it for a very specific case — two bodies tumbling near a gravitational balance point — and the answer turned out to be more interesting than we expected.

Before we start: the real JWST

The James Webb Space Telescope sits about 1.5 million kilometres from Earth, at the second Lagrange point L2 — a place where the gravity of the Sun and Earth and the centrifugal tug of the orbit conspire to let a spacecraft loiter. Webb is not rigidly fixed there; it traces a slow halo around the point and must constantly, delicately manage its attitude — which way it is pointing — to keep its mirror in shadow and its instruments cold.

Attitude is rotational. A telescope is a rigid body, and a rigid body free to rotate is one of the oldest problems in mechanics: Euler wrote its equations in the 1750s. Our project models a stripped-down cousin of the Webb situation — two rigid bodies near a Lagrange point, coupled only by gravity — and asks a question the mission engineers care about in a more applied form: if the thing starts tumbling, does the tumble stay orderly, or does it become unpredictable?

The wing-nut that flips itself

Start with a fact you can see on any video shot aboard the International Space Station. Take an object with three different widths — a wing-nut, a phone, a book — and spin it about its middle axis (the one that is neither the longest nor the shortest). It will not spin steadily. After a few rotations it suddenly flips over, spins the other way for a few rotations, flips again, and keeps doing this forever. Cosmonaut Vladimir Dzhanibekov noticed it in 1985; it is now called the Dzhanibekov effect, or the tennis-racket theorem, and it falls straight out of Euler’s equations. The middle axis is unstable: a body spinning about it sits on a knife-edge, and the slightest nudge sends it into the flipping dance.

Here is the subtle part, and it is the heart of this essay. The flipping is not chaotic. It looks dramatic, but a torque-free rigid body — even one flipping end over end — is what mathematicians call integrable. There is a closed-form solution: you can write down, in advance, exactly where the body will be at every future moment, using functions (the Jacobi elliptic functions) that were worked out in the 1820s. The flips are perfectly periodic. A small error in how you start stays a small error. The wing-nut is unpredictable to the eye but entirely predictable to the equations.

So a spectacular-looking instability and genuine unpredictability are different things. The middle-axis instability is local — it is about a knife-edge equilibrium — but the global motion threads its way around that knife-edge in a completely regular way. To get true unpredictability you need something more.

What it takes to break predictability

That “something more,” in our two-body system, is the gravity gradient.

A body near a much larger mass does not feel a single gravitational pull; the near side is pulled harder than the far side, and the difference exerts a twisting torque that depends on which way the body is oriented. This is the same effect that locks the Moon’s face toward Earth and that any large spacecraft must reckon with. It couples the body’s attitude to its position — and crucially, it destroys the special mathematical structure (a second conserved quantity) that made the torque-free tumble integrable.

Once integrability is gone, the door to chaos is open. Whether chaos actually walks through it, and how strongly, is a quantitative question — and that is what the Lyapunov exponent measures.

The number that measures unpredictability

Take your system, and take a second copy of it that starts an infinitesimal distance away — the smallest perturbation you like. Let both evolve. Track the distance between them. In a regular system that distance grows slowly, like a polynomial; in a chaotic system it grows exponentially, like eλt. That growth rate λ is the Lyapunov exponent. If it is positive, the system is chaotic: errors double every 1/λ or so, and prediction has a horizon. If it is zero (with the motion staying bounded), the system is regular — the wing-nut, the orderly tumble, the predictable case.

It is a beautifully operational definition. It does not ask philosophical questions about determinism; it just measures how fast two nearly-identical copies of the universe drift apart. We computed it for our two-body system, using an algorithm from 1980 (Benettin and colleagues) that propagates a little cloud of perturbations alongside the trajectory and watches how the cloud stretches.

We ran it — three cases

We checked the machinery against three situations whose answers we already knew, which is the only honest way to trust a chaos measurement:

A torque-free tumbling body — the wing-nut, integrable. The measured Lyapunov exponent came out essentially zero (a tiny 2.3 × 10 − 4 that keeps shrinking the longer we run): no chaos, as the closed-form solution demands.

A body librating — rocking gently back and forth about a gravity-gradient equilibrium, like a pendulum that never quite goes over the top. Here the exponent was 2.4 × 10 − 6, more than two hundred times smaller than the rocking frequency itself. Rocking is not diverging. Oscillation and unpredictability are, again, different things.

And the gravity-gradient Dzhanibekov case — the tumbling body with the orientation-position coupling switched on, sitting near that unstable middle axis. Here, and only here, the exponent came out genuinely positive: 1.5 × 10 − 5 per second. Small — the chaos is weak, a thin layer near the knife-edge — but unmistakably greater than zero. The coupling broke the integrability, and predictability went with it.

That hierarchy is the whole story in one line: a dramatic local instability is not chaos; a gentle oscillation is not chaos; only the coupling that destroys a conserved quantity produces real, measured unpredictability — and even then, here, only weakly.

A correction we had to make to ourselves

One honest aside, because it is the part of this work we are most glad we caught. When we first wrote up why the integrable case converges so slowly to zero, we blamed the mathematical coordinates we had chosen — a flat, slightly awkward way of representing rotations. We predicted that a more elegant, geometry-respecting method would make the slow convergence vanish.

We then built the elegant method and ran it. It gave the same slow convergence. Our explanation had been wrong: the slowness is not an artefact of our bookkeeping, it is intrinsic to the quasi-periodic motion itself — a property of the physics, not of our description of it. The elegant method did fix a different, real problem (it cleaned up the full spectrum of exponents), but not the one we had attributed to it.

We mention this not as a confession but because it is, in miniature, exactly the discipline the whole project is built on: a claim about the world is only as good as the test that could prove it wrong, and when the test speaks, the write-up changes — not the other way around. The prediction was falsified by an experiment we ran on ourselves, and the document now says so plainly.

What this tells us: determinism is not the same as foreknowledge

Step back from the spacecraft. The thing we measured is the rate at which a system forgets its own initial conditions — the speed at which the present stops determining the knowable future, even while it still determines the actual one.

This is, quietly, one of the more important conceptual shifts of the last century. Laplace’s demon is not refuted by chaos; given truly perfect information it would still succeed. What chaos shows is that perfect is doing impossible work in that sentence. In a system with a positive Lyapunov exponent, the information you would need to predict the distant future is not a little more than what you have — it is exponentially more, without limit, forever. Determinism survives. Foreknowledge does not.

And the line between the two is not vague. It is a number you can compute, for a tumbling telescope or a planetary orbit or a beating heart. For our two bodies near their balance point, the number is small and positive: the motion is deterministic, lawful, and — past a horizon of a few times 1/λ — genuinely beyond prediction. The same equations that fix the future also, in the same breath, hide it. There is something worth sitting with in that.

For the technically curious. Everything above is derived from first principles, with every number reproducible from open code. The chaos analysis lives in Chapter 07 of the project’s canonical documentation; the three calibration cases, the Lyapunov algorithm, and the self-correction described above are all there in full, with the experiments that produced each figure. The integrability of the torque-free top (the wing-nut) is the classical Jacobi-elliptic result; the gravity-gradient torque and its consequences are derived in Chapters 03 and 05.

References

  • L. D. Landau & E. M. Lifshitz, Mechanics (Course of Theoretical Physics Vol. 1), 3rd ed., 1976 — §37, the torque-free asymmetric top and its Jacobi-elliptic solution.
  • H. Goldstein, C. Poole & J. Safko, Classical Mechanics, 3rd ed., 2002 — §5.6, rigid-body motion and the polhode.
  • G. Benettin, L. Galgani, A. Giorgilli & J.-M. Strelcyn, “Lyapunov Characteristic Exponents for Smooth Dynamical Systems and for Hamiltonian Systems; A Method for Computing All of Them,” Meccanica 15, 9–30 (1980) — the algorithm.
  • C. Skokos, “The Lyapunov Characteristic Exponents and Their Computation,” Lecture Notes in Physics 790 (2010) — modern exposition.
  • P. C. Hughes, Spacecraft Attitude Dynamics, 2004 — gravity-gradient torque and attitude stability.

For the technical reader. Project repository: https://github.com/lege-artis/jwst-l2-coupled-dynamics. The chaos chapter: docs/canonical/en/07-lyapunov-spectrum.md (Czech: docs/canonical/cs/07-ljapunovovo-spektrum.md). The companion canonical-reference project on the mathematics of the Fourier transform — which sits behind much of the signal analysis here — is https://github.com/lege-artis/fourier.

Part of an occasional series on the physics of spinning things and the mathematics that lets us — and sometimes refuses to let us — see ahead.