The axis a leak of energy chooses

Throw an object spinning and, with no air and no gravity gradient to speak of, nothing exerts a torque on it. Its angular momentum is then fixed forever — magnitude and direction, in space, exactly. Its kinetic energy is also fixed, so long as the object is genuinely rigid.

No object is. Everything flexes a little, and flexing warms it, and warming it takes energy out of the motion and does not give it back. The angular momentum cannot go anywhere, because no torque acts. The energy can, and does.

That combination has a consequence that sounds too strong to be true and is: a freely spinning body has exactly one stable state, and it is a spin about the axis it has the most inertia about. Every other spin is temporary, including the ones a textbook calls stable.

The argument in three lines #

There is no need to solve anything to see where this goes.

Angular momentum is conserved, so $L$ is a fixed number. For a spin about a principal axis, $L = I\omega$ and $T = \tfrac{1}{2}I\omega^2 = L^2/2I$. Larger $I$ therefore means smaller energy at the same momentum. And internal dissipation only ever reduces $T$.

So a body that starts spinning about its axis of least inertia is sitting at the top of the available energy range, with nothing holding it there and a one-way process pushing it down. It will end at the bottom, spinning about the axis of greatest inertia — a flat spin, end over end for a pencil-shaped object — and it will stay there, because at the bottom there is nothing left to remove.

The reasoning is exactly the reasoning of what a system actually minimises, applied to a mechanical rather than a thermodynamic constraint: a quantity that can only decrease, subject to a quantity that cannot change, has a destination that can be read off without following the path.

The same statement as a picture #

The energy band has a geometric form which is worth having beside the arithmetic, because it says what the intermediate stages look like.

In the space of angular velocity components, fixing the angular momentum confines the state to an ellipsoid, and fixing the energy confines it to a second, different ellipsoid. The motion of a rigid body is the intersection of the two — a closed curve, traced once per wobble, which Poinsot called the polhode. There is a family of these curves, one for each energy the momentum allows, and they are nested: tight loops around the two poles at the ends of the greatest-inertia axis, tight loops around the poles of the least-inertia axis, and between them a pair of curves crossing at the intermediate axis.

Dissipation moves the state from one curve of the family to the next. The rapid motion is around a polhode; the slow motion is the polhode itself migrating. That separation is why the argument in the previous section works without solving anything: the fast motion conserves what it needs to, and the slow drift has only one direction available.

It also says what the crossing curves mean. The two curves that intersect at the intermediate axis are the boundary between the two families, and a state on one side circulates about one axis while a state on the other circulates about a different one. A body arriving at that boundary does not stop there — the intersection point is a saddle, which is the intermediate-axis instability restated — and passes into the other family. Watching the polhode migrate across that boundary is watching the tumble begin.

Making the leak honest #

Writing this as a simulation requires care, because the obvious way to add damping is wrong in a way that would produce the right answer for the wrong reason.

An internal process — a flexing antenna, a slosh of fuel, a joint with friction in it — cannot change the total angular momentum. Any damping torque added to the equations of motion must therefore be perpendicular to $\mathbf{L}$, and if it is not, the simulation is quietly applying an external torque and the flip it produces proves nothing.

The damping used here is $\mathbf{M} = -c\,\boldsymbol{\omega}_\perp$, the component of the angular velocity that is not parallel to the angular momentum. That choice has two properties, both of which are exactly what is wanted. It is orthogonal to $\mathbf{L}$ by construction, so $|\mathbf{L}|$ is conserved identically. And its power is $-c|\boldsymbol{\omega}_\perp|^2$, which is negative except when $\boldsymbol{\omega}$ is parallel to $\mathbf{L}$ — which happens only on a principal axis. The sink switches itself off precisely at the states it is driving the body towards, which is the signature of a relaxation rather than a driven process.

Both statements are checked against the run rather than assumed: the figure reports the momentum’s drift, and the energy is required to be non-increasing at every single step before anything is drawn.

The flip itself #

The shape of that curve is the part worth carrying away, because it is what makes the failure dangerous.

Nothing happens for a long time. The wobble grows exponentially from whatever it started at, which means an initial tilt a hundred times smaller merely postpones the event by a fixed number of doubling times rather than preventing it. Then, once the wobble is comparable with the spin, the transition takes about one doubling time and the body is in a completely different motion.

An exponential with a long time constant looks exactly like stability until it does not. A test on the ground that ran for an hour would have shown nothing at all.

How long #

The scaling has no threshold in it. There is no leak small enough to make a minor-axis spin permanent, only leaks small enough to make it last longer than the mission.

That is the lesson Explorer 1 delivered in 1958. The satellite was a pencil: about two metres long, fifteen centimetres across, with a moment of inertia about its long axis a hundred-odd times smaller than about a transverse one. It was spun about that long axis at 750 revolutions a minute, which is a spin about the axis of least inertia — stable by the rigid-body theorem, and the natural choice for a body that has to fit inside a rocket.

It also carried four flexible whip antennas. They flexed. Within a few hours the satellite was tumbling end over end, and the telemetry showed it: the signal strength varied with a period that no longer matched the spin. The theory had been available since the nineteenth century and the design had not used it. Every spin-stabilised satellite built since is spun about its axis of greatest inertia, and those that cannot be carry a nutation damper — a tube of fluid or a mass on a spring, put there deliberately to make the leak large, so that the relaxation happens quickly and predictably instead of slowly and at an inconvenient moment.

The numbers make the case rather than the story does. Explorer 1 massed about fourteen kilograms in a cylinder two metres long and fifteen centimetres across, which puts its moment of inertia about the long axis near 0.04 kg m² and about a transverse axis above 4 — a ratio of order a hundred, and therefore an available energy drop of the same factor. It was spun at 750 revolutions a minute, so a whip antenna at the end of the body was being swung through several g at about twelve hertz, flexing on every turn. There was no shortage of leak, and no ambiguity about where the energy would go once it started leaving. The only quantity nobody had estimated was the time, and the answer turned out to be a few hours.

Adding dissipation on purpose is the correct response, and it is worth noticing why. The leak cannot be removed, so the only remaining choice is over its size, and a large leak reaches the stable state during commissioning rather than during operation.

What the wobble does to a measurement #

Long before the flip, the growing wobble is doing something a designer notices first: it moves the direction the body is pointing.

A spin-stabilised satellite is used as a platform. Its antenna, its camera or its sensor is fixed to the body and points along the spin axis, and the whole point of spinning it is that the axis holds still in space. What holds still is the angular momentum, and the body axis coincides with it only when the spin is about a principal axis. As the wobble grows, the body axis sweeps a cone around the fixed momentum vector, and the half-angle of that cone is exactly the pointing error.

So the failure announces itself as a slow degradation of pointing rather than as a sudden event, and it does so on the same exponential schedule as everything else here: a tenth of a degree, then a fifth, then two-fifths. On Explorer 1 the announcement came as a modulation of the received signal strength, because the antenna pattern was being swung around. That is the general shape of it — the first symptom is not a loss of control but a periodic variation in something that had been steady, at the wobble frequency rather than the spin frequency, and the two are easy to confuse if only one is expected.

Two different theorems, often confused #

There is a well-known result about which spins of a rigid body are stable, and it is not this one.

The axis that will not hold is about the middle column of that table: a spin about the intermediate axis is unstable in a perfectly rigid body, on a timescale of a fraction of a turn, and the tennis racket flipping in mid-air is the demonstration. That instability is geometric. It needs no dissipation, no flexing and no time to speak of.

The instability here is in the right-hand column, and it is different in every respect. It affects the axis of least inertia, which the rigid theorem calls stable. It requires dissipation, so a genuinely rigid body would not show it. And it is slow — as slow as the leak is small.

Putting the two columns together leaves one row with two ticks in it. A real body has one stable spin, and everything else is either destroyed in half a turn or destroyed eventually.

The same shape, elsewhere #

The structure of the argument recurs whenever a conserved quantity and a dissipated one are in tension, and it is worth naming so it can be recognised.

A rotating star or a self-gravitating cloud loses energy by radiating while its angular momentum is fixed by having nowhere to send it, so it must contract in the direction along its spin axis and flatten — which is the reason a galaxy is a disc. A molecule vibrating in a solid loses energy to its neighbours at fixed total momentum and settles into the lowest mode available to it. A spinning coin on a table loses energy to friction and rolling and ends flat, which is the same end state as the satellite’s for the same reason.

In each case the conserved quantity restricts the destination and the dissipated one selects among what is left. Neither alone predicts anything; the pair does. It is the mechanical version of the argument that a system at fixed energy maximises its entropy, and it needs no statistics at all — entropy is a count reaches the same conclusion with many degrees of freedom, and this reaches it with three.

Reading the timescale backwards #

Because the relaxation time scales simply with the leak, an object observed tumbling carries information about how lossy it is — and that inversion is used routinely on objects nobody can visit.

An asteroid struck hard enough is left rotating about no principal axis at all, and internal friction then relaxes it towards a principal-axis spin on a timescale that depends on its size, its spin rate and how dissipative its interior is. The dependence is steep: slow, small bodies relax slowly. So a survey that finds tumbling among the slow small ones and principal-axis rotation among the rest is measuring, indirectly, both how long ago those bodies were last hit and how much energy their interiors absorb — which is a statement about whether they are solid rock or a loosely held pile of rubble.

The logic is the same one this essay has been making, run in reverse. Forwards, a known leak predicts a time. Backwards, an observed time constrains the leak, and the leak is a property of the material. A relaxation whose rate depends on one unknown is a measurement of that unknown, and the fact that the destination is fixed by conservation alone is what makes the rate the only thing left to learn.

Where the model stops #

The damping is a single number. Real dissipation in a spacecraft comes from specific mechanisms — a fuel slug moving in a tank, a boom bending at a particular frequency, a joint slipping — each with its own dependence on the wobble’s amplitude and frequency, and some of them are strongly resonant. The scaling verified here is the scaling of a linear sink and would not survive a mechanism that only works in a narrow band.

The body is treated as rigid apart from the leak. That is a contradiction taken carefully: the flexing that dissipates the energy also changes the moments of inertia, and for a body with large flexible appendages the change is not small. The figures here are for a body whose flexing is enough to lose energy and not enough to matter geometrically.

The moments are constant. A body that changes shape during the process — a satellite deploying a boom, a diver tucking — has a different problem, and the quantity that survives a change of shape is where that one is worked out.

And nothing here is about a body being driven. A wheel with a motor on it, a spacecraft with thrusters, or anything else with an external torque is outside the argument entirely, because the conservation law that fixes the destination has gone.

What the pictures cannot show #

The flip figure plots the three components of the angular velocity in the body’s own frame, which is where the equations are simple and where the motion is least intuitive. What a viewer would see is the body’s long axis sweeping out a cone that opens up over hours until the object is going end over end — and there is no way to put that on the same axes as the quantities that explain it.

The band figure shows the energy range and gives no sense of the path taken through it. A body relaxing does not slide smoothly down the band; it circulates rapidly around a polhode while the polhode itself migrates slowly, and the two motions differ in rate by whatever the damping constant is small. That separation of timescales is what makes the argument in the first section work, and it is invisible in a figure of the endpoints.

Where the ladder goes next #

The rotation ladder began with torque and the moment of inertia in the same push, further out and the mass, and where it sits, reached the geometric instability of the axis that will not hold, and then the gyroscope in the top that nods before it settles. This rung adds the one ingredient a rigid body cannot have — a way of losing energy — and finds that it decides the outcome by itself.

The rung after it is the body that is not merely leaky but genuinely deformable, where the shape it settles into and the spin it settles into are solved together. The habit to carry there is the one this rung is built on: when one quantity is conserved and another can only fall, the answer is at the bottom of the second subject to the first, and it can be written down before any equation of motion is solved.