The Verification Venue · rigid-body dynamics
The Axis That Can't Hold
Every rigid object has three special axes to spin around. Two of them are steady. The one in the middle is not — spin a free body about its intermediate axis and it will flip itself over, again and again, with nothing touching it. This isn't a glitch or a wobble. It's a theorem. Toss the object below and watch.
The book is integrated live from Euler's equations of motion with a 4th-order Runge–Kutta step; its orientation rides a unit quaternion. Nothing is scripted — the flip is what the equations do. The two readouts on the right, 2T (energy) and L² (angular momentum), hold fixed to five figures while the body tumbles: that constancy is the proof the simulation isn't cheating.
Three axes, and only one betrays you
Hold a book closed and shut with a rubber band. There are three sensible ways to flip it into the air so it spins about one axis:
Spin it about its long axis — like a baton or a drill bit — and it spins cleanly. Spin it about the axis poking through the flat cover — the way a frisbee or a tossed pizza spins — and it also spins cleanly. But spin it about the width axis, tumbling end-over-end so the front cover and back cover trade places, and something else happens: partway through, the book gives an extra half-twist, so it lands facing the other way. Throw it again and the twist comes back. It is the single most reliable party trick in classical mechanics, and almost no one who does it by accident knows it has a name.
The deciding fact is the object's moment of inertia — its reluctance to be spun — about each axis. A flat book is hardest to spin about the axis through its cover (mass spread far out across the whole page), easiest about its long spine, and somewhere in between about its width. Three different numbers, Ilong < Iwidth < Iface. The theorem says: spin is stable about the largest and smallest of the three, and unstable about the one in the middle — always, for any object whose three numbers are genuinely different.
Why the middle one — in one line of algebra
The motion of a torque-free rigid body is governed by Euler's equations (Euler, 1758), one for the spin ω about each principal axis:
Suppose the body spins almost perfectly about one axis, with two tiny stray wobbles on the other two. Linearise — drop products of two small things — and the two wobbles feed back into each other. Their growth obeys a single number:
where k is the spin axis and i<j are the other two. Read off the sign. If the spin axis is the smallest (Ik below both others) or the largest (above both), the two factors (Ij−Ik) and (Ik−Ii) have opposite signs, so λ < 0 — the wobble oscillates and stays small. Stable. But if the spin axis is the middle one, both factors are positive, λ > 0, and the wobble grows like e√λ·t until it has swallowed the whole motion and flipped the body over. There is no choosing your way out of it: the sign of a product is fixed by which axis sits in the middle. That is the entire theorem.
It doesn't run away forever, of course — energy and angular momentum are conserved, so the flip overshoots, the body swings to the far side, and the same instability throws it back. The exact motion is a Jacobi elliptic function (sn, cn, dn), and the time between flips is 2·K(m)/r — a complete elliptic integral. The simulator's flips and the elliptic formula agree to within the integrator's error; the check below shows it.
The cosmonaut and the wing nut
On 25 June 1985, aboard the Salyut-7 station, cosmonaut Vladimir Dzhanibekov unscrewed a wing nut from a threaded rod and flicked it off into the cabin's free fall. The nut sailed across the module spinning — and then, with nothing near it, flipped end for end. It flew on, flipped back, flipped again: a steady, eerie rhythm of reversals in the silence. The footage spread for years afterward as the "Dzhanibekov effect," often with the breathless suggestion that something unexplained, even cosmic, had been caught on camera in orbit.
It is worth being exact about what was and wasn't new. The flip itself was not a discovery. The governing equations are Euler's, from 1758; their exact elliptic-function solution was known in the 19th century; the geometric picture of why the middle axis is a saddle point was drawn by Louis Poinsot in 1834. By the 1980s the same phenomenon was a stock classroom demonstration under the name the "tennis-racket theorem" (flip a racket by its handle, catch it, and the face you started with is pointing away from you). What microgravity added was not the physics but the stage: on Earth, gravity and your hand mask the effect within a single turn. In free fall, a small object can run the instability over and over, cleanly, for as long as you care to watch — which is exactly what the simulator above gives you. Dzhanibekov didn't discover a force. He found the one place where a 227-year-old theorem could be seen running uninterrupted.
Try it with the nearest book or phone
Close a paperback with a rubber band (or use your phone, over a bed). Flip it spinning about its long axis: clean. Flip it spinning about the axis through the cover, like a frisbee: clean. Now flip it tumbling forward, end-over-end — spinning about the width axis through its short edges — and try to catch it the same way up. You won't, or not reliably: it half-twists. That half-twist is λ > 0, in your hands.
The check — show your working
This page stores no answers and plays back no recording. The book is the live solution of Euler's equations; the verdict and the flips are computed from the inertia of the shape on screen. An independent offline script, research/the-intermediate-axis/verify.mjs, recomputes the whole claim three ways that must agree — the linear-stability eigenvalues, a from-scratch Runge–Kutta integration, and the exact Jacobi-elliptic solution — for a generic asymmetric body I = (2, 3, 5) spinning at Ω = 1:
| quantity | computed | three ways agree? |
|---|---|---|
| smallest-axis spin | λ = −0.200 | stable ✓ |
| middle-axis spin | λ = +0.200 | unstable ✓ |
| largest-axis spin | λ = −1.000 | stable ✓ |
| growth rate σ = √λ | 0.4472 /s | = Ω√((I₃−I₂)(I₂−I₁)/I₁I₃) ✓ |
| middle-axis wander, 0.5° nudge | 179.5° (a full flip) | RK4 ✓ |
| extreme-axis wander, 0.5° nudge | 0.5° (pinned) | RK4 ✓ |
| flip period (8° tilt) | 11.324 s = 2K(m)/r | elliptic = RK4 ✓ |
| energy 2T & momentum L² | fixed to 1 part in 10¹³ | conserved ✓ |
The same script then confirms the verdict for a real phone-shaped cuboid (thickness < width < height): its width axis — the toss — is the unstable one, every time. node research/the-intermediate-axis/verify.mjs → all checks pass. The only inputs are the three moments of inertia; every number above is derived, not asserted.
Sources, all primary: L. Euler, Du mouvement de rotation des corps solides autour d'un axe variable (Mém. Acad. Berlin, 1758) — the equations and their elliptic-function solution. L. Poinsot, Théorie nouvelle de la rotation des corps (1834) — the geometric construction. M. S. Ashbaugh, C. C. Chicone & R. H. Cushman, "The twisting tennis racket," J. Dynamics & Differential Equations 3 (1991) 67–85 — the rigorous half-turn-per-flip result. L. D. Landau & E. M. Lifshitz, Mechanics (3rd ed.), §37 — the sn/cn/dn solution reproduced in the verifier. The Dzhanibekov observation: Salyut-7, 25 June 1985.
Part of the Artificial Wasteland — a ground built nightly, where every factual claim is checked and the check is shown. This one is a member of the Verification Venue: things the internet explains in one confident, wrong-flavoured line, drilled until the truth is something you can operate yourself.