Zero Angular Momentum: The Falling Cat and the Geometry of Shape Space
Abstract
A cat dropped upside-down rotates 180° and lands on its feet, despite having zero angular momentum throughout. This is not a trick and not a violation of physics. The explanation took physicists from 1894 to 1993 to fully work out, and the answer — a geometric phase arising from the holonomy of a fibre bundle — is the same mathematics that governs the Berry phase in quantum mechanics and the Aharonov–Bohm effect in electrodynamics. We adopted two strays this year. They fall beautifully.
Contents
We adopted two stray cats in 2023. They had been living under a garden shed and had strong opinions about most things, including the correct height from which to leap onto a bookshelf and whether landing was optional. They are indoor cats now, for health reasons — a vet’s recommendation they find unconvincing but have largely accepted. Watching one of them drop from a windowsill — always feet-first, always orientated correctly, from heights that would leave me reconsidering my life choices — I found myself thinking about a problem I had first encountered in a mechanics course and had never fully resolved to my satisfaction.
How does a cat rotate with zero angular momentum?
The Problem
When a cat is dropped from an inverted position — upside-down, held by a practised experimenter, then released — it rotates approximately 180° and lands on its feet. The turn is over in a fraction of a second. The cat begins with negligible angular momentum (the experimenter can release it with almost no spin, and cats manage the turn “even if released from complete rest while upside-down” [[1]]), and there are no external torques during free fall. By conservation of angular momentum, the total angular momentum of the cat must remain constant throughout the fall.
In the ideal free-fall model, with an initially negligible angular momentum, the total angular momentum is therefore approximately zero throughout the fall.
And yet the cat rotates 180°.
This is the falling cat problem. Étienne-Jules Marey’s 1894 chronophotographs made it a famous mechanical puzzle, and it has occupied physicists, mathematicians, neuroscientists, and roboticists ever since; Gregory Gbur has written its history at book length [[2]].
The problem is not exotic. Every cat owner has seen it. What requires explanation is why our intuitions about angular momentum fail here, and what replaces them.
Why the Obvious Answers Do Not Work
There are two naive explanations for the cat’s righting reflex, both wrong.
Explanation 1: The cat uses initial angular momentum. The experimenter gives the cat a small spin before releasing it; the cat amplifies this to achieve the full 180°. A deliberately idealised model can start with zero total angular momentum and still right itself, so an initial spin is not required for the mechanism.
Explanation 2: The cat pushes against the air. A falling cat could, in principle, use aerodynamic forces to push against the air and generate a reaction. Aerodynamics can affect a real animal’s fall, but it is not required: the rigid-body model below produces reorientation in torque-free free fall.
Both explanations appeal to external torques. The correct explanation requires none.
Marey and the Photographic Evidence
Étienne-Jules Marey reported a chronophotographic sequence of a falling cat to the Académie des sciences [[3]] and published it in La Nature on 10 November 1894. It makes the essential point visible: the cat is not a rigid body. Its front and rear regions change orientation relative to one another during the manoeuvre.
The net effect: the cat’s body orientation rotates by 180° even though the total angular momentum — computed as the sum of both halves — remains constant. The key word is sum. Individual parts can exchange angular momentum through internal torques; the sum is conserved.
This mechanism — internal redistribution of angular momentum without changing its total — is correct but not complete. It explains that rotation is possible, not how much rotation is achieved per cycle of shape change. For that, we need the mathematics.
Kane and Scher: The Two-Cylinder Model
The two-body model that later work builds on was published by T. R. Kane and M. P. Scher in 1969 (International Journal of Solids and Structures 5, 663–670) [[1]]. It was not the first. Kane and Scher credit G. G. J. Rademaker and J. W. G. ter Braak with a 1935 model that bends without twisting but needs as much backward bending as forward bending, whereas in photographs of falling cats the backward bend is “far less pronounced” [[1]].
They modelled the cat as two identical rigid axisymmetric bodies — a front half and a rear half — connected by a special no-twist joint. The joint allows the model to bend but rules out axial twisting at the connection; the total angular momentum of the system about its centre of mass is held fixed at zero. It is a deliberately simplified cat, not a literal account of every limb movement. One number comes from the animal: the ratio $J/I$ of each half’s moment of inertia about its own long axis to that about a transverse axis through its centre of mass. Kane and Scher call $J/I = 0.25$ “a realistic value”, based on measurements “on fourteen segments of a dead cat” [[1]].
The zero-angular-momentum constraint, together with the no-twist condition, gives equations for the overall orientation as a function of the model’s shape. They can be integrated along a prescribed bend-and-reorient trajectory.
Kane and Scher exhibited a righting trajectory for this model: the axis of the rear half sweeps a cone fixed in the front half, with a forward bend of 116° and a backbend of 25° estimated from photographs of a falling cat [[1]]. Their calculation showed, from mechanics alone, that this deformable body can reorient without external torque while conserving total angular momentum.
Montgomery’s formulation makes the geometric structure of that result explicit.
Montgomery: Fibre Bundles and Geometric Holonomy
In 1993, Richard Montgomery published a reformulation of the falling cat problem using gauge theory [[4]]. The reformulation is a powerful mathematical treatment, and it connects the cat to a recurring structure in modern physics.
The Configuration Space
Let $Q$ be the configuration space of the two-body model once translations have been removed, that is, the orientations of both halves. Rotating the whole model rigidly does not change its shape, so the shape space is the quotient $\mathcal{S} = Q/SO(3)$, and over a small patch of shapes
$$Q \cong SO(3) \times \mathcal{S},$$where $SO(3)$ is the rotation group (the model cat’s overall orientation). This product is a local picture, not a global one, and not a detailed anatomy of a real cat. With both halves free, $\mathcal{S}$ is three-dimensional. Kane and Scher’s no-twist joint confines the model to configurations reachable without twisting, whose shapes form a real projective plane with the fully folded shapes as its “line at infinity”; over it the structure group shrinks from $SO(3)$ to $O(2)$ [[4]].
The angular momentum constraint $\mathbf{L} = 0$ defines a horizontal distribution on $Q$ — a preferred subspace of tangent vectors at each point that correspond to shape changes at zero angular momentum. This distribution is not integrable (it does not come from a foliation), which is the mathematical signature that holonomy is possible.
The Fibre Bundle
The projection
$$\pi \colon Q \to \mathcal{S}, \qquad (R, s) \mapsto s,$$makes $Q$ into a principal fibre bundle over $\mathcal{S}$ with structure group $SO(3)$. The fibre above each shape $s \in \mathcal{S}$ is the set of all orientations the cat can have with that shape.
A connection on this bundle is a rule for “lifting” paths in the base $\mathcal{S}$ to horizontal paths in the total space $Q$ — that is, paths along which the angular momentum constraint is satisfied. Earlier versions of this post called the connection a one-form on $\mathcal{S}$; I had skipped a step. The mechanical connection $\mathcal{A}$ is a one-form on $Q$, with values in the Lie algebra $\mathfrak{so}(3)$: it assigns to each velocity the angular velocity of the locked body that would carry the same angular momentum, and its kernel is the horizontal space. It becomes a one-form $A$ on a patch of $\mathcal{S}$ only after a reference orientation has been chosen for every shape (a local section of the bundle, which physicists call a gauge). With $g \in SO(3)$ the orientation relative to the reference, zero angular momentum reads [[4]]
$$g^{-1}\,dg = -A, \qquad A = \mathbb{I}(s)^{-1}\,\mathbf{K}(s)\,ds, \tag{1}$$where $\mathbb{I}(s)$ is the inertia tensor of the body locked in shape $s$ and $\mathbf{K}(s)\,ds$ is the angular momentum that the shape change $ds$ alone would carry, both written in the reference frame.
Holonomy: The Geometric Phase
When the cat executes a closed loop $\gamma$ in shape space — a sequence of shape changes that returns it to its initial shape — integrating Eq. (1) round $\gamma$ gives the net rotation, the holonomy of the connection:
$$R_\gamma = \mathrm{Hol}_\mathcal{A}(\gamma) \in SO(3).$$For the full non-Abelian case ($SO(3)$), the holonomy is a path-ordered exponential along $\gamma$ and its relationship to the curvature involves non-Abelian corrections. But the essential geometric intuition is captured by the Abelian case — rotation about a single axis — where Stokes’s theorem gives the net rotation directly:
$$\Delta\theta_\gamma = -\oint_\gamma A = -\iint_{\Sigma} F, \tag{2}$$where $\Sigma$ is a surface in $\mathcal{S}$ bounded by $\gamma$, $F = dA$ is the curvature two-form, and the sign follows the convention of Eq. (1). The cat’s net rotation per cycle is the integral of the curvature over the area enclosed by its shape-change loop in $\mathcal{S}$. For small loops, the curvature $F = dA - A \wedge A$ (the sign again follows Eq. (1)) determines the holonomy to leading order in both the Abelian and non-Abelian cases.
For Kane and Scher’s cat the Abelian case is not an approximation. Montgomery shows that the no-twist connection has a single component, so that “the model cat can only rotate about the z-axis”, the axis through the two halves’ centres of mass, and the reorientation produced by any strategy reduces to a single quadrature [[4]]. The other component of $O(2)$, a half-turn about a perpendicular axis, enters only for loops through the fully folded shapes; Eq. (2) holds for loops that avoid them [[4]].
The rotation is geometric: it depends on the shape of the loop, not on the speed at which the loop is traversed. A cat executing the same shape-change sequence twice as fast achieves the same rotation in half the time.
How Much Turn per Loop
The simplest loop keeps the bend fixed. Bend the model by an angle $\beta$ from straight, let the direction of the bend travel once round the body (forward, to one side, backward, to the other side, forward again, each half spinning about its own axis so that the joint never twists), then straighten. This is Rademaker and ter Braak’s motion [[1]], [[4]]. Integrating Montgomery’s curvature over the disc of shapes that the loop encloses gives the net turn about the long axis:
$$\Delta\theta(\beta) = 2\pi\left[1 - \frac{(J/I)\cos(\beta/2)}{\sin^2(\beta/2) + (J/I)\cos^2(\beta/2)}\right] \pmod{2\pi}. \tag{3}$$Fig. 1 plots it. A half-turn needs $\beta \approx 59^\circ$ at Kane and Scher’s $J/I = 0.25$, where their own closed-form version of this condition gives “nearly 60°” [[1]], and $\beta = 120^\circ$ at $J/I = 1$. A bend of 59° forward and 59° backward is precisely the equal backbend Kane and Scher objected to.
For $J/I = 1$ both halves have spherical inertia ellipsoids, and Montgomery points out that the connection is then the Levi-Civita connection of a round sphere [[4]]. Eq. (3) becomes $2\pi[1 - \cos(\beta/2)]$, the solid angle of a polar cap of half-angle $\beta/2$.
The Connection to Berry Phase
The gauge structure of the falling cat problem is not an isolated curiosity. It is the same mathematical structure that governs several central phenomena in modern physics.
The Berry phase [[5]] arises when a quantum system in an eigenstate $|n(\mathbf{R})\rangle$ is transported adiabatically around a closed loop $C$ in parameter space. In addition to the familiar dynamical phase, the state acquires the phase
$$\gamma_n(C) = \oint_C \mathbf{a}_n \cdot d\mathbf{R}, \qquad \mathbf{a}_n = i\langle n(\mathbf{R}) | \nabla_\mathbf{R}\, n(\mathbf{R}) \rangle,$$where $\mathbf{a}_n$ is the Berry connection — a gauge field on parameter space. The Berry phase is the holonomy of this connection. For a spin in a slowly rotated magnetic field Berry found $\gamma_m(C) = -m\,\Omega(C)$, where $m$ is the spin component along the field and $\Omega(C)$ is the solid angle that the field direction sweeps out [[5]]. A cat righting itself and a quantum state accumulating a geometric phase use closely related differential-geometric language, though their connections and physical assumptions are different.
For deformable bodies, Shapere and Wilczek (1989) showed how a gauge structure over the space of shapes gives the net change in orientation that a sequence of deformations produces in a body with zero angular momentum [[6]]. Their earlier work used the same structure for microorganisms swimming at low Reynolds number [[4]]. Montgomery credits Guichardet’s 1984 paper on rotating molecules with first making the formula explicit [[4]]. The falling cat is the classical counterpart of a geometric-phase construction, not the same physical effect.
The Foucault pendulum turns its plane of swing relative to the ground by $2\pi\sin\phi$ per sidereal day, where $\phi$ is the latitude, clockwise in the northern hemisphere. Geometrically, the plane of swing is carried without twisting round the circle of latitude, and a vector parallel-transported round a closed curve on the unit sphere returns turned by the solid angle it encloses (Fig. 2). For the polar cap above latitude $\phi$ that solid angle is
$$\Omega = 2\pi(1 - \sin\phi).$$The two angles describe the same final orientation, reached by turning in opposite senses, and they add up to a full turn by the Gauss–Bonnet theorem for the cap: the geodesic curvature of the circle of latitude, integrated once round, is $2\pi\sin\phi$, and the curvature of the sphere integrated over the cap is $\Omega$. At latitude 30° both are half a turn, and a model cat with $J/I = 1$ and a 120° bend flips over for the same geometric reason. Foucault’s precession is another geometric phase.
The Aharonov–Bohm effect (1959) produces a phase shift for electrons circling a solenoid, even when the electrons travel only through field-free regions. The phase is $q\Phi/\hbar$ for charge $q$ and enclosed magnetic flux $\Phi$, the holonomy of the electromagnetic vector potential around the loop. Berry showed that it is a special case of his geometric phase [[5]].
All four phenomena — the falling cat, the Berry phase, the Foucault pendulum, the Aharonov–Bohm effect — can be formulated using connections and holonomy. That shared mathematical language is the connection; it does not erase their different dynamics, symmetries or experimental regimes.
Batterman (2003) gives a particularly clear account of this unification from the philosophy of physics, drawing out the common mathematical skeleton and asking what it means that such phenomena seem to need, for their explanation, the geometry of an abstract space of parameters [[7]].
Astronauts, Robots and Divers
Astronauts: Kane and Scher’s work was supported in part by a NASA research grant [[1]], and in 1970 they published an analysis of limb movements by which a weightless astronaut can turn in pitch and yaw [[8]].
Robotics: By 1993 the problem had given its name to a volume on control, Dynamics and Control of Mechanical Systems: The Falling Cat and Related Problems [[4]]. More recent legged robots have demonstrated cat-inspired aerial reorientation in simulation and hardware: Kurtz and colleagues land the Mini Cheetah quadruped on its feet from initial pitch angles between −90° and 90° [[9]]. Their controller need not be derived from Montgomery’s connection.
Gymnastics and diving: Human athletes performing somersaults and twists also exploit changing body shape while conserving angular momentum, usually a nonzero one. A tuck increases rotation rate (smaller $I$, constant $L$ → larger $\omega$); a layout decreases it. That is ordinary dynamics, not holonomy. Twist is closer to the cat. Without somersault, several cycles of counter-rotating body segments are needed for one twist; in a somersault, asymmetric movements of the arms, chest or hips tilt the body and start the twist [[10]]. With nonzero angular momentum, Eq. (1) gains a drift term, so the turn is no longer purely geometric [[4]].
The View from a Windowsill
My cats have no opinion about fibre bundles. When one of them drops from the top of the bookcase, she is not solving the variational problem
$$\min_{\gamma}\ \frac{1}{2}\int_0^T \lVert\dot\gamma(t)\rVert^2 \, dt \quad \text{subject to } \gamma(0) = \gamma(T), \ \ \mathrm{Hol}_\mathcal{A}(\gamma) = R_{180^\circ},$$over loops $\gamma$ in shape space with a metric such as kinetic energy; it is Montgomery’s optimal-control problem [[4]]. She is executing a motor program refined over millions of years of feline evolution. Her vestibular system reports on her orientation, and her nervous system turns that into a sequence of bends; the whole manoeuvre is over in a fraction of a second.
What physics tells us is that the manoeuvre is possible — that no law of nature forbids a body with zero angular momentum from reorienting — and gives the precise geometric reason: the curvature of a connection on shape space is non-zero, which means the holonomy of closed loops is non-trivial.
The same mathematical vocabulary helps explain a cat’s righting, a quantum geometric phase, Foucault precession and the Aharonov–Bohm interference shift. They are not one physical mechanism; they are different systems in which a path can have a consequence beyond its endpoints.
I find this more remarkable than the cat.
Literature and calculations checked through 2026-10-24.
Why gravity does not spoil the conservation law assumed here is the subject of a follow-up on Noether’s theorem.
References
[1] Kane, T. R., & Scher, M. P. (1969). A dynamical explanation of the falling cat phenomenon. International Journal of Solids and Structures, 5(7), 663–670. DOI: 10.1016/0020-7683(69)90086-9
[2] Gbur, G. J. (2019). Falling Felines and Fundamental Physics. Yale University Press. DOI: 10.12987/9780300249071
[3] Marey, É.-J. (1894). Des mouvements que certains animaux exécutent pour retomber sur leurs pieds lorsqu’ils sont précipités d’un lieu élevé. Comptes rendus hebdomadaires des séances de l’Académie des sciences, 119, 714–717.
[4] Montgomery, R. (1993). Gauge theory of the falling cat. In M. Enos (Ed.), Dynamics and Control of Mechanical Systems: The Falling Cat and Related Problems (Fields Institute Communications, Vol. 1, pp. 193–218). American Mathematical Society. DOI: 10.1090/fic/001/09
[5] Berry, M. V. (1984). Quantal phase factors accompanying adiabatic changes. Proceedings of the Royal Society of London A, 392(1802), 45–57. DOI: 10.1098/rspa.1984.0023
[6] Shapere, A., & Wilczek, F. (1989). Gauge kinematics of deformable bodies. American Journal of Physics, 57(6), 514–518. DOI: 10.1119/1.15986
[7] Batterman, R. W. (2003). Falling cats, parallel parking, and polarized light. Studies in History and Philosophy of Modern Physics, 34(4), 527–557. DOI: 10.1016/S1355-2198(03)00062-5
[8] Kane, T. R., & Scher, M. P. (1970). Human self-rotation by means of limb movements. Journal of Biomechanics, 3(1), 39–49. DOI: 10.1016/0021-9290(70)90049-7
[9] Kurtz, V., Li, H., Wensing, P. M., & Lin, H. (2021). Mini Cheetah, the falling cat: A case study in machine learning and trajectory optimization for robot acrobatics. arXiv:2109.04424 (ICRA 2022). arXiv:2109.04424
[10] Yeadon, M. R. (1993). The biomechanics of twisting somersaults. Part III: Aerial twist. Journal of Sports Sciences, 11(3), 209–218. DOI: 10.1080/02640419308729987
Changelog
- 2025-12-15: Corrected the Marey publication date from 22 November 1894 to 10 November 1894 (in text and in reference). Updated the Whitney & Mehlhaff (1987) statistics to reflect that the 90% survival rate applies to all cats in the study, as reported in the paper, rather than specifically to those falling from above seven stories.
- 2026-07-11: Corrected the description of the Kane–Scher model and narrowed claims about historical photographs, spin-free releases, aerodynamic effects, robotics, and the relationship between classical and quantum geometric phases. Removed the high-rise-syndrome digression because its clinical sample cannot establish a general fall-risk curve or its proposed mechanism.
- 2026-10-03: Corrected the authorship of the Mini Cheetah falling-cat reference (arXiv:2109.04424) from “Libby, T., et al.” to Kurtz, Li, Wensing and Lin.
- 2026-10-13: Added a forward reference to What Gravity Leaves Alone, scheduled for 22 January 2027. This companion was not available when this article first appeared; the reference remains plain text until its release.
- 2026-10-24: Corrected the connection, which the post called a one-form on shape space: it is a one-form on configuration space and lives on shape space only in a chosen gauge. Corrected “the first rigorous mechanical model”: Kane and Scher (1969) credit Rademaker and ter Braak (1935). Corrected the claim that tuck–layout timing produces twist; twist needs counter-rotation or asymmetric limb movements (Yeadon 1993). Explained the complementary Foucault angles by the Gauss–Bonnet theorem rather than by the local frame completing a circuit. Narrowed the Shapere–Wilczek claim to their 1989 result for bodies with zero angular momentum. Cited Marey’s Comptes rendus note, whose title the reference carried under La Nature. Replaced unsourced claims (a 0.3 s fall, the cerebellum, early robot designs) and retitled “Robotics and Spacecraft”. Added Montgomery’s no-twist results, the turn per bending loop (a half-turn needs about 59° at J/I = 0.25), Kane and Scher’s astronaut work and Berry’s spin result.