[Physics FAQ] - [Copyright]
Original by Michael Weiss.
End note added by Don Koks, 2013.
This old chestnut predates GR. The trail starts with Max Born's notion of an "SR rigid body", a relativistic replacement for the classical notion. It leads past Einstein's discussion in his great 1916 paper on GR, where he uses the rotating disk to introduce noneuclidean geometry. It then twists and turns as Eddington, Lorentz, and lesser lights attempt to compute the fate of the rigid disk.
In this entry, I summarize what I know of the literature, and invite others to fill in the gaps. Surely such a celebrated problem should have found a definitive resolution by now. But the tale I have to tell ends on an incomplete note. We look at SR first, then GR.
Born rigidity: In 1909, Born proposed a Lorentz-invariant definition of a "rigid body". Pauli's monograph on relativity  gives a nice summary of Born's idea, and the responses it drew from Ehrenfest, Herglotz, Noether, and von Laue. (Pais's Einstein bio suggests that Born's 1909 paper may have helped set Einstein on the road to riemannian geometry .)
We know already that rigidity and SR don't mix—just think of the barn and the pole! How could a physicist like Born, mathematically sophisticated, have made such an elementary error? A simple remark by Pauli clarifies things considerably:
If thus the concept of a RIGID BODY has no place in relativistic mechanics, it is nevertheless useful and natural to introduce the concept of a RIGID MOTION of a body. We shall denote those motions as rigid for which Born's condition (*) is satisfied.
Born thought he was defining a rigid body, but Pauli's rephrasing saves the mathematics while improving the physics. We have no rigid rods in SR, but if you accelerate every atom of an ordinary rod in just the right way, you can move the rod rigidly. And Born's definition is Lorentz invariant.
I won't plunge right into Born's definition (as Pauli does). Instead I'll approach it by thinking atomistically. Imagine our solid as made up of a large number of atoms A1, ..., An. Between any two nearby atoms Ai and Aj there is a "natural distance" dij, natural in the sense that if Ai and Aj are pushed together or pulled apart, stresses result, trying to restore the distance dij. Of course there are propagation delays, but if we start with the solid at rest in some inertial frame, and accelerate it gently, the resulting elastic waves in the solid should die out pretty quickly. Or we can pretend that exactly the right force is applied to each atom at all times, so that natural distances are preserved. Let the number of atoms tend to infinity (continuum approximation), let the stress/strain ratio tend to infinity, and apply forces gently enough so that the elastic waves can be ignored. In such a case we arrive at Born's definition. Born used coordinates, but I'll try for a coordinate-free rephrasing.
First, what is a solid? Is it just a swath in spacetime? This is not enough: the atomistic viewpoint suggests we should be able to "mark" a point inside the solid (call it a particle) and follow its worldline. An "event", as usual, is a point in spacetime. The events along a particle's worldline are parametrized by τ, the time on that particle's clock.
(*) Pick a particle (call it A) in the solid. Pick an event p at time τ on the worldline of A. Draw the spatial plane orthogonal to the worldline at p (i.e., the plane of simultaneity at that event).
Now pick another particle B. The plane of simultaneity intersects the world lines of these two particles A and B at two events; let s(τ,A,B) be the interval between these two events.
Suppose that for any A and B infinitesimally close to one another, ds(τ)/dτ = 0 for all τ. Then we say the body MOVES RIGIDLY.
If you don't like the notion of "infinitesimally close", there are ways to get around it, but I won't go into that. (The basic idea is that we are using particle world lines to transport the metric from one tangent plane to another.)
OK, now what? First, it should be pretty clear that Born's definition captures the idea that the body moves without internal stresses. Or if you prefer, you can say that we have nearly rigid motion when Young's modulus is large enough, and the accelerations are gentle enough, so that infinitesimal pieces of the body are barely deformed, when viewed in the comoving frame of reference. Born rigidity is then the limiting case.
If we accelerate a rod rigidly in the longitudinal direction, then the rod suffers the usual Lorentz contraction. More generally, rigid motion without any twisting corresponds to so-called Fermi–Walker transport (see MTW ). The acceleration of the front of the rod is less than the acceleration of the rear; this is a variation on Bell's Spaceship Paradox (see ; and the Relativity FAQ entry.)
Ehrenfest noted that a disk cannot be brought from rest into a state of rotation without violating Born's condition. Integrating τ out of Born's condition, we see that infinitesimally close particles must keep the same proper distance. So in the original rest frame, they suffer Lorentz contraction in the transverse direction but none in the radial direction. The circumference contracts but the radius doesn't. But in the original rest frame, the circumference is a circle, sitting in a spatial slice (t = constant) of ordinary flat Minkowski spacetime. In other words, we would have a "noneuclidean circle" sitting in ordinary Euclidean space. This is a contradiction.
The issue of spatial slices deserves a few words. The particles in a rotating disk (not assumed rigid) cannot agree on a global notion of simultaneity. For if you make a circuit around the edge, joining up the infinitesimal planes of simultaneity, when you return to your starting point, the planes no longer match up. This makes it problematic to talk about geometry "as seen by the particles" (or by observers standing on the disk).
I've talked about making complete circuits about the center, but both Ehrenfest's argument and the simultaneity problem have local versions. Say we have a ball bearing a light-year from the Sun. We cannot put the ball bearing in orbit around the Sun, keeping one face to the Sun, without violating Born's condition. Nor can the particles of the ball bearing agree on a notion of simultaneity.
The proofs are not hard. I'll sketch the local version of Ehrenfest's argument. Take a spatial slice in the rest frame and look at the metric. The simplest way to express it is to use polar coordinates inherited from when the ball bearing was at rest. Each particle was then labelled with coordinates (r,θ), and we can use these to label its whole worldline, and thus also the points in the spatial slice. The same "transverse versus radial" argument that Ehrenfest used shows that:
ds2 = dr2 + r2 dθ2/(1 − r2ω2)
where ω is angular speed. A routine computation shows that the curvature is nonzero. This contradicts the fact that the spatial slice is Euclidean space.
Pauli states: "It was further proved, independently, by Herglotz and Noether that a rigid body in the Born sense has only three degrees of freedom... Apart from exceptional cases, the motion of the body is completely determined when the motion of a single of its points is prescribed." I haven't looked up the papers by Herglotz and Noether, but I dare say it's similar to the argument above.
Max von Laue pointed out that a body made up of n point particles must have at least 3n degrees of freedom. Say we give an impulse to each particle at t = 0. Because of the finite speed of light, there can be no constraints relating the velocities of different particles. Rigid motion can occur in SR only through a conspiracy of forces.
So much for the rotating rigid disk in SR.
Einstein's 1916 paper on GR  makes no mention of elevators; instead, the Equivalence Principle is introduced via the rotating disk. Einstein reproduces Ehrenfest's argument, but with a different conclusion: since we are no longer assuming flat Minkowski space, Einstein asserts that geometry for the rigid rotating disk is noneuclidean. The Equivalence Principle now implies that geometry in a gravitational field will also be noneuclidean. (By "geometry", I mean spatial geometry, i.e., we're not concerned with the temporal components of the spacetime metric.)
Can we make any sense of Einstein's argument? The simplest interpretation makes a couple of assumptions:
dτ2 = dt2 − f(r) dz2 − g(r) dr2 − h(r) dθ2
Assumption (2) seems a lot more dubious than (1), but it does allow us to talk about spatial slices (t=constant), and hence the geometry of the spinning disk. We appeal to axial symmetry and steady-state conditions in making f,g, and h independent of θ and t. We have no such justification for leaving out z.
Einstein doesn't give an explicit formula for the spacetime metric of the rigid spinning disk, but here's one obvious candidate (assuming the angular speed is ω):
dτ2 = dt2 − dz2 − dr2 − r2 dθ2/(1 − r2ω2)
Turn to GR. Now all sorts of complications appear.
To settle the question definitively, it seems one has to perform a full-blown, hairy GR calculation. Perhaps someone has done this; perhaps someone has turned the vague notion of "infinitely rigid" into a formula for a stress–energy tensor, plugged that into the Einstein field equations, and solved. If the Gentle Reader knows of a reference, please let me know.
I have retained the above discussion for its historical interest, but I have always felt that the real question the physicists above meant to address is the description of a series of events in flat spacetime from the viewpoint of a rotating frame. (Sure, after sorting that out, we could then turn our attention to curved spacetime, but flat spacetime is tough enough.) I think the above discussion of the rotating disk got derailed from the start when it began to focus on the structure of the disk. The rotating disk is merely a device for helping us to picture and think about the rotating frame, but discussion of how the disk rotates, how it gets accelerated and so on, are completely irrelevant to the analysis of a rotating frame within the context of special relativity, which has to do with what is simultaneous with what; for that, we need only appeal to the Clock Postulate to construct planes of simultaneity at each event of interest. This really has nothing to do with general relativity, which is a theory of gravity. If one draws planes of simultaneity at various events in a rotating system such as a disk, then one can form a coherent picture of those events. Much care must be taken with the mathematical details, but these are an expected part of applying the machinery of special relativity.
The same idea is universally taken for granted in the standard Lorentz transform studied throughout special relativity. There too, we don't concern ourselves—no one does—with how the "primed frame" was ever made to move. It is simply taken as having always been moving, forever. Any other treatment of its motion would only mask and complicate the underlying ideas, which concern the Lorentz transform and not questions of how to accelerate physical objects. We are content to treat a constant-velocity primed frame as having had its state of motion forever, so we should do likewise for the rotating frame and not allow discussion of how the disk was set into motion derail the core issue, which is the analysis of events in a rotating frame.
Of course, a study of how objects are accelerated is certainly important and fruitful in special relativity. For example, when discussing Bell's Spaceship Paradox we would do well to ask ourselves how the string connecting the two spacecraft behaves; it has mass and it needs to be accelerated, and we can't just treat it as a series of massless events on a spacetime diagram that merrily go along for the ride as the spaceships accelerate. It interacts with them, so we shouldn't be surprised to find that eventually it breaks as it becomes more difficult to accelerate, when the ships' speeds get close to the speed of light. So a study of accelerating physical objects has its place in special relativity, and there's nothing wrong with asking how a disk is accelerated. But just as we do for the standard Lorentz transform, let's treat such an advanced dynamical question as secondary to simply asking how a rotating frame views the world. The above discussion of the rotating disc seems never to have managed that.
It might at first be thought that the rotating frame gives problems with the speed of light. For instance if we spin around on the spot, we see the Moon whizz around us in a huge circle, so isn't it travelling faster than light in our frame? It is travelling faster than the tabulated value of "c", but it is not travelling faster than light; after all, we agree that light is still escaping from its surface. In the language of special relativity, the Moon's world line always remains within its local light cone; it remains "timelike". In fact, this behaviour is no different to the well-accepted behaviour of light in an accelerated frame, where the measured speed of light depends upon where in that frame it currently is. This speed can in fact have any value, from zero to infinity. (See the FAQ entry Do moving clocks always run slowly? for further discussion on this.)
You can find it occasionally written, such as in one textbook and on at least one Wikipedia page—as far as I can ascertain beneath all that obscure and standard Wikipedia gravitas!—that after representing spacetime on the disk's edge by a cylinder (which is certainly valid), we can just cut the cylinder on a line parallel to the time axis, flatten the cylinder out, and draw some infinite lines of simultaneity. This is simply naïve. Special relativity is not about cutting spacetime into pieces to suit whatever analysis we wish to make a guess at. Rather, it's about making a straightforward-but-sometimes-hard analysis of the simultaneity of events.
 Wolfgang Pauli, Theory of Relativity, pages 130–134, Pergamon Press, 1958.
 Abraham Pais, Subtle is the Lord: the Science and Life of Albert Einstein, pg 214.
 Misner, Thorne, and Wheeler, Gravitation, pg 170.
 J.S.Bell, Speakable and Unspeakable in Quantum Mechanics, pg 67, Cambridge University Press, 1987.
 The Principle of Relativity, Dover.
 G. Cavalleri, Nuovo Cimento 53B pg 415.
 O. Gron, AJP Vol. 43 No. 10 pg 869 (1975)
 C. Berenda Phys. Rev. 62 pg 280 (1942)