Thomas Precession

Put a gyroscope in a storage ring. Do not touch it — no torque, no field, nothing acting on its spin. Let it go round. It comes back turned. And at $\beta = 0.6$ it comes back turned by exactly $90^\circ$ — a quarter turn, per lap, against the direction of travel. Nothing rotated it. The gyroscope's own instantaneous rest frames did, because the boosts that string them together do not compose into a boost. `aq-sr-thomas-wigner-rotation` proves the finite version: two non-collinear boosts leave behind a genuine rotation, $12.68^\circ$ at $\beta_1 = \beta_2 = 0.6$ and a right angle between them. This page owns the rate — what happens when you take not two boosts but a continuum of them, along a curved worldline. The answer is an angular velocity $\boldsymbol\omega_{\mathrm{T}}$ , and its most important property is not its size but its sign: on any circular orbit it points against the orbital angular momentum. It is retrograde. That minus sign is worth a factor of two in the spectrum of hydrogen, and in the muon $g-2$ ring it cancels 96.6% of the spin precession a naive calculation predicts. It is not a correction. It is one of the two big terms — the one nobody put there.

Key takeaways

Signature $(+,-,-,-)$ : $\mathrm{d}s^2 = c^2\mathrm{d}t^2 - \mathrm{d}x^2$ , $u\cdot u = +c^2$ , timelike positive. Conventions declared before the sign is derived, because the sign is the whole point.
The owned result — the RATE. For a particle with lab velocity $\mathbf v$ and lab acceleration $\mathbf a$ , a torque-free spin obeys $\;\mathrm{d}\mathbf S/\mathrm{d}t = \boldsymbol\omega_{\mathrm{T}}\times\mathbf S\;$ with $\;\boxed{\boldsymbol\omega_{\mathrm{T}} = \dfrac{\gamma^2}{\gamma+1}\,\dfrac{\mathbf a\times\mathbf v}{c^2}}$ . `aq-sr-thomas-wigner-rotation` quotes this once and cedes the derivation here.
THE SIGN, and it is not a convention. For a circular orbit $\mathbf a\times\mathbf v$ is antiparallel to $\mathbf L$ . The precession is RETROGRADE. The base for this page prints the rate three times and contradicts itself: its boxed card has $\mathbf a\times\mathbf v$ , its two derivations have $\mathbf v\times\mathbf a$ and $\mathbf v\times\mathrm{d}\mathbf v$ . Two independent machine checks — explicit $4\times4$ boost composition, and Fermi–Walker transport of a real spin — both return $\mathbf a\times\mathbf v$ . The derivations are wrong.
Circular motion, exactly. $\;\boxed{\boldsymbol\omega_{\mathrm{T}} = -(\gamma-1)\,\boldsymbol\omega_{\text{orb}}}\;$ — $\gamma^2\beta^2/(\gamma+1) \equiv \gamma-1$ is an identity, not an expansion. Per lap the spin turns $\;\boxed{2\pi(\gamma-1)}\;$ backwards; at $\beta = 0.6$ that is $\pi/2$ , a quarter turn per orbit.
The limits. $\gamma^2/(\gamma+1) \to \tfrac12$ as $\gamma\to1$ — the Thomas factor, which halves the spin–orbit coupling — and $= (\gamma-1) + \frac{1}{\gamma+1} \to \gamma - 1$ as $\gamma\to\infty$ . A sibling base prints this asymptote as $\gamma - \tfrac12$ . It is $\gamma - 1$ .
Key idea: nothing is torquing the gyroscope, and nothing needs to. The rotation is in the map between successive rest frames, not in the object. Boosts do not form a group; a particle on a curved worldline is dragged around a closed loop in velocity space; and the holonomy of that loop is a real, retrograde, measurable turn of a real axis.

The core insight

The Thomas precession is a holonomy. Velocity space in special relativity is hyperbolic — negatively curved — and a particle on a curved worldline traces a closed circuit in it. Parallel-transport an orientation around a closed circuit on a curved surface and it returns rotated by the enclosed curvature. There is no force in the story anywhere: the gyroscope does nothing but exist, honestly, in each frame it passes through.

Where this lives today

Every storage-ring spin measurement is built on this equation. In the muon $g-2$ experiments at Brookhaven and Fermilab the Thomas precession is no small correction inside the model — it cancels $96.6\%$ of the magnetic precession, and the residue is the signal. Fermilab's final result (June 2025) reaches $127$ parts per billion on $a_\mu$ ; a sign error in the Thomas term would not perturb that result, it would destroy it.

And in accelerator physics generally. The spin tune of a stored beam is $\nu_s = a\gamma$ , whose $\gamma$ -dependence is a direct fingerprint of the Thomas term's near-cancellation with the Larmor term. Resonant depolarisation exploits it to calibrate beam energy to parts in $10^{5}$ — it is how LEP knew its own energy well enough to weigh the $Z$ .

The physical picture

Take the electron round its Bohr orbit. At each instant jump into its rest frame with a pure boost $B(\mathbf v)$ ; a moment later its velocity is $\mathbf v + \mathrm{d}\mathbf v$ and you jump again. Were the two boosts collinear this would be uneventful — collinear boosts commute, compose into a pure boost, and land you on the axes you expected. They are not collinear. On a curved orbit $\mathrm{d}\mathbf v$ has a component perpendicular to $\mathbf v$ — that is what "curved" means — and `aq-sr-thomas-wigner-rotation` proves the consequence: a pure boost is a symmetric matrix, $(B_2B_1)^{\mathsf T} = B_1B_2$ , so the product of two non-commuting boosts is not symmetric, hence not a pure boost. What is left over is a rotation about $\mathbf v_1\times\mathbf v_2$ . That is canon, and it is not re-derived here. What is derived here is what happens over a whole orbit. Each infinitesimal step contributes an infinitesimal rotation, and they all share the same axis — perpendicular to the orbital plane — so they do not cancel. They add. After one lap the accumulated rotation is finite, and a gyroscope nobody touched is pointing somewhere new. And it is pointing backwards. Set up the orbit: $\mathbf r = R(\cos\omega t,\ \sin\omega t,\ 0)$ , so $\mathbf v = R\omega(-\sin\omega t,\ \cos\omega t,\ 0)$ and $\mathbf a = -\omega^2\mathbf r$ — the acceleration points inward, at the centre, as centripetal acceleration must. The orbital angular momentum is $\mathbf L \propto \mathbf r\times\mathbf v \propto +\hat{\mathbf z}$ . And

\mathbf a\times\mathbf v = -\omega^2\,\mathbf r\times\mathbf v \;\propto\; -\hat{\mathbf z} .

$\mathbf a\times\mathbf v$ is antiparallel to $\mathbf L$ . No convention was used and none can rescue it: the acceleration is centripetal, and its cross product with the velocity necessarily opposes the angular momentum. So whatever prefactor sits in front of $\mathbf a\times\mathbf v$ , the precession runs the wrong way round the orbit — and that, three articles later, is why the spin–orbit coupling in hydrogen comes out half the size a naive calculation gives.

Figure

PUNCHLINE — the precession runs BACKWARDS, and at β = 0.6 it is a quarter turn per lap. The particle (blue) goes anticlockwise; its orbital angular momentum L points out of the page. The centripetal acceleration a points INWARD, so a × v necessarily points INTO the page — antiparallel to L. No convention can change that: it follows from the geometry of a circle. The red gyroscope arrows show the rest-frame spin at four points of one orbit, and they rotate CLOCKWISE — against the motion. RETROGRADE. The turn per lap is exactly 2π(γ − 1), which at β = 0.6 (γ = 1.25) is exactly π/2 = 90.000000°, verified by numerically Fermi–Walker transporting a real spin all the way round. Nothing torqued it. In hydrogen the same formula gives 1.673 × 10⁻⁴ rad per orbit — tiny, and worth a factor of two in the spectrum

The core equations

Conventions first, because the sign is the result. $B(\mathbf v)$ is the symmetric pure boost carrying a rest four-vector to velocity $\mathbf v$ — the convention of `aq-sr-thomas-wigner-rotation`. $\boldsymbol\omega$ acts actively: $\delta\mathbf x = \boldsymbol\omega\times\mathbf x$ . $\mathbf S$ is the spin in the particle's instantaneous rest frame, compared against lab axes. With those three sentences fixed, everything below is forced.

\boxed{\;\frac{\mathrm{d}\mathbf S}{\mathrm{d}t} = \boldsymbol\omega_{\mathrm{T}}\times\mathbf S, \qquad\qquad \boldsymbol\omega_{\mathrm{T}} = \frac{\gamma^{2}}{\gamma+1}\,\frac{\mathbf a\times\mathbf v}{c^{2}}\;} \qquad (1)

where: $\mathbf v$ , $\mathbf a = \mathrm{d}\mathbf v/\mathrm{d}t$ are the lab velocity and acceleration, $\gamma = (1-v^2/c^2)^{-1/2}$ . Physical reading: there is no torque in this equation. It is not a dynamical law; it is a kinematic bookkeeping identity, saying that the frames the spin is honestly at rest in are themselves turning. Note $\mathbf a\times\mathbf v$ — not $\mathbf v\times\mathbf a$ ; the base prints both, in different cards, and the two derivations below settle it independently. Collinear acceleration kills it outright — $\mathbf a \parallel \mathbf v \Rightarrow \mathbf a\times\mathbf v = 0$ — because collinear boosts commute. A rocket accelerating dead ahead never twists its gyroscopes, however hard or however long it burns. Only a change in the direction of motion produces the effect, which is why it is invisible in every one-dimensional problem and why it took an orbiting electron to find it. For uniform circular motion, $|\mathbf a| = v\,\omega_{\text{orb}}$ with $\mathbf a\perp\mathbf v$ , so $|\mathbf a\times\mathbf v| = v^2\omega_{\text{orb}}$ and the prefactor collapses:

\frac{\gamma^{2}}{\gamma+1}\,\frac{v^{2}}{c^{2}} = \frac{\gamma^{2}\beta^{2}}{\gamma+1} = \frac{\gamma^{2}-1}{\gamma+1} = \gamma-1 \qquad\Longrightarrow\qquad \boxed{\;\boldsymbol\omega_{\mathrm{T}} = -(\gamma-1)\,\boldsymbol\omega_{\text{orb}}\;} \qquad (2)

where: the minus sign is the retrograde sense, carried over as a vector equation. Physical reading: $\gamma^2\beta^2 = \gamma^2 - 1$ exactly, so this is an identity, not an expansion — good at $\beta = 0.999$ as at $\beta = 0.001$ . (The base prints the magnitude correctly and then writes it as a vector equation without the minus sign, two cards below the card that has the sign right.) Integrate (2) over one period $T = 2\pi/\omega_{\text{orb}}$ :

\boxed{\;\Delta\phi_{\text{per orbit}} = -\,2\pi(\gamma-1)\;} \qquad (3)

where: negative means against the orbital sense. Physical reading: the gyroscope does not come home. At $\beta = 0.6$ , $\gamma = 1.25$ exactly and $\Delta\phi = -2\pi(0.25) = -\pi/2$ — a quarter turn per lap, retrograde, in exact rationals. For hydrogen, $\gamma - 1 = 2.66267\times10^{-5}$ gives $1.6730\times10^{-4}$ rad per orbit — the number `aq-sr-thomas-wigner-rotation` prints, recomputed here from a different formula.

Deriving the rate — twice, by two unrelated routes

Derivation AThe rate from the composition of boosts, with every convention nailed down before the sign appears

The rate comes from composing two infinitesimally different boosts, and the difficulty is entirely bookkeeping: the literature prints $\mathbf a\times\mathbf v$ and $\mathbf v\times\mathbf a$ as though interchangeable, and the base for this page prints both, in adjacent cards. The strategy here is to fix every convention first, in words, so that no step can quietly flip a sign; then compose two explicit matrices; then read the rotation off the antisymmetric part, which is a definition and not a choice. The non-closure theorem itself is owned by `aq-sr-thomas-wigner-rotation` and is used, not proved.

Step 1 — declare the conventions, in full. $B(\mathbf v)$ is the symmetric pure boost with $B^0{}_0 = \gamma$ , $B^0{}_i = B^i{}_0 = \gamma\beta_i$ , $B^i{}_j = \delta_{ij} + (\gamma-1)\beta_i\beta_j/\beta^2$ — canon 1902's convention, with $B(\mathbf v)^{-1} = B(-\mathbf v)$ . A small rotation acts on the spatial block as $R \approx \mathbb{1} + [\boldsymbol\omega]_\times$ , so that $\delta\mathbf x = \boldsymbol\omega\times\mathbf x$ . Why: this step is usually skipped, and it is where the sign is decided. With $\boldsymbol\omega$ pinned to the active rotation of a vector, "which way does it turn" has an answer that can be checked against a picture.

Step 2 — compose. Form the Lorentz transformation relating the rest frame at $t$ to the rest frame at $t + \mathrm{d}t$ :

\Lambda = B(\mathbf v + \mathrm{d}\mathbf v)\;B(-\mathbf v).

Were boosts closed under composition, $\Lambda$ would be a pure boost and its spatial block symmetric. It is not. Why: this is precisely the two-boost problem of `aq-sr-thomas-wigner-rotation`, taken in the limit where the second boost is infinitesimal — the same theorem, differentiated.

Step 3 — read off the rotation. The spatial block $S = \Lambda^{i}{}_{j}$ has antisymmetric part $A = \tfrac12(S - S^{\mathsf T})$ , and by Step 1 the rotation vector is $\mathrm{d}\boldsymbol\Omega = (A_{32},\,A_{13},\,A_{21})$ . Building $\Lambda$ explicitly for circular motion at $\beta = 0.6$ with $\mathrm{d}\mathbf v = \mathbf a\,\mathrm{d}t$ returns

\mathrm{d}\boldsymbol\Omega = \frac{\gamma^{2}}{\gamma+1}\,\frac{\mathbf a\times\mathbf v}{c^{2}}\,\mathrm{d}t \qquad\Longrightarrow\qquad \boldsymbol\omega_{\mathrm{T}} = \frac{\gamma^{2}}{\gamma+1}\,\frac{\mathbf a\times\mathbf v}{c^{2}} .

Why: $\mathbf a\times\mathbf v$ , and the machine says so. With $\mathbf v = 0.6c\,\hat{\mathbf x}$ and $\mathrm{d}\mathbf v \parallel \hat{\mathbf y}$ the computed $\mathrm{d}\boldsymbol\Omega$ is $-2.5\times10^{-7}\hat{\mathbf z}$ , while $(\gamma^2/(\gamma+1))(\mathbf v\times\mathrm{d}\mathbf v)/c^2$ is $+2.5\times10^{-7}\hat{\mathbf z}$ — the opposite sign, and it is what the base's two derivation cards print. The base's boxed equation, which has $\mathbf a\times\mathbf v$ , agrees with the matrices and with canon.

Step 4 — sanity-check the prefactor at both ends. $\gamma^2/(\gamma+1) \to \tfrac12$ as $\gamma\to1$ : the Thomas factor, and `aq-sr-thomas-factor-spin-orbit` cashes it. At the other end, divide out: $\gamma^2 = (\gamma+1)(\gamma-1) + 1$ , so

\frac{\gamma^{2}}{\gamma+1} = (\gamma-1) + \frac{1}{\gamma+1} \;\xrightarrow{\ \gamma\to\infty\ }\; \gamma - 1 .

Why: the ultra-relativistic asymptote is $\gamma - 1$ , not $\gamma - \tfrac12$ . A sibling base prints $\gamma - \tfrac12$ ; at $\gamma = 10$ the true value is $100/11 = 9.0909$ , and $\gamma - 1 = 9$ while $\gamma - \tfrac12 = 9.5$ . The error does not shrink — it tends to a constant $\tfrac12$ . It is corrected here. *Verified in verify_aq-sr-thomas-precession.py: $B(\mathbf v)$ checked symmetric and Lorentz ( $B^{\mathsf T}\eta B = \eta$ ); $\Lambda = B(\mathbf v+\mathrm{d}\mathbf v)B(-\mathbf v)$ built explicitly and its antisymmetric spatial part compared with both candidate signs, on a grid of speeds and orbital phases; $\mathbf a\times\mathbf v$ wins everywhere and $\mathbf v\times\mathbf a$ fails everywhere; and the identity $\gamma^2/(\gamma+1) = (\gamma-1) + 1/(\gamma+1)$ is confirmed symbolically, with $\gamma-\tfrac12$ shown to be wrong by $\tfrac12$ in the limit.*

Go deeper · Derivation B · 2 minThe same rate from Fermi-Walker transport of a real spin — no boosts, no composition, no conventions to get wrong

Derivation BThe same rate from Fermi-Walker transport of a real spin — no boosts, no composition, no conventions to get wrong

A derivation resting on a matrix convention invites the suspicion that its sign rests on the convention too. So here is a wholly independent route, in which no boost is ever composed with another: write down what "torque-free" means covariantly, integrate it along the worldline, and then simply look at where the spin points after one lap. The only inputs are that the spin is orthogonal to the four-velocity and has constant magnitude. If this route and the last agree — and they do, to ten digits — the sign is a fact about the world, not about the notation.

Step 1 — say what a torque-free gyroscope is. The spin four-vector $S^\mu$ satisfies $S\cdot u = 0$ (it is purely spatial in the rest frame) and $S\cdot S = -|\mathbf S|^2 = \text{const}$ . Differentiating the first: $\dot S\cdot u + S\cdot A = 0$ , where $A^\mu = \mathrm{d}u^\mu/\mathrm{d}\tau$ . A torque-free spin changes only as much as the constraint forces it to, so $\dot S^\mu \propto u^\mu$ ; fixing the constant with $u\cdot u = +c^2$ :

\boxed{\;\frac{\mathrm{d}S^\mu}{\mathrm{d}\tau} = -\frac{1}{c^{2}}\,(S\cdot A)\,u^\mu\;}

Why: this is Fermi–Walker transport, and it is the covariant definition of "not rotating". Note it has no free parameters and no reference to any boost: it follows from the two constraints and nothing else. In $(+,-,-,-)$ , $u\cdot u = +c^2$ — take that from $(-,+,+,+)$ and the sign of the whole equation flips.

Step 2 — integrate it round a circle. Take $\mathbf v(t) = \beta c(-\sin\omega t,\ \cos\omega t,\ 0)$ at $\beta = 0.6$ , build $u^\mu$ and $A^\mu$ , start the spin along $\hat{\mathbf x}$ in the rest frame, and integrate the boxed equation numerically for one full period, boosting $S^\mu$ back into the current rest frame at each instant to recover $\mathbf S$ . Why: the rest-frame spin is what an experiment measures. No approximation is made anywhere.

Step 3 — look at the answer.

\mathbf S(0) = (1,\,0,\,0) \qquad\longrightarrow\qquad \mathbf S(T) = (0,\,-1,\,0).

The spin has rotated by $-1.5707963268\ \mathrm{rad} = \mathbf{-90.000000^\circ}$ , and the orbital motion was anticlockwise. Why: that is a quarter turn, backwards, and it settles the sign for good. The predicted value is $-2\pi(\gamma-1) = -2\pi(0.25) = -\pi/2$ , and the integration returns it to ten significant figures. No boost was composed. No sign convention for $\varepsilon_{ijk}$ was invoked. The gyroscope simply came back turned.

Step 4 — extract the instantaneous rate and compare. Differentiating the rest-frame spin at $t = 0$ and solving $\mathrm{d}\mathbf S/\mathrm{d}t = \boldsymbol\omega\times\mathbf S$ gives $\boldsymbol\omega = -2.5\times10^{-3}\,\omega\,\hat{\mathbf z}$ for the test orbit, and

\frac{\gamma^{2}}{\gamma+1}\frac{\mathbf a\times\mathbf v}{c^{2}} = -2.5\times10^{-3}\,\omega\,\hat{\mathbf z}, \qquad -(\gamma-1)\boldsymbol\omega_{\text{orb}} = -2.5\times10^{-3}\,\omega\,\hat{\mathbf z} .

All three agree. Why: the boost-composition route, the Fermi–Walker route and the closed-form circular-orbit result are three independent computations that meet at the same vector, sign included. That is what it takes to be sure of a minus sign that the source material gets wrong three times out of four. *Verified in verify_aq-sr-thomas-precession.py: $u\cdot u = +c^2$ and $u\cdot A = 0$ asserted; $S\cdot u = 0$ and $|S|$ conserved along the whole integration; the one-orbit rotation returned as $-\pi/2$ at $\beta = 0.6$ to $10^{-9}$ ; the instantaneous $\boldsymbol\omega$ matched against $(\gamma^2/(\gamma+1))(\mathbf a\times\mathbf v)/c^2$ and against $-(\gamma-1)\boldsymbol\omega_{\text{orb}}$ over a sweep of $\beta$ ; and $\mathbf v\times\mathbf a$ asserted to FAIL at every point.*

Common trapThe base contradicts itself on the sign, in print, three cards apart

The AllQFT base for this page states the precession rate three times:

card	what it prints	verdict
eq[0], boxed	$\boldsymbol\omega_{\mathrm{T}} = \frac{\gamma^2}{\gamma+1}\frac{\mathbf a\times\mathbf v}{c^2}$	correct — agrees with canon and with both machine checks
DERIV[0] step 4	$\mathrm{d}\boldsymbol\Omega = \frac{\gamma^2}{\gamma+1}\frac{\mathbf v\times\mathrm{d}\mathbf v}{c^2}$	sign-flipped
DERIV[1] step 1	$\boldsymbol\omega_{\mathrm{T}} = \frac{\gamma^2}{\gamma+1}\frac{\mathbf v\times\mathbf a}{c^2}$	sign-flipped

and its own note on eq[0] papers over the contradiction with the phrase "sign convention: …when precession is defined as spin-relative-to-lab". It is not a convention. Two independent computations — explicit $4\times4$ boost composition, and Fermi–Walker transport of a real spin four-vector — both return $\mathbf a\times\mathbf v$ ; and the invariant statement needs no matrices at all: the centripetal acceleration points inward, so $\mathbf a\times\mathbf v$ opposes $\mathbf L$ , so the precession is retrograde — and were it not, it would add to the spin–orbit coupling instead of halving it, and hydrogen's fine structure would come out three times too large instead of twice. A separate defect, in the same base: eq[3] gives the circular-orbit magnitude as $(\gamma-1)\omega_{\text{orb}}$ — exactly right — and then writes it as a vector equation $\boldsymbol\omega_{\mathrm{T}} = (\gamma-1)\boldsymbol\omega_{\text{orb}}$ , dropping the minus sign that the card three above it had correct. Right scalar, wrong vector.

Common trapGravity Probe B did not measure the Thomas precession, and this article does not claim it did

It is often said that Gravity Probe B — four cryogenic gyroscopes in a $642\ \mathrm{km}$ polar orbit, reporting in 2011 — measured the Thomas precession. It did not. GP-B measured the geodetic (de Sitter) precession: $-6601.8 \pm 18.3$ mas/yr against a general-relativistic prediction of $-6606.1$ , plus frame-dragging at $-37.2 \pm 7.2$ against $-39.2$ (Everitt et al., Phys. Rev. Lett. 106, 221101). That is a general-relativistic effect. GP-B's gyroscopes were in free fall — zero proper acceleration — and equation (1) with $\mathbf a = 0$ gives exactly zero. The geodetic precession's full $3/2$ coefficient needs the spatial curvature of the Schwarzschild metric, which no amount of special relativity supplies and which the general-relativity track owns. A kinematic term of Thomas type does appear in its post-Newtonian expansion, but disentangling it is a GR calculation, not an SR one, and this page does not attempt it.

Where the Thomas precession IS measured, decisively: the fine structure of hydrogen (`aq-sr-thomas-factor-spin-orbit`) and storage rings (`aq-sr-bmt-equation` — where, as the worked example shows, it is not a correction but one of the two dominant terms).

Limiting cases

Case	Result	Meaning
$\mathbf a \parallel \mathbf v$	$\boldsymbol\omega_{\mathrm{T}} = 0$	straight-line acceleration never twists a gyroscope. Collinear boosts commute
$\gamma \to 1$	$\gamma^2/(\gamma+1) \to \tfrac12$	the Thomas factor. This is the number that halves the spin–orbit coupling
$\gamma \to \infty$	$\gamma^2/(\gamma+1) \to \gamma - 1$	not $\gamma - \tfrac12$ . Exactly, $(\gamma-1) + 1/(\gamma+1)$
circular orbit	$\boldsymbol\omega_{\mathrm{T}} = -(\gamma-1)\boldsymbol\omega_{\text{orb}}$	exact identity, all $\beta$ . Retrograde

Worked examples

1 · The muon storage ring — where the Thomas term cancels 96.6% of the physics

Muons circulate at $\gamma = 29.3$ in a uniform field of $B = 1.45\ \mathrm{T}$ (the BNL and Fermilab $g-2$ ring). Compute the Thomas precession, and compare it with the magnetic torque.

Step 1 — identify. $\omega_c = eB/(\gamma m_\mu)$ , with $e/m_\mu = 8.5061\times10^{8}\ \mathrm{C\,kg^{-1}}$ and $\gamma = 29.3034$ :

\omega_c = \frac{8.5061\times10^{8} \times 1.45}{29.3034} = \mathbf{4.2091\times10^{7}\ rad\,s^{-1}}, \qquad f_c = \mathbf{6.70\ MHz} \quad (\text{period } \mathbf{149.3\ ns}) .

That is the real ring: the muons go round every $149$ ns.

Step 2 — apply (2). The orbit is circular, so

|\boldsymbol\omega_{\mathrm{T}}| = (\gamma-1)\,\omega_c = 28.3034 \times 4.2091\times10^{7} = \mathbf{1.1913\times10^{9}\ rad\,s^{-1}}, \qquad\text{retrograde.}

The Thomas precession is $28$ times faster than the orbital motion. Nothing about this is a perturbation.

Step 3 — the boundary number. Now the magnetic torque. In the rest frame the muon sees $B' = \gamma B$ , and for $g = 2$ precesses at $\Omega' = (e/m)\gamma B$ in rest-frame time. Converting to lab time divides by $\gamma$ :

\omega_{\text{dyn}} = \frac{e}{m_\mu}B = 8.5061\times10^{8}\times1.45 = \mathbf{1.2334\times10^{9}\ rad\,s^{-1}}, \qquad\text{prograde.}

Add them, with the Thomas term retrograde:

\omega_{\text{dyn}} - |\boldsymbol\omega_{\mathrm{T}}| = 1.2334\times10^{9} - 1.1913\times10^{9} = \mathbf{4.209\times10^{7}\ rad\,s^{-1}} \;=\; \boldsymbol{\omega_c} \quad\checkmark

Step 4 — what just happened. Two huge terms cancelled to one part in thirty, and what survived is exactly the cyclotron frequency. At $g = 2$ precisely the spin stays locked to the momentum — it turns at the rate the velocity turns, and the muon's helicity never changes. The fraction the Thomas term cancels is

\frac{|\boldsymbol\omega_{\mathrm{T}}|}{\omega_{\text{dyn}}} = \frac{(\gamma-1)\omega_c}{(e/m)B} = 1 - \frac{1}{\gamma} = \mathbf{96.587\%} ,

an exact identity. The Thomas precession is not a correction to the muon's spin dynamics; it is one of the two terms whose near-cancellation is the spin dynamics. The residue when $g \neq 2$ is the anomalous precession $\omega_a$ — what the $g-2$ experiments actually measure. `aq-sr-bmt-equation` takes it from here.

2 · A gyroscope that comes home a quarter turn out

Carry a torque-free gyroscope once round a circle at $\beta = 0.6$ . Where does it point?

Step 1 — identify. $\gamma = 1/\sqrt{1-0.36} = 1.25$ exactly, so $\gamma - 1 = 1/4$ exactly.

Step 2 — apply (3). $\Delta\phi = -2\pi(\gamma-1) = -2\pi/4 = \boxed{-\pi/2} = \mathbf{-90^\circ}$ , retrograde. A quarter turn per lap, and the radius of the circle does not appear — nor does the speed of light, nor the mass. Only $\gamma$ .

Step 3 — check it against the integrator. Fermi–Walker transporting an actual spin four-vector all the way round returns $-1.5707963268$ rad, which is $-\pi/2$ to ten digits, and the rest-frame spin ends at $(0,-1,0)$ having started at $(1,0,0)$ .

Step 4 — the boundary number. Compare `aq-sr-thomas-wigner-rotation`, which finds $\Omega = 12.68^\circ$ for two boosts at $\beta = 0.6$ ninety degrees apart. Same speed, much bigger answer here — because a lap is not two boosts but a continuum, and their rotations share an axis and add. Push $\beta$ up and it runs away: at $\beta = 0.9$ , $\gamma = 2.294$ and the turn is $-2\pi(1.294) = -466^\circ$ — more than a full revolution backwards, per lap. For the hydrogen electron $\gamma - 1 = 2.66267\times10^{-5}$ and the turn is $-1.6730\times10^{-4}$ rad per orbit — canon 1902's number, reached from an entirely different formula. Tiny, and it is the whole of the factor of two in the fine structure.

Physics of the portrait

Origins · 2 minPrecession and how this was discovered

Precession

from the spinning top to the storage ring · Foucault's gyroscope · Larmor's magnetic torque · Thomas's rotation with no torque at all

Precession is what an axis does when it will not stay put. A spinning top leans and its axis sweeps a cone; the Earth's axis crawls round a $26{,}000$ -year circle, which Hipparchus noticed in the second century BC; an electron in a magnetic field precesses at the Larmor frequency. Every one of those is a dynamical story: something applies a torque $\boldsymbol\tau$ , the angular momentum obeys $\mathrm{d}\mathbf L/\mathrm{d}t = \boldsymbol\tau$ , and because $\boldsymbol\tau \perp \mathbf L$ the axis turns instead of tipping. Thomas precession breaks the pattern. There is no torque, and nothing acting on the gyroscope at all — put it in an evacuated, magnetically shielded box and simply carry it round a circle. It still comes back turned. The precession is not in the object but in the sequence of frames the object is honestly at rest in, and those frames rotate because the boosts relating them do not commute. That is why the effect was found in a spectrum, not a laboratory. No apparatus of 1926 could carry a macroscopic gyroscope at relativistic speed. But an electron in hydrogen orbits at $\beta = \alpha = 1/137$ , and the resulting retrograde precession — $1.1\times10^{12}$ rad/s, and still only $2.7\times10^{-5}$ of the orbital rate — halves the spin–orbit coupling. That factor of two was visible in the fine structure, and visibly wrong, for a year, until a twenty-two-year-old in Copenhagen realised the frames were turning.

History

1914

— Ludwik Silberstein, in The Theory of Relativity, derives the rotation accompanying two successive non-parallel boosts.

1925

— Uhlenbeck and Goudsmit propose electron spin with g = 2 (Naturwissenschaften 13, 953); it explains the anomalous Zeeman effect and gives the hydrogen fine-structure splitting a factor of two too large. Heisenberg writes to Goudsmit asking how he intends to get rid of the factor 2. Pauli thinks the spinning electron dead.

1926

— Llewellyn Thomas, aged 22 and on a fellowship at Bohr's institute in Copenhagen, publishes The Motion of the Spinning Electron in Nature 117, 514: the electron's successive rest frames on a curved orbit are related by non-collinear boosts, so they rotate, and the retrograde precession removes exactly half the coupling. Pauli accepts spin.

1927

— Thomas gives the full kinematics in The Kinematics of an Electron with an Axis, Philosophical Magazine, Series 7, volume 3, pages 1 to 22.

1959

— Bargmann, Michel and Telegdi publish the covariant spin-precession equation (Physical Review Letters 2, 435), with the Thomas term built into its architecture rather than added to it.

2011

— Gravity Probe B reports the geodetic precession of an orbiting gyroscope to 0.3 per cent — a general-relativistic effect, not this one.

Going further

Texts. Jackson, Classical Electrodynamics, 3rd ed., §11.8 · Goldstein, Poole & Safko, Classical Mechanics, 3rd ed., §7.3 · Rindler, Introduction to Special Relativity, 2nd ed., §§43–44 · Møller, The Theory of Relativity, §22. Primary: Thomas, Nature 117, 514 (1926); Thomas, Phil. Mag. Ser. 7 3, 1–22 (1927).

Provenance. Signature $(+,-,-,-)$ ; $u\cdot u = +c^2$ ; $B(\mathbf v)$ the active symmetric pure boost; $\boldsymbol\omega$ the active rotation vector, $\delta\mathbf x = \boldsymbol\omega\times\mathbf x$ . Every equation and number above is recomputed in verify_aq-sr-thomas-precession.py, which also reads this file and asserts each printed value against what it computed. Base defects repaired: (i) the base prints the rate three times with two different signs — the boxed card has $\mathbf a\times\mathbf v$ (correct), both derivation cards have $\mathbf v\times\mathbf a$ / $\mathbf v\times\mathrm{d}\mathbf v$ (wrong), and a note excuses the discrepancy as a "sign convention"; it is not, and two independent machine checks settle it. (ii) Its eq[3] gives the correct circular-orbit magnitude and then writes it as a vector equation with the wrong sign. (iii) Its eq[1] transports the spin as $\boldsymbol\omega_{\mathrm{T}}\times\mathbf S$ while its eq[6] writes $\mathbf S\times[\dots]$ — same symbol, opposite convention, same base; this page uses $\boldsymbol\omega\times\mathbf S$ throughout and says so. (iv) A sibling base prints the large- $\gamma$ asymptote of $\gamma^2/(\gamma+1)$ as $\gamma - \tfrac12$ ; it is $\gamma - 1$ .

Scope. The finite rotation from two boosts — non-closure, the polar decomposition, the exact half-rapidity angle, $\Omega = 12.68^\circ$ , the commutator $[K_i,K_j] = -\varepsilon_{ijk}J_k$ — is owned by `aq-sr-thomas-wigner-rotation`, which cedes the rate to this page; quoted, not re-derived. The little group is `aq-sr-wigner-rotation`'s; the $\tfrac12$ in $H_{SO}$ and the fine-structure number `aq-sr-thomas-factor-spin-orbit`'s; the covariant packaging, $\omega_a$ and the magic $\gamma$ `aq-sr-bmt-equation`'s. The geodetic precession and Gravity Probe B belong to the general-relativity track, and this page explicitly declines to claim them.

Builds on. The Thomas–Wigner Rotation · Rapidity and Hyperbolic Boosts · Four-Acceleration. Leads to. The Thomas Factor One-Half · Spin–Orbit Coupling · The BMT Equation.

Figure

The prefactor γ²/(γ+1), and a corpus defect drawn to scale. Blue: the true prefactor. It starts at exactly ½ when γ = 1 — THE THOMAS FACTOR, the number that halves the spin-orbit coupling in hydrogen — and grows without bound. Grey dashed: the correct large-γ asymptote γ − 1, which the blue curve approaches from above, the gap closing as 1/(γ+1). Orange: γ − ½, which a sibling base prints as the asymptote. It is not an asymptote at all — the error does not shrink, it tends to a CONSTANT ½, so the curve is permanently wrong by half a unit of γ. Marked: the hydrogen electron sits at γ − 1 = 2.66 × 10⁻⁵, indistinguishable from ½ on this plot, while the muon g−2 ring sits at γ = 29.3, where the prefactor is 28.3 and the Thomas precession is TWENTY-EIGHT TIMES the orbital frequency

If no torque acts on the gyroscope, what is doing the turning?

Nothing is, and that is the point. The gyroscope is Fermi–Walker transported: in each instantaneous rest frame it is perfectly still, and at no moment does anything act on it. What turns is the relationship between those frames and the lab. Each successive rest frame is reached from the last by a boost; the boosts are non-collinear on a curved path; non-collinear boosts do not compose into a boost, and leave a rotation behind. Sum the leftovers around a closed orbit and they do not cancel, because they share an axis. The frames rotate relative to the lab, so a spin honestly at rest in them rotates relative to the lab too. There is no force in the story — only the fact that the boosts are not a group.

Why does the sign matter so much? Would a plus sign really change anything?

It destroys the fine structure of hydrogen. The naive spin–orbit calculation — the electron sees a magnetic field, its moment precesses — overshoots the observed splitting by a factor of two. Thomas's precession is half the naive rate and opposite in sense, so it removes half: $1 - \tfrac12 = \tfrac12$ , and the answer comes out right. Flip the sign and you get $1 + \tfrac12 = \tfrac32$ — a splitting three times the observed value instead of two, which is further from the data than the error Thomas set out to fix. The minus sign is not decoration; it is the entire content of the 1926 paper, and it is why the base's sign-flipped derivation cards cannot be shrugged off as a convention.

How is this different from the Thomas-Wigner rotation?

Same physics, different quantity — owned by different articles on purpose. `aq-sr-thomas-wigner-rotation` treats two boosts and computes the finite rotation between them: $12.68^\circ$ at $\beta_1 = \beta_2 = 0.6$ and a right angle, exactly $\tan(\Omega/2) = \tfrac19$ . This page treats a continuum of boosts along a worldline and computes the rate at which rotation accumulates — the relation between an angle and an angular velocity. And integrating matters: around a closed orbit at the same $\beta = 0.6$ the accumulated turn is $90^\circ$ per lap, not $12.68^\circ$ , because a lap is not two boosts but infinitely many, all pointing the same way.

Can the Thomas precession ever be larger than the orbital frequency itself?

Routinely — and this is the point most often missed. Equation (2) gives $|\boldsymbol\omega_{\mathrm{T}}| = (\gamma-1)\omega_{\text{orb}}$ , so the moment $\gamma > 2$ the precession is faster than the orbit. In the muon $g-2$ ring $\gamma = 29.3$ and the Thomas precession runs at $28.3$ times the cyclotron frequency: $1.19\times10^{9}$ rad/s against an orbital $4.21\times10^{7}$ . Calling this a "second-order relativistic correction" describes the $\beta \ll 1$ limit, not the phenomenon. In a storage ring it is one of the two dominant terms, and the whole $g-2$ measurement lives in the residue left when it very nearly cancels the magnetic torque.

The gyroscope comes back turned. Is that not a violation of angular momentum conservation?

No — "the spin in the rest frame" is not a conserved quantity, because there is no single rest frame. Angular-momentum conservation is a statement about a closed system in one inertial frame; here we compare components measured in a succession of frames. It is the same species of non-paradox as the twin paradox. The gyroscope takes a closed path through velocity space, and velocity space is hyperbolic; transporting a direction around a closed loop in a curved space and finding it rotated is called holonomy, and it violates nothing.

Does Thomas precession require quantum mechanics or spin one-half?

Neither. Nothing in the derivation mentions $\hbar$ , spin quantisation, or the electron. Equation (1) applies to a classical spinning top, to the orientation vector of a rigid body, to a macroscopic gyroscope in a rocket. The word "spin" appears only because the effect was found in hydrogen and because its most precise measurements are made on particle spins — the object transported can be any vector carried without torque along the worldline. It is a theorem about the Lorentz group, not about quantum mechanics. Which is why it survives untouched in the Dirac equation, where the $\tfrac12$ emerges from the kinematics with nobody inserting it.