Thomas–Wigner Rotation

Boost east. Then boost north. You are now moving north-east, and your axes are pointing somewhere they should not be — turned, by an angle nobody put there. Nothing was rotated. Two pure boosts were applied, one after the other, each an ordinary change of inertial frame with no whiff of rotation about it. And yet the composite transformation contains a genuine spatial rotation: at $\beta_1 = \beta_2 = 0.6$ and a right angle between them, $12.68^\circ$ . Not a rounding error, not a coordinate artefact, and not something you can transform away. The consequence is structural. The set of pure Lorentz boosts is not closed under composition. Boosts do not form a group. The Lorentz group is the smallest thing containing them, and it necessarily contains rotations too — because the boosts generate them. Reverse the order and the rotation reverses: $-12.68^\circ$ . The two orderings differ by $25.36^\circ$ of honest, physical rotation, so the non-commutativity of boosts — a fact that looks like abstract algebra — is a turn you could measure with a gyroscope. Llewellyn Thomas measured it in 1926, in the hydrogen spectrum, and it was worth a factor of two.

Key takeaways

Signature $(+,-,-,-)$ , timelike-positive. A pure boost $B(\mathbf{v})$ is the symmetric $4\times4$ matrix taking a rest four-momentum to velocity $\mathbf{v}$ : $B^{0}{}_{0} = \gamma$ , $B^{0}{}_{i} = B^{i}{}_{0} = \gamma\beta_i$ , $B^{i}{}_{j} = \delta_{ij} + (\gamma-1)\beta_i\beta_j/\beta^2$ . The forward/inverse boost convention is that of Transverse Velocity Transformation.
The non-closure theorem (owned). $B^{\mathsf{T}} = B$ , so $(B_2B_1)^{\mathsf{T}} = B_1B_2$ . For non-collinear boosts $B_1B_2 \neq B_2B_1$ , so $B_2B_1$ is not symmetric — hence not a pure boost. Polar decomposition gives $\;\boxed{B(\mathbf{v}_2)\,B(\mathbf{v}_1) = R(\Omega)\,B(\mathbf{w})}\;$ with $R$ a genuine rotation about $\mathbf{v}_1 \times \mathbf{v}_2$ .
The exact angle (owned). $\;\tan\dfrac{\Omega}{2} = \dfrac{\sin\theta\,\tanh\frac{\varphi_1}{2}\tanh\frac{\varphi_2}{2}}{1 + \cos\theta\,\tanh\frac{\varphi_1}{2}\tanh\frac{\varphi_2}{2}}\;$ , with $\varphi_i = \operatorname{artanh}\beta_i$ and $\theta$ the angle between the boosts. Verified here against an explicit $4\times4$ matrix product, polar-decomposed.
The leading term is $O(\beta^2)$ , not $O(\beta^4)$ . $\;\Omega \approx \dfrac{|\mathbf{v}_1 \times \mathbf{v}_2|}{2c^2} = \dfrac{v_1v_2\sin\theta}{2c^2}$ . (The base article for this page claimed $\Omega \approx \tfrac12(\gamma_1-1)(\gamma_2-1)$ , which is $O(\beta^4)$ — wrong by a factor of $20\,000$ at $\beta = 0.01$ . It is corrected below.)
The algebra says the same thing in one line: $[K_x, K_y] = -J_z$ . The commutator of two boost generators is a rotation generator. Boosts cannot close among themselves; the rotations are forced.
Key idea: a pure boost is a symmetric matrix, and symmetric matrices do not close under multiplication unless they commute. That one sentence of linear algebra is the whole theorem — and its residue is a real rotation of a real gyroscope, whose accumulation around an orbit is the Thomas precession that supplies the $\tfrac12$ in atomic fine structure.

The core insight

The Thomas–Wigner rotation is not a correction, an approximation, or an artefact of a bad coordinate choice. It is the statement that the boosts are not a subgroup — and it is forced, because the product of two symmetric matrices is symmetric only when they commute.

Where this lives today

One of the most precise tests of the Standard Model in existence runs on this rotation. In the muon $g-2$ experiments at Brookhaven and Fermilab, muons circulate in a storage ring at $\gamma = 29.3$ while their spins precess, and what is measured is the difference between the spin-precession and cyclotron frequencies — a difference that exists partly because of the Thomas term. The Bargmann–Michel–Telegdi equation (1959) models it, and the Thomas contribution is not a correction inside it but part of its architecture. The non-closure theorem is also load-bearing in relativistic quantum theory. Because boosts do not close, one cannot classify particles by "how they transform under boosts" — one must use representations of the full Poincaré group. Wigner's 1939 answer is that a particle is an irreducible representation, labelled by mass and spin; the Wigner rotation is how the little group of a momentum acts when that momentum is boosted.

The physical picture

Take a spinning gyroscope, axis fixed, and do not touch it. Now change frames: boost to $\mathbf{v}_1$ , then to $\mathbf{v}_2$ , with $\mathbf{v}_2$ pointing somewhere other than along $\mathbf{v}_1$ . No torque, no rotation. But in the final frame the axis points in a different direction than the naive composition predicts, and the discrepancy is a rotation about $\mathbf{v}_1 \times \mathbf{v}_2$ . Where does the turn come from? From the transverse velocity rule. A boost along $\hat{x}$ leaves the transverse coordinates alone but divides the transverse velocity by $\gamma$ — Transverse Velocity Transformation's owned result. So when the second boost's velocity is resolved in the first boost's frame its transverse part is squeezed and its parallel part is not: the composed motion is not the direction you get by adding vectors. The velocity triangle does not close in the Euclidean way; it closes hyperbolically, and the angular defect of a hyperbolic triangle is what is left over — the rotation. At $\beta_1 = \beta_2 = 0.6$ with $\mathbf{v}_1 = 0.6c\,\hat{x}$ , $\mathbf{v}_2 = 0.6c\,\hat{y}$ : the composed velocity is $\mathbf{w} = (0.6,\ 0.48,\ 0)\,c$ , not $(0.6, 0.6, 0)c$ . The $y$ -component is $0.6/\gamma_1 = 0.6/1.25 = 0.48$ — the transverse rule, verbatim. And $|\mathbf{w}| = 0.768375\,c$ , with $\gamma_{\mathbf{w}} = 1.5625 = \gamma_1\gamma_2$ exactly. The rotation is what is left over once that velocity is accounted for. It is $12.68^\circ$ .

Figure

PUNCHLINE — boost east, then north, and your axes come back TURNED. Two pure boosts at β₁ = β₂ = 0.6, at right angles. LEFT: the composed velocity is w = (0.6, 0.48)c — NOT (0.6, 0.6)c, because the transverse rule divides the second boost's velocity by γ₁ = 1.25, giving 0.6⁄1.25 = 0.48. The velocity triangle does not close in the Euclidean way. RIGHT: what is left over is a genuine spatial rotation of the axes by Ω = 12.68°, about the v₁ × v₂ axis. Reverse the order and the rotation reverses to −12.68°, so the two orderings differ by 25.36°. No torque was applied. Nothing was rotated. The rotation was FORCED by the fact that the product of two symmetric matrices is not symmetric

The core equations

A pure boost is a symmetric matrix. That is the fact everything hangs on, so it is written first:

B(\mathbf{v}) = \begin{pmatrix} \gamma & \gamma\beta_x & \gamma\beta_y & \gamma\beta_z \\ \gamma\beta_x & & & \\ \gamma\beta_y & & \delta_{ij} + (\gamma-1)\dfrac{\beta_i\beta_j}{\beta^2} & \\ \gamma\beta_z & & & \end{pmatrix}, \qquad B^{\mathsf{T}} = B \qquad (1)

where: $\boldsymbol{\beta} = \mathbf{v}/c$ , $\gamma = (1-\beta^2)^{-1/2}$ . Physical reading: $B$ satisfies $\Lambda^{\mathsf{T}}\eta\Lambda = \eta$ with $\eta = \mathrm{diag}(+1,-1,-1,-1)$ (The Minkowski Metric), and it is symmetric — which is what distinguishes a pure boost from a boost-with-a-rotation. The composition, and this article's owned result:

\boxed{\;B(\mathbf{v}_2)\,B(\mathbf{v}_1) \;=\; R(\Omega\,\hat{\mathbf{n}})\;B(\mathbf{w}), \qquad \hat{\mathbf{n}} = \frac{\mathbf{v}_1 \times \mathbf{v}_2}{|\mathbf{v}_1 \times \mathbf{v}_2|}\;} \qquad (2)

where: $R$ is a genuine spatial rotation, $\mathbf{w}$ the composed velocity. Physical reading: this is the failure of closure, written out. Were the boosts closed the right-hand side would be $B(\mathbf{w})$ alone and $\Omega$ zero. It is not, unless $\mathbf{v}_1 \parallel \mathbf{v}_2$ . The polar decomposition is unique, so $\Omega$ is not a matter of choice. The exact angle:

\boxed{\;\tan\frac{\Omega}{2} \;=\; \frac{\sin\theta\;\tanh\dfrac{\varphi_1}{2}\,\tanh\dfrac{\varphi_2}{2}}{1 + \cos\theta\;\tanh\dfrac{\varphi_1}{2}\,\tanh\dfrac{\varphi_2}{2}}\;} \qquad (3)

where: $\varphi_i = \operatorname{artanh}\beta_i$ (rapidity, owned by Rapidity and Hyperbolic Boosts), $\theta$ the angle between the boosts. Physical reading: exact at all speeds and angles. Half-rapidities — the half is the fingerprint of the formula's spinor ( $SL(2,\mathbb{C})$ ) origin, not an approximation. And the same fact, in the Lie algebra. Write $B(\varphi\hat{x}) = \exp(\varphi K_x)$ and $R(\theta\hat{z}) = \exp(\theta J_z)$ ; then

[K_x,\ K_y] \;=\; -J_z \qquad\text{i.e.}\qquad [K_i,\ K_j] = -\varepsilon_{ijk}J_k \qquad (4)

where: $K_i$ , $J_i$ are the real $4\times4$ generators. Physical reading: the commutator of two boost generators is a rotation generator. That single line is non-closure, infinitesimally, and it is why the Lorentz algebra cannot be split into a boost part and a rotation part.

Deriving the structure

Derivation ANON-CLOSURE — the product of two symmetric matrices is not symmetric, and that is the whole theorem

The claim is that the set of pure boosts is not closed under composition. It sounds like a fact requiring calculation, and it is not: it is a single line of linear algebra, and the entire physical content of the Thomas–Wigner rotation is squeezed out of it. The strategy is to identify "pure boost" with "symmetric Lorentz matrix", observe that symmetry is precisely the property that fails to survive multiplication, and then name what is left over — not by guessing, but by the polar decomposition, which is unique.

Step 1 — a pure boost is symmetric; a symmetric positive-definite Lorentz matrix is a pure boost. Read (1): $B$ is manifestly symmetric, and the converse is the standard characterisation. Why: we now have a test. Multiply two boosts and ask whether the answer is symmetric. If it is not, it is not a pure boost.

Step 2 — apply the test. Transpose the product, using $B_i^{\mathsf{T}} = B_i$ and $(AB)^{\mathsf{T}} = B^{\mathsf{T}}A^{\mathsf{T}}$ :

\left(B_2 B_1\right)^{\mathsf{T}} = B_1^{\mathsf{T}} B_2^{\mathsf{T}} = B_1 B_2.

So $B_2B_1$ is symmetric if and only if $B_1B_2 = B_2B_1$ — if and only if the boosts commute. ∎ Why: that is the theorem, and it took one line. Collinear boosts commute (both are functions of the same $K$ ) and their product is a pure boost — the additivity result that Rapidity and Hyperbolic Boosts owns. Non-collinear boosts do not commute, so their product is not symmetric, so it is not a pure boost. The set is not closed.

Step 3 — name what is left over. Every invertible real matrix has a unique polar decomposition $\Lambda = R\,S$ with $R$ orthogonal, $S$ symmetric positive-definite, $S = (\Lambda^{\mathsf{T}}\Lambda)^{1/2}$ , $R = \Lambda S^{-1}$ . Apply it to $\Lambda = B_2B_1$ : because $\Lambda$ preserves $\eta$ , the orthogonal factor is block-diagonal, $R = \mathrm{diag}(1, R_{3\times3})$ — a spatial rotation — and the symmetric factor is a pure boost:

B(\mathbf{v}_2)B(\mathbf{v}_1) = R(\Omega)\,B(\mathbf{w}).

Why: uniqueness matters. $\Omega$ is not an artefact of how one chooses to split the transformation; the split is forced. And the rotation is genuine — $R$ acts on the spatial block alone, mixes no time, and would turn a gyroscope.

Step 4 — find the axis. The only vector the problem supplies besides $\mathbf{v}_1$ and $\mathbf{v}_2$ is $\mathbf{v}_1 \times \mathbf{v}_2$ ; and reflecting the configuration in the plane of the velocities must reverse the rotation while leaving the boosts alone. Hence $\hat{\mathbf{n}} \parallel \mathbf{v}_1 \times \mathbf{v}_2$ , and $\Omega = 0$ for parallel boosts. Why: the rotation lives in the plane the boosts span, and vanishes when there is no plane — which is why the effect is invisible in every one-dimensional problem.

Step 5 — the same fact infinitesimally. With real generators $K_x = E_{01} + E_{10}$ and $K_y = E_{02} + E_{20}$ ( $E_{ab}$ has a single $1$ in slot $ab$ ), multiply out:

K_xK_y = E_{12}, \qquad K_yK_x = E_{21} \qquad\Longrightarrow\qquad [K_x, K_y] = E_{12} - E_{21} = -J_z,

where $J_z = E_{21} - E_{12}$ generates rotations from $\hat{x}$ toward $\hat{y}$ . The commutator of two boosts is a rotation. Why: an algebra whose commutators leave the set cannot exponentiate to a closed subgroup — Lie theory's version of Step 2. The three rotations do close ( $[J_i,J_j] = \varepsilon_{ijk}J_k$ ). The three boosts do not. *Verified in verify_aq-sr-thomas-wigner-rotation.py: $B^{\mathsf{T}} = B$ and $B^{\mathsf{T}}\eta B = \eta$ for random $\mathbf{v}$ ; $B_2B_1$ is not symmetric for non-collinear boosts and is for collinear ones; its polar decomposition returns a block-diagonal $R = \mathrm{diag}(1, R_3)$ , $R_3 \in SO(3)$ , with axis parallel to $\mathbf{v}_1 \times \mathbf{v}_2$ ; and $[K_x,K_y] = -J_z$ entrywise.*

Go deeper · Derivation B · 2 minTHE EXACT ANGLE — and a numerical proof against an explicit 4x4 matrix product

Derivation BTHE EXACT ANGLE — and a numerical proof against an explicit 4x4 matrix product

Non-closure is qualitative; the rotation angle is what an experiment measures, and it must be got exactly. The formula below is stated in half-rapidities — the tell-tale of its origin in the spinor representation $SL(2,\mathbb{C})$ , where a boost of rapidity $\varphi$ is $\cosh(\varphi/2) + (\hat{\mathbf{n}}\cdot\boldsymbol\sigma)\sinh(\varphi/2)$ , and the product of two such objects splits into a Hermitian part (the boost) and a unitary part (the rotation) by inspection. Rather than reproduce that computation, this card states the result and proves it the hard way — by building $B(\mathbf{v}_2)B(\mathbf{v}_1)$ as an explicit $4\times4$ array of numbers, polar-decomposing it, and reading the angle off. That is a genuinely independent check, and it is the one the verification script performs.

Step 1 — the formula. With $\varphi_i = \operatorname{artanh}\beta_i$ and $\theta$ the angle between the boosts,

\boxed{\;\tan\frac{\Omega}{2} = \frac{\sin\theta\;\tanh\frac{\varphi_1}{2}\tanh\frac{\varphi_2}{2}}{1 + \cos\theta\;\tanh\frac{\varphi_1}{2}\tanh\frac{\varphi_2}{2}}\;}

Multiplying through by $\cosh\frac{\varphi_1}{2}\cosh\frac{\varphi_2}{2}$ gives the $\sinh/\cosh$ form seen in the literature; they are the same equation. Why: the $\tanh$ form makes the small-velocity limit fall out in one line.

Step 2 — check it at a right angle. Put $\theta = \pi/2$ and $\beta_1 = \beta_2 = 0.6$ . Then $\varphi = \operatorname{artanh}(0.6) = \ln 2 = 0.693147$ , so $\tanh(\varphi/2) = \tanh(0.346574) = 1/3$ exactly. The cosine term dies, and

\tan\frac{\Omega}{2} = \frac{1 \times \tfrac13 \times \tfrac13}{1 + 0} = \frac19 \qquad\Longrightarrow\qquad \Omega = 2\arctan\tfrac19 = 0.221314\ \mathrm{rad} = \mathbf{12.6804^\circ}.

Why: an exact rational, $\tan(\Omega/2) = 1/9$ , from an ordinary configuration. That is the number the verification script must reproduce from a matrix product — and it does.

Step 3 — prove it against the matrix. Build $\Lambda = B(0.6c\,\hat{y})\,B(0.6c\,\hat{x})$ numerically from equation (1). Form $S = (\Lambda^{\mathsf{T}}\Lambda)^{1/2}$ and $R = \Lambda S^{-1}$ . The result: $S$ is symmetric with $S^{0}{}_{0} = \gamma_{\mathbf{w}} = \mathbf{1.5625}$ and $\mathbf{w} = (0.6,\ 0.48,\ 0)c$ , $|\mathbf{w}| = 0.768375c$ ; and $R = \mathrm{diag}(1, R_3)$ with

\Omega = \arctan\frac{(R_3)_{yx}}{(R_3)_{xx}} = \mathbf{12.680383^\circ},

agreeing with Step 2 to every digit computed. ∎ Why: this is the point of the card. The half-angle formula is not quoted on authority; it is confirmed against the object it describes. Note the by-products: $\gamma_{\mathbf{w}} = 1.5625 = \gamma_1\gamma_2$ exactly, and $w_y = 0.48 = 0.6/\gamma_1$ — the transverse velocity rule, reappearing inside a matrix decomposition.

Step 4 — the small-velocity limit, and the order that matters. For $\beta \ll 1$ , $\varphi \approx \beta$ and $\tanh(\varphi/2) \approx \beta/2$ . The denominator $\to 1$ , so

\tan\frac{\Omega}{2} \approx \frac{\beta_1\beta_2\sin\theta}{4} \qquad\Longrightarrow\qquad \boxed{\;\Omega \approx \frac{\beta_1\beta_2\sin\theta}{2} = \frac{|\mathbf{v}_1 \times \mathbf{v}_2|}{2c^2}\;}

which is $O(\beta^2)$ . Why: the order is the point, and the base article got it wrong. It claimed $\Omega \approx \tfrac12(\gamma_1-1)(\gamma_2-1)$ ; but $\gamma - 1 \approx \beta^2/2$ , so that is $O(\beta^4)$ — a different power of the small parameter. At $\beta_1 = 0.01$ , $\beta_2 = 0.02$ , $\theta = 90^\circ$ the exact angle is $1.000\times10^{-4}$ rad, the base's expression $5.00\times10^{-9}$ : wrong by a factor of twenty thousand. The correct $O(\beta^2)$ term is what carries the factor $\tfrac12$ into atomic fine structure; an $O(\beta^4)$ effect would be invisible there.

Step 5 — the ultra-relativistic limit. As $\beta \to 1$ , $\tanh(\varphi/2) \to 1$ , and the formula collapses to $\tan(\Omega/2) \to \sin\theta/(1+\cos\theta) = \tan(\theta/2)$ . So

\boxed{\;\Omega \;\longrightarrow\; \theta\;} \qquad (\beta \to 1)

— the frame is turned by the entire angle between the two boosts. For orthogonal boosts that is a quarter turn: $\beta = 0.9$ gives $42.896^\circ$ ; $\beta = 0.99$ , $73.941^\circ$ ; $\beta = 0.99999$ , $89.488^\circ$ . Why: the effect is not small when the speeds are not small — which is why it matters in storage rings and not merely in atoms. Note the ceiling is $\theta$ , not $\pi/2$ : two nearly antiparallel ultra-relativistic boosts turn the frame through nearly a half turn. *Verified in verify_aq-sr-thomas-wigner-rotation.py: formula (3) against an explicit $4\times4$ polar decomposition on a grid of 20 speeds $\times$ 17 angles, agreeing to $10^{-9}$ rad; $\tan(\Omega/2) = 1/9$ , $\Omega = 12.680383^\circ$ , $\gamma_{\mathbf{w}} = 1.5625$ , $\mathbf{w} = (0.6, 0.48, 0)c$ , $|\mathbf{w}| = 0.768375c$ ; the small- $\beta$ limit confirmed $O(\beta^2)$ and the base's expression wrong by $2\times10^{4}$ ; $42.896^\circ$ , $73.941^\circ$ , $89.488^\circ$ ; and $\Omega \to -\Omega$ on reversal.*

Limiting cases

Case	Result	Meaning
$\mathbf{v}_1 \parallel \mathbf{v}_2$	$\Omega = 0$	the boosts commute, the product is a pure boost
$\beta_1,\beta_2 \ll 1$	$\Omega \approx \dfrac{\|\mathbf{v}_1\times\mathbf{v}_2\|}{2c^2}$	$O(\beta^2)$ — not $O(\beta^4)$ . This term carries the Thomas factor $\tfrac12$
$c \to \infty$	$\Omega \to 0$	Galilean boosts commute; the effect is purely relativistic
$\theta = 0$ or $\pi$	$\Omega = 0$	$\sin\theta = 0$ : no plane, no rotation
$\beta \to 1$	$\Omega \to \theta$	the frame turns through the whole angle between the boosts; orthogonal $\Rightarrow$ a quarter turn
order reversed	$\Omega \to -\Omega$	$B_1B_2$ and $B_2B_1$ differ by a rotation of $2\Omega$

Common trapTwo defects in the source article for this page, both corrected here

Defect 1 — the wrong power of β. The base's summary claimed the lowest-order Wigner angle is $\Omega \approx \tfrac12(\gamma_1-1)(\gamma_2-1)$ . Since $\gamma - 1 = \tfrac12\beta^2 + O(\beta^4)$ , that is $O(\beta^4)$ , whereas the true leading term $\Omega \approx |\mathbf{v}_1\times\mathbf{v}_2|/2c^2$ is $O(\beta^2)$ — which is what the base's own small-velocity card said. The base contradicted itself, and the summary was the wrong half. At $\beta_1 = 0.01$ , $\beta_2 = 0.02$ , $\theta = 90^\circ$ : exact $\Omega = 1.000\times10^{-4}$ rad; the base's expression gives $5.00\times10^{-9}$ . A factor of $2\times10^4$ . An $O(\beta^4)$ Thomas precession would be far too small to matter in atomic fine structure, and the factor of $\tfrac12$ would be unexplained.

Defect 2 — a truncated equation. The base's card "Rotation via Lorentz factors" ends, literally, in $\dots$ over $\dots$ . It cannot be evaluated, and has been discarded, not repaired: the exact half-angle formula (3) is complete and verifiable.

Common trapWhat the Thomas–Wigner rotation is NOT

It is not aberration. Aberration turns the apparent direction of a light ray; the Wigner rotation turns the axes of a frame, and survives after the velocities have been correctly composed. Relativistic Aberration of Light owns the other one.
It is not a coordinate artefact. The polar decomposition is unique and $R$ acts on the spatial block alone. A gyroscope carried around a closed circuit of boosts comes back genuinely turned.
It is not the Thomas precession, though it becomes it. $\Omega$ is a finite rotation from two boosts; the precession is the rate at which it accumulates for a continuously accelerating particle, and belongs to Thomas Precession and The Thomas Factor One-Half — named here, derived there.
Nothing you can see rotates. The frames rotate relative to one another. What is observable is the effect on a quantity carried with the frame — a spin, a gyroscope axis — which is why the discovery happened in a spectrum and not a photograph.

Worked examples

1 · The hydrogen atom, and the factor of two that Thomas found

In 1925 Uhlenbeck and Goudsmit's spinning electron gave the fine structure of hydrogen — a factor of two too large. The missing ingredient was the rotation on this page, accumulated around the orbit. Compute it.

Step 1 — identify. In the Bohr ground state the electron has $v = \alpha c$ with $\alpha = 1/137.036$ , and radius $a_0 = 5.29177 \times 10^{-11}\ \mathrm{m}$ :

v = \alpha c = \mathbf{2.18769\times10^{6}\ \mathrm{m\,s^{-1}}}, \qquad \omega_{\text{orb}} = \frac{v}{a_0} = \mathbf{4.1341\times10^{16}\ \mathrm{rad\,s^{-1}}}.

The acceleration is centripetal, $a = v^2/a_0$ , so $\mathbf{a} \parallel -\hat{\mathbf{r}}$ while $\mathbf{v}$ is tangential: they are perpendicular, and $\mathbf{v}\times\mathbf{a} \parallel \mathbf{L}$ .

Step 2 — apply the rotation, infinitesimally. Over $\mathrm{d}t$ the velocity turns from $\mathbf{v}$ to $\mathbf{v} + \mathbf{a}\,\mathrm{d}t$ , and the two boosts are non-collinear. Feeding this into (3) and keeping the full $\gamma$ gives the accumulated rotation rate

\boldsymbol{\omega}_{\mathrm{T}} = \frac{\gamma^2}{\gamma+1}\,\frac{\mathbf{a}\times\mathbf{v}}{c^2}, \qquad \frac{\gamma^2}{\gamma+1} \xrightarrow{\ \gamma\to1\ } \frac{1}{2}.

This is the Thomas precession, whose derivation is owned by Thomas Precession — quoted here once, in one form, because it is what the Wigner rotation becomes. Note the sense: $\mathbf{a}\times\mathbf{v}$ is antiparallel to $\mathbf{L}$ , so the precession is retrograde, and subtracts from the naive spin–orbit torque.

Step 3 — the boundary number. With $\gamma \approx 1$ , $\omega_{\mathrm{T}} = \tfrac12 v a/c^2 = v^3/(2a_0c^2)$ :

$\omega_{\mathrm{T}} = \frac{(2.18769\times10^6)^3}{2 \times 5.29177\times10^{-11} \times (2.99792\times10^8)^2} = \mathbf{1.1007\times10^{12}\ \mathrm{rad\,s^{-1}}}.$

Compare it to the orbit:

$\frac{\omega_{\mathrm{T}}}{\omega_{\text{orb}}} = \frac{v^3/(2a_0c^2)}{v/a_0} = \frac{v^2}{2c^2} = \frac{\alpha^2}{2} = \mathbf{2.6626\times10^{-5}} \ \ \checkmark$

— the Bohr radius cancels identically, and the answer is $\alpha^2/2$ exactly: the clean way to see this is an $O(\beta^2)$ effect, with no dependence on the orbit's size at all. Per orbit the electron's rest frame comes back turned by $2\pi \times 2.66257\times10^{-5} = \mathbf{1.673\times10^{-4}\ \mathrm{rad}}$ .

Step 4 — cash it out. Tiny, and decisive. The spin–orbit interaction computed naively in the electron's instantaneous rest frame overshoots the observed fine-structure splitting by exactly a factor of two; Thomas's retrograde precession, half the naive magnitude and opposite in sense, cancels half of it. The Thomas factor $\tfrac12$ is not a fudge — it is the $\gamma \to 1$ value of $\gamma^2/(\gamma+1)$ . Had the leading term been $O(\beta^4)$ , as the base claimed, the correction would be $\sim\alpha^4 \sim 3\times10^{-9}$ of the orbit — undetectable, and the factor of two would still be a mystery.

2 · Two boosts at a right angle, done exactly

Boost to $0.6c$ along $\hat{x}$ , then to $0.6c$ along $\hat{y}$ . Find the composed velocity, the rotation, and what happens if you swap the order.

Step 1 — identify. $\gamma_1 = \gamma_2 = 1/\sqrt{1-0.36} = 1.25$ , and $\varphi_1 = \varphi_2 = \operatorname{artanh}(0.6) = \ln 2 = 0.693147$ . So $\tanh(\varphi/2) = \tanh(\tfrac12\ln 2) = (2-1)/(2+1) = 1/3$ exactly.

Step 2 — the velocity. Compose with the rules of Transverse Velocity Transformation: the second boost's velocity is transverse to the first, so it is divided by $\gamma_1$ :

$\mathbf{w} = \left(0.6,\ \frac{0.6}{1.25},\ 0\right)c = \mathbf{(0.6,\ 0.48,\ 0)\,c}, \qquad |\mathbf{w}| = \sqrt{0.36 + 0.2304}\ c = \mathbf{0.768375\,c},$

with $\gamma_{\mathbf{w}} = 1/\sqrt{1-0.5904} = \mathbf{1.5625} = \gamma_1\gamma_2 = 1.25^2$ ✓. Note what it is not: the naive vector sum $(0.6, 0.6)c$ has $|\mathbf{v}| = 0.849c$ and the wrong direction entirely.

Step 3 — the boundary number. From equation (3) with $\theta = 90^\circ$ :

$\tan\frac{\Omega}{2} = \frac{1 \cdot \tfrac13 \cdot \tfrac13}{1 + 0} = \frac{1}{9} \qquad\Longrightarrow\qquad \Omega = 2\arctan\frac19 = \mathbf{12.6804^\circ}.$

Independently, building $B(0.6c\,\hat{y})B(0.6c\,\hat{x})$ as a $4\times4$ array and polar-decomposing it returns $\Omega = 12.680383^\circ$ and exactly the $\mathbf{w}$ above. Two different computations, one number.

Step 4 — swap the order. $B(0.6c\,\hat{x})\,B(0.6c\,\hat{y})$ gives the same speed, $|\mathbf{w}'| = 0.768375c$ — but $\mathbf{w}' = (0.48,\ 0.6,\ 0)c$ , a different direction, and a rotation of $\mathbf{-12.6804^\circ}$ . The two orderings differ by $25.36^\circ$ of physical rotation. With a gyroscope you could tell which way round you had done it. That is why boosts are not a group.

Figure

The rotation, as a function of speed — and the base's O(β⁴) error, drawn. Ω for two ORTHOGONAL boosts of equal speed β. Blue: the exact half-angle formula, verified point-by-point against an explicit 4×4 matrix product. It rises from the O(β²) line (grey dashed, Ω ≈ β²⁄2) at small β and saturates at 90° — a full quarter turn — as β → 1. Marked: 12.68° at β = 0.6, 42.90° at β = 0.9, 73.94° at β = 0.99. Orange, hugging the axis: the base article's claimed leading term ½(γ₁−1)(γ₂−1), which is O(β⁴) and is wrong by a factor of 20 000 at β = 0.01. It is not a worse approximation — it is the wrong POWER of the small parameter

Physics of the portrait

Origins · 2 minLlewellyn Hilleth Thomas and how this was discovered

Llewellyn Hilleth Thomas

1903–1992 · London · Cambridge · Copenhagen · Ohio State · Columbia · North Carolina

He was twenty-two, a Cambridge PhD student on a travelling fellowship at Bohr's institute in Copenhagen, and in a few weeks he solved a problem that had defeated the founders of quantum mechanics. In 1925 Uhlenbeck and Goudsmit proposed that the electron carries intrinsic spin with a $g$ -factor of $2$ . It explained the anomalous Zeeman effect beautifully — and gave the fine-structure splitting of hydrogen exactly twice the observed value. Heisenberg, Pauli and Bohr all knew the discrepancy; none could remove it, and Pauli thought the spinning electron dead because of it. Llewellyn Hilleth Thomas (born Lambeth, 21 October 1903) noticed that everyone had computed the spin–orbit coupling in the electron's instantaneous rest frame and then carried the answer back to the lab as if the rest frames at successive instants were related by a pure boost. They are not — the electron is on a curved orbit, the boosts are non-collinear, and the frame is rotating. The accumulated rotation is retrograde, at exactly half the naive spin-precession rate. Subtract it, and the factor of two disappears. He published it in Nature 117, 514 (1926), a note of barely a page, and gave the full kinematics in Philosophical Magazine 3, 1 (1927). Pauli accepted the spinning electron immediately afterwards. The kinematics was later placed on a general footing by Eugene Wigner, in his 1939 classification of the unitary representations of the Poincaré group (Ann. Math. 40, 149) — which is why the rotation carries both names, though Émile Borel had seen the underlying non-Euclidean geometry in 1913 and Ludwik Silberstein derived the rotation itself in his 1914 textbook. Thomas's later career went elsewhere: the Thomas–Fermi model of the atom, published in 1927 at twenty-three, independently of Fermi, and the ancestor of every density-functional calculation done today. He died in Raleigh, North Carolina, on 20 April 1992.

History

1913

— Émile Borel observes that the composition of non-collinear relativistic velocities involves the non-Euclidean geometry of velocity space, and that a kinematic rotation results.

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; the hydrogen fine-structure splitting comes out a factor of two too large, and no one can explain it.

1926

— Thomas, aged 22 and working in Copenhagen, publishes "The motion of the spinning electron" in Nature 117, 514, identifying the missing relativistic frame rotation and removing the factor of two.

1927

— Thomas gives the full treatment in "The kinematics of an electron with an axis", Philosophical Magazine 3, 1, and in the same year publishes the Thomas-Fermi model of the atom.

1939

— Wigner, in Annals of Mathematics 40, 149, classifies the little groups of the Poincaré group; the rotation becomes a general feature of relativistic kinematics and acquires his name.

1959

— Bargmann, Michel and Telegdi give the covariant equation for spin precession in electromagnetic fields, with the Thomas term built in; it is the working tool of every storage-ring g−2 measurement since.

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., §§14, 38. Primary: Thomas, Nature 117, 514 (1926) and Phil. Mag. 3, 1 (1927); Wigner, Ann. Math. 40, 149 (1939); Bargmann, Michel & Telegdi, Phys. Rev. Lett. 2, 435 (1959).

Provenance. Signature $(+,-,-,-)$ . $B(\mathbf{v})$ is the active pure boost; the generator sign in $[K_i,K_j] = -\varepsilon_{ijk}J_k$ is derived, not quoted. Three base defects found and repaired. (i) The base's summary gave the leading angle as $\tfrac12(\gamma_1-1)(\gamma_2-1)$ , which is $O(\beta^4)$ and contradicts the base's own small-velocity card; the correct $O(\beta^2)$ term is restored and the discrepancy quantified. (ii) The base's card "Rotation via Lorentz factors" is truncated to $\dots$ and has been discarded. (iii) The base printed the Thomas precession rate in two places, as $-\frac{\gamma^2}{\gamma+1}\frac{\mathbf{a}\times\mathbf{v}}{c^2}$ and $+\frac{\gamma^2}{\gamma+1}\frac{\mathbf{v}\times\mathbf{a}}{c^2}$ — consistent with each other, but both sign-flipped relative to Jackson §11.8 and relative to the sign this article derives from its own $4\times4$ polar decomposition. It is quoted here once, correctly. Every number printed on this page is recomputed in verify_aq-sr-thomas-wigner-rotation.py — the exact angles against an explicit $4\times4$ polar decomposition, the commutator entrywise, and the hydrogen chain from CODATA constants.

Builds on. Relativistic Velocity Addition · Transverse Velocity Transformation · Rapidity and the Additivity of Velocities · The Minkowski Metric. Leads to. Thomas Precession · Wigner Rotation · The Thomas Factor One-Half · Spin–Orbit Coupling.

If no torque acts, how can anything rotate?

Nothing rotates — the frames are rotated relative to one another. A gyroscope carried along is not torqued and does not change its state; but if you set up your axes by boosting from the lab in two steps in different directions, the axes you end up with are turned relative to those a single boost would have given you. The gyroscope, which has not moved relative to anything physical, now appears rotated relative to those axes. The rotation is a property of the map between frames, not of the object. It is observable nonetheless: carry the gyroscope around a closed loop in velocity space — as an orbiting electron does — and the accumulated rotation does not cancel. The axis comes back genuinely turned.

Why the HALF rapidities in the exact formula? Is that an approximation?

It is exact, and its origin is the double cover. The Lorentz group is covered twice by $SL(2,\mathbb{C})$ , in which a boost of rapidity $\varphi$ along $\hat{\mathbf{n}}$ is $\exp(\tfrac12\varphi\,\hat{\mathbf{n}}\cdot\boldsymbol{\sigma})$ — note the half. Multiply two such objects, split the product into a Hermitian factor and a unitary factor, and the unitary factor is the rotation; the halves come along for the ride. The same $\tfrac12$ is why a $2\pi$ rotation acts as $-1$ on a spinor. So $\tanh(\varphi/2)$ in (3) is no small-angle expansion — it is the spin- $\tfrac12$ structure of spacetime, showing up in a formula about coordinate axes.

Is the rotation the same as the "rotation" in the Terrell–Penrose effect?

No, and the coincidence of names is unfortunate. Terrell–Penrose is about what a camera records: light from different parts of a moving object leaves at different times, and the delays conspire to make a contracted sphere photograph as an undistorted sphere. It is an optical effect, about light-travel time, and it happens with a single boost. The Thomas–Wigner rotation is about the composition of two boosts and needs no light at all — it would be there in a universe with no photons. Both can be called "a moving thing looks rotated"; they have nothing else in common.

Does the Wigner rotation mean the velocity-addition law is wrong?

No — it means the velocity-addition law is incomplete as a description of the transformation. Compose two non-collinear boosts and the resulting velocity is exactly what the composition rules give: at $\beta_1 = \beta_2 = 0.6$ orthogonal, $\mathbf{w} = (0.6, 0.48, 0)c$ , the $0.48$ straight from the transverse rule. What velocity does not capture is the orientation of the frame you land in. A Lorentz transformation has six parameters; a velocity has three. Composing two boosts gives a general group element, and reading off only its velocity throws away the other three — of which, here, one is nonzero and is called $\Omega$ .

Why is velocity addition non-associative in three dimensions, if the Lorentz group is associative?

Because $\oplus$ on velocities is not the group operation. The group operation is matrix multiplication of Lorentz transformations, and it is perfectly associative. Velocity composition is what you get by multiplying two boosts and then throwing away the rotation — and that projection does not respect the group law. The result is a gyrogroup: associativity and commutativity hold only up to a rotation, and the rotation is $\Omega$ . "Velocity" is simply a bad coordinate on the Lorentz group in more than one dimension. In one dimension Velocity Composition and the Speed-of-Light Invariant rescues it with $\operatorname{artanh}$ ; in three there is no rescue, because the group is six-dimensional and three numbers will not do.

How big is the effect for a real relativistic particle — is it ever more than a curiosity?

It is routinely enormous. For two orthogonal boosts at $\beta = 0.9$ the rotation is $42.9^\circ$ ; at $\beta = 0.99$ , $73.9^\circ$ . In a storage ring a particle's velocity direction rotates continuously, so the Wigner rotation accumulates every turn and the resulting spin precession is not a perturbation but a dominant term — which is why the Bargmann–Michel–Telegdi equation exists. "Second-order relativistic effect" describes the $\beta \ll 1$ limit, not the phenomenon.