Thomas-Wigner Rotation and the Spin-Orbit Coupling Factor

Every fine-structure measurement in atomic physics. The $\tfrac12$ is not a curiosity that Dirac made obsolete — it is why the semiclassical estimate every undergraduate does works, and it is baked into the effective Hamiltonians that atomic and quantum-optics calculations still run on. Any time somebody writes $\xi(r)\,\mathbf S\cdot\mathbf L$ with $\xi = \frac{1}{2m^2c^2}\frac1r\frac{\mathrm dV}{\mathrm dr}$, the $2$ in that denominator is Thomas's.

The nuclear shell model — where the analogy is instructive precisely because it breaks. Nuclei have magic numbers (2, 8, 20, 28, 50, 82, 126) that no naive shell model reproduces. Goeppert-Mayer's insight in 1949 was that a large $\mathbf S\cdot\mathbf L$ term fixes them. But the nuclear spin–orbit coupling is not this one: it is far stronger than any electromagnetic estimate, and it has the opposite sign (it pushes $j = \ell+\tfrac12$ down, not up). It comes from the strong interaction, not from a boosted Coulomb field. The lesson is worth stating plainly: an $\mathbf S\cdot\mathbf L$ in a Hamiltonian is a shape, not a mechanism. Thomas's $\tfrac12$ belongs to the electromagnetic one and to no other.

Spintronics, with the same caveat. The Rashba and Dresselhaus couplings that drive spin transport in semiconductor heterostructures have exactly the form of equation (5) with the crystal's internal field in place of $\mathrm dV/\mathrm dr$. If you plug that field into the vacuum formula, you will underestimate the observed spin splitting by six orders of magnitude. The reason is that the $mc^2$ in the denominator is really a gap, and in a narrow-gap semiconductor the relevant gap is $\sim 1$ eV, not $511$ keV. The Thomas factor survives the transcription; the $c^2$ does not.

Storage rings, where the "$\tfrac12$" is a red herring. At $\gamma = 29.3$ — the muon $g-2$ ring — the kinematic prefactor is $28.3$, and the Thomas precession is twenty-eight times the cyclotron frequency. Nothing of the atomic story survives, and yet the physics is identical: the same $\boldsymbol\omega_T$, at a different $\gamma$. The covariant statement covering both is `aq-sr-bmt-equation`.

Ride with the electron. In its instantaneous rest frame the proton is not sitting still — it is orbiting you, once every $1.5\times10^{-16}$ s, and a circulating charge is a current loop. The loop puts a magnetic field through the electron. It is not a small field. Put the numbers in and

— comparable to the strongest superconducting magnets ever built, and the electron carries it around with it for free. The electron has a magnetic moment $\boldsymbol\mu = -\dfrac{ge}{2m}\mathbf S$; a moment in a field feels a torque; a torque on an angular momentum makes it precess. That is the dynamical precession $\boldsymbol\Omega_{\text{dyn}}$, and it is entirely correct. The energy of a precessing spin is $H = \boldsymbol\Omega\cdot\mathbf S$. Feed $\boldsymbol\Omega_{\text{dyn}}$ in and you get the naive Hamiltonian — and a fine-structure splitting twice what the spectrograph shows.

Now the part everyone missed. The frame you just did that calculation in is accelerating. Follow it round the orbit and you are not looking at one frame, you are looking at a continuously changing family of them, glued together by boosts whose directions keep swinging round. Boosts that point in different directions do not commute — canon `aq-sr-thomas-wigner-rotation` proves this: $B(\mathbf v_2)B(\mathbf v_1) = R(\Omega)\,B(\mathbf w)$, a boost and a leftover rotation. Take that rotation continuously around the orbit and the electron's rest frame is found to be turning, at the Thomas rate

— a rate derived, sign and all, by `aq-sr-thomas-precession`. Nothing torques it. It is what a frame does when you drag it around a bend in Minkowski space. So the lab sees the spin turn at $\boldsymbol\Omega_{\text{dyn}} + \boldsymbol\omega_T$. In a central field the acceleration is radial, and radial acceleration makes $\mathbf a\times\mathbf v$ point opposite to $\mathbf L$. The Thomas term therefore subtracts. And its size, once you put in $\mathbf a = -\hat{\mathbf r}(1/m)\,\mathrm dV/\mathrm dr$, is exactly one half of the $g=2$ dynamical term. Two minus one is one, and one is half of two. That is the whole story, and the factor of $\tfrac12$ is the arithmetic of it.

Everything below is SI. The electron has charge $-e$ ($e > 0$), mass $m$, spin $\mathbf S$, and $V(r)$ is its potential energy in the atom (so for hydrogen $V = -e^2/4\pi\varepsilon_0 r$ and $\mathrm dV/\mathrm dr > 0$).

(1) The field the electron actually sees. Transform the nucleus's electrostatic field into the electron's instantaneous rest frame. With $\mathbf B = 0$ in the lab,

where: $\mathbf E$ is the lab (nuclear) electric field at the electron and $\mathbf v$ its lab velocity. The $\gamma$ is a correction of order $\alpha^2 \approx 5\times10^{-5}$ in hydrogen and is dropped from here on. Note there is no factor of $c$ in the numerator — a Gaussian-unit $\mathbf B' = -\mathbf v\times\mathbf E/c$ leaking into an SI calculation is the single commonest way to lose the whole effect.

(2) The dynamical precession. The moment is $\boldsymbol\mu = -\dfrac{ge}{2m}\mathbf S$, and $\mathrm d\mathbf S/\mathrm dt = \boldsymbol\mu\times\mathbf B'$. Writing that as $\mathrm d\mathbf S/\mathrm dt = \boldsymbol\Omega\times\mathbf S$,

where: the sign is fixed by the electron's negative charge, and $g \approx 2$. This is where the base's card is broken. Its eq[5] prints $\boldsymbol\Omega_{\text{dyn}} = -\dfrac{ge}{2mc}\mathbf B'$ — with a spurious Gaussian $1/c$ and a spurious minus sign. Its own derivation step then quietly uses the correct SI expression. See the warning below.

(3) The kinematic precession — recalled, not re-derived. Owned by `aq-sr-thomas-precession`:

where: $\mathbf a$ is the lab acceleration. The order is $\mathbf a\times\mathbf v$ — not $\mathbf v\times\mathbf a$; the precession is retrograde. That sign is the difference between halving the coupling and tripling it.

(4) The spin turns at the sum.

where: $\mathbf L = m\,\mathbf r\times\mathbf v$. Read the bracket. It is $(g-1)/2$. The dynamical term carries $g/2$; the kinematic term carries $-\tfrac12$ and does not care what $g$ is — a gyroscope with no magnetic moment at all still precesses at $\boldsymbol\omega_T$. At $g = 2$ the bracket collapses to $\tfrac12$.

(5) The Hamiltonian. With $H = \boldsymbol\Omega\cdot\mathbf S$ (the relation that makes $\mathrm d\mathbf S/\mathrm dt = \boldsymbol\Omega\times\mathbf S$ come out of $[\,\cdot,H\,]$):

The $\tfrac12$ in front is the Thomas factor. Delete it and equation (5) becomes $H^{\text{naive}}_{SO}$, which is what Kronig, Uhlenbeck and Goudsmit all wrote down, and which is wrong by a factor of two.

1 · The 12-tesla magnet the electron carries with it

What magnetic field does the ground-state hydrogen electron actually see, and how fast do its two precessions run?

Step 1 — identify. Bohr orbit: $v = \alpha c = 2.1877\times10^{6}\ \mathrm{m/s}$, $r = a_0 = 5.2918\times10^{-11}\ \mathrm m$. The proton's field there is $|\mathbf E| = e/4\pi\varepsilon_0 a_0^2 = 5.1422\times10^{11}\ \mathrm{V/m}$.

Step 2 — apply. Equation (1): $|\mathbf B'| = |\mathbf v||\mathbf E|/c^2 = (2.1877\times10^{6})(5.1422\times10^{11})/(8.9875\times10^{16}) = \mathbf{12.5168\ T}$.

Step 3 — the boundary numbers. Equation (2) at $g = 2$: $\Omega_{\text{dyn}} = (e/m)|\mathbf B'| = (1.7588\times10^{11})(12.5168) = \mathbf{2.2015\times10^{12}\ rad/s}$. Equation (3): with $a = v^2/a_0 = 9.0442\times10^{22}\ \mathrm{m/s^2}$, $\omega_T = \tfrac12 av/c^2 = \mathbf{1.1008\times10^{12}\ rad/s}$, pointing the other way.

Step 4 — what it means. $\Omega_{\text{dyn}}/\omega_T = \mathbf{1.99992}$ — two, to five figures, and the deviation is exactly $(\gamma+1)/\gamma^2$, the $\gamma$-correction to the Thomas prefactor. Not noise. The lab sees $2.2015 - 1.1008 = 1.1008\times10^{12}$ rad/s: half the torque term. And note the scale. That $12.5$ T field is no perturbative wisp — it is stronger than an LHC dipole. Fine structure is a small effect not because the field is small but because the moment is: $\mu_B B' \approx 0.7\ \mathrm{meV}$ against $13.6$ eV, a ratio of order $\alpha^2$ — which is where the whole $\alpha^4$ scaling of fine structure comes from.

2 · Cash it out: hydrogen's $2P$ doublet, in megahertz

Take equation (5), evaluate it on hydrogen, and compare with the spectrograph.

Step 1 — identify. For $H_{SO} = \xi(r)\,\mathbf S\cdot\mathbf L$ we need $\langle\mathbf S\cdot\mathbf L\rangle = \tfrac{\hbar^2}{2}[\,j(j+1) - \ell(\ell+1) - s(s+1)\,]$. For $\ell = 1$: $j = \tfrac32$ gives $+\tfrac12\hbar^2$; $j = \tfrac12$ gives $-\hbar^2$. The gap is $\tfrac32\hbar^2$ — a ratio of $-2$ between the two shifts, which is why the $2P_{1/2}$ level moves down twice as far as the $2P_{3/2}$ moves up.

Step 2 — apply. The standard evaluation of $\langle \xi(r)\rangle$ with $\langle r^{-3}\rangle = \big[n^3 a_0^3\,\ell(\ell+\tfrac12)(\ell+1)\big]^{-1}$ gives, for $n=2$, $\ell=1$:

Step 3 — the boundary number. Divide by $h$: $\Delta \nu = \mathbf{10\,949\ MHz}$. Now delete the $\tfrac12$ and the answer doubles: $\mathbf{21\,899\ MHz}$.

Step 4 — what it means. The measured splitting is the sum of two of the most precisely known intervals in physics: the $2S_{1/2}$–$2P_{3/2}$ interval, $9911.200(12)$ MHz (Hagley & Pipkin 1994), plus the Lamb shift $2S_{1/2}$–$2P_{1/2}$, $1057.845(9)$ MHz (Lundeen & Pipkin 1981). Their sum is

The Thomas-corrected prediction is low by $0.18\%$. The uncorrected one is high by $\mathbf{99.6\%}$. And here is the check that closes the case: the Dirac equation — which knows nothing of rest frames, torques or Thomas, and simply solves a covariant wave equation — predicts a $2P$ fine-structure splitting of $\alpha^4mc^2/32$. The identical expression. Two derivations that share not a single step, agreeing to every digit, is not a coincidence. It is what it looks like when a semiclassical argument has been repaired correctly.

Texts. Jackson, Classical Electrodynamics, 3rd ed., §11.8 (classic, but check his units) · Griffiths, Introduction to Quantum Mechanics, ch. 6.3 · Commins, Electron Spin and Its History, Ann. Rev. Nucl. Part. Sci. 62 (2012). Primary: L. H. Thomas, Nature 117, 514 (1926) and Phil. Mag. 3, 1 (1927).

Data. $2S_{1/2}$–$2P_{3/2}$ $= 9911.200(12)$ MHz, Hagley & Pipkin, PRL 72, 1172 (1994). Lamb shift $2S_{1/2}$–$2P_{1/2}$ $= 1057.845(9)$ MHz, Lundeen & Pipkin (1981). Sum: $10\,969.0$ MHz.

Provenance. SI end to end. Signature $(+,-,-,-)$. The Thomas rate (3) is quoted from `aq-sr-thomas-precession`, which derives it twice; the non-closure of boosts is quoted from `aq-sr-thomas-wigner-rotation`, which owns it. This page derives the factor: that in any central potential the kinematic term is $-\tfrac12$ of the $g=2$ dynamical term, and the observable consequence in MHz. Base defects, corrected: eq[5] mixes Gaussian and SI and has the wrong sign; DERIV[1] step 2 is off by a factor of $\gamma$ and has the cross product backwards. Both are documented in the warnings and both are checked numerically in the verification script.

Claim-scope. The $s$-state ($\ell = 0$) fine structure — the Darwin term — is not derived here and is not part of $H_{SO}$. Neither is the relativistic kinetic correction $-p^4/8m^3c^2$, the same order and also in the full fine-structure formula; the $2P_{3/2}$–$2P_{1/2}$ splitting is used precisely because both cancel out of it, leaving the spin–orbit term alone. The Dirac equation is cited, not derived.

Builds on. Thomas Precession · The Thomas–Wigner Rotation · Four-Velocity. Leads to. The BMT Equation.