Relativistic Corrections to the Hydrogen Atom

The Bohr–Schrödinger spectrum $E_n = -13.6\,\text{eV}/n^2$ is a spectacular success and a deliberate lie: it treats the electron as a slow, spinless, pointlike particle. Look closely at a hydrogen line with a good spectrograph and each "level" fractures into a cluster of sublevels separated by a part in $10^{5}$ — the fine structure. The perturbative chapters already know how to compute it: three correction operators added to the Schrödinger Hamiltonian, each evaluated by hand. This article asks the deeper question those chapters defer — where do the three operators come from, and why exactly three? The answer is that all three fall out of one equation. When the four-component Dirac equation in a Coulomb field is block-diagonalized by the Foldy–Wouthuysen transformation — pushing the reduction to order $1/c^2$ — the even part of the reduced Hamiltonian contains precisely three new operators at order $\alpha^4 mc^2$ : the mass-velocity term $-\hat p^4/8m^3c^2$ , the spin-orbit term $\tfrac{1}{2m^2c^2}\tfrac1r\tfrac{dV}{dr}\,\mathbf S\!\cdot\!\mathbf L$ , and the Darwin term $\tfrac{\hbar^2}{8m^2c^2}\nabla^2V$ . Not two, not four — three, and no more, at this order. The teaching punchline lives in the middle term. The classic 1926 estimate of spin-orbit coupling — boost to the electron's rest frame, let its moment feel the orbiting proton's field — comes out twice too large, and is rescued only by a separate argument about Thomas precession. The Dirac equation supplies that factor of $\tfrac12$ automatically: it is already inside the FW coefficient, with no rest frame invoked. That the same reduction also produces the Darwin term (owned as a Zitterbewegung effect in The Darwin Term and Zitterbewegung) and the mass-velocity term (the perturbative version owned by Relativistic Kinetic Energy Correction) is the sign that relativity, spin, and their coupling are one structure, not three. This article owns that assembly; the closed-form sum they collapse to is developed in Exact Dirac Solution of the Hydrogen Atom.

Key takeaways

Fine structure is the set of $O(\alpha^4 mc^2)$ corrections to the gross Bohr levels; this article owns their Dirac origin via the Foldy–Wouthuysen reduction, not the perturbative bookkeeping.
The FW reduction to order $1/c^2$ yields exactly three even operators: mass-velocity $-\hat p^4/8m^3c^2$ , spin-orbit $\tfrac{1}{2m^2c^2}\tfrac1r\tfrac{dV}{dr}\mathbf S\!\cdot\!\mathbf L$ , and Darwin $\tfrac{\hbar^2}{8m^2c^2}\nabla^2V$ — each scaling as $\alpha^4$ (checked).
Owned punchline: the Thomas factor $\tfrac12$ in spin-orbit is automatic in the Dirac coefficient — no ad-hoc precession argument, unlike the classical estimate which overshoots by exactly $2\times$ .
The three combine into a formula depending only on $n$ and $j$ : $E^{(1)}_{nj} = \tfrac{(E_n^{(0)})^2}{2mc^2}\big(3 - \tfrac{4n}{j+1/2}\big)$ ; the $\ell$ -dependence cancels.
Numeric anchors (verified): the fine-structure scale is $\alpha^2\,\text{Ry} = 7.24\times10^{-4}$ eV; equivalently $\tfrac12 mc^2\alpha^4$ ; splittings grow as $Z^4$ .
Perturbative evaluation of $\langle 1/r^3\rangle$ and the $\mathbf S\!\cdot\!\mathbf L$ eigenvalue is owned by Spin-Orbit Coupling by name; I supply the operator and its coefficient from Dirac.
Key idea: One Foldy–Wouthuysen reduction of the Dirac–Coulomb Hamiltonian produces exactly the three fine-structure operators at order $\alpha^4$ , with the spin-orbit Thomas factor $\tfrac12$ built in — so the "patchwork" of corrections is really one relativistic structure.

The core insight

Reducing the Dirac equation in a Coulomb field to order $1/c^2$ (Foldy–Wouthuysen) yields exactly three even operators of order $\alpha^4 mc^2$ — mass-velocity, spin-orbit, and Darwin — with the spin-orbit Thomas factor $\tfrac12$ supplied automatically, and their sum depends only on $n$ and $j$ .

Where this lives today

Fine structure is no longer a curiosity — it is the metrology floor of atomic physics. Precision hydrogen spectroscopy (the source of the Rydberg constant and a competitive value of $\alpha$ ) works by measuring deviations from the Dirac fine structure: the QED Lamb Shift (which splits $2s_{1/2}$ from $2p_{1/2}$ , developed under that name), hyperfine structure from the proton spin, the finite proton radius, and recoil. Because the Dirac prediction is exact at the one-body level, each small effect is defined as a departure from the level scheme this article assembles — the proton-radius puzzle (electronic vs. muonic hydrogen) is literally a disagreement about one such correction. In heavy hydrogen-like ions the $Z^4$ growth turns fine structure from a $10^{-4}$ eV whisper into a keV effect that storage rings and electron-beam ion traps measure routinely, testing the Dirac framework where its own $Z\alpha$ expansion is about to fail.

The physical picture

Watch the electron near perihelion, at the inner reach of its orbit, where it moves fastest — $v/c \sim Z\alpha$ . Three different relativistic facts bite there at once. First, its kinetic energy is not $p^2/2m$ but $\sqrt{p^2c^2+m^2c^4}-mc^2$ ; the correction $-p^4/8m^3c^2$ is always negative, so relativity makes a fast electron slightly less bound than the Bohr formula says. Second, in the electron's instantaneous frame the proton circulates as a current loop, and the magnetic field it makes torques the electron's spin — spin-up-along- $\mathbf L$ and spin-down-along- $\mathbf L$ split apart. Third, the Dirac electron is never quite at a point: a jitter at the Compton scale smears it over a tiny ball, so it samples an average of the Coulomb potential rather than its value at a point. The subtle one is the second. The naive rest-frame estimate double-counts, because that frame is non-inertial — it is being swung in a circle, and a sequence of infinitesimal boosts along a curving path compounds into a net rotation, the Thomas precession. That precession undoes exactly half of the naive magnetic coupling. In the standard perturbative story you have to know this and insert the $\tfrac12$ by hand. The Dirac equation never boosts to a rest frame at all — it is Lorentz-covariant from the start — so the $\tfrac12$ is simply there, sitting in the coefficient the FW reduction hands you. The figure below is the whole article in one frame: three operators of utterly different physical character, all order $\alpha^4$ , adding up to a single ladder of levels labelled by $n$ and $j$ alone. That a mass correction, a magnetic coupling, and a contact smearing conspire to erase the $\ell$ -dependence is the first hint that the Dirac equation knows something the three separate perturbations do not — a hint cashed out fully in the exact spectrum.

Figure

Three Foldy–Wouthuysen operators at order α⁴ collapse to one n,j level

The math

Everything below flows from one object: the reduced Hamiltonian the Foldy–Wouthuysen transformation produces from the Dirac–Coulomb equation. The systematic FW machinery — the order-by-order unitary that decouples the large and small spinor components — is developed in Foldy–Wouthuysen Transformation and recalled here by name; we take its order- $1/c^2$ output as the starting point and read the three fine-structure operators straight off it. The free-particle expansion of $\sqrt{p^2c^2+m^2c^4}$ that seeds the mass-velocity term is the perturbative content of Relativistic Kinetic Energy Correction, recalled by name. The reduced even Hamiltonian is

H_{\rm FW} = mc^2 + \frac{p^2}{2m} + V \;\underbrace{-\;\frac{\hat p^4}{8m^3c^2}}_{\text{mass-velocity}} \;+\; \underbrace{\frac{1}{2m^2c^2}\frac1r\frac{dV}{dr}\,\mathbf S\!\cdot\!\mathbf L}_{\text{spin-orbit}} \;+\; \underbrace{\frac{\hbar^2}{8m^2c^2}\nabla^2 V}_{\text{Darwin}} \qquad (1)

where: $mc^2$ is the rest energy; $p^2/2m + V$ is the Schrödinger Hamiltonian; and the three braced operators are the fine-structure corrections. Read it as a hierarchy in $v/c$ : the first line is order $(v/c)^0$ and $(v/c)^2$ , and every braced term is order $(v/c)^4 \sim \alpha^4$ for a Coulomb-bound electron. There is nothing between the Schrödinger Hamiltonian and these three — that "exactly three" is the content of the reduction, not an assumption. The reason each braced operator sits at the same order $\alpha^4 mc^2$ is a scaling argument worth making explicit, because it is what makes fine structure a single, unified correction rather than a random collection. In a Coulomb-bound state the characteristic momentum is $p \sim m c\,\alpha$ (since $v/c\sim\alpha$ ) and the characteristic radius is the Bohr radius $a_0 = \hbar/(mc\alpha)$ . Feed those scales in:

\frac{\hat p^4}{8m^3c^2} \sim \frac{(mc\alpha)^4}{m^3c^2} = mc^2\,\frac{\alpha^4}{8}, \qquad \frac{\hbar^2}{8m^2c^2}\nabla^2V \sim \frac{\hbar^2}{m^2c^2}\frac{e^2}{a_0^3} = mc^2\,\frac{\alpha^4}{8} \qquad (2)

where: $e^2 = \alpha\hbar c$ in Gaussian units and $\nabla^2V \sim V/r^2 \sim e^2/a_0^3$ . Both operators land on $mc^2\,\alpha^4/8$ — verified symbolically below — and the spin-orbit term carries the same scaling through $\tfrac1r\tfrac{dV}{dr}\sim e^2/a_0^3$ . Since the gross structure is order $mc^2\alpha^2$ , the corrections are suppressed by $\alpha^2 \approx 5\times10^{-5}$ : hence "fine."

Derivation AThe spin-orbit coefficient and the automatic Thomas one-half

We show that the Foldy–Wouthuysen reduction delivers the spin-orbit operator with coefficient $\tfrac{1}{2m^2c^2}$ — the value that already contains the Thomas factor $\tfrac12$ — and we do it without ever transforming to the electron's rest frame. The strategy is to isolate the piece of the reduced Hamiltonian that couples spin to the potential gradient, using only the Dirac algebra and the Pauli identity, and then to contrast the result with the naive lab-frame estimate to locate where the classical argument doubles the answer. The one thing to watch: the factor of $\tfrac12$ is not put in — it emerges from the $1/8m^2c^2$ that the double commutator carries, combined with the split into symmetric and antisymmetric parts.

Step 1. The FW reduction's order- $1/c^2$ correction from the odd operator $\mathcal O = c\,\boldsymbol\alpha\!\cdot\!\mathbf p$ acting against the even potential $\mathcal E = V$ is the double commutator $-\tfrac{1}{8m^2c^2}[\mathcal O,[\mathcal O,\mathcal E]]$ . Written out, $[\mathcal O,[\mathcal O,V]] = c^2[\boldsymbol\alpha\!\cdot\!\mathbf p,[\boldsymbol\alpha\!\cdot\!\mathbf p, V]]$ , and using $[p_i,V] = -i\hbar\,\partial_i V$ twice this reduces to a sum of a spin-independent piece and a spin-dependent piece.

Step 2. Split the reduced operator into its symmetric and antisymmetric parts,

-\frac{1}{8m^2c^2}[\mathcal O,[\mathcal O,V]] = \frac{\hbar^2}{8m^2c^2}\nabla^2V \;+\; \frac{\hbar}{4m^2c^2}\,\boldsymbol\Sigma\!\cdot\!(\nabla V\times\mathbf p).

Why the split matters: the symmetric part is the Darwin term (no spin structure, just $\nabla^2V$ ), and the antisymmetric part is spin-orbit. The single coefficient $\tfrac{1}{8m^2c^2}$ feeds both — the Darwin coefficient is $\tfrac18$ , and the same prefactor gives spin-orbit its $\tfrac14$ , which becomes $\tfrac12$ after the spin operator is introduced.

Step 3. For a central potential $\nabla V = \tfrac{\mathbf r}{r}\tfrac{dV}{dr}$ , so $\nabla V\times\mathbf p = \tfrac1r\tfrac{dV}{dr}(\mathbf r\times\mathbf p) = \tfrac1r\tfrac{dV}{dr}\,\mathbf L$ . Substituting and using $\mathbf S = \tfrac{\hbar}{2}\boldsymbol\Sigma$ (so $\boldsymbol\Sigma = 2\mathbf S/\hbar$ ),

H_{\rm SO} = \frac{\hbar}{4m^2c^2}\,\frac1r\frac{dV}{dr}\,\boldsymbol\Sigma\!\cdot\!\mathbf L = \frac{1}{2m^2c^2}\,\frac1r\frac{dV}{dr}\,\mathbf S\!\cdot\!\mathbf L.

The coefficient is $\tfrac{1}{2m^2c^2}$ , boxed and final.

Step 4. Now the naive estimate, for contrast. In the electron's rest frame the field is $\mathbf B' = -\tfrac{1}{c^2}\mathbf v\times\mathbf E$ , and $H_{\rm naive} = -\boldsymbol\mu\!\cdot\!\mathbf B'$ with $\boldsymbol\mu = -\tfrac{g_s e}{2m}\mathbf S$ , $g_s=2$ . Carrying this through gives coefficient $\tfrac{1}{m^2c^2}$ — exactly twice the Dirac value. Why the classical argument is wrong by two: it treats the electron frame as inertial. Restoring the Thomas precession of that accelerating frame multiplies by $\tfrac12$ ; the Dirac equation, being covariant, never made the inertial-frame error, so its coefficient is already the correct $\tfrac{1}{2m^2c^2}$ . *Verified numerically: the ratio (Dirac coefficient)/(naive coefficient) $= \tfrac12$ exactly, and the operator's prefactor $\tfrac{1}{2m^2c^2}$ scales as $\alpha^4 mc^2$ on Coulomb states — both checked in the verifier.*

Go deeper · Derivation B · 1 minThe three operators sum to a formula in n and j alone

Derivation BThe three operators sum to a formula in n and j alone

We assemble the three first-order energy shifts and show the $\ell$ -dependence cancels, leaving a result that depends only on $n$ and $j$ . The route: evaluate the mass-velocity and spin-orbit shifts for $\ell\neq0$ (the spin-orbit vanishes for $\ell=0$ ), show their sum is already $\ell$ -free, then check the Darwin term seamlessly continues the same formula down to $\ell=0$ where spin-orbit died. The perturbative radial averages $\langle 1/r\rangle,\langle 1/r^2\rangle,\langle 1/r^3\rangle$ and the $\mathbf S\!\cdot\!\mathbf L$ eigenvalue are the machinery of Spin-Orbit Coupling and Relativistic Kinetic Energy Correction, recalled by name; I use their results, not re-derive them.

Step 1. The mass-velocity shift, using $\hat p^2 = 2m(E_n^{(0)}-V)$ to avoid $\langle p^4\rangle$ directly, is

\langle H_{\rm kin}\rangle = -\frac{(E_n^{(0)})^2}{2mc^2}\Big[\frac{4n}{\ell+\tfrac12} - 3\Big],

negative and $\ell$ -dependent. This is the content of Relativistic Kinetic Energy Correction; it is quoted here to be summed.

Step 2. The spin-orbit shift for $\ell\neq0$ , with $\langle 1/r^3\rangle = 1/[a_0^3 n^3\ell(\ell+\tfrac12)(\ell+1)]$ (Kramers, owned by Spin-Orbit Coupling) and $\langle\mathbf S\!\cdot\!\mathbf L\rangle = \tfrac{\hbar^2}{2}[j(j{+}1)-\ell(\ell{+}1)-\tfrac34]$ , is

\langle H_{\rm SO}\rangle = \frac{(E_n^{(0)})^2}{mc^2}\,\frac{n\big[j(j{+}1)-\ell(\ell{+}1)-\tfrac34\big]}{\ell(\ell+\tfrac12)(\ell+1)}.

Why we can use the coupled basis: $H_{\rm SO}\propto\mathbf S\!\cdot\!\mathbf L$ commutes with $\mathbf J^2,\mathbf L^2,\mathbf S^2$ , so it is already diagonal in $|n,\ell,j,m_j\rangle$ — no degenerate-perturbation matrix to diagonalize.

Step 3. Add them. For both branches $j=\ell\pm\tfrac12$ the algebra collapses to a single expression free of $\ell$ :

\langle H_{\rm kin}+H_{\rm SO}\rangle = \frac{(E_n^{(0)})^2}{2mc^2}\Big(3 - \frac{4n}{j+\tfrac12}\Big) \qquad (\ell\neq0).

The cancellation is exact and is the crux — two $\ell$ -dependent terms produce a $j$ -only sum.

Step 4. Close the $\ell=0$ gap with Darwin. There $H_{\rm SO}=0$ (no $\mathbf L$ ), but $\langle H_{\rm D}\rangle = \tfrac{\hbar^2}{8m^2c^2}\tfrac{e^2}{\epsilon_0}|\psi_{n00}(0)|^2$ with $|\psi_{n00}(0)|^2 = 1/(\pi n^3 a_0^3)$ , giving $\langle H_{\rm D}\rangle = \tfrac{(E_n^{(0)})^2}{2mc^2}\cdot 4n$ . Since $j=\tfrac12$ for $s$ -states, $4n/(j+\tfrac12)=4n$ , and the Darwin value is exactly what the (vanished) spin-orbit term would have supplied. The unified result

E^{(1)}_{nj} = \frac{(E_n^{(0)})^2}{2mc^2}\Big(3 - \frac{4n}{j+\tfrac12}\Big)

holds for all $\ell$ . Why this is a miracle and not an accident: it is the $O(\alpha^4)$ shadow of the exact Dirac spectrum, which depends on $n$ and $j$ by an exact hidden symmetry — developed in Exact Dirac Solution of the Hydrogen Atom. *Verified numerically: for $n=2$ , $E^{(1)}_{2,1/2}$ (the $2s_{1/2},2p_{1/2}$ value) $= -5.66\times10^{-5}$ eV and $E^{(1)}_{2,3/2}$ (the $2p_{3/2}$ value) $= -1.13\times10^{-5}$ eV, so the $2p_{3/2}-2p_{1/2}$ splitting is $4.53\times10^{-5}$ eV $= 0.365\ \text{cm}^{-1}$ ; and $E^{(1)}_{2,1/2}$ is identical for $2s_{1/2}$ and $2p_{1/2}$ — checked in the verifier.*

Limiting cases

case	result	meaning
$\ell = 0$ ( $s$ -states)	$\langle\mathbf S\!\cdot\!\mathbf L\rangle = 0$ ; $\langle H_{\rm D}\rangle = \tfrac{\pi\hbar^2 e^2}{2m^2c^2}\vert\psi(0)\vert^2$	spin-orbit vanishes, Darwin fills the gap so the $j$ -formula stays valid
$\ell \neq 0$	$\langle H_{\rm D}\rangle = 0$ ; $\langle H_{\rm SO}\rangle\neq0$	Darwin vanishes ( $\psi(0)=0$ ); kinetic and spin-orbit do the work
$\alpha\to0$ (nonrelativistic)	$E_{nj}\to -\tfrac{mc^2\alpha^2}{2n^2}$	corrections scale as $\alpha^4$ , vanish relative to gross structure — recover Bohr
high- $Z$ hydrogenic ion	$\Delta E_{\rm fs}\propto (Z\alpha)^4$	splittings grow as $Z^4$ ; the $1/c$ expansion fails as $Z\alpha\to1$
naive rest-frame estimate	spin-orbit coefficient $\tfrac{1}{m^2c^2}$	wrong by $2\times$ ; the Dirac/FW value $\tfrac{1}{2m^2c^2}$ has Thomas $\tfrac12$ built in

Common trapThe three operators come from three separate physical arguments

The most common misconception is that mass-velocity, spin-orbit, and Darwin are three independent corrections you happen to add together — a mass fix, a magnetic fix, a smearing fix, each with its own hand-crafted coefficient. They are not independent: they are the complete even part of one Foldy–Wouthuysen reduction at order $1/c^2$ , and their coefficients are locked together by the Dirac algebra. The tell is the spin-orbit $\tfrac12$ . In the "three separate arguments" picture you must import Thomas precession from special relativity to fix a factor of two; in the Dirac picture that $\tfrac12$ is already inside the coefficient, and the same $\tfrac{1}{8m^2c^2}$ that sets it also sets the Darwin term. If your derivation needs an external Thomas argument, you are doing the perturbative story (owned by Spin-Orbit Coupling), not the Dirac story — and you should expect exactly the $\tfrac12$ the covariant equation already knew.

Worked examples

Example 1 — the fine-structure energy scale (confidence builder). Estimate the size of the fine-structure corrections in hydrogen from first principles, and confirm they are $\alpha^2$ below the gross structure.

Step 1. Set the scale. Every correction is order $\tfrac12 mc^2\alpha^4$ . With $mc^2 = 5.110\times10^{5}$ eV and $\alpha = 1/137.036$ , compute $\alpha^2 = 5.325\times10^{-5}$ and $\alpha^4 = 2.836\times10^{-9}$ . Then $\tfrac12 mc^2\alpha^4 = \tfrac12(5.110\times10^{5})(2.836\times10^{-9})$ eV $= \boxed{7.24\times10^{-4}\ \text{eV}}$ .

Step 2. Cross-check against the Rydberg. The Rydberg energy is $\text{Ry} = \tfrac12 mc^2\alpha^2 = 13.606$ eV, so $\tfrac12 mc^2\alpha^4 = \text{Ry}\cdot\alpha^2 = 13.606\times5.325\times10^{-5}$ eV $= \boxed{7.24\times10^{-4}\ \text{eV}}$ — the two routes agree, as the verifier confirms.

Step 3. Read the suppression. The fine-structure shift is $\alpha^2 \approx 5\times10^{-5}$ of the $13.6$ eV gross scale — a part in $\sim20{,}000$ . This is why it took precise spectroscopy to see it, and why the word "fine" is apt: the structure is there in every line, buried $\alpha^2$ deep.

Example 2 — locating the $n=2$ sublevels (exam level). Compute the fine-structure shifts of the $n=2$ level of hydrogen and give the $2p_{3/2}$ – $2p_{1/2}$ splitting in eV and cm $^{-1}$ .

Step 1. Set up. The gross energy is $E_2^{(0)} = -13.606/4 = -3.4015$ eV, so $(E_2^{(0)})^2/(2mc^2) = (3.4015)^2/(2\cdot5.110\times10^{5})$ eV $= 1.132\times10^{-5}$ eV. The shift is $E^{(1)}_{2j} = (1.132\times10^{-5})\,(3 - 4\cdot2/(j+\tfrac12))$ eV $= (1.132\times10^{-5})(3 - 8/(j+\tfrac12))$ eV.

Step 2. Evaluate $j=\tfrac12$ (the $2s_{1/2}$ and $2p_{1/2}$ states). Here $j+\tfrac12 = 1$ , so the bracket is $3 - 8 = -5$ , and $E^{(1)}_{2,1/2} = (1.132\times10^{-5})(-5)$ eV $= \boxed{-5.66\times10^{-5}\ \text{eV}}$ . Both $2s_{1/2}$ and $2p_{1/2}$ get this identical shift — the degeneracy the Lamb shift later breaks.

Step 3. Evaluate $j=\tfrac32$ (the $2p_{3/2}$ state). Here $j+\tfrac12 = 2$ , so the bracket is $3 - 4 = -1$ , and $E^{(1)}_{2,3/2} = (1.132\times10^{-5})(-1)$ eV $= \boxed{-1.13\times10^{-5}\ \text{eV}}$ .

Step 4. Take the difference. The $2p_{3/2}$ – $2p_{1/2}$ fine-structure splitting is $(-1.13\times10^{-5}) - (-5.66\times10^{-5}) = \boxed{4.53\times10^{-5}\ \text{eV}}$ . Converting with $1\ \text{eV} = 8065.54\ \text{cm}^{-1}$ gives $0.365\ \text{cm}^{-1}$ , equivalently $\sim10.9$ GHz — the classic hydrogen fine-structure interval, matched to the exact Dirac spectrum in the verifier.

Figure

Three operators, one number — how the 2s and 2p shifts are built from kinetic, spin-orbit, and Darwin

Historical development

Origins · 1 minthomas precession and how this was discovered

thomas precession

1903–1992 · Cambridge and Ohio State, kinematic relativity

The story of the factor of two is the story of Llewellyn Thomas. In 1925 Goudsmit and Uhlenbeck's electron spin explained atomic doublets beautifully — except the fine-structure splitting came out twice too large, a discrepancy embarrassing enough to nearly sink the spin hypothesis. In 1926 Thomas, then twenty-three, showed the error was kinematic: the electron's rest frame is accelerating, and a succession of Lorentz boosts along a curved path composes into a net rotation — now the Thomas precession — that runs at half the orbital frequency and halves the naive spin-orbit coupling. It rescued spin. What makes the Dirac equation (1928) so striking is that it never needs Thomas's argument at all: the covariant equation, reduced to the nonrelativistic regime, produces the spin-orbit term with the correct $\tfrac12$ automatically. Thomas's precession is the classical shadow of a factor the relativistic wave equation carries intrinsically — a rare case where a subtle correction to the old theory turns out to be a built-in feature of the deeper one.

History

1916

— Sommerfeld derives the fine-structure formula from relativistic Bohr–Sommerfeld orbits — right answer, spinless (wrong) reason

1925

— Goudsmit and Uhlenbeck propose electron spin; doublet splitting comes out a factor 2 too large

1926

— Thomas explains the factor 2 as a kinematic precession of the accelerating electron frame

1928

— Dirac's equation yields all three fine-structure operators, with the Thomas one-half automatic

1947

— Lamb and Retherford measure the 2s₁⁄₂–2p₁⁄₂ splitting, exposing the limit of the Dirac picture

Going further

References. Griffiths, Introduction to Quantum Mechanics, 3rd ed., §7.3 · Sakurai & Napolitano, Modern Quantum Mechanics, 2nd ed., §5.3, §8.2 · Bjorken & Drell, Relativistic Quantum Mechanics, §4.3 (Foldy–Wouthuysen) · Shankar, Principles of Quantum Mechanics, 2nd ed., Ch. 17 · Bethe & Salpeter, Quantum Mechanics of One- and Two-Electron Atoms, §11–14 · Foldy & Wouthuysen, Phys. Rev. 78, 29 (1950).

Provenance. The reduced Hamiltonian (1) and the spin-orbit coefficient $\tfrac{1}{2m^2c^2}$ with the automatic Thomas $\tfrac12$ are verified against the Foldy–Wouthuysen reduction (Bjorken–Drell §4.3). Owned and shown here: (i) the FW reduction yields exactly three even $O(\alpha^4)$ operators, with the $\alpha^4$ scaling of the mass-velocity and Darwin operators checked symbolically ( $mc^2\alpha^4/8$ each); (ii) the spin-orbit Thomas $\tfrac12$ as the ratio (Dirac)/(naive) $=\tfrac12$ ; (iii) the collapse to the $j$ -only formula, with the $n=2$ shifts $E^{(1)}_{2,1/2}=-5.66\times10^{-5}$ eV, $E^{(1)}_{2,3/2}=-1.13\times10^{-5}$ eV and the $2p_{3/2}$ – $2p_{1/2}$ splitting $4.53\times10^{-5}$ eV $=0.365$ cm $^{-1}$ ; (iv) the fine-structure scale $\alpha^2\,\text{Ry}=7.24\times10^{-4}$ eV. The perturbative evaluation of $\langle 1/r\rangle,\langle1/r^2\rangle,\langle1/r^3\rangle$ and the $\mathbf S\!\cdot\!\mathbf L$ eigenvalue is owned by Spin-Orbit Coupling and Relativistic Kinetic Energy Correction by name; the closed-form Dirac spectrum they sum to is developed in Exact Dirac Solution of the Hydrogen Atom; the residual $2s_{1/2}$ – $2p_{1/2}$ splitting is the QED Lamb Shift, stated as beyond this article. All numbers CODATA-standard; verifier `verify_aq-qm-relativistic-corrections-hydrogen.py`.

Builds on: The Dirac Equation · Foldy–Wouthuysen Transformation · Nonrelativistic Reduction and the Pauli Equation · Spin from the Dirac Equation

Leads to: Spin-Orbit Coupling from Relativity · The Darwin Term and Zitterbewegung · Exact Dirac Solution of the Hydrogen Atom · Fine Structure of Hydrogen

If the perturbative chapters already compute fine structure, what does the Dirac derivation add?

It adds the why. The perturbative treatment (owned by Relativistic Kinetic Energy Correction, Spin-Orbit Coupling, and the Darwin Term article) writes down three operators and evaluates them, but it cannot say why there are exactly three, why their coefficients have the values they do, or why the spin-orbit term needs a Thomas factor of $\tfrac12$ . The Dirac derivation answers all three at once: the operators are the complete even part of one Foldy–Wouthuysen reduction at order $1/c^2$ , their coefficients are fixed by the Dirac algebra, and the $\tfrac12$ is built into the coefficient. Same numbers, deeper reason — and the reason matters when you go to the exact spectrum, where the same $(n,j)$ dependence is exact rather than accidental.

Why does the classical rest-frame estimate of spin-orbit coupling give exactly twice the right answer?

Because it treats the electron's instantaneous rest frame as if it were inertial. The electron is accelerating (moving in a curved orbit), so its rest frame is a sequence of frames related by boosts in slightly different directions. Composing such boosts produces a net rotation — the Thomas precession — that the naive calculation omits. That precession opposes the spin's Larmor precession and runs at a rate that halves the effective coupling. The factor is exactly $\tfrac12$ , not approximately, because it comes from the $\tfrac12$ in the relativistic velocity-addition/rotation kinematics. The Dirac equation never boosts to a rest frame, so it never makes the error, and its coefficient needs no correction.

Why do mass-velocity and spin-orbit each depend on ℓ, but their sum does not?

Each term separately carries $\ell$ through the radial averages: mass-velocity through $\langle 1/r\rangle,\langle 1/r^2\rangle$ (giving a $4n/(\ell+\tfrac12)$ factor) and spin-orbit through $\langle 1/r^3\rangle$ and the $\mathbf S\!\cdot\!\mathbf L$ eigenvalue. When you add them and substitute $j=\ell\pm\tfrac12$ , the $\ell$ -dependent pieces cancel algebraically, leaving $3-\tfrac{4n}{j+1/2}$ . This is not luck — it is the low-order manifestation of the exact Dirac–Coulomb symmetry that makes the full spectrum depend only on $n$ and $j$ . The cancellation is a promise that the three perturbations are secretly one object.

Do these corrections make hydrogen levels go up or down?

Down overall — but which state moves how far depends on $j$ . The combined shift $E^{(1)}_{nj}=\tfrac{(E_n^{(0)})^2}{2mc^2}(3-\tfrac{4n}{j+1/2})$ is negative for every physical state, because the bracket $3-\tfrac{4n}{j+1/2}$ is negative whenever $n>\tfrac34(j+\tfrac12)$ . Higher $j$ gives a less negative bracket, so $2p_{3/2}$ sits above the degenerate $2s_{1/2}/2p_{1/2}$ pair. The mass-velocity term alone always lowers; spin-orbit can raise or lower depending on spin alignment; the net at fixed $n$ is a small spreading of the level into a $j$ -labelled multiplet.

How does the fine-structure splitting scale with nuclear charge, and why does that matter?

It scales as $Z^4$ . The coupling strengths carry factors of $Z$ (through the potential and its gradient) and the radial averages carry more (through $\langle 1/r^3\rangle\propto Z^3$ ), combining to $(Z\alpha)^4 mc^2$ . This steep growth is why fine structure is a $10^{-4}$ eV whisper in hydrogen but a keV-scale effect in uranium's hydrogen-like ion — measurable in electron-beam ion traps and used to test QED in strong fields. It also warns you: as $Z\alpha\to1$ (around $Z\approx137$ ) the expansion parameter approaches unity and the perturbative $\alpha^4$ ordering breaks down; you must then use the exact Dirac spectrum, and eventually field theory.

Is the Darwin term really "filling in" for spin-orbit at ℓ=0, or is that a coincidence?

It is a genuine continuation, not a coincidence. For $\ell\neq0$ the combined kinetic+spin-orbit shift is $\tfrac{(E_n^{(0)})^2}{2mc^2}(3-\tfrac{4n}{j+1/2})$ . At $\ell=0$ spin-orbit vanishes, but $j=\tfrac12$ forces $4n/(j+\tfrac12)=4n$ , and the Darwin term evaluates to exactly $\tfrac{(E_n^{(0)})^2}{2mc^2}\cdot4n$ — the value the formula demands. Both terms trace to the same $\tfrac{1}{8m^2c^2}$ double commutator in the FW reduction (symmetric vs antisymmetric parts), so their agreement at $\ell=0$ is the reduction being self-consistent, and ultimately the exact spectrum's smooth $j$ -dependence.