Fine Structure of Hydrogen

In 1887 — the same year his interferometer failed to find the ether — Albert Michelson pointed the instrument at a hydrogen lamp and noticed that the red Balmer line, H $\alpha$ , is not one line. It is a doublet, about $0.014$ nm wide: a fifth-decimal-place crack in what was supposed to be the simplest spectrum in nature. Nothing in the Bohr ladder, and nothing in the Schrödinger solution that replaced it, predicts that crack. Both give $E_n=-13.6\,\mathrm{eV}/n^2$ — energies labeled by $n$ alone. The three preceding serials built the repair pieces one at a time: a relativistic correction to the kinetic energy, the spin-orbit coupling, and the Darwin contact term — three physically unrelated-looking perturbations, each of order $\alpha^2$ relative to the Bohr energies, each with its own explicit $\ell$ -dependence. This article assembles them. The assembly is where the magic happens: added together, every trace of $\ell$ cancels, both $j=\ell\pm\tfrac12$ branches collapse onto one expression, and the $\ell=0$ case — where spin-orbit is absent and Darwin takes its place — lands on exactly the same formula. What survives is a spectrum labeled by $n$ and $j$ only. That $j$ -only structure is not luck. It is the leading shadow of the Dirac equation, whose exact Coulomb spectrum depends on $n$ and $j$ and nothing else — and this article's second derivation expands that exact formula and watches the perturbative master formula fall out of it, term by term. The one degeneracy the collapse leaves standing, $2s_{1/2}$ with $2p_{1/2}$ , is a loaded gun: the next serial fires it. Every splitting, wavelength, and census below is machine-verified.

Key takeaways

Fine structure = three $O(\alpha^4 mc^2)$ corrections on top of the Bohr spectrum: relativistic kinetic (Relativistic Kinetic Energy Correction), spin-orbit (Spin-Orbit Coupling), Darwin (Darwin Term) — consumed here, not re-derived.
Each piece depends on $\ell$ ; the sum does not: $E^{(1)}_{nj}=-\dfrac{(E_n^{(0)})^2}{2mc^2}\Big(\dfrac{4n}{j+1/2}-3\Big)$ — both $j=\ell\pm\tfrac12$ branches and the $\ell=0$ (Darwin) case verified to collapse onto this one formula.
The scale: fine structure is smaller than gross structure by $\alpha^2\approx5.3\times10^{-5}$ — at $n=2$ the $2p_{3/2}$ – $2p_{1/2}$ split is $4.53\times10^{-5}$ eV $=10.95$ GHz, exactly $|E_2|\,\alpha^2/4$ .
$2s_{1/2}$ and $2p_{1/2}$ (same $n$ , same $j$ , different $\ell$ ) stay exactly degenerate — the residual degeneracy that the Dirac equation protects and only QED (Lamb Shift) breaks.
The exact Dirac energy $E(n,j)$ , expanded in powers of $\alpha$ , reproduces rest energy + Bohr levels + this master formula at order $\alpha^4$ (sympy-verified series) — perturbation theory and the exact relativistic answer agree.
On H $\alpha$ the splitting is $\Delta\lambda=0.0157$ nm: resolving it needs $R\approx42{,}000$ , which is why Michelson's interferometer saw it in 1887 while every grating before had missed it.
Key idea: three separate relativistic corrections conspire so that only $n$ and $j$ label hydrogen's levels — $j$ , not $\ell$ , is the good quantum number of the relativistic Coulomb problem.

The core insight

Three physically distinct $O(\alpha^4 mc^2)$ corrections — relativistic kinetic, spin-orbit, Darwin — when added together give a single formula depending only on $n$ and $j$ , because $j$ (not $\ell$ ) is the good quantum number for the Dirac–Coulomb problem.

Where this lives today

Fine structure is now a tool more than a puzzle. Astronomers read stellar magnetic fields and cosmic abundances through fine-structure multiplets; the far-infrared [C II] 158 μm line — the same physics in many-electron dress — is among the brightest single lines galaxies emit, tracked across cosmological distances. In the lab, the $Z^4$ scaling (via Hydrogenic Ions and Scaling with Nuclear Charge) turns hydrogenic ions into stress tests: in hydrogen-like uranium the "fine" structure reaches hundreds of eV and probes QED where $Z\alpha$ is no longer small. And $\alpha$ itself — named for exactly this splitting — is measured today to better than $10^{-10}$ : every precision run at these intervals is Michelson's 1887 afternoon with fourteen more digits.

The physical picture

Start with the zoom ladder. At magnification one, hydrogen's spectrum is the Bohr ladder: levels at $-13.6\,\mathrm{eV}/n^2$ , each a degenerate stack of $2n^2$ states (Degeneracy and the Role of the Magnetic Quantum Number counts the $2\ell+1$ part; Accidental Degeneracy and the SO(4) Symmetry explains the rest). Zoom in by $\alpha^2\approx1/18{,}770$ and each rung resolves into a cluster of sublevels — the fine structure. Zoom by roughly another $\alpha$ and even the surviving coincidences split (the Lamb shift, next serial). Hydrogen is a Russian doll of spectra, and $\alpha$ sets the scale ratio between its shells.

Figure

The zoom ladder: gross structure, then fine structure, labeled by n then by j

Now the cast of characters. Why three corrections, and why these three? Because at order $(v/c)^2$ the electron is doing three relativistically naive things at once. Its kinetic energy is not $p^2/2m$ — that costs the $p^4$ correction of Relativistic Kinetic Energy Correction. Its magnetic moment rides through the proton's electric field, which in the electron's frame is partly a magnetic field grabbing the spin — the coupling of Spin-Orbit Coupling, Thomas factor included. And its position is smeared over a Compton wavelength of jitter, which softens the potential right at the nucleus — the contact term of Darwin Term. Same small parameter, three different limbs of the Hamiltonian. The wonder is what the sum does. Each piece carries explicit $\ell$ 's — the kinetic piece through $\langle1/r\rangle,\langle1/r^2\rangle$ , the spin-orbit piece through $\langle1/r^3\rangle$ and $\mathbf{L}\cdot\mathbf{S}$ , the Darwin piece through its all-or-nothing $\delta_{\ell0}$ . Yet the total depends on $j$ alone. A label the ingredients barely mention turns out to be the only label the answer keeps. When a computed quantity is blind to something its parts can see, a symmetry is hiding — the same logic that flagged the $\ell$ -degeneracy of the Coulomb problem as SO(4)'s fingerprint applies here, with the Dirac equation's structure as the culprit.

Figure

The n=2 assembly: three ℓ-dependent pieces, one j-labeled outcome

The three ingredients, on the table

Everything here is quoted from its owner serial; this article's job is the sum. The unperturbed energies are the Bohr levels of The Bohr Energy Spectrum and Rydberg Constant, $E_n^{(0)}=-13.6\,\mathrm{eV}/n^2$ , and every expectation value below is evaluated on the hydrogen states of Hydrogenic Wavefunctions and Quantum Numbers, with the radial moments $\langle1/r\rangle$ , $\langle1/r^2\rangle$ , $\langle1/r^3\rangle$ supplied by Radial Probability Distribution and Expectation Values — none of that is re-derived here. The kinetic correction, owned by Relativistic Kinetic Energy Correction, evaluates $-\hat p^4/8m^3c^2$ on level $n$ and finds

E^{(1)}_{\mathrm{rel}} = -\frac{(E_n^{(0)})^2}{2mc^2}\left(\frac{4n}{\ell+\tfrac12}-3\right) \qquad (1)

where: the $\ell$ -dependence enters through the radial moments — low- $\ell$ states dive closer to the nucleus, move faster, and pay a larger relativistic penalty. The correction is negative for every state (the true relativistic kinetic energy grows slower with $p$ than $p^2/2m$ pretends), and it alone already breaks the $\ell$ -degeneracy of the Bohr levels. The spin-orbit term, owned by Spin-Orbit Coupling (Thomas factor and all), is diagonal in the coupled basis $|n\,\ell\,j\,m_j\rangle$ built in Addition of Angular Momenta, where $\langle\mathbf{L}\cdot\mathbf{S}\rangle=\tfrac{\hbar^2}{2}[j(j+1)-\ell(\ell+1)-\tfrac34]$ :

E^{(1)}_{\mathrm{SO}} = \frac{(E_n^{(0)})^2}{mc^2}\cdot\frac{n\left[j(j+1)-\ell(\ell+1)-\tfrac34\right]}{\ell(\ell+\tfrac12)(\ell+1)} \qquad (2)

where: the sign flips between the doublet partners — $j=\ell+\tfrac12$ is pushed up, $j=\ell-\tfrac12$ pulled down — and the $\ell(\ell+\tfrac12)(\ell+1)$ downstairs is the $\langle1/r^3\rangle$ moment's signature. For $\ell=0$ the numerator vanishes identically while the moment diverges; the formula simply does not apply there, which is precisely the gap the third ingredient fills. The Darwin term, owned by Darwin Term, is a pure contact interaction — $\propto\delta^3(\mathbf{r})$ — so it reads off $|\psi_{n00}(0)|^2=1/(\pi n^3a^3)$ and touches nothing else:

E^{(1)}_{\mathrm{D}} = \frac{(E_n^{(0)})^2}{2mc^2}\,4n\,\delta_{\ell 0} \qquad (3)

where: only $s$ -states have amplitude at the origin, so only they are shifted — and shifted up: the Zitterbewegung smearing dilutes the potential's most attractive point. Numerically, for the $1s$ state this is $+7.24\times10^{-4}$ eV against the kinetic piece's $-9.06\times10^{-4}$ eV (both computed in the owner serials and re-verified here); the two nearly cancel, and their residual is the fine-structure shift.

The assembly

Derivation AThe j-only collapse: one formula from three corrections

We add equations (1), (2), and (3) and prove the total depends on $n$ and $j$ alone. The strategy is exhaustive case analysis — there are exactly three cases: $j=\ell+\tfrac12$ , $j=\ell-\tfrac12$ (both with $\ell\ge1$ , Darwin absent), and $\ell=0$ (spin-orbit absent, Darwin present). The thing to watch is that the two doublet branches carry different $\mathbf{L}\cdot\mathbf{S}$ eigenvalues and must nevertheless land on the same function of $j$ — that is where the cancellation is least obvious and most magical. We work in units of $u\equiv(E_n^{(0)})^2/2mc^2$ to keep the algebra readable.

Step 1. Case $j=\ell+\tfrac12$ . Here $j(j+1)-\ell(\ell+1)-\tfrac34=\ell$ , so (2) contributes $+2n\ell/[\ell(\ell+\tfrac12)(\ell+1)]\,u=2n/[(\ell+\tfrac12)(\ell+1)]\,u$ . Adding (1):

E^{(1)} = u\left[3-\frac{4n}{\ell+\tfrac12}+\frac{2n}{(\ell+\tfrac12)(\ell+1)}\right] = u\left[3-\frac{4n}{\ell+\tfrac12}\cdot\frac{(\ell+1)-\tfrac12}{\ell+1}\right] = u\left[3-\frac{4n}{\ell+1}\right]

and since $j+\tfrac12=\ell+1$ in this branch, the bracket is $3-4n/(j+\tfrac12)$ .

Step 2. Case $j=\ell-\tfrac12$ . Now $j(j+1)-\ell(\ell+1)-\tfrac34=-(\ell+1)$ , so (2) contributes $-2n/[\ell(\ell+\tfrac12)]\,u$ . Adding (1):

E^{(1)} = u\left[3-\frac{4n}{\ell+\tfrac12}-\frac{2n}{\ell(\ell+\tfrac12)}\right] = u\left[3-\frac{4n}{\ell+\tfrac12}\cdot\frac{\ell+\tfrac12}{\ell}\right] = u\left[3-\frac{4n}{\ell}\right]

and here $j+\tfrac12=\ell$ : the same bracket $3-4n/(j+\tfrac12)$ again, from completely different intermediate fractions. Why the two branches must agree: nothing in first-order perturbation theory forces it — it is an algebraic identity special to the Coulomb moments. Swap in any other central potential's $\langle1/r^2\rangle$ and $\langle1/r^3\rangle$ and the branches land on different functions of $j$ ; alkali doublets are split for exactly that reason. The coincidence is the perturbative echo of the Dirac equation's exact $j$ -only spectrum, checked independently in the next derivation.

Step 3. Case $\ell=0$ . Spin-orbit is absent — $\langle\mathbf{L}\cdot\mathbf{S}\rangle=0$ identically for an $s$ -state, as Spin-Orbit Coupling's delicacy card explains — and (1) at $\ell=0$ gives $u[3-8n]$ . The Darwin term (3) adds $+4nu$ :

E^{(1)} = u\left[3-8n+4n\right]=u\left[3-4n\right]

which is the master bracket evaluated at $j+\tfrac12=1$ , i.e. $j=\tfrac12$ — the only $j$ an $s$ -state can have. The Darwin term contributes exactly the amount that makes the $\ell=0$ case fall in line: the "conspiracy" Darwin Term's ℓ→0 miracle card verified from its side.

Step 4. Collect all three cases:

E^{(1)}_{nj} = -\frac{(E_n^{(0)})^2}{2mc^2}\left(\frac{4n}{j+\tfrac12}-3\right) = -\frac{13.6\,\mathrm{eV}}{n^2}\,\frac{\alpha^2}{n^2}\left(\frac{n}{j+\tfrac12}-\frac34\right)

The second form uses $E_n^{(0)}=-\alpha^2mc^2/2n^2$ (the Bohr energy written the relativistic way, per The Bohr Radius and Characteristic Atomic Scales): fine structure is the gross structure times $\alpha^2/n^2$ times a bracket of order one. Every state moves down (the bracket is positive for all $j\le n-\tfrac12$ ), and higher $j$ moves less. Verified: both branch collapses and the ℓ=0 case symbolically (sympy, general n and ℓ); the 1s assembly numerically: −9.057×10⁻⁴ + 7.245×10⁻⁴ = −1.811×10⁻⁴ eV = master formula at n=1, j=1/2 (article verifier).

Read the master formula against the states it governs. At fixed $n$ , the levels fan out by $j$ : for $n=3$ , the sublevels $j=\tfrac12,\tfrac32,\tfrac52$ carry brackets $4n/1-3=9$ , $4n/2-3=3$ , $4n/3-3=1$ — a descending staircase with the highest $j$ least disturbed. And at fixed $j$ , the two towers $\ell=j\pm\tfrac12$ sit at identical energy:

E_{2s_{1/2}}=E_{2p_{1/2}}, \qquad E_{3p_{3/2}}=E_{3d_{3/2}}, \qquad \dots \qquad (4)

where: this is not an approximation surviving by neglect — it is exact at this order, and the exact Dirac spectrum protects it to all orders. The pairing fails only at the top of each manifold, $j=n-\tfrac12$ , whose sole parent is $\ell=n-1$ . Counting states: each $n$ still holds $\sum_\ell 2(2\ell+1)=2n^2$ states (verified programmatically for $n\le4$ ), now organized into $j$ -ladders instead of $\ell$ -towers. The label changed; the census didn't.

Go deeper · Derivation B · 2 minThe Dirac ladder check: expanding the exact energy

Derivation BThe Dirac ladder check: expanding the exact energy

The Dirac equation solves the relativistic Coulomb problem exactly (the solution itself belongs to the relativistic serials at the end of this track), and its spectrum is famously $\ell$ -blind. We quote that exact result and expand it in powers of $\alpha$ , checking that our perturbative master formula is precisely its order- $\alpha^4$ shadow. This is the strongest available audit of the assembly: an entirely independent route — no perturbation theory, no basis choices, no Thomas factor bookkeeping — must reproduce the same coefficient. With $k\equiv j+\tfrac12$ , the exact levels are

E^{\mathrm{Dirac}}_{nj} = mc^2\left[1+\left(\frac{\alpha}{\,n-k+\sqrt{k^2-\alpha^2}\,}\right)^2\right]^{-1/2}

Step 1. Expand the inner square root: $\sqrt{k^2-\alpha^2}=k-\alpha^2/2k+O(\alpha^4)$ , so the denominator is $n-\alpha^2/2k+O(\alpha^4)$ — the "effective principal quantum number" dips below $n$ by an amount controlled by $j$ alone. Already here the $\ell$ -blindness is manifest: $\ell$ appears nowhere in the exact formula.

Step 2. Expand the outer bracket to fourth order in $\alpha$ (binomial series, two terms deep — done symbolically, not by hand, in the verifier):

E^{\mathrm{Dirac}}_{nj} = mc^2\left[1-\frac{\alpha^2}{2n^2}-\frac{\alpha^4}{2n^4}\left(\frac{n}{k}-\frac34\right)+O(\alpha^6)\right]

Step 3. Read off the ladder. The leading term is the rest energy $mc^2$ . The $\alpha^2$ term is $-\alpha^2mc^2/2n^2=E_n^{(0)}$ — the entire Bohr spectrum arrives as the first correction to sitting still. The $\alpha^4$ term is exactly $-\,(E_n^{(0)})^2/2mc^2\cdot(4n/k-3)$ : our master formula, coefficient and all. Why this audit matters beyond reassurance: it certifies that the three perturbations were the complete order- $(v/c)^2$ list — nothing forgotten, nothing double-counted. If, say, the Thomas half had been dropped in the spin-orbit serial, Step 2's coefficient would disagree and the mismatch would appear here as a wrong bracket.

Step 4. What the exact formula adds beyond order $\alpha^4$ : the $O(\alpha^6)$ terms (tiny for hydrogen, respectable for high- $Z$ hydrogenic ions per Hydrogenic Ions and Scaling with Nuclear Charge, since the true expansion parameter is $Z\alpha$ ), and a hard boundary — the expression turns complex when $Z\alpha>k$ , the relativistic Coulomb problem's famous $Z\approx137$ pathology for $j=\tfrac12$ . Both stories belong to the relativistic serials; here they are one honesty line each. Verified: the full series expansion symbolically (sympy, atomic k): α⁰ coefficient = 1, α² coefficient = −1/2n², α⁴ coefficient = −(1/2n⁴)(n/k − 3/4), matching the master formula exactly (article verifier).

Limiting cases

case	result	meaning
$\alpha\to0$	$E^{(1)}_{nj}\to0$	fine structure is a relativistic effect; the Schrödinger spectrum is the $c\to\infty$ world
$j=n-\tfrac12$ (top of manifold)	bracket $=1$ : $E^{(1)}=-\,(E_n^{(0)})^2/2mc^2$	the circular-orbit states are disturbed least — slowest, farthest from the nucleus
$j=\tfrac12$ (bottom)	bracket $=4n-3$ : largest shift	penetrating states pay the full relativistic bill
$n=1$	single level $1s_{1/2}$ , shift $-1.81\times10^{-4}$ eV	no splitting — one $j$ only; the ground state just sinks
same $j$ , different $\ell$	exactly degenerate ( $2s_{1/2}=2p_{1/2}$ )	Dirac-protected; broken only by QED — the Lamb Shift serial
$Z\alpha\to k$	Dirac formula turns complex	perturbative story dead long before; relativistic serials' territory

Common trapThe three pieces are not optional courses you may take separately

Two standing errors, both fatal on exams. First: quoting the spin-orbit result alone for an $s$ -state. The spin-orbit formula is a $0/0$ at $\ell=0$ — $\langle\mathbf{L}\cdot\mathbf{S}\rangle$ vanishes while $\langle1/r^3\rangle$ diverges — and its naive " $\ell\to0$ limit" happens to equal the Darwin term's contribution, a coincidence explained in Darwin Term, not a license. Physically an $s$ -state has no orbital motion to couple to; what shifts it is the contact term. Second: adding the corrections in the uncoupled basis $|n\,\ell\,m_\ell\,m_s\rangle$ . Spin-orbit is not diagonal there, and within the degenerate $n$ -manifold the off-diagonal elements are first-order large — the situation Degenerate Perturbation Theory diagnoses. The coupled basis $|n\,\ell\,j\,m_j\rangle$ is not a convenience; it is the set of good states certified by the theorem of Good Quantum Numbers and Symmetry-Adapted Bases.

Worked examples

Example 1 — the n=2 manifold, assembled end-to-end. Place all three $n=2$ levels using the owner serials' formulas, and verify the master formula's verdict. Work in units of $u_2\equiv(E_2^{(0)})^2/2mc^2$ with $E_2^{(0)}=-3.401$ eV, so $u_2=1.132\times10^{-5}$ eV (computed).

Step 1. The $2s_{1/2}$ level ( $\ell=0$ , $j=\tfrac12$ ). Kinetic: bracket $4n/(\ell+\tfrac12)-3=16-3=13$ , so $-13\,u_2$ . Darwin: $+4n\,u_2=+8\,u_2$ . Spin-orbit: zero. Net: $-5\,u_2$ .

Step 2. The $2p_{1/2}$ level ( $\ell=1$ , $j=\tfrac12$ ). Kinetic: bracket $8/\tfrac32-3=\tfrac73$ , so $-\tfrac73\,u_2$ . Spin-orbit with $\langle\mathbf{L}\cdot\mathbf{S}\rangle=-\hbar^2$ , hence $2\langle\mathbf{L}\cdot\mathbf{S}\rangle/\hbar^2=-2$ (the $j=\ell-\tfrac12$ branch): $2n\cdot(-2)/[1\cdot\tfrac32\cdot2]\,u_2=-\tfrac83\,u_2$ . Darwin: zero. Net: $-\tfrac73-\tfrac83=-5\,u_2$ . Same as $2s_{1/2}$ , from entirely different bookkeeping — the degeneracy equation (4) live.

Step 3. The $2p_{3/2}$ level ( $\ell=1$ , $j=\tfrac32$ ). Kinetic: $-\tfrac73\,u_2$ as before. Spin-orbit with $\langle\mathbf{L}\cdot\mathbf{S}\rangle=+\tfrac{\hbar^2}{2}$ , hence $2\langle\mathbf{L}\cdot\mathbf{S}\rangle/\hbar^2=+1$ : $+2n\cdot1/3\,u_2=+\tfrac43\,u_2$ . Net: $-1\,u_2$ — the master bracket $4\cdot2/2-3=1$ . ✔

Step 4. The splitting: $2p_{3/2}-2p_{1/2}=4\,u_2=4.53\times10^{-5}$ eV $=10.95$ GHz (computed). It has a pretty closed form: $4u_2=(E_2^{(0)})^2\cdot2/mc^2=|E_2^{(0)}|\,\alpha^2/4$ exactly in this leading-order theory (verified to $10^{-9}$ relative) — the $\alpha^2$ suppression made flesh. The measured interval is $10\,969$ MHz — the extra $\sim20$ MHz above our $10\,949$ MHz is reduced mass plus QED, physics owned by later serials. Answer: $2s_{1/2}$ and $2p_{1/2}$ at $-5u_2$ , $2p_{3/2}$ at $-u_2$ ; split $4.53\times10^{-5}$ eV.

Example 2 — what Michelson saw: the Hα doublet. Convert the $n=2$ splitting into wavelength and decide what instrument resolves it.

Step 1. H $\alpha$ is the $n=3\to2$ Balmer line: $E_\gamma=13.6(1/4-1/9)=1.889$ eV, $\lambda=656.3$ nm (computed). The line's fine structure is dominated by the $n=2$ splitting (the $n=3$ brackets are smaller by the extra powers of $1/n$ : its full fan spans about a third as much).

Step 2. A shift $\delta E$ of either terminal level moves the line by $\Delta\lambda=\lambda\,\delta E/E_\gamma$ . For the dominant interval $\delta E=4.53\times10^{-5}$ eV: $\Delta\lambda=656.3\times(4.53\times10^{-5}/1.889)=0.0157$ nm (computed).

Step 3. Resolving power needed: $R=\lambda/\Delta\lambda\approx42{,}000$ (computed). A prism manages hundreds; 1880s gratings a few thousand in practical use on faint sources. Michelson's interferometer, reading fringe visibility as the mirror separation grew, had no such ceiling — which is how 1887 could see a splitting that theory would not explain until 1916 (and not correctly until 1928). His reported width, $\sim0.014$ nm, sits honestly near our leading-order $0.0157$ nm: the true H $\alpha$ pattern is seven components with Doppler blurring, so the "doublet separation" one reads off is slightly compressed. Answer: $\Delta\lambda\approx0.016$ nm, needing $R\approx4\times10^4$ — interferometer territory in 1887.

Origins · 1 minAlbert Abraham Michelson and how this was discovered

Albert Abraham Michelson

1852–1931 · Cleveland, Clark, Chicago

Michelson spent 1887 failing to find the ether — and, in the same building with the same interferometer, finding this article's subject. Turning the instrument from starlight to spectroscopy, he measured how interference fringes washed out as the path difference grew: the visibility curve, in modern language the Fourier transform of the spectral line shape. Sodium's D lines gave oscillating visibility (a doublet); then hydrogen's red H $\alpha$ did too — "a double line… perhaps 1/60 the distance of the sodium lines," about $0.014$ nm. Fine structure had been measured, twenty-nine years before Sommerfeld's relativistic orbits first computed its scale and forty-one before Dirac explained it. Michelson had no theory and wanted none ("a grand underlying theory" he famously doubted the need of); he catalogued the splitting the way he catalogued everything — as metrology, the fringe count as the fact. The 1907 Nobel Prize (the first to an American scientist) cited precisely this: precision optical instruments and the investigations performed with them. The deepest irony in the story: the man who never accepted relativity delivered, with the H $\alpha$ doublet, one of the first numbers relativity would ever have to explain — and the visibility-curve method he invented to do it is the direct ancestor of Fourier-transform spectroscopy, which today resolves that doublet as routine calibration.

History

1887

— Michelson's interferometric visibility curves reveal Hα as a close doublet — fine structure measured, unexplained

1916

— Sommerfeld's relativistic old-quantum orbits produce a splitting formula in n and a new constant α

1925

— Goudsmit and Uhlenbeck's spin, plus Thomas's factor, rebuild the doublet mechanism inside quantum mechanics

1928

— Dirac's equation delivers the exact j-only spectrum; the perturbative pieces become its expansion

1938

— Williams's precision Hα interferometry maps the full seven-component fine-structure pattern

1947

— the residual 2s–2p degeneracy falls: the Lamb shift opens the QED era

Going further

Griffiths & Schroeter §7.3 — the three corrections and the master formula (Eq. 7.66) at the source register.
Sakurai & Napolitano §5.3 — fine structure with the symmetry logic foregrounded.
Cohen-Tannoudji, Diu & Laloë Vol. II, Complement XII — the full degenerate-perturbation bookkeeping.
Bethe & Salpeter §§4–5 — the classic exhaustive treatment, including the Dirac expansion.
Shankar 2nd ed. Ch. 17 — a compact assembly with the Dirac comparison.

Provenance note: Equations and numerical factors follow Griffiths §7.3 (master formula Eq. 7.66), Sakurai & Napolitano §5.3, Bethe & Salpeter §§4–5, and Cohen-Tannoudji Vol. II Complement XII. A verifier should: (1) confirm the Thomas factor of 1/2 in the spin-orbit term; (2) recompute $\langle 1/r^3\rangle = 1/[a^3 n^3 \ell(\ell+1/2)(\ell+1)]$ and $|\psi_{n00}(0)|^2 = 1/(\pi a^3 n^3)$ ; (3) verify the cancellation of $\ell$ across both $j=\ell\pm 1/2$ branches; (4) check the ground-state shift $E_{1,1/2}^{(1)}\approx -1.81\times10^{-4}\,\mathrm{eV}$ using $E_1^{(0)}=-13.6\,\mathrm{eV}$ , $mc^2=511\,\mathrm{keV}$ ; (5) confirm agreement with the $O(\alpha^4)$ expansion of the exact Dirac energy. Constants: $\alpha=1/137.036$ , $a=0.529\,\text{\AA}$ (CODATA). — All five items are executed in this article's verify script (items 1–2 as consumed-input audits against the owner serials' formulas; items 3–5 symbolically and numerically). Edition note: the individual corrections are OWNED by Relativistic Kinetic Energy Correction, Spin-Orbit Coupling, and Darwin Term — this article quotes their results and owns only the assembly, the $j$ -only collapse, and the Dirac-expansion audit. The Thomas-precession derivation and the exact Dirac solution stay with their owners (the spin-orbit serial's portrait band and the relativistic serials respectively). The old wiki carries a same-title page ("fine-structure-of-hydrogen"); that collision is cutover-gate business, not this article's.

Why is it called "fine" structure — fine compared to what?

Compared to the gross structure: the Bohr spacings of order eV. Fine-structure intervals are $\alpha^2\approx5\times10^{-5}$ times smaller — tens of μeV, GHz in frequency. The name is literal spectroscopy: lines that looked single under coarse dispersion revealed fine subdivision under better instruments. The hierarchy continues downward — the Lamb shift finer by roughly another power of $\alpha$ , hyperfine structure suppressed by $m_e/m_p$ instead (later serials).

Each correction depends on ℓ — how can the sum possibly not?

Because the $\ell$ -dependences are not independent: all three pieces are built from Coulomb expectation values ( $\langle1/r\rangle,\langle1/r^2\rangle,\langle1/r^3\rangle,|\psi(0)|^2$ ) whose specific forms conspire. Substitute the two $\mathbf{L}\cdot\mathbf{S}$ eigenvalues and the Coulomb moments, and the branch algebra collapses — $\ell+1$ appears where $j+\tfrac12$ should for one branch, $\ell$ for the other, and both are $j+\tfrac12$ in their respective branches. In any non-Coulomb potential the conspiracy fails and levels genuinely depend on both $\ell$ and $j$ — which is why sodium's doublets don't follow the hydrogen pattern.

Is the degeneracy of 2s₁/₂ and 2p₁/₂ exact or just first-order?

Exact — within the Dirac theory of a point Coulomb charge, at every order of the $\alpha$ expansion. What finally splits them ( $\sim1058$ MHz) is physics outside the Dirac equation: the quantized radiation field's radiative corrections. That is the Lamb shift, the next serial's whole subject — and why this degeneracy is famous rather than merely tidy.

Why does j label the answer when the perturbations were computed with ℓ and s machinery?

Because $\hat J^2$ and $\hat J_z$ commute with the full fine-structure Hamiltonian while $\hat L_z$ and $\hat S_z$ separately do not — spin-orbit conserves only the total. The good-states theorem of Good Quantum Numbers and Symmetry-Adapted Bases then hands the degenerate manifold to $(j,m_j)$ . The deeper reason $\ell$ drops out of the energies (not just off the diagonal) is the Dirac equation's extra structure — the relativistic analogue of the SO(4) accident that made the Bohr levels $\ell$ -blind in the first place.

Why does the measured 10 969 MHz differ from the computed 10 949?

Reduced mass and QED. The electron-proton pair orbits its common center of mass ( $m\to\mu=m_em_p/(m_e+m_p)$ , about one part in 1836), and radiative corrections add small pieces on top — both sub-percent. This article's numbers are the point-nucleus, infinite-proton-mass, order- $\alpha^4$ values, the right register for the formula being verified; metrology-grade values belong to the precision serials.

What happens to fine structure for large Z?

Everything scales up brutally: the expansion parameter is really $Z\alpha$ , so fine structure grows as $Z^4$ against the gross structure's $Z^2$ . In hydrogen-like iron ( $Z=26$ ) the $2p$ splitting is already a factor $26^4\approx4.6\times10^5$ larger than hydrogen's; in hydrogen-like uranium the perturbative expansion itself is dying, and one must use the exact Dirac spectrum plus all-order QED. The $j=\tfrac12$ Dirac formula turns complex at $Z\alpha>1$ ( $Z\gtrsim137$ ) — the famous relativistic Coulomb catastrophe, told properly in the relativistic serials.

Michelson measured the splitting in 1887 — why did explaining it take until 1928?

The explanation needed three revolutions that didn't exist yet: relativity (1905), quantum orbits (1913-16), and spin (1925). Sommerfeld's 1916 relativistic orbits got the formula's shape right by a fortunate cancellation — his integer $k$ playing the role Dirac's $j+\tfrac12$ would later justify — and the correct mechanism arrived only with spin, Thomas precession, and finally the Dirac equation. Forty-one years from measurement to explanation: precision data patiently outwaiting theory.