Lamb Shift

Fine Structure of Hydrogen ended with a loaded gun: after every order- $\alpha^4$ correction is paid, $2s_{1/2}$ and $2p_{1/2}$ sit at exactly the same energy — and not by approximation. The exact Dirac spectrum depends on $n$ and $j$ only, so the pair is degenerate to all orders in $\alpha$ . Relativistic quantum mechanics, taken at its word, says there is nothing between those two levels. Nothing, forever. There is something. It is about $1057.8$ MHz — four millionths of an electron-volt, a microwave photon 28 centimeters long — and it is called the Lamb shift. Its existence means the picture of one electron in a fixed Coulomb field is incomplete: the electromagnetic field is itself a quantum system, jittering with zero-point fluctuations even in vacuum, and the electron cannot opt out. The jitter smears the electron over a fraction of its Compton wavelength; a smeared charge feels a softened Coulomb pit; and only $s$ -states — the ones that actually sit in the pit — notice. So $2s_{1/2}$ floats up off its Dirac-protected perch while $2p_{1/2}$ barely moves. This article does three things. It makes the "there should be nothing" case airtight, so the $1057.8$ MHz reads as the anomaly it is. It runs the Welton smearing estimate end-to-end — the vacuum-fluctuation cousin of the Darwin Term — and watches $\alpha^5\ln(1/\alpha)$ physics land within shouting distance of the measured number. And it takes apart the 1947 measurement: a metastable beam, a quench detector, and a resonance ridden downhill to zero field. Every number is machine-verified; the corrupt export base was healed and rebuilt (edition note at the foot).

Key takeaways

The Dirac–Coulomb spectrum depends only on $(n,j)$ : $2s_{1/2}$ and $2p_{1/2}$ should be degenerate to all orders in $\alpha$ . Measured: $2s_{1/2}$ sits $1057.8$ MHz ( $4.37\times10^{-6}$ eV) above $2p_{1/2}$ .
The cause is beyond any single-particle equation: the quantized radiation field's zero-point fluctuations. The electron jitters, the Coulomb singularity gets diluted, and only states with $|\psi(0)|^2\neq0$ — $s$ -states — are shifted (up).
Welton's heuristic makes this quantitative: $\langle(\delta r)^2\rangle=\tfrac{2\alpha}{\pi}(\hbar/mc)^2\ln(\dots)$ smearing plus the Darwin-style contact evaluation gives $\Delta E\sim\alpha^5mc^2\ln(1/\alpha)/n^3$ — computed honestly here: $1000$ - $1300$ MHz depending on the log's cutoffs.
Bethe's 1947 renormalization arithmetic (quoted, structure only) sharpens that to $\approx1040$ MHz; modern QED and experiment agree at the kHz level.
The measurement rode two levers: 2s metastability ( $\tau\approx0.12$ s against 2p's $1.6$ ns — a factor $10^{7.9}$ ) and Zeeman tuning of the resonance, extrapolated to zero field.
Scaling: $\propto Z^4\alpha^5/n^3$ times logs — the $1s$ shift is $8173$ MHz (ratio $7.7$ , not $8$ : the log softens $1/n^3$ ); in muonic hydrogen the contact factor grows by $(186)^3\approx6\times10^6$ , which is why the proton's size shows up there.
Key idea: the vacuum is not empty — its zero-point jitter smears the electron, weakens the Coulomb pit only where $s$ -states live, and splits a degeneracy the Dirac equation says is exact.

The core insight

The Dirac–Coulomb spectrum leaves states of equal $n,j$ but different $\ell$ degenerate; coupling the bound electron to the quantized radiation field (vacuum fluctuations + self-energy, renormalized) breaks this degeneracy, raising $s$ -states because only they have amplitude at the nucleus.

Where this lives today

The Lamb shift's second career is metrology. Frequency-comb spectroscopy of the $1s$ – $2s$ line (at $10^{-15}$ relative precision) tests bound-state QED only as well as the $1s$ Lamb shift — $8173$ MHz, now computed through two-loop and recoil terms — is known; inverted, the same equations extract the Rydberg constant and the proton charge radius. The muonic-hydrogen program made that inversion dramatic: with the contact factor amplified $6.4\times10^6$ -fold (computed), the 2010 CREMA measurement returned a proton radius five standard deviations from the accepted value, and a decade of re-measurement ended with the muon vindicated and CODATA revised. At the strong-field frontier, hydrogen-like uranium's $1s$ Lamb shift is $\sim460$ eV, measured in storage rings against all-order QED; and the Welton picture returns in circuit QED, where engineered vacua shift artificial-atom levels by design.

The physical picture

Hold the "nothing should be here" note firmly first. The fine-structure assembly of Fine Structure of Hydrogen was checked against the exact Dirac energy, whose formula contains $n$ and $j+\tfrac12$ and nothing else — expand it to any order and $2s_{1/2}$ , $2p_{1/2}$ ride together. This is not the fragile degeneracy of the Bohr levels; it is structural. Finding a splitting there is finding daylight between two sides of an algebraic identity: the equation itself must be missing physics. The missing physics is the field. Quantize the electromagnetic field and every mode keeps a zero-point half-quantum even in "empty" space. A bound electron is permanently bathed in these fluctuations, kicked at every frequency from the atomic scale up to its own Compton frequency. The slow kicks it follows adiabatically; the fast ones it cannot, and their residue is a jitter of its position, $\delta r$ , uncorrelated with the orbit.

Figure

Vacuum jitter smears the electron — and only s-states sit where it matters

The payoff structure is one this chapter has met before. In Darwin Term, Zitterbewegung smeared the electron over a Compton wavelength and produced a $\delta^3(\mathbf{r})$ contact correction — order $\alpha^4$ , already inside the Dirac equation. Here the smearing source is different (the field's zero-point motion, not the interference of Dirac components) and the smear is smaller by $\alpha\ln(1/\alpha)$ — but the evaluation grammar is identical: Taylor-expand the averaged potential, keep $\tfrac16\langle(\delta r)^2\rangle\nabla^2V$ , collect $|\psi(0)|^2$ . The Lamb shift is the Darwin term's little sibling, paid by the vacuum — same contact structure, one more power of $\alpha$ , which is why it splits $2s_{1/2}$ from $2p_{1/2}$ and why it is ten times smaller than the fine-structure interval. One more feature deserves its own sentence: the logarithm. The jitter integral $\int d\omega/\omega$ runs over every field frequency between the atomic scale (below which the electron follows the kick — no net smear) and the Compton scale (above which the nonrelativistic picture dies). Neither cutoff is sharp, so the answer carries $\ln$ of their ratio — the signature that all frequency decades in between contribute equally.

There should be nothing here

The claim to beat, quoted from Fine Structure of Hydrogen and re-audited in this article's verifier:

E^{\mathrm{Dirac}}_{nj} = mc^2\left[1+\left(\frac{\alpha}{\,n-(j+\tfrac12)+\sqrt{(j+\tfrac12)^2-\alpha^2}\,}\right)^2\right]^{-1/2} \qquad (1)

where: $\ell$ appears nowhere — the formula cannot tell $2s_{1/2}$ from $2p_{1/2}$ even in principle (a symbolic free-symbols audit in the verifier confirms the only labels are $n$ and $j+\tfrac12$ ). Any measured splitting between those states is therefore a clean signal of physics beyond the relativistic one-electron equation: no basis subtlety, no missed correction term, no reduced-mass bookkeeping can manufacture it. And the measurement says:

\Delta E(2s_{1/2}-2p_{1/2}) = h\times1057.8\,\mathrm{MHz} = 4.37\times10^{-6}\,\mathrm{eV} \qquad (2)

where: the modern value is $1057.845$ MHz with sub-kHz uncertainties; the sign puts $2s_{1/2}$ above $2p_{1/2}$ . For scale: one part in $10^6$ of the $n=2$ binding energy, a tenth of the $2p$ fine-structure interval, wavelength $28$ cm — hardware territory, which is exactly how it was found.

Derivation AWelton smearing: the vacuum pays a Darwin-style contact bill

We estimate the shift of an $s$ -state by letting the zero-point field jiggle the electron and evaluating the smeared Coulomb potential — the same contact-term grammar as Darwin Term, with the vacuum as the smearing source. This is an honest heuristic, not QED: we keep every step's logic visible, flag the two cutoffs that produce the logarithm, and expect the right power law, sign, $n$ -dependence, and order of magnitude — not the exact coefficient. That humility is the point: when a back-of-envelope gets within thirty percent of a Nobel-grade number, the mechanism is confirmed.

Step 1. Response to one mode. A field mode of angular frequency $\omega$ carries a zero-point electric amplitude; a free-ish electron driven by $\mathbf{E}_\omega e^{-i\omega t}$ oscillates with $m\,\ddot{\delta\mathbf{r}}=-e\mathbf{E}_\omega$ , i.e. $\delta r_\omega = eE_\omega/m\omega^2$ . High-frequency modes shake it weakly but there are many of them; summing $|\delta r_\omega|^2$ over the vacuum's mode density gives

\langle(\delta r)^2\rangle = \frac{2\alpha}{\pi}\left(\frac{\hbar}{mc}\right)^2\int\frac{d\omega}{\omega}

Why the integral is logarithmic: each decade of frequency contributes equally — the mode density's $\omega^2$ growth, the zero-point energy's $\omega$ , and the response's $1/\omega^4$ conspire to exactly $1/\omega$ . Physics must supply both cutoffs by hand: below the atomic frequency ( $\sim$ the Rydberg scale over $\hbar$ ) the electron follows the field adiabatically and accumulates no net smear; above $\omega\sim mc^2/\hbar$ the nonrelativistic response is wrong. The ratio of cutoffs is what the logarithm measures.

Step 2. The smeared potential. With jitter uncorrelated across directions, $\langle V(\mathbf{r}+\delta\mathbf{r})\rangle = V(\mathbf{r})+\tfrac16\langle(\delta r)^2\rangle\nabla^2V$ — the linear term averages away, the quadratic term survives with the isotropic $\tfrac16$ . For the Coulomb potential, $\nabla^2(1/r)=-4\pi\delta^3(\mathbf{r})$ — the identity owned and proven in The Radial Equation's smeared-source audit, consumed here — so the entire correction is a contact term at the nucleus:

\Delta E = \frac16\langle(\delta r)^2\rangle\cdot4\pi\frac{e^2}{4\pi\epsilon_0}\,|\psi(0)|^2

Step 3. Who pays. $|\psi_{ns}(0)|^2=1/(\pi n^3a^3)$ from Hydrogenic Wavefunctions and Quantum Numbers — for $2s$ , $1/(8\pi a^3)$ — while every $\ell\ge1$ state vanishes as $r^\ell$ at the origin (Radial Probability Distribution and Expectation Values' small- $r$ law). The shift is positive: the smear dilutes the potential's most attractive point, so the state that lived there is less bound. $2s_{1/2}$ rises; $2p_{1/2}$ stays.

Step 4. Assemble and count powers. Substituting $\langle(\delta r)^2\rangle$ with the log written as $L$ :

\Delta E_{ns} = \frac{4\alpha}{3\pi}\left(\frac{\hbar}{mc}\right)^2\frac{e^2}{4\pi\epsilon_0}\frac{L}{\pi n^3a^3}\;\sim\;\frac{8}{3\pi}\,\frac{\alpha^5}{n^3}\,mc^2\,L\times\mathcal{O}(1)

Power counting: $(\hbar/mc)^2\propto$ two powers of $\alpha$ against $a^2$ , the Coulomb factor and $1/a^3$ supply the rest — five powers of $\alpha$ total, one more than fine structure. Numerically (verifier, both honest cutoff choices): $L=\ln(1/\alpha^2)$ gives $1335$ MHz; $L=\ln(mc^2/16.64\,\mathrm{Ry})$ — borrowing Bethe's average excitation energy for the low cutoff — gives $1047$ MHz. The measured $1057.8$ MHz sits between them. Verified: the mode-response formula dimensionally and numerically; the smearing evaluation end-to-end in SI-Gaussian bookkeeping; both log choices computed; sign and 1/n³ scaling confirmed (article verifier).

What the heuristic cannot fix is the logarithm's inside or the additive constants — that took renormalization. Bethe's 1947 move, stated in structure only (the calculation and its story live at the electromagnetic track's radiative-shift article): compute the electron's second-order self-energy from emitting and reabsorbing virtual photons; recognize the divergent piece as the same divergence a free electron has, already hiding inside the measured mass; subtract it, and a finite remainder survives, proportional to $|\psi(0)|^2$ with a definite average excitation energy inside the log:

\Delta E_{2s} = \frac{8}{3\pi}\,\frac{\alpha^3}{n^3}\,\mathrm{Ry}\,\ln\!\frac{mc^2}{\langle E\rangle_{\mathrm{av}}}\Big|_{\langle E\rangle_{\mathrm{av}}=16.64\,\mathrm{Ry}} \approx 1040\,\mathrm{MHz} \qquad (3)

where: the verifier re-runs this arithmetic and lands at $1047$ MHz — one percent from the measured value, using nothing but the structure above. Two smaller QED tributaries complete the modern bookkeeping, quoted with owners: vacuum polarization (the Uehling term — the photon briefly a virtual $e^+e^-$ pair screening the nucleus) contributes about $-27$ MHz, opposite in sign; and the electron's anomalous magnetic moment $a_e=\alpha/2\pi$ slightly rescales the spin-orbit physics of the $2p$ states. Their derivations belong to the quantum-field-theory register beyond this track's relativistic serials.

The measurement: leverage, not brute force

Why was this hard? Optical attacks on the $2s$ – $2p$ interval face GHz-scale Doppler broadening — the splitting drowns. The 1947 genius was to stop looking at light and drive the transition with microwaves, using two levers the atom itself provides.

Figure

The quench-dip method: a beam that reports its own excitation

\tau_{2p}=1.60\,\mathrm{ns} \qquad\text{vs}\qquad \tau_{2s}\approx0.12\,\mathrm{s} \qquad (4)

where: the $2p$ state radiates to $1s$ almost instantly (Einstein coefficient $6.26\times10^8\,\mathrm{s}^{-1}$ ), while $2s\to1s$ by one photon is forbidden by the angular-momentum selection rules (their derivation lives in the time-dependent serials) — the $2s$ atom must emit two photons, and waits an eighth of a second. That ratio, $7.6\times10^7$ (computed), is the experiment's signal-to-background: a beam can be flown, sorted, and interrogated entirely on the question "is the atom still excited?"

Go deeper · Derivation B · 2 minAnatomy of the 1947 experiment: metastable beam, quench detector, resonance ridden to zero field

Derivation BAnatomy of the 1947 experiment: metastable beam, quench detector, resonance ridden to zero field

We reconstruct the measurement's logic block by block, with the numbers that make each block work. The goal is to see how a $10^{-6}$ -relative-energy splitting was measured to a few percent in 1947 (and to a part in $10^4$ by 1953) using war-surplus microwave hardware — and why the method needed the Zeeman Effect's physics as a tuning knob rather than an enemy.

Step 1. Make metastables. A tungsten oven dissociates H $_2$ at $\sim2500$ K; the thermal beam moves at $v\approx8$ km/s. Cross-bombardment by electrons ( $\sim10.8$ eV, just above the $n=2$ threshold) excites a small fraction to $n=2$ . Over the $\sim10$ cm flight to the detector ( $13$ μs, computed), every $2p$ atom decays within its first millimeters ( $\sim8000$ lifetimes elapse — computed), while $2s$ atoms lose only $10^{-4}$ of their population. The beam that arrives excited is purely metastable $2s$ .

Step 2. Detect excitation, not light. A $2s$ atom carries $10.2$ eV of internal energy — more than the work function of a tungsten surface — so it ejects an electron on impact; ground-state atoms do not. The detector current is the metastable count. This inverts the problem: instead of resolving a faint emission line, count atoms in a binary state.

Step 3. Drive and quench. In the interrogation region the beam crosses a microwave field. On resonance ( $2s_{1/2}\to2p_{1/2}$ at $\sim1$ GHz, or $2s_{1/2}\to2p_{3/2}$ near $10$ GHz), atoms transfer to $2p$ — and are gone in nanoseconds, radiating to the ground state before the detector. Resonance appears as a dip in detector current: the beam is quenched. The $28$ cm wavelength (computed) is exactly the radar band the postwar lab had mastered.

Step 4. Tune with the Zeeman fan, extrapolate to zero. The 1947 microwave frequency was fixed by available oscillators; the levels were tuned instead, by sweeping a magnetic field — the level fans of the Zeeman Effect serial, used here as an instrument. Plotting resonant field against frequency for several oscillator settings and extrapolating to $B=0$ gave the field-free interval: roughly $1000$ MHz in the first report, $1057.77\pm0.10$ MHz by the 1953 series. The Zeeman effect — historically spectroscopy's problem with magnets — became the calibration dial. Why extrapolation beats shielding: at the few-MHz precision sought, no 1947 magnetic shield could guarantee a clean zero field along the whole interrogation path — but a deliberately applied, measured field with a theoretically known level slope turns the contaminant into the independent variable. The method also split apart the hyperfine components (nuclear-spin structure, owner two serials ahead), which would otherwise smear the resonance. *Verified: flight-time/lifetime contrasts, quench arithmetic, the 28 cm wavelength, and the metastability ratio 7.6×10⁷ (article verifier); the resonance-vs-field slopes are the Zeeman serial's physics, cited not recomputed.*

Limiting cases

case	result	meaning
$\ell\ge1$ states	leading shift $\approx0$	no amplitude at the origin — contact physics passes them by; only small non-contact residues
$n$ -dependence	$\Delta E\propto\|\psi_{ns}(0)\|^2\propto1/n^3$	$1s$ : $8173$ MHz vs $2s$ : $1058$ MHz — ratio $7.7$ , the log softening the naive $8$ (computed)
power of $\alpha$	$\alpha^5mc^2\ln(1/\alpha)$	one photon loop costlier than fine structure's $\alpha^4$ : a tenth of the $2p$ interval
$Z$ scaling	$\sim Z^4\alpha^5mc^2\ln(1/Z\alpha)/n^3$	hydrogen-like uranium: shifts of tens-of-eV scale, and the $Z\alpha$ expansion itself dies
heavier lepton	contact factor $\times(\mu/m_e)^3\approx6.4\times10^6$ for muonic H	the shift graduates from QED test to proton-size ruler (computed reduced mass $186\,m_e$ )
classical field limit	no zero-point fluctuations → no shift	the Lamb shift is direct spectroscopic evidence that the field is quantized

Common trapThe Lamb shift is not a fine-structure correction you forgot

Three standing confusions. First: no improvement within the Dirac equation — better bases, higher orders of the $(v/c)$ expansion, exact solutions — can produce this splitting; equation (1) is $\ell$ -blind to all orders, so any "derivation" of the Lamb shift from single-particle relativistic corrections is wrong by construction. Second: the bare self-energy is divergent; quoting the loop integral without the free-electron subtraction (mass renormalization) yields infinity, not $1058$ MHz — the finite answer is a difference between bound and free divergences. Third: the sign. Students expect "vacuum attraction" to bind the electron more; in fact the smearing dilutes the Coulomb singularity, so $s$ -states are less bound — $2s_{1/2}$ moves up. If your heuristic gives the wrong sign, you averaged the potential incorrectly.

Worked examples

Example 1 — the Welton estimate, end-to-end with honest cutoffs. Compute the $2s$ shift from the smearing formula and compare against $1057.8$ MHz.

Step 1. The smear. $\langle(\delta r)^2\rangle=\tfrac{2\alpha}{\pi}(\hbar/mc)^2L$ with $\hbar/mc=386$ fm. Taking $L=\ln(1/\alpha^2)=9.84$ : rms jitter $\approx82$ fm (computed) — $1.5\times10^{-3}$ of a Bohr radius; tiny, but the Coulomb curvature at the origin is infinite, so tiny is enough.

Step 2. The contact evaluation. $\Delta E=\tfrac16\langle(\delta r)^2\rangle\cdot4\pi(e^2/4\pi\epsilon_0)|\psi_{2s}(0)|^2$ with $|\psi_{2s}(0)|^2=1/(8\pi a^3)$ . Running the numbers (verifier): $\Delta E=1335$ MHz.

Step 3. The cutoff sensitivity, stated rather than hidden. Swap the crude $L=\ln(1/\alpha^2)$ for the sharper $\ln(mc^2/16.64\,\mathrm{Ry})$ — Bethe's average excitation energy marking where the electron stops following the kick — and the same arithmetic gives $1047$ MHz. The spread $1047$ – $1335$ MHz is the heuristic's error bar; the measured $1057.8$ sits inside it. Answer: $\sim1000$ – $1300$ MHz, right sign, right $n^{-3}$ , right order — mechanism confirmed; the coefficient left to renormalized QED.

Example 2 — the n=2 neighborhood, audited in MHz. Place every small splitting near $n=2$ on one ruler and check the hierarchy of Fine Structure of Hydrogen's pyramid.

Step 1. Fine structure: $2p_{3/2}-2p_{1/2}=4.53\times10^{-5}$ eV $=10\,949$ MHz at leading order (quoted from the assembly serial, re-computed in the verifier; measured $10\,969$ ).

Step 2. Lamb: $2s_{1/2}-2p_{1/2}=1057.8$ MHz — a factor $10.4$ below fine structure (computed), numerically consistent with one extra factor of $\alpha\ln(1/\alpha)$ times an $\mathcal{O}(1)$ coefficient (checked: the ratio over $\alpha\ln(1/\alpha)$ is $2.7$ ).

Step 3. Hyperfine, for calibration: the $1s$ splitting is $1420.4$ MHz (the 21-cm line — owner serials two ahead). Different physics — $(m_e/m_p)\alpha^4$ vs $\alpha^5\ln\alpha^{-1}$ — happens to land within a factor $1.3$ of the Lamb shift at these quantum numbers. Ordering audit (computed): $10\,949 > 1420 > 1058$ MHz. Answer: fine : hyperfine(1s) : Lamb $\approx$ 10.4 : 1.34 : 1 — three rungs, three mechanisms, one atom.

Origins · 1 minWillis Eugene Lamb Jr. and how this was discovered

Willis Eugene Lamb Jr.

1913–2008 · Columbia, Stanford, Oxford, Tucson

The 1947 discovery story is told at the radiative-shift article of the electromagnetism track; what belongs here is the decade of craft that turned a 10% anomaly into a part-in- $10^4$ ruler. Lamb's instrument was built on the atom's own asymmetries: the $2s$ state's quarter-second stubbornness against the $2p$ 's nanosecond death, the 10.2 eV of stored energy that let a metastable atom announce itself by kicking an electron off a tungsten plate, and the Zeeman fan that let a fixed-frequency oscillator walk the resonance instead of the reverse. With Retherford he assembled these into the quench-dip method, then spent six years sharpening it: the Lamb–Retherford papers of 1950–1953 chase systematics one by one — stray fields, hyperfine overlaps, line-shape asymmetries, the extrapolation to zero field — down to $1057.77\pm0.10$ MHz. Lamb the theorist audited Lamb the experimentalist throughout, publishing his own bound-state calculations of the very number his beam produced. The style he built — drive a forbidden interval directly, detect state survival, extrapolate the perturbing field away — became the template for precision atomic physics, running today in the muonic-hydrogen experiments that turned his shift into a caliper for the proton itself.

History

1928

— Dirac's equation: exact Coulomb spectrum in n and j; 2s and 2p at j=1/2 locked degenerate

1938

— Pasternack infers from optical data that 2s sits ~0.03 cm⁻¹ high — an anomaly without a mechanism

1940

— Breit and Teller establish 2s metastability: two-photon decay, lifetime ~1/8 s

1947

— Lamb and Retherford drive 2s→2p with microwaves: ~1000 MHz where theory said zero

1950

— Lamb–Retherford I–IV begin the systematics campaign toward 1057.77 ± 0.10 MHz

2010

— CREMA measures the muonic-hydrogen Lamb shift; the proton radius puzzle opens

2019

— electron- and muon-based radii converge; the Lamb shift stands as the proton's ruler

Going further

Griffiths & Schroeter — the afterword's Lamb-shift discussion at the course register.
Bethe, Phys. Rev. 72, 339 (1947) — the two-page renormalization estimate; still readable in one sitting.
Lamb & Retherford, Phys. Rev. 72, 241 (1947) and the 1950–53 series — the measurement papers.
Welton, Phys. Rev. 74, 1157 (1948) — the vacuum-jitter heuristic run here.
Cohen-Tannoudji, Diu & Laloë — the radiative-corrections complement for the systematic treatment.
Bjorken & Drell, Relativistic Quantum Mechanics — the field-theory register this article points toward.

Provenance note: Dirac level formula, $|\psi_{nS}(0)|^2=1/(\pi n^3 a_0^3)$ , and contact-term structure are standard (Griffiths, Sakurai, Shankar). The renormalized self-energy and Bethe logarithm follow Bethe, Phys. Rev. 72, 339 (1947); Welton's heuristic from Phys. Rev. 74, 1157 (1948). Numbers — $2S_{1/2}$ – $2P_{1/2}\approx 1058\,\mathrm{MHz}$ , $1S$ shift $\approx 8173\,\mathrm{MHz}$ , Bethe estimate $\approx 1040\,\mathrm{MHz}$ , fine-structure interval $\approx 10969\,\mathrm{MHz}$ , $a_e=\alpha/2\pi\approx 0.00116$ — are standard verifiable values. A verifier should recompute the second-order self-energy sum and confirm the $\alpha^5 mc^2$ scaling and sign (S above P), and check the measured splitting against Lamb–Retherford and modern CODATA values. — The scaling, sign, both cutoff estimates, the Bethe-structure arithmetic ( $1047$ MHz), and every experimental number are executed in this article's verify script. Edition note (base heal): this article's AllQFT export was one of the four genuinely corrupt double-escape bases of the QM track (390 `\\`-before-letter runs, every LaTeX command doubled; served math was broken from Step-1 until the pilot's display-heal restage). For this rebuild the raw base was healed (`\\`→`\` before LaTeX tokens; audit: 390 runs → 0, no legitimate matrix row-separators in this base) and every algebraic claim carried over was independently re-verified rather than trusted. The Bethe self-energy derivation in the base is retained here only as quoted structure (equation 3); its full form belongs to the QED register, and the 1947 discovery narrative is owned by the electromagnetism track's radiative-shift article.

Is the vacuum really "fluctuating," or is that just a cartoon for the maths?

The operational content is real: the quantized field's ground state has nonzero $\langle E^2\rangle$ though $\langle E\rangle=0$ , and anything coupled to the field inherits that variance — the Lamb shift, spontaneous emission, and the Casimir effect are its measured consequences. The cartoon part is picturing classical fields flickering in time; the fluctuations are quantum variances, not histories. Welton's model treats them as a classical drive because, for this observable, the variance is all that enters — which is why a cartoon gets within 30% of the answer.

Why doesn't the Darwin term already account for this? Both are contact terms from smearing.

Same grammar, different source, different order. The Darwin smear is Zitterbewegung — structure inside the Dirac equation, amplitude fixed at $\hbar/mc$ , no logarithm — and it is already spent in the fine-structure assembly, where it cannot split $2s_{1/2}$ from $2p_{1/2}$ (the master formula stays $j$ -only). The Lamb smear comes from the field's zero-point motion, physics the Dirac equation doesn't contain, smaller by $\alpha\ln(1/\alpha)$ and carrying the cutoff logarithm. The kinship is why the evaluation looks familiar; the extra $\alpha$ is why it took twenty more years to see.

If the self-energy calculation diverges, why is the Lamb shift finite and predictable?

Because the divergence is not a property of the hydrogen atom — a free electron's self-energy diverges identically, and that infinity is already inside the measured mass. Subtracting the free-electron expression (mass renormalization) leaves the difference — how being bound changes the self-interaction — which is finite, with the atomic physics entering through $|\psi(0)|^2$ and the excitation spectrum in the logarithm. The Lamb shift was the first number to survive this surgery and match experiment: 1947 as QED's quantitative birthday.

Why is 2s metastable in the first place — what exactly forbids 2s→1s?

One-photon electric-dipole decay requires Δℓ=±1 (the selection rules derived in the time-dependent serials); $s\to s$ has Δℓ=0 and is dipole-forbidden, and higher multipoles vanish or are suppressed to invisibility between these particular states. The escape route is two-photon emission — a second-order process, suppressed by additional powers of $\alpha$ and phase-space factors — giving $\tau\approx0.12$ s. That eight-order-of-magnitude gap against the 2p lifetime is not a nuisance; it is the entire experimental design.

Does the 2p₁/₂ state shift at all?

Slightly: it has no contact term ( $|\psi(0)|^2=0$ ), but the anomalous moment ( $a_e=\alpha/2\pi$ ) rescales its spin-orbit energy, vacuum polarization reaches it weakly, and small non-contact self-energy residues remain — tens of MHz against the 2s state's $\sim1$ GHz. The measured 1057.8 MHz is the difference, dominated by s-state physics, so "2s moves, 2p barely does" is honest to a few percent.

The hyperfine 21-cm line is 1420 MHz and the Lamb shift is 1058 MHz — coincidence?

Complete coincidence, and a useful warning. Hyperfine scales as $(m_e/m_p)\alpha^4mc^2$ (nuclear-moment physics), the Lamb shift as $\alpha^5mc^2\ln(1/\alpha)$ (radiative physics) — different mechanisms, different quantum numbers ( $F$ vs $j$ ), different states (21 cm lives in $1s$ ). Hydrogen's constants just happen to land them within 35%. Confusing them is a classic exam disaster: one is the vacuum shifting a level, the other a spin-flip within a level.

What would the world look like if the field were classical — no zero-point fluctuations?

Then equation (1) would be exact for point charges: no Lamb shift, no spontaneous emission, no Casimir force. The 1057.8 MHz is one of the cleanest single numbers separating "quantum matter in a classical field" from full QED — which is why this article sits at the boundary between the track's approximation methods and its relativistic finale.