Relativistic Kinetic Energy Correction

The hydrogen spectrum of the previous chapter is a triumph built on a borrowed part: the kinetic energy $p^2/2m$ is Newton's, not Einstein's. The electron does not care about the convenience. In the ground state it moves at $v/c\approx\alpha\approx1/137$ — slow enough that the Bohr energies are excellent, fast enough that the fourth decimal place is wrong. And spectroscopy reads fourth decimal places for breakfast: interferometers were resolving hydrogen's line splittings before anyone could explain them. This article installs the correct kinetic energy, $T=\sqrt{p^2c^2+m^2c^4}-mc^2$ , expands it in powers of $(p/mc)^2$ , and keeps the first new term: $\hat{H}'_{\mathrm{rel}}=-\hat{p}^4/8m^3c^2$ . That operator is the simplest of the three corrections that together build hydrogen's fine structure, and it teaches the cleanest lessons. First lesson: when naive perturbation theory is legal in a degenerate spectrum — $\hat{p}^4$ is a rotational scalar, so the good-states theorem of the previous serial certifies $(\ell,m_\ell)$ as good labels and no degenerate diagonalization is needed. Second lesson: the operator-square trick, $\hat{p}^2=2m(\hat{H}_0-V)$ , which converts a hopeless-looking $\langle\hat{p}^4\rangle$ into two radial moments that the hydrogen chapter already computed. The result, $E^{(1)}_{\mathrm{rel}}=-(E_n^2/2mc^2)\left[4n/(\ell+\tfrac12)-3\right]$ , lowers every level, hits penetrating s-states hardest, and — quietly, before spin has even been mentioned — begins dismantling hydrogen's accidental $\ell$ -degeneracy: the kinetic correction alone drops $2s$ below $2p$ by $29$ GHz. Every number and every symbolic identity below is machine-verified, including the master formula checked against explicit wavefunction integrals for all $(n,\ell)$ with $n\le3$ .

Key takeaways

The exact kinetic energy expands as $T=\dfrac{p^2}{2m}-\dfrac{p^4}{8m^3c^2}+\dfrac{p^6}{16m^5c^4}-\cdots$ (sympy-verified); the second term is the leading relativistic perturbation $\hat{H}'_{\mathrm{rel}}$ .
The expansion parameter is $(v/c)^2\sim(Z\alpha)^2$ : for hydrogen $\alpha^2=5.33\times10^{-5}$ — fine structure is fourth-decimal-place physics (The Bohr Radius and Characteristic Atomic Scales supplies $v_1/c=\alpha$ ).
$\hat{p}^4$ is a rotational scalar commuting with $\hat{L}^2,\hat{L}_z$ — by the good-states theorem of Good Quantum Numbers and Symmetry-Adapted Bases, the $|n\,\ell\,m_\ell\rangle$ basis is already good and plain first-order theory applies despite the $n^2$ -fold degeneracy.
The operator-square trick $\hat{p}^2=2m(\hat{H}_0-V)$ gives $E^{(1)}_{\mathrm{rel}}=-\frac{1}{2mc^2}[E_n^2-2E_n\langle V\rangle+\langle V^2\rangle]$ ; the moments $\langle1/r\rangle=1/n^2a$ and $\langle1/r^2\rangle=1/[n^3(\ell+\tfrac12)a^2]$ are consumed from Radial Probability Distribution and Expectation Values.
Result: $E^{(1)}_{\mathrm{rel}}=-\dfrac{E_n^2}{2mc^2}\left[\dfrac{4n}{\ell+\frac12}-3\right]$ — negative for every state, largest at low $\ell$ (verified against explicit integrals, $n\le3$ , all $\ell$ ).
Computed shifts: $1s$ : $-9.06\times10^{-4}$ eV; $2s$ : $-1.47\times10^{-4}$ eV; $2p$ : $-2.64\times10^{-5}$ eV. Kinetic energy alone already splits $2s$ below $2p$ by $1.21\times10^{-4}$ eV $=29.2$ GHz — the accidental degeneracy of Accidental Degeneracy and the SO(4) Symmetry does not survive relativity.
Key idea: replace $p^2/2m$ by the relativistic kinetic energy and keep $-\hat{p}^4/8m^3c^2$ ; a scalar perturbation needs no degenerate machinery, and the operator-square trick reduces its shift to two known radial moments.

The core insight

Replace $p^2/2m$ with the relativistic kinetic energy and keep the first correction $-\hat{p}^4/8m^3c^2$ ; since this commutes with $\hat{\mathbf{L}}$ , its first-order shift is an expectation value computed via $\hat{p}^2=2m(\hat{H}_0-V)$ , giving $-\frac{E_n^2}{2mc^2}\big[\frac{4n}{\ell+1/2}-3\big]$ .

Where this lives today

The $(Z\alpha)$ expansion is working infrastructure. Astrophysicists identify highly charged iron in the solar corona and in X-ray binaries by lines whose positions are dominated by exactly the corrections of this chapter, scaled by $Z^4$ ; laboratory ion traps (EBITs) now measure hydrogenic and lithium-like uranium, where the "correction" is a third of the binding energy and tests bound-state QED rather than perturbation theory. At the precision frontier the traffic runs the other way: optical atomic clocks compare transitions with different sensitivities to $\alpha$ , turning this chapter's $\alpha^2$ -scaling into a drift detector — current bounds hold any fractional change of $\alpha$ below about $10^{-18}$ per year. And every quantum-chemistry package carries a descendant of $-\hat{p}^4/8m^3c^2$ (in scalar-relativistic form) because gold's color and mercury's liquidity are, at bottom, this article's physics applied at $Z\approx80$ .

The physical picture

Plot the two kinetic energies against momentum. Newton's is a parabola, climbing without conscience. Einstein's, $\sqrt{p^2c^2+m^2c^4}-mc^2$ , hugs the parabola at small $p$ and then bends below it — at large momentum it grows only linearly, $pc$ . A relativistic particle always has less kinetic energy than the parabola claims for the same momentum. So wherever a bound electron moves fastest, the borrowed Newtonian bookkeeping has been overcharging it — and correcting the books lowers the energy. Every level drops.

Figure

The bent curve: relativistic kinetic energy falls below the parabola, and the deficit is the perturbation

Who drops most? The correction is $-\hat{p}^4/8m^3c^2$ : the fourth power weights the high-momentum tail of a state far more aggressively than $\langle p^2\rangle$ does. High momentum lives at small $r$ , where the Coulomb well is deep and the electron sprints. The states that penetrate — low $\ell$ , above all s-states, whose radial density at small $r$ behaves as $r^{2\ell+2}$ per Radial Probability Distribution and Expectation Values — collect the largest correction. Circular states ( $\ell=n-1$ ) idle far from the nucleus and barely notice. An $\ell$ -dependent shift inside a level that was $\ell$ -degenerate: the correction does not merely move the ladder, it splits the rungs. That splitting has a pedigree. The $n^2$ -fold coincidence of hydrogen's levels was diagnosed in Accidental Degeneracy and the SO(4) Symmetry as the fingerprint of a hidden symmetry of the pure $1/r$ problem. The relativistic correction is not a small deformation of that symmetry — it is symmetry-breaking with a measured budget: $2s$ and $2p$ , exactly degenerate all through the hydrogen chapter, part company by $29$ GHz from the kinetic term alone. (Rotational symmetry survives untouched: the $2\ell+1$ states of each $\ell$ stay locked together, per Degeneracy and the Role of the Magnetic Quantum Number.)

The expansion parameter

Before any operator algebra, price the effect. The natural speed scale of hydrogen was computed in The Bohr Radius and Characteristic Atomic Scales: the ground-state electron moves at $v_1/c=\alpha$ , and for a hydrogenic ion of charge $Z$ (the $Z$ -ladder of Hydrogenic Ions and Scaling with Nuclear Charge) at $Z\alpha$ . The kinetic correction is second order in this ratio:

\frac{E^{(1)}_{\mathrm{rel}}}{E_n}\sim\left(\frac{v}{c}\right)^2\sim\frac{(Z\alpha)^2}{n^2},\qquad \alpha=\frac{e^2}{4\pi\varepsilon_0\hbar c}=\frac{1}{137.036} \qquad (1)

where: $\alpha$ is the fine-structure constant — dimensionless, the pure number that measures how relativistic an atom is allowed to be (its story is this article's portrait card). For hydrogen, $\alpha^2=5.33\times10^{-5}$ : corrections in the tens of microelectronvolts on eV-scale levels. Small — but this is the "fine structure" scale that gives the constant its name, and interferometric spectroscopy resolves it easily. The same formula warns where perturbation theory will die: at $Z\alpha\sim1$ the expansion parameter reaches unity, and no amount of resummation saves a series built on $v\ll c$ .

Derivation AFrom the exact energy to the p-to-the-fourth perturbation

We expand the exact relativistic kinetic energy in powers of $(p/mc)^2$ and identify the new Hamiltonian: the familiar Coulomb problem plus a small scalar operator. The only subtlety worth watching is bookkeeping — which term is $\hat{H}_0$ , which is the perturbation, and how big the discarded remainder is. The expansion itself is machine-verified.

Step 1. Define the kinetic energy from the relativistic energy-momentum relation: $T=\sqrt{p^2c^2+m^2c^4}-mc^2$ , the total energy minus the rest energy. Factor the rest energy out of the radical: $T=mc^2\left[\sqrt{1+x}-1\right]$ with $x=(p/mc)^2$ .

Step 2. Expand with the binomial series $\sqrt{1+x}=1+\tfrac12x-\tfrac18x^2+\tfrac1{16}x^3-\cdots$ , legitimate because $x=(v/c)^2\sim(Z\alpha)^2\ll1$ for the atoms in this chapter:

T=\frac{p^2}{2m}-\frac{p^4}{8m^3c^2}+\frac{p^6}{16m^5c^4}-\cdots \qquad (2)

Why the promotion to operators is safe here: each term is a power of $\hat{p}^2$ alone, and all powers of one operator commute with each other — no ordering ambiguity arises in the kinetic series itself. (Ordering questions do arise for the spin and contact terms of the next two serials, which is one reason they are subtler.)

Step 3. Identify the pieces. The Hamiltonian becomes

\hat{H}=\underbrace{\frac{\hat{p}^2}{2m}+V(r)}_{\hat{H}_0}\;\underbrace{-\;\frac{\hat{p}^4}{8m^3c^2}}_{\hat{H}'_{\mathrm{rel}}}+\mathcal{O}\!\left(\frac{p^6}{m^5c^4}\right) \qquad (3)

$\hat{H}_0$ is the solved Coulomb problem of The Hydrogen Atom: Coulomb Radial Equation; $\hat{H}'_{\mathrm{rel}}$ is the perturbation. The discarded $p^6$ term is smaller by another factor $(Z\alpha)^2$ — order $\alpha^6mc^2$ , below even the Lamb shift's $\alpha^5$ scale, safely ignorable at this order. Verified: series coefficients $1/2m$ , $-1/8m^3c^2$ , $+1/16m^5c^4$ extracted symbolically from the exact square root (article verifier, section 1).

Why plain perturbation theory is legal

Hydrogen's level $n$ is $n^2$ -fold degenerate, and Degenerate Perturbation Theory just spent an article warning that first-order formulas fail inside a degenerate manifold unless the basis is right. Here is the first payoff of the good-states theorem of Good Quantum Numbers and Symmetry-Adapted Bases: $\hat{p}^4=(\hat{p}^2)^2$ is built from $\hat{p}^2$ , a rotational scalar, so

[\hat{H}'_{\mathrm{rel}},\hat{L}^2]=0,\qquad[\hat{H}'_{\mathrm{rel}},\hat{L}_z]=0 \qquad (4)

where: $\hat{L}^2$ and $\hat{L}_z$ also commute with $\hat{H}_0$ and distinguish the degenerate states of level $n$ by their $(\ell,m_\ell)$ eigenvalues. The theorem then guarantees every off-diagonal element of $\hat{H}'_{\mathrm{rel}}$ between distinct $(\ell,m_\ell)$ pairs vanishes: the W-matrix is already diagonal, the $|n\,\ell\,m_\ell\rangle$ states are good, and the shift is the plain expectation value $\langle n\ell m_\ell|\hat{H}'_{\mathrm{rel}}|n\ell m_\ell\rangle$ — no diagonalization, no drama. Rotational invariance also forces the answer to be independent of $m_\ell$ (the axis-choice doctrine of Degeneracy and the Role of the Magnetic Quantum Number), so the shift can depend on $n$ and $\ell$ only.

Go deeper · Derivation B · 1 minThe operator-square trick and the master formula

Derivation BThe operator-square trick and the master formula

We evaluate $E^{(1)}_{\mathrm{rel}}=-\langle\hat{p}^4\rangle/8m^3c^2$ without ever touching a momentum-space integral. The strategy is the chapter's most reusable trick: on an eigenstate of $\hat{H}_0$ , the operator $\hat{p}^2$ can be traded for $2m(\hat{H}_0-V)$ , and two such trades convert $\langle\hat{p}^4\rangle$ into radial moments that Radial Probability Distribution and Expectation Values has already computed and verified. One honesty stop about hermiticity is owed along the way.

Step 1. Split the fourth power symmetrically: $\langle\psi|\hat{p}^4|\psi\rangle=\langle\hat{p}^2\psi|\hat{p}^2\psi\rangle$ , using hermiticity of $\hat{p}^2$ to send one factor leftward. Each factor now acts directly on an eigenstate: $\hat{p}^2|\psi_{n\ell m}\rangle=2m(E_n-V)|\psi_{n\ell m}\rangle$ . Why the honesty stop: for s-states, whose wavefunctions have a cusp at the origin, $\hat{p}^4$ alone is a delicate operator — naive double integration by parts generates boundary terms at $r=0$ . The symmetric split above never differentiates more than twice against any wavefunction, so it is safe for every $\ell$ including zero; that is why this derivation writes $\langle\hat{p}^2\psi|\hat{p}^2\psi\rangle$ and not $\langle\psi|\hat{p}^4\psi\rangle$ . (The same small- $r$ delicacy, taken seriously as physics rather than bookkeeping, becomes the Darwin term — owner named at the end.)

Step 2. Expand the square of numbers, since $E_n-V(r)$ is now an ordinary function inside the integral:

E^{(1)}_{\mathrm{rel}}=-\frac{1}{2mc^2}\left[E_n^2-2E_n\langle V\rangle+\langle V^2\rangle\right] \qquad (5)

Step 3. Insert the Coulomb potential $V=-e^2/4\pi\varepsilon_0r$ and cash the two moments from Radial Probability Distribution and Expectation Values — consumed with their credentials, not re-derived: $\langle1/r\rangle=1/n^2a$ (the $\ell$ -blind virial echo; it gives $\langle V\rangle=2E_n$ ) and $\langle1/r^2\rangle=1/[n^3(\ell+\tfrac12)a^2]$ (the Hellmann-Feynman moment, where $\ell$ enters).

Step 4. Collapse. With $\langle V\rangle=2E_n$ and $\langle V^2\rangle=4E_n^2\,\frac{n}{\ell+1/2}$ (three lines of algebra from $E_n=-e^2/8\pi\varepsilon_0an^2$ ), the bracket becomes $E_n^2[1-4+4n/(\ell+\tfrac12)]$ :

E^{(1)}_{\mathrm{rel}}=-\frac{E_n^2}{2mc^2}\left[\frac{4n}{\ell+\tfrac12}-3\right] \qquad (6)

The bracket exceeds $1$ for every allowed $(n,\ell)$ , so the shift is strictly negative: every level drops, s-states ( $\ell=0$ , bracket $8n-3$ ) hardest, circular states ( $\ell=n-1$ , bracket $\to1$ for large $n$ ) least. Verified: the symbolic collapse of (5) to (6) using the two quoted moments; AND independently, $\langle(E_n-V)^2\rangle$ computed by explicit sympy integrals of the hydrogen radial functions for every $(n,\ell)$ with $n\le3$ — all six cases match (6) exactly (article verifier, sections 2-3).

What it does to the spectrum

Numbers, all computed. For the ground state, $E^{(1)}_{\mathrm{rel}}(1s)=-9.06\times10^{-4}$ eV — a $6.7\times10^{-5}$ relative shift, which is $\tfrac54\alpha^2$ : equation (1)'s promise kept to the coefficient. In the $n=2$ manifold the correction reads $-1.47\times10^{-4}$ eV for $2s$ and $-2.64\times10^{-5}$ eV for $2p$ :

E^{(1)}_{\mathrm{rel}}(2s)-E^{(1)}_{\mathrm{rel}}(2p)=-1.21\times10^{-4}\ \mathrm{eV}=-h\times29.2\ \mathrm{GHz} \qquad (7)

where: the sign says the kinetic correction pushes $2s$ below $2p$ — the penetrating state pays more. This single number is the death certificate of the accidental degeneracy: no spin, no magnetic field, no external agent, just honest kinematics splitting what SO(4) had locked together. It is not yet the observable splitting — the next two serials add comparable spin-orbit and contact pieces, and the sum reorganizes dramatically (the $j$ -only miracle owned by Fine Structure of Hydrogen) — but it sets the scale and the mechanism.

Figure

The n = 2 manifold under the kinetic correction alone: the accidental degeneracy breaks by 29 GHz

Limiting cases

case	result	meaning
$p\ll mc$	$T\to p^2/2m$ ; correction $\to0$	Newton recovered; the parabola is the small- $p$ tangent of the true curve
$\ell=0$	bracket $=8n-3$ : maximal shift	s-states penetrate to the deep, fast region near the nucleus
$\ell=n-1$ (circular)	bracket $=4n/(n-\tfrac12)-3\to1$	slowest, farthest states — minimal correction, but never zero
$n\to\infty$	shift $\propto1/n^3$ (for fixed $\ell$ : $E_n^2\cdot n\sim n^{-3}$ )	Rydberg states are nonrelativistic; the correction dies faster than the binding
$Z$ grows	shift/level $\sim\tfrac54(Z\alpha)^2$ : $6.7\times10^{-5}$ (H) $\to0.045$ (Fe $^{25+}$ ) $\to0.17$ ( $Z=50$ )	perturbation theory visibly failing; Dirac equation required (QM11's serials)
$m_\ell$ within fixed $\ell$	no dependence	scalar perturbation: the axis-choice doctrine holds, $2\ell+1$ states move together

Common trapNo degenerate machinery is needed here — and p⁴ is an operator, not a number

Two opposite mistakes cancel many students' confidence. First: seeing hydrogen's $n^2$ -fold degeneracy, they reach for the W-matrix of Degenerate Perturbation Theory — unnecessary, because $\hat{p}^4$ commutes with $\hat{L}^2$ and $\hat{L}_z$ , whose eigenvalues distinguish the degenerate states; the good-states theorem certifies $|n\,\ell\,m_\ell\rangle$ and the plain expectation value is the whole calculation. Second, the reverse sin: treating $\hat{p}^2$ as the number $2m(E_n-V(r))$ everywhere. The trade is licensed only against an eigenstate of $\hat{H}_0$ — one factor acting rightward on $|\psi\rangle$ , one leftward on $\langle\psi|$ via hermiticity. And for s-states specifically, avoid writing $\langle\psi|\hat{p}^4\psi\rangle$ with four derivatives stacked on a cusped wavefunction: use the symmetric split $\langle\hat{p}^2\psi|\hat{p}^2\psi\rangle$ , which is what the derivation above does and why its answer is trustworthy at $\ell=0$ .

Worked examples

Example 1 — the $n=2$ manifold, end to end. Compute every kinetic shift at $n=2$ and the induced splitting.

Step 1. Shared prefactor: $E_2=-3.4014$ eV, so $E_2^2/2mc^2=(3.4014)^2/(2\times510999)\ \mathrm{eV}=1.132\times10^{-5}$ eV.

Step 2. Brackets: $2s$ ( $\ell=0$ ): $4\cdot2/\tfrac12-3=13$ . $\,2p$ ( $\ell=1$ ): $4\cdot2/\tfrac32-3=\tfrac73=2.333$ .

Step 3. Shifts: $E^{(1)}(2s)=-1.132\times10^{-5}\times13=-1.47\times10^{-4}$ eV; $E^{(1)}(2p)=-1.132\times10^{-5}\times2.333=-2.64\times10^{-5}$ eV. Both negative; the s-state's is $5.6\times$ larger.

Step 4. The splitting: $\Delta=1.21\times10^{-4}$ eV. In frequency, $\Delta/h=29.2$ GHz — a radio-frequency-technology number, enormous by precision-spectroscopy standards. Answer: $2s$ drops $147\ \mu$ eV, $2p$ drops $26\ \mu$ eV; the kinetic term alone splits them by $29.2$ GHz, $2s$ below. (What experiment sees at $n=2$ is the reassembled structure once spin-orbit and Darwin join — Fine Structure of Hydrogen owns that sum, in which, remarkably, $2s_{1/2}$ rejoins $2p_{1/2}$ exactly.)

Example 2 — riding the $Z$ -ladder until the method breaks. For hydrogenic ions the velocity ratio is $Z\alpha$ (Hydrogenic Ions and Scaling with Nuclear Charge), and the relative shift of the ground state is $\tfrac54(Z\alpha)^2$ . Walk it upward.

Step 1. Hydrogen, $Z=1$ : $v/c=0.0073$ , relative shift $6.7\times10^{-5}$ . Textbook perturbation theory: the correction is real but tiny.

Step 2. Hydrogenic iron, $Z=26$ (solar-corona spectroscopy's workhorse): $v/c=0.19$ , relative shift $4.5\%$ . The "correction" is now larger than typical chemistry. First-order theory still lands within a percent or so of truth, but the discarded $p^6$ term — smaller by another $(Z\alpha)^2\approx3.6\%$ — is knocking.

Step 3. $Z=50$ : $v/c=0.37$ , relative shift $17\%$ . The expansion parameter $(Z\alpha)^2=0.13$ is no longer small; successive orders shrink slowly, and "fine structure" is not fine — it is restructuring the spectrum. $Z=82$ (hydrogenic lead): $v/c=0.60$ , nominal relative shift $45\%$ : the series is decorative at best.

Step 4. The wall: at $Z=137$ , $Z\alpha=1$ — the formal end of the road, where the electron's would-be speed reaches $c$ and the whole $(v/c)$ expansion is meaningless. The honest instrument for high $Z$ is the exact Dirac equation (the relativistic serials of QM11), for which this article's formula is the first term of the $(Z\alpha)^2$ expansion — Fine Structure of Hydrogen exhibits that expansion explicitly. Answer: the perturbative description degrades from $10^{-5}$ -level (H) through percent-level (Fe $^{25+}$ ) to unusable ( $Z\gtrsim50$ ), with a formal singularity at $Z\alpha=1$ . (One reduced-mass footnote, per the honesty rules of The Hydrogen Atom: Coulomb Radial Equation: for muonic atoms the same formulas apply with $m\to m_\mu$ -reduced mass, and the relativistic corrections scale accordingly — the muonic Lamb-shift story belongs to Lamb Shift.)

Origins · 1 minFine Structure Constant and how this was discovered

Fine Structure Constant

1916–2018 · Munich, Geneva, Paris

The number $1/137.036$ entered physics through exactly this article's subject. In 1916 Arnold Sommerfeld, working in Munich with Bohr's planetary model, let the electron's orbits be relativistic ellipses: the perihelion precessed, energies picked up corrections in powers of a new dimensionless combination $\alpha=e^2/4\pi\varepsilon_0\hbar c$ — the ratio of the ground orbit's speed to light — and the single Balmer lines split into the "fine structure" that gave the constant its name. His formula, derived with no spin and no wave mechanics, agreed with experiment; decades later it turned out to match the Dirac equation's expansion exactly, a coincidence-with-a-reason (his integer orbit label and the modern $j+\tfrac12$ play the same algebraic role) that physicists still shake their heads over. The constant itself outgrew its birthplace: it measures the strength of electromagnetism as a pure number no unit reform can touch — the 2018 SI overhaul fixed $e$ , $h$ , and $c$ by decree, and $\alpha$ alone stayed a measured fact of nature, currently $1/137.035999$ . Eddington tried to derive 137 from numerology and failed publicly; Feynman called it "a magic number that comes to us with no understanding"; precision spectroscopy — the direct descendant of the fine-structure splittings in this chapter — now reads it to eleven digits, and drifting-constant searches use atomic clocks to ask whether it is even constant. No one has yet computed it from anything deeper.

History

1887

— Michelson's interferometer resolves the Balmer lines into close doublets: fine structure measured before it is named

1905

— Einstein's special relativity fixes the kinetic energy this article expands

1916

— Sommerfeld's relativistic orbits introduce α and reproduce the splittings within the old quantum theory

1926

— Schrödinger tries a fully relativistic wave equation first, gets the fine structure wrong, and publishes the nonrelativistic equation instead

1928

— Dirac's equation delivers the exact spectrum; the p⁴ term re-emerges as its leading kinetic correction

2018

— the SI redefinition fixes e, h, c exactly; α remains the measured dial of electromagnetism

Going further

Griffiths & Schroeter, Introduction to Quantum Mechanics, 3rd ed., §7.3.1 — the source register for this derivation.
Sakurai & Napolitano, Modern Quantum Mechanics, 2nd ed., §5.3 — the correction inside the full fine-structure program.
Shankar, Principles of Quantum Mechanics, 2nd ed., Ch. 17 — with the hydrogen integrals done explicitly.
Bethe & Salpeter, Quantum Mechanics of One- and Two-Electron Atoms — the professional reference for all hydrogenic corrections.

Provenance note: Equations cross-checked against Griffiths Introduction to Quantum Mechanics 3rd ed. §7.3.1 (relativistic correction $-p^4/8m^3c^2$ and shift $E^{(1)}_{r}=-\frac{E_n^2}{2mc^2}[4n/(\ell+1/2)-3]$ ), Sakurai & Napolitano §5.3, and Bethe–Salpeter. A verifier should: (1) confirm the binomial expansion of $\sqrt{(pc)^2+(mc^2)^2}$ ; (2) recompute $\langle p^4\rangle$ via $p^2=2m(E_n-V)$ using $\langle1/r\rangle=1/(n^2a)$ and $\langle1/r^2\rangle=1/[n^3(\ell+1/2)a^2]$ ; (3) check the order $\alpha^2/n^2$ and the numerical estimate $\sim10^{-5}E_n$ for $n=2$ . The constant $\alpha\approx1/137.036$ and $E_n=-13.6\,\mathrm{eV}/n^2$ are standard textbook values. — All three verifier items are executed in this article's verify script, plus a check the base does not request: the master formula is confirmed against explicit sympy integrals of the hydrogen radial functions for every $(n,\ell)$ with $n\le3$ . The two radial moments are consumed by name from Radial Probability Distribution and Expectation Values (its Kramers–Pasternack machinery, never re-derived); the good-states legality argument consumes the theorem of Good Quantum Numbers and Symmetry-Adapted Bases. Edition note: the base's two derivations are retained as this article's two cards, with the hermiticity subtlety at $\ell=0$ made explicit (symmetric split) and the spectrum consequences ( $2s$ – $2p$ kinetic splitting $29.2$ GHz, $Z$ -ladder breakdown) added as computed exhibits. Spin-orbit, Darwin, and the $j$ -only assembly are owned by the next three serials and are only named here.

If the electron moves at less than one percent of c, why does anyone care about relativity here?

Because spectroscopy's resolving power outruns the smallness. The shift is parts in $10^5$ of eV-scale energies — tens of microelectronvolts, or tens of GHz — while interferometric and laser spectroscopy resolve kilohertz. Michelson could see hydrogen's fine structure in 1887; the theory chapters you have just read could not explain it until relativity (and then spin) was installed. "Small" in physics means small compared to what you can measure, and here the measurement wins by six orders of magnitude.

Is −p⁴/8m³c² hermitian? I have heard p⁴ is dangerous for s-states.

The operator is symmetric, and the danger is real but locatable: on wavefunctions with a cusp at the origin (hydrogen s-states), stacking four derivatives and integrating by parts twice can generate boundary terms at $r=0$ , making naive $\langle\psi|\hat{p}^4\psi\rangle$ manipulations untrustworthy. The cure used in the derivation is the symmetric split $\langle\hat{p}^2\psi|\hat{p}^2\psi\rangle$ : never more than two derivatives against any state, perfectly finite for all $\ell$ . The result agrees with the careful treatments (and with the Dirac expansion), which is the real certificate.

Why does the answer depend on ℓ but not on m?

Both facts are symmetry speaking. The perturbation is a rotational scalar, so it cannot tell orientations apart: all $2\ell+1$ states of fixed $\ell$ shift identically (the axis-choice doctrine — $m$ is a choice of quantization axis, not physics). But different $\ell$ have genuinely different radial habitats — penetration versus centrifugal exclusion — and $\hat{p}^4$ prices exactly that difference through $\langle1/r^2\rangle$ 's $1/(\ell+\tfrac12)$ factor. Rotational degeneracy is protected; the accidental $\ell$ -degeneracy, protected only by the Coulomb problem's hidden symmetry, is not.

Does this correction alone give the observed fine structure of hydrogen?

No — it is one of three comparable pieces. Spin-orbit coupling (next serial) contributes at the same $\alpha^4mc^2$ order for $\ell\ge1$ , and the Darwin term (the serial after) replaces spin-orbit's role at $\ell=0$ . Individually each piece depends on $\ell$ ; their sum, astonishingly, depends only on $n$ and $j$ — the $j$ -only collapse owned by Fine Structure of Hydrogen. In particular the $29$ GHz kinetic splitting of $2s$ / $2p$ computed here is not what experiment sees at $n=2$ : after the sum, $2s_{1/2}$ and $2p_{1/2}$ are exactly degenerate again (until the Lamb shift).

Why is the expansion in (Zα)² and not just α²?

Because the electron's speed scales with the nuclear charge. The Bohr-model velocity ratio is $v_n/c=Z\alpha/n$ : a stronger nucleus binds tighter orbits that move faster. Everything relativistic therefore enters through $Z\alpha$ — hydrogen's $\alpha^2\approx5\times10^{-5}$ becomes $4.5\%$ for Fe $^{25+}$ and order-one for the heaviest ions, where the perturbative chapter you are reading hands over to the exact Dirac treatment. The $Z$ -scaling of everything else (energies $Z^2$ , radii $1/Z$ ) is the subject of the hydrogenic-ions serial, consumed here.

Should I use the reduced mass in these formulas?

Yes, the same reduced-mass discipline as the whole hydrogen chapter: $m$ is really $\mu=m_em_N/(m_e+m_N)$ , a $5\times10^{-4}$ effect for hydrogen — comparable to the fine structure itself in precision work, which is why spectroscopic tables are careful about it. For exotic atoms the distinction is dramatic: muonic hydrogen's $\mu\approx186\,m_e$ rescales every energy in this article by that factor and every radius down by it, which is exactly what makes the muonic Lamb shift such a sensitive proton-size probe (that story belongs to Lamb Shift).

How do I remember which states feel the correction most?

Follow the momentum. The correction weights $\langle p^4\rangle$ — the high-momentum tail — and high momentum lives where the well is deep, at small $r$ . States that penetrate (low $\ell$ ; radial density $\sim r^{2\ell+2}$ near the origin) sample that region; circular states ( $\ell=n-1$ ) are excluded by the centrifugal barrier. Hence the bracket $4n/(\ell+\tfrac12)-3$ : maximal $8n-3$ for s-states, approaching $1$ for circular states. One slogan: relativity taxes penetration.