Gauge Transformations and Covariant Gauge Conditions

The Four-Potential ended with a confession: $A^\mu$ is not unique. Add the four-gradient of any scalar, $A^\mu\to A^\mu+\partial^\mu\chi$ , and the field tensor — the physics — does not move. That article stated the freedom and moved on; this one makes the freedom the subject, because how you manage it decides whether relativistic electrodynamics stays relativistic on paper. A gauge condition is a rule for picking one potential out of each equivalence class, and rules can be frame-proof or frame-tied. The Lorenz condition $\partial_\mu A^\mu=0$ is a Lorentz scalar: impose it in one frame and every boosted observer finds it still holding. The Coulomb condition $\nabla\cdot\mathbf{A}=0$ — the workhorse of atomic physics — is not: we will boost a potential that satisfies it and compute, with numbers, the amount by which the moving observer finds it violated. The second half of the article spends the freedom rather than merely taming it. Even after Lorenz, a residue survives — transformations whose gauge function solves the free wave equation — and spending that residue on a source-free field strips the four-potential down to two transverse components. That is the cleanest classical answer to a question quantum mechanics will make urgent: light has four potential components but only two polarizations, and gauge freedom is where the other two go. The counting, the covariance of the conditions, and the taxonomy of gauges assembled here are precisely the classical scaffolding on which quantum electrodynamics will be erected — the track's endgame, seeded now.

Key takeaways

The gauge orbit $A^\mu\to A^\mu+\partial^\mu\chi$ leaves $F^{\mu\nu}$ untouched; a gauge condition is a rule slicing one representative out of each orbit.
Conditions inherit — or fail to inherit — Lorentz covariance: $\partial_\mu A^\mu=0$ is a scalar equation, true in all frames if true in one; $\nabla\cdot\mathbf{A}=0$ is frame-tied, and a boost visibly breaks it (we compute the violation: $1.8\times10^{-8}$ V·s/m² for a 1 nC charge seen at $0.6c$ from 1 m).
In Lorenz gauge, Maxwell's equations collapse to $\Box A^\mu=\mu_0 J^\mu$ — the four wave equations The Four-Potential assembled; the 3+1 attainability and persistence proofs live in E9's Lorenz Gauge and are cited, not repeated.
Lorenz gauge is not a complete fixing: any $\chi$ with $\Box\chi=0$ preserves it. This residual freedom is a feature — spent on a free plane wave it removes the time component, leaving exactly two transverse polarizations.
Degrees-of-freedom ledger for light: $4$ potential components $-\,1$ (Lorenz) $-\,1$ (residual) $=2$ physical states — the covariant accounting behind E8's transversality.
Key idea: the four-potential is defined only up to $A^\mu\to A^\mu+\partial^\mu\chi$ ; the Lorenz condition is the covariant way to spend the freedom, and what survives it is exactly the two polarizations of light.

The core insight

The four-potential is defined only up to $A^\mu\to A^\mu+\partial^\mu\chi$ ; the Lorenz gauge $\partial_\mu A^\mu=0$ is the unique standard choice that is itself a Lorentz-scalar statement, and it turns Maxwell's equations into $\Box A^\mu=\mu_0 J^\mu$ while leaving a residual $\Box\chi=0$ freedom that pares the photon down to its two transverse states.

Where this lives today

Every quantum field theory calculation on Earth runs inside a covariant gauge from this article's table. The photon propagator that begins every QED computation is written in the ξ family — Feynman gauge for compactness, Landau gauge for renormalization work — and the residual-freedom bookkeeping of Derivation B becomes, after quantization, the Gupta–Bleuler condition selecting physical photon states, then the ghost fields of non-abelian gauge theory, then the BRST symmetry that modern textbooks wield as the master tool. Numerical relativity solves Einstein's equations in "harmonic gauge," the gravitational sibling of Lorenz's condition, chosen for exactly the reason Lorenz's was: it turns constraint-tangled equations into wave equations. And gravitational-wave astronomers strip their strain tensors to two transverse polarizations by precisely the spend-the-residue argument of Derivation B, one rank higher. The management of redundancy — this article's whole subject — turned out to be one of physics' most transferable skills.

Origins · 1 minVladimir Fock and how this was discovered

Vladimir Fock

1898–1974 · Leningrad

Fock is the physicist who made gauge freedom relativistic and quantum at once. In 1926, within months of Schrödinger's papers, he wrote the wave equation for a charged particle in an electromagnetic field and noticed the structure this article lives on: the equation keeps its form under $A^\mu\to A^\mu+\partial^\mu\chi$ only if the wavefunction simultaneously rotates its phase, $\psi\to e^{iq\chi/\hbar}\psi$ . He called it "gradient invariance" — the first statement that gauge transformations of the potential and phase transformations of matter are one symmetry, the seed of minimal coupling and ultimately of the Standard Model's architecture. Weyl's earlier "gauge" (a failed rescaling of lengths) supplied the name; Fock supplied the correct physics. His covariant instincts ran deep: the Fock space of QED, the Hartree–Fock method, and the harmonic (Lorenz-sibling) coordinates he championed in general relativity all bear his stamp. Soviet officialdom found his relativity too idealist; Fock defended Einstein's physics — while quarreling with Einstein's interpretation — with the same stubbornness he brought to index bookkeeping.

History

1867

— Ludvig Lorenz publishes retarded potentials with the condition now (mis)filed under Lorentz's name

1904

— Hendrik Lorentz systematizes the potential formulation of electrodynamics; the eponym confusion is seeded

1918

— Weyl's eichinvarianz: gauge as a fundamental principle, wrong theory, right idea

1926

— Fock's gradient invariance: gauge transformation of

A^\mu

= phase rotation of matter

1929

— Weyl reforges his principle in quantum form; minimal coupling becomes the law of light–matter interaction

1950

— Gupta and Bleuler quantize the photon inside the Lorenz slice, spending the residue on the state space

2001

— Jackson and Okun's forensic history restores Lorenz's name to Lorenz's condition

The physical picture

Picture the space of all four-potentials — every smooth assignment of $(V/c,\mathbf{A})$ to spacetime. This space is vastly bigger than the space of physical situations, because through every potential runs an orbit: the family $A^\mu+\partial^\mu\chi$ for all scalars $\chi$ , every member producing the identical field tensor, the identical forces on every charge, the identical experimental record. Physics does not live at points of potential-space; it lives on orbits. A gauge condition is a surface cutting across the orbits, meeting each (ideally) once — a choice of representative, like choosing sea level as the zero of altitude. Nothing measurable can depend on the choice; everything computational does. Relativity adds the twist this chapter cares about. Different inertial observers each have their own natural slicings, and a condition written in one observer's 3+1 language — "the spatial divergence of $\mathbf{A}$ vanishes" — picks a surface tilted with respect to the one a moving observer would draw from the same words. The Lorenz condition is special because it is written in no observer's language: $\partial_\mu A^\mu$ is a full spacetime divergence, a Lorentz scalar, so all observers agree on which potentials satisfy it. The slice is parallel to the boosts, so to speak: relativistic calculations can hand potentials between frames without leaving the gauge. Coulomb-gauge potentials cannot travel that way, and the first worked example below prices the failure.

Figure

Gauge orbits and gauge slices — physics lives on orbits, conditions choose representatives

One more instinct to install before the mathematics. Gauge freedom is often introduced apologetically, as untidiness to be mopped up. The modern view inverts this: the redundancy is load-bearing. It is what lets the theory carry a massless spin-1 field in a manifestly covariant package — four components for covariance, two of them removable for masslessness — and, when this track reaches the action principle next article, the demand that the redundancy be respected will turn out to dictate how light couples to matter and why charge is conserved. In quantum electrodynamics the freedom catalogued here must be fixed, tracked, and cancelled with full covariant bookkeeping; every tool for that job is a covariant gauge condition from this article's taxonomy.

The freedom, and the conditions that manage it

The starting point is the statement The Four-Potential certified: for any smooth scalar field $\chi(x)$ ,

A^\mu\;\longrightarrow\;A'^\mu=A^\mu+\partial^\mu\chi \qquad\Longrightarrow\qquad F'^{\mu\nu}=F^{\mu\nu}, \qquad (1)

where: the invariance is one line deep — $F^{\mu\nu}$ picks up $\partial^\mu\partial^\nu\chi-\partial^\nu\partial^\mu\chi=0$ because derivatives commute (The Electromagnetic Field Tensor, gauge-invariance step). In 3+1 language the same move reads $V\to V-\partial_t\chi$ , $\mathbf{A}\to\mathbf{A}+\nabla\chi$ : the whole apparatus E9's Gauge Transformations built, whose invariance and completeness theorems — every pair of potentials for the same field differs by exactly such a $\chi$ — we import wholesale. Equation (1) says the map $A\mapsto F$ is many-to-one, and the fibers (orbits) are labeled by one free scalar function. A gauge condition is an equation $\mathcal{C}[A]=0$ intended to intersect each orbit once. The workhorse taxonomy:

condition	equation	covariant?	typical habitat
Lorenz	$\partial_\mu A^\mu=0$	yes — Lorentz scalar	radiation, relativity, QED
Coulomb	$\nabla\cdot\mathbf{A}=0$	no — one frame's divergence	atomic physics, bound states
temporal	$A^0=0$	no — one frame's time axis	canonical quantization
axial	$n_\mu A^\mu=0$ , $n$ fixed	only if $n^\mu$ is a physical four-vector	gauge-theory technology
Feynman-family	$\partial_\mu A^\mu$ unfixed, weighted in the action	yes — scalar weighting	quantized perturbation theory

where: "covariant" asks a precise question — is the condition itself preserved by Lorentz transformation of the potential? For the Lorenz row, $\partial_\mu A^\mu$ is a contraction of a covariant gradient with a contravariant four-vector, hence a scalar field (Four-Vectors and the Metric Tensor's contraction doctrine): zero in one frame, zero in all. For the Coulomb row the answer is no, and because that "no" is the crux of the article, Derivation A computes it rather than asserting it. Why anyone bothers with the Lorenz slice is already on record: on it, the gauge-blind master equation of Covariant Form of Maxwell's Equations — $\Box A^\nu-\partial^\nu(\partial\cdot A)=\mu_0J^\nu$ — drops its second term, leaving

\Box A^\mu=\mu_0 J^\mu, \qquad (2)

where: four uncoupled wave operators carry the four potential components, sources on the right, relativity manifest, solved once and for all by the invariant Green function of Covariant Wave Equation and Retarded Solutions three articles ahead. The 3+1 groundwork — that the Lorenz slice is always reachable from any potential, and that it persists in time exactly when charge is conserved — was proven by E9's Lorenz Gauge; those theorems are cited here as finished goods. What E9 could not say, and this chapter can, is that the slice is frame-proof: reachable-and-persistent is now a statement all observers share, because the condition is a scalar and the wave equation it must solve, $\Box\chi=-\partial\cdot A$ , is built from invariant machinery.

Derivation ACovariance of gauge conditions — Lorenz passes, Coulomb fails by a computable amount

We prove that the Lorenz condition transforms as a Lorentz scalar and therefore holds in every frame or none (per orbit representative), and then show by direct construction that the Coulomb condition does not survive a boost: a potential with $\nabla\cdot\mathbf{A}=0$ in the lab acquires a nonzero spatial divergence for a moving observer. The strategy for the second half is to pick the cleanest possible victim — the static point charge, whose Coulomb-gauge potential is also Lorenz — boost it, and differentiate. Watch that the two conditions coincide for statics and are pulled apart by the boost: that is the whole story in miniature.

Step 1. Lorenz is a scalar. Under $x^\mu\to\bar x^\mu=\Lambda^\mu{}_\nu x^\nu$ , the four-gradient transforms covariantly and $A^\mu$ contravariantly, so their contraction is invariant:

\bar\partial_\mu\bar A^\mu(\bar x)=\partial_\nu A^\nu(x).

If $\partial\cdot A=0$ at every event in one frame, the same function — evaluated by any boosted observer at the same events — vanishes. The Lorenz slice is one surface, drawn identically by all observers. Why this is not a triviality: the statement is about the condition, not the potential. $\bar A^\mu$ is a different list of four functions from $A^\mu$ ; what is preserved is the property " $\partial\cdot A=0$ ", because that property is expressed as the vanishing of a scalar field.

Step 2. Set up the Coulomb-gauge victim. In the charge's rest frame $S$ : $V=q/4\pi\varepsilon_0 r$ , $\mathbf{A}=0$ . This satisfies $\nabla\cdot\mathbf{A}=0$ trivially — and also $\partial\cdot A=\tfrac{1}{c^2}\partial_tV+\nabla\cdot\mathbf{A}=0$ , since nothing depends on time. For statics the two gauges coincide; only a boost can separate them.

Step 3. Boost to $\bar S$ , where the charge moves at $+v\hat{\mathbf{x}}$ . The potential transforms as a four-vector:

\bar A^0=\gamma A^0=\frac{\gamma V}{c}, \qquad \bar A^x=\gamma\beta A^0=\frac{\gamma\beta V}{c}, \qquad \bar A^y=\bar A^z=0,

with the argument re-expressed in barred coordinates through $x=\gamma(\bar x-v\bar t)$ , $y=\bar y$ , $z=\bar z$ . The moving observer sees a vector potential pointing along the motion, proportional to the (boosted) Coulomb potential — the uniform-motion Liénard–Wiechert pair that The Four-Potential obtained by exactly this boost.

Step 4. Differentiate. Evaluate $\bar\nabla\cdot\bar{\mathbf{A}}=\partial_{\bar x}\bar A^x$ at the event $\bar t=0$ , $\bar{\mathbf{x}}=(a,0,0)$ — a point at distance $\gamma a$ ahead of the charge in rest-frame coordinates:

\partial_{\bar x}\bar A^x =\frac{\gamma\beta}{c}\,\frac{\partial V}{\partial x}\frac{\partial x}{\partial\bar x} =\frac{\gamma\beta}{c}\left(-\frac{q}{4\pi\varepsilon_0(\gamma a)^2}\right)\gamma =-\frac{\beta q}{4\pi\varepsilon_0\,c\,a^2}.

For $q=1$ nC, $a=1$ m, $\beta=0.6$ ( $\gamma=1.25$ ): $|\bar\nabla\cdot\bar{\mathbf{A}}|=0.6\times8.99/(2.998\times10^8)=1.8\times10^{-8}$ V·s/m² — not zero. The barred observer, applying the words "Coulomb gauge" in her own frame, finds the inherited potential disqualified; to restore $\bar\nabla\cdot\bar{\mathbf{A}}=0$ she must perform her own gauge transformation, different at every boost. Meanwhile $\bar\partial_\mu\bar A^\mu=0$ still holds exactly, by Step 1 — the same potential remains a citizen of the Lorenz slice in every frame. Verified: the boosted pair recomputed against the uniform-motion Liénard–Wiechert potentials of E9 (matches, as it must); the divergence evaluated symbolically (sympy) and numerically ( $1.799\times10^{-8}$ V·s/m²); the Lorenz condition checked to vanish identically for the boosted pair at arbitrary events.

Go deeper · Derivation B · 2 minSpending the residue — from the Lorenz slice to two transverse polarizations

Derivation BSpending the residue — from the Lorenz slice to two transverse polarizations

Lorenz gauge does not nail the potential: transformations with $\Box\chi=0$ move you along the slice. Here we spend that residual freedom on the free electromagnetic field and count what survives. The strategy is to work in the plane-wave sector — every free field is a superposition, and gauge bookkeeping respects superposition — impose Lorenz, then use one harmonic $\chi$ to strip the time component. The result is the covariant ledger behind a fact E8 discovered with curls: light is transverse, with exactly two polarization states. Watch what is honest and what is frame-tied at the end; the counting is invariant even though the final gauge is not.

Step 1. The residue. If $\partial\cdot A=0$ and $A'^\mu=A^\mu+\partial^\mu\chi$ , then $\partial\cdot A'=\Box\chi$ : the transformed potential stays on the slice exactly when $\Box\chi=0$ . The leftover freedom is itself a free wave — gauge ripples travelling at $c$ alongside the physics.

Step 2. Plane-wave sector. Take $A^\mu=\mathrm{Re}\!\left[\varepsilon^\mu e^{ik\cdot x}\right]$ with $k\cdot k=0$ (the free dispersion $\omega=c|\mathbf{k}|$ of The Electromagnetic Wave Equation, in invariant dress). The Lorenz condition becomes algebra:

\partial_\mu A^\mu=0\;\Longrightarrow\;k_\mu\varepsilon^\mu=0.

One linear constraint: four amplitudes drop to three. Take the concrete quadruple $\varepsilon^\mu=(3,1,2,3)$ with $k^\mu=(k,0,0,k)$ : check $k\cdot\varepsilon=k(3)-k(3)=0$ — admissible.

Step 3. Spend the residue. Choose $\chi=\mathrm{Re}\!\left[\tfrac{\lambda}{i}\,e^{ik\cdot x}\right]$ ; then $\Box\chi=0$ automatically (null $k$ ), and the transformation shifts the polarization four-vector by a multiple of $k$ :

\varepsilon^\mu\;\longrightarrow\;\varepsilon^\mu+\lambda k^\mu.

Pick $\lambda$ to kill the time component: $\lambda=-\varepsilon^0/k^0=-3/k$ gives $\varepsilon'^\mu=(0,1,2,0)$ . With $\varepsilon'^0=0$ , the surviving Lorenz constraint reads $\mathbf{k}\cdot\boldsymbol{\varepsilon}'=0$ — the amplitude is transverse, and its longitudinal part (here the $z$ -component, along $\mathbf{k}$ ) has been gauged away with the time part in one stroke.

Step 4. Count, and check invariance of the physics. Ledger: $4$ components $-1$ (Lorenz) $-1$ (residual) $=2$ — for $\mathbf{k}\parallel\hat{\mathbf{z}}$ , the states $\hat{\mathbf{x}}$ and $\hat{\mathbf{y}}$ , exactly the two polarizations Monochromatic Plane Waves in Vacuum extracted from the curl equations. The gauge-blind scalar $\varepsilon\cdot\varepsilon$ audits the bookkeeping: before, $3^2-1^2-2^2-3^2=-5$ ; after, $0-1^2-2^2-0=-5$ . Physics untouched, dead weight removed. Honesty clause: the final condition $\varepsilon^0=0$ (radiation gauge) is frame-tied — a boosted observer must re-spend her own residue to reach it — but the count is a Lorentz-invariant statement, because both subtractions were made with covariant equations ( $k\cdot\varepsilon=0$ ; shift by $\lambda k^\mu$ ). Verified: $k\cdot k=0$ , $k\cdot\varepsilon=0$ , $\varepsilon\cdot\varepsilon=\varepsilon'\cdot\varepsilon'=-5$ , $k\cdot\varepsilon'=0$ recomputed numerically; counting logic cross-checked against Jackson §6.3/§12 and Zangwill Ch. 22; transversality statement matched by name to the E8 plane-wave triad.

Figure

The degrees-of-freedom ledger — four potential components stripped to two polarizations

Fixing gauge inside the dynamics — the ξ family

There is a third way to manage the freedom, and although its full machinery belongs to the action principle of the next article, its shape belongs in this taxonomy. Instead of imposing $\partial\cdot A=0$ as an external constraint, one modifies the dynamics to penalize departures from it, replacing the master equation by

\Box A^\nu-\Big(1-\tfrac{1}{\xi}\Big)\,\partial^\nu(\partial\cdot A)=\mu_0 J^\nu, \qquad (3)

where: $\xi$ is a dimensionless gauge-weighting parameter. Every value of $\xi$ yields the same $F^{\mu\nu}$ for conserved sources — the parameter multiplies only the orbit direction — and the choice $\xi=1$ (Feynman gauge) makes the operator collapse to a pure d'Alembertian on all four components, the smoothest possible bookkeeping. This is the form in which gauge fixing enters quantum electrodynamics: not a constraint appended to the theory but a term added to its Lagrangian density, $-(\partial\cdot A)^2/2\xi\mu_0$ . How terms get added to Lagrangians, and why this one is legitimate, is Relativistic Lagrangian and Action for Charged Particles' business; here it stands as the taxonomy's final row — the covariant condition that is not a condition at all, but a weight.

Limiting cases

case	result	meaning
statics	Lorenz and Coulomb conditions coincide ( $\partial_tV=0$ )	the gauges only separate for time-dependent or boosted problems
free field, Lorenz + residue spent	$A^0=0$ , $\mathbf{k}\cdot\mathbf{A}=0$ , two states	radiation gauge; E8 transversality recovered covariantly
$\xi=1$ in the weighted dynamics	$\Box A^\nu=\mu_0J^\nu$ componentwise	Feynman gauge — Lorenz-gauge physics, no explicit constraint
$\xi\to0$	constraint enforced infinitely stiffly	Landau gauge; $\partial\cdot A=0$ exact
$\chi$ constant	$A^\mu$ unchanged	the only gauge transformation that is no transformation — freedom is in gradients only
non-simply-connected regions	orbits acquire topological labels	flux through excluded holes survives gauging — the Aharonov–Bohm seed (QM track)

Common trapLorenz is not Lorentz — and a gauge condition is not a gauge fixing

Two distinct traps, both classics. First, the name: the condition $\partial_\mu A^\mu=0$ belongs to Ludvig Lorenz (Copenhagen, 1867), not Hendrik Lorentz of the transformations — different men, one letter apart, and the confusion is so entrenched that generations of textbooks credited the wrong one. The coincidence is admittedly poetic: Lorenz's condition is the one that respects Lorentz's symmetry. Second, the logic: imposing the Lorenz condition does not produce a unique potential — the residual $\Box\chi=0$ family still roams the slice, and forgetting it causes real errors: you may find your "fixed" potential differing from a colleague's by a harmonic gauge wave, and neither of you is wrong. Complete fixing requires spending the residue too (as Derivation B does for free fields), or weighting it away in the ξ formalism. When comparing potentials — never fields — across references, first ask which slice, then ask where on the slice.

Worked examples

Example 1 — pricing the Coulomb gauge's frame dependence. A 1 nC charge sits at rest; its potential $(V=q/4\pi\varepsilon_0 r,\ \mathbf{A}=0)$ satisfies both Coulomb and Lorenz conditions. An observer flies past at $\beta=0.6$ ( $\gamma=1.25$ ) along $x$ and inherits the boosted potential.

Step 1. She measures, at her event $(\bar t=0,\ \bar x=1.0\ \text{m},\ \bar y=\bar z=0)$ , the inherited $\bar A^x=\gamma\beta V/c$ with $V$ evaluated at rest-frame distance $\gamma a=1.25$ m: $V=8.99/1.25=7.19$ V, so $\bar A^x=1.25\times0.6\times7.19/(2.998\times10^8)=1.80\times10^{-8}$ V·s/m.

Step 2. Her divergence audit (Derivation A, Step 4): $\bar\nabla\cdot\bar{\mathbf{A}}=-\beta q/4\pi\varepsilon_0 ca^2=-1.8\times10^{-8}$ V·s/m² $\neq0$ : Coulomb gauge broken by the boost. Her Lorenz audit: $\tfrac{1}{c^2}\partial_{\bar t}\bar V+\bar\nabla\cdot\bar{\mathbf{A}}=0$ exactly — the time derivative of her scalar potential (the charge is moving, so $\bar V$ ticks) cancels the divergence to all decimal places.

Step 3. The moral, priced: the violation scales as $\beta q/ca^2$ — first order in the boost, long-ranged ( $1/a^2$ ), and proportional to the very potential the gauge was supposed to organize. To re-impose her own Coulomb gauge she must solve $\bar\nabla^2\bar\chi=-\bar\nabla\cdot\bar{\mathbf{A}}$ — a fresh Poisson problem at every velocity. The Lorenz slice travels; the Coulomb slice must be rebuilt in every frame.

Example 2 — gauge-stripping a plane wave. A free wave moves along $+\hat{\mathbf{z}}$ with null wavevector $k^\mu=(k,0,0,k)$ and (Lorenz-admissible) polarization $\varepsilon^\mu=(3,1,2,3)$ in some units.

Step 1. Verify admissibility: $k\cdot\varepsilon=k(3)-k(3)=0$ . ✓ Lorenz holds; three components are still in play, including an unphysical time part and a longitudinal part.

Step 2. Spend the residue with $\lambda=-\varepsilon^0/k^0=-3/k$ : the shift $\varepsilon^\mu\to\varepsilon^\mu+\lambda k^\mu$ subtracts $3$ from the time slot and $3$ from the $z$ slot simultaneously: $\varepsilon'^\mu=(0,1,2,0)$ . One move removed both unphysical directions — because for a null $k$ they were the same direction along the orbit.

Step 3. Read the physics: transverse amplitude $\boldsymbol{\varepsilon}'_\perp=(1,2)$ in the $xy$ -plane — linear polarization at angle $\arctan(2/1)=63.4°$ to $\hat{\mathbf{x}}$ , amplitude $\sqrt{5}$ . Invariant audit: $\varepsilon\cdot\varepsilon=9-1-4-9=-5$ before, $-(1+4)=-5$ after — the gauge-proof content, $-|\boldsymbol{\varepsilon}_\perp|^2$ , never moved. The four-potential carried $\sqrt{5}$ 's worth of light and $3$ 's worth of bookkeeping.

Going further

Reading. Griffiths, Introduction to Electrodynamics, 4th ed., §10.1 and §12.3 — the potential formulation and its relativistic dress. Jackson, Classical Electrodynamics, 3rd ed., §6.3 and §11.9 — gauge transformations, the Lorenz condition, and the covariant form; see also Jackson & Okun, Rev. Mod. Phys. 73, 663 (2001) for the naming forensics. Zangwill, Modern Electrodynamics, §15.3 and Ch. 22 — gauge freedom with modern bookkeeping. Purcell & Morin, 3rd ed., Ch. 6 appendix.

Provenance. Built from the AllQFT base article for this topic (gauge-orbit framing, Lorenz/Coulomb/temporal taxonomy, residual-freedom counting, and the naming pitfall retained; base's gauge function $\lambda$ renamed $\chi$ to match this track's usage). A verifier should: (1) confirm the scalar-transformation argument for $\partial\cdot A$ ; (2) recompute the boosted point-charge potential and its spatial divergence ( $-\beta q/4\pi\varepsilon_0ca^2$ ; $1.8\times10^{-8}$ V·s/m² at $q=1$ nC, $a=1$ m, $\beta=0.6$ ) and check the Lorenz condition survives the same boost; (3) rerun the plane-wave ledger ( $k\cdot\varepsilon=0$ , shift by $\lambda k^\mu$ , $\varepsilon\cdot\varepsilon=-5$ invariant, final $(0,1,2,0)$ ); (4) verify equation (3) reduces correctly at $\xi=1$ and $\xi\to0$ . Division of labor for this edition: the 3+1 gauge machinery (invariance, completeness, continuity-built-in) is owned by E9's Gauge Transformations; Lorenz-slice attainability and persistence by Lorenz Gauge; the Coulomb gauge's instantaneous-potential acquittal by Coulomb Gauge; the freedom's statement and the $\Box A^\mu=\mu_0J^\mu$ assembly by The Four-Potential; the action-principle standing of the ξ term by Relativistic Lagrangian and Action for Charged Particles (next-but-one); covariant retarded solutions by Covariant Wave Equation and Retarded Solutions. This article owns the covariance classification of gauge conditions, the boosted-Coulomb violation computation, the residue-spending polarization count, and the taxonomy.

Builds on: The Four-Potential · The Electromagnetic Field Tensor · Four-Vectors and the Metric Tensor · The Four-Current Density · Lorentz Transformations

Leads to: Relativistic Dynamics of a Charged Particle · Relativistic Lagrangian and Action for Charged Particles · Covariant Wave Equation and Retarded Solutions · The Dual Field Tensor

Is gauge freedom a symmetry of nature or just redundancy in our description?

Redundancy — but redundancy with teeth. A gauge transformation maps a potential to another description of the same physical state; nothing observable changes, so unlike a rotation or boost it does not relate different experiments. In that sense it is not a symmetry of nature. But the requirement that physics be blind to the redundancy is enormously restrictive: it forbids a photon mass term, dictates the coupling structure to matter (minimal coupling — next articles), and ties into charge conservation. Modern usage says it cleanly: gauge "symmetry" is a constraint on theories, not on states. The practical corollary is this article's discipline — only gauge-invariant quantities ( $F^{\mu\nu}$ , its invariants, action integrals, phase differences) may carry physical claims.

Do I have to use the Lorenz gauge for relativistic problems?

No — physics is gauge-independent, and any slice yields the same fields. What the Lorenz choice buys is bookkeeping that survives boosts: the condition itself is a scalar, so a potential prepared on the slice stays on it in every frame, and intermediate quantities keep their four-vector character. In Coulomb gauge the fields still come out right in any single frame — E9's Coulomb Gauge article acquitted its instantaneous potential of causality violation precisely because the fields repair the appearance — but every change of frame demands a fresh gauge repair (Example 1 prices it). The professional rule: cross frames in Lorenz dress, exploit Coulomb dress for one-frame bound-state work, and translate at the end through gauge-invariant quantities only.

Why are Lorenz and Lorentz different people, and who actually owns the condition?

Ludvig Lorenz (1829–1891), Danish, published the retarded potentials and the condition $\partial_tV/c^2+\nabla\cdot\mathbf{A}=0$ in 1867 — decades before relativity. Hendrik Lorentz (1853–1928), Dutch, built the electron theory and the transformations this chapter runs on, and also used the condition, which is how his name stuck to it. The 2001 Jackson–Okun review settled the paternity: the condition is Lorenz's; the covariance of the condition is Lorentz's mathematics; the poetic justice — Lorenz's condition being Lorentz-invariant — belongs to nobody and delights everybody. This track spells them apart deliberately: watch the "t".

Is the residual gauge freedom after imposing the Lorenz condition ever physically important?

Constantly. Classically, it is why two correct references can print different Lorenz-gauge potentials for the same field — they sit at different points along the slice, differing by a harmonic gauge wave; comparing potentials without checking this wastes afternoons. It is also the freedom Derivation B spends to exhibit transversality. Quantum mechanically it becomes load-bearing: in covariant quantization all four photon polarizations are provisionally kept, and the residual freedom reappears as the condition (Gupta–Bleuler) that strips timelike and longitudinal quanta from physical states in matched, cancelling pairs. The pattern — impose a covariant condition, then spend a residue — is the same at both levels; only the arena changes from function space to Hilbert space.

If potentials are pure bookkeeping, why does quantum mechanics seem to care about them so much?

Because quantum matter couples to potentials locally and to fields only through them — yet in a gauge-invariant way. The resolution of the apparent paradox is that quantum phases care about gauge-invariant functionals of the potential that are not local functions of the fields: the line integral $\oint A_\mu dx^\mu$ around a closed loop equals the flux through it and is gauge-proof, even where fields vanish along the loop itself. That is the Aharonov–Bohm arrangement, and it shows the honest hierarchy: not "fields fundamental, potentials bookkeeping," but "gauge equivalence classes fundamental; fields are one convenient set of class invariants, and not quite a complete set in nontrivial topology." The minimal-coupling construction in the next-but-one article is where the potential's quantum role is built classically.

What breaks if I try to fix the gauge completely with one covariant condition?

A theorem's worth of stubbornness: no single covariant condition both slices every orbit exactly once and preserves manifest locality — some residue or some frame-tie always survives. Lorenz leaves the harmonic residue; axial gauges $n\cdot A=0$ fix completely but drag a preferred four-vector $n^\mu$ into every formula; radiation gauge fixes free fields but is frame-tied. In the quantized theory this stubbornness deepens into the Gribov problem for non-abelian fields (slices that re-intersect orbits far from the origin). The working resolution is the ξ formalism: stop trying to slice geometrically, weight the orbit direction in the dynamics, and let cancellations guarantee that no observable notices the weight. That pragmatism, not a perfect slice, is what actually powers QED calculations.

Does the two-polarization count depend on the frame, since the radiation gauge itself is frame-tied?

No — and the ledger shows why. The two subtractions that produce the count are covariant statements: $k\cdot\varepsilon=0$ is a scalar equation, and "shift by $\lambda k^\mu$ " is a statement about the orbit direction, which is itself built covariantly. What is frame-tied is only the labeling of the two survivors: my choice $\varepsilon^0=0$ picks particular representatives that a boosted observer would not pick; she re-spends her residue and lands on her own transverse pair, related to mine by a gauge shift plus the aberration of $\mathbf{k}$ (Relativistic Doppler Effect and Aberration, ahead, handles how $k^\mu$ itself boosts). Two states in every frame; which two vectors represent them is gauge dress. The invariant $\varepsilon\cdot\varepsilon$ audit in Example 2 is the frame-proof receipt.