Gauge Transformations

The chapter opener ended with a mild embarrassment. Potentials were introduced to make the two homogeneous Maxwell equations automatic — $\mathbf{B}=\nabla\times\mathbf{A}$ and $\mathbf{E}=-\nabla V-\partial\mathbf{A}/\partial t$ — and the move worked: six field components collapsed into four potential components, and two of the four Maxwell equations dissolved into identities. But the bookkeeping came out loose. The map from potentials to fields is many-to-one: infinitely many distinct pairs $(V,\mathbf{A})$ describe exactly the same electromagnetic field, everywhere and for all time. You met a static slice of this already, in magnetostatics, where $\mathbf{A}$ could absorb any gradient. Electrodynamics doubles the plot: now the scalar potential joins the conspiracy, and the two shifts must be choreographed together or the electric field breaks. This article measures the slack exactly. One arbitrary scalar function $\lambda(\mathbf{r},t)$ — the gauge function — generates the entire redundancy: $\mathbf{A}\to\mathbf{A}+\nabla\lambda$ with the compensating $V\to V-\partial\lambda/\partial t$ leaves both fields untouched, and every pair of potential descriptions of the same field is connected by exactly such a $\lambda$ . That converse is the sharp part, and it is what licenses everything the rest of the chapter does: because the freedom is total and harmless, you may spend it, imposing one scalar condition on the potentials to make their equations of motion as pleasant as possible. Two ways of spending it dominate physics — the Coulomb gauge and the Lorenz gauge, each the subject of its own article next — and the retarded potentials, the point-charge potentials, and every radiation formula downstream are all written in the currency this article mints. Beyond the chapter, the same transformation, read as a symmetry principle rather than a nuisance, is the structural template for the Standard Model.

Key takeaways

The potentials behind given fields are not unique: $\mathbf{A}'=\mathbf{A}+\nabla\lambda$ and $V'=V-\partial\lambda/\partial t$ , for any smooth scalar $\lambda(\mathbf{r},t)$ , produce identical $\mathbf{E}$ and $\mathbf{B}$ .
The two shifts are locked together: the gradient added to $\mathbf{A}$ would contaminate $\mathbf{E}$ through $-\partial\mathbf{A}/\partial t$ , and the shift of $V$ is precisely the antidote. In statics the $V$ shift sleeps; time-dependence wakes it.
The freedom is complete as well as harmless: any two potential descriptions of the same field are connected by one shared gauge function $\lambda$ — no residue, no exceptions (on simply connected domains).
Fixing a gauge means imposing one scalar condition ( $\nabla\cdot\mathbf{A}=0$ , the Coulomb gauge; $\nabla\cdot\mathbf{A}=-\mu_0\varepsilon_0\,\partial V/\partial t$ , the Lorenz gauge) to pick a convenient representative from each equivalence class.
Physical statements must be gauge-invariant: the fields, forces, energies, and loop integrals $\oint\mathbf{A}\cdot d\boldsymbol\ell$ qualify; the pointwise values of $V$ and $\mathbf{A}$ do not.
The consistency of the whole potential formulation quietly requires charge conservation — continuity is gauge freedom's structural partner, not an extra assumption.
Key idea: the transformation $\mathbf{A}\to\mathbf{A}+\nabla\lambda$ , $V\to V-\partial\lambda/\partial t$ leaves $\mathbf{E}$ and $\mathbf{B}$ unchanged for any scalar $\lambda(\mathbf{r},t)$ — and it exhausts the freedom, so only gauge-invariant quantities are physical.

The core insight

The transformation $\mathbf{A}\to\mathbf{A}+\nabla\lambda$ , $V\to V-\partial\lambda/\partial t$ leaves $\mathbf{E}$ and $\mathbf{B}$ unchanged for any scalar $\lambda(\mathbf{r},t)$ ; it exhausts the redundancy of the potentials, and only gauge-invariant quantities are physical.

Where this lives today

Gauge freedom graduated from nuisance to organizing principle. The Standard Model is built by demanding invariance under transformations of exactly this type — the electromagnetic $\lambda$ generalized to matrix-valued functions — and reading the interactions off as the price of the demand; every collider cross-section computed for the LHC is evaluated in some gauge (Feynman gauge for loops, often axial gauges for jets) with the final number checked gauge-invariant, the modern industrial version of this article's doctrine. In precision QED, the electron's magnetic moment is currently computed through five loops — tens of thousands of gauge-dependent integrals conspiring to a gauge-invariant 13 significant figures that agree with experiment. The discipline bites in numerical work too: numerical-relativity and electromagnetics codes must fix a gauge to run at all, a poor slice can crash a simulation with pure coordinate pathologies, and gauge-invariant extraction of observables is a named pipeline step. And the Aharonov–Bohm loop integral of equation (3) is now engineering: superconducting flux qubits store quantum information in gauge-invariant flux states threading a ring — the $\oint\mathbf{A}\cdot d\boldsymbol\ell$ that in 1959 was a philosophical scandal is read out billions of times per second in quantum processors.

Origins · 1 minHendrik Antoon Lorentz and how this was discovered

Hendrik Antoon Lorentz

1853–1928 · Leiden — who systematized the potentials of electrodynamics, and lent the wrong name to the right gauge

Lorentz was the great organizer of classical electron theory: from his chair at Leiden he rebuilt Maxwell's theory around charged particles and fields, derived the force law that bears his name, and in his 1904 encyclopedia article and lectures set out the potential formulation of electrodynamics in essentially the modern form — including the transformation $\mathbf{A}\to\mathbf{A}+\nabla\lambda$ , $V\to V-\partial_t\lambda$ as the general statement of the potentials' freedom, and the systematic use of the condition that decouples their wave equations. His authority was such that the condition and the gauge were christened "Lorentz's" — though the Dane Ludvig Lorenz had used it in 1867, when Hendrik Antoon was fourteen; the community's slow correction to "Lorenz gauge" is the next article's opening story. Lorentz shared the 1902 Nobel Prize with Zeeman, chaired the Solvay councils that midwifed quantum theory, and was the physicist Einstein said he admired above all others. That the transformations of special relativity also carry his name is a separate, fully deserved credit — the man is under-cited only in the one place he is over-cited.

History

1867

— Ludvig Lorenz writes electrodynamics in terms of retarded potentials, implicitly using the freedom and the condition later misattributed

1904

— H. A. Lorentz's encyclopedia treatment systematizes the potential formulation and the general gauge transformation

1918

— Hermann Weyl attempts to unify gravity and electromagnetism via local scale ("Eich-", gauge) invariance — wrong theory, immortal name

1926

— Vladimir Fock shows the quantum wavefunction must rotate its phase in step with the classical gauge shift: gauge symmetry couples to matter

1929

— Weyl recasts his idea as phase invariance of quantum electrodynamics: the modern gauge principle

1954

— Yang and Mills generalize the gauge function to non-commuting symmetries, the template for the strong and electroweak forces

1959

— Aharonov and Bohm predict interference shifts from the gauge-invariant loop integral of A where B vanishes; confirmed by Chambers in 1960 and definitively by Tonomura in 1986

The physical picture

Think of the potentials as an altitude map and the fields as the terrain's slopes. Nobody hiking the landscape ever measures altitude — only steepness, at every point, in every direction. You can raise the whole map by a constant, or warp its reference level from place to place and moment to moment, and so long as the warp is choreographed so every slope comes out the same, the two maps describe one landscape. The gauge function $\lambda$ is exactly such a warp. Two physicists using different $\lambda$ 's disagree about $V$ and $\mathbf{A}$ at every single point of spacetime — and compute identical forces on every charge, identical fluxes, identical energies. Neither is wrong. They are using different coordinates on the same physics. The magnetostatics version of this story — met three chapters ago — had only one knob: add $\nabla\lambda$ to $\mathbf{A}$ and $\mathbf{B}$ survives, because the curl of a gradient dies identically. Electrodynamics is where the choreography becomes visible. The electric field now reads $\mathbf{A}$ too, through $-\partial\mathbf{A}/\partial t$ , so a time-dependent $\nabla\lambda$ added to $\mathbf{A}$ would change $\mathbf{E}$ — unless the scalar potential simultaneously absorbs $-\partial\lambda/\partial t$ , whose gradient cancels the intrusion term for term. One scalar function, two coordinated shifts, both fields blind to it. The diagram below wires up the cancellation.

Figure

The gauge choreography — one scalar function shifts both potentials, and the two routes into E cancel exactly while B never sees a gradient

How much freedom is that, really? Count. The potentials carry four functions of space and time, $(V, A_x, A_y, A_z)$ . The gauge function eats one. A gauge condition — one scalar equation — pins one more. Two functional degrees of freedom remain, and these are physical: they are the two transverse polarizations that will become the radiation field later in this chapter, and the two polarization states of the photon in the quantum theory. The potentials over-describe the physics by exactly the amount $\lambda$ can reach, no more. The picture to carry forward is a space of descriptions sliced into equivalence classes. Every point is a pair $(V,\mathbf{A})$ ; two points connected by some $\lambda$ lie on the same gauge orbit, and all points of an orbit describe one physical field history. A gauge condition is a surface slicing through the space, meeting each orbit in (ideally) one point — the chosen representative. The Coulomb slice and the Lorenz slice cut differently, and each intersects every orbit, which is why you may always demand either condition. That freedom-spending is the next two articles' business; this article's business is the orbit structure itself.

Figure

Gauge orbits and gauge slices — each orbit is one physical field described infinitely many ways; a gauge condition picks a representative from every orbit

The transformation, measured exactly

Everything rests on the field definitions delivered by the chapter opener, Scalar and Vector Potentials:

\mathbf{B}=\nabla\times\mathbf{A},\qquad \mathbf{E}=-\nabla V-\frac{\partial\mathbf{A}}{\partial t}. \qquad (1)

where: $V$ and $\mathbf{A}$ are the scalar and vector potentials; the pair of definitions makes $\nabla\cdot\mathbf{B}=0$ and Faraday's law automatic. Read (1) as a division of labor inside $\mathbf{E}$ : the $-\nabla V$ piece is curl-free by construction, while $-\partial\mathbf{A}/\partial t$ carries all of the curl that Faraday's law demands — and note that $\mathbf{A}$ now reports to both fields, which is exactly why its freedom must be chaperoned. The claim to be established, then spent for the rest of the chapter, is the gauge transformation:

\mathbf{A}'=\mathbf{A}+\nabla\lambda,\qquad V'=V-\frac{\partial\lambda}{\partial t}, \qquad (2)

where: $\lambda(\mathbf{r},t)$ is any smooth scalar field — the gauge function — with no equation of motion, no boundary condition, no physical content whatsoever. The primed and unprimed potentials produce identical fields (Derivation A), and every pair of potential descriptions of the same fields is related by exactly one such $\lambda$ , up to an additive constant (Derivation B). Note the units: $\lambda$ carries volt-seconds, so $\nabla\lambda$ has the units of $\mathbf{A}$ (V·s/m) and $\partial\lambda/\partial t$ the units of $V$ — the dimensional bookkeeping already hints that one scalar can feed both potentials.

Invariance is only half the theorem, and the less interesting half. If some other residual freedom existed — some way two potential pairs could agree on the fields without being $\lambda$ -related — then gauge fixing could silently discard physics. The converse closes that door, and it is where the time-dependent theory earns its keep beyond the magnetostatic version.

Go deeper · Derivation B · 2 minCompleteness — the gauge function exhausts the freedom

Derivation BCompleteness — the gauge function exhausts the freedom

The goal is the converse theorem: if $(V,\mathbf{A})$ and $(V',\mathbf{A}')$ generate the same $\mathbf{E}$ and $\mathbf{B}$ everywhere and for all time, then there exists a single scalar $\lambda(\mathbf{r},t)$ with $\mathbf{A}'=\mathbf{A}+\nabla\lambda$ and $V'=V-\partial\lambda/\partial t$ simultaneously. The strategy is to name the differences $\boldsymbol\alpha \equiv \mathbf{A}'-\mathbf{A}$ and $\beta \equiv V'-V$ , feed each field equality in turn, and watch the two conditions weld into one shared gauge function. The magnetostatic half of the argument — curl-free means gradient — is recalled from the magnetostatics chapter, where it was proved with its simply-connected fine print; the electrodynamic content here is the gluing: showing the scalar that generates $\boldsymbol\alpha$ is the same scalar whose time derivative generates $\beta$ .

Step 1. Equal magnetic fields force $\boldsymbol\alpha$ to be a gradient. $\mathbf{B}'=\mathbf{B}$ gives $\nabla\times\boldsymbol\alpha = \mathbf{0}$ at every instant. On a simply connected region, a curl-free field is a gradient — the converse theorem proved in Gauge Freedom and the Coulomb Gauge — applied now at each time separately:

\boldsymbol\alpha(\mathbf{r},t) = \nabla\lambda(\mathbf{r},t).

Why $\lambda$ inherits time dependence: the theorem is applied instant by instant, producing a potential for $\boldsymbol\alpha$ at each $t$ ; smoothness of $\boldsymbol\alpha$ in time lets the family be chosen smoothly, giving one scalar field of space and time. The simply-connected caveat is real — around a topological hole, $\boldsymbol\alpha$ could be curl-free yet not a gradient, the loophole the Aharonov–Bohm ring exploits.

Step 2. Equal electric fields tie $\beta$ to that same $\lambda$ . $\mathbf{E}'=\mathbf{E}$ reads $-\nabla\beta - \partial_t\boldsymbol\alpha = \mathbf{0}$ . Insert $\boldsymbol\alpha = \nabla\lambda$ and commute the derivatives:

\nabla\!\left(\beta + \frac{\partial\lambda}{\partial t}\right) = \mathbf{0} \qquad\Longrightarrow\qquad \beta + \frac{\partial\lambda}{\partial t} = f(t),

a function of time alone — spatially constant, because its gradient vanishes everywhere. Why this is almost, but not yet, the theorem: we have $V'-V = -\partial_t\lambda + f(t)$ ; the stray $f(t)$ says the scalar shift is the time derivative of $\lambda$ only up to a spatially uniform offset.

Step 3. Absorb the stray function. Define $\tilde\lambda(\mathbf{r},t) = \lambda(\mathbf{r},t) - \int_0^t f(t')\,dt'$ . Since the correction carries no space dependence, $\nabla\tilde\lambda = \nabla\lambda$ — Step 1 is undisturbed — while $\partial_t\tilde\lambda = \partial_t\lambda - f(t)$ soaks up the offset exactly:

\mathbf{A}' = \mathbf{A} + \nabla\tilde\lambda, \qquad V' = V - \frac{\partial\tilde\lambda}{\partial t}.

Why this closes the theorem: one scalar now generates both shifts. The only remaining ambiguity in $\tilde\lambda$ itself is an additive constant, which moves neither potential. So the gauge orbits of the picture section are exactly the equivalence classes: same fields if and only if $\lambda$ -related.

Step 4. What was time-dependent about it. In statics the theorem was one application of curl-free-means-gradient. Here it ran twice — once per field — and the two runs had to agree on a single $\lambda$ ; the spatially-uniform $f(t)$ was the entire slack between them, and it was pure bookkeeping. The freedom of electrodynamic potentials is precisely one scalar function of space and time, no more, no less. The counting is self-consistent: 4 potential functions − 1 gauge function − 1 gauge condition = 2 physical polarizations.

Spending the freedom: gauge conditions

Nothing physical distinguishes points on one orbit, so calculation strategy may. A gauge condition is one scalar equation imposed on $(V,\mathbf{A})$ , chosen purely for convenience — the art is matching the slice to the problem. The two workhorses each get a full article next; here is the menu with the price tags, including two limiting slices that appear in specialist work:

gauge condition	what it buys	where it shines
Coulomb: $\nabla\cdot\mathbf{A}=0$	$V$ obeys an instantaneous Poisson equation	statics; radiation theory and QED, where $\mathbf{A}$ carries exactly the two transverse polarizations
Lorenz: $\nabla\cdot\mathbf{A}=-\mu_0\varepsilon_0\,\partial_t V$	$V$ and $\mathbf{A}$ obey decoupled, symmetric wave equations	everything time-dependent in this chapter: retarded potentials, Liénard–Wiechert, radiation; manifestly relativistic
temporal (Weyl): $V=0$	one potential eliminated outright	Hamiltonian and lattice formulations of field theory
axial: $A_z=0$	one component of $\mathbf{A}$ eliminated	certain quantization and plasma problems

Two cautions about the table. First, a gauge condition does not always kill the freedom completely — a residual family of $\lambda$ 's may preserve the condition; each gauge article prices its own residue. Second, attainability is a theorem, not a hope: one must prove the slice meets every orbit by exhibiting the repairing $\lambda$ . Those proofs belong to the next two articles — the Coulomb repair is a Poisson problem (solved in the magnetostatics twin, revisited dynamically in Coulomb Gauge), the Lorenz repair a wave equation (Lorenz Gauge). What survives the slicing unchanged — the gauge-invariant ledger — is worth stating as doctrine:

\text{invariant:}\quad \mathbf{E},\ \mathbf{B},\ \mathbf{F}=q(\mathbf{E}+\mathbf{v}\times\mathbf{B}),\ u,\ \mathbf{S},\ \oint_{\mathcal C}\mathbf{A}\cdot d\boldsymbol\ell = \Phi_{\text{enc}}. \qquad (3)

where: $u$ and $\mathbf{S}$ are the field energy density and Poynting vector (built from fields only), and the loop integral of $\mathbf{A}$ is invariant because the gradient $\nabla\lambda$ integrates to zero around any closed curve — Stokes converts it to the enclosed flux $\Phi_{\text{enc}}$ , a field quantity. That last entry is the deepest: in quantum mechanics the phase $\tfrac{q}{\hbar}\oint\mathbf{A}\cdot d\boldsymbol\ell$ shifts electron interference fringes even when the electrons never touch $\mathbf{B}$ — the Aharonov–Bohm effect — so gauge-invariant combinations of potentials carry physics that the fields at the particle's location do not. Gauge freedom demotes the potentials' pointwise values, not the potentials themselves.

Continuity: the structural partner

There is a quieter theorem sitting inside the potential formulation, and it deserves to be seen once in bare three-dimensional form. When the field definitions (1) are inserted into the two source Maxwell equations, the chapter opener assembled the coupled system

\nabla^2 V + \frac{\partial}{\partial t}(\nabla\cdot\mathbf{A}) = -\frac{\rho}{\varepsilon_0}, \qquad \left(\nabla^2\mathbf{A} - \mu_0\varepsilon_0\frac{\partial^2\mathbf{A}}{\partial t^2}\right) - \nabla\!\left(\nabla\cdot\mathbf{A} + \mu_0\varepsilon_0\frac{\partial V}{\partial t}\right) = -\mu_0\mathbf{J}. \qquad (4)

where: $\rho$ and $\mathbf{J}$ are the sources; these two equations carry all of Maxwell (the other two are identities), in any gauge — the ugly $\nabla(\cdots)$ block is precisely the Lorenz-gauge combination waiting to be zeroed two articles from now. The theorem: the system (4) is solvable only for sources that conserve charge, and conversely a violation of continuity cannot even be written in potential form.

Go deeper · Derivation C · 2 minContinuity is built in — the potential system polices its own sources

Derivation CContinuity is built in — the potential system polices its own sources

The goal is to show that the coupled potential equations (4) force the continuity equation $\partial\rho/\partial t + \nabla\cdot\mathbf{J} = 0$ identically — not as a consequence of the sources' good behavior but as an algebraic identity of the left-hand sides, so that charge conservation is a precondition for any potential description at all. The strategy: take $\varepsilon_0\,\partial_t$ of the scalar equation, add $\tfrac{1}{\mu_0}\nabla\cdot$ of the vector equation, and watch every potential term annihilate its partner. This is the potential-side restatement of the consistency audit performed on Maxwell's equations directly in the Maxwell chapter — recalled, not repeated; what is new here is where the cancellation comes from: the terms that die are exactly the gauge-condition combinations, which is the sense in which continuity is gauge freedom's partner.

Step 1. Differentiate the scalar equation. Multiplying the first of (4) by $-\varepsilon_0$ and applying $\partial_t$ :

-\varepsilon_0\frac{\partial}{\partial t}\nabla^2 V - \varepsilon_0\frac{\partial^2}{\partial t^2}(\nabla\cdot\mathbf{A}) = \frac{\partial\rho}{\partial t}.

Why this move: the continuity equation needs $\partial_t\rho$ ; the scalar equation is the only place $\rho$ lives.

Step 2. Take the divergence of the vector equation. Dividing the second of (4) by $\mu_0$ and applying $\nabla\cdot$ , using $\nabla\cdot(\nabla^2\mathbf{A}) = \nabla^2(\nabla\cdot\mathbf{A})$ and $\nabla\cdot\nabla(\cdots) = \nabla^2(\cdots)$ :

\frac{1}{\mu_0}\Big[\nabla^2(\nabla\cdot\mathbf{A}) - \mu_0\varepsilon_0\frac{\partial^2}{\partial t^2}(\nabla\cdot\mathbf{A}) - \nabla^2(\nabla\cdot\mathbf{A}) - \mu_0\varepsilon_0\nabla^2\frac{\partial V}{\partial t}\Big] = -\nabla\cdot\mathbf{J}.

The two $\nabla^2(\nabla\cdot\mathbf{A})$ terms cancel on the spot, leaving

-\varepsilon_0\frac{\partial^2}{\partial t^2}(\nabla\cdot\mathbf{A}) - \varepsilon_0\,\nabla^2\frac{\partial V}{\partial t} = -\nabla\cdot\mathbf{J}.

Why the cancellation is the point: what survived on the left is, term for term, the same pair of objects produced in Step 1 — mixed derivatives of exactly the combination $\nabla\cdot\mathbf{A}$ and $\partial_t V$ that gauge conditions manipulate.

Step 3. Subtract. Step 1 minus Step 2 (derivatives commute on smooth potentials):

\frac{\partial\rho}{\partial t} + \nabla\cdot\mathbf{J} = -\varepsilon_0\,\partial_t\nabla^2 V - \varepsilon_0\,\partial_t^2(\nabla\cdot\mathbf{A}) + \varepsilon_0\,\partial_t^2(\nabla\cdot\mathbf{A}) + \varepsilon_0\,\nabla^2\partial_t V = 0.

Why this is stronger than it looks: the right side vanished identically — for any potentials whatsoever, in any gauge, before any equation is solved. A source with $\partial_t\rho + \nabla\cdot\mathbf{J} \neq 0$ makes the system (4) inconsistent: no $(V,\mathbf{A})$ exists for it. Continuity is not checked after the fact; it is the admission ticket.

Step 4. The symmetry reading. In the Lagrangian formulation this identity is Noether's theorem for the gauge symmetry: because the action is blind to $\lambda$ , the source it couples to must be conserved. The full four-dimensional version — one line of tensor antisymmetry, $\partial_\nu\partial_\mu F^{\mu\nu}=0$ — belongs to the relativity chapter; the three-dimensional grind above is the same fact wearing work clothes.

Limiting cases

case	result	meaning
static $\lambda=\lambda(\mathbf{r})$	$\mathbf{A}\to\mathbf{A}+\nabla\lambda$ , $V$ untouched	the magnetostatic freedom of the E4 twin — the electrodynamic transformation's sleeping limit
constant $\lambda$	nothing moves	the identity element: gauge transformations form a group under addition of gauge functions
$\lambda = \lambda_0(\mathbf{r})\,t$	$V\to V-\lambda_0$ , $\mathbf{A}\to\mathbf{A}+t\,\nabla\lambda_0$	a constant-in-time offset of $V$ traded for a linearly growing $\mathbf{A}$ — the zero-of-potential convention is itself a gauge choice
$\lambda$ with $\nabla\lambda=0$ , $\partial_t\lambda=$ const	$V\to V-$ const	the familiar "potentials are defined up to a constant" of circuit theory is the tiniest corner of gauge freedom
covariant packaging	$A^\mu\to A^\mu+\partial^\mu\lambda$ , $A^\mu=(V/c,\mathbf{A})$	both shifts are one four-gradient; the relativity chapter's four-potential makes the choreography a single statement

Common trapQuoting the value of V or A at a point is not physics

The potentials are not measurable and not unique: before a gauge is fixed, "the vector potential here is 107 V·s/m" has no more content than "the longitude here is 40°" before a prime meridian is chosen. Two errors follow from forgetting this. First, mixing gauges mid-calculation: the formula $\mathbf{E}=-\nabla V-\partial_t\mathbf{A}$ holds in every gauge, but $V$ from a Coulomb-gauge solution combined with $\mathbf{A}$ from a Lorenz-gauge solution computes garbage. Second, reading gauge artifacts as physics: in the Coulomb gauge $V$ updates instantaneously across the universe — a statement about the coordinates, not about signal speed, as the next article prosecutes in detail. The discipline: fix the gauge once, compute everything in it, and let only gauge-invariant quantities out the door.

Worked examples

1 · A static point charge with zero scalar potential (confidence builder)

Setup. Nothing dramatizes non-uniqueness like describing the most familiar field in physics with the "wrong" potential. Claim: the pair

V'(\mathbf{r},t) = 0, \qquad \mathbf{A}'(\mathbf{r},t) = -\frac{1}{4\pi\varepsilon_0}\frac{q\,t}{r^2}\,\hat{\mathbf{r}}

describes the static Coulomb field of a point charge $q$ at the origin — no scalar potential at all, and a vector potential that grows forever.

Step 1. Compute the fields. The electric field: $\mathbf{E} = -\nabla V' - \partial_t\mathbf{A}' = -\partial_t\mathbf{A}' = \frac{1}{4\pi\varepsilon_0}\frac{q}{r^2}\hat{\mathbf{r}}$ — the Coulomb field, exactly. The magnetic field: $\mathbf{A}'$ is radial with a spherically symmetric magnitude, and the curl of such a field vanishes (its circulation vanishes on every closed loop: on a loop lying in a sphere $\mathbf{A}'\cdot d\boldsymbol{\ell}=0$ because $\mathbf{A}'$ is radial, which kills the radial component of the curl; on a loop in a plane through the origin the contributions are $f(r)\,dr$ , an exact differential whose integral around any closed path is zero, which kills the tangential components), so $\mathbf{B} = \nabla\times\mathbf{A}' = \mathbf{0}$ . Both fields correct, for all $t$ .

Step 2. Find the gauge function connecting this to the standard description $(V = \frac{q}{4\pi\varepsilon_0 r},\ \mathbf{A}=\mathbf{0})$ . We need $V' = V - \partial_t\lambda = 0$ , so $\partial_t\lambda = V$ , giving

\lambda(\mathbf{r},t) = \frac{1}{4\pi\varepsilon_0}\frac{q\,t}{r}.

Check the vector shift it induces: $\nabla\lambda = \frac{qt}{4\pi\varepsilon_0}\nabla(1/r) = -\frac{qt}{4\pi\varepsilon_0 r^2}\hat{\mathbf{r}} = \mathbf{A}' - \mathbf{A}$ . ✓ Both shifts, one $\lambda$ , as Derivation B guarantees.

Step 3. Numbers, to feel the gauge dependence. Take $q = 1$ nC, $r = 0.5$ m, $t = 3$ s. The physical field is $E = kq/r^2 = (8.99\times10^9)(10^{-9})/0.25 = \mathbf{36.0\ V/m}$ in both descriptions. The bookkeeping wildly disagrees: the standard gauge says $V = 18.0$ V, $A = 0$ ; the exotic gauge says $V' = 0$ , $A' = 108\ \text{V}\,\cdot\,\text{s/m}$ — and growing by $36$ V·s/m every second. An instrument that could read " $\mathbf{A}$ at a point" would see a runaway; every real instrument reads $\mathbf{E}$ and sees a time-independent 36 V/m forever.

The lesson: "the electric field is static" is physics; "the potentials are static" is a gauge choice. Even which potential does the work — $V$ in one description, $\mathbf{A}$ in the other — is coordinates.

2 · Length gauge versus velocity gauge — the transformation AMO physics uses daily (exam level)

Setup. In strong-field atomic physics, a laser field at the atom is modeled (in the dipole approximation — the field is taken uniform over the atom, honest for atoms far smaller than the wavelength) as $\mathbf{E}(t) = E_0\cos(\omega t)\,\hat{\mathbf{z}}$ . Two standard potential descriptions coexist in the literature. Velocity gauge: $V_1 = 0$ , $\mathbf{A}_1 = -\tfrac{E_0}{\omega}\sin(\omega t)\,\hat{\mathbf{z}}$ . Length gauge: $V_2 = -E_0\,z\cos(\omega t)$ , $\mathbf{A}_2 = \mathbf{0}$ . Verify both, find the connecting $\lambda$ , and put numbers on a real laser.

Step 1. Verify the velocity gauge: $\mathbf{E} = -\partial_t\mathbf{A}_1 = \tfrac{E_0}{\omega}\,\omega\cos(\omega t)\,\hat{\mathbf{z}} = E_0\cos(\omega t)\,\hat{\mathbf{z}}$ . ✓ And $\mathbf{B} = \nabla\times\mathbf{A}_1 = 0$ since $\mathbf{A}_1$ is uniform in space.

Step 2. Verify the length gauge: $\mathbf{E} = -\nabla V_2 = E_0\cos(\omega t)\,\hat{\mathbf{z}}$ . ✓ Here the field is delivered by a quasi-static-looking potential ramp — the "length" name comes from the interaction energy $qE_0 z$ it hands a charge.

Step 3. The connecting gauge function. Need $\nabla\lambda = \mathbf{A}_2 - \mathbf{A}_1 = \tfrac{E_0}{\omega}\sin(\omega t)\,\hat{\mathbf{z}}$ , so $\lambda = \tfrac{E_0}{\omega}\,z\sin(\omega t)$ . Check the scalar shift: $V_2 = V_1 - \partial_t\lambda = -E_0 z\cos(\omega t)$ . ✓ One $\lambda$ , both shifts — this is the Göppert-Mayer transformation of AMO theory.

Step 4. Numbers. A Ti:sapphire pulse at intensity $10^{14}\ \mathrm{W/cm^2}$ , wavelength 800 nm: $E_0 = \sqrt{2I/c\varepsilon_0} = 2.74\times10^{10}$ V/m and $\omega = 2.35\times10^{15}$ rad/s. The velocity-gauge amplitude is $A_0 = E_0/\omega = 1.17\times10^{-5}$ V·s/m; the length-gauge potential across one ångström is $V_2^{\max} = E_0 z = \mathbf{2.74\ V}$ — an energy swing of 2.74 eV for an electron, nearly twice the 1.55 eV photon energy, which is why $10^{14}\ \mathrm{W/cm^2}$ tears atoms apart. Any ionization rate, being physical, must come out gauge-invariant; truncated approximations need not, and chasing that discrepancy is a live methods literature.

Honesty note. The uniform oscillating $\mathbf{E}$ with $\mathbf{B}=0$ is not an exact vacuum Maxwell solution (Ampère–Maxwell would demand a curl of $\mathbf{B}$ ) — it is the leading term of the true laser field for $z \ll \lambda_{\text{laser}} = 800$ nm $\gg$ 1 Å. The gauge transformation itself is exact; only the field being described is approximate.

Going further

References. Griffiths, Introduction to Electrodynamics, 4th ed., §10.1.2–10.1.3 · Jackson, Classical Electrodynamics, 3rd ed., §6.2–6.3 · Zangwill, Modern Electrodynamics, §15.2–15.3 · Aharonov & Bohm, Phys. Rev. 115, 485 (1959) Key results in this article: the gauge transformation $\mathbf{A}\to\mathbf{A}+\nabla\lambda$ , $V\to V-\partial\lambda/\partial t$ leaves $\mathbf{E},\mathbf{B}$ invariant via $\nabla\times\nabla\lambda=0$ and the commutation of $\nabla$ with $\partial_t$ ; the wave equation for reaching the Lorenz gauge is $\Box\lambda=-(\nabla\cdot\mathbf{A}+\mu_0\varepsilon_0\partial_t V)$ ; covariantly, $F^{\mu\nu}=\partial^\mu A^\nu-\partial^\nu A^\mu$ is invariant under $A^\mu\to A^\mu+\partial^\mu\lambda$ , and $\partial_\nu J^\nu=0$ follows from the antisymmetry of $F^{\mu\nu}$ . All expressions use standard SI with $\mu_0,\varepsilon_0,c$ . The Lorenz-gauge repair equation and the covariant tensor statements are treated at statement level here, with full derivations in the companion articles. Worked-example numbers: 36.0 V/m; $E_0 = 2.74\times10^{10}$ V/m, $A_0 = 1.17\times10^{-5}$ V·s/m, 2.74 eV vs 1.55 eV.

Builds on: Scalar and Vector Potentials · Maxwell's Equations in Vacuum · Vector Identities and Product Rules · Gauge Freedom and the Coulomb Gauge (magnetostatics) Leads to: Coulomb Gauge · Lorenz Gauge · Inhomogeneous Wave Equations for Potentials · Retarded Potentials

If potentials are pure bookkeeping, why does the Aharonov–Bohm effect see A where B is zero?

Because what the electron reads is not the pointwise value of $\mathbf{A}$ (gauge-dependent, meaningless) but the closed-loop integral $\oint\mathbf{A}\cdot d\boldsymbol\ell$ (gauge-invariant, equal to the enclosed flux). Outside a long solenoid $\mathbf{B}=0$ but the loop integral around it equals the confined flux — a nonlocal, invariant fact about the region's topology. The effect therefore does not contradict gauge freedom; it sharpens the doctrine. The invariant content of the potentials is more than the local fields (it includes flux-around-holes information), while the gauge-variant part remains exactly as unphysical as this article says. Note the fine print in Derivation B: curl-free implies gradient only on simply connected regions, and the AB ring is precisely not one.

How is this article different from the gauge article back in magnetostatics?

Division of labor. The magnetostatics twin (Gauge Freedom and the Coulomb Gauge) owns the static story: $\mathbf{A}\to\mathbf{A}+\nabla\lambda$ with $V$ untouched, the converse theorem for curl-free fields with its simply-connected fine print, the attainability of $\nabla\cdot\mathbf{A}=0$ by solving a Poisson problem, and the uniform-field menagerie. This article owns the electrodynamic structure: the coordinated $V$ shift and why its sign is forced (Derivation A), the completeness converse when everything depends on time — two applications of the static theorem glued by a single $\lambda$ (Derivation B) — the gauge-orbit/slice picture, the gauge menu, and the continuity identity (Derivation C). The static twin's freedom is this article's transformation with $\partial_t\lambda = 0$ : one limiting row of the table.

Does the gauge function itself satisfy any equation?

None whatsoever, and that is the point: $\lambda$ is arbitrary — any smooth scalar field of space and time generates a legal transformation. Equations for $\lambda$ appear only when you want something: to move a given potential pair onto a chosen slice. Demanding the Lorenz condition of the transformed pair forces $\Box\lambda = -(\nabla\cdot\mathbf{A}+\mu_0\varepsilon_0\partial_t V)$ — a wave equation with known source, always solvable, as the Lorenz article proves. Demanding the Coulomb condition forces a Poisson equation. In each case the existence of solutions is what makes the gauge attainable, and the multiplicity of solutions (homogeneous solutions of the same operator) is exactly the residual freedom that survives.

If lambda is arbitrary, could I use a discontinuous or multivalued one?

Only at your peril, and sometimes at your profit. The derivations assumed $\lambda$ twice differentiable so that curls of gradients vanish and derivatives commute; a kink would inject delta-function artifacts into the potentials that must then cancel in the fields — legal with distributions, pointless in most problems. Multivalued gauge functions are genuinely useful in multiply connected regions: the standard potential descriptions outside a flux-carrying solenoid differ by exactly such a $\lambda$ , and in quantum mechanics the demand that the matter phase $e^{iq\lambda/\hbar}$ stay single-valued quantizes the allowed multivaluedness — the germ of flux quantization in superconductors. Within this chapter, take $\lambda$ smooth and single-valued.

Why do we say charge conservation is a partner of gauge symmetry rather than a consequence of Maxwell's equations?

Both statements are true and illuminate different faces. The Maxwell chapter derived continuity by combining the two source equations — a consequence. Derivation C shows something sharper: in the potential formulation the left-hand sides identically satisfy the continuity combination, so a non-conserving source cannot even be coupled to potentials — the theory refuses to be written. In the Lagrangian formulation this becomes Noether's second theorem: local gauge invariance of the action forces the coupled matter current to be conserved before any equation of motion is imposed. "Partner" is the honest word — neither causes the other; both are shadows of one structure, which is what makes promoting gauge invariance to a principle so powerful: conservation laws come bundled.

In the worked example, A grows without bound as t increases. Is that not unphysical?

It is un-measurable, which is the only standard potentials answer to. Every field computed from the runaway $\mathbf{A}'$ is the static Coulomb field at all times; no experiment distinguishes the descriptions. The runaway does violate aesthetic side conditions people impose silently — potentials that stay bounded, vanish at infinity, share the fields' time-independence — and those side conditions are precisely what a good gauge choice enforces. The practical moral: gauge freedom permits monstrous descriptions, gauge fixing is how you decline them.

Do the two remaining physical degrees of freedom mean the photon has two polarizations?

Yes — the counting survives quantization intact. Classically: four potential components, minus one function removable by $\lambda$ , minus one gauge condition, leaves two functional degrees of freedom, which for radiation are the two transverse polarization amplitudes of the wave chapter. Quantum mechanically the same ledger says the photon has two polarization states, not three or four — the massless vector particle's little-group representation, in field-theory language. The counting is not a coincidence: gauge redundancy is precisely how a manifestly local, four-component description coexists with a two-state physical particle. A massive photon would break the ledger — it gains a longitudinal state, and gauge freedom is lost (current experimental bound on the photon mass: below $10^{-54}$ kg, from solar-wind magnetohydrodynamics and torsion-balance work).