Potentials and Gauge Freedom
Maxwell's equations come as four coupled equations for six numbers —
the three components of \mathbf{E} and the three of
\mathbf{B} at every point of space. That is a lot of bookkeeping. Yet two
of those four equations, \nabla\cdot\mathbf{B}=0 and
\nabla\times\mathbf{E}=-\partial\mathbf{B}/\partial t, contain no charges
or currents at all — they are pure statements about the shape the fields are allowed to take.
On this page we do something that at first looks like a bookkeeping trick and turns out to be one of
the deepest ideas in physics: we trade the six field components for four
potentials, a single scalar \phi and a vector
\mathbf{A}, chosen so that those two source-free equations are satisfied
automatically, for free, always.
The price of that convenience is a strange and wonderful freedom. The potentials
(\phi,\mathbf{A}) that produce a given \mathbf{E}
and \mathbf{B} are not unique — a whole infinite family of
them describes exactly the same physics. Sliding between members of that family is called a
gauge transformation, and the freedom to do it is gauge freedom.
That single idea — the potentials carry surplus information that no experiment can pin down —
is what this page is about. It rests on the two source-free
Maxwell equations,
so keep the curl and divergence handy.
Where the potentials come from
Two small facts from vector calculus do all the work. First: a divergence-free field is the
curl of something. Since \nabla\cdot\mathbf{B}=0 everywhere — there
are no magnetic monopoles — we are guaranteed a vector field \mathbf{A}, the
vector potential, with
\mathbf{B} = \nabla\times\mathbf{A}.
This is not an approximation or a special case; it is a theorem. And it automatically satisfies
\nabla\cdot\mathbf{B}=0, because the divergence of any curl is identically
zero: \nabla\cdot(\nabla\times\mathbf{A})=0. One of Maxwell's equations has
just become a definition rather than a constraint.
Now feed \mathbf{B}=\nabla\times\mathbf{A} into Faraday's law,
\nabla\times\mathbf{E}=-\partial\mathbf{B}/\partial t:
\nabla\times\mathbf{E} = -\frac{\partial}{\partial t}\left(\nabla\times\mathbf{A}\right) = -\nabla\times\frac{\partial\mathbf{A}}{\partial t} \;\;\Longrightarrow\;\; \nabla\times\left(\mathbf{E}+\frac{\partial\mathbf{A}}{\partial t}\right)=0.
Second fact: a curl-free field is the gradient of something. The combination
\mathbf{E}+\partial\mathbf{A}/\partial t has zero curl, so it must be minus
the gradient of a scalar \phi, the scalar potential.
Rearranging gives the electric field in terms of both potentials:
-
The magnetic field is the curl of the vector potential:
\mathbf{B}=\nabla\times\mathbf{A}.
-
The electric field gets a piece from each potential:
\mathbf{E}=-\nabla\phi-\frac{\partial\mathbf{A}}{\partial t}.
-
Written this way, \nabla\cdot\mathbf{B}=0 and
\nabla\times\mathbf{E}=-\partial\mathbf{B}/\partial t hold
identically — two of Maxwell's four equations come along for free.
In electrostatics the vector potential is constant in time, so the second term drops and we recover
the familiar \mathbf{E}=-\nabla\phi of the
electric potential. The
-\partial\mathbf{A}/\partial t piece is what's new: it is how a
changing magnetic vector potential drives an electric field — Faraday's law, hiding in plain
sight.
The surplus: gauge freedom
Here is the catch. The vector potential is defined only through its curl, and
curl throws information away: many different \mathbf{A}'s
have the same curl. Specifically, take any scalar function
\chi(\mathbf{r},t) you like and shift the potentials by
\mathbf{A}\;\longrightarrow\;\mathbf{A}' = \mathbf{A}+\nabla\chi, \qquad \phi\;\longrightarrow\;\phi' = \phi-\frac{\partial\chi}{\partial t}.
This paired shift is a gauge transformation, and \chi is
the gauge function. The claim is that the new potentials
(\phi',\mathbf{A}') give exactly the same
\mathbf{E} and \mathbf{B}. Let's check both.
The magnetic field. The curl of a gradient is always zero,
\nabla\times(\nabla\chi)=0, so
\mathbf{B}' = \nabla\times\mathbf{A}' = \nabla\times\mathbf{A} + \nabla\times(\nabla\chi) = \nabla\times\mathbf{A} + 0 = \mathbf{B}.
The electric field. The two new pieces cancel exactly, because gradient and
time-derivative commute:
\mathbf{E}' = -\nabla\phi' - \frac{\partial\mathbf{A}'}{\partial t} = -\nabla\phi + \nabla\frac{\partial\chi}{\partial t} - \frac{\partial\mathbf{A}}{\partial t} - \frac{\partial(\nabla\chi)}{\partial t} = \mathbf{E} + \left(\nabla\frac{\partial\chi}{\partial t} - \frac{\partial(\nabla\chi)}{\partial t}\right) = \mathbf{E}.
So no measurement of \mathbf{E} or \mathbf{B} —
no force on any charge, no reading on any meter — can tell one gauge from another. The potentials carry
surplus information: a whole function's worth of it, one free
\chi at every point of spacetime.
See it: the same field, many potentials
Nothing makes gauge freedom feel real like watching it. Below is a uniform magnetic
field pointing straight out of the screen, \mathbf{B}=B\,\hat{\mathbf{z}}
— the same everywhere, unchanging. The arrows show one choice of vector potential
\mathbf{A} that produces it. Drag the slider: you are adding the gradient of
the gauge function \chi=\tfrac{1}{2}B\,xy, sweeping from the
symmetric gauge (a tidy swirl, \mathbf{A}=\tfrac{1}{2}B(-y,\,x,\,0))
at \lambda=0 to the Landau gauge (a lop-sided shear,
\mathbf{A}=B(0,\,x,\,0)) at \lambda=1.
The two pictures could hardly look more different — one all rotation, the other all shear — and yet if
you compute the curl of either, you get the same answer everywhere:
\nabla\times\mathbf{A}=B\,\hat{\mathbf{z}}. The swirliness that becomes
the magnetic field is identical; only the surplus, the part that curls to nothing, has changed.
That surplus is the gradient \nabla\chi you added.
Spending the freedom: gauge conditions
Gauge freedom is not a nuisance to be tolerated — it is a resource to be spent.
Because we may impose one scalar condition on \mathbf{A} without changing
any physics, we choose the condition that makes our particular problem easiest. Two choices dominate.
The Coulomb gauge demands that the vector potential be divergence-free:
\nabla\cdot\mathbf{A}=0 \qquad\text{(Coulomb gauge).}
With this choice the equation for \phi collapses to
Poisson's
equation, \nabla^2\phi=-\rho/\varepsilon_0, solved
instantaneously by the charge distribution — the scalar potential everywhere responds at once
to the charge, exactly as in electrostatics. That "instantaneous" feel makes the Coulomb gauge
beautifully convenient for statics and slowly varying problems, where it strips the
maths down to the bare minimum.
The Lorenz gauge (after Ludvig Lorenz, not Hendrik Lorentz — a famous near-collision
of names) demands instead
\nabla\cdot\mathbf{A}+\mu_0\varepsilon_0\frac{\partial\phi}{\partial t}=0 \qquad\text{(Lorenz gauge).}
This looks fussier, but it is magic. Substituting it into the two source Maxwell equations
decouples them: \phi and each component of
\mathbf{A} separate cleanly, and each obeys an identical
wave equation with a source on the right,
\Box\,\phi=\frac{\rho}{\varepsilon_0}, \qquad \Box\,\mathbf{A}=\mu_0\mathbf{J}, \qquad \Box\equiv\nabla^2-\mu_0\varepsilon_0\frac{\partial^2}{\partial t^2}.
Four tidy wave equations, one per potential component, sourced by charge and current. This is where
electromagnetic radiation becomes obvious — waves fall right out — and the Lorenz
condition is Lorentz-covariant, meaning it looks the same in every inertial frame, so
it is the natural language for relativity. Two gauges, one physics; you pick the one whose equations
you'd rather solve.
A worked check: a pure-gauge potential does nothing
Suppose someone hands you the potentials
\mathbf{A}=\nabla\chi and \phi=-\partial\chi/\partial t
for some function \chi — a potential built entirely out of a gauge
function, with nothing underneath. What fields does it produce? By our formulas,
\mathbf{B}=\nabla\times\mathbf{A}=\nabla\times(\nabla\chi)=0,
\mathbf{E}=-\nabla\phi-\frac{\partial\mathbf{A}}{\partial t}=\nabla\frac{\partial\chi}{\partial t}-\frac{\partial(\nabla\chi)}{\partial t}=0.
Both fields vanish. A "pure-gauge" potential is physically empty — it is exactly a gauge
transformation of the zero potential, and describes a region with no fields at all. Concretely, take
\chi=k\,x for a constant k. Then
\mathbf{A}=\nabla\chi=(k,0,0) is a uniform, non-zero vector potential
everywhere — and yet \mathbf{B}=\nabla\times\mathbf{A}=0. A constant
\mathbf{A} is invisible. It is a stark reminder that a non-zero potential
need not mean a non-zero field.
For a long time the potentials were regarded as pure scaffolding — handy for calculation, but with the
"real" physics living entirely in \mathbf{E} and
\mathbf{B}. Then quantum mechanics arrived and unsettled the verdict. In the
Aharonov–Bohm effect, a beam of electrons is split around a thin, perfectly shielded
solenoid and recombined. Outside the solenoid the magnetic field \mathbf{B}
is exactly zero — the electrons never touch any field. Yet switching the current in
the solenoid on or off shifts the interference pattern. The electrons "know" about a
magnetic flux they never physically encountered.
What they respond to is the vector potential \mathbf{A}, which is
non-zero outside the solenoid even though its curl there is not. The quantum phase of the electron
picks up \oint\mathbf{A}\cdot d\boldsymbol{\ell} around its path — a
gauge-invariant quantity (it equals the enclosed flux) even though \mathbf{A}
itself is not. So the potentials are more than bookkeeping: in the quantum world they carry genuine,
observable consequences that the fields alone cannot express. The scaffolding turned out to be part of
the building.
The single most common confusion here: the potentials are not unique, but the fields
are. Two people can write down completely different (\phi,\mathbf{A})
for the very same situation — that is gauge freedom — and neither is "more correct". What they must
agree on is \mathbf{E} and \mathbf{B}, because
those are what push on charges and register on instruments.
A specific trap follows: a non-zero \mathbf{A} does not mean a
non-zero magnetic field. We saw a uniform \mathbf{A}=(k,0,0) whose
curl is zero — pure gauge, no field. So never read off "there's a vector potential here, therefore
there's a magnetic field here". Only \nabla\times\mathbf{A} is the field.
And a corollary for problem-solving: quantities like \nabla\cdot\mathbf{A}
or the value of \phi at a point are gauge-dependent — they
change when you change gauge — so any physical answer you compute must come out gauge-independent, or
you've made a slip.
Why bother? Four numbers instead of six
Step back and count the winnings. We replaced six field components with
four potential components (\phi and the three of
\mathbf{A}). Two of Maxwell's four equations are now satisfied
identically, leaving only the two source equations to solve. Choose the Lorenz gauge
and even those decouple into four independent wave equations. Fewer objects, fewer equations, cleaner
structure — and, as the Aharonov–Bohm effect shows, the potentials become indispensable
the moment you step into quantum mechanics, where the whole theory of charged particles and the entire
edifice of gauge field theory (the Standard Model included) is built on
\mathbf{A} and \phi, not on
\mathbf{E} and \mathbf{B}. The humble trick of
introducing potentials turns out to be the doorway to modern physics.