Magnetic Vector Potential and Gauge Freedom

Statement

Because the magnetostatic field satisfies \(\nabla\cdot\mathbf{B}=0\) everywhere, it can be written as the curl of a vector potential, \(\mathbf{B}=\nabla\times\mathbf{A}\). This \(\mathbf{A}\) is fixed only up to the gradient of an arbitrary scalar, \(\mathbf{A}\to\mathbf{A}+\nabla\chi\) (gauge freedom). Imposing the Coulomb gauge \(\nabla\cdot\mathbf{A}=0\) reduces the magnetostatic Ampère law \(\nabla\times\mathbf{B}=\mu_0\mathbf{J}\) to a vector Poisson equation \(\nabla^2\mathbf{A}=-\mu_0\mathbf{J}\).

Why it matters

The vector potential converts a curl equation for \(\mathbf{B}\) into a Poisson equation for \(\mathbf{A}\) — the same operator already solved once for electrostatics — so the entire machinery of the Coulomb problem (Green's functions, multipole expansion, uniqueness theorems) transfers directly to magnetostatics. Each Cartesian component of \(\mathbf{A}\) then follows from a single integral rather than a set of coupled curl equations.

Gauge freedom is not a defect but the first appearance of a principle that organises all of modern physics: the potentials carry redundant information, and the observable content lives only in gauge-invariant combinations. The same \(\mathbf{A}\) becomes a dynamical variable in electrodynamics, the object that couples to the quantum phase of a charged particle in the Aharonov–Bohm effect, and the template for Yang–Mills gauge theory.

Assumptions

Derivation

Result

Limiting cases

• Source-free region (\(\mathbf{J}=0\)): the Coulomb-gauge equation becomes Laplace's equation \(\nabla^2\mathbf{A}=0\); each component is harmonic, mirroring vacuum electrostatics.
• Uniform field \(\mathbf{B}=B\hat{\mathbf{z}}\): a valid Coulomb-gauge potential is \(\mathbf{A}=\tfrac12\mathbf{B}\times\mathbf{r}=\tfrac12 B s\,\hat{\boldsymbol{\phi}}\), symmetric and divergence-free.
• Localised current, far field: expanding the integral for \(\mathbf{A}\) gives the magnetic dipole term \(\mathbf{A}=\frac{\mu_0}{4\pi}\frac{\mathbf{m}\times\hat{\mathbf{r}}}{r^2}\), the leading multipole once the monopole term vanishes.
• Static limit of electrodynamics: dropping \(\partial^2\mathbf{A}/\partial t^2\) in the Lorenz-gauge wave equation \(\Box\mathbf{A}=-\mu_0\mathbf{J}\) returns exactly \(\nabla^2\mathbf{A}=-\mu_0\mathbf{J}\).

Breaks when

Failure modes

• Treating \(\mathbf{A}\) as physical/unique. Reporting "the value of \(\mathbf{A}\)" without a gauge or reference point — \(\mathbf{A}\) is defined only up to \(\nabla\chi\), so only \(\nabla\times\mathbf{A}\) and closed-loop integrals of \(\mathbf{A}\) are meaningful.
• Dropping \(\nabla(\nabla\cdot\mathbf{A})\) before choosing the gauge. Writing \(\nabla\times(\nabla\times\mathbf{A})=-\nabla^2\mathbf{A}\) as if it were an identity; it holds only after \(\nabla\cdot\mathbf{A}=0\) is imposed.
• Sign slip in the Poisson equation. Writing \(\nabla^2\mathbf{A}=+\mu_0\mathbf{J}\); the identity puts \(-\nabla^2\mathbf{A}\) on the current side, so the physical equation carries the minus sign, exactly like \(\nabla^2 V=-\rho/\varepsilon_0\).
• Componentwise Laplacian in curvilinear coordinates. Applying \(\nabla^2 A_\phi=-\mu_0 J_\phi\) in cylindrical coordinates and forgetting the \(-A_\phi/s^2\) and cross terms from the rotating basis; \(\nabla^2\mathbf{A}\) is the vector Laplacian.
• Using \(\nabla\times\mathbf{B}=\mu_0\mathbf{J}\) in a time-varying problem. Omitting the displacement current and then "deriving" a static Poisson equation that no longer applies.
• Confusing gauge fixing with a physical constraint. Believing \(\nabla\cdot\mathbf{A}=0\) restricts the physics; it only labels the redundancy — the measurable \(\mathbf{B}\) is identical in any gauge.

Discussion

The deep point is that \(\nabla\cdot\mathbf{B}=0\) is precisely the integrability condition guaranteeing that a potential exists: just as an irrotational field (\(\nabla\times\mathbf{E}=0\)) is a gradient, a solenoidal field (\(\nabla\cdot\mathbf{B}=0\)) is a curl. Electrostatics and magnetostatics are the two halves of the Helmholtz decomposition, and each Maxwell divergence/curl constraint is what licenses one potential. That symmetry is why the same Poisson operator, the same Green's function, and the same uniqueness theorems recur. Taking the curl of the integral solution reproduces the Biot–Savart law, closing the loop with the prior result used in step 3.

The Coulomb gauge is a choice of convenience, not of nature. Its virtue is that it makes \(\mathbf{A}\) transverse (\(\nabla\cdot\mathbf{A}=0\) means \(\mathbf{k}\cdot\tilde{\mathbf{A}}=0\) in Fourier space) and produces an instantaneous Poisson equation; its cost is that \(\mathbf{A}\) is then not manifestly relativistic. The Lorenz gauge \(\nabla\cdot\mathbf{A}+\mu_0\varepsilon_0\,\partial V/\partial t=0\) trades that instantaneity for a Lorentz-covariant wave equation. Neither changes a single measurable field — the freedom to choose is exactly the gauge freedom exposed in step 2.

Physically, \(\mathbf{A}\) is the field that enters the canonical momentum \(\mathbf{p}=m\mathbf{v}+q\mathbf{A}\), and its line integral is the enclosed magnetic flux, \(\oint\mathbf{A}\cdot d\boldsymbol{\ell}=\Phi_B\). This is why a solenoid produces a nonzero \(\mathbf{A}\) outside itself even though \(\mathbf{B}=0\) there — a fact with no analogue for \(\mathbf{B}\) alone.

Viewed geometrically, \(\mathbf{A}\) is a connection on a \(U(1)\) fibre bundle and \(\mathbf{B}\) (more precisely the two-form \(F=d\mathbf{A}\)) is its curvature; a gauge transformation \(\mathbf{A}\to\mathbf{A}+\nabla\chi\) is a change of local phase convention \(\psi\to e^{iq\chi/\hbar}\psi\) for a charged wavefunction. The Aharonov–Bohm effect shows that the loop integral \(\frac{q}{\hbar}\oint\mathbf{A}\cdot d\boldsymbol{\ell}=\frac{q}{\hbar}\Phi_B\) — gauge invariant — is genuinely physical even where \(\mathbf{B}=0\). Demanding that this local phase symmetry be dynamical is the organising principle from which the full structure of electromagnetism, and by generalisation to non-abelian groups the whole Standard Model, is reconstructed.

Common misconceptions. (i) "\(\mathbf{A}=0\) means no field" — false; conversely a pure gauge \(\mathbf{A}=\nabla\chi\) has \(\mathbf{B}=0\) yet nonzero \(\mathbf{A}\). (ii) "The Coulomb gauge is more correct" — no gauge is more correct; all give identical \(\mathbf{B}\) and identical forces. (iii) "\(\mathbf{A}\) has no physical reality" — its gauge-invariant loop integrals do, as quantum interference demonstrates.

Worked examples

Example 1 — Vector potential of an infinite straight wire.

Example 2 — Symmetric potential for a uniform field.

Problems

1. Gauge shift. Given the symmetric gauge \(\mathbf{A}=\tfrac{B}{2}(-y,\,x,\,0)\) for \(\mathbf{B}=B\hat{\mathbf{z}}\), find a scalar \(\chi\) that transforms it to the Landau gauge \(\mathbf{A}'=(0,\,Bx,\,0)\), and confirm \(\mathbf{B}\) is unchanged.
   Solution

   Compute the difference componentwise: \(\mathbf{A}'-\mathbf{A}=\big(0-(-\tfrac{B}{2}y),\ Bx-\tfrac{B}{2}x,\ 0\big)=\big(\tfrac{B}{2}y,\ \tfrac{B}{2}x,\ 0\big)\). This equals \(\nabla\chi\) with \(\chi=\tfrac12 Bxy\), since \(\partial_x\chi=\tfrac{B}{2}y\), \(\partial_y\chi=\tfrac{B}{2}x\), \(\partial_z\chi=0\). Both potentials give \(\nabla\times\mathbf{A}=\nabla\times\mathbf{A}'=B\hat{\mathbf{z}}\) because \(\nabla\times(\nabla\chi)=0\). Note \(\nabla\cdot\mathbf{A}'=\partial_y(Bx)=0\), so the Landau gauge is also a Coulomb gauge here.
2. Poisson check. For the infinite wire, \(A_z=-\frac{\mu_0 I}{2\pi}\ln(s/s_0)\) outside the wire. Verify it satisfies \(\nabla^2 A_z=0\) for \(s>0\), consistent with \(\mathbf{J}=0\) there.
   Solution

   \(\nabla^2 A_z=\frac{1}{s}\frac{d}{ds}\!\left(s\frac{dA_z}{ds}\right)\). With \(\frac{dA_z}{ds}=-\frac{\mu_0 I}{2\pi s}\), we get \(s\frac{dA_z}{ds}=-\frac{\mu_0 I}{2\pi}\) (constant), whose \(s\)-derivative is zero. Hence \(\nabla^2 A_z=0\) for all \(s>0\). Correct: outside the wire \(\mathbf{J}=0\), so \(\nabla^2\mathbf{A}=0\). The current appears only as a delta-function source on the axis, or equivalently via the boundary condition at the wire surface.
3. Numerical \(A\) from the integral. A short current element \(I\,d\boldsymbol{\ell}=(5.0\ \text{A})(2.0\ \text{mm})\hat{\mathbf{z}}\) sits at the origin. Estimate \(|\mathbf{A}|\) at a point \(10\ \text{cm}\) away using \(\mathbf{A}\approx\frac{\mu_0}{4\pi}\frac{I\,d\boldsymbol{\ell}}{r}\).
   Solution

   \(\frac{\mu_0}{4\pi}=1.0\times10^{-7}\ \text{T·m·A}^{-1}\). \(I\,d\ell=(5.0)(2.0\times10^{-3})=1.0\times10^{-2}\ \text{A·m}\). \(r=0.10\ \text{m}\). Then \(A=\frac{(1.0\times10^{-7})(1.0\times10^{-2})}{0.10}=1.0\times10^{-8}\ \text{T·m}\), directed along \(\hat{\mathbf{z}}\) (parallel to the current). Units: \(\frac{(\text{T·m·A}^{-1})(\text{A·m})}{\text{m}}=\text{T·m}\). ✓
4. Solenoid exterior. An ideal infinite solenoid, \(n=1000\) turns/m, \(I=2.0\ \text{A}\), radius \(R=2.0\ \text{cm}\), has uniform \(\mathbf{B}=\mu_0 n I\,\hat{\mathbf{z}}\) inside and \(\mathbf{B}=0\) outside. Find \(A_\phi(s)\) for \(s>R\) using \(\oint\mathbf{A}\cdot d\boldsymbol{\ell}=\Phi\).
   Solution

   Interior field \(B=\mu_0 n I=(4\pi\times10^{-7})(1000)(2.0)=2.51\times10^{-3}\ \text{T}\). Enclosed flux \(\Phi=B\pi R^2=(2.51\times10^{-3})\pi(0.020)^2=3.16\times10^{-6}\ \text{Wb}\). By symmetry \(\mathbf{A}=A_\phi(s)\hat{\boldsymbol{\phi}}\), and \(\oint\mathbf{A}\cdot d\boldsymbol{\ell}=2\pi s\,A_\phi=\Phi\) for \(s>R\) (all flux enclosed), so \(A_\phi=\frac{\Phi}{2\pi s}=\frac{B R^2}{2s}\). At \(s=5.0\ \text{cm}\): \(A_\phi=\frac{3.16\times10^{-6}}{2\pi(0.05)}=1.0\times10^{-5}\ \text{T·m}\). Note \(\mathbf{A}\neq0\) outside even though \(\mathbf{B}=0\) — the basis of the Aharonov–Bohm effect.
5. Curl-of-curl identity. Starting from Ampère's law and \(\mathbf{B}=\nabla\times\mathbf{A}\), show explicitly that without the Coulomb gauge the equation for \(\mathbf{A}\) is \(\nabla(\nabla\cdot\mathbf{A})-\nabla^2\mathbf{A}=\mu_0\mathbf{J}\), and state why choosing \(\nabla\cdot\mathbf{A}=0\) is always permitted.
   Solution

   Substitute \(\mathbf{B}=\nabla\times\mathbf{A}\) into \(\nabla\times\mathbf{B}=\mu_0\mathbf{J}\): \(\nabla\times(\nabla\times\mathbf{A})=\mu_0\mathbf{J}\). Apply the identity \(\nabla\times(\nabla\times\mathbf{A})=\nabla(\nabla\cdot\mathbf{A})-\nabla^2\mathbf{A}\), giving \(\nabla(\nabla\cdot\mathbf{A})-\nabla^2\mathbf{A}=\mu_0\mathbf{J}\). Gauge freedom: replacing \(\mathbf{A}\to\mathbf{A}+\nabla\chi\) changes \(\nabla\cdot\mathbf{A}\to\nabla\cdot\mathbf{A}+\nabla^2\chi\) but leaves \(\mathbf{B}\) untouched. To make \(\nabla\cdot\mathbf{A}=0\), solve \(\nabla^2\chi=-\nabla\cdot\mathbf{A}\), a Poisson equation that always has a solution with suitable fall-off. The first term then vanishes, yielding \(\nabla^2\mathbf{A}=-\mu_0\mathbf{J}\). The choice is a labelling of redundancy, not a physical restriction, because every gauge yields the same measurable \(\mathbf{B}\).