The Magnetic Vector Potential

Definition and the defining equation

Gauss's law for magnetism states that the magnetic field is divergence-free everywhere:

∇ · B = 0

A theorem of vector calculus (Helmholtz decomposition) says that any smooth, divergence-free field can be written as the curl of some other field. We name that field the magnetic vector potential A and define:

B = ∇ × A

where the symbols mean:

• B — the magnetic flux density (magnetic field), units tesla (T = kg·s⁻²·A⁻¹).
• A — the magnetic vector potential, a vector field, units tesla·metre (T·m), equivalently V·s/m or Wb/m.
• ∇ × — the curl operator, which measures the local circulation of a vector field.

The construction is self-consistent because the divergence of any curl is identically zero: ∇ · (∇ × A) = 0. So writing B = ∇ × A guarantees ∇ · B = 0 automatically — the absence of magnetic monopoles is built into the very existence of A.

Why the vector potential matters

There are three distinct reasons physicists use A rather than B directly.

1. It simplifies solving for B. In magnetostatics, the field of a steady current density J is given (in the Coulomb gauge) by an integral that mirrors the ordinary Coulomb potential:

A(r) = (μ₀ / 4π) ∫ J(r′) / |r − r′| d³r′

There is no cross product in this integrand — A simply points in the direction of the current that produces it. Compare this with the Biot–Savart law for B, which carries a J × (r − r′) numerator. It is usually far easier to compute A and then take one curl than to attack B head-on.

2. It packages Maxwell's equations. Introducing the scalar potential φ through E = −∇φ − ∂A/∂t reduces the four Maxwell equations to two equations for the potentials (φ, A). In the Lorenz gauge these become uncoupled wave equations, which is the natural language of electromagnetic radiation.

3. It is the object quantum mechanics couples to. The Schrödinger, Pauli and Dirac equations all couple to the potentials via minimal coupling, not to E and B directly. This is not a mere convenience — the Aharonov–Bohm effect proves that A carries measurable physics that B cannot express locally.

Gauge freedom — A is not unique

The vector potential is defined only up to the gradient of an arbitrary scalar function f(r, t). Consider the transformation:

A → A′ = A + ∇f

Taking the curl of both sides, B′ = ∇ × A′ = ∇ × A + ∇ × (∇f) = ∇ × A = B, because the curl of a gradient is identically zero. So A and A′ produce exactly the same magnetic field. This non-uniqueness is called gauge freedom, and the family of A's giving the same B is a gauge orbit.

To keep the scalar potential φ unchanged as well (so that E is also invariant), the full gauge transformation is:

A → A + ∇f      φ → φ − ∂f/∂t

Every measurable quantity — B, E, forces, energies, scattering cross sections — is invariant under this transformation. This gauge invariance is one of the deepest principles in physics: promoting it to a local symmetry is exactly the logic that generates the electromagnetic, weak, and strong interactions in the Standard Model.

Fixing the gauge: Coulomb vs Lorenz

Because A is underdetermined, we impose an extra condition on its divergence to make a specific problem well-posed. The two standard choices are compared below.

Property	Coulomb (transverse) gauge	Lorenz gauge
Condition	∇·A = 0	∇·A + (1/c²)∂φ/∂t = 0
Scalar potential equation	∇²φ = −ρ/ε₀ (instantaneous Poisson)	□φ = −ρ/ε₀ (retarded wave eq.)
Vector potential equation	∇²A = −μ₀J_transverse	□A = −μ₀J
Lorentz covariance	Not manifest (frame-dependent)	Manifestly covariant
Best used for	Magnetostatics, bound states, QM atoms	Radiation, antennas, relativity, QED
Named for	Charles-Augustin de Coulomb	Ludvig Lorenz (1867), not H. A. Lorentz

Here □ = ∇² − (1/c²)∂²/∂t² is the d'Alembertian (wave) operator, ρ is charge density (C/m³), J is current density (A/m²), ε₀ = 8.854 × 10⁻¹² F/m, μ₀ = 1.25663706 × 10⁻⁶ T·m/A, and c = 1/√(ε₀μ₀) = 2.998 × 10⁸ m/s. A common exam trap: the Lorenz gauge is named after the Danish physicist Ludvig Lorenz, distinct from the Dutch Hendrik Lorentz of the Lorentz force and transformations.

Canonical momentum and minimal coupling

For a particle of charge q and mass m moving in an electromagnetic field, the Lagrangian is L = ½mv² + qv·A − qφ. The momentum conjugate to position is therefore not the kinetic momentum mv but the canonical momentum:

p = ∂L/∂v = mv + qA

Solving for the kinetic momentum gives mv = p − qA, so the Hamiltonian of a charged particle is:

H = (p − qA)² / 2m  +  qφ

where every symbol is: p the canonical momentum (kg·m/s), q the charge (C), A the vector potential (T·m), m the mass (kg), φ the scalar potential (V). This replacement p → p − qA is called minimal coupling. Quantizing by p → −iħ∇ turns it into the operator substitution −iħ∇ → −iħ∇ − qA, which is precisely how A enters the Schrödinger equation, the Landau-level spectrum of a charge in a uniform B, and the Aharonov–Bohm phase. Because it is the canonical momentum (not mv) that is conjugate to position, it is p — not the kinetic momentum — that is conserved when the system is translationally symmetric.

The Aharonov–Bohm effect: where A becomes real

Consider a long, thin solenoid carrying flux Φ. Outside an ideal solenoid the magnetic field B is exactly zero, yet the vector potential A circulates around it and is nonzero everywhere. In 1959, Yakir Aharonov and David Bohm predicted that an electron beam split around such a solenoid would show an interference shift even though it never enters any region of nonzero B. The phase difference between the two paths is:

Δφ = (q/ħ) ∮ A · dl = qΦ/ħ

where Φ = ∮ A · dl is the enclosed magnetic flux (units Wb = T·m²), q is the electron charge (−e), and ħ = 1.055 × 10⁻³⁴ J·s. The line integral ∮ A · dl is gauge-invariant even though A itself is not, because a gauge change adds ∮ ∇f · dl = 0 around a closed loop. Akira Tonomura's group at Hitachi confirmed the effect experimentally in 1986 using electron holography with a shielded toroidal magnet, closing the loophole of stray fields. The message is stark: the electron responds to a region it never visits, through A. B alone, evaluated along the electron's path, is zero and cannot account for the result.

Worked example: A of a long straight wire

Take an infinite straight wire along the z-axis carrying steady current I. By symmetry A points along z and depends only on the cylindrical radius s. In the Coulomb gauge the vector potential is:

A_z(s) = −(μ₀ I / 2π) · ln(s / s₀)

where s₀ is an arbitrary reference radius (its choice is pure gauge — it shifts A by a constant, leaving B alone). Taking the curl in cylindrical coordinates, B = ∇ × A has only an azimuthal component:

B_φ = −∂A_z/∂s = μ₀ I / (2π s)

which is exactly Ampère's law for a wire. Notice how the vector potential points along the current (simple, one component) while B wraps around it (azimuthal) — the direction bookkeeping is handled entirely by the single curl.

Vector potential of common sources

Source	Vector potential A	Notes
Infinite straight wire, current I	A_z = −(μ₀I/2π)·ln(s/s₀)	Along wire; curl gives B = μ₀I/2πs
Magnetic dipole m	A = (μ₀/4π)·(m × r̂)/r²	Falls off as 1/r²; azimuthal
Uniform field B₀ ẑ (symmetric gauge)	A = ½ B₀ (−y, x, 0)	Used for Landau levels
Uniform field B₀ ẑ (Landau gauge)	A = (0, B₀x, 0)	Same B, different A — gauge choice
Ideal solenoid, flux Φ (outside)	A_φ = Φ/(2πs)	B = 0 outside, yet A ≠ 0 → Aharonov–Bohm
Plane EM wave	A = A₀ cos(k·r − ωt)	E = −∂A/∂t, B = ∇ × A

The two rows for a uniform field make gauge freedom concrete: the symmetric and Landau gauges give completely different-looking A yet identical B = B₀ẑ. They differ by the gradient of f = ½B₀xy.

JavaScript — vector potential and gauge checks

const mu0 = 1.25663706212e-6;  // T·m/A
const hbar = 1.054571817e-34;  // J·s
const e = 1.602176634e-19;     // C

// A_z of an infinite straight wire (Coulomb gauge), s in metres
function wireA(I, s, s0 = 1) {
  return -(mu0 * I / (2 * Math.PI)) * Math.log(s / s0);  // tesla·metre
}

// B from the wire's A: B_phi = -dA_z/ds = mu0 I / (2 pi s)
function wireB(I, s) {
  return mu0 * I / (2 * Math.PI * s);  // tesla
}

console.log(`Wire A_z at s=0.05 m, I=10 A: ${wireA(10, 0.05).toExponential(3)} T·m`);
console.log(`Wire B at s=0.05 m, I=10 A:   ${wireB(10, 0.05).toExponential(3)} T`);  // ~4.0e-5 T

// Gauge invariance: adding grad f leaves B unchanged.
// Uniform field B0 z: symmetric A = 0.5*B0*(-y, x, 0), Landau A = (0, B0*x, 0)
// Both give curl A = (0,0,B0). They differ by grad(0.5*B0*x*y).
function curlZ_uniform(B0) { return B0; }  // both gauges yield the same Bz
console.log(`Both gauges give Bz = ${curlZ_uniform(0.3)} T`);

// Aharonov–Bohm phase from enclosed flux Phi (weber)
function ahaBohmPhase(Phi, charge = -e) {
  return (charge / hbar) * Phi;  // radians
}

// One flux quantum for an electron: Phi0 = h/e -> phase = 2*pi
const h = 2 * Math.PI * hbar;
const Phi0 = h / e;  // ~4.136e-15 Wb
console.log(`Flux quantum Phi0 = ${Phi0.toExponential(3)} Wb`);
console.log(`AB phase for Phi0: ${(ahaBohmPhase(Phi0) / Math.PI).toFixed(2)} pi rad`);  // -2 pi

// Canonical vs kinetic momentum: p = m v + q A
function canonicalMomentum(m, v, q, A) {
  return m * v + q * A;  // per component, kg·m/s
}

Where the vector potential shows up

• Magnetostatics. Computing B for coils, solenoids and dipoles is easier via A because the source integral has no cross product.
• Radiation theory. Antenna fields, Liénard–Wiechert potentials and the Poynting flux are all derived from retarded A in the Lorenz gauge.
• Quantum mechanics. Minimal coupling p → p − qA gives Landau levels, the quantum Hall effect, and the Aharonov–Bohm phase.
• Superconductivity. The London equation J = −(1/μ₀λ²)A ties the supercurrent directly to A and produces the Meissner effect; flux is quantized in units h/2e.
• Gauge field theory. A is the prototype gauge field; its local U(1) symmetry generalizes to the SU(2)×U(1) and SU(3) fields of the Standard Model.
• Berry phase and topology. The AB phase is the archetype of a geometric phase, connecting electromagnetism to topological insulators and Berry curvature.

Common misconceptions

• "A is just math with no physics." True classically, but the Aharonov–Bohm effect (confirmed 1986) shows A produces measurable phase shifts where B = 0. In quantum mechanics it is physical.
• "A is unique." No — A → A + ∇f leaves B unchanged. Any statement about "the" vector potential must specify a gauge.
• "The Lorenz gauge is named after H. A. Lorentz." It is named after Ludvig Lorenz (no 't'), a different physicist. The Lorentz force and transformations are Hendrik Lorentz's.
• "Canonical momentum equals mv." For a charged particle p = mv + qA. Confusing the two corrupts the Hamiltonian and the conservation laws.
• "∮A·dl is gauge-dependent because A is." The closed-loop line integral equals the enclosed flux Φ and is gauge-invariant, since ∮∇f·dl = 0. That is why the AB phase is observable.
• "You need magnetic monopoles to source A." No — steady currents source A. Monopoles would instead break the global existence of A (B could no longer be a pure curl everywhere).