Vector Potentials

The scalar potential compressed curl-free fields into one function. Now for the other half of the Helmholtz split. Magnetic fields obey $\nabla\cdot\mathbf{B}=0$ — always, everywhere, with no exception ever observed — and the theorem's second corollary says a divergence-free field is a curl: there exists a vector potential $\mathbf{A}$ with $\mathbf{B}=\nabla\times\mathbf{A}$ . This article builds that object, and with it closes the vector-calculus chapter by handing the next tiers their working machinery. Be warned: $\mathbf{A}$ plays by stranger rules than $V$ , and the strangeness is the interesting part. The scalar potential was ambiguous by a constant; the vector potential is ambiguous by the gradient of any function whatsoever — $\mathbf{A}$ and $\mathbf{A}+\nabla\lambda$ produce identical fields. This enormous slack, called gauge freedom, looks at first like a defect. It is in fact one of the deepest structural features of physics: "choosing a gauge" is how every practical calculation tames $\mathbf{A}$ , and gauge symmetry is the organizing principle on which the entire Standard Model is built. And $\mathbf{A}$ earns its keep concretely. Its circulation around a loop is the magnetic flux through the loop — one line integral replacing a surface integral. In quantum mechanics, electrons respond to $\mathbf{A}$ even in regions where $\mathbf{B}=0$ (the Aharonov–Bohm effect), settling an old argument about whether $\mathbf{A}$ is "real." E4 will compute with it daily, and E9 will promote it, alongside $V$ , into the four-potential of relativistic electrodynamics. This is the chapter's exit door, and it opens onto everything.

Key takeaways

A divergence-free field is a curl: $\nabla\cdot\mathbf{F}=0$ guarantees a vector potential $\mathbf{A}$ with $\mathbf{F}=\nabla\times\mathbf{A}$ — the magnetostatic charter, since $\nabla\cdot\mathbf{B}=0$ always.
$\mathbf{A}$ is not unique: $\mathbf{A}\to\mathbf{A}+\nabla\lambda$ leaves the field untouched for any scalar $\lambda$ — gauge freedom, the vector analogue of the scalar potential's arbitrary constant.
Fixing the slack is choosing a gauge; in the Coulomb gauge $\nabla\cdot\mathbf{A}=0$ , magnetostatics reduces to Poisson's equation $\nabla^2\mathbf{A}=-\mu_0\mathbf{J}$ , solved by $\mathbf{A}=\frac{\mu_0}{4\pi}\int\frac{\mathbf{J}(\mathbf{r}')}{|\mathbf{r}-\mathbf{r}'|}\mathrm{d}\tau'$ — the same Coulomb-kernel machine as ever.
The flux-circulation identity $\Phi_B=\oint\mathbf{A}\cdot\mathrm{d}\boldsymbol{\ell}$ (via Stokes) makes flux a property of the loop alone; a uniform field has $\mathbf{A}=\tfrac12\mathbf{B}\times\mathbf{r}$ among infinitely many gauge-equivalent choices.
Outside an ideal solenoid $\mathbf{B}=0$ yet $\mathbf{A}\neq0$ — and quantum charges notice (Aharonov–Bohm), so the potential is physical in loops even where the field vanishes.
Key idea: every divergence-free field is the curl of a vector potential, unique only up to a gradient — gauge freedom — and the potential's circulation around any loop equals the field's flux through it.

The core insight

A divergence-free field can always be written as $\mathbf{F}=\nabla\times\mathbf{A}$ ; the vector potential is defined only up to a gradient ( $\mathbf{A}\to\mathbf{A}+\nabla\lambda$ , gauge freedom), and its circulation around any closed loop equals the flux of the field through that loop.

Where this lives today

The vector potential is the native language of quantum devices. In a superconducting ring, the condensate's phase must return to itself around the loop, forcing the enclosed flux — $\oint\mathbf{A}\cdot\mathrm{d}\boldsymbol{\ell}$ — to be quantized in units of $h/2e=2.07\times10^{-15}$ Wb; counting and manipulating those flux quanta is the operating principle of SQUID magnetometers and of the flux qubits in superconducting quantum computers. Every Josephson-junction circuit diagram in a quantum-computing lab is, at bottom, bookkeeping of $\mathbf{A}$ 's loop integrals. In fundamental physics the promotion is total. Quantum mechanics couples matter to electromagnetism through $\mathbf{A}$ (the substitution $\mathbf{p}\to\mathbf{p}-q\mathbf{A}$ ), not through the fields; lattice QCD simulations store the gauge potential's parallel transporters on every link of a spacetime grid; and the Standard Model's three forces are all "gauge fields" — direct generalizations of $\mathbf{A}$ whose gauge freedom, far from being slack to apologize for, dictates the very existence and form of the interactions. The ambiguity this article taught you to manage is, by consensus of a century's physics, the deepest design principle we know. The chapter that began with gradients ends at the doorway of gauge theory — and E4 will now put $\mathbf{A}$ to work on honest magnets.

The physical picture

The scalar potential was an altitude map; the vector potential wants a different metaphor. Since $\mathbf{B}=\nabla\times\mathbf{A}$ , run the curl picture backwards: wherever magnetic flux flows, $\mathbf{A}$ circulates around it, the way water circulates around the core of a whirlpool. Around a solenoid — a concentrated tube of flux — the vector potential runs in rings, strongest near the tube, falling off as $1/s$ outside like the swirl around a bathtub drain. The field $\mathbf{B}$ punches through surfaces; its potential $\mathbf{A}$ wraps around rims. The Stokes theorem is the dictionary between them: flux through equals circulation around.

Figure

The vector potential circulates around magnetic flux — rings of A wrapping a solenoid's field tube

But here is where the story turns strange and modern. Suppose you have one valid $\mathbf{A}$ . Add to it the gradient of any scalar function $\lambda(\mathbf{r})$ : the curl of a gradient vanishes identically, so $\mathbf{A}+\nabla\lambda$ produces exactly the same $\mathbf{B}$ . Not one potential but an infinite family, differing by arbitrary gradient fields — and no experiment measuring $\mathbf{B}$ can tell family members apart. This is gauge freedom. The practical response is to choose: impose one extra condition (a "gauge") that picks a convenient representative, the way choosing sea level pins down altitude. The profound response came later: quantum mechanics found that phases of wavefunctions shift under gauge transformations in a compensating way, and demanding that consistency — locally, for every symmetry — generates the interactions of the Standard Model. The slack in $\mathbf{A}$ is not a bug; it is the blueprint.

Figure

Gauge freedom — two very different vector potentials, one identical magnetic field

Existence, honestly earned

The claim to prove: if $\nabla\cdot\mathbf{F}=0$ everywhere (in a suitably hole-free region — for divergence the requirement is that closed surfaces in the domain bound volumes inside it), then some $\mathbf{A}$ exists with

\mathbf{F}=\nabla\times\mathbf{A} \qquad (1)

where: the converse direction is free of charge — $\nabla\cdot(\nabla\times\mathbf{A})=0$ is an identity, so only divergence-free fields could possibly be curls. Divergence-free is thus the exact admission criterion, mirroring how curl-free was the exact criterion for scalar potentials. One route to existence is already in hand: the Helmholtz construction of the previous article manufactures $\mathbf{W}$ with $\nabla\times\mathbf{W}=\mathbf{F}$ from the field's curl data. But it is worth seeing a bare-hands construction that uses nothing but integration — it shows where the divergence-free condition is spent, and it produces the gauge phenomenon before your eyes.

Derivation ABuilding a vector potential by direct integration

The goal is existence by explicit construction: given $\nabla\cdot\mathbf{F}=0$ , produce a concrete $\mathbf{A}$ with $\nabla\times\mathbf{A}=\mathbf{F}$ . The strategy exploits gauge freedom in advance: since the answer is non-unique anyway, we are entitled to hunt for a convenient representative — try one with $A_z=0$ . That converts the three curl equations into two integrations and one consistency check, and the check is precisely where $\nabla\cdot\mathbf{F}=0$ gets spent. Watch that step: it is the whole theorem.

Step 1. Set $A_z=0$ and write out the curl's components:

F_x=\frac{\partial A_z}{\partial y}-\frac{\partial A_y}{\partial z}=-\frac{\partial A_y}{\partial z},\qquad F_y=\frac{\partial A_x}{\partial z},\qquad F_z=\frac{\partial A_y}{\partial x}-\frac{\partial A_x}{\partial y}

Step 2. Integrate the first two along $z$ from a base plane $z=0$ :

A_x(x,y,z)=\int_0^zF_y(x,y,z')\,\mathrm{d}z',\qquad A_y(x,y,z)=-\int_0^zF_x(x,y,z')\,\mathrm{d}z'+g(x,y)

Why the free function $g$ appears: integrating in $z$ determines $A_y$ only up to a $z$ -independent function — this is gauge slack surfacing already, and we will spend it in Step 4 to satisfy the third equation.

Step 3. Impose the third equation. Compute $\partial_xA_y-\partial_yA_x=-\int_0^z\big(\partial_xF_x+\partial_yF_y\big)\mathrm{d}z'+\partial_xg$ . Now spend the hypothesis: $\nabla\cdot\mathbf{F}=0$ means $\partial_xF_x+\partial_yF_y=-\partial_zF_z$ , so the integral collapses by the fundamental theorem of calculus:

\partial_xA_y-\partial_yA_x=\int_0^z\partial_{z'}F_z\,\mathrm{d}z'+\partial_xg=F_z(x,y,z)-F_z(x,y,0)+\partial_xg

Why this is the crux: without zero divergence, the leftover integral would depend on the whole history of $F$ along $z$ and no choice of $g(x,y)$ — a function of two variables — could absorb it. Divergence-free is exactly the condition that reduces the leftover to a $z$ -independent deficit.

Step 4. Kill the deficit with the free function: choose $g(x,y)=\int_0^xF_z(x',y,0)\,\mathrm{d}x'$ , so $\partial_xg=F_z(x,y,0)$ . Then $\partial_xA_y-\partial_yA_x=F_z$ — all three curl equations hold, and an explicit vector potential exists.

Step 5. Read off the moral about uniqueness: every choice made here — the $A_z=0$ ansatz, the base plane, the constant of the $g$ integration — was arbitrary. Different choices give different $\mathbf{A}$ 's, all with the same curl; any two answers differ by a curl-free (hence gradient) field. Existence and gauge freedom arrive in the same construction. Verified: matches the standard construction (Griffiths Problem 5.31, Zangwill §10.5); test on $\mathbf{F}=B_0\hat{\mathbf{z}}$ : the recipe yields $A_x=0$ , $A_y=B_0x$ — the Landau gauge of the figure, whose curl is indeed $B_0\hat{\mathbf{z}}$ .

Gauge freedom, and how to spend it

The non-uniqueness deserves its own display, because it is the article's second protagonist:

\mathbf{A}'=\mathbf{A}+\nabla\lambda \quad\Longrightarrow\quad \nabla\times\mathbf{A}'=\nabla\times\mathbf{A} \qquad (2)

where: $\lambda(\mathbf{r})$ — any differentiable scalar ("gauge function"). Conversely, two potentials for the same field differ by a curl-free field, which (in simply-connected space) is a gradient — so (2) captures the entire ambiguity. Compare the scalar story: $V$ was loose by a constant (one number); $\mathbf{A}$ is loose by a whole function of position. The practical taming is a gauge condition, and the workhorse of magnetostatics is the Coulomb gauge:

\nabla\cdot\mathbf{A}=0 \qquad (3)

where: reachability is guaranteed — given any $\mathbf{A}_0$ , solve $\nabla^2\lambda=-\nabla\cdot\mathbf{A}_0$ (a Poisson equation, always solvable with decaying sources) and $\mathbf{A}_0+\nabla\lambda$ obeys (3). The reward is structural: in the Coulomb gauge, Ampère's law $\nabla\times\mathbf{B}=\mu_0\mathbf{J}$ becomes, via the double-curl identity $\nabla\times(\nabla\times\mathbf{A})=\nabla(\nabla\cdot\mathbf{A})-\nabla^2\mathbf{A}$ ,

\nabla^2\mathbf{A}=-\mu_0\mathbf{J} \quad\Longrightarrow\quad \mathbf{A}(\mathbf{r})=\frac{\mu_0}{4\pi}\int\frac{\mathbf{J}(\mathbf{r}')}{|\mathbf{r}-\mathbf{r}'|}\,\mathrm{d}\tau' \qquad (4)

where: the second expression solves the first by the delta-function identity, exactly as in the Helmholtz construction — three Poisson equations, one per Cartesian component. Read (4) physically: the vector potential is built from currents the way $V$ is built from charges, with the same $1/|\mathbf{r}-\mathbf{r}'|$ kernel; as a rule of thumb $\mathbf{A}$ runs parallel to the current that sources it. This equation is E4's central tool — differentiate it once and the Biot–Savart law falls out. Finally, the identity that makes $\mathbf{A}$ a daily convenience rather than a curiosity. Apply the Stokes theorem to $\mathbf{B}=\nabla\times\mathbf{A}$ over any surface spanning a loop:

\Phi_B=\int_{\mathcal{S}}\mathbf{B}\cdot\mathrm{d}\mathbf{a}=\oint_{\partial\mathcal{S}}\mathbf{A}\cdot\mathrm{d}\boldsymbol{\ell} \qquad (5)

where: the flux through a loop is the circulation of $\mathbf{A}$ around it — manifestly surface-independent (it never mentions a surface), manifestly gauge-invariant (the added $\oint\nabla\lambda\cdot\mathrm{d}\boldsymbol{\ell}$ vanishes around any closed loop). Equation (5) is the precise sense in which $\mathbf{A}$ 's loops are physical even though its pointwise values are not: gauge transformations shuffle the pointwise values but cannot touch any closed-loop circulation.

Go deeper · Derivation B · 1 minGauge invariance of every physical statement about A

Derivation BGauge invariance of every physical statement about A

The goal is to establish the two halves of the gauge story as theorems: (i) the transformation (2) exhausts the ambiguity in $\mathbf{A}$ , and (ii) the flux-circulation identity (5) — the physical content of $\mathbf{A}$ — is untouched by it. The strategy is short and structural, leaning on this chapter's two integral theorems; the step to watch is the converse in (i), which quietly reuses the scalar-potential theorem of the previous article.

Step 1. The transformation preserves the field: $\nabla\times(\mathbf{A}+\nabla\lambda)=\nabla\times\mathbf{A}+\nabla\times(\nabla\lambda)=\nabla\times\mathbf{A}$ , since the curl of a gradient vanishes identically. Any gradient may be added free of charge.

Step 2. The converse — nothing else may be. If $\nabla\times\mathbf{A}_1=\nabla\times\mathbf{A}_2$ , their difference $\mathbf{K}=\mathbf{A}_1-\mathbf{A}_2$ is curl-free; in simply-connected space the scalar-potential theorem promotes it to a gradient, $\mathbf{K}=\nabla\lambda$ . Why this matters: it certifies that a gauge condition plus boundary behaviour can pin $\mathbf{A}$ completely — the freedom is exactly one scalar function, no more.

Step 3. Reachability of the Coulomb gauge: given $\mathbf{A}_0$ , demand $\nabla\cdot(\mathbf{A}_0+\nabla\lambda)=0$ , i.e. $\nabla^2\lambda=-\nabla\cdot\mathbf{A}_0$ — a Poisson equation for $\lambda$ , solvable by the Coulomb-kernel integral for decaying sources. Every field configuration admits a divergence-free potential.

Step 4. Gauge invariance of the loop content: under (2), $\oint\mathbf{A}\cdot\mathrm{d}\boldsymbol{\ell}\to\oint\mathbf{A}\cdot\mathrm{d}\boldsymbol{\ell}+\oint\nabla\lambda\cdot\mathrm{d}\boldsymbol{\ell}$ , and the second term is the fundamental theorem for gradients around a closed path: $\lambda(\text{end})-\lambda(\text{start})=0$ . Circulations of $\mathbf{A}$ — hence, by the Stokes theorem, all fluxes of $\mathbf{B}$ — are gauge-blind. Why this is the right notion of "physical": exactly the gauge-invariant combinations survive into measured quantities; the Aharonov-Bohm phase is $q/\hbar$ times this same loop integral. *Verified: consistent with Griffiths §5.4.1-5.4.2 and Zangwill §10.5; the symmetric-vs-Landau example of the figure realizes Step 2 with $\lambda=\tfrac12B_0xy$ , whose gradient is precisely the difference of the two potentials.*

Which statement answers which question

you want	use	notes
does $\mathbf{A}$ exist for this field?	check $\nabla\cdot\mathbf{F}=0$	exact criterion; $\mathbf{B}$ always qualifies
$\mathbf{A}$ from currents	equation (4)	Coulomb gauge; $\mathbf{A}\parallel\mathbf{J}$ rule of thumb
flux through a loop	equation (5)	one line integral, no surface needed
a uniform field's potential	$\mathbf{A}=\tfrac12\mathbf{B}\times\mathbf{r}$	one gauge choice among many
switch gauges	$\mathbf{A}\to\mathbf{A}+\nabla\lambda$	fields, fluxes, physics unchanged
impose Coulomb gauge	solve $\nabla^2\lambda=-\nabla\cdot\mathbf{A}_0$	always reachable

Limiting cases

Case	Result	Meaning
uniform $\mathbf{B}=B_0\hat{\mathbf{z}}$	$\mathbf{A}=\tfrac12\mathbf{B}\times\mathbf{r}$ or $B_0x\,\hat{\mathbf{y}}$	same field, gauge-related potentials
inside an ideal solenoid	$\mathbf{A}=\tfrac12B s\,\hat{\boldsymbol{\phi}}$	rings growing linearly with $s$
outside an ideal solenoid	$\mathbf{A}=\dfrac{\Phi}{2\pi s}\,\hat{\boldsymbol{\phi}}$ , $\mathbf{B}=0$	the potential remembers the enclosed flux
straight wire	$\mathbf{A}=-\dfrac{\mu_0I}{2\pi}\ln(s/s_0)\,\hat{\mathbf{z}}$	parallel to the current; the cylindrical logarithm again
$\lambda=$ const in (2)	$\mathbf{A}$ unchanged	the scalar story's additive constant, embedded
time-dependent fields	$\mathbf{E}=-\nabla V-\partial\mathbf{A}/\partial t$	$\mathbf{A}$ joins the electric field's bookkeeping (E9)

Common trapDo not read A pointwise — its values are gauge noise, its loops are physics

The most tempting misuse of the vector potential is treating its local value as physically meaningful. It is not: at any single point you can make $\mathbf{A}$ vanish, double, or point at the moon by a gauge transformation — the figure's symmetric and Landau gauges assign wildly different vectors to the same point for the same uniform field. Three corollaries. First, never compare $\mathbf{A}$ 's from different calculations without checking they use the same gauge; textbook answers for the "same" problem legitimately differ by gradients. Second, energy- and momentum-like expressions containing bare $\mathbf{A}$ (such as the canonical momentum $\mathbf{p}=m\mathbf{v}+q\mathbf{A}$ you will meet later) are gauge-dependent bookkeeping; only gauge-invariant combinations reach experiments. Third — the flip side — loop quantities are untouchable by gauge: $\oint\mathbf{A}\cdot\mathrm{d}\boldsymbol{\ell}=\Phi_B$ is as physical as anything in the theory, which is precisely why the Aharonov–Bohm phase, flux quantization, and inductance are all written with it. The slogan: $\mathbf{A}$ 's derivatives and loops are physics; its pointwise values are coordinates.

Worked examples

1 · One uniform field, two potentials (confidence builder)

A uniform field $\mathbf{B}=0.20\ \mathrm{T}\,\hat{\mathbf{z}}$ fills space. (a) Verify $\mathbf{A}_1=\tfrac12\mathbf{B}\times\mathbf{r}$ works, and evaluate it at $P=(0.10,\ 0.20,\ 0)$ m. (b) Show $\mathbf{A}_2=Bx\,\hat{\mathbf{y}}$ gives the same field, and find the gauge function connecting them.

Step 1. Unpack $\mathbf{A}_1=\tfrac12\mathbf{B}\times\mathbf{r}=\tfrac{B}{2}(-y,\ x,\ 0)$ . Its curl's $z$ -component: $\partial_xA_y-\partial_yA_x=\tfrac{B}{2}+\tfrac{B}{2}=B$ ✓ (the $x$ and $y$ components vanish). Half the field comes from each of the two swirling components — the symmetric gauge spreads the work democratically.

Step 2. Numbers at $P$ : $\mathbf{A}_1=\tfrac{0.20}{2}(-0.20,\ 0.10,\ 0)=\boxed{(-0.020,\ 0.010,\ 0)\ \mathrm{T\,m}}$ — a ring-shaped potential of magnitude $\tfrac12Bs=0.022$ T·m at $s=0.224$ m from the axis.

Step 3. The alternative: $\mathbf{A}_2=Bx\,\hat{\mathbf{y}}$ has $\partial_xA_y-\partial_yA_x=B-0=B$ ✓. Same field — yet at $P$ , $\mathbf{A}_2=(0,\ 0.020,\ 0)$ T·m, a different vector at the same point. Neither is "the" potential; both are the field's legitimate representatives.

Step 4. The connecting gauge function: $\mathbf{A}_2-\mathbf{A}_1=\tfrac{B}{2}(y,\ x,\ 0)=\nabla\lambda$ with $\boxed{\lambda=\tfrac12Bxy}$ — check: $\nabla(\tfrac12Bxy)=\tfrac{B}{2}(y,x,0)$ ✓. Gauge freedom in the flesh: an entire function's worth of slack, harmless to every measurable.

2 · The solenoid — flux, rings of A, and a ghost outside (exam level)

A long solenoid has $n=1000$ turns per metre, current $I=2.0$ A, radius $R=5.0$ cm. Find (a) the flux it carries, (b) the vector potential at $s=2.0$ cm inside and $s=10$ cm outside, and (c) state what is remarkable about the outside result.

Step 1. The field inside (from E4's standard result, quoted here): $B=\mu_0nI=4\pi\times10^{-7}\times1000\times2.0=2.51\times10^{-3}\ \mathrm{T}$ , axial and uniform; outside, $\mathbf{B}=0$ . The flux: $\Phi=B\pi R^2=2.51\times10^{-3}\times\pi\times(0.050)^2=\boxed{2.0\times10^{-5}\ \mathrm{Wb}}$ .

Step 2. Symmetry dictates the potential's shape: currents run in $\hat{\boldsymbol{\phi}}$ , so (rule of thumb from equation (4)) $\mathbf{A}=A(s)\,\hat{\boldsymbol{\phi}}$ — rings, as in the hero figure. The flux-circulation identity (5) on a coaxial circle of radius $s$ gives $A(s)\cdot2\pi s=\Phi_{\mathrm{enc}}(s)$ .

Step 3. Inside ( $s=2.0$ cm): the circle encloses $\Phi_{\mathrm{enc}}=B\pi s^2$ , so $A=\tfrac12Bs=\tfrac12\times2.51\times10^{-3}\times0.020=\boxed{2.5\times10^{-5}\ \mathrm{T\,m}}$ — the uniform-field ring potential of example 1, as it must be where $\mathbf{B}$ is uniform.

Step 4. Outside ( $s=10$ cm): the circle encloses the full flux regardless of radius, so $A=\dfrac{\Phi}{2\pi s}=\dfrac{1.97\times10^{-5}}{2\pi\times0.10}=\boxed{3.1\times10^{-5}\ \mathrm{T\,m}}$ , falling as $1/s$ .

Step 5. The remarkable part: outside the solenoid $\mathbf{B}=0$ exactly, yet $\mathbf{A}\neq0$ — and cannot be gauged away, because its circulation $\oint\mathbf{A}\cdot\mathrm{d}\boldsymbol{\ell}=\Phi\neq0$ is gauge-invariant. Classically a charge out there feels no force and the ghost seems harmless. Quantum mechanically, an electron encircling the solenoid acquires the measurable phase $q\Phi/\hbar$ — the Aharonov–Bohm effect, confirmed decisively by Tonomura's 1986 experiments. The potential this chapter just built is not scaffolding; in the quantum theory it is the more fundamental object.

Origins · 1 minJames Clerk Maxwell and how this was discovered

James Clerk Maxwell

1831–1879 · Cambridge and the Cavendish Laboratory

Maxwell took the vector potential more seriously than anyone before or for a century after. Faraday had intuited an "electrotonic state" — a condition of space around magnets, remembered by circuits when flux changed. In his 1856 paper on Faraday's lines of force, the 24-year-old Maxwell made it mathematics: the electrotonic state is $\mathbf{A}$ , with $\mathbf{B}=\nabla\times\mathbf{A}$ and induced EMF from $-\partial\mathbf{A}/\partial t$ . In the Treatise (1873) he called $\mathbf{A}$ the electromagnetic momentum, its loop integral the momentum a circuit would surrender when flux collapsed — a mechanical reading vindicated, much later, by the canonical momentum $\mathbf{p}=m\mathbf{v}+q\mathbf{A}$ of quantum theory. After his death the practical men rebelled: Heaviside and Hertz recast the theory in fields alone, Heaviside dismissing the potentials as "metaphysics" to be "murdered." For seventy years they were. Then Aharonov and Bohm (1959) and the gauge theories of the Standard Model restored Maxwell's instinct: the potential, modulo gauge, is where the physics lives.

History

1845

— Franz Neumann's theory of induction introduces the vector potential as a calculational device for mutual inductance.

1856

— Maxwell identifies Faraday's electrotonic state with A, writing B as its curl.

1873

— Maxwell's Treatise names A the electromagnetic momentum and builds electrodynamics around the potentials.

1885

— Heaviside recasts the theory in E and B alone, campaigning to banish the potentials as unphysical scaffolding.

1959

— Aharonov and Bohm predict a measurable electron phase from A in field-free regions; Tonomura's 1986 experiments settle it — the loops of A are real.

Going further

References. Griffiths, Introduction to Electrodynamics, 4th ed., §5.4 (magnetic vector potential) and §10.1 (potential formulation) · Jackson, Classical Electrodynamics, 3rd ed., §5.4 and §6.2-6.3 · Zangwill, Modern Electrodynamics, §10.5 · Purcell & Morin, Electricity and Magnetism, 3rd ed., §6.3 · Feynman Lectures, Vol. II, Ch. 15 (the vector potential, including the AB discussion)

Provenance. Existence construction, gauge analysis, and the Coulomb-gauge Poisson reduction follow Griffiths §5.4 and Zangwill §10.5; the flux-circulation identity is the Stokes theorem applied to $\mathbf{B}=\nabla\times\mathbf{A}$ . A verifier should: (1) run the $A_z=0$ construction on $\mathbf{F}=B_0\hat{\mathbf{z}}$ and recover the Landau gauge; (2) check both example-1 potentials give curl $=B\hat{\mathbf{z}}$ and that $\lambda=\tfrac12Bxy$ connects them; (3) recompute the solenoid numbers — $B=\mu_0nI=2.51$ mT, $\Phi=2.0\times10^{-5}$ Wb, $A_{\mathrm{in}}=\tfrac12Bs=2.5\times10^{-5}$ T·m at 2 cm, $A_{\mathrm{out}}=\Phi/2\pi s=3.1\times10^{-5}$ T·m at 10 cm; (4) confirm $\oint\nabla\lambda\cdot\mathrm{d}\boldsymbol{\ell}=0$ underlies gauge invariance of flux. The Aharonov-Bohm phase $q\Phi/\hbar$ and flux quantum $h/2e=2.07\times10^{-15}$ Wb are standard (Feynman II-15; Tonomura 1986).

Builds on: The Curl of a Vector Field · The Divergence of a Vector Field · The Del Operator (Nabla) · Second Derivatives and the Laplacian · The Gradient Operator Leads to: The Curl (Stokes) Theorem · The Divergence (Gauss) Theorem · Surface Integrals and Flux

Is the vector potential physically real, or just mathematical scaffolding?

The modern answer is precise: its gauge-invariant content is real; its pointwise values are convention. Classically you could just barely maintain the scaffolding view — forces involve only $\mathbf{B}$ . Quantum mechanics broke the tie: in the Aharonov–Bohm setup, electrons passing on either side of a shielded solenoid interfere with a phase shift $q\Phi/\hbar=\frac{q}{\hbar}\oint\mathbf{A}\cdot\mathrm{d}\boldsymbol{\ell}$ , though they never touch a region with $\mathbf{B}\neq0$ . Something physical lives outside the solenoid, and it is the (gauge-invariant loop content of the) vector potential. Tonomura's 1986 experiments — flux trapped in a superconductor-shielded toroidal magnet — closed every loophole. The refined slogan: nature computes with $\mathbf{A}$ , but only asks gauge-invariant questions of it.

How do I choose a gauge, and does the choice ever matter?

Choose for convenience, exactly as you choose coordinates — and no measurable quantity ever depends on the choice. The Coulomb gauge $\nabla\cdot\mathbf{A}=0$ is magnetostatics' default (it turns Ampère's law into Poisson equations); the Landau gauge $\mathbf{A}=Bx\,\hat{\mathbf{y}}$ simplifies quantum problems with one translation symmetry; the symmetric gauge suits rotational symmetry; and in E9 the Lorenz gauge (a different animal, mixing $V$ and $\mathbf{A}$ ) makes relativity manifest. What can go wrong is inconsistency: mixing formulas derived in different gauges within one calculation, or comparing your $\mathbf{A}$ with a textbook's without checking their condition. Discipline: declare the gauge once at the top of a calculation, and verify any imported formula shares it.

Why does A tend to point along the current that produces it?

Read it off the Coulomb-gauge solution $\mathbf{A}=\frac{\mu_0}{4\pi}\int\frac{\mathbf{J}(\mathbf{r}')}{|\mathbf{r}-\mathbf{r}'|}\mathrm{d}\tau'$ : the integrand is the current density itself, merely weighted by the scalar kernel $1/|\mathbf{r}-\mathbf{r}'|$ — so each current element contributes a piece of $\mathbf{A}$ parallel to itself, and the total inherits the current's general direction (a straight wire's $\mathbf{A}$ runs along the wire; a solenoid's circular currents make circular $\mathbf{A}$ ). Contrast $\mathbf{B}$ , which the Biot–Savart cross product forces to wrap around currents. The parallelism is a Coulomb-gauge statement — other gauges can tilt it — but as physical intuition for sketching potentials it is the single most useful rule of thumb in E4.

The scalar potential had a clear meaning — work per charge. Does A have one?

Yes, though it took quantum mechanics to make it crisp. Maxwell's own reading — $\mathbf{A}$ as "electromagnetic momentum per unit charge" — is vindicated by the canonical momentum $\mathbf{p}=m\mathbf{v}+q\mathbf{A}$ : a charge in a potential carries momentum bookkeeping beyond $m\mathbf{v}$ , and when flux collapses (switching off a solenoid), the impulse delivered to a nearby charge is exactly $-q\,\Delta\mathbf{A}$ — the field pays out the stored momentum. In quantum mechanics the meaning sharpens further: $\mathbf{A}$ sets the phase gradient of charged wavefunctions, $\frac{q}{\hbar}\mathbf{A}\cdot\mathrm{d}\boldsymbol{\ell}$ of phase per step, which is why loops of $\mathbf{A}$ produce interference shifts. Per unit charge: work-per-step for $V$ , phase-per-step (or momentum) for $\mathbf{A}$ — the pair will fuse into one four-vector in E9.

If magnetic monopoles existed, what would happen to the vector potential?

Then $\nabla\cdot\mathbf{B}=\mu_0\rho_m\neq0$ at monopoles, and no globally smooth $\mathbf{A}$ could exist — the existence theorem's hypothesis fails exactly where a monopole sits. Dirac explored the loophole in 1931: one can keep $\mathbf{A}$ everywhere except along a thin "Dirac string" running from the monopole to infinity, and the string is undetectable precisely when the monopole charge obeys $qg=n\frac{h}{4\pi}$ — quantization of electric charge derived from the mere possibility of a monopole, one of the most seductive arguments in physics. No monopole has ever been observed ( $\nabla\cdot\mathbf{B}=0$ stands unbroken), but the analysis previews a deep theme: topological obstructions to potentials carry physics, a motif running from the wire's multivalued scalar potential through modern topological materials.

Why is the flux through a loop expressible without choosing a surface, and why do inductors care?

Because equation (5) writes flux as $\oint\mathbf{A}\cdot\mathrm{d}\boldsymbol{\ell}$ — a rim quantity with no surface in sight; the Stokes theorem guarantees every spanning surface would have agreed anyway (that was the surface-independence of the curl-theorem article, now automated). Circuit engineering leans on this constantly: the flux linkage of a coil, $N\Phi$ , and hence its inductance $L=N\Phi/I$ , is well-defined for arbitrarily contorted windings because flux is a property of each turn's rim. Mutual inductance calculations (Neumann's formula, historically the birthplace of $\mathbf{A}$ ) are literally double loop-integrals of the $1/|\mathbf{r}-\mathbf{r}'|$ kernel — two circulations of the vector potential shaking hands. When E6 derives induction machinery, every "flux through the circuit" will silently be this gauge-invariant loop integral.

Both potentials in hand, what exactly gets carried into the rest of the track?

The complete kit for representing fields. Statics: $\mathbf{E}=-\nabla V$ (curl-free sector) and $\mathbf{B}=\nabla\times\mathbf{A}$ (divergence-free sector) — E1 through E5 run on these. Full electrodynamics (E9): the two homogeneous Maxwell equations ( $\nabla\cdot\mathbf{B}=0$ and Faraday) are identically solved by writing $\mathbf{B}=\nabla\times\mathbf{A}$ , $\mathbf{E}=-\nabla V-\partial\mathbf{A}/\partial t$ , with the gauge transformation growing a partner: $V\to V-\partial\lambda/\partial t$ alongside $\mathbf{A}\to\mathbf{A}+\nabla\lambda$ . Four potential functions, minus one gauge function, carry the six field components — and in relativity $(V/c,\mathbf{A})$ assembles into a four-vector while $(\mathbf{E},\mathbf{B})$ must share a tensor. The chapter's parting claim, redeemed across the rest of the track: potentials are not a computational trick; they are the variables in which electrodynamics is most itself.