Inhomogeneous Wave Equations for Potentials

The last article ended with four equations that deserve a longer look than a victory lap. In the Lorenz gauge, the scalar potential and each Cartesian component of the vector potential obey one and the same equation — $\Box^2 V = -\rho/\varepsilon_0$ , $\Box^2\mathbf{A} = -\mu_0\mathbf{J}$ — the inhomogeneous wave equation. Every radiation formula in this chapter, every antenna pattern, every field of every moving charge is a statement about solutions of this single scalar problem, run four times with different sources. Before solving it (next article), it pays to understand it as an object: what kind of equation it is, what it promises, and what it demands. What it promises is causality with teeth. The wave operator's light-cone structure forces a domain of dependence: the potential at a point, now, can depend only on data and sources inside its backward light cone — nothing outside can matter, provably, with an energy argument doing the enforcement. In three dimensions the promise sharpens into something almost miraculous: an impulse from a point source produces a response confined to an infinitely thin expanding shell — no precursor, no wake. Your radio works because the wave equation in 3D forgets nothing and smears nothing. What it demands is data. A Poisson equation wants boundary values; a wave equation wants initial conditions and a statement about infinity — the outgoing-radiation condition that quietly selects "retarded" behavior before any Green function is chosen. This article builds the operator's anatomy, proves the causal structure, constructs the point-impulse response — which turns out to be a delta function riding the light cone, retarded or advanced — and hands the two candidates to the next two articles: one to verify and adopt, one to prosecute and (mostly) decline.

Key takeaways

In the Lorenz gauge all of electrodynamics reduces to four copies of one scalar problem: the inhomogeneous wave equation $\Box^2\psi = -f$ , with $\Box^2 = \nabla^2 - \tfrac{1}{c^2}\partial_t^2$ the d'Alembertian, $\psi \in \{V, A_x, A_y, A_z\}$ and $f$ the matching source.
The operator is hyperbolic: it has characteristics — light cones — and its solutions obey a strict domain-of-dependence theorem: $\psi(\mathbf{r},t)$ is fixed entirely by data and sources in the backward cone $|\mathbf{r}-\mathbf{r}'| \le c(t-t')$ .
Static limit: $\Box^2\to\nabla^2$ and the equations collapse to the Poisson equations of electro- and magnetostatics; quasistatic corrections enter at order $(L/c\tau)^2$ .
The point-impulse response (Green function) is a spherical delta shell on the light cone: $G_{\text{ret}} = \delta(t - t' - \boldsymbol{\mathscr{r}}/c)/4\pi\mathscr{r}$ — and its time-mirror, the advanced twin. Construction here; verification and physical selection in the next two articles.
In three dimensions Huygens's principle holds sharply: signals have clean rear edges as well as fronts — a brief source produces a shell of finite thickness, hollow inside.
Well-posedness needs initial data ( $\psi$ , $\partial_t\psi$ at $t=0$ ) plus decay/outgoing conditions at infinity (the Sommerfeld radiation condition) — the replacements for electrostatics' boundary screens.
Key idea: the Lorenz-gauge potentials obey four identical inhomogeneous wave equations whose hyperbolic structure enforces causality — response confined to the light cone, propagating at $c$ , with the retarded and advanced shell solutions as the elementary responses.

The core insight

In the Lorenz gauge the coupled potential equations separate into four identical inhomogeneous wave equations — the d'Alembertian of each potential equals minus its source — whose light-cone structure confines every response to speed $c$ .

Where this lives today

The inhomogeneous wave equation is arguably the most-solved PDE in engineering. Every FDTD (finite-difference time-domain) electromagnetic solver — the tool that designs phone antennas, radar cross-sections, and photonic chips — is a discretization of exactly this structure, and its two great practical headaches are exactly this article's two theorems in disguise: the CFL stability limit (time step short enough that numerical influence respects the light cone) and absorbing boundary conditions (perfectly matched layers, the industrial descendants of Sommerfeld's outgoing condition, soaking up radiation so the finite box imitates infinity). The four-copies structure is also why the same codebase solves acoustics and elastodynamics with swapped constants. The operator's reach extends far beyond electromagnetism. Linearized general relativity puts the metric perturbation in a Lorenz-analog (harmonic) gauge and finds $\Box \bar h^{\mu\nu} = -16\pi G\,T^{\mu\nu}/c^4$ — the same equation, tensor-valued: LIGO's waveforms and the quadrupole formula are this chapter's mathematics with heavier indices. And the strong-Huygens sharpness has a modern observational edge: gravitational-wave signals and gamma-ray bursts from the same neutron-star merger arriving within seconds across $10^{15}$ light-seconds is a live test that both messengers ride the same cones this article drew.

Origins · 1 minJean le Rond d'Alembert and how this was discovered

Jean le Rond d'Alembert

1717–1783 · Paris — who wrote the wave equation and split it along its characteristics

Abandoned as an infant on the steps of Saint-Jean-le-Rond and raised by a glazier's wife, d'Alembert became co-editor of the Encyclopédie and the mathematical engine of the French Enlightenment. His 1747 memoir on the vibrating string produced the first partial differential equation of physics — the wave equation — together with its complete solution $u = F(x - ct) + H(x + ct)$ : the insight that the operator factors along characteristics, so that arbitrary profiles ride left and right at fixed speed. Both halves of this article are his legacy compounded: the $u = rG$ reduction hands the 3D point response to his 1D formula, and the light-cone causality of Derivation A is the geometric maturation of his characteristics. He fought bitterly with Euler and Daniel Bernoulli over what "arbitrary function" could mean — a quarrel that seeded the eventual theory of distributions, without which the delta-shell Green function of this article could not even be written. The operator wears his name: the d'Alembertian, the wave equation's Laplacian-with-a-clock.

History

1747

— d'Alembert derives and solves the one-dimensional wave equation for the vibrating string

1818

— Poisson's spherical-means formula solves the 3D initial-value problem, exposing the shell structure

1858

— Riemann's retarded-potential memoir (published 1867) treats the inhomogeneous equation for electrodynamics

1867

— Lorenz builds electrodynamics on retarded solutions of these equations, independent of Maxwell

1894

— Volterra and later Hadamard classify when Huygens's principle holds sharply — odd dimensions ≥ 3

1912

— Sommerfeld formulates the radiation condition, making "outgoing at infinity" a precise boundary statement

1966

— Yee's staggered-grid algorithm (FDTD) turns the equations into the workhorse of computational electromagnetics

2017

— GW170817: gravitational and electromagnetic signals from one merger arrive within seconds — the shared cone structure tested across 130 million light-years

The physical picture

Start from what you already trust. Electrostatics taught you $\nabla^2 V = -\rho/\varepsilon_0$ : the potential is the accumulated $1/r$ response to charge, adjusting everywhere the moment the source moves, because the Laplacian has no clock in it. The wave equation is that law with a conscience installed: subtract $\tfrac{1}{c^2}\partial_t^2$ and the response can no longer keep up instantaneously — adjustments ripple outward at $c$ , no faster, ever. The single most useful way to hold the operator in your head is as a competition: $\nabla^2\psi$ measures how much $\psi$ at a point lags the average of its neighbors (the curvature-as-disagreement reading from the boundary-value chapter), and $\tfrac{1}{c^2}\partial_t^2\psi$ is the acceleration with which $\psi$ responds to that disagreement. The wave equation says the response is the disagreement — a field of coupled oscillators, each dragged toward its neighbors' consensus, with $c$ set by the stiffness-to-inertia ratio of empty space, $c = 1/\sqrt{\mu_0\varepsilon_0}$ . The "inhomogeneous" part is where electrodynamics happens. The source term $-f$ commandeers the oscillator at each point where charge or current lives, and the equation's job is to reconcile the commandeering with the neighbors' consensus — the reconciliation propagating outward along the cone. Everything downstream in this chapter is a study of that reconciliation: the retarded integral is its explicit formula, Jefimenko's equations are its field-level consequences, and radiation is its far-field residue.

Figure

The backward light cone — the potential at one event can depend only on data and sources inside the cone; the energy argument of Derivation A patrols the boundary

One more structural remark before the theorems, because it costs nothing here and pays off for an entire later chapter. The same operator acts on all four quantities. Not similar operators — the same one, with the sources lining up as $(\rho c, \mathbf{J})$ against $(V/c, \mathbf{A})$ . When one differential operator refuses to distinguish four objects, physics is telling you they are components of a single thing. The relativity chapter will make it official: $(V/c, \mathbf{A})$ is a four-vector $A^\mu$ , $(c\rho, \mathbf{J})$ is the four-current $J^\mu$ , and this article's four equations are one covariant statement, $\Box^2 A^\mu = -\mu_0 J^\mu$ with $\partial_\mu A^\mu = 0$ . Here that stands as a preview at statement level; the tensor machinery belongs to The Four-Potential.

The operator and its ledger

Collect the objects. The d'Alembertian (Griffiths's notation):

\Box^2 \equiv \nabla^2 - \mu_0\varepsilon_0\frac{\partial^2}{\partial t^2} = \nabla^2 - \frac{1}{c^2}\frac{\partial^2}{\partial t^2}, \qquad (1)

where: the sign convention matters when comparing books — many relativity texts define $\Box = \tfrac{1}{c^2}\partial_t^2 - \nabla^2$ (the negative of this), so a formula ported between conventions flips the sign of every source term. Griffiths's $\Box^2$ is "Laplacian minus clock," which keeps the statics limit visually obvious. The Lorenz-gauge system, quoted from the last article:

\Box^2 V = -\frac{\rho}{\varepsilon_0}, \qquad \Box^2 A_x = -\mu_0 J_x, \qquad \Box^2 A_y = -\mu_0 J_y, \qquad \Box^2 A_z = -\mu_0 J_z. \qquad (2)

where: the Cartesian-component form is not cosmetic — only in Cartesian coordinates does the vector Laplacian act componentwise (curvilinear unit vectors twist, mixing components), so "four scalar copies" is a statement you may only cash in rectangular coordinates or after the dust settles. Each copy is the generic problem

\Box^2 \psi(\mathbf{r},t) = -f(\mathbf{r},t), \qquad (3)

where: $\psi$ stands for any of the four potentials and $f$ for its source (in units where the analogy to Poisson's $\nabla^2\psi = -f$ is exact). Because the operator is linear, sources superpose: solve for a point source, integrate over the actual distribution — the strategy of every Green-function method you have met, now with time in the game. And because the operator carries a clock, the character of the problem changes genus: Poisson is elliptic (boundary-value: give me screens and I hand you equilibrium), the wave equation is hyperbolic (initial-value: give me a starting state and I evolve it). The data that make (3) well-posed are

\psi(\mathbf{r},0), \quad \frac{\partial\psi}{\partial t}(\mathbf{r},0) \quad\text{everywhere} \qquad + \qquad \text{outgoing behavior at infinity}, \qquad (4)

where: the last item — the Sommerfeld radiation condition, $\lim_{r\to\infty} r\left(\partial_r\psi + \tfrac{1}{c}\partial_t\psi\right) = 0$ — says no radiation sneaks in from infinity. It replaces electrostatics' grounded screens, and it is the precise, honest place where "retarded" gets selected: the equations themselves are time-symmetric, and the asymmetry enters as data. That prosecution is Advanced Potentials and Causality; here we simply post the requirement.

Derivation ADomain of dependence — nothing outside the light cone can matter

The goal is the theorem that gives the wave equation its causal teeth: the value $\psi(\mathbf{r}_0, t_0)$ is completely determined by the initial data and the sources within the backward light cone $|\mathbf{r}-\mathbf{r}_0| \le c(t_0 - t)$ ; equivalently, changing data or sources outside the cone changes nothing at the apex. The strategy is the classical energy argument: for the difference $u$ of two solutions with identical data and sources inside the cone, define a local energy on shrinking spheres and show it cannot increase — trapping $u$ at zero all the way up to the apex. This is also the uniqueness theorem the previous article's constraint-propagation argument borrowed, so the chapter's IOU gets paid here.

Step 1. Reduce to the homogeneous problem. Let $\psi_1, \psi_2$ solve (3) with the same $f$ inside the cone and the same initial data on the cone's base. Their difference $u = \psi_1 - \psi_2$ satisfies $\Box^2 u = 0$ inside the cone with $u = \partial_t u = 0$ on the base. Why differences: linearity converts a dependence question ("what does the apex read?") into a vanishing question ("can zero data grow?") — the standard maneuver, and the reason the theorem holds for any source, however violent, as long as it is the same source inside the cone.

Step 2. Build the shrinking-sphere energy. For $0 \le t \le t_0$ define

E(t) = \frac{1}{2}\int_{B(t)} \left[\frac{1}{c^2}\left(\frac{\partial u}{\partial t}\right)^2 + |\nabla u|^2\right] d^3r, \qquad B(t) = \left\{\,|\mathbf{r}-\mathbf{r}_0| \le c(t_0 - t)\,\right\},

the wave energy inside the sphere that shrinks at speed $c$ toward the apex. On the base, $E(0) = 0$ because both $\partial_t u$ and $\nabla u$ vanish there.

Step 3. Differentiate and let the boundary fight the flux. Two terms: the volume integrand's time derivative, and the leakage from the shrinking boundary (Leibniz):

\frac{dE}{dt} = \int_{B(t)} \left[\frac{1}{c^2}\frac{\partial u}{\partial t}\frac{\partial^2 u}{\partial t^2} + \nabla u\cdot\nabla\frac{\partial u}{\partial t}\right] d^3r \;-\; \frac{c}{2}\oint_{\partial B(t)}\left[\frac{1}{c^2}\left(\frac{\partial u}{\partial t}\right)^2 + |\nabla u|^2\right] da.

In the volume term, integrate the second piece by parts and use $\Box^2u = 0$ (i.e. $\tfrac{1}{c^2}\partial_t^2 u = \nabla^2 u$ ): the volume contributions cancel exactly, leaving only surface terms:

\frac{dE}{dt} = \oint_{\partial B(t)}\left[\frac{\partial u}{\partial t}\frac{\partial u}{\partial n} - \frac{c}{2}\left(\frac{1}{c^2}\left(\frac{\partial u}{\partial t}\right)^2 + |\nabla u|^2\right)\right] da.

Step 4. The boundary term is never positive. Estimate the cross term with the arithmetic–geometric inequality, $\left|\partial_t u\,\partial_n u\right| \le \tfrac{c}{2}\left[\tfrac{1}{c^2}(\partial_t u)^2 + (\partial_n u)^2\right]$ , and note $(\partial_n u)^2 \le |\nabla u|^2$ :

\frac{dE}{dt} \le 0.

Why the sphere must shrink at exactly $c$ : the Leibniz leakage term carries the factor $c$ that makes the estimate close. Shrink slower and energy can outrun the boundary — the estimate fails; the failure is physical, since signals genuinely travel at $c$ . The cone's slope is not an assumption but the unique speed at which the argument seals.

Step 5. Close the trap. $E(0) = 0$ , $E \ge 0$ , $dE/dt \le 0$ force $E(t) = 0$ for all $t \le t_0$ : $u$ has zero gradient and zero time-derivative throughout the cone, and being zero on the base, vanishes identically — including at the apex. So $\psi_1 = \psi_2$ at $(\mathbf{r}_0, t_0)$ : the apex value depends on nothing outside the cone. As corollaries: (i) uniqueness for the initial-value problem (the theorem the Lorenz-gauge persistence argument used), and (ii) finite propagation speed — a source switched on at the origin cannot affect $\psi$ at distance $r$ before $t = r/c$ . Verified: standard energy-method result (Evans, Partial Differential Equations, §2.4; Zangwill §20.2); consistent with the explicit retarded solutions of the next article, which visibly involve only cone data.

The elementary response: a delta on the cone

Superposition reduces everything to one question: what does a point flash produce — a source $f = \delta^3(\mathbf{r})\,\delta(t)$ , one impulse at the origin at one instant? Its response $G$ is the Green function of the operator, and the full solution of (3) is then the sum of flashes weighted by the actual source history. The construction is short and beautiful, and it stops — deliberately — one step before the finish line.

Go deeper · Derivation B · 2 minThe point-impulse response — spherical shells, retarded and advanced

Derivation BThe point-impulse response — spherical shells, retarded and advanced

The goal is every solution of $\Box^2 G = -\delta^3(\mathbf{r})\,\delta(t)$ that respects the symmetry of the problem, and the answer will be a pair: an expanding and a contracting spherical delta shell — the retarded and advanced Green functions. The strategy has three moves: symmetry reduces $G$ to a function of $r$ and $t$ ; away from the origin the radial wave equation is solved exactly by the substitution $u = rG$ (the same trick that solved the spherical wave in the wave chapter); and the delta's strength is fixed by matching to the known static limit. What this derivation does not do — postponed by design — is verify that the retarded choice, integrated over a real source, solves the full time-dependent problem (next article's flagship), or decide between the twins (the article after).

Step 1. Spend the symmetry. The operator and the source are invariant under rotations about the origin and translations in neither space nor time beyond the impulse's location, so $G = G(r, t)$ with $r = |\mathbf{r}|$ . For $r > 0$ the source vanishes and

\frac{1}{r}\frac{\partial^2 (rG)}{\partial r^2} - \frac{1}{c^2}\frac{\partial^2 G}{\partial t^2} = 0 \qquad\Longrightarrow\qquad \frac{\partial^2 u}{\partial r^2} - \frac{1}{c^2}\frac{\partial^2 u}{\partial t^2} = 0, \quad u \equiv rG,

using the spherical-Laplacian identity $\nabla^2 G = \tfrac{1}{r}\partial_r^2(rG)$ for radial functions. Why the substitution works: multiplying by $r$ absorbs the geometric dilution of spherical spreading, converting the 3D radial problem into d'Alembert's 1D wave equation on the half-line — whose complete solution has been in hand since the wave chapter.

Step 2. Write the general radial solution. By d'Alembert, $u = F(t - r/c) + H(t + r/c)$ for arbitrary profiles $F, H$ , so

G(r,t) = \frac{F(t - r/c)}{r} + \frac{H(t + r/c)}{r}.

The first term is a profile propagating outward at $c$ (whatever $F$ carries at the origin appears at radius $r$ a time $r/c$ later); the second propagates inward.

Step 3. Fix the profiles with the impulse. The source acts only at $r=0$ , $t=0$ , and the response must echo the impulse's time dependence: take $F(\xi) = \alpha\,\delta(\xi)$ and $H(\xi) = \beta\,\delta(\xi)$ and determine the constants by the static-limit match: integrating the defining equation over time, $\int G\,dt$ must solve $\nabla^2\left(\int G\,dt\right) = -\delta^3(\mathbf{r})$ — the Poisson point problem, whose solution $1/4\pi r$ has been known since electrostatics. Since $\int \delta(t \mp r/c)\,dt = 1$ ,

\int G\,dt = \frac{\alpha + \beta}{r} \stackrel{!}{=} \frac{1}{4\pi r} \qquad\Longrightarrow\qquad \alpha + \beta = \frac{1}{4\pi}.

Why only the sum is fixed: the equation is time-symmetric, and the impulse constrains total strength, not its split between the outgoing and ingoing branches. Mathematics offers a one-parameter family; the physically distinguished endpoints are all-outgoing and all-ingoing.

Step 4. Name the endpoints. All-outgoing ( $\beta = 0$ ) and all-ingoing ( $\alpha = 0$ ):

G_{\text{ret}}(r,t) = \frac{\delta(t - r/c)}{4\pi r}, \qquad\qquad G_{\text{adv}}(r,t) = \frac{\delta(t + r/c)}{4\pi r}.

The retarded response is an infinitely thin spherical shell of influence expanding at $c$ from the flash — zero before the light cone arrives, zero after it passes, geometric dilution $1/r$ on the shell itself. The advanced response is its time mirror: a shell converging on the flash from the infinite past. Both are legitimate solutions of the defining equation; general superpositions $\alpha G_{\text{ret}} + \beta G_{\text{adv}}$ with $\alpha+\beta = 1/4\pi\cdot 4\pi = 1$ ... (in the normalized sense of Step 3) are too. Which one nature's problems call for — and the precise sense in which that is a statement about data, not dynamics — is the next two articles' business. Verified: the retarded kernel reproduces the static $1/4\pi r$ on time-integration (Step 3's own check); inserting $G_{\text{ret}}$ into the source integral and differentiating is the verification derivation of the next article, matching Griffiths §10.2 and Jackson §6.4.

The payoff of the delta-shell structure deserves its own name: Huygens's principle in the strong sense. Because the 3D response lives only on the cone $t = r/c$ — not inside it — signals in three dimensions have sharp rear edges as well as sharp fronts. A source that acts for a microsecond produces, at any observer, a disturbance lasting exactly a microsecond (plus the source's light-crossing time): no afterglow, no reverberation. This is special to odd space dimensions ≥ 3: in 2D (a struck drumhead, ripples on a pond) the response fills the cone's interior and every signal drags an infinite tail. Radio, and arguably intelligible speech, are gifts of living in three dimensions.

Figure

Strong Huygens in 3D — a source active for a finite interval produces a hollow shell of influence: sharp front, sharp rear, silent interior

Limiting cases

case	result	meaning
statics ( $\partial_t \to 0$ )	$\nabla^2 V = -\rho/\varepsilon_0$ , $\nabla^2\mathbf{A} = -\mu_0\mathbf{J}$	the Poisson equations of chapters past: electrostatics and magnetostatics are the wave equations with the clock unplugged
slow sources (size $L$ , timescale $\tau$ , $L \ll c\tau$ )	Poisson + corrections of order $(L/c\tau)^2$	quasistatics quantified: a 50 Hz, 1 m transformer errs by $\sim3\times10^{-14}$ ; a 3 GHz, 10 cm circuit board has $L/c\tau \approx 1$ — waves or bust
source-free region	$\Box^2 V = 0$ , $\Box^2\mathbf{A} = 0$	the free wave equation of the wave chapter — the potentials inherit its whole solution theory (d'Alembert, plane-wave completeness)
harmonic time dependence $e^{-i\omega t}$	$(\nabla^2 + k^2)\tilde\psi = -\tilde f$ , $k = \omega/c$	the inhomogeneous Helmholtz equation: one frequency at a time, the workhorse reduction for antennas and cavities
covariant packaging	$\Box^2 A^\mu = -\mu_0 J^\mu$ , $\partial_\mu A^\mu = 0$	four equations, one four-vector statement — preview cashed in the relativity chapter

Common trapThese clean wave equations are a Lorenz-gauge privilege — and the potentials' propagation is still gauge bookkeeping

Two scope fences. First: $\Box^2 V = -\rho/\varepsilon_0$ is not a gauge-independent law. In the Coulomb gauge $V$ obeys an instantaneous Poisson equation and only $\mathbf{A}$ waves (with the nonlocal transverse current as source); quote the wave equation for $V$ outside the Lorenz gauge and you compute fiction. Second: even inside the Lorenz gauge, the fact that the potentials propagate at $c$ is bookkeeping, not observation — the previous article's gauge waves are potentials propagating at $c$ with zero measurable content. The gauge-invariant causal statements are about the fields (their sourced wave equations, and eventually Jefimenko's formulas). The Lorenz gauge's virtue is that its bookkeeping parallels the physics; it remains bookkeeping.

Worked examples

1 · The Helmholtz reduction and the outgoing spherical wave (confidence builder)

Setup. A source oscillates at a single frequency: $\rho(\mathbf{r},t) = \text{Re}\left[\tilde\rho(\mathbf{r})\,e^{-i\omega t}\right]$ . Show the wave equation becomes a Helmholtz equation, solve it for a point oscillator $\tilde\rho = q_0\,\delta^3(\mathbf{r})$ , and put numbers on the phase structure at $f = 1$ GHz.

Step 1. Reduce. Writing $V = \text{Re}[\tilde V(\mathbf{r})e^{-i\omega t}]$ , time derivatives become $-i\omega$ factors: $\Box^2 V = -\rho/\varepsilon_0$ becomes

\left(\nabla^2 + k^2\right)\tilde V = -\frac{\tilde\rho}{\varepsilon_0}, \qquad k = \frac{\omega}{c}.

One frequency, no clock: the price of the reduction is that causality hides in a boundary condition (Step 3).

Step 2. Solve for the point source. Away from the origin, the radial substitution $u = r\tilde V$ gives $u'' + k^2 u = 0$ : $u = \alpha e^{ikr} + \beta e^{-ikr}$ . Matching the delta's strength as in Derivation B (the $k \to 0$ limit must recover $q_0/4\pi\varepsilon_0 r$ ) fixes $\alpha + \beta = q_0/4\pi\varepsilon_0$ :

\tilde V(r) = \frac{q_0}{4\pi\varepsilon_0}\frac{e^{ikr}}{r} \quad (\text{choosing } \beta = 0).

Step 3. Justify the choice. Restore the clock: $e^{ikr}e^{-i\omega t} = e^{i(kr - \omega t)}$ — crests move outward; the discarded $e^{-ikr}$ branch is an incoming wave. The Sommerfeld radiation condition, $r(\partial_r\tilde V - ik\tilde V) \to 0$ , kills the incoming branch mechanically: the frequency domain's version of choosing $G_{\text{ret}}$ .

Step 4. Numbers. At $f = 1$ GHz: $k = 2\pi f/c = 20.96$ rad/m, $\lambda = 30.0$ cm. At $r = 1$ m the phase lag is $kr = 20.96$ rad $= 3.34$ full cycles: the potential there reports the oscillator 3.34 periods ago. Equal-phase shells are spaced 30 cm apart, marching outward at $c$ — retardation made visible one frequency at a time.

The lesson: the Helmholtz equation is the wave equation with causality relocated into the boundary condition — and the $e^{ikr}/r$ outgoing kernel is $G_{\text{ret}}$ wearing frequency-domain clothes.

2 · The hollow shell — support of the solution, computed exactly (exam level)

Setup. A source occupies a ball of radius $a = 1$ m at the origin and is active only during $0 < t < \tau = 1\ \mu$ s (before and after: nothing, and all initial data vanish). Using the causal structure — Derivation A plus the delta-shell response — determine exactly where the potential can be nonzero at $t = 5\ \mu$ s, and how the answer evolves.

Step 1. Support from the cones. Each source event $(\mathbf{r}', t')$ with $|\mathbf{r}'| \le a$ , $0 < t' < \tau$ influences the field only on its light cone $|\mathbf{r} - \mathbf{r}'| = c(t - t')$ — on, not inside, by the delta-shell structure (strong Huygens). The support at time $t$ is the union of these spheres.

Step 2. Outer front. The largest reach comes from the earliest events ( $t' \to 0$ ) at the ball's far edge: $r_{\text{out}} = a + ct = 1 + (2.998\times10^8)(5\times10^{-6}) = 1 + 1499 = \mathbf{1500\ m}$ .

Step 3. Rear edge. The last influence at small radii comes from the latest events ( $t' \to \tau$ ) at the near edge: $r_{\text{in}} = c(t - \tau) - a = 1199.2 - 1 = \mathbf{1198\ m}$ . For $r < 1198$ m the field is identically zero again — every source cone has already swept past.

Step 4. The shell. The disturbance at $t = 5\ \mu$ s occupies the spherical shell $1198\ \text{m} \le r \le 1500\ \text{m}$ : thickness $c\tau + 2a = 299.8 + 2 = \mathbf{302\ m}$ , constant for all later times as the shell expands. An observer at any radius hears: silence — a 302-m-thick burst sweeping past at $c$ (duration $\tau + 2a/c \approx 1.007\ \mu$ s) — silence forever after.

Step 5. The counterfactual. In two space dimensions (or for the electrostatic Poisson equation, which has no cones at all) Step 3 fails: the response fills the cone's interior, the "silence after" never comes, and the observer hears an infinite decaying tail. The clean rear edge is a theorem about $G_{\text{ret}}$ 's delta structure, special to 3D.

The lesson: causal support can be computed before solving anything — the operator's geometry already fixes where the answer is allowed to live. When the retarded integrals arrive next article, check them against exactly this bookkeeping.

Going further

References. Griffiths, Introduction to Electrodynamics, 4th ed., §10.1 · Jackson, Classical Electrodynamics, 3rd ed., §6.4–6.5 · Zangwill, Modern Electrodynamics, §20.1–20.5 · Evans, Partial Differential Equations, §2.4 (energy methods and domain of dependence)

Provenance. All equations match Griffiths 4th ed. §10.1 (Eqs. 10.4–10.19) and Jackson 3rd ed. §6.3–6.4. A verifier should: (1) re-derive the two coupled potential equations by substituting $\mathbf{E}=-\nabla V-\partial_t\mathbf{A}$ , $\mathbf{B}=\nabla\times\mathbf{A}$ into the inhomogeneous Maxwell equations; (2) confirm cross-term cancellation under the Lorenz condition; (3) verify $\mu_0\varepsilon_0=1/c^2$ with $c=2.998\times10^8\,\mathrm{m/s}$ ; (4) check the covariant reduction with metric $(+,-,-,-)$ . Numerical constants ( $c$ , $\mu_0$ , $\varepsilon_0$ ) are standard CODATA/textbook values. Addendum for this edition: per the chapter's division of labor, verifier items (1)–(2) — the assembly of the coupled system and its Lorenz-gauge decoupling, the base's first derivation — are owned by Scalar and Vector Potentials and Lorenz Gauge respectively and are quoted here as equation (2); the base's "existence of the Lorenz gauge" derivation is likewise owned by Lorenz Gauge. This edition owns the operator anatomy, the domain-of-dependence/uniqueness theorem (Derivation A — the result the Lorenz-gauge persistence argument borrows), the Green-function construction halting at the retarded/advanced pair (Derivation B; verification belongs to Retarded Potentials, selection to Advanced Potentials and Causality), the strong-Huygens analysis, and the well-posedness/Sommerfeld data discussion. Worked-example numbers ( $k = 20.96$ rad/m, $kr = 3.34$ cycles; shell fronts 1500 m / 1198 m, thickness 302 m; quasistatic ratios $2.8\times10^{-14}$ and $\approx1$ ) recomputed for this edition.

Builds on: Scalar and Vector Potentials · Gauge Transformations · Lorenz Gauge · The Electromagnetic Wave Equation Leads to: Retarded Potentials · Advanced Potentials and Causality · Jefimenko's Equations · Liénard–Wiechert Potentials · Electric Dipole Radiation

What exactly does "inhomogeneous" mean here — inhomogeneous in what?

It is the standard PDE usage: the equation has a source term independent of the unknown. $\Box^2\psi = 0$ is homogeneous (every term contains $\psi$ ); $\Box^2\psi = -f$ is inhomogeneous ( $f$ sits there regardless of $\psi$ ). The word has nothing to do with inhomogeneous media — everything in this chapter is vacuum. The structure to internalize: general solution = any particular solution driven by $f$ + arbitrary homogeneous solution. The retarded integral (next article) supplies the particular solution; the homogeneous freedom is fixed by initial data and the Sommerfeld condition, and in the quiescent-past setting it is zero. The same grammar organized Poisson problems (particular integral + harmonic function); the wave equation just has a richer homogeneous family — all of free radiation.

Why is it significant that one operator governs all four potentials?

Because operators are the fingerprints of geometry. In Newtonian mechanics, time and space transform separately and nothing forces temporal and spatial derivatives into a fixed package; the d'Alembertian locks them together with one constant, $c$ , and treats $(V/c, \mathbf{A})$ as four components of one object driven by four components $(c\rho, \mathbf{J})$ of another. Lorentz transformations are precisely the linear maps that preserve this operator, so "the same $\Box^2$ acts on all four" is the statement that electrodynamics has Lorentz symmetry, seen before relativity is formally introduced. The relativity chapter will run the logic forward: promote $A^\mu$ and $J^\mu$ to four-vectors, and the four equations of this article compress into $\Box^2 A^\mu = -\mu_0 J^\mu$ — one line containing all of Maxwell.

Is the domain-of-dependence theorem not circular — do we not already know signals travel at c?

No prior knowledge is used, and that is the theorem's value. Derivation A manipulates only the equation itself: an energy functional, integration by parts, one inequality. Nothing about Green functions, retarded integrals, or electromagnetic phenomenology enters. The conclusion — influence confined to the cone — is therefore a property of the operator, inherited by anything the operator governs: potentials, fields, gravitational perturbations, acoustic pressure (with $c_s$ in place of $c$ ). This is also why the theorem is the right tool for the previous article's constraint-propagation argument and for numerical stability analysis: it holds for every solution, including ones no formula can express. The retarded integral, when it arrives, is one explicit solution consistent with the theorem — not its source.

Where exactly does the retarded-versus-advanced choice enter, if the equation is time-symmetric?

In the data, and nowhere else — this article is deliberately careful about it. The operator $\Box^2$ is invariant under $t \to -t$ ; Derivation B accordingly produced a symmetric pair of Green functions and could fix only their combined strength. The asymmetry enters through condition (4): demanding quiescent past (nothing before sources act) or outgoing behavior at infinity (Sommerfeld) selects $G_{\text{ret}}$ ; the mirror conditions would select $G_{\text{adv}}$ . So "nature uses retarded potentials" is really "the problems nature hands us have retarded-type data" — sources switch on, radiation leaves, infinity does not conspire to send waves inward. Why the universe supplies that kind of data is a genuine arrow-of-time question, and it is prosecuted properly in Advanced Potentials and Causality, two articles from now.

What is special about three dimensions in the strong Huygens principle?

The Green function's support. Solve the wave equation in $n$ space dimensions and examine the point-impulse response: for odd $n \ge 3$ it is concentrated on the light cone (in 3D, the clean $\delta(t - r/c)/4\pi r$ ); for even $n$ — and for $n = 1$ — it fills the cone's interior, decaying like an afterglow. The mechanism is the method of descent: a 2D problem is a 3D problem with a line source, and the line keeps delivering delayed contributions forever. Consequences are physical and audible: a lightning strike heard through the air (3D) is a crack; the same energy fed into a membrane or shallow water (2D) rings with a long tail. Dispersive media break the sharpness in any dimension by making $c$ frequency-dependent — vacuum's dispersion-free cones are what keep a light pulse a pulse over gigaparsecs.

How do these equations know about charge conservation?

Through the gauge condition's self-consistency, and the operator makes the check one line. Apply $\partial_t$ to the $V$ equation (times $\mu_0\varepsilon_0$ ) and $\nabla\cdot$ to the $\mathbf{A}$ equation, and add: the left side assembles $\Box^2(\nabla\cdot\mathbf{A} + \mu_0\varepsilon_0\partial_t V)$ — the d'Alembertian of the Lorenz combination — while the right side assembles $-\mu_0(\partial_t\rho + \nabla\cdot\mathbf{J})$ . If charge is conserved, the Lorenz condition can hold for all time (the previous article's persistence theorem); if it is not, the system is inconsistent — no solution respects the gauge. The covariant version is even terser: the four-divergence of $\Box^2A^\mu = -\mu_0 J^\mu$ under $\partial_\mu A^\mu = 0$ reads $0 = \partial_\mu J^\mu$ . Sources that fail continuity cannot be coupled to the potential formulation at all — a fact already met, from the gauge side, in Gauge Transformations.

Do I ever need the advanced Green function for a real calculation?

Yes — as scaffolding, and occasionally as physics. Scaffolding: scattering theory and QED build propagators (Feynman's, notably) as specific combinations of retarded and advanced kernels, because the boundary conditions natural to particle physics ("positive frequencies forward in time") differ from radiation theory's quiescent past. The radiative field $F_{\text{ret}} - F_{\text{adv}}$ , half of Dirac's classical electron program, needs both. Physics: the Wheeler–Feynman absorber theory takes half-retarded-plus-half-advanced as fundamental and recovers ordinary retarded phenomenology from the response of distant matter — the boldest answer ever offered to the previous FAQ's arrow-of-time question, and the next-but-one article's subject. For garden-variety antenna and radiation problems, though: retarded only, with the advanced twin standing as the reminder that causality was your input, not your output.