Covariant Wave Equation and Retarded Solutions

The E9 chapter solved electrodynamics: retarded potentials, verified against the wave equation, feeding the entire radiation ladder. But its machinery was built in one frame — a chosen $t$ , a chosen $r$ , a "retarded time" defined by that observer's clock — and the relativity chapter has since taught us to distrust any construction that names a frame. So a debt has been accumulating quietly for seventeen articles: is the retarded solution itself relativistic? If two observers each solve $\Box A^\mu=\mu_0J^\mu$ with their own retarded prescription, do they describe the same field? This article pays the debt, and the payment is beautiful. The retarded Green function — E9's spherical-shell impulse response — can be rewritten as a manifestly invariant object, $D_{\text{ret}}(x)=\theta(x^0)\,\delta(x\cdot x)/2\pi$ : a delta function on the light cone gated by a step function into the future. The delta factor is a Lorentz scalar by inspection. The step function looks frame-tied — until you notice it only ever acts on null separations, and the invariant-order theorem of Spacetime and Minkowski Geometry says null order is absolute. The retarded prescription is the same prescription for every observer because causal order on the light cone is frame-proof — the certificate that article promised the radiation chapter, now cashed where it counts. With the invariant kernel in hand, everything E9 built collapses out of one integral: the retarded potentials for general sources, and — evaluated on a point charge's worldline — the Liénard–Wiechert potential in its covariant form $A^\mu\propto u^\mu/u\cdot(x-r)$ , the third route to that celebrated result. Sharp signals, frame-proof causality, and one Green function that quantum field theory will inherit almost unchanged.

Key takeaways

In Lorenz gauge the problem is $\Box A^\mu=\mu_0J^\mu$ — four wave equations, solved once and for all by one Green function: $\Box D=\delta^4(x)$ .
The retarded solution is manifestly invariant: $D_{\text{ret}}(x)=\tfrac{1}{2\pi}\theta(x^0)\,\delta(x\cdot x)$ — supported exactly on the forward light cone.
$\delta(x\cdot x)$ is a scalar by construction; $\theta(x^0)$ is legitimate because it acts only on null separations, whose time order is absolute (Spacetime and Minkowski Geometry's invariant-order theorem, cashed here) — retardation is frame-proof.
Unpacking $\delta(x^2)=[\delta(ct-r)+\delta(ct+r)]/2r$ shows the invariant kernel is E9's impulse response $\delta(t-r/c)/4\pi r$ ; convolved with any source it reproduces the retarded potentials exactly.
On-cone support is special to 3+1 dimensions: signals propagate sharply at $c$ , with no afterglow — Huygens' principle as an algebraic accident of our dimension.
Evaluated on a point-charge worldline, the integral collapses to the covariant Liénard–Wiechert potential $A^\mu=\mu_0qc\,u^\mu/[4\pi\,u\cdot(x-r)]$ — E9's result by a third route; its full covariant anatomy is the E11 treatment's business.
Key idea: one invariant Green function — a delta on the forward light cone — solves covariant electrodynamics, and the absoluteness of null causal order is what makes "retarded" a frame-independent word.

The core insight

The covariant wave equation $\Box A^\mu=\mu_0J^\mu$ is solved by the invariant retarded Green function $D_{\text{ret}}(x-x')=\tfrac{1}{2\pi}\theta(x^0-x'^0)\,\delta[(x-x')^2]$ — supported on the forward light cone, frame-proof because null order is absolute — and every retarded potential of E9, including Liénard–Wiechert, is its convolution with the source.

Where this lives today

The invariant kernel is the classical object that quantum field theory promoted into its central tool. QED's photon propagator is this article's $D$ with one modification — Feynman's $i\epsilon$ prescription, which gates the cone not by $\theta(x^0)$ but by a rule that mixes retarded and advanced pieces for virtual quanta while preserving macroscopic causality; the commutator functions that enforce "no signaling outside the cone" in QFT are differences of the retarded and advanced kernels met here. Numerical relativity and computational electrodynamics live downstream too: FDTD solvers respect a discrete version of the cone (the Courant condition is nothing but "the numerical domain of dependence must contain the physical cone"), and gravitational-wave extraction uses retarded Green functions on curved backgrounds where the sharp-Huygens property fails — tails scatter off curvature, an observable relic of how special the flat 3+1 delta-on-the-cone really is. And pulsar timing arrays operate Example 1 at galactic scale: every millisecond-pulsar tick arrives on a forward cone thousands of years old, and the correlated delays across many cones are how nanohertz gravitational waves were finally seen.

Origins · 1 minJulian Schwinger and how this was discovered

Julian Schwinger

1918–1994 · New York and Cambridge, Massachusetts

Schwinger is the physicist who made Green functions the language of relativistic field theory rather than a technique within it. A child prodigy who published at sixteen and ran wartime radar-waveguide theory at the MIT Radiation Lab — where computing field responses to sources was the daily bread that sharpened his Green-function instincts — he returned to fundamental physics convinced that the response function, not the field configuration, was the primary object. His covariant formulation of QED (1948–51, the work sharing the 1965 Nobel with Feynman and Tomonaga) is built on exactly the objects of this article: invariant kernels supported by causal structure, generalized into the "Schwinger functions" and proper-time representations that still carry his name. The famous Schwinger effect — vacuum sparking into electron-positron pairs in a strong field — came from treating the electromagnetic background by these methods. Legendarily formal in style (he lectured in polished paragraphs, worked alone, and reputedly used Feynman diagrams only in private), he supervised over seventy PhD students, four of them future Nobel laureates. His dictum fits this article as a caption: the Green function is the physics — the rest is source bookkeeping.

History

1867

— Ludvig Lorenz publishes retarded potentials: source news travelling at finite speed

1898

— Liénard, then Wiechert (1900), evaluate the retarded potentials of a point charge

1909

— Ritz and Einstein publicly split over retarded-versus-advanced as law or thermodynamics

1938

— Dirac's classical electron paper wields the invariant

\theta

-gated cone functions

1945

— Wheeler and Feynman's absorber theory makes the advanced kernel earn honest work

1948

— Schwinger's covariant QED: the retarded kernel's quantum descendants take over physics

The physical picture

Picture an event where charge twitches — a spark at one point of spacetime. Where can the potential respond? The Green function is the answer to exactly this question: the field's response to a unit impulse, from which every response is assembled by superposition (the fields are linear; the covariant system said so). E9's Inhomogeneous Wave Equations for Potentials answered in 3+1 language: an expanding spherical shell, all the response concentrated at radius $r=ct$ , weight falling as $1/r$ . Stack the snapshots of that expanding shell into spacetime and the picture becomes geometric: the shell traces the forward light cone of the spark. The 3+1 answer was a cone all along, sliced by one observer into spheres. That restacking is the whole article. On the cone, weight $1/r$ ; off the cone — inside or outside — exactly nothing. "On the cone" is the invariant statement $(x-x')^2=0$ ; "forward" is $x^0>x'^0$ . So the response function must be built from a delta pinning separations to the cone and a step function selecting its future half:

D_{\text{ret}}\;\sim\;\theta(\text{future})\,\times\,\delta(\text{on the cone}),

and the factor $1/r$ will turn out to live inside the delta function's normalization — the first derivation makes the bookkeeping exact. Nothing about a cone names a frame; the only worry is the word "future," and that worry is precisely what the Minkowski-geometry article's causal-order theorem was built to retire: for null (and timelike) separations, every inertial observer agrees which event came first. Retardation — effect after cause, always — is not one observer's convention. It is light-cone geometry.

Figure

The impulse response is a light cone — E9's expanding shell, restacked into spacetime

Hold one more expectation before the algebra: the on-cone sharpness is a gift of living in three spatial dimensions. Solve the same wave equation in a two-dimensional world and the impulse response fills the interior of the cone — a flash of light would be followed by an infinite, decaying afterglow, and "the speed of light" would mark only the front, not the signal. Every crisp image on your retina, every clean radio pulse, exploits the algebraic accident that the 3+1 Green function is a delta on the cone rather than a function inside it. The FAQ returns to why.

The problem, and its invariant solution

In Lorenz gauge the covariant system reduces (per The Four-Potential, with the gauge landscape mapped by Gauge Transformations and Covariant Gauge Conditions) to

\Box A^\mu=\mu_0 J^\mu, \qquad (1)

where: $\Box=\partial_\nu\partial^\nu$ is the d'Alembertian — a Lorentz scalar operator, the fact Four-Vectors and the Metric Tensor engineered — and the four components decouple: one scalar problem, solved four times. Linearity reduces everything to the unit-impulse problem

\Box\,D(x)=\delta^4(x), \qquad (2)

where: $\delta^4(x)=\delta(x^0)\delta^3(\mathbf{x})$ is the four-dimensional delta — itself a Lorentz scalar, since the Jacobian of a Lorentz transformation is one. Given any $D$ solving (2), the potential is the superposition $A^\mu(x)=\mu_0\!\int d^4x'\,D(x-x')\,J^\mu(x')$ . Equation (2) has many solutions differing by free waves; physics selects among them by a causal boundary condition, and the selection is where relativity has to be consulted.

Derivation AThe invariant kernel — the retarded Green function is a delta on the forward light cone

We show that E9's retarded impulse response and the manifestly invariant expression $\theta(x^0)\delta(x\cdot x)/2\pi$ are the same object, then audit the expression's Lorentz credentials — including the step function, whose invariance is not automatic and rests on the causal-order theorem. The strategy is deliberately economical: the hard analytic work (solving the impulse problem by spherical symmetry) was done in 3+1 language by E9's Inhomogeneous Wave Equations for Potentials and is quoted, not repeated; this derivation's work is the rewriting and the certificate. Watch the $1/r$ weight — the signature of energy conservation on expanding shells — emerge from nothing but the delta function's composition rule.

Step 1. Quote the 3+1 result. The retarded solution of (2) — response after the impulse, spherical shell at $r=ct$ — is, per Inhomogeneous Wave Equations for Potentials (in coordinates $x^0=ct$ ):

D_{\text{ret}}(x)=\frac{\delta(x^0-r)}{4\pi r},\qquad r=|\mathbf{x}| .

Step 2. Rewrite via delta composition. The invariant interval factors as $x\cdot x=(x^0)^2-r^2=(x^0-r)(x^0+r)$ , so the composition rule $\delta(f)= \sum_{\text{roots}}\delta(x-x_i)/|f'(x_i)|$ gives

\delta(x\cdot x)=\frac{\delta(x^0-r)+\delta(x^0+r)}{2r} .

Why the $1/r$ was never a separate ingredient: the shell weight is the derivative of the interval across its zero — geometry of the cone supplies the amplitude law that 3+1 language attributes to energy spreading.

Step 3. Gate and assemble. Multiplying by $\theta(x^0)$ kills the $\delta(x^0+r)$ root (the backward cone) and leaves exactly Step 1's kernel:

\frac{1}{2\pi}\,\theta(x^0)\,\delta(x\cdot x) =\frac{\delta(x^0-r)}{4\pi r} =D_{\text{ret}}(x). \qquad (3)

The advanced solution is the same object gated to the past, $D_{\text{adv}}=\theta(-x^0)\delta(x\cdot x)/2\pi$ — the pairing whose physics (and whose absorber-theory afterlife) E9's Advanced Potentials and Causality owns.

Step 4. The covariance audit — the article's point. $\delta(x\cdot x)$ is a scalar: its argument is the invariant interval. $\theta(x^0)$ , by itself, is not: for spacelike separations, observers disagree about temporal order, and a naive step function would be frame-dependent. But in the product (3), $\theta$ is only ever evaluated where $\delta(x\cdot x)$ has support — on null separations — and there the invariant-order theorem of Spacetime and Minkowski Geometry applies: the time order of null-separated events is absolute (the forward cone maps to the forward cone under every orthochronous Lorentz transformation). So $\theta(x^0)\,\delta(x\cdot x)$ is invariant as a whole: every observer's "retarded prescription" gates the same half of the same cone. This is the frame-consistency certificate for the entire E9 retarded machinery — promised by the geometry article as the reason retarded potentials would survive the relativity chapter, and delivered here at the level of the solution kernel itself. Verified: composition-rule algebra checked (both roots, Jacobian $2r$ ); (3) reproduces the IWEQ kernel exactly; the certificate's logic cross-checked against the geometry article's boost-bound proof and Zangwill §23; factors match Jackson §12.11 (his $D_r$ , mod SI bookkeeping).

Go deeper · Derivation B · 2 minConvolve and collapse — E9's retarded potentials, and Liénard–Wiechert by the third route

Derivation BConvolve and collapse — E9's retarded potentials, and Liénard–Wiechert by the third route

We feed the invariant kernel a general source and recover E9's retarded potentials exactly; then we feed it the most important special source — a point charge on an arbitrary worldline — and watch the four-dimensional integral collapse, in two lines, onto the covariant Liénard–Wiechert potential. The strategy is delta-function bookkeeping: each delta spends one integral. Watch what each collapse costs: the general case spends $\delta(x^2)$ on the time integral (producing "evaluate at retarded time"); the point-charge case spends it on the proper-time integral, and the Jacobian it leaves behind is the invariant $u\cdot(x-r)$ — the covariant ancestor of E9's famous $(1-\hat{\mathbf{n}}\cdot\boldsymbol\beta)$ factor.

Step 1. General source. With $d^4x'=c\,dt'\,d^3x'$ :

A^\mu(x)=\mu_0\!\int d^4x'\,D_{\text{ret}}(x-x')J^\mu(x') =\frac{\mu_0}{4\pi}\!\int d^3x'\,c\!\int dt'\,\frac{\delta\!\big(c(t-t')-R\big)}{R}\,J^\mu(x'),

with $R=|\mathbf{x}-\mathbf{x}'|$ , using form (3). The $t'$ integral spends the delta at $t'=t-R/c$ (cost: a factor $1/c$ ):

A^\mu(t,\mathbf{x})=\frac{\mu_0}{4\pi}\int d^3x'\,\frac{J^\mu(t-R/c,\,\mathbf{x}')}{R}\,. \qquad (4)

This is E9 verbatim: the scalar slot is Retarded Potentials' $V$ , the space slots its $\mathbf{A}$ — that article's verification against the wave equation and its Lorenz-condition theorem ( $\partial\cdot A=0$ holds exactly when $\partial\cdot J=0$ ) transfer to the covariant solution unchanged, and are recalled, not re-proved. The whole E9 chapter is downstream of formula (4); formula (4) is downstream of a delta on a cone.

Step 2. Point charge. The covariant current of a charge on worldline $r^\mu(\tau)$ is the proper-time integral $J^\mu(x)=qc\int d\tau\,u^\mu(\tau)\,\delta^4\big(x-r(\tau)\big)$ (the ladder principle of Proper Time and Four-Velocity choosing $\tau$ as the invariant parameter). Insert into the convolution; the $\delta^4$ spends the $d^4x'$ integral:

A^\mu(x)=\frac{\mu_0qc}{2\pi}\int d\tau\;u^\mu(\tau)\,\theta\big(x^0-r^0(\tau)\big)\,\delta\!\big[(x-r(\tau))^2\big].

Step 3. Spend the last delta on $\tau$ . Its argument vanishes where the worldline pierces the field point's light cone; the $\theta$ selects the past piercing, which is unique for any subluminal worldline (the uniqueness theorem E9's Liénard–Wiechert Potentials proved and this construction inherits). The composition rule bills the Jacobian $\big|\tfrac{d}{d\tau}(x-r)^2\big|=2\,u\cdot(x-r)$ :

A^\mu(x)=\frac{\mu_0\,q\,c\;u^\mu}{4\pi\;u\cdot(x-r)}\Bigg|_{\tau=\tau_{\text{ret}}} . \qquad (5)

Two lines, manifestly covariant, no frame chosen — the third route to Liénard–Wiechert, joining E9's apparent-volume and 3+1-delta routes (both recalled by name). Reading the invariant denominator — how $u\cdot(x-r)$ factors into "retardation correction × distance," the gradient lemma $\partial_\mu\tau_{\text{ret}}$ , the uniform-motion reduction, and the field tensor that follows — is the anatomy owned by the E11 treatment Liénard–Wiechert Potentials: Covariant Formulation, to which (5) is handed as the starting object. Verified: the collapse bookkeeping (which delta spends which integral, and each Jacobian) rechecked symbolically; (4) matched line-by-line against the E9 statement; (5)'s 3+1 reduction spot-checked at closest approach for uniform motion ( $V=\gamma q/4\pi\varepsilon_0 b$ — the boost-route value of The Four-Potential, worked example 2 below).

Figure

One kernel, two collapses — the whole retarded machinery downstream of a delta on a cone

Limiting cases

case	result	meaning
static source	(4) → $V=q/4\pi\varepsilon_0r$ , $\mathbf{A}$ steady	retardation invisible when nothing changes; Coulomb recovered
slowly varying source	expand $J(t-R/c)$ in $R/c$	quasistatics + first radiation corrections — E9's zone hierarchy
point charge, uniform motion	(5) → boosted Coulomb	The Four-Potential's boost route, cross-checked
$\theta(x^0)\to\theta(-x^0)$	advanced solution	runs the movie backward; physics of the choice owned by E9's causality article
2+1 dimensions	support fills the cone's interior	afterglow; sharp Huygens is a 3+1 privilege
massive photon ( $\Box+m^2$ )	support leaks inside the cone	dispersion; sharp cone ⟺ masslessness

Common trapThe interval delta has two roots — and the step function is doing load-bearing causal work

The compact form $\delta(x\cdot x)$ seduces people into two distinct errors. First: forgetting that the interval factors through both cones, $\delta(x^2)=[\delta(x^0-r)+\delta(x^0+r)]/2r$ . Drop the $\theta(x^0)$ gate and your "solution" is the half-sum of retarded and advanced kernels — a legitimate solution of equation (2), but one that makes potentials respond before sources move; every convolution you build afterwards will smear cause and effect symmetrically and no error message will appear. Second: treating the gate as trivial. $\theta(x^0)$ alone is not Lorentz-invariant — on spacelike separations observers genuinely disagree about order — and quoting " $\theta(x^0)\delta(x^2)$ is invariant" without the null-support argument is a gap, not a proof; the invariance rests specifically on the absoluteness of light-cone order (Derivation A, Step 4). A useful drill for both traps at once: check any Green-function formula against the static limit and against a moving observer. The half-sum fails the first honestly (it still gives Coulomb — which is why statics never catches causality errors) but pollutes any time-dependent problem; a mis-gated kernel fails the second. In 3+1 electrodynamics, causality lives entirely in that one unglamorous $\theta$ .

Worked examples

Example 1 — retardation you can time. A charge $q=1.0$ nC at the origin is switched into a new configuration at $t=0$ (say, moved abruptly by a millimeter). An observer at $r=30$ cm monitors the potential.

Step 1. Before and long after the switch, formula (4) is Coulomb: $V=q/4\pi\varepsilon_0r=29.96$ V at the old position's distance.

Step 2. The kernel's support says the observer's potential cannot respond before the forward cone of the switch event arrives: $\Delta t=r/c=1.0$ ns. For exactly one nanosecond the observer's instruments — potential, fields, everything — hold the old values, unaware. No adjustable constant appears anywhere in that statement; the delay is pure cone geometry.

Step 3. The frame-proof reading: any moving observer present at the same monitoring event also finds the news undelivered, because "the monitoring event is outside the switch's forward cone" is an invariant statement (Derivation A, Step 4). Compare the trap the E9 Coulomb Gauge article defused: gauge-dependent potentials can look instantaneous, but the invariant kernel of the Lorenz-gauge solution wears causality on its support.

Example 2 — the covariant LW potential, evaluated. A charge $q=1.0$ nC moves uniformly at $\beta=0.8$ ( $\gamma=5/3$ ) along $x$ . Find the scalar potential at the moment of closest approach, at perpendicular distance $b=10$ cm.

Step 1. For uniform motion, formula (5) can be resummed into present-position form (the reduction owned by the E11 covariant-LW treatment; equivalently, boost the Coulomb potential per The Four-Potential): at closest approach the denominator's invariant evaluates to $u\cdot(x-r)=c\,b/\gamma\cdot\gamma=cb$ ... done honestly via the boost: $V=\gamma\,q/4\pi\varepsilon_0 b$ .

Step 2. Numbers: $V=\tfrac{5}{3}\times8.99\times10^9\times10^{-9}/0.10=1.50\times10^{2}$ V — versus $89.9$ V for a static charge at the same distance: the moving charge's potential at broadside is $\gamma$ -enhanced (the pancaking of Transformation of Electric and Magnetic Fields, seen in the potential). The vector potential rides along: $A_x=\beta V/c=4.0\times10^{-7}$ V·s/m.

Step 3. The consistency web, which is the real lesson: this one number is now certified by three independent routes — E9's Liénard–Wiechert evaluation, the potential-boost of The Four-Potential, and the invariant-kernel collapse (5) — built respectively from 3+1 geometry, from transformation theory, and from cone-supported Green functions. Three formalisms, one answer, no conspiracy: that is what "the retarded machinery is covariant" means in practice.

Going further

Reading. Jackson, Classical Electrodynamics, 3rd ed., §12.11 — the invariant Green function. Zangwill, Modern Electrodynamics, §23. Landau & Lifshitz, Classical Theory of Fields, §62–63. Barut, Electrodynamics and Classical Theory of Fields and Particles, Ch. 4 — the proper-time collapse in detail.

Provenance. Built from the AllQFT base article for this topic (invariant kernel, cone support, covariant LW collapse retained), conventions locked to this track's $(+,-,-,-)$ dictionary. A verifier should: (1) recheck the composition-rule unpacking $\delta(x^2)=[\delta(x^0-r)+\delta(x^0+r)]/2r$ and the identity (3) against the IWEQ kernel; (2) confirm the null-support invariance argument for $\theta(x^0)\delta(x^2)$ ; (3) rerun both collapses of Derivation B (general source → E9's (4); point charge → (5) with Jacobian $2u\cdot(x-r)$ ); (4) recompute the examples ( $V=29.96$ V and $\Delta t=1.0$ ns at 30 cm; $V=\gamma q/4\pi\varepsilon_0b=150$ V and $A_x=4.0\times10^{-7}$ V·s/m at $\beta=0.8$ , $b=10$ cm, vs $89.9$ V static). Division of labor for this edition: the 3+1 impulse response and domain-of-dependence are owned by E9's Inhomogeneous Wave Equations for Potentials (quoted at Step 1); the retarded potentials' verification and Lorenz-condition theorem by Retarded Potentials; the two 3+1 LW routes, the retarded-time uniqueness theorem, and the beaming ledger by Liénard–Wiechert Potentials; the retarded-versus-advanced physics by Advanced Potentials and Causality; the causal-order theorem cashed here by Spacetime and Minkowski Geometry; the covariant LW potential's full anatomy — invariant denominator unpacked, $\partial_\mu\tau_{\text{ret}}$ lemma bank, uniform-motion present-position form, and the fields that follow — by E11's Liénard–Wiechert Potentials: Covariant Formulation and Fields of a Relativistic Moving Charge, to which equation (5) is handed. This edition owns the invariant rewriting, the covariance certificate at kernel level, and the two collapses.

Builds on: The Four-Potential · Gauge Transformations and Covariant Gauge Conditions · Covariant Form of Maxwell's Equations · Spacetime and Minkowski Geometry · Four-Vectors and the Metric Tensor

Leads to: Relativistic Doppler Effect and Aberration · Multipole Expansion of Radiation Fields · Liénard–Wiechert Potentials: Covariant Formulation

Is the retarded time that different observers compute the same event on the source's worldline?

Yes — and this is the sharpest way to state the article's certificate. The retarded event is defined geometrically: the intersection of the source worldline with the field point's backward light cone. Both ingredients — the worldline and the cone — are frame-independent objects; only their coordinates differ between observers. So all observers agree on which emission event is responsible for the field at a given field event; they merely label its time differently. Contrast a fake alternative: "the source position 5 nanoseconds ago" names a frame (whose 5 ns?) and would be genuinely ambiguous. The cone construction never does. The uniqueness of the intersection — exactly one retarded event for any subluminal worldline — is E9's theorem, and it is what makes formula (5) a definition rather than a choice.

Why does the wave equation in three spatial dimensions produce a sharp cone while other dimensions leave afterglow?

The clean way to see it: solve in $n$ spatial dimensions and the impulse response is (schematically) the $\big(\tfrac{n-3}{2}\big)$ -th derivative of a delta on the cone for odd $n\geq3$ , but a function with an inverse-square-root tail inside the cone for even $n$ — the famous even/odd split of Huygens' principle. A slick argument for the even-dimension failure: a 2D world is a 3D world with sources extended uniformly along the suppressed axis, so a 2D "flash" is really an infinite 3D line source — and distant segments of the line keep delivering signal forever, which is the afterglow. Three spatial dimensions sit in the privileged odd family (and, among them, are the lowest with stable orbits and interesting chemistry). Practical corollaries are real: dispersion-free communication, clean pulse propagation, and the fact that "the sound of thunder" — governed by the same operator — arrives as a rumble mostly because of atmospheric scattering and the extended source, not the kernel; in flatland it would rumble intrinsically.

Is choosing the retarded kernel over the advanced one a law of nature?

Not at the level of equation (2) — both gates solve it, as does any mixture, and the electrodynamic laws themselves are time-symmetric. The retarded choice is a boundary condition encoding which problems we pose: "fields produced by these sources, given no incoming radiation." E9's Advanced Potentials and Causality owns the full discussion — including the Wheeler–Feynman absorber program, in which emitters use the symmetric half-sum and the universe's absorbers conspire to make the net field retarded, tying the choice to thermodynamic arrows. What this article adds is a consistency fact often left unstated: whichever gate you choose, the choice is frame-proof (both halves of the cone are invariant), so "retarded" is a possible law-like statement — relativity does not force it, but relativity also never garbles it.

How does the Feynman propagator of QED differ from the retarded kernel built here?

By its gate, chosen for a different question. The retarded kernel answers a classical initial-value question: field now, given sources before. QED's central objects are transition amplitudes, and Feynman's propagator gates the cone by frequency instead of time order: positive-frequency parts propagate forward, negative-frequency (antiparticle) parts backward — implemented by the $i\epsilon$ shift of the poles rather than a $\theta$ function. The bookkeeping consequences: the Feynman function has support (exponentially small) even at spacelike separation, which alarms newcomers but violates nothing — signals are carried by commutators, which are retarded-minus-advanced and vanish exactly outside the cone, this article's kernels doing the enforcement. All the standard functions — retarded, advanced, Feynman, Wightman — are one cone-supported family differing by gates; classical electrodynamics uses the member with the classical question built in.

Can the same Green-function method deliver the fields directly, skipping potentials?

Yes — apply the same kernel to the wave equations satisfied by the fields themselves, and the result is Jefimenko's equations: $\mathbf{E}$ and $\mathbf{B}$ as retarded integrals over charge, current, and their time derivatives. E9's Jefimenko's Equations owns those formulas and their lessons (notably: fields depend on retarded rates, which is why quasi-static intuition fails in the radiation zone). The potential route remains the workhorse for three reasons: it is four scalar problems instead of six coupled components; the Lorenz-gauge system is exactly the clean form (1); and the covariant collapse to (5) has no equally compact field-level analogue — the fields of a point charge come most efficiently by differentiating (5), the program the E11 covariant-LW pair executes. Method summary: potentials for structure and covariance, Jefimenko when you want fields' causal anatomy exhibited.

Does the invariant-kernel construction survive inside a material medium?

The method survives; the kernel does not. In a linear dispersive medium the wave operator acquires the medium's response — effectively $n(\omega)$ in the frequency domain — and the Green function's support fills in behind a front: the front still travels at $c$ (medium response cannot outrun vacuum causality — the Kramers–Kronig/front-velocity story of E8's absorption-dispersion article), but signal energy arrives spread over the Sommerfeld–Brillouin precursors rather than as a clean delta. Covariance also loses its manifest character: the medium's rest frame is physically distinguished, so kernels are invariant only under the trivial group unless you carry the medium four-velocity as an ingredient — precisely the bookkeeping the next-but-one article (Tensor Formulation of Fields in Media) sets up with $u^\mu_{\text{medium}}$ entering the constitutive relations. The vacuum cone kernel built here is the reference object all of that machinery must reduce to as $n\to1$ .

The collapse in Derivation B treated the charge as exactly pointlike — where does that idealization bite?

Two places, one benign and one famous. Benign: for extended charge distributions, formula (4) simply integrates the kernel over the structure, and "the" retarded time becomes a range — nothing conceptual changes, the cone bookkeeping just gains an extra integral. Famous: for the self-field of a point charge, equation (5) is singular on the worldline itself, and asking "what force does the charge's own retarded field exert back on it" produces the divergent self-energy and the radiation-reaction pathologies that E9's Radiation Reaction Force and Abraham–Lorentz Equation dissected — with the E11 classical-electron set performing the full autopsy. The Green-function viewpoint localizes the disease precisely: the kernel is perfectly finite between distinct points; every pathology of classical electron theory lives at the single point where field point meets source point on the cone's tip. Renormalization, classical and quantum, is the art of subtracting that tip.