Retarded Potentials

The machinery is assembled; this is the article where it produces the answer. The Lorenz-gauge potentials obey four copies of the inhomogeneous wave equation, and the last article built that operator's elementary response — a spherical delta shell expanding at $c$ . Superpose the shells over a real source and out comes the most quietly beautiful formula in classical electrodynamics: the potential now, here is the familiar static integral — same $1/|\mathbf{r}-\mathbf{r}'|$ kernel, same superposition — with one change. Each source element is evaluated not at the present time but at its own retarded time $t_r = t - |\mathbf{r}-\mathbf{r}'|/c$ : the moment its news, traveling at $c$ , had to leave in order to arrive at the observation event. The formula looks like a guess a good student might make — "statics, but delayed" — and that is exactly why it must be earned. The flagship derivation of this article differentiates the retarded integral honestly, retarded-time dependence and all, and watches the wave equation emerge through a near-miraculous cancellation. The same honest differentiation immediately issues a warning: the guess that works for potentials fails for fields. Retarded Coulomb is not the electric field; the leftover terms are radiation itself, and chasing them is the next four articles. Everything in radiation theory descends from the two integrals earned here. Specialize to a point charge and the Liénard–Wiechert potentials fall out; differentiate and Jefimenko's equations appear; expand in source size over wavelength and the multipole ladder — dipole radiation first — unrolls. This is the causality spine of the chapter, and this article is its base vertebra: the exact, checkable statement that electromagnetic news travels at $c$ , no faster, from every element of every source.

Key takeaways

The retarded potentials solve $\Box^2 V = -\rho/\varepsilon_0$ and $\Box^2\mathbf{A} = -\mu_0\mathbf{J}$ : V(\mathbf{r},t) = \frac{1}{4\pi\varepsilon_0}\int\frac{\rho(\mathbf{r}',t_r)}{\mathscr{r}}\,d^3r', \qquad \mathbf{A}(\mathbf{r},t) = \frac{\mu_0}{4\pi}\int\frac{\mathbf{J}(\mathbf{r}',t_r)}{\mathscr{r}}\,d^3r', with $\mathscr{r} = |\mathbf{r}-\mathbf{r}'|$ and $t_r = t - \mathscr{r}/c$ .
The structure is "statics with stale news": the static kernels survive intact; only the sampling time changes, and it changes per source element — near elements report fresh, far elements report old.
The formula is verified, not guessed: differentiating under the integral (the derivatives act on $t_r$ too) reproduces the wave equation exactly, with the dangerous middle terms cancelling.
The pair satisfies the Lorenz condition if and only if charge is conserved — continuity is the glue holding $V$ and $\mathbf{A}$ in the same gauge.
The identical recipe with $t_a = t + \mathscr{r}/c$ (advanced potentials) also solves the equations; retardation is selected by causal data, prosecuted properly next article.
The fields are NOT the static laws with $t \to t_r$ : differentiation generates extra $\dot\rho$ and $\dot{\mathbf{J}}$ terms — Jefimenko's equations — and those terms are the radiation.
Key idea: the dynamical potential is the static integral with each source element evaluated at its own retarded time $t_r = t - \mathscr{r}/c$ — every contribution arrives having traveled at exactly the speed of light.

The core insight

The dynamical potential is the static Coulomb/Biot–Savart integral with the source evaluated not at the present time but at the retarded time $t_r = t - |\mathbf{r}-\mathbf{r}'|/c$ , so that every contribution arrives having traveled at exactly the speed of light.

Where this lives today

The retarded integral is production code. The method-of-moments solvers behind every antenna design (NEC and its descendants), the boundary-element codes for radar cross-sections and electromagnetic compatibility, and the transient solvers for lightning and pulsed-power all evaluate exactly equation (1)'s integrals — the engineering literature calls the frequency-domain kernel $e^{ikr}/4\pi r$ and the discipline of tracking $t_r$ across a meshed source is their central numerical difficulty. When your phone negotiates with a tower, both ends were designed by machines summing stale news over past light cones. The archive picture also runs the largest experiments in science. Very-long-baseline interferometry — the Event Horizon Telescope's black-hole images — is the art of correlating one wavefront's retarded arrival times across the Earth; pulsar timing arrays read decades of retarded pulses to hunt nanohertz gravitational waves; and every deep-space link budgets its $\mathscr{r}/c$ in hours. Even the definition of the meter now runs through this article: distance is retardation, $c$ times a measured delay. Physics upstream, the retarded structure is the classical skeleton of the photon propagator — QED's perturbation series is built on Green functions whose real part is precisely the retarded/advanced pair of this chapter.

Origins · 1 minBernhard Riemann and how this was discovered

Bernhard Riemann

1826–1866 · Göttingen — who wrote the retarded potential in 1858, then put it in a drawer

Riemann is mathematics' Riemann — the hypothesis, the surfaces, the curvature tensor that would later hold up general relativity — but in 1858 he presented to the Göttingen Academy a short paper on electrodynamics containing the retarded potential: the proposal that the electric potential propagates from its sources at a finite speed, which he computed from electrical units and found "very nearly the velocity of light." He withdrew the paper (apparently dissatisfied with a mathematical point), and it appeared only posthumously in 1867 — in the same volume of Annalen der Physik, by editorial coincidence, as Ludvig Lorenz's independent and more complete retarded-potential electrodynamics. Riemann died of tuberculosis at 39, a year before publication. The episode is a study in how ideas arrive before their time: a decade ahead of any experimental hint that electromagnetic influence propagates, two decades before Hertz, Riemann had already written down the mathematical object this article verifies — the stale-news integral, with $c$ inside it.

History

1858

— Riemann presents the retarded potential to the Göttingen Academy; withdrawn, published posthumously 1867

1867

— Lorenz's retarded-potential electrodynamics appears, deriving light as electromagnetic disturbance independent of Maxwell

1888

— Hertz's spark-gap experiments confirm that electromagnetic news indeed travels at c

1898

— Liénard specializes the retarded integrals to a moving point charge; Wiechert independently in 1900

1912

— Schott's Electromagnetic Radiation systematizes retarded-potential radiation theory

1962

— Panofsky and Phillips's textbook presents the explicit retarded field formulas later named for Jefimenko

1966

— Jefimenko's Electricity and Magnetism states the causal field equations in their modern textbook form

2019

— the Event Horizon Telescope images M87*: planet-scale correlation of retarded arrival times

The physical picture

Every charge in the universe is a broadcaster, ceaselessly announcing its presence in all directions at the speed of light. What you measure here and now is not a survey of where the charges are — it is an archive of their announcements, each one exactly as stale as its distance demands. The lamp across the room reports itself from ten nanoseconds ago; the Sun from eight minutes ago; a radio galaxy from half a universe ago. The retarded integral is nothing but the bookkeeping of this archive: weight each announcement by the static $1/\mathscr{r}$ , and date each announcement by its own $t_r = t - \mathscr{r}/c$ . The geometric object doing the sampling is the observation event's past light cone. An extended source is never seen as a snapshot: the far side of a star is sampled earlier than the near side, by the light-crossing time of the star. Whether that skew matters is the single most useful engineering estimate in the subject — compare the light-crossing time of the source, $d/c$ , with the timescale $\tau$ on which the source changes. A 50 Hz transformer: $d/c \sim 0.3$ ns against $\tau \sim 20$ ms — retardation is a part-in- $10^8$ affair, and statics wins (this is the quasistatic license, priced honestly in the Maxwell chapter). An AM broadcast array: 100 m of antenna at 1 MHz gives $d/c = 334$ ns against a 1000 ns period — a third of a cycle across the source, and the entire technology of phased arrays lives inside that skew.

Figure

The past light cone does the sampling — each source element contributes from its own retarded time, near elements fresh, far elements stale

The formulas themselves, assembled by superposing the last article's delta shells: each source element $\rho(\mathbf{r}',t')\,d^3r'\,dt'$ launches a shell $G_{\text{ret}} = \delta(t - t' - \mathscr{r}/c)/4\pi\mathscr{r}$ ; integrating the shells against the source history, the time integral collapses on the delta, pinning $t' = t_r$ :

V(\mathbf{r},t) = \frac{1}{4\pi\varepsilon_0}\int \frac{\rho(\mathbf{r}',t_r)}{\mathscr{r}}\,d^3r', \qquad \mathbf{A}(\mathbf{r},t) = \frac{\mu_0}{4\pi}\int \frac{\mathbf{J}(\mathbf{r}',t_r)}{\mathscr{r}}\,d^3r', \qquad t_r = t - \frac{\mathscr{r}}{c}. \qquad (1)

where: $\mathscr{r} = |\mathbf{r}-\mathbf{r}'|$ is the source-to-field distance and the retarded time $t_r$ depends on the integration variable — it is a different time for every source element, which is the technical heart of everything difficult and everything interesting below. Read (1) against its static ancestors: the kernels, the constants, the superposition are untouched; the only novelty is the dating of the source. That such a minimal edit upgrades electrostatics and magnetostatics into the full causal theory is the chapter's central elegance — but "looks right" is not a proof, and the wave equation must now be checked, because plausible retarded guesses fail in general (the warning below kills the most tempting one).

Derivation AVerification — the retarded potential satisfies the wave equation

The goal is the flagship theorem: the retarded $V$ of (1) satisfies $\Box^2 V = -\rho/\varepsilon_0$ exactly (the identical argument runs for each component of $\mathbf{A}$ ). This must be done by brute differentiation under the integral, because the integrand depends on the field point $\mathbf{r}$ twice: through the explicit kernel $1/\mathscr{r}$ , and through the retarded time inside $\rho$ — and the second dependence is what a naive check forgets. The strategy: compute $\nabla V$ , then $\nabla^2 V$ , watching three species of term appear; the delta-function identity $\nabla^2(1/\mathscr{r}) = -4\pi\delta^3(\boldsymbol{\mathscr{r}})$ , owned by the vector-calculus chapter, produces the source; and the two dangerous middle terms cancel against each other — the cancellation is the theorem. Notation: dots denote $\partial\rho/\partial t$ evaluated at $t_r$ ; the chain rule gives $\nabla\rho(\mathbf{r}',t_r) = \dot\rho\,\nabla t_r = -\dfrac{\dot\rho}{c}\,\nabla\mathscr{r} = -\dfrac{\dot\rho}{c}\,\hat{\boldsymbol{\mathscr{r}}}$ .

Step 1. First gradient. Differentiating under the integral, product rule on $\rho(t_r)/\mathscr{r}$ :

\nabla V = \frac{1}{4\pi\varepsilon_0}\int\left[-\frac{\dot\rho}{c}\frac{\hat{\boldsymbol{\mathscr{r}}}}{\mathscr{r}} - \rho\,\frac{\hat{\boldsymbol{\mathscr{r}}}}{\mathscr{r}^2}\right]d^3r'.

Why two terms: the second is the static gradient (Coulomb's field, essentially); the first exists only because the sampling time moves when the field point moves — step one meter closer and every announcement you read is one light-meter fresher. This term has the longer range ( $1/\mathscr{r}$ against $1/\mathscr{r}^2$ ) — the first scent of radiation.

Step 2. The divergence, term by term. Applying $\nabla\cdot$ to the first integrand, with $\nabla\dot\rho = -(\ddot\rho/c)\hat{\boldsymbol{\mathscr{r}}}$ by the same chain rule and $\nabla\cdot(\hat{\boldsymbol{\mathscr{r}}}/\mathscr{r}) = 1/\mathscr{r}^2$ :

-\frac{1}{c}\nabla\cdot\left[\dot\rho\,\frac{\hat{\boldsymbol{\mathscr{r}}}}{\mathscr{r}}\right] = \frac{\ddot\rho}{c^2}\frac{1}{\mathscr{r}} - \frac{\dot\rho}{c}\frac{1}{\mathscr{r}^2}.

For the second integrand, with $\nabla\cdot(\hat{\boldsymbol{\mathscr{r}}}/\mathscr{r}^2) = 4\pi\delta^3(\boldsymbol{\mathscr{r}})$ — the delta identity of the vector-calculus chapter, recalled not re-proved:

-\nabla\cdot\left[\rho\,\frac{\hat{\boldsymbol{\mathscr{r}}}}{\mathscr{r}^2}\right] = \frac{\dot\rho}{c}\frac{1}{\mathscr{r}^2} - 4\pi\rho\,\delta^3(\boldsymbol{\mathscr{r}}).

Why the delta appears with $\rho$ un-retarded: the delta fires only at $\mathscr{r}=0$ , where $t_r = t$ — the one place in the integral where the news is perfectly fresh. The source term of the wave equation will thus carry the present, local charge density, exactly as it must.

Step 3. Assemble, and watch the cancellation. Summing the four pieces inside the integral:

\nabla^2 V = \frac{1}{4\pi\varepsilon_0}\int\left[\frac{\ddot\rho}{c^2\,\mathscr{r}} \;\underbrace{-\;\frac{\dot\rho}{c\,\mathscr{r}^2} + \frac{\dot\rho}{c\,\mathscr{r}^2}}_{=\,0}\; - 4\pi\rho\,\delta^3(\boldsymbol{\mathscr{r}})\right]d^3r' = \frac{1}{c^2}\frac{\partial^2 V}{\partial t^2} - \frac{\rho(\mathbf{r},t)}{\varepsilon_0}.

The first term is recognized as $\tfrac{1}{c^2}\partial_t^2 V$ because $\partial t_r/\partial t = 1$ at fixed positions, so time-differentiating (1) twice simply plants $\ddot\rho$ in the integrand; the delta collapses the last term onto the field point. Why the cancellation is the whole ballgame: the two $\dot\rho/\mathscr{r}^2$ terms came from completely different physics — one from the retardation of the gradient, one from the static kernel meeting the retarded density — and they annihilate exactly. Change the recipe in any way (advance instead of retard by a different speed, retard by $2\mathscr{r}/c$ , use $1/\mathscr{r}^2$ weighting) and the cancellation fails: the wave equation is satisfied by essentially this formula and its advanced mirror alone. The "obvious guess" was one honest derivative away from being wrong.

Step 4. Rearranged, the theorem: $\Box^2 V = \nabla^2 V - \tfrac{1}{c^2}\partial_t^2 V = -\rho(\mathbf{r},t)/\varepsilon_0$ . ∎ The same computation with $\mu_0 J_i$ in place of $\rho/\varepsilon_0$ verifies each component of $\mathbf{A}$ . Together with the domain-of-dependence uniqueness of the previous article (quiescent past ⇒ this is the solution), the retarded potentials are established, not conjectured. Verified: reproduces Griffiths §10.2.1 (the Ex. 10.1 computation) and Jackson §6.5; static limit ( $\dot\rho = 0$ ) collapses every retardation term and recovers the Coulomb integral exactly.

The consistency check: one gauge, two integrals

Equations (1) were solved separately — four independent scalar problems. But the Lorenz gauge was the standing assumption, and the gauge is a relation between $V$ and $\mathbf{A}$ . Nothing in the separate solving enforced it; it must be checked, and the check turns out to hinge on physics, not algebra.

Go deeper · Derivation B · 2 minThe retarded pair satisfies the Lorenz condition — if charge is conserved

Derivation BThe retarded pair satisfies the Lorenz condition — if charge is conserved

The goal is $\nabla\cdot\mathbf{A} + \mu_0\varepsilon_0\,\partial_t V = 0$ for the pair (1) — the certificate that the two retarded integrals really are potentials of the same gauge, so that $\mathbf{E} = -\nabla V - \partial_t\mathbf{A}$ may legally combine them. The strategy is to push the divergence through the retarded integral for $\mathbf{A}$ , converting field-point derivatives into source-point derivatives (they act oppositely on $\mathscr{r}$ , but identically through $t_r$ — the bookkeeping subtlety of the whole computation), discard a surface term, and watch the continuity equation appear inside the integral. Notation: $\nabla$ acts on $\mathbf{r}$ , $\nabla'$ on $\mathbf{r}'$ ; note $\nabla\mathscr{r} = -\nabla'\mathscr{r} = \hat{\boldsymbol{\mathscr{r}}}$ .

Step 1. Divergence under the integral. For one integrand of $\mathbf{A}$ , the product rule splits the divergence into kernel and retardation pieces:

\nabla\cdot\left[\frac{\mathbf{J}(\mathbf{r}',t_r)}{\mathscr{r}}\right] = \frac{1}{\mathscr{r}}\left(\nabla\cdot\mathbf{J}\right)\Big|_{\text{via }t_r} + \mathbf{J}\cdot\nabla\frac{1}{\mathscr{r}}, \qquad \left(\nabla\cdot\mathbf{J}\right)\Big|_{\text{via }t_r} = -\frac{1}{c}\,\dot{\mathbf{J}}\cdot\hat{\boldsymbol{\mathscr{r}}}.

Step 2. Trade field-point for source-point derivatives. The genuinely clever move: compute $\nabla'\cdot\mathbf{J}(\mathbf{r}',t_r)$ , which receives two contributions — the explicit dependence of $\mathbf{J}$ on its first argument (the physical divergence, written $(\nabla'\cdot\mathbf{J})_{t_r}$ , continuity's subject) and the $t_r$ dependence, which enters with the opposite sign to Step 1's since $\nabla' \mathscr{r} = -\hat{\boldsymbol{\mathscr{r}}}$ :

\nabla'\cdot\left[\mathbf{J}(\mathbf{r}',t_r)\right] = (\nabla'\cdot\mathbf{J})_{t_r} + \frac{1}{c}\,\dot{\mathbf{J}}\cdot\hat{\boldsymbol{\mathscr{r}}}.

Combining the two displays to eliminate the $\dot{\mathbf{J}}$ term and reassembling with the kernel ( $\nabla(1/\mathscr{r}) = -\nabla'(1/\mathscr{r})$ ):

\nabla\cdot\left[\frac{\mathbf{J}(t_r)}{\mathscr{r}}\right] = \frac{1}{\mathscr{r}}(\nabla'\cdot\mathbf{J})_{t_r} - \nabla'\cdot\left[\frac{\mathbf{J}(t_r)}{\mathscr{r}}\right].

Why this identity is the workhorse: it isolates the physical divergence of the current (first term) from a pure source-coordinate divergence (second term), which Gauss's theorem will convert to a surface integral at infinity — zero for localized sources. Every retarded-integral manipulation in the coming articles reuses exactly this trade.

Step 3. Integrate, discard the surface, add the scalar piece. Integrating Step 2 over all source points kills the total-divergence term; meanwhile $\mu_0\varepsilon_0\,\partial_t V$ time-differentiates (1) directly ( $\partial t_r/\partial t = 1$ ):

\nabla\cdot\mathbf{A} + \mu_0\varepsilon_0\frac{\partial V}{\partial t} = \frac{\mu_0}{4\pi}\int \frac{1}{\mathscr{r}}\Big[(\nabla'\cdot\mathbf{J}) + \dot\rho\Big]_{t_r}\,d^3r'.

Step 4. Invoke continuity. The bracket is the continuity combination evaluated at the retarded time; charge conservation makes it vanish pointwise, hence the integral vanishes and the Lorenz condition holds. The converse reading deserves equal billing: for a hypothetical charge-non-conserving source, the retarded integrals (1) would still solve their separate wave equations — but they would sit in different gauges, and assembling fields from them would be meaningless. Continuity is the glue binding the scalar and vector integrals into one potential description. Verified: Griffiths §10.2 (Problem 10.8) and Jackson §6.5; consistent with the Lorenz-gauge persistence theorem, which made the same physics appear at the level of the equations rather than the solutions.

Limiting cases

case	result	meaning
static sources	$t_r$ -dependence drops; (1) → Coulomb and magnetostatic integrals	electro- and magnetostatics are the fresh-news limit of the archive
quasistatic ( $d/c\tau \ll 1$ )	$\rho(t_r) \approx \rho(t) + (t_r - t)\dot\rho + \cdots$	Taylor-expanding the staleness: leading corrections of order $(d/c\tau)$ , organized into the multipole ladder ahead
far zone ( $r \gg d, \lambda$ )	$V, \mathbf{A} \sim \dfrac{(\text{source at } t - r/c)}{r}$	the $1/r$ tail carrying time variation: the raw material of radiation
point charge on a trajectory	integral collapses with Jacobian $(1 - \hat{\boldsymbol{\mathscr{r}}}\cdot\mathbf{v}/c)^{-1}$	the Liénard–Wiechert potentials, two articles ahead — the factor is not the naive $q/4\pi\varepsilon_0\mathscr{r}$
$t_r \to t_a = t + \mathscr{r}/c$	advanced potentials — equally valid solutions	the time-mirror twin; selection by causal data, prosecuted next article

Common trapYou may not retard the fields — the shortcut that works for potentials fails one derivative later

The seductive move: "fields are computed from potentials, potentials are statics-with- $t_r$ , so surely $\mathbf{E}$ is Coulomb-with- $t_r$ and $\mathbf{B}$ is Biot–Savart-with- $t_r$ ." False, and Derivation A shows exactly why: $\nabla$ and $\partial_t$ act on the retarded-time dependence inside the integrals, generating additional $\dot\rho$ , $\dot{\mathbf{J}}$ terms with $1/\mathscr{r}$ range that the naive replacement omits. Those "extra" terms are not corrections — they are the radiation: drop them and no antenna works, no light propagates, no energy leaves any source, and the formula even fails dimensional consistency checks against Faraday's law. The correct field expressions, with every term accounted, are Jefimenko's equations (two articles ahead). Curiously, the magnetic naive guess is far better than it deserves to be — the leading retardation corrections cancel for slowly varying currents, the quasistatic vindication also owed to Jefimenko's article. Until then: retard potentials freely; never retard a field formula.

Worked examples

1 · The retardation budget — when does $t_r$ matter? (confidence builder)

Setup. The universal first estimate: compare the light-crossing time of the source, $d/c$ , to the source's timescale $\tau$ (for oscillations, the period). Retardation across the source skews the sampling by $d/c$ ; if $d/c \ll \tau$ , the archive is effectively a snapshot.

Step 1. Power engineering. A transformer, $d = 0.1$ m, at 50 Hz ( $\tau = 20$ ms): skew fraction $= \dfrac{d/c}{\tau} = \dfrac{3.3\times10^{-10}\ \text{s}}{2\times10^{-2}\ \text{s}} = 1.7\times10^{-8}$ . Retardation is seventeen parts per billion of a cycle: circuit theory's license to ignore $t_r$ entirely, consistent with the quasistatic ordering of the Maxwell chapter.

Step 2. Broadcast engineering. An AM antenna array, $d = 100$ m, at 1 MHz ( $\tau = 1\ \mu$ s): skew $= \dfrac{334\ \text{ns}}{1000\ \text{ns}} = 0.33$ — a third of a full cycle across the array. Retardation is not a correction here; it is the design variable: phased arrays steer their beams precisely by arranging which elements' news arrives in phase in which direction.

Step 3. The crossover rule. Skew fraction $= d/(c\tau) = d/\lambda$ for oscillating sources: retardation across a source matters exactly when the source is not small compared to the wavelength. That single ratio, $d/\lambda$ , will organize the entire radiation zoo — the dipole approximation of the chapter's finale is the regime $d/\lambda \ll 1$ where only the first correction survives.

The lesson: before any retarded integral, compute $d/\lambda$ . It tells you whether you are doing circuit theory, antenna theory, or something in between.

Figure

The switched-on wire — only the segment whose news has arrived contributes, and the fields ride the truncation

2 · Switching on an infinite wire — watching magnetostatics establish itself (exam level)

Setup. An infinite straight wire on the $z$ -axis carries no current until $t = 0$ , then constant $I_0 = 10$ A: the standard test of the retarded machinery (Griffiths Ex. 10.2). Find $\mathbf{A}$ , then the fields, at distance $s$ ; evaluate at $s = 0.5$ m.

Step 1. Only part of the wire has reported in. At time $t$ , the observation point has received news only from wire elements within $\mathscr{r} < ct$ , i.e. $|z| < \sqrt{c^2t^2 - s^2}$ (nothing at all until $t = s/c = 1.67$ ns). The retarded integral truncates itself:

A_z(s,t) = \frac{\mu_0 I_0}{4\pi}\,2\!\int_0^{\sqrt{c^2t^2-s^2}} \frac{dz}{\sqrt{s^2+z^2}} = \frac{\mu_0 I_0}{2\pi}\,\ln\!\left(\frac{ct + \sqrt{c^2t^2 - s^2}}{s}\right), \qquad s < ct.

Step 2. Differentiate for the fields — both derivatives now see the time dependence through the moving limit:

\mathbf{E} = -\frac{\partial A_z}{\partial t}\hat{\mathbf{z}} = -\frac{\mu_0 I_0 c}{2\pi\sqrt{c^2t^2 - s^2}}\,\hat{\mathbf{z}}, \qquad \mathbf{B} = -\frac{\partial A_z}{\partial s}\hat{\boldsymbol\varphi} = \frac{\mu_0 I_0}{2\pi s}\,\frac{ct}{\sqrt{c^2t^2 - s^2}}\,\hat{\boldsymbol\varphi}.

Step 3. Numbers at $s = 0.5$ m. News arrives at $t = 1.67$ ns. At $t = 3.34$ ns ( $ct = 1$ m): $B = \dfrac{\mu_0 I_0}{2\pi s}\cdot\dfrac{1}{\sqrt{1 - 0.25}} = 4.0\ \mu\text{T}\times1.155 = \mathbf{4.62\ \mu T}$ , and a startling $E = \mathbf{693\ V/m}$ pointing against the current's direction — the wire, establishing its magnetic field, drives a fierce inductive kick (this is the physics of inductive spiking in fast switches). By $ct = 100$ m ( $t = 334$ ns): $B = 4.0\ \mu\text{T}\times1.0000125$ — magnetostatics to 13 ppm — while $E$ has collapsed to 6 V/m and continues to die as $1/t$ .

Step 4. Audit against the causal bookkeeping. Before $s/c$ : nothing, exactly — the domain-of-dependence theorem, visible. As $t \to \infty$ : $B \to \mu_0 I_0/2\pi s$ , the Ampère-law answer, and $E \to 0$ : magnetostatics is not an assumption but the late-time attractor of the retarded dynamics. And note $\mathbf{A}$ itself diverges logarithmically as $t\to\infty$ — a reminder that potentials carry gauge slack while their derivatives, the fields, settle cleanly.

The lesson: the retarded integral handles even violently non-static sources (a step function!) and hands back both the transient — real, measurable, 693 V/m of it — and the familiar static endpoint.

Going further

References. Griffiths, Introduction to Electrodynamics, 4th ed., §10.2 · Jackson, Classical Electrodynamics, 3rd ed., §6.4–6.5 · Zangwill, Modern Electrodynamics, §20.3 · Purcell & Morin, Electricity and Magnetism, 3rd ed., Ch. 5–9 (causality discussion)

Provenance. Equations and structure follow Griffiths §10.2, Jackson §6.4–6.5, and Zangwill §20.3. A verifier should: (1) confirm the retarded Green's function $\delta(t-r/c)/(4\pi r)$ solves $\Box^2 G=-\delta^3\delta$ ; (2) reproduce the convolution collapsing to $t_r=t-\mathcal{R}/c$ ; (3) verify the Lorenz-gauge consistency derivation using $\nabla f(t_r)=-\nabla'f(t_r)|_t$ plus continuity; (4) check the static-limit reductions to Coulomb and Biot–Savart. Constant: $\epsilon_0\mu_0=1/c^2$ (exact). Addendum for this edition: per the chapter's division of labor, verifier item (1) — the Green-function construction, the base's first derivation — is owned by Inhomogeneous Wave Equations for Potentials and recalled here; item (2) appears as the short convolution collapse in the picture section rather than a full derivation. This edition owns the flagship verification-by-differentiation that the retarded potentials satisfy the wave equation (Derivation A, the Griffiths §10.2.1 computation with the middle-term cancellation made explicit) and the Lorenz-condition consistency proof (Derivation B, with the $\nabla\leftrightarrow\nabla'$ trade documented as the reusable workhorse); the advanced twin is stated and handed to Advanced Potentials and Causality; the fields-may-not-be-retarded audit is carried in the warning with receipts deferred to Jefimenko's Equations. Worked examples (retardation budgets $1.7\times10^{-8}$ and $0.33$ ; wire turn-on $B = 4.62\ \mu$ T and $E = 693$ V/m at $ct = 1$ m, 13 ppm settling at $ct = 100$ m) recomputed for this edition. ⚠ Slug note: bare `retarded-potentials` collides with an old physics-wiki row — REPLACE at global cutover per recon §c-A; the aq- prefix protects until then.

Builds on: Inhomogeneous Wave Equations for Potentials · Lorenz Gauge · Scalar and Vector Potentials · Gauge Transformations Leads to: Advanced Potentials and Causality · Jefimenko's Equations · Liénard–Wiechert Potentials · Fields of a Moving Point Charge · Electric Dipole Radiation

How is this article different from the Green-function construction in the previous one?

Division of labor along the line between building and certifying. The previous article owns the operator: its causal structure, and the construction that the point-impulse response must be a delta shell — retarded or advanced, with only the total strength fixed. This article owns the physics-facing results: the explicit source integrals (1), the verification that the retarded choice genuinely satisfies the full inhomogeneous wave equation (Derivation A — construction from a Green function motivates the formula but the honest check is differentiation, which is also where the fields-may-not-be-retarded lesson lives), and the gauge certificate binding $V$ to $\mathbf{A}$ (Derivation B). The next article completes the trilogy by prosecuting the retarded-versus-advanced selection. Construction, certification, selection: three articles, zero overlap.

Why can I not pull the retarded time out of the integral as a common factor?

Because there is no such thing as "the" retarded time — $t_r = t - \mathscr{r}/c$ contains the integration variable through $\mathscr{r} = |\mathbf{r}-\mathbf{r}'|$ , so each source element is sampled at its own moment. Factoring it out amounts to replacing the past light cone by a flat time slice, which is legitimate only in the regimes where the difference is negligible: sources small compared to a wavelength ( $d/\lambda \ll 1$ , worked example 1's budget), or the far-zone approximation where $\mathscr{r} \approx r - \hat{\mathbf{r}}\cdot\mathbf{r}'$ and the leading $t - r/c$ can be factored while the correction term $\hat{\mathbf{r}}\cdot\mathbf{r}'/c$ is kept to first order — precisely the bookkeeping that generates the multipole radiation expansion later in the chapter. The skew you are tempted to discard is, in that expansion, the entire signal.

Does the retarded potential of a moving point charge equal q/4πε₀ times one over the retarded distance?

No — and the failure is one of the great traps of the subject, flagged here because the temptation arises the moment you meet (1). For a point charge the density is a moving delta function, and integrating a delta whose argument contains the integration variable (through $t_r$ ) produces a Jacobian: the volume integral collapses not to $q/\mathscr{r}$ but to $q/[\mathscr{r}(1 - \hat{\boldsymbol{\mathscr{r}}}\cdot\mathbf{v}/c)]$ , the famous Liénard–Wiechert denominator. The physical reading (the "information-collecting sphere" argument, built carefully two articles ahead) is that a charge moving toward you crowds more of its announcements into your collection window — a geometric compression, present even at constant velocity. The naive $q/4\pi\varepsilon_0\mathscr{r}_{\text{ret}}$ is wrong by exactly that factor, which can be arbitrarily large for ultrarelativistic motion.

What happens to the retarded integral for field points inside the source?

Nothing special — a reassurance with content. The $1/\mathscr{r}$ singularity is integrable in three dimensions (spherical shells contribute $\mathscr{r}^2\,d\mathscr{r}/\mathscr{r} \propto \mathscr{r}\,d\mathscr{r}$ , vanishing at zero), exactly as in electrostatics, and the delta identity in Derivation A already handled the field point rigorously — that is where the source term $-\rho(\mathbf{r},t)/\varepsilon_0$ came from. Physically, an interior point simply reads its archive at all depths: nearby source elements report almost-fresh, the far side of the source reports staler, and the self-consistency of the whole (fields inside antennas, current distributions responding to their own retarded fields) is the hard integral-equation problem that antenna solvers iterate numerically. The formula (1) is exact throughout; only its self-consistent coupling to the source's dynamics is work.

Is it obvious that the two separately-solved integrals can be combined into E = −∇V − ∂A/∂t?

It is not, and Derivation B exists precisely because it is not. The field formula presumes $V$ and $\mathbf{A}$ describe the same physics in the same gauge — but each retarded integral was manufactured independently, from its own scalar wave equation. The certificate is the Lorenz condition, and the check reveals it holds only by grace of charge conservation: the continuity combination appears under the integral and must vanish. Feed the machinery a fictional source that creates charge from nothing and the two integrals desynchronize — each still solves its equation, but $-\nabla V - \partial_t\mathbf{A}$ computes gauge garbage, not a field. This is the solution-level echo of a pattern the chapter has met twice (the potential system's consistency identity; the gauge's persistence theorem): electrodynamics interlocks with charge conservation at every level of its structure.

Causality was fed in by choosing the retarded solution. Is that circular reasoning about the speed of light?

No, but the logic deserves care. Two separate facts: (i) the wave equation permits no influence faster than $c$ — that is the domain-of-dependence theorem, proved from the operator alone, no solution choice involved; (ii) among the influences the equation permits, nature uses the past-directed ones — that is the retarded choice, and it is genuinely an input, encoded in data (quiescent past, no incoming radiation) rather than derived from the dynamics. So the speed limit is a theorem; the arrow is a boundary condition. Nothing circular — but the honest accounting shows exactly what would have to be different for advanced effects: not different equations, different data. Why the universe supplies retarded-type data is the arrow-of-time question, and it is the next article's entire subject, absorber theory and all.

Do the retarded potentials describe near fields, or only radiation?

Everything — near zone, far zone, statics, transients — with no approximation whatsoever. Equation (1) is exact for any localized source: worked example 2's switching wire is violently non-radiative near-field physics (plus a transient), handled without ceremony. The zones enter only when you expand the exact integrals: close in ( $r \ll \lambda$ ) the staleness is negligible and quasistatic behavior dominates; far out ( $r \gg \lambda$ ) the $1/r$ tail with its time-varying numerator takes over, and that surviving piece is what the radiation articles isolate. It is worth internalizing the priority: the retarded potentials are the master formula; "near field" and "radiation zone" are asymptotic readings of it, not separate theories. When in doubt — odd geometries, intermediate zones, ultrawideband pulses — return to (1) and integrate.

Retarded Potentials

Where this lives today

The physical picture

The consistency check: one gauge, two integrals

Limiting cases

Worked examples

1 · The retardation budget — when does $t_r$ matter? (confidence builder)

2 · Switching on an infinite wire — watching magnetostatics establish itself (exam level)

Going further

Before this

Where this leads

Retarded Potentials

Where this lives today

The physical picture

The consistency check: one gauge, two integrals

Limiting cases

Worked examples

1 · The retardation budget — when does trt_rtr​ matter? (confidence builder)

2 · Switching on an infinite wire — watching magnetostatics establish itself (exam level)

Going further

Before this

Where this leads

1 · The retardation budget — when does $t_r$ matter? (confidence builder)