Retarded and Advanced Green's Functions for the Wave Equation

A Green's function answers a single sharp question: what does spacetime do in response to one unit impulse — a point flash delivered at one place at one instant? For the wave operator this has a subtle twist. Because the wave equation is hyperbolic and second order in time, the differential equation alone does not pin the answer down; you must also say what happens in the infinite past or future. Demand that no response precede its cause and you select the retarded Green's function $G^{\mathrm{ret}}$ , the workhorse of every radiation calculation. Run time backward for the advanced Green's function $G^{\mathrm{adv}}$ , mathematically just as legitimate and indispensable in scattering theory, field-theory commutators, and the Wheeler–Feynman absorber picture. The physical heart of the matter is finite propagation speed. The heat kernel spreads instantaneously, but the three-dimensional wave Green's function is concentrated exactly on the light cone $|\mathbf{r}-\mathbf{r}'|=c(t-t')$ : a flash is felt at a distant point only at the one instant the spherical wavefront sweeps past, and then silence returns. That is Huygens' principle in its strict form, and it is why we get crisp images, undistorted speech, and clean signalling in a three-dimensional world. This article takes the correct impulse response, derives it two independent ways, exposes the pole prescription that encodes causality, and follows the same kernel into the retarded potential, the Pauli–Jordan commutator function, and the dimension-dependence that makes two-dimensional waves reverberate while three-dimensional ones ring true.

Key takeaways

The retarded and advanced Green's functions $G^{\mathrm{ret/adv}}$ are the fundamental solutions of the wave operator $\Box=c^{-2}\partial_t^2-\nabla^2$ , obeying $\Box G=\delta^3(\mathbf{r}-\mathbf{r}')\,\delta(t-t')$ .
In three dimensions $G^{\mathrm{ret}}=\delta(\tau-R/c)/4\pi R$ is supported exactly on the forward light cone — a razor-thin spherical shell, the strong form of Huygens' principle.
Causality selects the retarded branch: the field at time $t$ depends only on sources at the earlier retarded time $t'=t-R/c$ , encoding finite-speed propagation.
The two branches are picked out by which way the frequency contour skirts the poles $\omega=\pm c|\mathbf{k}|$ — a prescription $\omega\to\omega\pm i\varepsilon$ that is the time-domain face of a boundary condition in the far past or future.
Their difference $G^{\mathrm{ret}}-G^{\mathrm{adv}}$ is the source-free Pauli–Jordan function governing field commutators; their average is the time-symmetric kernel of Wheeler–Feynman electrodynamics.
Huygens' principle holds cleanly only in odd space dimensions $\ge 3$ ; in 2D and every even dimension a tail lingers behind the wavefront, so a point flash rings on forever.
Key idea: the wave operator's Green's function is fixed only after a choice of temporal boundary condition; causality selects the retarded solution, which in 3D lives entirely on the light cone $G^{\mathrm{ret}}=\delta(t-t'-R/c)/4\pi R$ , so the field is read straight off the sources at the retarded time.

The core insight

The wave operator's Green's function is fixed only after a choice of temporal boundary condition; causality selects the retarded solution, which in 3D lives entirely on the light cone $G^{\mathrm{ret}}=\delta(t-t'-R/c)/4\pi R$ with $R=|\mathbf{r}-\mathbf{r}'|$ , so the field is read off from sources at the retarded time.

Where this lives today

The retarded Green's function is the computational core of every field that radiates. In electrodynamics the Liénard–Wiechert potentials — retarded potentials for a point charge in arbitrary motion — are what accelerator physicists integrate to predict synchrotron spectra at light sources like the ESRF and APS, and what antenna engineers build phased arrays from. In gravitational-wave astronomy the same structure appears one order up: the LIGO–Virgo–KAGRA strain is, to leading order, the retarded field of a distant binary's changing quadrupole, arriving a light-cone travel time after the merger. Radio astronomers reading a pulsar are literally sampling the retarded time — the Crab pulsar's beam we time today left the source about 6,500 years ago. Time-domain boundary-element and FDTD solvers propagate pulses by discretizing exactly this kernel, its clean 3D support keeping sharp fronts from smearing. And the advanced solution, long dismissed as unphysical, resurfaces in time-symmetric formulations probed by tests of the arrow of time — a reminder that the mathematics never forced causality on us; we did.

The physical picture

Strike a single point in empty three-dimensional space for one instant. The disturbance leaves as an infinitely thin spherical shell expanding at speed $c$ . A microphone a distance $R$ away registers silence until $t=R/c$ , then one sharp click as the shell sweeps through, then silence again. That click is the retarded Green's function — a Dirac delta riding the light cone, weighted by $1/4\pi R$ . The advanced Green's function is the exact time-reverse: an incoming shell that converges on the point and vanishes at the firing instant. Neither fills the interior; the response is on the surface and nowhere else. The first figure draws this in spacetime, where the retarded/advanced split becomes the forward/backward light cones of one event.

Figure

The light cone of a point impulse in spacetime: the retarded Green's function lives on the forward cone (blue), the advanced one on the backward cone (amber), the interior stays silent. A detector at separation R clicks once, when the cone sweeps past at τ = R/c — for c = 3.0×10⁸ m/s and R = 6.0×10⁸ m that click lands at τ = 2.0 s

Contrast this with diffusion, whose heat kernel is a Gaussian nonzero everywhere the instant the source turns on. The wave Green's function has bounded support: nothing outside the cone, and in 3D nothing inside it either. This "support on the cone, not in it" is the geometric fingerprint of Huygens' principle, a feature of odd space dimensions. Change the dimension and the response changes character: in 1D the Green's function is a Heaviside step filling the interior, so once a wavefront passes the disturbance never leaves; in 2D a point flash produces not a click but a thud with a fading tail decaying like $1/\tau$ . The second figure sets the three cases against a real time axis.

Figure

Detector response to a point impulse at separation R versus time, by spatial dimension: 3D gives a single sharp click on the cone (blue spike), 2D an onset with an endlessly decaying tail (green), 1D a step that never returns (amber). For sound at c = 343 m/s and R = 686 m the front arrives at τ = R/c = 2.0 s

The pole-displacement language ties these threads together. In frequency space the wave operator becomes $\omega^2-c^2k^2$ , whose zeros at $\omega=\pm ck$ block the inverse-transform integral, so you must say how to skirt them. Pushing both poles into the lower half $\omega$ -plane forces the time integral to enclose nothing for $t<t'$ — that vanishing is causality, the retarded choice; the mirror gives the advanced function.

The core equations

Everything begins with the impulse response: the wave operator acting on $G$ returns a unit "kick" at $(\mathbf{r}',t')$ , and $G$ is the disturbance it produces everywhere else. Because the operator is second order in time, this has a family of solutions differing by homogeneous waves; a temporal boundary condition singles one out.

\Box\, G(\mathbf{r},t;\mathbf{r}',t') = \delta^{3}(\mathbf{r}-\mathbf{r}')\,\delta(t-t'),\qquad \Box \equiv \frac{1}{c^2}\frac{\partial^2}{\partial t^2} - \nabla^2 \qquad (1)

where: $\Box$ — the d'Alembertian (wave operator); $c$ — propagation speed; $G$ — the Green's function or impulse response; $\delta^3,\ \delta$ — the spatial and temporal Dirac deltas. Since empty space is homogeneous, $G$ depends only on the separations $\mathbf{R}=\mathbf{r}-\mathbf{r}'$ and $\tau=t-t'$ , with $R=|\mathbf{R}|$ ; we write $G(\mathbf{R},\tau)$ from here on. Imposing the two temporal boundary conditions — vanishing before, or after, the impulse — produces the two solutions that name this article. In 3D each is a delta pinned to a light cone, scaled by the inverse-distance spreading of a spherical wave.

G^{\mathrm{ret}}(\mathbf{R},\tau) = \frac{\delta\!\big(\tau - R/c\big)}{4\pi R},\qquad G^{\mathrm{adv}}(\mathbf{R},\tau) = \frac{\delta\!\big(\tau + R/c\big)}{4\pi R} \qquad (2)

where: $\delta(\tau-R/c)$ — support on the forward (retarded) light cone; $\delta(\tau+R/c)$ — support on the backward (advanced) cone; $1/4\pi R$ — the geometric factor from the static Poisson kernel. The retarded response at separation $R$ is a single spike arriving at $\tau=R/c$ , and nothing else. To see why only these two survive, transform to frequency–wavevector space, where plane waves $e^{i(\mathbf{k}\cdot\mathbf{R}-\omega\tau)}$ diagonalize the operator — turning $\Box$ into a number and the delta sources into unity, and $\tilde G$ into a rational function.

\tilde{G}(\mathbf{k},\omega) = \frac{-c^2}{\omega^2 - c^2|\mathbf{k}|^2} \qquad (3)

where: $\mathbf{k}$ — wavevector; $\omega$ — angular frequency; $\tilde G$ — the spacetime Fourier transform of $G$ . The denominator vanishes at $\omega=\pm c|\mathbf{k}|$ : the poles that leave the inverse transform ambiguous until a contour rule is chosen. Displacing the poles by $\pm i\varepsilon$ before inverting selects the boundary condition — causality reduced to a sign.

G^{\mathrm{ret/adv}}(\mathbf{R},\tau) = -c^2\!\int\!\frac{\mathrm{d}^3k\,\mathrm{d}\omega}{(2\pi)^4}\,\frac{e^{i(\mathbf{k}\cdot\mathbf{R}-\omega\tau)}}{(\omega\pm i\varepsilon)^2 - c^2|\mathbf{k}|^2} \qquad (4)

where: $\varepsilon\to 0^+$ — infinitesimal pole displacement; $+i\varepsilon$ — retarded (poles in the lower half-plane, so $G^{\mathrm{ret}}=0$ for $\tau<0$ ); $-i\varepsilon$ — advanced (poles in the upper half-plane, $G^{\mathrm{adv}}=0$ for $\tau>0$ ). The sign of $\varepsilon$ is a boundary condition, not an approximation. Once the impulse response is known, any source follows by superposition. Collapsing the temporal delta in 3D turns the spacetime convolution into a single spatial integral at the retarded time.

\phi(\mathbf{r},t)=\phi_{\mathrm{hom}}+\frac{1}{4\pi}\int \frac{s\big(\mathbf{r}',\,t-|\mathbf{r}-\mathbf{r}'|/c\big)}{|\mathbf{r}-\mathbf{r}'|}\,\mathrm{d}^3r' \qquad (5)

where: $s$ — the source density on the right of $\Box\phi=s$ ; $\phi_{\mathrm{hom}}$ — a homogeneous solution fixed by initial data; $t-|\mathbf{r}-\mathbf{r}'|/c$ — the retarded time. This is the retarded potential: each source element contributes what it did $R/c$ ago, weighted by $1/4\pi R$ . Finally, the two branches combine into two objects central to field theory: their difference is source-free, their average solves the same inhomogeneous equation but treats past and future symmetrically.

\Delta \equiv G^{\mathrm{ret}}-G^{\mathrm{adv}} = \frac{\delta(\tau-R/c)-\delta(\tau+R/c)}{4\pi R},\qquad \bar G \equiv \tfrac12\big(G^{\mathrm{ret}}+G^{\mathrm{adv}}\big) \qquad (6)

where: $\Delta$ — the Pauli–Jordan (commutator) function, obeying $\Box\Delta=0$ ; $\bar G$ — the time-symmetric (principal-value) Green's function, obeying $\Box\bar G=\delta^3\delta$ . The first governs microcausality in QFT; the second underlies Wheeler–Feynman electrodynamics.

Derivation AConstructing the retarded Green's function by contour integration

The goal is to carry the Fourier representation of equation (4) to the closed form of equation (2), doing the frequency integral first with the retarded $+i\varepsilon$ prescription and then the wavevector integral, and to watch causality appear on its own — the vanishing for $\tau<0$ falling out of where the poles sit rather than being imposed by hand. This is the derivation showing the light-cone delta and the $1/4\pi R$ factor are forced, not assumed.

Step 1. Separate the integrals and place the poles. Writing the kernel as a frequency integral inside a wavevector integral, $G^{\mathrm{ret}}=-c^2\!\int\!\frac{\mathrm{d}^3k}{(2\pi)^3}e^{i\mathbf{k}\cdot\mathbf{R}}\!\int\!\frac{\mathrm{d}\omega}{2\pi}\frac{e^{-i\omega\tau}}{(\omega+i\varepsilon)^2-c^2k^2}$ , the poles sit at $\omega=\pm ck-i\varepsilon$ , both just below the real axis. Why: the $+i\varepsilon$ shift is the entire retarded boundary condition, pushing both poles into the lower half-plane.

Step 2. Close the contour by the sign of $\tau$ . Since $e^{-i\omega\tau}$ decays in the lower half-plane for $\tau>0$ and the upper for $\tau<0$ , Jordan's lemma closes the contour downward for $\tau>0$ and upward for $\tau<0$ ; the upward loop encloses no poles, so $G^{\mathrm{ret}}=0$ before the impulse. Why: causality emerges automatically — we never demanded it, only chose which side of the axis the poles lie on.

Step 3. Sum the residues for $\tau>0$ . Closing below (clockwise, a factor $-2\pi i$ ) and collecting the residues at $\omega=\pm ck$ gives $\int\frac{\mathrm{d}\omega}{2\pi}\frac{e^{-i\omega\tau}}{\omega^2-c^2k^2}=-\frac{\sin(ck\tau)}{ck}$ . Why: the symmetric pole pair combines into a sine — the free oscillation of a single Fourier mode of the wave equation.

Step 4. Do the angular wavevector integral. Substituting the sine and integrating $e^{i\mathbf{k}\cdot\mathbf{R}}$ over solid angle (using $\int\mathrm{d}\Omega\,e^{i\mathbf{k}\cdot\mathbf{R}}=4\pi\sin(kR)/(kR)$ ) gives $G^{\mathrm{ret}}=\frac{c}{2\pi^2 R}\int_0^\infty \sin(kR)\,\sin(ck\tau)\,\mathrm{d}k$ . Why: the angular integral is where the inverse-distance factor $1/R$ is born, the same spreading as in the static Coulomb problem.

Step 5. Evaluate the radial integral. Writing the product of sines as a difference of cosines and using $\int_0^\infty\cos(ka)\,\mathrm{d}k=\pi\delta(a)$ gives $\frac{\pi}{2}\big[\delta(R-c\tau)-\delta(R+c\tau)\big]$ ; only the first fires for $R,\tau>0$ , and $\delta(R-c\tau)=c^{-1}\delta(\tau-R/c)$ , leaving $G^{\mathrm{ret}}=\delta(\tau-R/c)/4\pi R$ . Why: the discarded second delta is precisely the advanced piece the retarded pole placement threw away. *Two quick checks confirm the result: a temporal Fourier transform of $G^{\mathrm{ret}}$ returns the Helmholtz outgoing kernel $e^{ikr}/4\pi r$ , and integrating over all $\tau$ recovers the static Poisson kernel $1/4\pi R$ .*

Go deeper · Derivation B · 1 minThe real-space route: spherical reduction to a one-dimensional wave

Derivation BThe real-space route: spherical reduction to a one-dimensional wave

The goal is to reach the same $\delta(\tau-R/c)/4\pi R$ without any Fourier machinery, by exploiting spherical symmetry away from the source. The key structural fact is that the radial Laplacian collapses onto a one-dimensional wave operator for the combination $u=RG$ , so the 3D problem becomes a d'Alembert problem whose outgoing solution, matched to the static singularity at the origin, fixes the answer uniquely. This route makes the light-cone shell a one-dimensional pulse dressed by geometric spreading.

Step 1. Reduce the Laplacian. For a spherically symmetric function the identity $\nabla^2 f=R^{-1}\partial_R^2(Rf)$ turns the source-free wave equation into $c^{-2}\partial_\tau^2(RG)=\partial_R^2(RG)$ , a one-dimensional wave equation for $u=RG$ . Why: multiplying by $R$ absorbs the curvature of spherical spreading, leaving a flat operator we already know from d'Alembert.

Step 2. Keep the outgoing solution. The general solution $u=f(\tau-R/c)+g(\tau+R/c)$ loses $g$ under the retarded condition (nothing incoming from infinity), so $RG=f(\tau-R/c)$ . Why: the outgoing/incoming split of a 1D wave is exactly the retarded/advanced split of the shell, made visible before any Fourier transform.

Step 3. Match the source at the origin. As $R\to0$ the time-dependence drops out and $G$ must reproduce the static Poisson kernel $1/4\pi R$ , forcing $f(\tau)=\delta(\tau)/4\pi$ and hence $G^{\mathrm{ret}}=\delta(\tau-R/c)/4\pi R$ . Why: near the source the wave equation reduces to Poisson's, so electrostatics pins the shell's weight. *This agrees with the contour-integration result of the previous derivation; notably, the $1/4\pi R$ weight is fixed here by an entirely independent route — the static limit alone — confirming that the geometric factor is not an artifact of the Fourier method.*

Go deeper · Derivation C · 1 minMethod of descent and the tail that breaks Huygens in two dimensions

Derivation CMethod of descent and the tail that breaks Huygens in two dimensions

The goal is to build the 2D retarded Green's function from the 3D one by the method of descent — treating a 2D source as a uniform line of 3D point sources stacked along the missing third axis — and to expose the surprise it produces: the sharp shell of 3D smears into a front followed by a slowly decaying tail, the concrete failure of Huygens' principle in even dimensions. This is why a struck drumhead rings while a point clap in open air merely clicks.

Step 1. Stack sources along the missing axis. A source independent of $z$ is effectively 2D, so $G_2^{\mathrm{ret}}(\boldsymbol{\rho},\tau)=\int_{-\infty}^{\infty}\frac{\delta(\tau-\sqrt{\rho^2+z^2}/c)}{4\pi\sqrt{\rho^2+z^2}}\,\mathrm{d}z$ , with $\boldsymbol{\rho}$ the in-plane separation and $R=\sqrt{\rho^2+z^2}$ the true 3D distance. Why: descent trades one dimension for an integral over the ignorable coordinate, superposing point-source shells into a line source.

Step 2. Change to the distance variable. With $\mathrm{d}z=R\,\mathrm{d}R/\sqrt{R^2-\rho^2}$ and the two branches $z\gtrless 0$ contributing equally, $G_2^{\mathrm{ret}}=\frac{2}{4\pi}\int_\rho^\infty\frac{\delta(\tau-R/c)}{\sqrt{R^2-\rho^2}}\,\mathrm{d}R$ . Why: the light-cone delta now acts on its natural variable, and $R=\rho$ is the closest a stacked source can be.

Step 3. Fire the delta and read off the tail. The delta selects $R=c\tau$ , allowed only when $c\tau\ge\rho$ , giving $G_2^{\mathrm{ret}}=\dfrac{c\,\Theta(c\tau-\rho)}{2\pi\sqrt{c^2\tau^2-\rho^2}}$ — singular at the front $c\tau=\rho$ and nonzero for all later times, decaying as $1/2\pi\tau$ . Why: the 2D response never switches back off, because contributions keep arriving from ever-more-distant parts of the stacked line — the tail that marks even dimensions. *The result is singular at the wavefront and falls off as $\sim(c\tau)^{-1}$ far behind it. Descending one further dimension gives the one-dimensional case, the Heaviside step $\tfrac{c}{2}\Theta(c\tau-|x|)$ .*

Limiting cases

The wave Green's function changes its nature with the spatial dimension and the temporal condition imposed. The table collects the canonical forms and their meaning.

Case	Result	Meaning
1D (d'Alembert)	$G^{\mathrm{ret}}_1 = \tfrac{c}{2}\,\Theta\!\big(c\tau - \vert x-x'\vert\big)$	A Heaviside step filling the cone interior; strong Huygens fails, and once the front passes the disturbance never leaves.
2D (membrane)	$G^{\mathrm{ret}}_2 = \dfrac{c\,\Theta(c\tau-\rho)}{2\pi\sqrt{c^2\tau^2 - \rho^2}}$	A front at $c\tau=\rho$ followed by a tail decaying as $1/\tau$ ; even-dimensional waves reverberate.
3D (strong Huygens)	$G^{\mathrm{ret}}_3 = \dfrac{1}{4\pi R}\,\delta\!\big(\tau - R/c\big)$	Support exactly on the cone $R=c\tau$ : sharp wavefronts, no tail, clean signalling.
Static / zero-frequency	$\displaystyle\int_{-\infty}^{\infty}\! G^{\mathrm{ret}}_3\,\mathrm{d}\tau = \dfrac{1}{4\pi R}$	Time-integrating recovers the Poisson kernel; a steady source gives a Coulomb-like potential.
Helmholtz / single frequency	$G_\omega = \dfrac{e^{ikR}}{4\pi R},\ k=\omega/c$	The Fourier component of $G^{\mathrm{ret}}$ is the outgoing Helmholtz kernel; advanced gives $e^{-ikR}/4\pi R$ .

Worked examples

1 · Retarded potential of an oscillating point source (confidence builder)

A small antenna at the origin drives $s=q_0\,\delta^3(\mathbf{r}')\cos(\omega t)$ at $f=100\ \mathrm{MHz}$ . Find the field at an observer $r=30\ \mathrm{m}$ away, and the time delay and phase lag.

Step 1. Apply the retarded potential. With a point source, equation (5) collapses the spatial integral: $\phi(r,t)=\dfrac{q_0}{4\pi r}\cos\!\big(\omega(t-r/c)\big)$ — a delayed, weakened copy of the source. The amplitude factor is $1/4\pi r=1/(4\pi\cdot 30)=\mathbf{2.65\times10^{-3}\ m^{-1}}$ times $q_0$ .

Step 2. Compute the retarded delay. The observer feels what the source did $r/c$ ago: $\Delta t=r/c=30/(2.998\times10^8)=\mathbf{100\ ns}$ . Since $T=1/f=10\ \mathrm{ns}$ , this is exactly $10.0$ periods — the field replays the source from ten cycles earlier.

Step 3. Compute the phase lag. The lag is $\varphi=\omega\Delta t=kr=2\pi r/\lambda$ with $\lambda=c/f=2.998\ \mathrm{m}$ , giving $\varphi=2\pi(30/2.998)=\mathbf{62.9\ rad}$ , i.e. $10.007$ cycles. Sitting almost exactly ten wavelengths out, the delayed field is essentially back in step. Delay = 100 ns = 10.0 periods; phase lag = 62.9 rad ≈ 10 wavelengths; amplitude ∝ 1/4πr = 2.65×10⁻³ m⁻¹.

2 · The sharp click versus the ringing tail (dimension matters)

A point impulse fires at separation $R=686\ \mathrm{m}$ with wave speed $c=343\ \mathrm{m/s}$ . Describe what a detector records in 3D, 2D, and 1D.

Step 1. Find the front time. In every dimension the disturbance first arrives when the cone reaches the detector, at $\tau_{\mathrm{front}}=R/c=686/343=\mathbf{2.00\ s}$ . Before this the detector is in the causal shadow, recording zero.

Step 2. The 3D response. Here $G_3=\delta(\tau-R/c)/4\pi R$ : the detector hears one sharp click at τ = 2.00 s, weight $1/4\pi R=1.16\times10^{-4}$ — silence, spike, silence. A point impulse in open 3D air is a clap, not a rumble.

Step 3. The 2D and 1D responses. In 2D, $G_2=c\,\Theta(c\tau-R)/2\pi\sqrt{c^2\tau^2-R^2}$ diverges at the front and keeps sounding: at $\tau=4.00\ \mathrm{s}$ it is still $343/(2\pi\cdot1188\ \mathrm{m})=\mathbf{0.046}$ , decaying as $1/2\pi\tau$ . In 1D, $G_1=\tfrac{c}{2}\Theta(c\tau-R)$ jumps to $c/2$ at $\tau=2.00\ \mathrm{s}$ and stays. 3D: a lone click at 2.00 s; 2D: a front at 2.00 s then an endless tail; 1D: a permanent step at 2.00 s.

3 · The difference is source-free: checking the Pauli–Jordan function

Show that $\Delta=G^{\mathrm{ret}}-G^{\mathrm{adv}}$ solves the homogeneous wave equation and vanishes outside the light cone, and interpret the result.

Step 1. Subtract the two inhomogeneous equations. Both satisfy $\Box G=\delta^3\delta$ with the same right-hand side, so $\Box\Delta=0$ : $\Delta$ is source-free. From equation (6), $\Delta=[\delta(\tau-R/c)-\delta(\tau+R/c)]/4\pi R$ is nonzero only on the cones $\tau=\pm R/c$ , is odd in $\tau$ , and vanishes for spacelike separations $|\tau|<R/c$ .

Step 2. Read the physics. $\Delta$ is a freely propagating superposition of outgoing and incoming shells carrying no source; in QFT it is the Pauli–Jordan function setting the commutator $[\phi(x),\phi(y)]\propto\Delta(x-y)$ . Its vanishing outside the light cone is microcausality: operators at spacelike separation commute, so nothing influences anything faster than light. $\Box\Delta=0$ , supported on the cone, odd in $\tau$ — the kernel of relativistic causality.

Physics of the portrait

The retarded Green's function is where a century of mathematical physics converges: a device invented for electrostatics, sharpened into the engine of radiation theory, and finally read as the statement of causality itself.

Origins · 1 minretarded Green's function and how this was discovered

retarded Green's function

the causal impulse response G = δ(τ − R/c) ⁄ 4πR · fundamental solution of the wave operator

The idea of a "Green's function" begins with George Green's privately printed 1828 essay, which introduced the reciprocal function now bearing his name to solve electrostatics — a static theory with no light cone in sight. The dynamical, retarded version awaited the wave theory of the field. Bernhard Riemann and, decisively, Ludvig Lorenz in 1867 wrote the electromagnetic potentials as integrals over sources at the retarded time $t-R/c$ , making finite propagation speed explicit for the first time. Gustav Kirchhoff cast the retarded solution as a rigorous surface integral in 1882, and Liénard (1898) and Wiechert (1900) specialized it to a moving charge — the Liénard–Wiechert potentials behind synchrotron light and antennas. The quantum era gave the same kernel new meaning: Jordan and Pauli in 1928 built the relativistic commutator function from $G^{\mathrm{ret}}-G^{\mathrm{adv}}$ , and Wheeler and Feynman in 1945 founded absorber electrodynamics on the time-symmetric average. What Green built for static charges became, in the wave equation, the mathematical form of cause preceding effect.

History

1828

— Green's essay founds the Green's-function method in potential theory

1867

— Lorenz writes the electromagnetic field as retarded potentials over past sources

1882

— Kirchhoff casts the retarded wave solution as a rigorous surface integral

1898

— Liénard, and Wiechert in 1900, give the retarded potentials of a moving point charge

1945

— Wheeler and Feynman found absorber electrodynamics on the time-symmetric combination

Going further

Texts. Stone & Goldbart, Mathematics for Physics, wave-equation Green functions · Arfken & Weber, Mathematical Methods for Physicists, Green's-function chapter · Hassani, Mathematical Physics, Green's functions in $m$ dimensions · Jackson, Classical Electrodynamics, §6.4 and ch. 14 · Morse & Feshbach, Methods of Theoretical Physics, vol. I, ch. 7 · Barton, Elements of Green's Functions and Propagation. Key consistency relations: the contour integration yields the light-cone delta with $G^{\mathrm{ret}}=0$ for $\tau<0$ ; the descent integral produces the 2D tail; the static limit gives $1/4\pi R$ and the single-frequency component is $e^{ikR}/4\pi R$ . The worked numbers are: delay $r/c=100\ \mathrm{ns}$ at $r=30\ \mathrm{m}$ ; wavefront arrival $R/c=2.0\ \mathrm{s}$ at $R=686\ \mathrm{m}$ , $c=343\ \mathrm{m/s}$ .

Prerequisites. The Wave Equation · d'Alembert's Solution · The Helmholtz Equation · Classification of Second-Order Linear PDEs · Boundary Conditions and Well-Posedness. Leads to. The Method of Characteristics · Sturm–Liouville Theory and Eigenfunction Expansions · Separation of Variables in Spherical Coordinates.

Ask the tutor: Ask me to walk through the contour integral pole by pole, or to show exactly how the 2D tail grows out of stacking 3D light-cone shells.

Why is the retarded solution "the physical one" if the advanced solution solves the same equation?

Because the differential equation is time-symmetric but the world we model is not. Both $G^{\mathrm{ret}}$ and $G^{\mathrm{adv}}$ satisfy $\Box G=\delta^3\delta$ exactly; nothing in the equation prefers one. The preference is a boundary condition matching experience: fields radiate outward from sources switched on in the past, so we demand the response vanish before the source acts, selecting the retarded branch. The advanced solution is not wrong — it answers a differently-posed problem (fields converging onto a future sink) and is essential in scattering theory. Causality is an input, not a theorem, and the two Green's functions faithfully offer both choices.

How can causality "emerge automatically" from the contour derivation if we never put it in?

We do put it in — quietly, as the sign of $i\varepsilon$ . Choosing $\omega\to\omega+i\varepsilon$ pushes both poles into the lower half of the frequency plane. Evaluating the inverse transform for $\tau<0$ , Jordan's lemma forces you to close in the upper half-plane, where there are now no poles, so the integral is exactly zero. That vanishing "for free" is the payoff of the boundary condition smuggled in through the pole placement. Flip to $-i\varepsilon$ and the function vanishes for $\tau>0$ instead — the advanced case. So the $i\varepsilon$ is not a computational nicety; it is the boundary condition in disguise.

Why is the 3D Green's function a razor-thin shell rather than a filled-in ball?

Because in three dimensions the wave equation supports perfectly sharp, undistorted propagation — strong Huygens' principle. The radial reduction $u=RG$ obeys a 1D wave equation whose solution $f(\tau-R/c)$ is a rigidly translating pulse; a delta source stays a delta, riding out at speed $c$ with only the geometric $1/R$ dilution. Physically, every point on an expanding wavefront radiates wavelets that reconstruct the advancing front and cancel behind it. This cancellation is special to odd dimensions; in even dimensions it is incomplete and you get the filled-in tail instead.

What is the relationship between the Green's function and the QFT propagator?

They are the same object seen through different boundary conditions. The retarded and advanced Green's functions are two ways of routing the frequency contour around the poles $\omega=\pm ck$ . Quantum field theory adds a third — the Feynman prescription — sending positive-frequency (particle) modes forward in time and negative-frequency (antiparticle) modes backward, giving the Feynman propagator $G_F$ . All satisfy the same inhomogeneous equation and differ only in the $i\varepsilon$ rule; their differences are homogeneous solutions like $\Delta=G^{\mathrm{ret}}-G^{\mathrm{adv}}$ . So the two Green's functions and the Feynman propagator are one rational function with three inequivalent contour choices.

How does this reduce to d'Alembert's solution in one dimension?

Directly, and it is a good sanity check. In one dimension the retarded Green's function is not a delta on the cone but the Heaviside step $G_1^{\mathrm{ret}}=\tfrac{c}{2}\Theta(c\tau-|x-x'|)$ . Convolving a source with this step, or initial data with its derivatives, reproduces d'Alembert's formula: the solution at $(x,t)$ averages the initial displacement over the characteristics $x\pm ct$ plus an integral of the initial velocity between them. The "filled interior" of the 1D Green's function is why d'Alembert integrates over a whole interval rather than sampling a single retarded point — one dimension has no strong Huygens' principle.

Does using the retarded potential mean an accelerating charge "feels" its own past field?

In a precise sense, yes, and that is the origin of radiation reaction. The retarded potential builds the field from the source's behaviour at the retarded time. For an accelerating point charge, its own retarded field evaluated back at the charge does not fully cancel — the leftover is the radiation-reaction force (Abraham–Lorentz, $\propto\dot{\mathbf a}$ ) that drains energy into radiation. Wheeler and Feynman's absorber theory recovers the same force from the time-symmetric $\bar G$ plus the advanced response of future absorbers. Either way, the retardation is no bookkeeping trick: the delay between cause and self-effect is real and measurable in the energy an antenna loses.