Causal, Retarded, and Advanced Green's Functions

A Green's function is the impulse response of a linear system: solve $L_t\,G(t,t')=\delta(t-t')$ once, and every driven problem $L_t\,x(t)=f(t)$ is solved by a single convolution, $x(t)=\int G(t,t')\,f(t')\,\mathrm{d}t'$ , plus a homogeneous solution. For a self-adjoint spatial operator the boundary conditions in space pin $G$ down uniquely. The time operator is different. As an initial-value problem it is not self-adjoint, its kernel is two-dimensional, and that leftover freedom — the freedom to add any solution of $L\,G=0$ — is exactly the freedom to decide whether an effect is allowed to precede its cause. Demand that the response never runs ahead of the force and you select the retarded Green's function $G_R(t-t')$ , supported on $t\ge t'$ . Reverse the arrow and you get the mathematically legal but acausal advanced function $G_A$ , supported on $t\le t'$ . The two differ by a homogeneous solution and obey $G_A(t,t')=G_R(t',t)$ . The cleanest way to control which one you get is the Fourier transform: $\tilde G(\omega)=1/p(-i\omega)$ , where $p$ is the characteristic polynomial of $L$ . Its poles sit on or near the real axis, and the inverse transform must say how the contour steps around them. An infinitesimal shift $\omega\to\omega+i\epsilon$ pushes the poles off the axis and — through Jordan's lemma and contour closure — encodes retarded, advanced, or Feynman boundary conditions. This one analytic structure recurs everywhere: it is the backbone of linear-response theory and the fluctuation–dissipation theorem, the origin of the Kramers–Kronig dispersion relations, the mechanism of retarded electromagnetic potentials, and — with the Feynman $i\epsilon$ — the propagator of quantum field theory. Master the contour bookkeeping and you have mastered causality in linear systems.

Key takeaways

A Green's function $G(t,t')$ inverts a linear time operator $L$ , solving $L\,G=\delta(t-t')$ ; the homogeneous solution you add fixes the boundary conditions, and hence the physics.
The retarded function $G_R$ vanishes for $t<t'$ (effect follows cause); the advanced $G_A$ vanishes for $t>t'$ (time-reversed); their difference is a homogeneous solution.
In frequency space causality is a rule for stepping around poles: pushing the poles of $\tilde G(\omega)$ into the lower half-plane ( $\omega\to\omega+i\epsilon$ ) gives $G_R$ , the upper half-plane gives $G_A$ .
The Feynman propagator uses a mixed prescription — positive frequencies forward, negative frequencies backward — giving the time-ordered two-point function at the heart of QFT.
Causality is equivalent to analyticity of $\tilde G_R(\omega)$ in the upper half-plane, and that single fact powers the Kramers–Kronig relations linking dispersion to absorption.
The same object unifies driven classical oscillators, electromagnetic radiation, quantum scattering, and field propagators under one contour picture.
Key idea: the equation $L\,G=\delta$ has no unique solution; the homogeneous piece you add — encoded as an $i\epsilon$ pole prescription in frequency space — is what selects causality. Retarded, advanced, and Feynman propagators are one $\tilde G(\omega)$ with three different contour deformations.

The core insight

The differential equation $L\,G=\delta$ does not have a unique solution; the choice of homogeneous solution added — encoded analytically as an $i\epsilon$ pole prescription in frequency space — is what selects causality. Retarded, advanced, and Feynman propagators are the same $\tilde G(\omega)$ with different contour deformations.

Where this lives today

The $i\epsilon$ contour is a working tool, not a museum piece. In materials science, Kramers–Kronig analysis routinely extracts a full complex dielectric function from a reflectivity measurement — measure absorption over a wide band and the dispersion relation hands you the refractive index, the exact causality-enforced link derived above. In gravitational-wave astronomy the LIGO–Virgo–KAGRA detectors model a black-hole merger's ringdown through quasinormal modes: the complex poles of a retarded Green's function in the lower half-plane, the first worked example applied to spacetime itself. In quantum many-body theory, retarded and time-ordered Green's functions are the central objects of the GW approximation and dynamical mean-field theory, computed daily for photoemission spectra and quasiparticle lifetimes. At the frontier, the analytic structure of Feynman propagators — the placement and pinching of $i\epsilon$ poles — drives the modern amplitudes program (unitarity cuts, the optical theorem, Landau singularities) behind LHC cross-sections. A live open problem is the pole structure of finite-temperature and out-of-equilibrium propagators, where the clean upper-half-plane analyticity of the vacuum function deforms and the right contour is still being mapped.

The physical picture

Picture a struck bell. The retarded Green's function is the bell's response to one sharp tap at $t'$ : silence before, ringing after. There is no physical world in which the bell rings before you strike it — that acausal solution is the advanced Green's function, perfectly valid algebra but reversed in time. Real damping breaks the symmetry between the two: a struck bell rings down, it does not ring up out of nothing. The first figure puts the two responses of a damped oscillator side by side across the moment of the impulse.

Figure

Impulse response of a damped oscillator with ω₀ = 5 rad/s and γ = 1/s, so ω_d = √(ω₀² − γ²) = 4.90 rad/s and quality factor Q = ω₀/2γ = 2.5: the retarded G_R (blue) rings down only after the tap at t = 0, while the advanced G_A (amber, dashed) is its acausal mirror living entirely before the tap

Now go to frequency space. Every natural mode of the system is a pole of $\tilde G(\omega)$ . Damping moves those poles off the real axis; for a stable system they sit in the lower half-plane. For an ideal undamped oscillator the poles land exactly on the axis, and the inverse transform is ambiguous until you say which side to pass them — the $i\epsilon$ prescription is nothing but a whisker of damping that resolves the tie. Closing the inverse-transform contour then becomes the question "which way does time run?": for $t>t'$ the kernel $e^{-i\omega(t-t')}$ decays in the lower half-plane, so only poles pushed below the axis contribute, and the response is nonzero only after the source. That is causality, drawn as geometry. The second figure is the master diagram of the whole subject.

Figure

The complex ω-plane is where the boundary condition is chosen: the retarded prescription ω → ω + iε puts both poles at ±ω₀ − iε just below the real axis (blue), so closing downward for t greater than 0 encloses them and gives G_R = −2πi Σ Res; the advanced prescription puts the poles above (amber), and the Feynman rule splits them one up and one down

The Feynman propagator is the compromise relativity forces on us. Antiparticles make the clean split impossible: positive-energy excitations must move forward in time, negative-energy ones backward. So the Feynman contour does not push both poles the same way — it weaves between them, sending the $+\omega_0$ pole down and the $-\omega_0$ pole up. The result is not a strictly causal response but the time-ordered product, and that is precisely the object Feynman diagrams are built from.

The core equations

A Green's function is the inverse of a differential operator: feed the operator a unit impulse and read off the response. Everything downstream computes or reshapes this one object.

L_t\,G(t,t') = \delta(t-t') \qquad (1)

where: $L_t$ — a linear time operator, e.g. $\partial_t^2+2\gamma\partial_t+\omega_0^2$ ; $G(t,t')$ — the response at time $t$ to a unit impulse at $t'$ ; $\delta(t-t')$ — the impulse localized at $t=t'$ . For time-translation-invariant $L$ , $G$ depends only on the elapsed time $t-t'$ . Because $L$ is linear, an arbitrary forcing is a superposition of impulses and the response is the matching superposition of Green's functions — a single convolution. The homogeneous term carries the initial data, and it is where the retarded/advanced choice lives.

x(t)=x_{\mathrm{hom}}(t)+\int_{-\infty}^{\infty} G(t,t')\,f(t')\,\mathrm{d}t' \qquad (2)

where: $f(t')$ — the source term; $x_{\mathrm{hom}}$ — any solution of $L\,x_{\mathrm{hom}}=0$ . Adding $x_{\mathrm{hom}}$ leaves $L\,x=f$ intact, so $G$ is not unique until a boundary condition is imposed. Fourier transforming turns the operator into multiplication by its characteristic polynomial, so inverting $L$ becomes ordinary division. The zeros of that polynomial — the natural frequencies — become the poles that carry all the physics.

\tilde G(\omega)=\frac{1}{p(-i\omega)},\qquad p(D)=L \qquad (3)

where: $p$ — the characteristic polynomial of $L$ , with $p(-i\omega)=-\omega^2+\omega_0^2$ for the undamped oscillator; $\omega$ — the Fourier variable. Throughout, the inverse transform is $G(t)=\int\frac{\mathrm{d}\omega}{2\pi}\,\tilde G(\omega)\,e^{-i\omega t}$ . Transforming back is where the ambiguity surfaces: the poles sit on the real axis and the integral is undefined until the contour is told how to pass them. An infinitesimal imaginary shift of $\omega$ does exactly that, and its sign labels the boundary condition.

G(t)=\int_{-\infty}^{\infty}\frac{\mathrm{d}\omega}{2\pi}\;\frac{e^{-i\omega t}}{-(\omega\pm i\epsilon)^2+\omega_0^2} \qquad (4)

where: $\epsilon$ — a positive infinitesimal displacing the poles off the axis, $+$ giving retarded and $-$ advanced; $\omega_0$ — the natural frequency. The $+i\epsilon$ moves both poles into the lower half-plane, forcing $G_R(t)=0$ for $t<0$ . Carrying out the contour integral for the undamped oscillator gives the causal ringdown — a pure sine switched on by a step, the canonical result to memorize.

G_R(t)=\frac{\sin(\omega_0 t)}{\omega_0}\,\Theta(t) \qquad (5)

where: $\Theta(t)$ — the Heaviside step enforcing $G_R=0$ for $t<0$ ; $\sin(\omega_0 t)/\omega_0$ — the homogeneous solution with $G(0)=0,\ \dot G(0)=1$ . The unit jump in the velocity at $t=0$ is what regenerates the delta source. The advanced function is the retarded one time-reversed: swap the arguments and the support flips to $t\le t'$ . Their difference is a source-free ringing solving $L\,G=0$ — the non-uniqueness of equation (2) made concrete.

G_A(t,t')=G_R(t',t),\qquad G_A(t)=-\frac{\sin(\omega_0 t)}{\omega_0}\,\Theta(-t) \qquad (6)

where: $\Theta(-t)$ — support for $t<0$ ; $G_R-G_A=\sin(\omega_0 t)/\omega_0$ is a homogeneous solution valid for all $t$ . The two functions solve the same equation and differ only in which way causality points. The bridge from a pole prescription to measurable content is the Sokhotski–Plemelj identity, which splits the $i\epsilon$ into a principal value (dispersive, reactive) and a delta (absorptive, on-shell).

\frac{1}{x\pm i\epsilon}=\mathrm{P}\,\frac{1}{x}\mp i\pi\,\delta(x) \qquad (7)

where: $\mathrm{P}$ — the Cauchy principal value; $\delta(x)$ — the delta from the imaginary shift. This is the algebraic engine behind energy conservation in scattering and the real/imaginary split of any response function. Analyticity of $\tilde G_R$ in the upper half-plane — causality restated — ties its real and imaginary parts together through a principal-value integral: the Kramers–Kronig relations.

\mathrm{Re}\,\tilde G_R(\omega)=\frac{1}{\pi}\,\mathrm{P}\!\int_{-\infty}^{\infty}\frac{\mathrm{Im}\,\tilde G_R(\omega')}{\omega'-\omega}\,\mathrm{d}\omega' \qquad (8)

where: $\mathrm{Re},\mathrm{Im}$ — dispersive and absorptive parts; $\mathrm{P}\int$ — a principal-value integral over all frequencies. Refraction without absorption somewhere is impossible: the two are halves of one analytic function. The relativistic generalization sends positive frequencies forward and negative frequencies backward through a single $+i\epsilon$ in the mass term — the Feynman propagator, one internal line of every diagram.

D_F(x-y)=\int\frac{\mathrm{d}^4p}{(2\pi)^4}\,\frac{i\,e^{-ip\cdot(x-y)}}{p^2-m^2+i\epsilon} \qquad (9)

where: $p^2=p_0^2-\mathbf{p}^2$ — the Minkowski square of the four-momentum; $m$ — the mass, fixing poles at $p_0=\pm\sqrt{\mathbf{p}^2+m^2}$ ; $i\epsilon$ — the prescription that time-orders the fields. The same $\pm i\epsilon$ bookkeeping as the oscillator, now in four dimensions.

Derivation AThe retarded oscillator Green's function by contour integration

The goal is to compute $G_R(t)$ for the undamped oscillator $\ddot G+\omega_0^2 G=\delta(t)$ from its frequency-space form, and to watch causality emerge not as an extra assumption but as a forced consequence of which half-plane the contour closes in. We push both poles below the real axis, split the cases $t<0$ and $t>0$ , use Jordan's lemma to fix the closure, collect the residues, and confirm the answer reproduces the delta source. The whole argument is sign-and-half-plane bookkeeping, and getting that right is the entire skill.

Result (causal ringdown).

G_R(t)=\frac{\sin(\omega_0 t)}{\omega_0}\,\Theta(t)

Step 1. Write the inverse transform with the retarded shift, $G_R(t)=\int_{-\infty}^{\infty}\frac{\mathrm{d}\omega}{2\pi}\,\frac{e^{-i\omega t}}{\omega_0^2-(\omega+i\epsilon)^2}$ , whose denominator factors as $-(\omega+i\epsilon-\omega_0)(\omega+i\epsilon+\omega_0)$ ; the poles sit at $\omega=\pm\omega_0-i\epsilon$ , both a hair below the real axis. Why: the sign of the $i\epsilon$ is the boundary condition in disguise — poles in the lower half-plane will force the response to vanish before the impulse.

Step 2. Fix the closure from the kernel: with $\omega=a+ib$ , the factor $e^{-i\omega t}=e^{-iat}e^{bt}$ decays as $b\to+\infty$ for $t<0$ and as $b\to-\infty$ for $t>0$ . So for $t<0$ close upward, where no poles live, giving $G_R(t)=0$ . Why: this step is causality itself — the response is zero before the source purely because the poles were placed on the far side.

Step 3. For $t>0$ close downward, encircling both poles clockwise, so $\oint=-2\pi i\sum\mathrm{Res}$ . Why: the clockwise minus sign must be tracked — dropping it turns a decaying response into a growing one and breaks stability.

Step 4. The residues of $\frac{1}{2\pi}\cdot\frac{-e^{-i\omega t}}{(\omega-\omega_0)(\omega+\omega_0)}$ are $\frac{-e^{-i\omega_0 t}}{2\pi(2\omega_0)}$ at $\omega=\omega_0$ and $\frac{e^{+i\omega_0 t}}{2\pi(2\omega_0)}$ at $\omega=-\omega_0$ ; now let $\epsilon\to0$ . Why: taking $\epsilon\to0$ only after assigning the poles to a half-plane is the whole trick — the contour topology is already fixed.

Step 5. Sum: $G_R(t)=-2\pi i\,\frac{-e^{-i\omega_0 t}+e^{+i\omega_0 t}}{2\pi(2\omega_0)}=-i\,\frac{e^{+i\omega_0 t}-e^{-i\omega_0 t}}{2\omega_0}$ . Why: the two residues combine into the pure sine a struck oscillator rings with.

Step 6. Use $\sin x=(e^{ix}-e^{-ix})/2i$ to get $G_R(t)=\sin(\omega_0 t)/\omega_0$ for $t>0$ , then multiply by $\Theta(t)$ . Why: the step is not cosmetic — its derivative regenerates the delta source, closing the loop. *Verified: substituting $G_R=\sin(\omega_0 t)\Theta(t)/\omega_0$ gives $\dot G_R=\cos(\omega_0 t)\Theta(t)$ (the delta term dies because $\sin 0=0$ ) and $\ddot G_R=-\omega_0^2 G_R+\delta(t)$ , so $\ddot G_R+\omega_0^2 G_R=\delta(t)$ . Matches Stone & Goldbart Ch. 9 and Riley–Hobson–Bence Ch. 21.*

Go deeper · Derivation B · 1 minThe Sokhotski–Plemelj identity and the anatomy of the iε shift

Derivation BThe Sokhotski–Plemelj identity and the anatomy of the iε shift

The goal is to show what an $i\epsilon$ is once you stop treating it as a formal token: a nascent distribution that resolves, as $\epsilon\to0$ , into a principal value plus a delta. This is why the same prescription that enforces causality in time also enforces energy conservation in scattering — the principal value is the reactive part, the delta the absorptive, on-shell part. We rationalize the denominator, then take the distributional limit of each piece.

Result (the identity).

\frac{1}{x\pm i\epsilon}=\mathrm{P}\,\frac{1}{x}\mp i\pi\,\delta(x)

Step 1. Multiply by the conjugate denominator: $\frac{1}{x\pm i\epsilon}=\frac{x\mp i\epsilon}{x^2+\epsilon^2}$ , separating a real and an imaginary piece. Why: the two limit to genuinely different distributions, so keeping them apart prevents conflating dispersion with absorption.

Step 2. The real part $\frac{x}{x^2+\epsilon^2}$ tends to $1/x$ away from the origin and is odd through it, i.e. the Cauchy principal value $\mathrm{P}\,1/x$ . Why: the principal value is the finite, symmetric integration through a pole — the reactive response that survives.

Step 3. The imaginary part $\frac{\epsilon}{x^2+\epsilon^2}$ is a Lorentzian of fixed area $\int_{-\infty}^{\infty}\frac{\epsilon\,\mathrm{d}x}{x^2+\epsilon^2}=\pi$ , so as $\epsilon\to0$ it collapses to $\pi\,\delta(x)$ . Why: this delta is the absorptive part — the on-shell energy-conserving delta in scattering, the peak of $\mathrm{Im}\,n(\omega)$ in optics.

Step 4. Assemble $\frac{1}{x\pm i\epsilon}=\mathrm{P}\,\frac1x\mp i\pi\,\delta(x)$ , the delta's sign set by the shift's. Why: flipping $\epsilon$ flips the absorptive term — exactly the difference between outgoing-wave ( $+i\epsilon$ ) and incoming-wave ( $-i\epsilon$ ) boundary conditions. *Verified: the area $\int\epsilon/(x^2+\epsilon^2)\,\mathrm{d}x=\pi$ is standard, and the real/imaginary split reproduces the Arfken Ch. 11 statement of the identity. Applying it to $\tilde G(\omega)=1/(\omega_0^2-(\omega+i\epsilon)^2)$ recovers the delta contributions at $\omega=\pm\omega_0$ .*

Go deeper · Derivation C · 1 minThe Kramers–Kronig relations from causality

Derivation CThe Kramers–Kronig relations from causality

The goal is to derive the dispersion relations directly from $G_R(t)=0$ for $t<0$ , showing that "no response before the cause" and "analytic in the upper half-plane" are one fact in two guises. With analyticity in hand, Cauchy's theorem on a contour hugging the real axis and closing over the top does the rest, and the real/imaginary split gives the two Kramers–Kronig formulas tying refraction to absorption. This is the deepest payoff of the subject: a kinematic taboo becomes a hard, testable constraint on a material's spectrum.

Result (dispersion relation).

\tilde G_R(\omega)=\frac{1}{i\pi}\,\mathrm{P}\!\int_{-\infty}^{\infty}\frac{\tilde G_R(\omega')}{\omega'-\omega}\,\mathrm{d}\omega'

Step 1. Because $G_R(t)=0$ for $t<0$ , its transform $\tilde G_R(\omega)=\int_0^{\infty}G_R(t)\,e^{i\omega t}\,\mathrm{d}t$ starts at zero, and for $\mathrm{Im}\,\omega>0$ the factor $e^{-(\mathrm{Im}\,\omega)t}$ makes it converge and holomorphic throughout the upper half-plane. Why: this is the pivot — causality is not merely suggestive of analyticity, it is logically equivalent to it.

Step 2. Take $\oint\frac{\tilde G_R(\omega')}{\omega'-\omega}\,\mathrm{d}\omega'$ along the real axis, indented by a small semicircle around $\omega'=\omega$ and closed over the top; enclosing no singularity, it vanishes. Why: closing upward is legitimate only because Step 1 kills $\tilde G_R$ on the large arc there.

Step 3. The large arc contributes nothing, and the small semicircle contributes half a residue, $-i\pi\,\tilde G_R(\omega)$ , its sign fixed by the clockwise indentation. Why: this half-residue — a pole sitting on the path — is what turns an equality-to-zero into a relation for $\tilde G_R$ itself.

Step 4. Collecting, $\mathrm{P}\!\int\frac{\tilde G_R(\omega')}{\omega'-\omega}\,\mathrm{d}\omega'-i\pi\,\tilde G_R(\omega)=0$ — the boxed relation; the real/imaginary split gives the two Kramers–Kronig integrals. Why: that split is the usable form — predict a material's refraction across the spectrum from its absorption, enforced by causality alone. *Verified: the structure appears in Jackson Ch. 7 and Arfken Ch. 11; the real/imaginary split reproduces the standard dielectric dispersion relations, and applied to a single-pole susceptibility it returns the Lorentzian pair $\mathrm{Re}\,\chi,\ \mathrm{Im}\,\chi$ exactly.*

Limiting cases

The pole prescription is everything: fix the poles' half-plane and you have fixed the boundary condition. With the convention $G(t)=\int\frac{\mathrm{d}\omega}{2\pi}\,\tilde G(\omega)\,e^{-i\omega t}$ and characteristic poles at $\pm\omega_0$ , the four canonical choices are:

Case	Result	Meaning
Retarded	$\tilde G_R(\omega)=\dfrac{1}{-(\omega+i\epsilon)^2+\omega_0^2}$	both poles pushed to the lower half-plane; $G_R(t)=0$ for $t<0$ ; response strictly follows cause
Advanced	$\tilde G_A(\omega)=\dfrac{1}{-(\omega-i\epsilon)^2+\omega_0^2}$	both poles in the upper half-plane; $G_A(t)=0$ for $t>0$ ; time-reversed and acausal
Feynman	$\tilde G_F(\omega)=\dfrac{1}{-\omega^2+\omega_0^2-i\epsilon}$	$+\omega_0$ pole down, $-\omega_0$ pole up; positive frequencies forward, negative backward; time-ordered
Principal value	$\tilde G_P(\omega)=\mathrm{P}\,\dfrac{1}{-\omega^2+\omega_0^2}$	the symmetric average $\tfrac12(G_R+G_A)$ ; a standing wave with no net causal direction
Damped system	poles at $\pm\omega_d-i\gamma$ automatically	with $\gamma>0$ the poles leave the axis on their own; no $i\epsilon$ needed, causality is built in

Common trapThe retarded function is analytic in the upper half-plane, not the lower

The single most common error is a sign slip in the convention. With the physicists' convention $e^{-i\omega t}$ used here, the retarded function $\tilde G_R(\omega)$ is analytic in the upper half-plane, and its poles sit below the axis — even though "retarded" intuitively feels like it should point downward. Switch to the engineers' convention $e^{+i\omega t}$ and every half-plane flips: the half-plane you close in, the sign of the causal $i\epsilon$ , and the direction of analyticity all reverse together. There is no convention-free answer to "which sign of $i\epsilon$ is retarded?" — the only safe check is operational: compute $G(t)$ and confirm it actually vanishes for $t<t'$ before you trust it. A second trap: do not confuse the retarded prescription (both poles the same side) with the Feynman one (poles split); they agree on the pole locations on the real axis but differ entirely in the response they produce.

Worked examples

1 · The impulse response of a damped oscillator (fully worked)

A mass on a spring obeys $\ddot x+2\gamma\dot x+\omega_0^2x=f(t)$ with $\omega_0=5\ \mathrm{rad/s}$ and $\gamma=1\ \mathrm{s^{-1}}$ . Find the retarded Green's function, its ring frequency and quality factor, and evaluate the response at a few times to confirm it is causal and decaying.

Step 1. Transform and locate the poles. With $L=\partial_t^2+2\gamma\partial_t+\omega_0^2$ the characteristic factor is $p(-i\omega)=-\omega^2-2i\gamma\omega+\omega_0^2$ , vanishing at $\omega=\pm\omega_d-i\gamma$ with $\omega_d=\sqrt{\omega_0^2-\gamma^2}$ . Numerically $\omega_d=\sqrt{25-1}=\sqrt{24}=\mathbf{4.90\ rad/s}$ . Both poles already lie below the real axis, so no $i\epsilon$ is needed — the damping supplies causality for free.

Step 2. Close downward for $t>0$ and collect the two residues exactly as in Derivation A; the extra factor $e^{-\gamma t}$ rides along from the imaginary part of each pole. The result is the damped causal ringdown

G_R(t)=\frac{e^{-\gamma t}\sin(\omega_d t)}{\omega_d}\,\Theta(t).

Step 3. Read off the physics. The ring frequency is $\omega_d=4.90\ \mathrm{rad/s}$ (barely shifted, since damping is light), and the quality factor is $Q=\omega_0/2\gamma=\mathbf{2.5}$ , so the amplitude falls by $e^{-\pi/(2Q)}\approx0.53$ per half-period — the visibly decaying ring of the first figure.

Step 4. Evaluate at sample times. At $t=0^-$ the response is exactly zero (causality). The first peak is at $t=\tfrac1{\omega_d}\arctan(\omega_d/\gamma)=0.28\ \mathrm{s}$ , where $G_R=\mathbf{0.151\ s}$ ; then $G_R(0.10\,\mathrm{s})=0.087\ \mathrm{s}$ , $G_R(0.50\,\mathrm{s})=0.079\ \mathrm{s}$ , with the first zero crossing at $t=\pi/\omega_d=0.64\ \mathrm{s}$ . Causal support, ring frequency 4.90 rad/s, Q = 2.5, all confirmed numerically.

Step 5. Check the undamped limit: $\gamma\to0$ gives $\omega_d\to\omega_0$ and $G_R\to\sin(\omega_0 t)\Theta(t)/\omega_0$ , recovering equation (5) — the $i\epsilon$ was only the ghost of this real damping.

2 · Retarded time and the field of a moving charge

An electron at position $\mathbf{x}'$ radiates; an observer sits $r=3.0\ \mathrm{m}$ away. Using the retarded Green's function of the wave operator, at what earlier instant must the charge's motion be evaluated to give the field "now", and how does this differ from the advanced answer?

Step 1. In Lorenz gauge Maxwell's equations read $\Box A^\mu=\mu_0 J^\mu$ , and the retarded Green's function of the d'Alembertian is $G_R(\mathbf{x},t)=\dfrac{\delta\!\left(t-|\mathbf{x}|/c\right)}{4\pi|\mathbf{x}|}$ , whose delta selects only the source on the past light cone.

Step 2. Solving $t-|\mathbf{x}-\mathbf{x}'|/c=0$ gives the retarded time $t_{\mathrm{ret}}=t-r/c$ , a delay $r/c=3.0/(2.998\times10^8)=\mathbf{1.0\times10^{-8}\ s}=10.0\ \mathrm{ns}$ . The field measured now was emitted $10.0$ ns ago — the origin of the Liénard–Wiechert potentials.

Step 3. The advanced function instead uses $\delta(t+|\mathbf{x}|/c)$ , giving $t_{\mathrm{adv}}=t+10.0\ \mathrm{ns}$ : dependence on where the charge will be. Nature picks the retarded root; the advanced partner underlies Wheeler–Feynman absorber theory. Retarded = 10.0 ns into the past, advanced = 10.0 ns into the future — same magnitude, opposite arrow.

3 · Why advanced and retarded differ by a free oscillation

Show explicitly that $G_R-G_A$ for the undamped oscillator solves the homogeneous equation, and interpret it.

Step 1. Subtract equations (5) and (6): $G_R(t)-G_A(t)=\frac{\sin(\omega_0 t)}{\omega_0}\big[\Theta(t)+\Theta(-t)\big]=\frac{\sin(\omega_0 t)}{\omega_0}$ for all $t$ , since the two step functions tile the whole line. The difference is a plain, source-free sine.

Step 2. Confirm it is homogeneous: $\ddot{}\big(\sin\omega_0 t/\omega_0\big)+\omega_0^2\big(\sin\omega_0 t/\omega_0\big)=-\omega_0\sin\omega_0 t/1+\omega_0\sin\omega_0 t=0$ . It carries no delta because it is smooth through $t=0$ .

Step 3. Interpret. The non-uniqueness of equation (2) is not abstract — it is this specific free oscillation. Choosing $G_R$ over $G_A$ is choosing to add exactly enough of it to move all the response to $t>0$ . The retarded–advanced ambiguity is one homogeneous mode, $\sin(\omega_0 t)/\omega_0$ , and nothing more.

Physics of the portrait

The retarded and advanced Green's functions are where a two-century argument about the arrow of time was finally made computational. The name "Green's function" honours a self-taught miller's son; the causal structure was assembled by a chain of physicists wrestling with radiation, dispersion, and antimatter.

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

retarded and advanced Green's functions

the causal inverse of a differential operator, with G_R supported on t ≥ t′ and G_A on t ≤ t′ · causality, analyticity, and the iε prescription

The Green's function is named for George Green, a Nottingham miller with almost no formal schooling, whose privately printed 1828 Essay introduced inverting a differential operator with a point-source solution — a work so obscure it was nearly lost until Lord Kelvin rescued it. The causal refinement came from radiation physics: Lorenz, Kirchhoff, and Sommerfeld found the wave equation admits both retarded and advanced potentials, and Sommerfeld's radiation condition first stated cleanly that outgoing waves are the physical choice. In 1926–27 Hendrik Kramers and Ralph Kronig made causality quantitative, showing the analyticity forced by "no response before the stimulus" links dispersion to absorption. The boldest use of the advanced solution came in 1945, when Wheeler and Feynman built an absorber theory from a time-symmetric half-advanced, half-retarded interaction. Feynman and Stueckelberg then folded both arrows into one object — the propagator with its $i\epsilon$ prescription — reinterpreting antiparticles as negative-energy solutions running backward in time. The modern synthesis: retarded, advanced, and Feynman are one function $\tilde G(\omega)$ read with different contours, and the contour is the choice of what causality means for the system at hand.

History

1828

— George Green's Essay introduces the Green's function to invert a differential operator

1867

— Ludvig Lorenz writes the retarded potentials of electromagnetism

1912

— Sommerfeld states the radiation condition selecting outgoing waves

1926

— Kramers and Kronig derive the dispersion relations from causality

1945

— Wheeler and Feynman propose the time-symmetric absorber theory of radiation

1948

— Feynman and Stueckelberg introduce the propagator with its iε prescription

Going further

Texts. Stone & Goldbart, Mathematics for Physics, Ch. 9 (Green Functions) · Arfken, Weber & Harris, Mathematical Methods for Physicists, 7th ed., Ch. 10 and Ch. 11 · Hassani, Mathematical Physics, 2nd ed., Ch. 21–22 (Green's functions in one dimension and time) · Byron & Fuller, Mathematics of Classical and Quantum Physics, Ch. 7 · Riley, Hobson & Bence, Mathematical Methods for Physics and Engineering, 3rd ed., Ch. 21 · Jackson, Classical Electrodynamics, 3rd ed., Ch. 6 (retarded potentials) · Peskin & Schroeder, An Introduction to Quantum Field Theory, Ch. 2 (Feynman propagator).

Provenance. The oscillator Green's function and its contour derivation follow Stone & Goldbart Ch. 9 and Riley–Hobson–Bence Ch. 21; a verifier should recompute the residue sum giving $\sin(\omega_0 t)/\omega_0$ and confirm vanishing for $t<0$ by upper-half-plane closure. The damped result $G_R=e^{-\gamma t}\sin(\omega_d t)/\omega_d$ with $\omega_d=\sqrt{\omega_0^2-\gamma^2}$ , and the numbers $\omega_d=4.90\ \mathrm{rad/s}$ , $Q=2.5$ , first peak $0.151$ at $t=0.28\ \mathrm{s}$ , are elementary and were checked directly. Sokhotski–Plemelj and Kramers–Kronig follow Arfken Ch. 11 and Hassani Ch. 22; the Feynman propagator pole structure follows Peskin & Schroeder Ch. 2; retarded potentials and the $r/c=10.0\ \mathrm{ns}$ delay at $3.0\ \mathrm{m}$ follow Jackson Ch. 6. All special-function values (e.g. $\int\epsilon/(x^2+\epsilon^2)\,\mathrm{d}x=\pi$ ) are standard.

Prerequisites. Fourier and Laplace transforms · Contour integration and the residue theorem · Dirac delta and distributions · Linear ODEs and homogeneous solutions. Leads to. WKB Approximation · Method of Steepest Descent and Saddle Points · Method of Stationary Phase.

Ask the tutor: Ask me to walk you through closing the contour for any operator you like — pick the equation and I will place the poles, choose the half-plane, and hand you back the retarded, advanced, and Feynman Green's functions side by side.

Why isn't the Green's function of a time operator unique when the spatial one usually is?

For a self-adjoint spatial operator, boundary conditions at both ends of an interval pin the Green's function down completely — nowhere for an extra homogeneous solution to hide. The time operator posed as an initial-value problem is not self-adjoint in that way, and its kernel (solutions of $L\,G=0$ ) has full dimension equal to the order of the equation. You can add any of those and still solve $L\,G=\delta$ . That leftover freedom is not a defect to remove with more mathematics — it is the physical input, where you tell the equation whether effects may precede causes. Retarded, advanced, and Feynman are three ways of spending it.

Is the advanced Green's function just a mathematical curiosity, or does it appear in real physics?

It is real, though nature never uses it alone. The famous case is the Wheeler–Feynman absorber theory (1945), which reproduces radiation reaction from a time-symmetric mix of half-advanced and half-retarded fields, the future absorber's advanced response conspiring to give a net causal result. Advanced functions also appear whenever you integrate boundary data backward from a final condition, and the pair $(G_R,G_A)$ builds the spectral function $G_R-G_A$ central to many-body theory. So the advanced function is legitimate and useful; what nature forbids is a lone advanced response, a bell that rings before it is struck.

What is the iε prescription really doing, physically?

It is an infinitesimal amount of damping. A perfectly undamped oscillator has poles sitting exactly on the real frequency axis, and the inverse Fourier transform genuinely does not exist until you say which side of each pole the contour passes. Adding $+i\epsilon$ is mathematically identical to giving the system a vanishing friction $\gamma=\epsilon$ , which nudges the poles just below the axis and makes the response causal. In the damped worked example the poles move off the axis by a finite amount $\gamma$ and no $i\epsilon$ is needed at all — the $i\epsilon$ is only the ghost of the damping you removed by idealizing to zero friction. Reading it this way demystifies the whole business: the prescription is not a trick, it is a limit of real physics.

Why is causality equivalent to analyticity in a half-plane?

Because the Fourier transform of a function supported only on $t\ge0$ automatically extends to a holomorphic function in a half-plane. If $G_R(t)=0$ for $t<0$ , then $\tilde G_R(\omega)=\int_0^\infty G_R(t)e^{i\omega t}\,\mathrm{d}t$ , and giving $\omega$ a positive imaginary part inserts a decaying factor $e^{-\mathrm{Im}(\omega)t}$ that only improves convergence — so the integral defines an analytic function throughout the upper half-plane, with no poles there. The converse holds too: analyticity in the upper half-plane, plus suitable decay, forces the time-domain function to vanish for $t<0$ . This two-way equivalence is why a purely kinematic statement (effects follow causes) becomes a rigid analytic constraint (Kramers–Kronig) that experiments can and do test.

How can a retarded Green's function have its poles below the axis yet be analytic above it?

No contradiction — those are statements about two different regions. "Analytic in the upper half-plane" means holomorphic where $\mathrm{Im}\,\omega>0$ ; the poles live where $\mathrm{Im}\,\omega<0$ . A function can have all its poles in one half-plane and be perfectly smooth in the other, and that is exactly $\tilde G_R$ . The physical reading is neat: the poles below the axis are the decaying natural modes (each a negative imaginary frequency, ringing down), and the emptiness above is causality. Confusion comes from mixing conventions — with $e^{-i\omega t}$ the retarded poles are below and analyticity is above; the engineers' $e^{+i\omega t}$ flips both together.

Does the same machinery really carry over unchanged to quantum field theory?

The bookkeeping is identical; only the setting grows. In the oscillator you have one pair of poles at $\pm\omega_0$ and a choice of how to skirt them; in field theory each momentum mode is an oscillator, and the Feynman propagator $i/(p^2-m^2+i\epsilon)$ applies the same $\pm i\epsilon$ rule to the pole at $p_0=\pm\sqrt{\mathbf{p}^2+m^2}$ for every mode at once. The one genuinely new ingredient is the choice of prescription: relativity plus antiparticles forces the Feynman (time-ordered) contour rather than the strictly retarded one, because positive-energy particles must go forward and negative-energy states (antiparticles) backward. But the contour-closing, residue-collecting, half-plane logic of Derivation A is exactly what you use to do the $p_0$ integral in any loop calculation. Learn it once here and you have learned it for QFT.