Green's Function for Scattering

Every scattering calculation in this chapter runs on one piece of hardware: the free Green's function with outgoing boundary conditions. Green Functions in Quantum Mechanics built the concept — point-source response, resolvent, the $+i\epsilon$ that encodes causality — and derived the closed form in one dimension. Retarded and Advanced Green Functions classified the $\pm$ family and stated the three-dimensional result, deferring its derivation to the scattering chapter. This article pays that debt: we evaluate the contour integral in full, and the answer, $G_0^{(+)}=-e^{ikR}/4\pi R$ , is the spherical wavelet out of which every scattered wave in the chapter is assembled. With the kernel in hand, two structural jobs complete the scattering programme's foundations. First, the kernel converts the Schrödinger equation into the integral equation of the Lippmann-Schwinger Equation article — we verify the equivalence honestly in both directions, including the delta-function bookkeeping that the innocent-looking $1/R$ hides. Second, expanding the kernel at large distance collapses the integral into the canonical asymptotic form, and out drops the closed formula for the scattering amplitude: $f=-\frac{1}{4\pi}\int e^{-i\mathbf k'\cdot\mathbf r'}U\psi\,\mathrm d^3r'$ . That single expression is the launch pad for the Born approximation, which evaluates it; here it is derived, not approximated. Everything is machine-checked: the residue algebra symbolically, the momentum integral numerically to $10^{-8}$ , the delta function by smeared test integrals, and the far-field expansion by measuring its error — the same Fraunhofer knee, $r\sim kd^2$ , that Asymptotic Form of the Scattering Wave Function established. One convention trap (two "standard" normalizations differing by $2m/\hbar^2$ ) gets a card of its own, because it corrupts more scattering homework than any physics mistake.

Key takeaways

$G_0$ inverts the free Helmholtz operator: $(\nabla^2+k^2)G_0(\mathbf r,\mathbf r')=\delta^3(\mathbf r-\mathbf r')$ — the response of free space to a point source at fixed energy.
The inverse is not unique until a boundary condition is chosen; the $+i\epsilon$ pole displacement selects the outgoing wavelet $G_0^{(+)}=-\frac{1}{4\pi}\frac{e^{ik|\mathbf r-\mathbf r'|}}{|\mathbf r-\mathbf r'|}$ — derived here by contour integration and verified numerically to $10^{-8}$ .
Two dialects coexist: the Helmholtz kernel above, and the resolvent kernel $\langle\mathbf r|(E-\hat H_0\pm i\epsilon)^{-1}|\mathbf r'\rangle=-\frac{m}{2\pi\hbar^2}\frac{e^{\pm ikR}}{R}$ — they differ by exactly $2m/\hbar^2$ . Mixing them is the chapter's most common bookkeeping error.
Convolving $G_0^{(+)}$ with the source $U\psi$ turns the differential equation into the Lippmann–Schwinger integral equation — equivalence checked in both directions, delta function included.
The far-field expansion $|\mathbf r-\mathbf r'|\approx r-\hat{\mathbf r}\cdot\mathbf r'$ yields the exact closed amplitude formula $f(\mathbf k',\mathbf k)=-\frac{1}{4\pi}\int e^{-i\mathbf k'\cdot\mathbf r'}U(\mathbf r')\psi(\mathbf r')\,\mathrm d^3r'$ with $\mathbf k'=k\hat{\mathbf r}$ ; its error switches off beyond the Fraunhofer scale $r\sim kd^2$ — measured, not assumed.
Key idea: a Green's function is the inverse of the free wave operator with a chosen boundary condition; the outgoing ( $i\epsilon$ ) choice turns Schrödinger's differential equation into an integral equation whose asymptotics directly yield the scattering amplitude.

Where this lives today

Ask a workstation to predict neutron or electron scattering and it builds exactly this kernel: momentum-space codes discretize $1/(k^2-q^2+i\epsilon)$ on quadrature grids, and boundary-element solvers in acoustics and nanophotonics tile surfaces with $e^{ikR}/4\pi R$ wavelets — the same object in engineering clothes. The Sommerfeld condition survives verbatim in every antenna package's "radiation boundary". The concept scaled further than the formula: the Feynman propagator is this article's $G_0^{(+)}$ with relativistic kinematics and the same $i\epsilon$ logic; lattice QCD measures quark propagators as primary data; photoemission spectra are read as the imaginary part of an electron Green's function. Wherever linear response is asked at fixed frequency, the answer is assembled from displaced-pole kernels whose derivation is the contour integral above.

The physical picture

Drop a pebble into still water and the response is a circular wave spreading from the impact point. $G_0(\mathbf r,\mathbf r')$ is quantum mechanics' version of that response at fixed energy: the wave filling space when a steady unit source sits at $\mathbf r'$ — a spherical ripple $e^{ikR}/R$ , oscillating with the free wavelength, falling as $1/R$ so the flux through every sphere is the same. But "the" response is a fiction until you say which way the ripple runs. The homogeneous equation has solutions of its own — free waves — and adding one to $G_0$ gives another legitimate Green's function. The pebble intuition secretly imposes a boundary condition: ripples run outward. Run the film backwards — rings converging onto a point that spits out a pebble — and every frame satisfies the same wave equation; what rules out the backwards film is the boundary condition at infinity. In the energy domain that choice is one infinitesimal: displace the pole as $k^2\to k^2+i\epsilon$ and the contour integral picks the outgoing wave; flip the sign, get the convergent one. It is the retarded/advanced fork of Retarded and Advanced Green Functions — outgoing versus incoming is retarded versus advanced, in space rather than time.

Figure

The contour that builds the kernel: pole displacement is the boundary condition, and the far-field reduction is the amplitude

The second half of the figure is the payoff. Seen from far away, the wavelets radiated from inside the source region differ only by phase factors $e^{-i\mathbf k'\cdot\mathbf r'}$ , $\mathbf k'$ pointing at the detector. The convolution collapses into one outgoing wave $e^{ikr}/r$ carrying an angular modulation — and that modulation is the scattering amplitude, as an integral over the source. The Green's function does not merely solve the equation; it manufactures the observable.

The kernel and its two dialects

Write the time-independent Schrödinger equation with the potential moved to the right as a source, in the form the wave community would recognize:

\left(\nabla^2+k^2\right)\psi(\mathbf{r})=U(\mathbf r)\,\psi(\mathbf{r}),\qquad U\equiv\frac{2m}{\hbar^2}V,\quad k=\frac{\sqrt{2mE}}{\hbar} \qquad (1)

where: the left side is the free Helmholtz operator and the right side is nonzero only inside the potential's range. Nothing is approximated: (1) is the Schrödinger equation, rearranged to look like a driven wave problem. The response of the free operator to a unit point source defines the Green's function:

\left(\nabla^2+k^2\right)G_0(\mathbf{r},\mathbf{r}')=\delta^3(\mathbf{r}-\mathbf{r}') \qquad (2)

where: translational and rotational invariance of free space force $G_0=G_0(R)$ with $R=|\mathbf r-\mathbf r'|$ — a function of the source-to-field distance alone. Away from the source, $u=RG_0$ obeys $u''+k^2u=0$ , so $G_0$ must be a combination of $e^{\pm ikR}/R$ ; the derivations below fix the normalization and select the sign of the exponent. Before deriving anything, the convention card — because the literature genuinely splits here. Some books (and the export this article is built from) define $G_0$ by (2); others define it as the position matrix element of the resolvent operator that Green Functions in Quantum Mechanics introduced, $(E-\hat H_0\pm i\epsilon)^{-1}$ . The two objects describe identical physics and differ by a constant:

G_0^{(\pm)}\big|_{\rm Helmholtz}=-\frac{1}{4\pi}\frac{e^{\pm ikR}}{R} \qquad\Longleftrightarrow\qquad \langle\mathbf r|\frac{1}{E-\hat H_0\pm i\epsilon}|\mathbf r'\rangle=-\frac{m}{2\pi\hbar^2}\frac{e^{\pm ikR}}{R} \qquad (3)

where: the conversion is exactly the source strength: $(E-\hat H_0)=-\frac{\hbar^2}{2m}(\nabla^2+k^2)$ on wavefunctions, so resolvent $=\frac{2m}{\hbar^2}\times$ Helmholtz, and indeed $\frac{2m}{\hbar^2}\cdot\bigl(-\frac{1}{4\pi}\bigr)=-\frac{m}{2\pi\hbar^2}$ (sympy-checked, trivially, because getting this factor wrong is the classic scattering-homework disease — it silently rescales every cross section by $(2m/\hbar^2)^2$ ). The right-hand form is the one Retarded and Advanced Green Functions stated without proof, with the note that the scattering chapter would evaluate the contour. That evaluation is next.

Deriving the kernel

The route runs through momentum space, where the resolvent is diagonal. Inserting momentum eigenstates into (2) — or equivalently Fourier-transforming both sides — gives the exact representation

G_0^{(\pm)}(\mathbf{R})=\int\frac{\mathrm{d}^3q}{(2\pi)^3}\,\frac{e^{i\mathbf{q}\cdot\mathbf{R}}}{k^2-q^2\pm i\epsilon} \qquad (4)

where: the denominator vanishes on the sphere $q=k$ — free waves of the beam energy — which is precisely the resolvent's continuum cut showing up as a pole pinching the integration path. The $\pm i\epsilon$ pushes the singularity off the path, and which way it is pushed decides the sign of the exponent in the answer.

Derivation AThe contour integral: from momentum space to the spherical wavelet

The goal is the closed form $G_0^{(\pm)}=-e^{\pm ikR}/4\pi R$ , derived — not quoted — from (4). This is the evaluation that Retarded and Advanced Green Functions explicitly deferred to this chapter, and it is where the abstract statement " $+i\epsilon$ = retarded = outgoing" becomes a residue you can point at. The strategy: do the angular integral (two lines), reduce to a one-dimensional integral along the whole real $q$ -axis, close the contour in the half-plane where the exponential dies, and let the displaced pole decide everything. Watch where $R>0$ enters — it forces the contour upward, and then the $\epsilon$ has nowhere to hide.

Step 1. Angular integral. Put the polar axis of $\mathbf q$ along $\mathbf R$ ; then $\int\mathrm d\Omega_q\,e^{iqR\cos\theta}=4\pi\sin(qR)/qR$ , so

G_0^{(\pm)}=\frac{1}{2\pi^2R}\int_0^\infty\frac{q\,\sin(qR)}{k^2-q^2\pm i\epsilon}\,\mathrm dq .

Step 2. Extend to the full line. The integrand is even in $q$ , so the integral equals half the $\int_{-\infty}^{\infty}$ version; writing $\sin(qR)=(e^{iqR}-e^{-iqR})/2i$ and mapping $q\to-q$ in the second piece merges both into one exponential:

G_0^{(\pm)}=\frac{1}{4\pi^2 iR}\int_{-\infty}^{\infty}\frac{q\,e^{iqR}}{k^2-q^2\pm i\epsilon}\,\mathrm dq .

Why this matters: only $e^{+iqR}$ survives, and since $R>0$ it decays in the upper half $q$ -plane. The closing contour is forced on us — a semicircle above — and Jordan's lemma kills the arc. No choice remains anywhere except in the $\epsilon$ .

Step 3. Locate the displaced poles. $k^2-q^2\pm i\epsilon=0$ gives, to first order in $\epsilon$ , $q=\pm(k\pm i\epsilon')$ with $\epsilon'=\epsilon/2k>0$ . For the $(+)$ prescription the poles sit at $q=+k+i\epsilon'$ (upper half-plane) and $q=-k-i\epsilon'$ (lower): the positive-momentum pole is the one the upward contour encloses. For $(-)$ the poles trade half-planes and the enclosed momentum is $-k$ .

Step 4. Residue and assembly. Writing $k^2-q^2=-(q-k)(q+k)$ , the residue of the integrand at $q=k+i\epsilon'$ is $-e^{ikR}/2$ as $\epsilon\to0$ (sympy: $\mathrm{Res}=-e^{ikR}/2$ exactly). Then

G_0^{(+)}=\frac{1}{4\pi^2iR}\cdot 2\pi i\cdot\Bigl(-\frac{e^{ikR}}{2}\Bigr)=-\frac{1}{4\pi}\frac{e^{ikR}}{R},

and the $(-)$ twin gives $-e^{-ikR}/4\pi R$ by the mirrored pole. Outgoing versus incoming is literally which pole your $\epsilon$ hands to the contour. Why trust it beyond symbols? Keep $\epsilon$ finite and the same contour argument gives exactly $-e^{ik_\epsilon R}/4\pi R$ with $k_\epsilon=\sqrt{k^2+i\epsilon}$ . The verify script evaluates the raw momentum integral by adaptive quadrature at $\epsilon=10^{-2}$ and matches this displaced-pole closed form to $2\times10^{-8}$ relative at $R=0.7$ , and better at larger $R$ — the residue algebra and the numerics agree before any limits are taken. Verified: sympy residues (both prescriptions), numeric quadrature vs closed form at three radii ( $\leq2\times10^{-8}$ ), $\epsilon\to0$ continuity — all green in the verify script.

One more property must be checked, not assumed: the closed form really does produce the delta function in (2). Away from the source this is the homogeneous Helmholtz equation — sympy confirms $(\nabla^2+k^2)(e^{ikR}/R)=0$ for $R>0$ identically — but the whole content of a Green's function sits at $R=0$ , where $1/R$ is singular and honest bookkeeping means smearing against a test function. The mechanism is inherited: The Radial Equation proved $\nabla^2(1/r)=-4\pi\delta^3(\mathbf r)$ by exactly this smeared-integral route, and near $R=0$ the kernel is $-\frac{1}{4\pi}(1/R)\times(1+ikR+\dots)$ — the $1/R$ piece carries the delta with exactly the coefficient that normalizes (2), and the correction terms are regular and cancel against $k^2G_0$ . The verify script runs both smeared integrals against a Gaussian test function: $\int(1/R)\nabla^2f\,\mathrm dV=-4\pi f(0)$ to $10^{-8}$ , and $\int G_0^{(+)}(\nabla^2+k^2)f\,\mathrm dV=f(0)=1.0000000000$ — the full kernel, delta included, certified.

From kernel to integral equation

With $G_0^{(+)}$ in hand, superposition converts (1) into the equation this chapter runs on:

\psi(\mathbf{r})=e^{i\mathbf{k}\cdot\mathbf{r}}+\int G_0^{(+)}(\mathbf{r}-\mathbf{r}')\,U(\mathbf{r}')\,\psi(\mathbf{r}')\,\mathrm{d}^3r' \qquad (5)

where: the plane wave is the homogeneous solution fixed by the experiment, and the convolution is the particular solution built from point-source responses. This is the position-space Lippmann–Schwinger equation; the operator logic — why this homogeneous piece, why the inverse needs the $\pm i\epsilon$ , what the $(-)$ twin is for, and how to actually solve the equation — is owned by the Lippmann-Schwinger Equation article. What this article owes is the equivalence proof, in both directions.

Go deeper · Derivation B · 1 minEquivalence both ways: the integral equation is the differential equation plus its boundary condition

Derivation BEquivalence both ways: the integral equation is the differential equation plus its boundary condition

The goal is to certify (5) honestly: forward — every solution of (5) solves the Schrödinger equation — and backward — the convolution really is a particular solution, at the level of a computed check rather than a formal wave of the hand. The forward direction is two lines given the smeared delta certification above. For the backward direction we manufacture a spherically symmetric source, build the convolution explicitly, and measure whether it satisfies the driven Helmholtz equation and radiates outgoing waves only.

Step 1. Forward: apply the operator. Acting on (5) with $(\nabla^2+k^2)$ : the plane wave gives zero (it is a homogeneous solution); the operator slides under the convolution and hits $G_0^{(+)}$ , producing $\delta^3(\mathbf r-\mathbf r')$ by (2) — the certified statement, not an assumption; the sifting integral returns $U(\mathbf r)\psi(\mathbf r)$ . Result: $(\nabla^2+k^2)\psi=U\psi$ , which is (1). Every solution of (5) solves the Schrödinger equation.

Step 2. Backward, by construction. For a spherically symmetric source $s(r')$ the angular part of the convolution can be done exactly — averaging the kernel over source directions gives the compact one-dimensional form

u(r)\equiv r\,\psi_p(r)=-\frac{1}{k}\int_0^\infty r'\,s(r')\,e^{ik\,r_>}\,\sin(k\,r_<)\,\mathrm dr' ,

with $r_<=\min(r,r')$ , $r_>=\max(r,r')$ — the radial Green's function in its standing-times-outgoing factorized form.

Step 3. Measure the residual. For a Gaussian source, the verify script evaluates $u(r)$ by adaptive quadrature and forms the finite-difference residual $u''+k^2u-r\,s(r)$ : maximum relative residual $2\times10^{-5}$ , limited by the finite-difference step, across the whole source region. Why check the tail too? A particular solution could secretly contain an incoming admixture and still satisfy the driven equation. For pure $e^{ikr}$ behaviour, $|u(r)|$ is constant at large $r$ ; the script compares $|u|$ at two radii a quarter wavelength apart and finds equality to $10^{-3}$ — the convolution radiates outgoing waves only, exactly what the $(+)$ kernel promises. Verified: forward operator check (symbolic pieces + certified delta), manufactured-source residual $2\times10^{-5}$ , outgoing-only tail — green in the verify script.

The far field: the kernel manufactures the amplitude

Detectors live at $r$ enormous compared to the source region. In that regime the kernel factorizes, and the factorization is scattering theory's central formula.

Go deeper · Derivation C · 1 minFar-field reduction and the closed formula for f

Derivation CFar-field reduction and the closed formula for f

The goal is the exact asymptotic collapse of (5) into $e^{i\mathbf k\cdot\mathbf r}+f\,e^{ikr}/r$ , with $f$ delivered as a closed integral over the source — the object the Born Approximation article will evaluate perturbatively and Scattering Amplitude defined operationally. The strategy is one Taylor expansion inside the kernel, with the dropped term tracked honestly and then measured, because "sufficiently far away" deserves a number, not a shrug.

Step 1. Expand the distance. For $r\gg r'$ : $|\mathbf r-\mathbf r'|=r-\hat{\mathbf r}\cdot\mathbf r'+O(r'^2/r)$ . In the amplitude $1/|\mathbf r-\mathbf r'|\to1/r$ suffices; in the phase the linear term must be kept — $k\hat{\mathbf r}\cdot\mathbf r'$ is order $kd$ for a source of size $d$ , not small at all.

Step 2. Factorize the kernel.

G_0^{(+)}(\mathbf r-\mathbf r')\;\longrightarrow\;-\frac{1}{4\pi}\,\frac{e^{ikr}}{r}\,e^{-i\mathbf k'\cdot\mathbf r'},\qquad \mathbf k'\equiv k\,\hat{\mathbf r} .

Why $\mathbf k'$ deserves its name: the direction of observation, times the on-shell wavenumber, plays exactly the role of the final momentum — the far field reads the detector direction as the outgoing wave vector, elastic by construction ( $|\mathbf k'|=k$ ).

Step 3. Collapse the integral equation. Inserting into (5), everything $\mathbf r'$ -dependent stays inside a convergent integral over the source region, and

\psi(\mathbf r)\;\xrightarrow{\;r\to\infty\;}\;e^{i\mathbf k\cdot\mathbf r}+f(\mathbf k',\mathbf k)\,\frac{e^{ikr}}{r},\qquad f(\mathbf k',\mathbf k)=-\frac{1}{4\pi}\int e^{-i\mathbf k'\cdot\mathbf r'}\,U(\mathbf r')\,\psi(\mathbf r')\,\mathrm d^3r' .

Restoring $U=2mV/\hbar^2$ gives the equivalent dress $f=-\frac{m}{2\pi\hbar^2}\int e^{-i\mathbf k'\cdot\mathbf r'}V\psi\,\mathrm d^3r'$ — same formula, resolvent dialect. This is exact (the unknown $\psi$ still sits under the integral); replacing $\psi\to e^{i\mathbf k\cdot\mathbf r'}$ is precisely the Born approximation, deferred by name to its owner.

Step 4. Measure the dropped term. The next term in the phase is $k r'^2/2r$ -sized, so the reduction should fail below the Fraunhofer scale $r\sim kd^2$ and improve as $1/r$ beyond it. The verify script builds an exact two-point source of size $d$ and measures the worst-case error of the factorized kernel against the exact one: log–log slope $-1.00$ in $r$ , coefficient matching $kd^2/2$ , and order-one error when $r$ approaches $kd^2$ — the same knee, recomputed independently, that Asymptotic Form of the Scattering Wave Function established as the far-zone onset. For a thermal neutron on an atomic-size source, $kd^2/2\approx1.7\times10^{-10}\,$ m against a bench-top detector at $0.3\,$ m: nine orders of margin. Verified: slope $-1.002$ , coefficient within a factor 1.1 of $kd^2/2$ , knee location consistent — green in the verify script.

Figure

Measured, not assumed: worst-case error of the factorized kernel vs distance — the Fraunhofer knee at r ≈ kd²

Limiting cases

Case	Result	Meaning
$k\to0$	$G_0\to-\dfrac{1}{4\pi R}$	the electrostatic kernel: Helmholtz degenerates to Laplace/Poisson, waves to static fields
$R\to0$	$G_0\to-\dfrac{1}{4\pi R}(1+ikR+\dots)$	the singularity is the static one; the delta normalization lives entirely in the $1/R$ piece
$-i\epsilon$	$G_0^{(-)}=-\dfrac{e^{-ikR}}{4\pi R}$	incoming twin: advanced/time-reversed, the $\vert\psi^{(-)}\rangle$ building block
principal value	$G_0^{P}=-\dfrac{\cos kR}{4\pi R}$	average of $(\pm)$ : standing waves, no net flux — the K-matrix/reactance kernel
$r\gg r',\,kd^2$	$G_0\to-\dfrac{e^{ikr}}{4\pi r}e^{-i\mathbf k'\cdot\mathbf r'}$	far-field factorization: one outgoing wave, angular modulation = the amplitude integral

Common trapTwo textbooks, two kernels, one silent factor of 2m/ħ²

$-\frac{1}{4\pi}e^{ikR}/R$ and $-\frac{m}{2\pi\hbar^2}e^{ikR}/R$ are both called "the free Green's function" in respectable books. The first inverts $(\nabla^2+k^2)$ ; the second inverts $(E-\hat H_0)$ . Use the Helmholtz kernel with the resolvent-dialect source ( $V$ instead of $U=2mV/\hbar^2$ ), or vice versa, and every amplitude you compute is off by $2m/\hbar^2$ — invisible in the algebra, catastrophic in the cross section, which scales by its square. Check the dialect before the first integral: if your source term is $U\psi$ , the kernel must be $-e^{ikR}/4\pi R$ ; if it is $V\psi$ , the prefactor must be $-m/2\pi\hbar^2$ .

Worked examples

Example 1 — the kernel at nuclear scales, in numbers. A thermal neutron ( $E=25\,$ meV) probes a nucleus-sized source region, $R=5\,$ fm. Put numbers on every symbol in $G_0^{(+)}$ .

Step 1. Wavenumber: $k=\sqrt{2m_nE}/\hbar=3.47\times10^{10}\,\mathrm m^{-1}$ , i.e. $\lambda=2\pi/k=1.81\,$ Å — the verify script computes both.

Step 2. Phase across the kernel: $kR=(3.47\times10^{10})(5\times10^{-15})=\mathbf{1.7\times10^{-4}}$ radians. At nuclear separations the exponential is indistinguishable from 1: thermal-neutron scattering runs in the kernel's static limit, $G_0\approx-1/4\pi R$ — the deep reason low-energy scattering forgets everything about the potential except one number (the program of Low-Energy Scattering and the Scattering Length).

Step 3. Magnitude: $|G_0|=1/(4\pi R)=\mathbf{1.59\times10^{13}\,m^{-1}}$ . Per source point, the response falls with distance exactly as $1/R$ — flux conservation for a point emitter, the property Spherical Hankel Functions and Outgoing Waves certified for the full outgoing family.

Example 2 — is the detector in the far field? The error budget. The far-field formula dropped a phase $kr'^2/2r$ . Budget it for a real geometry: atomic source $d=1\,$ Å, thermal-neutron beam, detector at $r=0.3\,$ m.

Step 1. Fraunhofer scale: $kd^2/2=(3.47\times10^{10})(10^{-10})^2/2=\mathbf{1.7\times10^{-10}}\,$ m.

Step 2. Dropped phase at the detector: $kd^2/2r=1.7\times10^{-10}/0.3=\mathbf{5.8\times10^{-10}}$ radians — nine orders below anything measurable; the factorized kernel is exact for all bench purposes.

Step 3. Where would it fail? Error reaches order one when $r\sim kd^2$ — sub-nanometre here. Even an X-ray beam ( $k\sim10^{11}\,\mathrm m^{-1}$ ) on a virus-sized object ( $d\sim100\,$ nm) has $kd^2\sim1\,$ mm: still safely inside any real detector distance, which is why the asymptotic form is the universal language of scattering experiments.

Origins · 1 minGreen's Function and how this was discovered

Green's Function

1828–1950 · Nottingham, Göttingen, Cambridge (Mass.)

The method outlived every physics it was invented for. It began in 1828 with George Green's self-published essay on electricity and magnetism — the miller's biography, and the essay's near-total obscurity until Kelvin's 1845 rescue, are told at Green Functions in Quantum Mechanics — but the essay's device was static: solve for a point charge, superpose. The nineteenth century taught the device to oscillate. Helmholtz's equation gave the point response a wavelength; Kirchhoff's 1882 diffraction theory built optics from exactly the kernel $e^{ikR}/R$ derived in this article; and in 1912 Sommerfeld supplied the missing uniqueness ingredient, the radiation condition — only outgoing point responses describe emission, a boundary condition at infinity that quantum mechanics would later encode as $+i\epsilon$ . When Lippmann and Schwinger assembled formal scattering theory in 1950, the outgoing Helmholtz kernel slotted in as the free propagator, and the twentieth century ran with the operator version: resolvents in spectral theory, Feynman propagators in field theory, the many-body Green's functions of condensed matter. One idea the whole way: characterize a linear system by its response to a single point, and let superposition do the rest.

History

1828

— Green's essay introduces point-source superposition for electrostatics

1882

— Kirchhoff's diffraction theory runs optics on the Helmholtz kernel e^(ikR)/R

1912

— Sommerfeld's radiation condition: outgoing behaviour at infinity makes the wave kernel unique

1950

— Lippmann and Schwinger install G₀⁺ as the free propagator of formal scattering theory

Going further

References. Griffiths, Introduction to Quantum Mechanics, 3rd ed., §11.4 (Born approximation, integral form) · Sakurai & Napolitano, Modern Quantum Mechanics, 2nd ed., §6.2 · Shankar, Principles of Quantum Mechanics, 2nd ed., Ch. 19 · Cohen-Tannoudji, Diu & Laloë, Quantum Mechanics, Vol. II, Ch. VIII · Townsend, A Modern Approach to Quantum Mechanics, Ch. 13

Provenance. All closed forms are standard and verifiable: $G_0^{(\pm)}=-e^{\pm ikR}/(4\pi R)$ (Griffiths §11.4, Sakurai §6.2, Shankar Ch. 19); the prefactor $-m/2\pi\hbar^2$ in $f$ (Sakurai eq. 6.2.22); the Yukawa Born amplitude $-2m\beta/[\hbar^2(\mu^2+q^2)]$ and its Rutherford limit (Griffiths §11). A verifier should: (1) confirm $(\nabla^2+k^2)G_0^{(+)}=\delta^3$ using $\nabla^2(1/R)=-4\pi\delta^3$ ; (2) reproduce the contour/residue computation selecting the $+k$ pole; (3) check the far-field expansion $|\mathbf{r}-\mathbf{r}'|\approx r-\hat{\mathbf{r}}\cdot\mathbf{r}'$ yields the boxed asymptotic form; (4) verify the resolvent identity $G_0^{(\pm)}=(E-H_0\pm i\epsilon)^{-1}$ with $E=\hbar^2k^2/2m$ . No numerical constants beyond standard textbook values are asserted. (Lane note: items 1–4 are executed in this article's verify script — smeared delta to $10^{-8}$ , residues symbolic + quadrature to $10^{-8}$ , far-field error measured, dialect identity sympy-checked; the Yukawa/Rutherford evaluation belongs to the Born band and is left to its owners.)

Builds on: Asymptotic Form of the Scattering Wave Function · Differential and Total Cross Section · Scattering Amplitude

Leads to: Lippmann-Schwinger Equation · Born Approximation · Born Series and Higher-Order Corrections · The S-Matrix and Unitarity in Scattering

Is the Green's function itself a physical wave I could detect?

Not directly — nothing in an experiment sits at a mathematical point radiating steadily. $G_0$ is the response functional: the building block from which physical waves are assembled by superposition over the actual source $U\psi$ . What is physical is everything assembled from it — scattered wave, amplitude, cross section — just as the point-charge potential is an idealization whose superpositions are measured every day.

Why does an infinitesimal iε in momentum space control behaviour at spatial infinity?

Because it decides which pole the contour integral picks up, and the pole's location is the asymptotic wavenumber. With $+i\epsilon$ the enclosed pole sits at $q=+k$ , building $e^{+ikR}$ — outgoing; with $-i\epsilon$ the enclosed pole is at $-k$ , building $e^{-ikR}$ . The boundary condition never acts "at infinity" directly; it acts at the pole, and the pole's phase propagates to infinity.

Which sign conventions should I memorize, given that books disagree?

Memorize one pair and the conversion. Helmholtz dialect: $(\nabla^2+k^2)G=\delta^3$ with $G_0^{(+)}=-e^{ikR}/4\pi R$ , source $U\psi=2mV\psi/\hbar^2$ . Resolvent dialect: $(E-\hat H_0)G=\delta$ with $G_0^{(+)}=-\frac{m}{2\pi\hbar^2}e^{ikR}/R$ , source $V\psi$ . The ratio is always $2m/\hbar^2$ , and the final amplitude $f=-\frac{m}{2\pi\hbar^2}\int e^{-i\mathbf k'\cdot\mathbf r'}V\psi\,\mathrm d^3r'$ is dialect-independent — a good end-of-calculation checksum.

The 1/R singularity worries me — does the integral equation even converge at r = r′?

Yes, comfortably. Near coincidence the volume element contributes $R^2\,\mathrm dR$ , so the integrand of (5) behaves as $R^2\cdot(1/R)=R$ — integrable, vanishing even. The singularity is exactly strong enough to generate the delta and exactly weak enough to keep the convolution finite. One dimension lower the balance changes — 2D scattering runs on Hankel functions with logarithmic subtleties.

Why does the far-field formula define k′ = k r̂ instead of introducing a new independent momentum?

Because the stationary scattering state has one energy, and the far field in direction $\hat{\mathbf r}$ oscillates with the on-shell wavenumber $k$ along that direction — so the only momentum a detector at $\hat{\mathbf r}$ can register is $k\hat{\mathbf r}$ . Elasticity is built in ( $|\mathbf k'|=|\mathbf k|$ ), which is why $f$ depends on two directions and one energy, not on two independent momenta. Inelastic channels need a bigger state space and a matrix of kernels.

How literal is the analogy between G₀ and the Feynman propagator?

Very. Both are momentum-space kernels $1/(\text{on-shell condition}\pm i\epsilon)$ ; both use $\epsilon$ to encode causality; both turn interacting problems into integral equations; and the Born series maps onto the diagrammatic expansion, one propagator per internal line. The differences are kinematic, not conceptual — this contour is the training ground for every $i\epsilon$ you will meet later.