Scalar and Vector Diffraction Theory

Geometrical optics says a wall with a hole casts a sharp-edged beam. Waves disagree — light bends into the shadow, shadows carry fringes, and a bright spot sits dead center behind an opaque disk (the 1818 discovery that converted the doubters). Diffraction theory predicts all of it, and this article builds the machinery the way electrodynamics demands: from boundary-value principles, failures on display. The engine is a single identity. Green's theorem plus the outgoing Helmholtz Green's function turns "the field on a surface" into "the field anywhere inside" — Huygens' picture of wavelets, made exact. Every named formulation is one identity with one choice inside it: Kirchhoff guesses both the field and its slope on the screen (and contradicts himself — a theorem this article proves), while Rayleigh and Sommerfeld repair the contradiction with an image trick borrowed from electrostatics: the Method of Images, replayed for waves. Downstream, the integral collapses through the Fresnel and Fraunhofer approximations until the far field becomes a Fourier transform of the aperture — the fact that unifies telescope resolution, antenna patterns, and X-ray crystallography. Then the honest part. Scalar theory treats light as one number per point; light is a vector field with boundary conditions. For apertures comparable to the wavelength the scalar answer is not slightly off — it is wrong by orders of magnitude. Bethe's exact small-hole solution, derived here end-to-end from effective dipoles, gives transmission $\propto (ka)^4$ : a hole ten times smaller than the wavelength passes ten thousand times less light than geometry promises. That fourth power runs your microwave oven's window.

Key takeaways

One identity rules everything: $U(\mathbf{r}) = \oint_S[G\,\partial_n U - U\,\partial_n G]\,da$ — Green's theorem with the outgoing Green's function $G = e^{ikR}/4\pi R$ . Every diffraction formula is a choice of $G$ and of which boundary data to trust.
Kirchhoff's choice (impose both $U$ and $\partial_n U$ on the screen) is mathematically inconsistent — a Helmholtz field vanishing with its normal derivative on a finite patch vanishes identically — yet works excellently for apertures many wavelengths wide.
The Rayleigh–Sommerfeld repair builds image Green's functions on the screen plane (Method of Images, waves edition): $G_-$ vanishes there (needs only $U$ , obliquity $\cos\theta$ ), $G_+$ 's normal derivative vanishes (needs only $\partial_n U$ ). Kirchhoff is exactly their average.
The working ladder: Fresnel (quadratic phase, near field, Fresnel number $N_F = a^2/\lambda z$ ) → Fraunhofer ( $N_F \ll 1$ ): the far field is the Fourier transform of the aperture — slit $\mathrm{sinc}^2$ , circular aperture Airy disk $\theta = 1.22\,\lambda/D$ .
Scalar theory ignores polarization and edge currents; for $a \lesssim \lambda$ it fails badly. The vector fix (Smythe formula) needs only the tangential aperture field.
Bethe's small hole, derived end-to-end: a circular hole with $ka \ll 1$ transmits as an effective magnetic dipole (plus a weaker electric one), giving $T = \frac{64}{27\pi^2}(ka)^4$ — four powers of smallness below the geometric guess.
Key idea: diffraction is a boundary-value problem — one Green's-function identity, with the taxonomy set by which Green's function and which boundary data, and a vector correction that takes over when holes shrink to the wavelength.

The core insight

Every diffraction formula is the single Green's-theorem identity $U(\mathbf{r}) = \oint_S[G\,\partial_n U - U\,\partial_n G]\,da$ ; the whole taxonomy — Kirchhoff vs Rayleigh–Sommerfeld, scalar vs vector — is just the choice of Green's function and of which boundary data you trust.

Where this lives today

Fourier optics — the Fraunhofer rung industrialized — is the working language of imaging science: lens design, adaptive optics, ptychography, and computational imaging all treat pupils as spatial-frequency filters, exactly as Derivation B's transform says. The same integral, relabeled, is aperture-antenna engineering: a reflector's pattern is the transform of its illumination — why antenna engineers and astronomers quote the same $\lambda/D$ rules and fight the same sidelobes (the kinship with Antenna Arrays and Directivity is not an analogy; it is the same mathematics with sums for integrals). The subwavelength end is where the modern action is. Bethe's fourth power sets the baseline for near-field scanning optical microscopy (NSOM probes at $a \approx \lambda/10$ pay the $(ka)^4$ tax on every photon), for EMC engineering (every slot in a shielded enclosure is a Bethe budget), and for the celebrated violation: Ebbesen's 1998 discovery that holes surrounded by periodic corrugations in real metals transmit orders of magnitude more than Bethe's law — extraordinary optical transmission, mediated by surface plasmons a perfect conductor does not support. The rule and its engineered violations now power plasmonic sensors, color filters, and recording heads; meanwhile EUV lithography at 13.5 nm runs full vector diffraction codes daily — the scalar-to-vector boundary of this article, drawn in silicon.

Origins · 1 minGustav Kirchhoff and how this was discovered

Gustav Kirchhoff

1824–1887 · Heidelberg and Berlin — who made Huygens' wavelets a theorem, and whose diffraction integral outlived its own contradiction

Kirchhoff is one of the century's great systematizers: the circuit laws bearing his name, the blackbody law that seeded quantum theory, spectroscopy with Bunsen — and, in 1882, the memoir turning Huygens' wavelet picture into a boundary integral derived from the wave equation. Fresnel had assembled wavelets with brilliant but ad hoc rules (his obliquity factor was guessed); Kirchhoff derived the integral from Green's theorem, obliquity factor and all, and it immediately became — and remains — the workhorse of practical optics. The irony took a decade to surface: Poincaré (1892) and Sommerfeld (1896, with the first rigorous diffraction solution, the half-plane) showed his boundary assumptions were self-contradictory — no field can satisfy them. The image repair preserved his integral's structure while restoring logic, and the verdict has stood: a formula can be wrong as an exact statement and indispensable as an asymptotic one. His integral computes every telescope point-spread function and antenna pattern to this day.

History

1678

— Huygens proposes the wavelet construction for wave propagation

1818

— Fresnel's memoir: wavelets plus interference; Poisson's "absurd" bright spot is observed by Arago, converting the doubters

1882

— Kirchhoff derives the diffraction integral from Green's theorem, obliquity factor included

1892

— Poincaré exposes the inconsistency of Kirchhoff's boundary conditions

1896

— Sommerfeld solves the conducting half-plane exactly — the first rigorous diffraction solution — and, with Rayleigh, repairs the scalar theory via image Green's functions

1944

— Bethe solves the subwavelength circular hole: transmission collapses as the fourth power

1998

— Ebbesen's extraordinary optical transmission: plasmon-assisted holes beat Bethe's law in real metals

The physical picture

Huygens' 1678 picture: freeze the wave at the aperture, let every open point emit a spherical wavelet, add the wavelets downstream with their phases. Diffraction theory is the accounting of that superposition — and the account has one deep line item. A wave equation is second order, so the field inside a region is fixed by boundary information — but how much? Specify both the field and its normal slope everywhere on a closed surface and you have overdetermined an elliptic problem: Helmholtz solutions are analytic, and analytic functions do not forgive contradictory Cauchy data. Physically, a screen lets you control one datum: a conductor pins the tangential field; nothing pins both field and slope. Kirchhoff's 1882 formula ignores this — it assumes the aperture carries the undisturbed incident wave and the dark side carries exactly nothing, field and slope alike. No actual field can do that. The miracle worth understanding is not that Kirchhoff is wrong (he is, provably) but that his integral works to fractions of a percent for wide apertures: the contradiction is confined to a rim of width $\sim\lambda$ . The Rayleigh–Sommerfeld repair is an old friend in new clothes. In electrostatics, the Method of Images built a Green's function that vanishes on a plane by reflecting the source and flipping its sign. Do exactly that with $e^{ikR}/4\pi R$ and the boundary term containing the unknown datum drops out of Green's identity — a formula needing only what you actually know. Same trick, same plane, now with phases. Far downstream the machinery simplifies into wave optics' most-used fact: the far-field pattern is the Fourier transform of the aperture. Wide apertures make narrow beams; sharp edges make broad wings; periodic structure makes discrete orders. And when the hole shrinks toward the wavelength the scalar picture quietly dies — vector boundary conditions take over, and transmission collapses by the Bethe fourth power.

Figure

The construction that powers everything — Green's identity on the half-space behind the screen, with the image source that builds the Rayleigh-Sommerfeld Green's function and kills the unknown boundary datum

The master identity

Factor the time dependence $e^{-i\omega t}$ out of a source-free Maxwell field and each Cartesian component obeys the scalar Helmholtz equation. Scalar theory promotes one representative component to the whole story:

(\nabla^2 + k^2)\,U(\mathbf{r}) = 0, \qquad k = \frac{\omega}{c} = \frac{2\pi}{\lambda}. \qquad (1)

where: $U$ — the complex amplitude of one field component. This is already the theory's original sin, priced honestly in Derivation C: components of $\mathbf{E}$ are not independent near boundaries — a conductor's edge couples them — so (1) per component is legitimate only where boundaries are many wavelengths away. Pair (1) with its outgoing point response, the free-space Green's function

G(\mathbf{r}, \mathbf{r}') = \frac{e^{ikR}}{4\pi R}, \qquad R = |\mathbf{r} - \mathbf{r}'|, \qquad (2)

where: the $+ikR$ sign is the Sommerfeld radiation condition (waves go out), and the $1/4\pi R$ normalization is fixed by the $k \to 0$ limit, where (2) must collapse to the Coulomb Green's function of the electrostatics chapter — the first of several places this theory leans on machinery the potential chapters already own.

Derivation AThe Helmholtz-Kirchhoff integral and the inconsistency theorem

The goal is two-sided: first the master identity — the field anywhere inside a closed surface, written entirely in terms of $U$ and $\partial_n U$ on that surface — and then the proof that Kirchhoff's use of it is self-contradictory. The route is Green's second identity applied to the pair $(U, G)$ , the same manipulation the electrostatic Green-function formalism runs (recalled from the E2 potential chapter, with $k^2$ terms now along for the ride), followed by an analyticity argument for the contradiction.

Step 1. Green's second identity for the volume $V$ bounded by $S$ (outward normal), applied to $U$ and $G$ :

\int_V \big(U\,\nabla'^2 G - G\,\nabla'^2 U\big)\,dV' = \oint_S \big(U\,\partial_{n'} G - G\,\partial_{n'} U\big)\,da'.

Step 2. Feed in the two Helmholtz equations: $\nabla'^2 U = -k^2 U$ (source-free in $V$ ) and $\nabla'^2 G = -k^2 G - \delta^3(\mathbf{r} - \mathbf{r}')$ (the point source sits at the observation point). The $k^2$ terms cancel identically and the delta sifts:

U(\mathbf{r}) = \oint_S \Big[G\,\frac{\partial U}{\partial n'} - U\,\frac{\partial G}{\partial n'}\Big]\,da'.

Why this is Huygens made exact: the field at $P$ is a superposition of point-source wavelets $G$ launched from every boundary element, weighted by the boundary data — no approximation yet, and no aperture yet. It is the wave-equation edition of the boundary-value representation the potential chapters use for Laplace fields.

Step 3. Kirchhoff's boundary choice. Close the surface with the screen plane plus a large hemisphere (whose contribution dies by the radiation condition). On the aperture $\Sigma$ , set $U$ and $\partial_n U$ to the unobstructed incident values; on the dark side of the screen, set both to zero. The result — with the incident spherical or plane wave inserted and the radiation-zone gradient $\partial_n G \approx ik\cos\theta\,G$ — is the workhorse Fresnel–Kirchhoff formula, with the celebrated obliquity factor $\tfrac12(\cos\theta + \cos\theta_0)$ .

Step 4. The inconsistency theorem. Suppose a Helmholtz solution really did vanish together with its normal derivative on a finite patch of the screen. Helmholtz solutions are real-analytic in the interior of any source-free region ( $U$ is a convolution of boundary data against the analytic kernel $G$ — elliptic regularity). Extend the field analytically across the patch: zero Cauchy data on a patch of positive area forces every derivative along and across the patch to vanish there, so the Taylor expansion at any patch point is identically zero, and by analytic continuation $U \equiv 0$ everywhere in the connected region. A nonzero diffracted field with Kirchhoff's boundary data is therefore a contradiction: no field satisfies Kirchhoff's assumptions.

Step 5. Why the formula survives its own contradiction. The identity of Step 2 is exact; only the inserted data are wrong, and wrong in a controlled place — within $\sim\lambda$ of the rim, where the true field smoothly interpolates rather than jumps. The corrupted strip carries a fraction $\sim\lambda/a$ of the integral: at $a = 100\lambda$ , a percent-level rim correction. Kirchhoff is an excellent large-aperture asymptotic wearing the costume of an exact solution. Verified: Step 2 against Jackson §10.5 and Born & Wolf §8.3 (and its $k \to 0$ collapse to the electrostatic boundary representation); the analyticity/uniqueness argument is the standard elliptic Cauchy rigidity (Born & Wolf §8.3.2); the $\tfrac12(\cos\theta + \cos\theta_0)$ factor rechecked against the RS average in Derivation B.

Go deeper · Derivation B · 2 minThe Rayleigh-Sommerfeld repair, and the ladder down to Fraunhofer

Derivation BThe Rayleigh-Sommerfeld repair, and the ladder down to Fraunhofer

The goal is the consistent scalar theory: build Green's functions adapted to the screen plane so that only one boundary datum is ever needed, then walk the result down the standard ladder — Fresnel's quadratic phase, the Fraunhofer Fourier transform, and the single-slit pattern as the executable check. The image construction is recalled by name from Method of Images: reflect the observation point through the plane $z = 0$ and combine.

Step 1. The two adapted Green's functions. With $\mathbf{r}'_{\rm im}$ the mirror image of the source point and $R'' = |\mathbf{r} - \mathbf{r}'_{\rm im}|$ :

G_\mp = \frac{e^{ikR}}{4\pi R} \mp \frac{e^{ikR''}}{4\pi R''}.

On the plane $z = 0$ , $R = R''$ : the minus combination vanishes there (Dirichlet), the plus combination has vanishing normal derivative (Neumann) — the same geometry that grounded conductors exploit in statics, with phases now riding along.

Step 2. Rayleigh–Sommerfeld I. Use $G_-$ in the master identity: the $G\,\partial_n U$ term dies on the whole plane, and on the plane $\partial_{n'}G_- = 2\,\partial_{n'}G$ (image doubling). Only the aperture field $U$ is needed — the datum an illuminated hole actually provides:

U_{\rm I}(\mathbf{r}) = \frac{1}{i\lambda}\int_\Sigma U(\mathbf{r}')\,\frac{e^{ikR}}{R}\,\cos\theta\,da',

keeping the radiation-zone gradient ( $kR \gg 1$ ), with $\cos\theta$ the tilt of the line-of-sight from the screen normal, and $k/2\pi = 1/\lambda$ absorbing the constants.

Step 3. Rayleigh–Sommerfeld II and the Kirchhoff verdict. Using $G_+$ instead kills the $U\,\partial_n G$ term and leaves $U_{\rm II} = -\frac{1}{2\pi}\int_\Sigma \partial_{n'}U\,\frac{e^{ikR}}{R}\,da'$ — the complementary consistent theory, obliquity $\cos\theta_0$ for plane-wave illumination. Each RS formula demands exactly one Cauchy datum: well-posed. Feeding both the same (Kirchhoff) aperture data, the Kirchhoff field is identically the average $\tfrac12(U_{\rm I} + U_{\rm II})$ — which is where its $\tfrac12(\cos\theta + \cos\theta_0)$ factor comes from, and why all three agree on-axis for large apertures, where the obliquity factors all approach one.

Step 4. The Fresnel rung. Put the screen at $z = 0$ , observe at distance $z$ ; expand $R = z\sqrt{1 + (\boldsymbol{\rho} - \boldsymbol{\rho}')^2/z^2} \approx z + \frac{(\boldsymbol{\rho} - \boldsymbol{\rho}')^2}{2z}$ in the phase (keep $R \approx z$ in the amplitude). The controlling object is the Fresnel number

N_F = \frac{a^2}{\lambda z},

which counts the half-period zones the aperture presents to the observer: $N_F \gtrsim 1$ means the quadratic phase matters (near field, on-axis intensity oscillates as zones alternate sign); $N_F \ll 1$ means it doesn't.

Step 5. The Fraunhofer rung and the slit. For $N_F \ll 1$ drop the quadratic aperture phase; what survives is a pure linear phase, and the integral becomes the spatial Fourier transform of the aperture field evaluated at transverse frequency $k\sin\theta$ :

U(\theta) \propto \int_\Sigma U(\mathbf{r}')\,e^{-ik\,\hat{\mathbf{r}}\cdot\mathbf{r}'}\,da'.

Executed for a uniform slit of width $a$ : $U \propto \int_{-a/2}^{a/2}e^{-ikx'\sin\theta}dx' = a\,\frac{\sin\beta}{\beta}$ with $\beta = \tfrac12 ka\sin\theta$ , hence $I(\theta) = I_0\,\mathrm{sinc}^2\beta$ — zeros exactly at $a\sin\theta = m\lambda$ , central lobe of full width $2\lambda/a$ . For a circular aperture of diameter $D$ the transform is a Bessel pair and the first dark ring sits at $\sin\theta = 1.22\,\lambda/D$ , the Airy radius, with $1.22 = 3.8317/\pi$ from the first zero of $J_1$ (computed, not memorized). Verified: image-doubling factor and RS-I/II forms against Goodman §3.5 and Born & Wolf §8.11; Kirchhoff = average confirmed algebraically; slit zeros and the $J_1$ zero $3.8317 \Rightarrow 1.2197$ recomputed numerically for this edition; the sinc² and Airy curves in the figure below are computed from these forms, not sketched.

Figure

The two canonical Fraunhofer patterns computed from the Fourier transform — slit sinc-squared with zeros at integer multiples, and circular-aperture Airy with its first zero pushed out to 1.22, the origin of the telescope resolution rule

Go deeper · Derivation C · 2 minVector diffraction and the Bethe hole, end to end

Derivation CVector diffraction and the Bethe hole, end to end

The goal is the honest boundary between scalar success and scalar failure: why the theory above cannot polarize, the Smythe formula that repairs it, and — the payoff — Bethe's exact small-hole transmission $T = \frac{64}{27\pi^2}(ka)^4$ , assembled end-to-end from effective dipoles with every factor audited. The dipole radiation is recalled by name from Electric Dipole Radiation: Exact Fields and Zone Structure and Magnetic Dipole and Electric Quadrupole Radiation; the small-sphere analogy runs next door in Scattering by Conducting and Dielectric Spheres.

Step 1. Why scalar theory cannot polarize. Each Cartesian component was promoted to an independent Helmholtz scalar — but at a conducting screen the boundary condition $\hat{\mathbf{n}}\times\mathbf{E} = 0$ couples components: rim currents regenerate field components the incident wave never had. For $a \gg \lambda$ the rim is a correction; for $a \lesssim \lambda$ the rim is the physics. The consistent vector statement (Smythe's form) writes the transmitted field entirely from the tangential aperture field,

\mathbf{E}(\mathbf{r}) = \frac{1}{2\pi}\,\nabla\times\int_\Sigma \big(\hat{\mathbf{n}}\times\mathbf{E}(\mathbf{r}')\big)\,\frac{e^{ikR}}{R}\,da',

divergence-free by construction — but it needs the true $\hat{\mathbf{n}}\times\mathbf{E}$ in the hole, which for a small hole is nothing like the incident field. Bethe's 1944 contribution was to solve for it.

Step 2. What drives a small hole. With the hole closed, a normally incident wave reflects: at the screen the tangential $\mathbf{E}$ vanishes while the tangential $\mathbf{H}$ doubles, $\mathbf{H}_{\rm scr} = 2\mathbf{H}_0$ . A hole with $ka \ll 1$ sees these quasi-static fields as uniform, and the static aperture problem — solvable exactly in oblate spheroidal coordinates — yields the effective moments seen from the far side:

\mathbf{m}_{\rm eff} = -\frac{4a^3}{3}\,\mathbf{H}_{\rm scr,\parallel}, \qquad \mathbf{p}_{\rm eff} = \frac{2a^3}{3}\,\varepsilon_0\,\mathbf{E}_{\rm scr,\perp},

the aperture polarizabilities. The magnetic one is twice the electric one — and at normal incidence, with no normal $\mathbf{E}$ to drive $\mathbf{p}$ , the hole transmits as a pure magnetic dipole: the little tangential- $\mathbf{H}$ antenna sketched below.

Step 3. Price the radiation. The transmitted power is the magnetic dipole's radiated power into the dark half-space. A tangential magnetic dipole on a conducting plane has a parallel image: moment doubled, radiation confined to a half-space — net factor $2$ relative to the same moment in free space. With $m_0 = \tfrac{4a^3}{3}(2H_0) = \tfrac{8a^3}{3}H_0$ and the harmonic magnetic-dipole power (recalled, by duality from the electric formula):

P_{\rm trans} = 2\times\frac{\mu_0\,\omega^4 m_0^2}{12\pi c^3} = \frac{32\,\mu_0\,\omega^4 a^6 H_0^2}{27\pi c^3}.

Step 4. The transmission coefficient. Divide by the power geometrically incident on the hole, $P_{\rm geo} = \tfrac12 Z_0 H_0^2 \cdot \pi a^2$ with $Z_0 = \mu_0 c$ :

T \equiv \frac{P_{\rm trans}}{P_{\rm geo}} = \frac{64}{27\pi^2}\,(ka)^4.

Why four powers: the moment is volumetric ( $a^3$ ), power squares it ( $a^6$ ), radiation prices $\omega^4$ , and normalizing by the hole's area leaves $(ka)^4$ — Rayleigh bookkeeping, because a subwavelength hole IS a Rayleigh scatterer made of absence. Scalar theory, blind to the boundary quench, predicts $T$ of order one — wrong by $(ka)^{-4}$ .

Step 5. Babinet's bonus. The complementary problem — a conducting disk where the hole was — is tied to the hole by the electromagnetic Babinet principle: swap $\mathbf{E} \leftrightarrow Z_0\mathbf{H}$ and the diffracted fields exchange roles. One solution, two geometries: Bethe's coefficients also govern the small conducting disk. Verified: the chain of Step 3–4 was run symbolically, giving $T = \frac{64}{27\pi^2}(ka)^4$ exactly — coefficient $64/27\pi^2 = 0.240$ , matching Bethe (Phys. Rev. 66, 163 (1944)) and Jackson §10.9; the polarizabilities $\alpha_m = 2\alpha_e$ against Collin and Jackson; numerically $T(ka = 0.314) = 2.3\times10^{-3}$ , $T(ka = 0.628) = 3.7\times10^{-2}$ .

Figure

Bethe's fourth-power law computed on log-log axes — a subwavelength hole transmits as an effective magnetic dipole, collapsing four orders of magnitude below the geometric expectation by the time the radius reaches a twentieth of the wavelength

Limiting cases

case	result	meaning
$\lambda \ll a$ , $N_F \gg 1$	geometrical optics + penumbra fringes of width $\sqrt{\lambda z}$	rays win; diffraction is an edge effect
$N_F = a^2/\lambda z \sim 1$	Fresnel regime: quadratic phase, on-axis intensity oscillates with zone count	the near field of an aperture
$N_F \ll 1$	Fraunhofer: pattern = Fourier transform of aperture; slit $\mathrm{sinc}^2$ , Airy $1.22\lambda/D$	the far field; antenna/telescope regime
aperture $\gg \lambda$	Kirchhoff ≈ RS-I ≈ RS-II (differ only in obliquity factors)	the inconsistency is a rim effect
$a \lesssim \lambda$	scalar theory fails; vector BCs and edge currents dominate	switch to Smythe / exact solutions
$ka \ll 1$	Bethe: $T = \frac{64}{27\pi^2}(ka)^4$ , magnetic-dipole transmission	four powers below geometric

Common trapKirchhoff's boundary conditions describe no field that can exist — and scalar answers for small holes are wrong by orders of magnitude, not percent

Two distinct failure modes, routinely conflated. The Kirchhoff inconsistency is mathematical: imposing $U = 0$ and $\partial_n U = 0$ on the dark side over-determines an elliptic problem — Derivation A shows the only such field is zero everywhere. The formula stays excellent for $a \gg \lambda$ ; treat it as a controlled asymptotic and switch to Rayleigh–Sommerfeld when consistency matters (back-propagation, near-aperture fields, careful phase work). The scalar failure is physical and violent: for $a \lesssim \lambda$ the vector boundary conditions quench transmission by $(ka)^4$ — a scalar estimate at $a = \lambda/20$ overestimates by $\sim400$ , and the error grows as the hole shrinks. Different fixes for different diseases: images repair Kirchhoff's logic; Bethe's dipoles replace scalar physics.

Worked examples

1 · The Airy budget of a space telescope (confidence builder)

Setup. Hubble's mirror: $D = 2.4$ m, observing at $\lambda = 550$ nm. Find the diffraction-limited angular resolution and the physical spot size at the focal plane ( $f = 57.6$ m effective).

Step 1. The Airy radius. $\theta = 1.22\,\lambda/D = 1.22 \times 550\times10^{-9}/2.4 = 2.80\times10^{-7}$ rad $= \mathbf{0.058}$ arcsec — the number every Hubble proposal is written against, and it is nothing but the first zero of $J_1$ ( $3.8317$ ) divided by $\pi$ , scaled by $\lambda/D$ .

Step 2. The focal-plane spot. $r_{\rm Airy} = f\theta = 57.6 \times 2.80\times10^{-7} = \mathbf{16\ \mu m}$ — matched to the detector pixel scale by design; undersample and you waste the mirror, oversample and you waste photons.

Step 3. The Fourier reading. Resolution $\propto 1/D$ because the mirror is a low-pass spatial-frequency filter of cutoff $\propto D/\lambda$ . A 6.5 m mirror (JWST) at the same wavelength resolves $2.7\times$ finer — same $J_1$ zero, bigger transform window.

The lesson: "diffraction-limited" is a Fourier-transform statement, and the $1.22$ is computable, not folklore.

2 · Why the microwave oven window is safe (exam level)

Setup. An oven door mesh has circular holes of radius $a = 1$ mm; the magnetron runs at 2.45 GHz ( $\lambda = 12.2$ cm). Estimate the power transmission per hole and the safety margin, then check the visible-light contrast.

Step 1. The size parameter. $ka = 2\pi a/\lambda = 2\pi \times 0.001/0.122 = \mathbf{0.0515}$ — deep Bethe regime, quasi-static hole.

Step 2. Bethe transmission. $T = \frac{64}{27\pi^2}(ka)^4 = 0.240 \times (0.0515)^4 = \mathbf{1.7\times10^{-6}}$ — each hole passes about two millionths of the power geometry would suggest. Summed over $\sim10^3$ holes, the leaked fraction of magnetron power stays far below $10^{-5}$ ; the measured leakage of a healthy oven ( $< 1$ mW/cm² at 5 cm) reflects exactly this fourth-power strangulation.

Step 3. The optical contrast. For green light through the same hole, $ka = 2\pi \times 0.001/(550\times10^{-9}) = 1.1\times10^4$ — the opposite limit: $T \approx 1$ , plain geometric transmission. The same mesh is opaque at 12 cm and transparent at 550 nm, a factor $\sim10^6$ apart in $ka$ — which is why you can watch your food while the microwaves cannot watch you.

Step 4. The scalar-theory counterfactual. A scalar estimate (aperture field ≈ incident field) predicts $T$ of order unity for the microwaves too — overestimating the leakage by six orders of magnitude. The mesh is safe because light is a vector field with boundary conditions.

The lesson: Bethe's $(ka)^4$ is consumer-product engineering — the window is a live demonstration of vector diffraction.

Going further

References. Jackson, Classical Electrodynamics, 3rd ed., §10.5–10.9 · Born & Wolf, Principles of Optics, 7th ed., Ch. 8 · Goodman, Introduction to Fourier Optics, 3rd ed., Ch. 3–4 · Stratton, Electromagnetic Theory (1941), Ch. VIII · Zangwill, Modern Electrodynamics, Ch. 21 · Bethe, Phys. Rev. 66, 163 (1944)

Provenance. Equations and structure follow Jackson §10.5–10.9, Born & Wolf Ch. 8, and Goodman Ch. 3–4. A verifier should reconfirm: the Green's-function normalization $1/4\pi R$ and its $k\to0$ Poisson limit; the Helmholtz–Kirchhoff identity via Green's second identity; the RS-I/RS-II obliquity factors $\cos\theta$ , $\cos\theta_0$ and their average reproducing Kirchhoff; the single-slit $\mathrm{sinc}^2$ with first minimum at $a\sin\theta=\lambda$ ; the Airy disk coefficient $1.22\,\lambda/D$ ; the grating/Bragg condition $d\sin\theta=m\lambda$ ; and Bethe's small-hole transmission $\tfrac{64}{27\pi^2}(ka)^4$ (Bethe, Phys. Rev. 66, 163 (1944); Jackson §10.9). All numbers are standard textbook values; none fabricated. Addendum for this edition: the Bethe coefficient was re-derived end-to-end rather than quoted — $\alpha_m = 4a^3/3 = 2\alpha_e$ , screen-doubling $H \to 2H_0$ , tangential-dipole image factor 2 — and the symbolic chain reproduces $\frac{64}{27\pi^2}(ka)^4$ exactly (computer-algebra verified; anchors $T(0.314) = 2.3\times10^{-3}$ , $T(0.0515) = 1.7\times10^{-6}$ ). The Airy constant was recomputed from the first $J_1$ zero ( $3.8317/\pi = 1.2197$ ); both Fraunhofer curves are computed, not sketched. Division of labor: Green's-identity and image machinery recalled from the E2 potential chapter (Method of Images, Uniqueness Theorems for the Potential); dipole radiation from Electric Dipole Radiation: Exact Fields and Zone Structure and its magnetic companion; aperture-antenna kinship deferred to Antenna Arrays and Directivity; the small-sphere analogy to Scattering by Conducting and Dielectric Spheres; extinction bookkeeping to Optical Theorem. Gratings, imaging theory, holography, and Fourier-optics practice are the Optics track's story — this article owns only the electrodynamic foundations.

Builds on: Helmholtz Equation and Green's Functions · Electric Dipole Radiation: Exact Fields and Zone Structure · Multipole Expansion of Radiation Fields · Fourier Analysis Leads to: Antenna Arrays and Directivity · Center-Fed Linear Antenna · Vector Spherical Harmonics

If Kirchhoff's assumptions are impossible, why does anyone still teach and use his formula?

Because it is a superb asymptotic wearing the costume of an exact solution, and most optics lives in its domain. The master identity is exact; only the inserted data are wrong, and the error concentrates within a wavelength of the rim — a fraction $\sim\lambda/a$ of the integral, i.e. $10^{-4}$ -level for a camera aperture at $a \sim 10^4\lambda$ . Moreover Kirchhoff is exactly the average of the two consistent RS formulas, so at small angles all three coincide. Use Kirchhoff freely for large apertures and moderate angles; reach for RS when you need well-posedness (near fields, back-propagation, wide angles), and for vector codes when geometry approaches $\lambda$ .

What is physically different between Rayleigh-Sommerfeld I and II — when would I prefer one?

They trust different data. RS-I propagates the aperture field $U$ (Dirichlet datum, obliquity $\cos\theta$ ); RS-II propagates its normal slope (Neumann, obliquity $\cos\theta_0$ ). Each is exact given its datum exactly; in practice you know neither perfectly and choose the more defensible guess. Behind a thin illuminated aperture, "field ≈ incident" (RS-I) is natural and standard in Fourier optics; RS-II suits pinned-slope boundaries (a pressure-release surface in acoustics). Their discrepancy, vanishing as $a/\lambda \to \infty$ , is itself the error bar: where they disagree materially, no scalar theory deserves trust.

Does the Poisson-Arago bright spot follow from this machinery, and why was it so convincing?

Directly. Behind an opaque disk every rim point sits at the same distance from the on-axis observer, so the rim wavelets arrive in phase and add: the axis behind a perfect disk is bright, comparable to the unobstructed wave. Poisson derived this from Fresnel's theory in 1818 as a reductio ad absurdum — no shadow has a bright heart, surely — and Arago promptly measured it. In Derivation B's zone language it is two lines: block every Fresnel zone except the rim and the survivors add coherently. (The spot is exquisitely sensitive to rim roughness — marks of size $\sim\sqrt{\lambda z}$ wash it out — why coins work poorly and precision disks work well.)

Why does the hole transmit as a magnetic dipole rather than an electric one?

Because of what a conducting screen does to the driving fields. Close the hole: the standing wave at the screen has zero tangential $\mathbf{E}$ but doubled tangential $\mathbf{H}$ (and doubled normal $\mathbf{E}$ ). The hole is driven by what survives at its location — and normal incidence supplies no normal $\mathbf{E}$ at all, leaving doubled tangential $\mathbf{H}$ as the only driver. The static aperture problem assigns $\alpha_m = 4a^3/3$ , twice $\alpha_e = 2a^3/3$ even when both are driven (oblique incidence). So the small hole is a little magnetic-dipole antenna, polarization signature to match — one of the cleanest confirmations of the vector theory, and the reason waveguide coupling holes sit where the mode's magnetic field is strongest (a placement rule Rectangular Waveguides uses).

How does this connect to the antenna patterns and array factors elsewhere in the chapter?

They are the same Fourier statement in two costumes. Derivation B says: far field = transform of the aperture excitation. Antenna Arrays and Directivity says: far field = element pattern × array factor, the transform of the discrete excitation. A continuous aperture is the $N \to \infty$ limit of an array; illumination taper plays exactly the role of apodizing a pupil (trading main-lobe width against sidelobes); the rule $D \sim 4\pi A/\lambda^2$ is the antenna reading of the diffraction limit; grating lobes of a sparse array are a grating's diffraction orders. Learn either formalism and you own both — why this article and the antenna pair cross-reference rather than duplicate.

Is Babinet's principle exact for electromagnetic fields, and what does it buy?

The electromagnetic version is exact for infinitely thin, perfectly conducting complementary screens: if screen B is the negative of screen A (metal where A had holes), the two diffracted fields are related by the duality swap $\mathbf{E} \to Z_0\mathbf{H}$ , $Z_0\mathbf{H} \to -\mathbf{E}$ . (The naive scalar version is only approximate — another scalar casualty.) It buys real leverage: Bethe's hole instantly yields the small conducting disk; slot antennas are designed as duals of dipoles (a slot radiates like a magnetic dipole with $E$ and $H$ exchanged); frequency-selective surfaces build band-stop screens from band-pass ones. It is the same electric–magnetic duality that organizes the channels of Multipole Expansion of Radiation Fields, here acting on boundary geometry.

Where exactly is the boundary between "use scalar theory" and "must go vector"?

Measure everything in wavelengths and ask what the boundaries do. Scalar theory needs field components to decouple, which holds when every edge and feature is many $\lambda$ away or many $\lambda$ in size: camera optics, telescope pupils, radio dishes, standard slits and gratings — all safely scalar. Warning signs, in order of onset: features below $\sim10\lambda$ (percent-level polarization dependence), high numerical apertures (longitudinal focal components), metallic structure at $\lambda$ scale (edge currents, resonances), anything subwavelength (Bethe regime — wrong by powers). A practical audit: compute with RS-I and RS-II; if they differ noticeably, or polarization matters at all, the scalar chapter is over. Practice then goes numerical (method of moments, FDTD) with Smythe/Stratton–Chu as the analytic frame — while the concepts (zones, transforms, $N_F$ ) survive as vocabulary.

Scalar and Vector Diffraction Theory

Where this lives today

The physical picture

The master identity

Limiting cases

Worked examples

1 · The Airy budget of a space telescope (confidence builder)

2 · Why the microwave oven window is safe (exam level)

Going further

Before this

Where this leads