The Helmholtz Equation

Almost every wave in physics is really two problems wearing one coat: a question about time — how fast does it oscillate? — and a question about space — what shape does it take? The Helmholtz equation is what remains after you strip the first away. Feed a wave equation, a diffusion equation, or the time-dependent Schrödinger equation a single pure frequency, factor the harmonic time dependence $e^{-i\omega t}$ out front, and the time derivative collapses into a number. What is left is the elliptic equation $(\nabla^2 + k^2)\psi = 0$ , with all the physics of the source compressed into one parameter: the wavenumber $k$ , equal to $\omega/c$ for a wave, $\sqrt{i\omega/D}$ for diffusion, or $\sqrt{2mE}/\hbar$ for a quantum particle. That single equation is the spine of an enormous amount of mathematical physics. On a bounded domain — a cavity, a drumhead, a waveguide cross-section — it becomes an eigenvalue problem: nontrivial solutions exist only for a discrete ladder of $k^2$ , and those eigenvalues are the resonant frequencies of the object. On an unbounded domain the ladder dissolves into a continuum, and the hard question flips from "which frequencies are allowed?" to "which way does the energy flow?" — answered by the Sommerfeld radiation condition, causality in the frequency domain. In the high-frequency limit the same equation turns back into geometry, its phase fronts obeying the eikonal equation of ray optics. Separation of variables, Sturm–Liouville theory, and Green's functions all meet here; this article treats the Helmholtz equation as the hub it is.

Key takeaways

The Helmholtz equation $(\nabla^2+k^2)\psi=0$ is the spatial skeleton left when you factor a single frequency $e^{-i\omega t}$ out of the wave or diffusion equation; $k=\omega/c=2\pi/\lambda$ .
It is elliptic — its principal symbol is Laplace's — but the $k^2$ term turns monotone harmonic solutions into oscillating standing waves.
On a bounded domain it is an eigenvalue problem with a discrete spectrum: a Dirichlet box rings only at $k_{lmn}^2=\pi^2(l^2/a^2+m^2/b^2+n^2/c^2)$ .
The sign of $k^2$ is decisive: $k^2>0$ gives propagating waves, $k^2<0$ the exponentially decaying modified-Helmholtz (Yukawa) regime.
On $\mathbb{R}^3$ the outgoing Green's function is $e^{ikr}/4\pi r$ , and the Sommerfeld radiation condition discards the unphysical incoming wave to make scattering well-posed.
Key idea: factor $u=\psi(\mathbf{r})e^{-i\omega t}$ out of a wave or diffusion equation and the spatial part always satisfies $(\nabla^2+k^2)\psi=0$ ; the value and sign of $k^2$ decide whether solutions oscillate or decay, and on bounded domains force $k^2$ onto a discrete eigenvalue ladder.

The core insight

Factor harmonic time dependence $u(\mathbf{r},t)=\psi(\mathbf{r})e^{-i\omega t}$ out of a wave or diffusion equation and the spatial part always satisfies $(\nabla^2+k^2)\psi=0$ ; the value and sign of $k^2$ — set by frequency and the parent equation — control whether solutions oscillate or decay, and on bounded domains force $k^2$ to take discrete eigenvalues.

Where this lives today

The Helmholtz equation is the daily bread of computational physics. Every finite-element solver used to design an antenna, a filter, a photonic crystal, or a concert hall is at heart a discretized Helmholtz solver, and the frequency-domain form is preferred because it replaces time-stepping with one large linear-algebra problem per frequency. Its numerical difficulty is a live frontier: at high wavenumber the discretized operator is indefinite and notoriously hard to precondition — the "high-frequency Helmholtz problem" — and building fast solvers for it is an active area in numerical analysis. In seismic imaging, full-waveform inversion reconstructs the subsurface by solving the Helmholtz equation millions of times; in photonics, the same equation with spatially varying $k^2(\mathbf{r})$ predicts the band gaps of photonic crystals and the modes of nanophotonic cavities. And the inverse spectral question Kac made famous — whether the spectrum determines the shape — was answered "no" in 1992 by Gordon, Webb, and Wolpert, who exhibited two differently shaped drums with identical Helmholtz spectra.

The physical picture

Picture a clamped drumhead. The full wave equation lets it flop around any way at all, but ask a sharper question — in what shapes can it ring at one single pure pitch? — and you are solving Helmholtz. Each pitch is an eigenvalue $k^2$ , each shape an eigenfunction, and the clamped rim forbids all but a discrete ladder of pitches: a free membrane could ring at any frequency, so the boundary is what quantizes. The first figure shows this directly — the three lowest modes of a square membrane and the exact wavenumbers at which it may resonate.

Figure

Quantization on a bounded square membrane: clamping the rim forces k onto the discrete set kₘₙ = (π/a)√(m²+n²), so the three lowest modes ring only at k₁₁ = π√2/a, k₂₁ = π√5/a, and k₂₂ = 2π√2/a — the drum has no notes in between

Now flip the sign of the constant term and the character of the equation changes entirely. The $+k^2$ term behaves like a restoring spring spread through space, bending solutions back toward zero to produce oscillations of wavelength $2\pi/k$ . Replace it with $-\kappa^2$ and the spring becomes a sink — solutions decay exponentially, and the ringing cavity becomes a screened electrostatic potential. The second figure puts the two regimes side by side: the outgoing oscillation of genuine Helmholtz against the evanescent Yukawa tail of its modified cousin.

Figure

The sign of k² decides everything: for k² greater than 0 the radial solution is an outgoing oscillation e^(ikr)/r of wavelength λ = 2π/k that carries energy to infinity (left), while for k² = −κ² it becomes the evanescent Yukawa tail e^(−κr)/r that dies over one screening length 1/κ and carries nothing (right)

On an infinite domain nothing quantizes, so the physics moves to a different question: which way does energy flow? Because both $e^{ikr}/r$ and $e^{-ikr}/r$ solve the equation, the mathematics alone cannot decide. The radiation condition rules out the incoming wave — it is the frequency-domain face of causality — and is the analytic heart of every scattering and antenna calculation.

The core equations

The equation is the whole subject in one line: Laplace's equation with a single extra term, from which every qualitative feature — oscillation, quantization, resonance, radiation — flows. Read it as the eigenvalue statement $-\nabla^2\psi=k^2\psi$ .

\nabla^2 \psi + k^2 \psi = 0 \qquad (1)

where: $\psi(\mathbf{r})$ — spatial amplitude (a field, pressure, displacement, or wavefunction); $\nabla^2$ — the Laplacian $\partial_x^2+\partial_y^2+\partial_z^2$ ; $k$ — the wavenumber. Reading: $\psi$ is an eigenfunction of $-\nabla^2$ with eigenvalue $k^2$ ; the equation is elliptic because its principal symbol is $\nabla^2$ , exactly Laplace's. The equation is what a time-harmonic wave leaves behind. Assume a single frequency, so all time dependence lives in one exponential, and the parent equation supplies $k$ .

u(\mathbf{r},t) = \psi(\mathbf{r})\,e^{-i\omega t} \qquad (2)

where: $\omega$ — angular frequency; $\psi(\mathbf{r})$ — the spatial part, which satisfies equation (1). Reading: this ansatz is the bridge from dynamics to Helmholtz — the second time derivative produces $-\omega^2$ , which becomes $+k^2$ . For a nondispersive wave of speed $c$ , wavenumber, frequency, and wavelength are locked together, assigning a numerical $k$ to any source.

k = \frac{\omega}{c} = \frac{2\pi}{\lambda} \qquad (3)

where: $c$ — phase speed; $\lambda$ — wavelength. Reading: doubling the frequency doubles $k$ and halves the wavelength; every Helmholtz solution wiggles on the scale $\lambda=2\pi/k$ . On an unbounded domain, equation (1) alone does not fix a unique solution — you must say what happens at infinity. The Sommerfeld radiation condition demands the field look like an outgoing wave far away, discarding the incoming solution.

\lim_{r\to\infty} r^{(d-1)/2}\left(\frac{\partial}{\partial r} - ik\right)\psi = 0 \qquad (4)

where: $r=|\mathbf{r}|$ — radial distance; $d$ — spatial dimension. Reading: $\partial_r-ik$ annihilates the outgoing term $e^{ikr}/r^{(d-1)/2}$ but not the incoming one, so the vanishing limit keeps only the wave carrying energy to infinity. Driving the equation with a point source defines the Green's function, from which all inhomogeneous problems are built by superposition — solving it once solves radiation and scattering for any source.

(\nabla^2 + k^2)\,G(\mathbf{r},\mathbf{r}') = -\delta^{(3)}(\mathbf{r}-\mathbf{r}'), \qquad G = \frac{e^{ik|\mathbf{r}-\mathbf{r}'|}}{4\pi|\mathbf{r}-\mathbf{r}'|} \qquad (5)

where: $G$ — outgoing free-space Green's function; $\delta^{(3)}$ — 3D Dirac delta; $|\mathbf{r}-\mathbf{r}'|$ — source-to-field distance. Reading: a unit point source radiates $e^{ikr}/4\pi r$ ; as $k\to0$ this collapses to the Coulomb kernel $1/4\pi r$ . On a bounded domain the spectrum is discrete, and for the rectangular Dirichlet box completely explicit — these eigenvalues are the resonant wavenumbers, and $c$ times them are the frequencies.

k_{lmn}^2 = \pi^2\!\left(\frac{l^2}{a^2}+\frac{m^2}{b^2}+\frac{n^2}{c^2}\right), \qquad \psi_{lmn}=\sin\frac{l\pi x}{a}\sin\frac{m\pi y}{b}\sin\frac{n\pi z}{c} \qquad (6)

where: $a,b,c$ — box edge lengths; $l,m,n$ — positive integers labelling the mode. Reading: each pinned face quantizes one direction independently, the total eigenvalue is the Pythagorean sum, and $\omega_{lmn}=c\,k_{lmn}$ are the pitches at which the cavity rings.

Derivation AFrom the wave equation to Helmholtz, and the rectangular cavity spectrum

The goal is to do the two foundational moves in one place: first show that a single-frequency assumption turns the hyperbolic wave equation into the elliptic Helmholtz equation, then solve it in the simplest bounded geometry — a rectangular box with every face held at zero — to watch the discrete spectrum emerge from the boundary conditions. The payoff is the eigenvalue formula $k_{lmn}^2=\pi^2(l^2/a^2+m^2/b^2+n^2/c^2)$ and a first-principles account of why a closed cavity resonates only at a discrete ladder of frequencies.

Step 1. Substitute the ansatz $u=\psi(\mathbf{r})e^{-i\omega t}$ into $\partial_t^2 u=c^2\nabla^2 u$ . The time derivative acts only on the exponential, $\partial_t^2 u=-\omega^2\psi e^{-i\omega t}$ , while the Laplacian acts only on $\psi$ ; cancelling $e^{-i\omega t}$ leaves $\nabla^2\psi+(\omega^2/c^2)\psi=0$ , and $k=\omega/c$ gives equation (1). Why: the sign is the punchline — $-\omega^2$ from the time side becomes $+k^2$ on the spatial side, so the constant term is positive and solutions oscillate rather than decay.

Step 2. Now solve in the box $0\le x\le a,\ 0\le y\le b,\ 0\le z\le c$ with $\psi=0$ on every face, seeking a product $\psi=X(x)Y(y)Z(z)$ . Dividing by $XYZ$ gives $X''/X+Y''/Y+Z''/Z=-k^2$ . Why: a function of $x$ alone can equal a function of $y,z$ alone only if each is constant, so the equation splits into three independent ordinary problems.

Step 3. Set each ratio to a constant, $X''/X=-\alpha^2$ and so on, with $\alpha^2+\beta^2+\gamma^2=k^2$ . Then $X(0)=0$ forces $X=\sin\alpha x$ , and $X(a)=0$ forces $\alpha a=l\pi$ ; likewise $\beta b=m\pi$ , $\gamma c=n\pi$ . Why: the boundary conditions do the quantizing — a free interval allows any $\alpha$ , but pinning both ends selects the discrete $\alpha=l\pi/a$ .

Step 4. Assemble: $\psi_{lmn}=\sin(l\pi x/a)\sin(m\pi y/b)\sin(n\pi z/c)$ with $k_{lmn}^2=\pi^2(l^2/a^2+m^2/b^2+n^2/c^2)$ , exactly equation (6), and $\omega_{lmn}=c\,k_{lmn}$ . Why: the allowed wavenumbers form a Pythagorean lattice; letting $b,c\to\infty$ recovers a continuum in those directions and the 1D string spectrum $k_l=l\pi/a$ — a clean consistency check.

Go deeper · Derivation B · 1 minThe outgoing free-space Green's function by Fourier transform

Derivation BThe outgoing free-space Green's function by Fourier transform

The aim is to construct the fundamental solution $G(\mathbf{r})=e^{ikr}/4\pi r$ of the point-driven equation $(\nabla^2+k^2)G=-\delta^{(3)}(\mathbf{r})$ from Fourier analysis, and to see where the outgoing character comes from: not from the algebra, which is symmetric between incoming and outgoing, but from a deliberate choice of contour that pushes the poles off the real axis. This is the most important solution in the subject — scattering, radiation, and integral-equation methods are all built by convolving a source against this kernel, and the same $i\epsilon$ choice encodes causality.

Step 1. Fourier transform both sides. Under $\nabla^2\to-q^2$ the equation becomes algebraic, $(-q^2+k^2)\tilde G=-1$ , so $\tilde G(\mathbf{q})=1/(q^2-k^2)$ . Why: the transform trades a differential equation for a division; all the difficulty is now in inverting past the poles at $q=\pm k$ .

Step 2. Invert with $d^3q=q^2\,dq\,d\Omega$ and the angular integral $\int d\Omega\,e^{i\mathbf{q}\cdot\mathbf{r}}=4\pi\sin(qr)/(qr)$ , which collapses the inversion to $G=\dfrac{1}{2\pi^2 r}\displaystyle\int_0^\infty\frac{q\sin(qr)}{q^2-k^2}\,dq$ . Why: spherical symmetry means only the radial magnitude matters, and the standard angular integral turns the plane wave into a $\sin(qr)/qr$ kernel.

Step 3. Since the integrand is even, $\int_0^\infty=\tfrac{1}{2i}\int_{-\infty}^\infty q\,e^{iqr}/(q^2-k^2)\,dq$ , with poles on the contour at $q=\pm k$ ; the outgoing solution displaces them by $k\to k+i\epsilon$ . Why: the real-axis poles make the bare integral ambiguous, and that ambiguity is physical — the ways to skirt the poles are the retarded, advanced, and Feynman prescriptions, and $k\to k+i\epsilon$ picks the outgoing (retarded) wave.

Step 4. For $r>0$ , $e^{iqr}$ decays in the upper half plane; closing there encloses only $q=k+i\epsilon$ , residue $e^{ikr}/2$ , so $\int_{-\infty}^\infty=\pi i\,e^{ikr}$ and $G=\dfrac{1}{2\pi^2 r}\cdot\dfrac{1}{2i}\cdot\pi i\,e^{ikr}=\dfrac{e^{ikr}}{4\pi r}$ , equation (5). Why: the contour choice, not the algebra, makes the answer $e^{+ikr}$ rather than $e^{-ikr}$ — that is the source of outgoingness. As a check, $k\to0$ gives $1/4\pi r$ , the Coulomb Green's function.

Go deeper · Derivation C · 1 minThe high-frequency limit and the eikonal equation of ray optics

Derivation CThe high-frequency limit and the eikonal equation of ray optics

This derivation shows that the Helmholtz equation contains geometrical optics as its short-wavelength limit: when $k$ is large compared with the scale on which the medium varies, the equation degenerates into rules for rays and the amplitude carried along them. The goal is to extract, at leading and next-to-leading order in $k$ , the eikonal equation $|\nabla S|^2=1$ governing phase fronts and the transport equation conserving intensity along rays — the bridge from wave optics to ray tracing, and the classical analogue of the WKB approximation.

Step 1. Write $\psi=A(\mathbf{r})\,e^{ikS(\mathbf{r})}$ with $S$ the optical path (the eikonal) and $A$ a slow amplitude, and compute $\nabla^2\psi=\big[\nabla^2 A+2ik\,\nabla A\cdot\nabla S+ikA\,\nabla^2 S-k^2A\,|\nabla S|^2\big]e^{ikS}$ . Why: separating a fast phase from a slow amplitude organizes the equation by powers of $k$ , since each phase derivative brings down a factor of $k$ .

Step 2. Insert into $\nabla^2\psi+k^2\psi=0$ and collect the highest order $k^2$ : $k^2A(1-|\nabla S|^2)=0$ , so $|\nabla S|^2=1$ , the eikonal equation. Why: the dominant balance keeps only the phase; the wavefronts $S=\text{const}$ propagate normal to themselves at unit optical speed, their normals are the rays, and $|\nabla S|=1$ is Fermat's principle in differential form.

Step 3. The next order $k^1$ gives $2\nabla A\cdot\nabla S+A\,\nabla^2 S=0$ , i.e. $\nabla\cdot(A^2\nabla S)=0$ . Why: this is energy conservation along ray tubes — intensity $A^2$ times ray direction $\nabla S$ is divergence-free. The neglected $\nabla^2 A$ term at order $k^0$ is the first diffraction correction, small precisely when $k$ is large.

Limiting cases

The behaviour of $(\nabla^2+k^2)\psi=0$ turns entirely on the value and sign of $k^2$ — one equation spanning static fields, propagating waves, screened potentials, and ray optics.

Case	Result	Meaning
Static limit $k\to0$	$\nabla^2\psi=0$	Recovers Laplace's equation; wavelength $2\pi/k\to\infty$ , solutions become harmonic with the mean-value property — electrostatics, steady heat.
Propagating $k^2>0$	$(\nabla^2+k^2)\psi=0$	Genuine Helmholtz; oscillatory standing or travelling waves of wavelength $\lambda=2\pi/k$ , governing resonance, waveguides, and scattering.
Evanescent $k^2=-\kappa^2$	$(\nabla^2-\kappa^2)\psi=0$	Modified (screened) Helmholtz; solutions decay as $e^{-\kappa r}/r$ — the Yukawa/Debye potential, no propagation, a waveguide below cutoff.
High frequency $k\to\infty$	$\psi\sim A\,e^{ikS}$	WKB / geometrical-optics limit; phase fronts $S=\text{const}$ obey the eikonal equation $\|\nabla S\|^2=1$ , recovering ray tracing.
Bounded vs unbounded	discrete $k_n^2$ vs continuum	A finite domain quantizes $k^2$ into a Sturm–Liouville ladder; $\mathbb{R}^n$ admits every $k^2\ge0$ , so quantization yields to the radiation condition.

Common trapConfusing the Helmholtz equation with the modified Helmholtz equation, and forgetting the radiation condition

Two traps recur. The first is sign blindness: $(\nabla^2+k^2)\psi=0$ and $(\nabla^2-\kappa^2)\psi=0$ look almost identical but describe opposite worlds — the first oscillates and radiates, the second decays and screens, and no algebra converts one into the other because the character is fixed by the sign of the constant term. Writing $k^2<0$ and then using $e^{ikr}$ with imaginary $k$ hides an exponential that either blows up or dies; decide the regime up front. The second trap is treating the unbounded problem as well-posed without extra input. On $\mathbb{R}^3$ the homogeneous equation has both outgoing and incoming solutions, so a scattering problem has infinitely many answers until the Sommerfeld radiation condition keeps only the outgoing one. Omit it and your scattered field is defined only up to an arbitrary incoming wave — not wrong, but ambiguous.

Worked examples

1 · Resonant frequency of a rectangular microwave cavity (fully worked)

A closed rectangular microwave cavity has inner dimensions a = 4.0 cm, b = 2.0 cm, d = 5.0 cm. Using the box eigenvalues of equation (6), find the resonant frequency of the dominant mode.

Step 1. The resonant frequencies are $f_{lmn}=c\,k_{lmn}/2\pi=(c/2)\sqrt{l^2/a^2+m^2/b^2+n^2/d^2}$ . The lowest electromagnetic resonance is the TE $_{101}$ mode, one variation along the long pair of dimensions and none along the short, so $(l,m,n)=(1,0,1)$ .

Step 2. Insert the numbers with $a=0.040$ m and $d=0.050$ m: $1/a^2=625\ \mathrm{m^{-2}}$ and $1/d^2=400\ \mathrm{m^{-2}}$ , so $\sqrt{625+400}=\sqrt{1025}=32.0\ \mathrm{m^{-1}}$ . Then $k_{101}=\pi\times32.0=100.6\ \mathrm{m^{-1}}$ .

Step 3. Convert with $c=2.998\times10^8$ m/s: $f_{101}=(c/2)\times32.0=(1.499\times10^8)(32.0)=\mathbf{4.80\ GHz}$ . This sits in the C-band — a centimetre-scale cavity rings in the microwave, which is why such cavities are the workhorses of radar and accelerator RF. The dominant TE₁₀₁ mode resonates at f ≈ 4.80 GHz.

2 · Cutoff and evanescence in a rectangular waveguide (fully worked)

A WR-90 rectangular waveguide has broad-wall width a = 22.86 mm. Its dominant TE₁₀ mode obeys the transverse Helmholtz problem with cutoff wavenumber k_c = π/a. Find the cutoff frequency, then find how strongly a 5.0 GHz signal (below cutoff) is attenuated over 5.0 cm.

Step 1. The cutoff is where the transverse eigenvalue exhausts the wavenumber. With $k_c=\pi/a=137.4\ \mathrm{m^{-1}}$ , the cutoff frequency is $f_c=c/2a=(2.998\times10^8)/(2\times0.02286)=\mathbf{6.56\ GHz}$ . Above $f_c$ the axial wavenumber $k_z=\sqrt{k^2-k_c^2}$ is real and the mode propagates; below it, $k_z$ is imaginary and the mode is evanescent — the modified-Helmholtz regime.

Step 2. At $f=5.0$ GHz the free-space wavenumber is $k=2\pi f/c=104.8\ \mathrm{m^{-1}}$ , below $k_c=137.4\ \mathrm{m^{-1}}$ — confirming we are below cutoff. The field decays as $e^{-\kappa z}$ with $\kappa=\sqrt{k_c^2-k^2}=\sqrt{137.4^2-104.8^2}=\mathbf{88.9\ m^{-1}}$ .

Step 3. The attenuation length is $1/\kappa=1/88.9=0.0112$ m $=\mathbf{1.12\ cm}$ . Over $z=5.0$ cm the amplitude falls by $e^{-\kappa z}=e^{-4.45}=0.0117$ , a factor of about $85$ , or $20\log_{10}(0.0117)=\mathbf{-38.6\ dB}$ — which is exactly how a cutoff waveguide serves as a calibrated microwave attenuator. f_c = 6.56 GHz; at 5 GHz the mode is evanescent with decay length 1.12 cm, giving −38.6 dB over 5 cm.

Physics of the portrait

The equation carries the name of the polymath who first extracted it from a problem in acoustics — the resonance of air in cavities and pipes — and who, more than anyone, embodied the nineteenth-century unification of physics, physiology, and mathematics.

Origins · 1 minHermann von Helmholtz and how this was discovered

Hermann von Helmholtz

1821–1894 · the acoustics of resonators and the mathematics of standing waves

Hermann von Helmholtz was arguably the last physicist to master the whole of physics, and much of physiology besides. Trained as a military surgeon, he made his first great mark in 1847 with a rigorous statement of the conservation of energy, Über die Erhaltung der Kraft. He then spent two decades on the physics and physiology of perception — inventing the ophthalmoscope, measuring the speed of nerve impulses, and building a complete theory of hearing and musical tone. It was in that acoustic work, above all his 1863 Die Lehre von den Tonempfindungen (On the Sensations of Tone) and the papers around 1860 on air vibrations in open pipes and resonators, that $(\nabla^2+k^2)\psi=0$ became a central tool: the resonant modes of a cavity of air are precisely its eigenfunctions, and the sharply tuned "Helmholtz resonator" — a rigid cavity with a neck that rings at one frequency — is the physical embodiment of a single Helmholtz eigenvalue. His Berlin institute later trained a generation including Hertz and Michelson. The equation's name is fitting shorthand for a career spent finding the mathematics inside physical sensation.

History

1860

— Helmholtz analyzes standing air waves in open pipes and resonators, giving the equation its name

1877

— Rayleigh's Theory of Sound systematizes cavity and membrane resonance as an eigenvalue problem

1882

— Kirchhoff casts scalar diffraction as a surface integral of the Helmholtz Green's function

1911

— Weyl proves the asymptotic law for how eigenvalues of the Laplacian accumulate

1912

— Sommerfeld formulates the radiation condition that selects outgoing waves

1966

— Kac poses "Can one hear the shape of a drum?", framing the inverse spectral problem

Going further

Texts. Arfken, Weber & Harris, Mathematical Methods for Physicists, 7th ed., Ch. 9 · Hassani, Mathematical Physics, 2nd ed., Ch. 22–24 · Stone & Goldbart, Mathematics for Physics, Ch. 7–8 · Byron & Fuller, Mathematics of Classical and Quantum Physics, Ch. 7 · Nakahara, Geometry, Topology and Physics (spectral viewpoint) · Morse & Feshbach, Methods of Theoretical Physics, Vol. I · Jackson, Classical Electrodynamics, 3rd ed., Ch. 9.

Provenance. The time-harmonic reduction and elliptic classification follow Arfken §9 and Stone & Goldbart Ch. 7–8. The box eigenvalues $k_{lmn}^2=\pi^2(l^2/a^2+m^2/b^2+n^2/c^2)$ and the separation construction are standard (Arfken §9.4, Hassani Ch. 22). The Green's function $e^{ikr}/4\pi r$ with the $i\epsilon$ contour and the radiation condition follow Jackson §9 and Morse & Feshbach Vol. I; a verifier should recompute the Fourier-inversion residue and confirm the $k\to0$ Coulomb limit $1/4\pi r$ . The eikonal reduction $|\nabla S|^2=1$ and its transport equation cross-check against Born & Wolf. Worked numbers are recomputable: TE $_{101}$ at $f=(c/2)\sqrt{1/a^2+1/d^2}=4.80$ GHz for $a=4$ cm, $d=5$ cm; WR-90 cutoff $f_c=c/2a=6.56$ GHz, with below-cutoff decay $\kappa=88.9\ \mathrm{m^{-1}}$ at 5 GHz. All mode indices and the $1/4\pi$ prefactor are exact textbook values.

Builds on: The Wave Equation · The Heat (Diffusion) Equation · Laplace's and Poisson's Equations · Classification of Second-Order Linear PDEs · Separation of Variables in Cartesian Coordinates. Leads to: Separation of Variables in Cylindrical Coordinates · Separation of Variables in Spherical Coordinates · Sturm–Liouville Theory and Eigenfunction Expansions · Fundamental Solutions and Green's Functions for PDEs · Boundary Conditions and Well-Posedness.

Ask the tutor: Ask me to separate the Helmholtz equation in cylindrical or spherical coordinates, work the Born approximation from the $e^{ikr}/4\pi r$ Green's function, or compute the mode spectrum of a cavity or drum of your choice.

Why does factoring out e^(−iωt) turn a hyperbolic equation into an elliptic one?

The type of a PDE is fixed by its highest-derivative (principal) part alone. The wave equation is hyperbolic because $+\partial_t^2$ and $-c^2\nabla^2$ have opposite signs, giving real characteristics along which signals travel. Assuming one frequency replaces $\partial_t^2$ by the number $-\omega^2$ — it stops being a derivative. What remains, $\nabla^2\psi+k^2\psi=0$ , has only same-sign spatial second derivatives in its principal part, the definition of elliptic. The physics matches: hyperbolic equations propagate in time, elliptic ones describe equilibrium spatial patterns, and a single-frequency standing wave is exactly such a fixed pattern.

What is the difference between the Helmholtz and modified Helmholtz equations?

Only a sign, but it changes everything. The Helmholtz equation $(\nabla^2+k^2)\psi=0$ has a positive constant term, so solutions oscillate with wavelength $2\pi/k$ and can carry energy as travelling waves. The modified (screened) form $(\nabla^2-\kappa^2)\psi=0$ has a negative term, and solutions decay as $e^{-\kappa r}/r$ — the Yukawa potential of a massive field, Debye screening in a plasma, or a waveguide mode below cutoff. The first has real wavenumber and propagates; the second has imaginary wavenumber and is evanescent. You meet the modified form whenever the frequency is too low, the field too massive, or the medium too lossy for a wave to exist.

Why can't I solve the scattering problem without the radiation condition?

Because the equation alone has no unique answer on an infinite domain. Far from a scatterer the field satisfies the homogeneous Helmholtz equation, which has two independent spherical solutions: the outgoing $e^{ikr}/r$ and the incoming $e^{-ikr}/r$ . Both are legitimate, so you can add any amount of the incoming wave and still satisfy every equation and boundary condition. Causality breaks the tie: a source sends waves out, it does not conspire with infinity to have waves converge onto it. The Sommerfeld radiation condition encodes exactly that and restores uniqueness. Without it, "the scattered field" is not well defined.

How does the Helmholtz equation quantize frequencies on a bounded domain?

On a finite domain with homogeneous boundary conditions it becomes the eigenvalue problem $-\nabla^2\psi=k^2\psi$ with $\psi$ (or its normal derivative) vanishing on the boundary. Sturm–Liouville theory guarantees solutions only for a discrete, increasing sequence $k_1^2<k_2^2<\cdots$ , with orthogonal eigenfunctions forming a complete basis. For the box, each boundary pins a sine to zero at both ends, forcing an integer number of half-wavelengths to fit — and that integer is the quantization. On an unbounded domain nothing is pinned, so every $k^2\ge0$ is allowed and the spectrum is continuous. A drum's overtones, a laser cavity's modes, and an atom's energy levels are all this mechanism.

In what sense is geometrical optics a limit of the Helmholtz equation?

When the wavelength is far smaller than the scale over which the medium changes, the wave barely diffracts and behaves like a bundle of rays. Writing $\psi=A\,e^{ikS}$ for large $k$ separates a fast phase from a slow amplitude; the leading power of $k$ gives the eikonal equation $|\nabla S|^2=1$ , whose surfaces $S=\text{const}$ are wavefronts and whose normals are rays — Fermat's principle in local form. The next order gives a transport equation conserving intensity along ray tubes. The neglected amplitude-Laplacian term is the first diffraction correction, important again when the wave is squeezed to a wavelength-sized region — a focus, edge, or caustic — where ray optics fails.

Is the same Green's function e^(ikr)/4πr used for both radiation and scattering?

Yes — it is the most reused object in the subject. For radiation you convolve a known source against $e^{ikr}/4\pi r$ to get the field directly; the far-field $e^{ikr}/r$ behaviour and the radiation condition are already built in. For scattering you rewrite $(\nabla^2+k^2)\psi=V\psi$ as the Lippmann–Schwinger integral equation, in which the same Green's function propagates the wave between scattering events; iterating gives the Born series, and to first order the Born approximation, where the amplitude is the Fourier transform of the potential. The only change is the source — a prescribed current versus the self-consistent $V\psi$ — while the outward-carrying kernel is identical.