Whispering Gallery Modes: How Sound and Light Cling to a Curved Wall

Rayleigh's puzzle: not an echo, but a wave hugging the wall

When John William Strutt, Lord Rayleigh, investigated the whispering gallery of St Paul's, the obvious explanation — sound focused across the dome by reflection from the curved ceiling — failed. A whisper is heard best not across the diameter but along the wall, and it travels equally well the short way or the long way round the circle. Rayleigh's insight, published in The Theory of Sound (1878) and refined in a 1910 Philosophical Magazine paper, was that the sound does not cross the enclosed space at all.

Instead, a wave launched nearly tangent to a concave wall strikes it at grazing incidence, reflects, travels a short chord, strikes again, and repeats. Each reflection off the rigid stone loses almost no energy, and because the surface curves away, the wave is continually 'caught' before it can spread into the room. The disturbance is trapped in a thin annular shell hugging the wall — a running wave that clings to the curve rather than a beam crossing the void. The acoustic intensity falls off exponentially with distance inward from the wall, so all the energy stays within a metre or two of the stonework.

The governing equation and the mode condition

Both the acoustic and optical problems reduce to the same scalar Helmholtz equation, the time-independent form of the wave equation for a field ψ (pressure for sound, an electric-field component for light):

• ∇²ψ + k²ψ = 0, with wavenumber k = 2π/λ = ω/v, where v is the wave speed (343 m/s for air, c/n for light in glass).
• In a circle of radius a, separate variables as ψ(r,φ) = R(r)·e^imφ. The angular factor forces the azimuthal index m to be an integer so the wave closes on itself after one lap.
• The radial part obeys Bessel's equation, so R(r) = J_m(kr) — a Bessel function of order m. Whispering gallery modes are the solutions where J_m(kr) piles almost all its amplitude just inside the boundary r = a, in the region m/k ≲ r ≤ a.

The resonance (mode) condition is that the wave interferes constructively after one round trip: the round-trip optical/acoustic path length must equal an integer number of wavelengths. To leading order this is

• 2πa n ≈ m λ, i.e. m ≈ 2πa/λ (with n = 1 for sound in air). The integer m counts the number of wavelengths — equivalently the number of intensity maxima — around the circumference.
• For St Paul's (a ≈ 17 m) and a whistled λ ≈ 0.17 m (ν ≈ 2 kHz), m ≈ 630 — hundreds of wavelengths strung around the ring. For a silica microsphere with a = 50 μm and λ = 1.55 μm, m ≈ 294.

Why the wave clings: the caustic and grazing reflection

The intuition lives in the ray picture. A ray launched at a shallow angle θ to the wall reflects with the same angle, and geometry guarantees each chord touches an inner circle — a caustic — of radius r_c = a·cos θ. No ray ever penetrates inside r_c; the whole family of reflections is confined to the annulus between the caustic and the wall. As θ → 0 (true grazing incidence), r_c → a and the wave is squeezed into an ever-thinner shell hard against the surface.

Crucially, at a concave wall the surface curves toward the ray between bounces, so a wave that starts tangent stays tangent — the geometry refocuses it rather than letting it escape, which is exactly why a flat or convex wall shows no such effect. In the wave picture, the field is oscillatory (a Bessel function) outside the caustic and evanescent (exponentially decaying) inside it. For light, the outer boundary supplies the second confinement: at grazing incidence the angle exceeds the critical angle θ_c = arcsin(1/n) ≈ 43.6° for silica (n = 1.45), so total internal reflection traps the beam with essentially perfect efficiency. A tiny evanescent tail does leak outside the glass — this is how light is coupled in and out with a tapered fibre.

Quality factor: how long the wave survives

The figure of merit is the dimensionless quality factor Q = ω·(energy stored)/(power lost) = ω·τ, where τ is the photon (or phonon) lifetime and ω = 2πν. Equivalently Q = ν/Δν, the ratio of resonant frequency to linewidth, and the wave oscillates roughly Q/(2π) times before decaying (the number of round trips is smaller by the mode index m, since each lap spans m wavelengths).

• For the acoustic gallery, absorption in air and imperfect reflection off stone cap Q at a few hundred to ~10³. A whisper survives a handful of laps — enough to circle a 100 m dome and be heard clearly, not enough to ring for seconds.
• For an optical microsphere, the loss is astonishingly low. Fused silica has an intrinsic absorption near 1.55 μm of order 0.2 dB/km, and a molten-then-cooled sphere has surface roughness of a few nanometres. Measured Q values reach 8 × 10⁹ in silica microspheres and above 10¹⁰–10¹¹ in crystalline CaF₂ or MgF₂ resonators.
• A Q of 10¹⁰ at ν = 194 THz (λ = 1.55 μm) means a photon lifetime τ = Q/ω ≈ 8 μs — during which light travelling at c/n ≈ 2 × 10⁸ m/s covers ≈ 1.6 km and circles a 50 μm sphere over 5 million times.

Total loss adds in reciprocal: 1/Q = 1/Q_abs + 1/Q_scatt + 1/Q_rad + 1/Q_coupling. The radiative (tunnelling) loss Q_rad grows exponentially with size, so even a modest sphere makes bending loss utterly negligible; real-world Q is set by surface scattering and water absorbed onto the glass.

Controlling variables and the mode zoo

A whispering gallery resonator is not one mode but a dense family, labelled by three integers:

• Azimuthal m — wavelengths around the equator, m ≈ 2πan/λ; sets the resonant frequency. Adjacent m differ in frequency by the free spectral range, FSR = c/(2πan) — about 660 GHz for a 50 μm silica sphere, and roughly 5 nm in wavelength at 1.55 μm.
• Radial q — number of intensity maxima in the radial direction; q = 1 modes hug the surface most tightly and radiate least.
• Polar ℓ − |m| — how the field spreads north–south of the equator on a sphere; the fundamental sits in a thin equatorial belt.

The key scaling levers: making the resonator bigger raises m and shrinks FSR while pushing Q up (less bending loss); smoother surfaces cut scattering loss; and the refractive-index contrast n sets both the critical angle and the evanescent decay length. Because the resonant frequency depends on both a and n, whispering gallery resonators are exquisitely sensitive transducers: a temperature change shifts n and a, a single virus binding to the surface perturbs the evanescent field, and a picometre wavelength shift is readily resolved thanks to the sub-MHz linewidth.

From cathedrals to photonics: where the effect is used

Architectural acoustics. The famous galleries — St Paul's (London), the Gol Gumbaz mausoleum in Bijapur (India, ⌀ 38 m, said to echo eleven times), the Temple of Heaven's Echo Wall in Beijing, and the Whispering Gallery in Grand Central Terminal's tiled vault — are all curved masonry surfaces that channel speech along the wall. The effect is a nuisance in concert-hall design, where a domed ceiling can spray sound unevenly around the perimeter.

Photonics and metrology. Optical whispering gallery resonators — microspheres, microtoroids, microdisks and crystalline rings — are among the highest-Q optical cavities ever made and now anchor a real technology stack:

• Frequency combs (microcombs): pump a nonlinear microresonator hard and its modes lock into a spectrum of thousands of evenly spaced lines — a self-referenced optical ruler on a chip, used for optical clocks, spectroscopy and coherent telecom.
• Ultra-narrow lasers: Er- or rare-earth-doped microspheres lase with microwatt thresholds and kilohertz linewidths.
• Biosensing: label-free detection down to single nanoparticles and viruses by watching the resonance shift or split as a particle enters the evanescent field.
• Cavity optomechanics: the mode's radiation pressure couples to the resonator's mechanical vibrations, enabling ground-state cooling of micromechanical motion.

Subtleties and common misconceptions

It is not focusing, and not a simple echo. The single most common error is to imagine the dome reflecting sound to a focus across the room. Whispering gallery modes carry energy along the wall in a thin shell; the centre of the dome is nearly silent. That is why you must speak into the wall and listen at the wall.

Convex won't do — concavity is essential. Only a wall that curves toward the wave recaptures it each bounce. A pillar (convex) scatters sound outward; a bowl (concave) holds it in.

The modes are running waves, not standing beams. A single travelling mode e^imφ circulates one way. Two counter-propagating modes (say +m and −m, degenerate on a perfect sphere) can be split into a standing-wave doublet by any surface defect that back-scatters light — the observable mode splitting that biosensors exploit to count particles.

There is always a leak. Even below the critical angle, an evanescent field extends a fraction of a wavelength outside the surface, and a curved boundary allows slow tunnelling of energy to infinity (radiative loss). Perfect trapping is impossible; the art is making every other loss channel smaller. And the whispering-gallery idea is universal — the same grazing-confinement physics governs seismic waves circling the Earth, microwaves in metallic cavities, and electrons in curved quantum wells.

Acoustic vs. optical whispering gallery modes: same geometry, two length scales.

Property	Acoustic (dome gallery)	Optical (silica microsphere)
Wave type	Longitudinal pressure wave	Transverse EM wave
Confinement	Reflection off rigid concave wall	Total internal reflection at glass–air
Wave speed	c ≈ 343 m/s (air)	c/n ≈ 2.06 × 10⁸ m/s (n ≈ 1.45)
Typical radius a	≈ 17 m	≈ 20–100 μm
Mode index m	hundreds (audio λ ≈ 0.3 m)	≈ 100–500 (λ ≈ 1.5 μm)
Quality factor Q	≈ 10²–10³	10⁹–10¹¹