Cutoff Frequencies and Dispersion in Waveguides

The rectangular-waveguide article ended with a ladder of numbers — cutoff frequencies, one per mode — and a promissory note: that the ladder's dynamics would get its own article. This is that article, and it closes the electromagnetic-waves chapter with the strangest honest physics in it. An empty metal pipe, containing nothing but vacuum, behaves as a dispersive medium: below a threshold frequency it transmits nothing at all — not weakly, nothing, with fields dying exponentially at calculable decibels per centimeter — while above threshold it propagates waves whose crests outrun light and whose energy crawls behind it, the two speeds locked by the exact reciprocity $v_p v_g = c^2$ . None of this requires new field theory. The dispersion relation $\omega^2 = \omega_c^2 + c^2k^2$ — a hyperbola where vacuum has a straight line — contains everything, and this article spends it three ways. First, the velocities: phase above $c$ , group below, their product pinned, with the phase-and-group-velocity article's taxonomy (recalled by name) finally getting its promised waveguide instance. Second, the picture that makes it all obvious: the guided mode is a plane wave zigzagging between the walls; the phase pattern racing along the axis is a projection effect (the scissors-intersection illusion, made rigorous), the energy physically travels the longer zigzag path at exactly $c$ , and cutoff is the angle going perpendicular — the wave bouncing across the pipe, going nowhere. Third, the forbidden zone below cutoff, where the axial wavenumber turns imaginary and the guide becomes a calculable exponential attenuator — so calculable that below-cutoff pipes are the primary standard by which attenuation itself is defined. There is also a joke the universe seems to be in on: the hyperbola is precisely the dispersion relation of a relativistic massive particle. Confine a massless photon and it acquires an effective mass $\hbar\omega_c/c^2$ — a working preview of how mass arises from confinement, gauge fields, and gaps everywhere in modern physics.

Key takeaways

One relation runs the article: $\omega^2 = \omega_c^2 + c^2k^2$ per mode — a hyperbola asymptotic to the light line, with a gap $\omega_c$ (the rectangular article's ladder, recalled) below which no real $k$ exists.
Velocities from the hyperbola: $v_p = c/\sqrt{1 - \omega_c^2/\omega^2} > c$ (chord) and $v_g = c\sqrt{1 - \omega_c^2/\omega^2} < c$ (tangent), with $v_pv_g = c^2$ exactly — the promised waveguide instance of the velocity-taxonomy article. Guide wavelength $\lambda_g = \lambda_0/\sqrt{1 - \omega_c^2/\omega^2}$ stretches accordingly.
The zigzag theorem: TE $_{10}$ decomposes exactly into two plane waves bouncing at angle $\cos\theta = kc/\omega$ off the walls. Phase speed above $c$ is the projection of oblique wavefronts onto the axis (nothing travels with it); energy zigzags at $c$ , netting $v_g = c\cos\theta$ along the guide. Cutoff = perpendicular bounce.
Below cutoff: $k = i\kappa$ with $\kappa = \sqrt{\omega_c^2 - \omega^2}/c$ — evanescent, zero average power flow, attenuation of tens of dB per centimeter (WR-90 at 5 GHz: 7.7 dB/cm). Calculable from geometry alone: the basis of piston attenuators, the metrology standard for attenuation.
The massive-photon reading: the hyperbola is $E^2 = (pc)^2 + (mc^2)^2$ with $m = \hbar\omega_c/c^2$ ; confinement gives the photon a mass ( $5\times10^{-41}$ kg in WR-90), and $v_pv_g = c^2$ is the relativistic identity in disguise.
Waveguide dispersion spreads pulses (GVD diverges near cutoff): the quantitative reason single-mode bands carry guard margins, and a working instance of the group-delay budgets radar and telecom live by.
KEY IDEA: a hollow guide is vacuum with a mass gap — dispersion $\omega^2 = \omega_c^2 + c^2k^2$ makes phase race ( $> c$ ) and energy crawl ( $< c$ ) with $v_pv_g = c^2$ , turns the below-gap regime into an exact exponential attenuator, and is explained entirely by a plane wave bouncing down the pipe at an angle.

The core insight

Each waveguide mode obeys $\omega^2 = \omega_c^2 + c^2k^2$ : above cutoff it propagates with $v_p = c/\sqrt{1-\omega_c^2/\omega^2}$ and $v_g = c\sqrt{1-\omega_c^2/\omega^2}$ locked by $v_pv_g = c^2$ — the bouncing-plane-wave geometry makes both obvious — while below cutoff the axial wavenumber is imaginary and the mode is an exponentially calculable evanescent attenuator.

Where this lives today

The hyperbola is working infrastructure at every scale. Accelerator RF systems live on its velocity structure: disc-loaded traveling-wave structures deliberately slow the phase velocity to match particle speeds (bare guides, with $v_p > c$ always, can never surf a particle — the entire technology of iris-loaded structures exists to fix this article's chord), while the group velocity sets cavity fill times that gate every pulse. Telecommunications squeezed below-cutoff physics into filters and multiplexers; EMC engineering runs on cutoff vents and choke flanges; and the band-edge dispersion penalty is a live constraint in the millimeter-wave backhaul links now carrying mobile traffic. The massive-photon reading graduated from analogy to research tool. Plasma physics is this dispersion relation with $\omega_p$ as the gap — fusion diagnostics time microwave pulses through plasma exactly as worked example 1 timed the radar run. In circuit QED, microwave photons in engineered waveguide structures acquire designer dispersion (band gaps, slow-light regions, negative- $v_g$ bands) that sets how superconducting qubits talk to each other. And the conceptual export runs deepest: "confinement opens a gap, and the gap acts as a mass" is the pattern behind photonic-crystal band gaps, the Meissner effect's massive photon in superconductors, and — in its grandest form — mass generation in gauge theory. The pipe on the radar bench is the tabletop member of that family.

Origins · 1 minLéon Brillouin and how this was discovered

Léon Brillouin

1889–1969 · Paris and New York

Brillouin is the physicist of wave-in-structure problems — the man whose name marks the zones, the scattering, and the velocity analysis that between them cover most of what waves do in confined and periodic media. Born into French physics royalty (his father Marcel and grandfather Éleuthère both held chairs), he wrote his 1914 analysis of signal propagation in dispersive media with Sommerfeld — the precursor work that settled how relativity survives anomalous dispersion, and the direct ancestor of this chapter's velocity honesty — days before being drafted into the war that delayed its fame. His 1926 WKB method (with Wentzel and Kramers) turned evanescent-zone tunneling into a calculational tool; his 1930 Brillouin zones organized every wave-in-lattice problem from electrons in crystals to photonic band structure — the periodic generalization of this article's mode ladder; Brillouin light scattering probes materials with the acoustic waves he analyzed; and his wartime move to America (fleeing Vichy France, where he had briefly and miserably run state broadcasting) produced waveguide and traveling-wave-tube work that fed directly into radar practice. His late-career campaign — Science and Information Theory (1956) — attacked Maxwell's demon with negentropy bookkeeping, helping weld thermodynamics to information decades before that fusion became fashionable. Few physicists have equipped so many different fields with their working vocabulary.

History

1893

— J.J. Thomson's analysis of oscillations in conductors plants the guided-wave seed

1897

— Rayleigh solves the hollow-tube modes: cutoffs and the dispersion relation, forty years early

1914

— Sommerfeld and Brillouin dissect dispersive signal propagation: fronts at c, group velocity dethroned in anomalous bands

1936

— Southworth and Barrow demonstrate practical hollow-pipe transmission; the hyperbola meets hardware

1940

s — Radar engineering builds the band-edge guard rails into the WR standards; piston attenuators define the dB

1946

— Traveling-wave tubes and loaded structures engineer the phase velocity below c to surf electrons

1960

s — Plasma physics adopts the identical gapped dispersion with the plasma frequency as cutoff

2000

s — Engineered waveguide lattices give microwave photons designer dispersion for quantum hardware

The physical picture

Start with the picture, because it converts every formula in this article into geometry. Take the TE $_{10}$ mode of the rectangular article (recalled by name — its field map, not re-derived) and notice what its cross-guide structure is: a standing wave, $\sin(\pi x/a)$ , between the side walls. Every standing wave is two traveling waves in disguise. Decompose (Derivation B does it exactly): the mode is a single plane wave — ordinary, vacuum-speed, obeying $\omega = c|\mathbf{k}|$ — bouncing between the walls at angle $\theta$ from the guide axis, its reflections off the perfect walls interfering into the mode pattern. All the "strange" dispersion is then projection geometry. The wavefronts are tilted by $\theta$ ; where they intersect the axis, the intersection point slides faster than the wave moves — the scissors effect: close a pair of scissors and the crossing point outruns both blades — giving $v_p = c/\cos\theta > c$ with nothing physical traveling at that speed. The energy, meanwhile, physically zigzags at exactly $c$ but progresses along the axis only as $v_g = c\cos\theta < c$ . Multiply: $v_pv_g = c^2$ , before any algebra. Frequency turns the zigzag dial. High above cutoff, the transverse wavenumber $\pi/a$ is a small fraction of the total $\omega/c$ : shallow bounce, $\theta \to 0$ , both velocities $\to c$ , the mode barely notices the walls. Lower the frequency and the fixed transverse structure eats a growing share of the wavenumber budget (the guided-waves article's constraint $\gamma^2 + k^2 = \omega^2/c^2$ , recalled): the bounce steepens. At $\omega = \omega_c$ exactly, $\theta = 90°$ — the wave bounces perpendicular to the axis, a pure standing wave rattling across the pipe, transporting nothing: $v_g = 0$ , $\lambda_g \to \infty$ , and the "propagating mode" has become a transverse resonance. Below that, the geometry has no real angle to offer, and the mathematics answers with an imaginary axial wavenumber: fields that decay exponentially down the pipe, storing reactive energy near the source but carrying no average power — the same evanescent physics as total internal reflection's forbidden zone (the Brewster/TIR article, recalled; the FAQ keeps the chapter's three decay lengths distinct). The deepest reading is the massive-particle one. Multiply the dispersion relation by $\hbar^2$ : $(\hbar\omega)^2 = (\hbar\omega_c)^2 + (\hbar k c)^2$ is exactly $E^2 = (mc^2)^2 + (pc)^2$ with rest energy $mc^2 = \hbar\omega_c$ . A photon in a guide behaves in every kinematic respect like a relativistic particle of mass $\hbar\omega_c/c^2$ : it cannot exist below its rest energy (cutoff), its group velocity is the relativistic particle velocity ( $v_g = pc^2/E$ ), and $v_pv_g = c^2$ is the textbook relativistic identity between phase and particle velocity of a de Broglie wave. Confinement manufactured a mass gap — a preview, at bench scale, of how gaps and masses arise from confinement and symmetry breaking across modern physics, and the reason plasma physicists, accelerator designers, and particle theorists all recognize this chapter's hyperbola on sight.

Figure

The zigzag dial, computed for WR-90 — shallow bounce far above cutoff, 41 degrees at 10 gigahertz, perpendicular rattle at cutoff, and no real angle below

The dispersion relation, spent

Per mode, with $\omega_c$ from the rectangular article's ladder (recalled, never re-solved):

\omega^2 = \omega_c^2 + c^2k^2 \qquad\Longleftrightarrow\qquad k(\omega) = \frac{1}{c}\sqrt{\omega^2 - \omega_c^2}, \qquad (1)

where: the hyperbola replaces vacuum's straight line $\omega = ck$ (the wave-equation article's reference dispersion): asymptotic to the light line at high frequency, gapped by $\omega_c$ at low. Everything below is a reading of (1). The velocities (Derivation A):

v_p = \frac{\omega}{k} = \frac{c}{\sqrt{1 - \omega_c^2/\omega^2}}, \qquad v_g = \frac{d\omega}{dk} = c\,\sqrt{1 - \frac{\omega_c^2}{\omega^2}}, \qquad v_p\,v_g = c^2, \qquad (2)

where: the chord exceeds $c$ , the tangent stays below, and the product is exact at every frequency — the hyperbolic-family reciprocity the velocity-taxonomy article flagged and deferred here by name. The same square root stretches the axial wavelength:

\lambda_g = \frac{2\pi}{k} = \frac{\lambda_0}{\sqrt{1 - \omega_c^2/\omega^2}}, \qquad (3)

where: $\lambda_g$ is the crest spacing a slotted-line probe actually measures — 3.97 cm at 10 GHz in WR-90 against the free-space 3.0 cm, the number the rectangular article computed as bookkeeping and this one explains: tilted wavefronts intersect the axis at stretched intervals. Below the gap (Derivation C), no real $k$ exists and the mode evanesces:

k = i\kappa, \qquad \kappa = \frac{1}{c}\sqrt{\omega_c^2 - \omega^2}, \qquad \text{attenuation} = 8.686\,\kappa\ \ \text{dB/m}, \qquad (4)

where: the fields fall as $e^{-\kappa z}$ carrying zero time-averaged power (the same conjugate-pair Poynting audit as TIR's forbidden zone), and the numbers are ferocious: WR-90 driven at 5 GHz attenuates at 7.7 dB/cm — five centimeters buys 38 dB, and 100 dB fits in 13 cm of empty pipe. Because (4) contains only geometry and frequency — no material properties at all — a below-cutoff tube is an absolutely calculable attenuator, which is why piston attenuators (a sliding probe in a cut-off circular guide) served for decades as the primary standard defining the decibel scale of RF attenuation. Finally the massive-particle dictionary, exact by inspection of (1):

(\hbar\omega)^2 = (\hbar\omega_c)^2 + (\hbar k\,c)^2 \quad\equiv\quad E^2 = (mc^2)^2 + (pc)^2, \qquad m = \frac{\hbar\omega_c}{c^2}, \qquad (5)

where: the guided photon's effective rest mass in WR-90 is $\hbar\omega_c/c^2 = 4.8\times10^{-41}$ kg ( $3\times10^{-5}$ eV — five orders below an electron), its group velocity is the relativistic $pc^2/E$ , and cutoff is the statement that a particle cannot be realized below its rest energy. The dictionary is not decoration: plasma waves obey the same hyperbola with $\omega_p$ as the gap (the absorption article's Drude limit), and the velocity-taxonomy article's pulsar example was this identical mathematics with the galaxy as the waveguide.

Figure

The dispersion diagram of WR-90, computed — one gapped hyperbola per mode, all asymptotic to the light line, with the forbidden zone and single-mode band visible at a glance

Derivation AVelocities from the hyperbola — the reciprocity, the limits, and the guide wavelength

The goal is the complete velocity package (2)–(3) from the dispersion relation (1), with the limiting behavior at both ends read honestly — delivering the waveguide instance the phase-and-group-velocity article deferred here by name. The strategy is one differentiation and disciplined limit-taking; the physics is in the readings.

Step 1. Phase velocity by division. From (1), $k = \sqrt{\omega^2 - \omega_c^2}/c < \omega/c$ always: the axial wavenumber is deficient relative to vacuum (part of the budget went transverse), so

v_p = \frac{\omega}{k} = \frac{c}{\sqrt{1 - \omega_c^2/\omega^2}} > c \quad\text{at every frequency above cutoff}.

Why no alarm is warranted: the velocity-taxonomy article's verdict, recalled — phase patterns carry no information, and this particular $v_p > c$ is literally a projection artifact (Derivation B). The measurable consequence is benign and useful: stretched $\lambda_g$ , equation (3), which is how slotted-line metrology infers frequency.

Step 2. Group velocity by differentiation. Differentiate $\omega^2 = \omega_c^2 + c^2k^2$ : $2\omega\,d\omega = 2c^2k\,dk$ , so

v_g = \frac{d\omega}{dk} = \frac{c^2k}{\omega} = c\sqrt{1 - \frac{\omega_c^2}{\omega^2}} < c,

and the product $v_pv_g = (\omega/k)(c^2k/\omega) = c^2$ drops out in one line — exact, frequency-independent, the hyperbolic family's signature. Why the group velocity is the honest one here: the guide is lossless and dispersive-but-transparent, exactly the jurisdiction where the taxonomy article's energy-transport theorem applies; a direct Poynting-over-energy audit of the TE₁₀ fields (the rectangular article's power integral divided by its stored energy per length) reproduces $c^2k/\omega$ , closing the loop.

Step 3. The two ends of the band. Far above cutoff ( $\omega \gg \omega_c$ ): expand — $v_g \approx c(1 - \omega_c^2/2\omega^2)$ , $v_p \approx c(1 + \omega_c^2/2\omega^2)$ : both approach $c$ symmetrically, the mode forgetting its confinement (shallow zigzag). At cutoff ( $\omega \to \omega_c^+$ ): $v_g \to 0$ , $v_p \to \infty$ , $\lambda_g \to \infty$ — infinite crest spacing because the wavefronts lie parallel to the axis, sweeping it everywhere at once; zero delivery because the bounce is perpendicular. Numbers for WR-90 at 10 GHz: obliquity $\sqrt{1 - (6.557/10)^2} = 0.755$ , so $v_p = 3.97\times10^8$ m/s, $v_g = 2.26\times10^8$ m/s, product $c^2$ ✓, $\lambda_g = 3.97$ cm.

Step 4. The dispersion penalty. Because $v_g$ depends on frequency, pulses spread: differentiating again, the GVD parameter is $\beta_2 = d^2k/d\omega^2 = -\omega_c^2/[c\,(\omega^2 - \omega_c^2)^{3/2}]$ , which diverges at cutoff. WR-90 at 10 GHz: $\beta_2 \approx -5.3\times10^{-20}\ \text{s}^2/\text{m}$ — a 100 ps pulse's dispersion length $T^2/|\beta_2|$ is only 19 cm; even a 1 ns pulse spreads meaningfully in a 19 m run. This is orders of magnitude harsher than optical fiber (the taxonomy article's 4.6 km for 10 ps) and is the quantitative content of the rectangular article's warning to stay 25% above cutoff: waveguide dispersion, not loss, sets the fidelity floor near the band edge. Verified: (2)–(3) match Griffiths §9.5.2 (Eqs. 9.192–9.197) and Jackson §8.3; the energy-velocity audit matches the standard $v_E = c^2k/\omega$ result for guide modes (Jackson Eq. 8.52-family); WR-90 numbers recomputed (0.755, 3.97/2.26 ×10⁸, λ_g = 3.97 cm); $\beta_2$ formula by direct differentiation, checked at 10 GHz = −5.3×10⁴ ps²/m.

Go deeper · Derivation B · 2 minThe zigzag theorem — the mode is a bouncing plane wave, exactly

Derivation BThe zigzag theorem — the mode is a bouncing plane wave, exactly

The goal is the chapter's best picture made rigorous: the TE $_{10}$ mode is two vacuum plane waves reflecting off the side walls, with every dispersion result of Derivation A re-derived as projection geometry. This is not an analogy — the decomposition is an algebraic identity — and it welds this chapter's two halves (free waves and guided waves) into one object.

Step 1. Split the standing wave. The mode's field (rectangular article, recalled): $E_y = E_0\sin(\pi x/a)\,e^{i(kz - \omega t)}$ . Write the sine as exponentials:

E_y = \frac{E_0}{2i}\left[e^{i(\pi x/a + kz - \omega t)} - e^{i(-\pi x/a + kz - \omega t)}\right]

— exactly two plane waves with wavevectors $\mathbf{k}_\pm = (\pm\pi/a,\ 0,\ k)$ , equal amplitudes, opposite transverse components: an upward- and a downward-slanting wave, each reflecting into the other at the walls (the sign flip between them is the PEC reflection $r = -1$ of the conductor-reflection article, recalled).

Step 2. Check they are legal vacuum waves. Each has $|\mathbf{k}_\pm|^2 = (\pi/a)^2 + k^2 = \omega^2/c^2$ by the dispersion relation (1) — so each component obeys the vacuum $\omega = c|\mathbf{k}|$ exactly. The guide adds no new wave physics; it only selects a pair of ordinary waves whose interference respects the walls ( $E_y = 0$ at $x = 0, a$ — the sine's zeros, automatically). Why this is the fulcrum: every "anomalous" guide property must now be a property of tilted ordinary waves, i.e. geometry.

Step 3. Read the angle. Each component travels at angle $\theta$ from the axis with $\cos\theta = k/|\mathbf{k}| = kc/\omega = \sqrt{1 - \omega_c^2/\omega^2}$ and $\sin\theta = (\pi/a)/(\omega/c) = \omega_c/\omega$ . The frequency dial is explicit: $\omega \gg \omega_c$ gives grazing propagation ( $\theta \to 0$ ); $\omega = \omega_c$ gives $\sin\theta = 1$ — perpendicular bounce, the standing rattle; below, no real $\theta$ exists (the sine would exceed 1 — Snell's-wall logic, and indeed the same complexification as TIR's).

Step 4. Re-derive the velocities geometrically. Phase: the component wavefronts advance at $c$ along $\mathbf{k}_\pm$ ; their intersection with the axis moves at $c/\cos\theta = v_p$ — the scissors point, faster than either blade, carrying nothing. Group: energy rides the components at exactly $c$ but along the zigzag; the axial component of that velocity is $c\cos\theta = v_g$ . Product: $c^2$ , by $\cos\theta\cdot(1/\cos\theta)$ — the reciprocity is literally a trigonometric identity. Path-length reading of the group delay: a photon entering a length- $L$ guide travels $L/\cos\theta$ of actual path; at 10 GHz in WR-90 ( $\theta = 41°$ ) the zigzag adds 32% to every meter, which is precisely the 53 ns the 12 m run of worked example 1 clocks against free space's 40 ns.

Step 5. What the picture buys beyond this article. Wall losses: the bounce rate per meter grows as $\tan\theta/a$ , which is why attenuation climbs steeply toward cutoff (each bounce taxes the fields through the skin-depth article's $R_s$ ) — the loss half of the band's lower guard rail. Mode conversion at discontinuities: a bend tilts the walls mid-bounce, misdirecting the component waves — the geometric seed of the rectangular article's junction-reactance story. And the quantum reading: the two components are the $\pm$ transverse-momentum states of a confined photon, the mode their superposition — the particle-in-a-box stationary state, drawn in real space. Verified: decomposition is the elementary exponential identity; component legality checked against (1); angle forms $\cos\theta = kc/\omega$ , $\sin\theta = \omega_c/\omega$ match Griffiths §9.5.2's bouncing-wave discussion and Feynman II-24's treatment; path-length delay 1/0.755 = 1.32 consistent with worked example 1's 53 vs 40 ns.

Go deeper · Derivation C · 2 minBelow cutoff — evanescence, zero net power, and the attenuator you can machine

Derivation CBelow cutoff — evanescence, zero net power, and the attenuator you can machine

The goal is the forbidden zone (4) done honestly: the imaginary wavenumber, the proof that no average power flows, the decibel arithmetic of below-cutoff attenuation, and why this — uniquely among attenuation mechanisms — is calculable enough to serve as a primary standard. The strategy mirrors the TIR evanescent derivation (Brewster/TIR article, recalled): same mathematics, new geometry, sharper engineering payoff.

Step 1. The imaginary root. For $\omega < \omega_c$ , (1) gives $k^2 < 0$ : $k = i\kappa$ with $\kappa = \sqrt{\omega_c^2 - \omega^2}/c$ real. The field of a mode driven at the guide mouth is

E \propto e^{-\kappa z}\,e^{-i\omega t}

— no oscillation along $z$ at all (contrast the conductor article's $k = \kappa$ half-wave: here the propagating part is entirely gone), pure exponential collapse. The growing root is excluded by the same no-sources-at-infinity argument as always.

Step 2. Zero average power, by the standard conjugate audit. The transverse fields slave to the mode profile through factors of $k$ (guided-waves article): with $k$ purely imaginary, the transverse $E$ and $B$ acquire a 90° relative phase, and $\langle S_z\rangle = \tfrac{1}{2\mu_0}\mathrm{Re}[E_yB_x^*] \propto \mathrm{Re}(ik...) = 0$ . The stored energy near the mouth is reactive — sloshing at $2\omega$ , netting nothing — so a below-cutoff guide presents its source not with a resistance but with a reactance (an inductive one for TE modes), exactly how microwave engineers model it in junction equivalent circuits.

Step 3. The decibel arithmetic. Amplitude attenuation in dB per length is $20\log_{10}(e)\,\kappa = 8.686\,\kappa$ . Two WR-90 numbers: driven at 5 GHz, $\kappa = \tfrac{2\pi}{c}\sqrt{6.557^2 - 5^2}\times10^9 = 88.9\ \text{m}^{-1}$ → 7.7 dB/cm; driven at 6 GHz (only 8.5% below cutoff), still 4.8 dB/cm. The scale is set by $c/\omega_c \sim$ the guide width: any below-cutoff pipe kills signals within a few of its own diameters. This is the other half of why hollow-pipe systems are so quiet — everything but the intended band is not merely unamplified but exponentially executed — and why the oven-door choke grooves and the ventilation holes in shielded rooms are safe: apertures behave as tiny waveguides far below their cutoffs, attenuating leakage by this article's exponent (a 5 mm vent hole at 2.45 GHz is a guide with $f_c \sim 35$ GHz: dozens of dB per centimeter of depth).

Step 4. The metrology payoff. Look at what (4) doesn't contain: no conductivity, no permittivity, no material property — only $\omega$ , and $\omega_c$ , which is pure geometry (a diameter, measurable to microns). A "piston attenuator" — a receiving probe sliding inside a circular guide driven far below cutoff — therefore attenuates by a factor computable from first principles to parts per million per dB: move the piston a measured distance, and the signal drops by an exactly known number of decibels. For most of the twentieth century this was the primary realization of the attenuation scale in national standards laboratories: the decibel, defined by a machined tube and this derivation. Few formulas in engineering are trusted with that job. The validity fine print, honestly: the exactness holds for a single mode in a perfect geometry; real standards budget for higher evanescent modes near the launcher (they die even faster) and machining tolerance — both correctable, neither material-dependent. *Verified: $\kappa$ and the zero-flux audit match Griffiths §9.5.2's below-cutoff discussion and Jackson §8.3; WR-90 attenuations recomputed (88.9 m⁻¹ → 7.72 dB/cm at 5 GHz; 4.81 dB/cm at 6 GHz); the reactive-load reading matches standard waveguide equivalent-circuit practice (Pozar Ch. 4); piston-attenuator metrology per the classic NBS/NPL realizations of the dB scale.*

Limiting cases

Case	Result	Meaning
$\omega \gg \omega_c$	$v_p, v_g \to c$ ; $\theta \to 0$	grazing zigzag: the mode forgets the pipe
$\omega \to \omega_c^+$	$v_g \to 0$ , $v_p, \lambda_g \to \infty$	perpendicular rattle; nothing delivered
$\omega < \omega_c$	$k = i\kappa$ , $\langle S_z\rangle = 0$	evanescent; dB/cm from geometry alone
$\omega \to 0$ (deep sub-cutoff)	$\kappa \to \omega_c/c$	attenuation saturates: a few dB per guide-width
any $\omega > \omega_c$	$v_pv_g = c^2$	hyperbolic-family reciprocity, exact
$\hbar$ -dressed	$E^2 = (pc)^2 + (mc^2)^2$ , $m = \hbar\omega_c/c^2$	confinement mass: 5×10⁻⁴¹ kg in WR-90
near cutoff, pulses	$\beta_2 \to \infty$	GVD divergence: the band's lower guard rail
plasma ( $\omega_c \to \omega_p$ )	same hyperbola	ionosphere, pulsar delays — one family

Common trapBelow cutoff is not absorption, and near cutoff is not usable — the two edges of the band fail differently

Two distinct misreadings of the same square root. First: below-cutoff attenuation is not dissipation. Nothing absorbs the energy — the ideal guide is lossless — the mode simply cannot propagate, and the evanescent field near the source stores reactive energy that returns to the generator (which sees a reactance, not a resistance). Consequences: below-cutoff attenuators reflect rather than absorb (systems needing echo-free attenuation must pad them with matched lossy elements), and no amount of "more power" pushes a below-cutoff signal through — the exponent is set by geometry, not by field strength (until breakdown physics changes the problem entirely). Second: just above cutoff the mode propagates but is barely usable — group velocity collapsing, GVD diverging (a 100 ps pulse smears within 20 cm of WR-90 at 10 GHz... at band-edge frequencies far worse), wall losses climbing as the zigzag steepens, and the guide wavelength swelling so that every flange gap and probe becomes electrically significant. "Propagates" and "transmits faithfully" part company at the band edge — that is the real content of the 25% guard margin, and why band edges are specified conservatively in every waveguide catalog. Keep the failure modes straight: below cutoff the pipe is a mirror; just above, it is a funhouse.

Worked examples

1 · Radar plumbing timing — the zigzag toll (confidence builder)

A 10 GHz radar's antenna sits 12 m of WR-90 away from its receiver. (a) Find $v_p$ , $v_g$ , $\lambda_g$ , and the bounce angle. (b) Compute the signal delay through the run and compare with 12 m of free space and 12 m of RG-58 coax. (c) Which velocity does the radar's range calibration care about?

Step 1. The obliquity factor at 10 GHz: $\sqrt{1 - (6.557/10)^2} = 0.755$ . So $v_p = c/0.755 = \mathbf{3.97\times10^8}$ m/s, $v_g = 0.755c = \mathbf{2.26\times10^8}$ m/s (product: $c^2$ ✓), $\lambda_g = 3.0\ \text{cm}/0.755 = \mathbf{3.97\ cm}$ , and the zigzag angle $\theta = \arccos(0.755) = \mathbf{41°}$ .

Step 2. Delay through the run: energy travels at $v_g$ , so $t = 12/2.26\times10^8 = \mathbf{53.0\ ns}$ . Free space would take 40.0 ns — the 13 ns excess is the zigzag toll, exactly the extra path $L/\cos\theta$ of the bouncing photon. RG-58 coax ( $v = 2.0\times10^8$ m/s, the TEM article's number): 60.0 ns — the waveguide beats the cable here despite the zigzag, because coax pays the dielectric's $\sqrt{\varepsilon_r}$ toll at every frequency while the guide's toll shrinks with frequency headroom.

Step 3. Calibration: a radar converts echo time to range, so the plumbing's contribution uses the group delay — 53.0 ns ≈ 8 m of apparent range offset, calibrated out as a constant. But the phase-sensitive parts of the system (monopulse comparators, phased-array feeds) budget path lengths in $\lambda_g$ — phase velocity's jurisdiction. One waveguide run, both members of the velocity taxonomy on duty, neither optional: confuse them and the radar is miscalibrated by meters or mis-phased by radians.

2 · Designing with the forbidden zone (exam level)

(a) How much attenuation does a 5 cm plug of WR-90 provide to an interfering 5 GHz signal? (b) A shielded-room vent is an array of 5 mm × 5 mm × 30 mm deep channels; estimate its leakage attenuation at 2.45 GHz. (c) Why can these numbers be trusted without knowing what the walls are made of?

Step 1. WR-90 at 5 GHz: $\kappa = \tfrac{2\pi}{c}\sqrt{(6.557)^2 - (5)^2}\times10^9 = 88.9\ \text{m}^{-1}$ , i.e. $8.686\times88.9/100 = 7.72$ dB/cm. Five centimeters: $\mathbf{39\ dB}$ — a thousandfold power rejection from a hand-width of empty pipe, while the 8.2–12.4 GHz band sails through untouched. This frequency-selective wall is the mode-filter half of every waveguide diplexer.

Step 2. The vent channel is a square waveguide with $a = 5$ mm: cutoff $f_c = c/2a = 30$ GHz — the 2.45 GHz interferer sits twelve times below cutoff, deep in the saturated regime where $\kappa \approx \omega_c/c = 2\pi f_c/c$ (the drive frequency barely matters). $\kappa \approx 628\ \text{m}^{-1}$ → 54.6 dB/cm... computing honestly: $\kappa = \tfrac{2\pi}{c}\sqrt{30^2 - 2.45^2}\times10^9 = 626\ \text{m}^{-1}$ , so 5.4 dB/mm, and the 30 mm depth delivers $\mathbf{\sim160\ dB}$ — total annihilation, with airflow intact. This "waveguide-below-cutoff vent" is standard EMC practice: honeycomb panels on every shielded enclosure are thousands of these channels in parallel, each executing this arithmetic.

Step 3. Trust: equation (4) contains frequency and geometry — nothing else. No conductivity (the walls only need to be good enough conductors to enforce the boundary conditions, and the skin-depth article showed any metal is, overwhelmingly, at these frequencies), no permittivity, no calibration factor. That material-blindness is why the piston attenuator could define the decibel scale, and why an EMC engineer quotes a vent's attenuation from a drawing before anything is built. The exponential is as trustworthy as the machining — a rare and precious situation in engineering, and the reason below-cutoff design feels less like attenuation and more like arithmetic.

Going further

References. Griffiths, Introduction to Electrodynamics, 4th ed., §9.5.2 (Eqs. 9.192–9.197) · Jackson, Classical Electrodynamics, 3rd ed., §8.3–8.4 · Zangwill, Modern Electrodynamics, Ch. 19 · Pozar, Microwave Engineering, 4th ed., Ch. 3–4 · Feynman Lectures, Vol. II, Ch. 24

Provenance. Dispersion relation, velocities, and the $v_pv_g = c^2$ reciprocity cross-checked against Griffiths §9.5.2, Jackson §8.3, and Feynman II-24 (bouncing-wave decomposition); below-cutoff attenuation against Pozar Ch. 4 and standard piston-attenuator metrology practice. A verifier should: (1) derive $v_p$ , $v_g$ from $\omega^2 = \omega_c^2 + c^2k^2$ and confirm the product; (2) recompute the WR-90 numbers — obliquity 0.755 at 10 GHz, $v_p = 3.97\times10^8$ m/s, $v_g = 2.26\times10^8$ m/s, $\lambda_g = 3.97$ cm, $\theta = 41°$ ; (3) verify $\kappa = 88.9$ m⁻¹ → 7.72 dB/cm for WR-90 at 5 GHz and the vent-channel estimate (5 mm square, $f_c = 30$ GHz, ~5.4 dB/mm at 2.45 GHz); (4) check the effective mass $\hbar\omega_c/c^2 = 4.8\times10^{-41}$ kg and $\beta_2 = -\omega_c^2/[c(\omega^2 - \omega_c^2)^{3/2}] = -5.3\times10^4$ ps²/m at 10 GHz. Addendum for this edition: scope split executed — the mode catalog and cutoff ladder are recalled by name from Rectangular Waveguides (never re-solved), the framework from Guided Waves and Transmission Lines; this article owns the dispersion dynamics: velocities with the $c^2$ reciprocity (delivering the deferral from Phase and Group Velocity), the exact zigzag decomposition with its geometric re-derivations, below-cutoff evanescence with the metrology story, the GVD/band-edge analysis, and the massive-photon dictionary. E8 chapter closes here; the potentials-and-radiation chapter opens next.

Builds on: Rectangular Waveguides · Guided Waves and Transmission Lines · Phase and Group Velocity · The Electromagnetic Wave Equation

Leads to: Scalar and Vector Potentials · Intensity and Time-Averaged Quantities · Momentum and Radiation Pressure

What does this article own versus the rest of the waveguide trio?

The closing share: the dynamics of the eigenvalues the other two produced. The framework article proved modes and cutoffs must exist and owned the TEM/transmission-line world; the rectangular article solved the box — eigenfunctions, the cutoff ladder, TE₁₀'s fields and currents, power ratings. This article takes each mode's $\omega_c$ as input and derives what it does: the hyperbolic dispersion relation and its velocities ( $v_pv_g = c^2$ , delivered here as promised by the velocity-taxonomy article), the exact bouncing-plane-wave decomposition and its geometric readings, evanescence below cutoff with the attenuator arithmetic, band-edge dispersion penalties, and the massive-photon dictionary. If you need a cutoff value, read one back; if you need why the guide behaves as a gapped vacuum at that cutoff, you are home.

The chapter now has three different exponential decays — TIR's evanescent wave, the conductor's skin depth, and below-cutoff modes. Are they the same thing?

Same mathematics, three distinct physical reasons — worth keeping firmly separated. TIR evanescence (Brewster/TIR article): lossless geometry problem — the tangential wavenumber demanded by phase matching exceeds what medium 2 can supply at that frequency; energy sloshes but nets zero; nothing dissipates. Skin depth (conductor articles): a dissipative decay — the medium absorbs; $k$ and $\kappa$ are equal (half wave, half collapse) and Joule heat is generated. Below-cutoff decay (here): lossless again, and even purer than TIR — the axial wavenumber is entirely imaginary (no oscillation along the decay direction at all); the guide geometry simply offers no propagating channel, and the stored field is reactive. The diagnostic questions: Is energy dissipated? (only the conductor case.) Does the field oscillate along the decay direction? (TIR: yes, along the surface; conductor: yes, one radian per e-fold; cutoff: no.) What sets the decay length? (TIR: angle excess; conductor: $\sigma\omega$ ; cutoff: pure geometry.) Three decays, one exponential family, three different masters.

If the group velocity goes to zero at cutoff, does a signal at exactly the cutoff frequency ever arrive?

In the idealization, no — and the idealization fails instructively. Exactly at $\omega_c$ the mode is a transverse standing resonance with $v_g = 0$ : energy rattles across the pipe and makes no axial progress. But "exactly at cutoff" is a measure-zero fiction for any real signal: a pulse has bandwidth, and its spectral content above $\omega_c$ propagates (slowly, dispersively) while content below evanesces. What actually arrives from a near-cutoff pulse is a badly mangled, long-delayed, low-frequency-filtered remnant — the waveguide's version of the Sommerfeld–Brillouin transient analysis (velocity-taxonomy article), and a vivid instance of why band edges are unusable rather than merely slow. There is also a genuine research corner here: engineered "slow-light" waveguide structures deliberately operate near photonic band edges to exploit the diverging light–matter interaction time, accepting the dispersion penalty this article quantifies — the band edge as a feature, bought with exactly the $\beta_2$ divergence that plumbing avoids.

Is the effective photon mass in a waveguide physically real or just a formal analogy?

Operationally real by every kinematic test, with one honest caveat about what "mass" means. The guided photon cannot exist below its rest energy $\hbar\omega_c$ (pair-production threshold analog: cutoff), its energy–momentum relation is exactly relativistic with $m = \hbar\omega_c/c^2$ , its group velocity obeys the massive-particle $v = pc^2/E$ , and if you modulate the guide adiabatically the mode responds like a particle with that inertia. The caveat: this is an effective mass of a confined excitation, not a Lorentz-invariant property of the photon field — step outside the guide and the photon is massless again; the "mass" lives in the boundary conditions, and a boosted observer sliding along the guide sees a different $\omega_c$ ... rather, the guide frame is preferred, breaking the full Lorentz story. That structure — masslessness in the bulk theory, effective mass from environment — is precisely the pattern of the Meissner effect (photons massive inside superconductors) and, with symmetry breaking supplying the "environment," of the Higgs mechanism. The waveguide is the honest tabletop member of the family: real enough to time with worked example 1, formal enough to keep the word "effective" attached.

Why do accelerators need slow-wave structures if waveguides already carry their power?

Because surfing requires matching phase velocity to the particle, and a bare guide's $v_p$ exceeds $c$ at every frequency — the wave's crests forever outrun any particle, delivering zero net acceleration (the particle sees an oscillating field averaging away). Bare rectangular guide is therefore perfect for transporting RF power (this chapter's job) and useless for coupling it to beams. The fix is periodic loading: irises, discs, or corrugations spaced along the structure diffract the mode into spatial harmonics, bending the dispersion curve so that one harmonic's phase velocity dips below $c$ and can be set equal to the (near- $c$ ) beam velocity — a particle riding that harmonic sees a constant accelerating field. The design mathematics is this article's hyperbola upgraded to a band structure — Brillouin's zones, fittingly, applied to Brillouin-adjacent hardware — and the same trick in reverse (beam faster than the wave) is the traveling-wave-tube amplifier in every satellite. The chapter's closing irony: the guide's "defect," $v_p > c$ , is exactly what a century of accelerator engineering exists to repair.

Does dielectric filling change the story, and why do engineers sometimes fill guides deliberately?

Filling with a dielectric $\varepsilon_r$ replaces $c \to c/\sqrt{\varepsilon_r}$ throughout: every cutoff drops by $\sqrt{\varepsilon_r}$ (the guided-waves article's budget with a slower light speed), the hyperbola compresses, and above cutoff both velocities scale down while $v_pv_g = c^2/\varepsilon_r$ keeps the reciprocity in the medium's light speed. The deliberate uses: miniaturization — a filled guide is $\sqrt{\varepsilon_r}$ smaller for the same band, which is the entire premise of substrate-integrated waveguides (the rectangular article's PCB-embedded guides, $\varepsilon_r \sim 3$ –10); pressure windows — thin dielectric plugs seal pressurized guide runs while passing the band; and dispersion engineering — partial (slab) loading bends the dispersion curve asymmetrically, a knob for matching or broadbanding. The costs: dielectric loss (loss tangent now taxes the whole field volume — the conductor articles' other loss channel), reduced power handling (breakdown fields in solids are volume-quality-limited, and any void arcs), and thermal limits. The vacuum guide remains the high-power king; the filled guide, the compact courtier — one more trade governed entirely by how $\omega_c$ and $c$ enter this article's single equation.