Phase and Group Velocity

Watch a group of ocean swells roll toward a beach and you can see, with the naked eye, one of the subtlest facts in wave physics: individual crests travel faster than the group they belong to. A crest is born at the back of the packet, climbs through it, and dies at the front — because for deep-water waves the crest speed is exactly twice the group speed. Nothing material does this; it is pure interference bookkeeping, and light in a dispersive medium plays the same game. The previous articles built the reason. A medium with $n(\omega)$ assigns each frequency its own crest speed $c/n(\omega)$ — the phase velocity — and any real signal, being a superposition of frequencies (a wave packet), moves at the speed its interference pattern reconstructs itself: the group velocity $v_g = d\omega/dk$ , the slope of the dispersion curve rather than the ratio $\omega/k$ . Where dispersion is gentle, $v_g$ is the honest speed of energy and information, and this article proves it — energy transport velocity equals group velocity in a transparent medium, damping included in the audit. Where dispersion is violent — inside the anomalous band of an absorption line, flagged by the absorption article and settled here — $v_g$ blows past $c$ , turns negative, and stops meaning anything kinematic at all, while relativity stands entirely unthreatened: the front of any signal moves at exactly $c$ , always, because no medium can respond faster than its electrons can be told the field has arrived. The taxonomy this article builds — phase, group, energy, signal, front — is permanent equipment. It prices pulse spreading in optical fibers, converts pulsar arrival-time smears into a survey of interstellar plasma (worked to the second below), explains the swell forecast, and stands guard over every "faster-than-light" press release.

Key takeaways

Phase velocity $v_p = \omega/k = c/n(\omega)$ is the crest speed of one infinite monochromatic wave — a ratio on the dispersion diagram. Group velocity $v_g = d\omega/dk$ is the slope — the speed of a packet's envelope, derived from two-wave beats and confirmed for full Gaussian packets.
Conversions: $v_g = \dfrac{c}{n + \omega\,dn/d\omega}$ (the group index $n_g = n + \omega\,dn/d\omega$ ); normal dispersion gives $v_g < v_p$ , and crests outrun their packet.
In a transparent dispersive medium, energy velocity equals group velocity: $\langle S\rangle/\langle u\rangle = v_g$ , provided $\langle u\rangle$ includes the energy parked in the medium's oscillators — the honest energy density $\tfrac{1}{4}\tfrac{d(\omega\varepsilon)}{d\omega}|E_0|^2 + \cdots$ .
Inside an anomalous-dispersion band $v_g$ exceeds $c$ or turns negative — and there it ceases to describe energy or signal transport (the packet reshapes; its peak is not a conserved object). Measured "superluminal group velocities" are real and carry no information faster than light.
The front velocity is exactly $c$ in every medium: a sharp front is built from $\omega \to \infty$ components, and every material responds as vacuum at infinite frequency ( $n \to 1$ ). Sommerfeld and Brillouin's precursors are the residue of this argument.
Second-order dispersion (GVD, $d^2k/d\omega^2$ ) spreads packets: the fiber-optics penalty, the pulsar astronomer's tool, the chirped-pulse amplifier's resource.
KEY IDEA: on the dispersion diagram, phase velocity is the chord $\omega/k$ and group velocity the tangent $d\omega/dk$ ; the tangent is the speed of envelopes and energy wherever the medium is transparent, and where it misbehaves, the invariant speed limit is carried by the front — always exactly $c$ .

The core insight

Phase velocity $\omega/k$ moves the crests; group velocity $d\omega/dk$ moves the envelope, and equals the energy-transport velocity in transparent media; where anomalous dispersion drives $v_g$ beyond $c$ it stops describing transport altogether, while the signal front travels at exactly $c$ in every medium.

Where this lives today

Group-velocity engineering is a working industry at both extremes. Slow light: electromagnetically induced transparency parks a laser on a hyper-steep normal-dispersion slope manufactured between absorption lines, reaching $n_g \sim 10^7$ — light at 17 m/s in a cold sodium cloud (Hau 1999), and "stopped light" when the slope is switched off with the pulse inside: the basis of quantum-memory prototypes, where a photon's state is parked in atomic coherence and released on demand. Fast light: gain-assisted anomalous dispersion gives the negative group velocities of Derivation C — now a precision test bed for the information-velocity theorems rather than a controversy. The workhorse applications are less exotic and larger. Fiber-optic networks are dispersion-managed systems: spans of opposite-sign $\beta_2$ concatenated so pulses breathe but arrive intact — millions of kilometers of deployed group-velocity budgeting. Chirped-pulse amplification (Nobel 2018) weaponizes GVD: stretch a femtosecond pulse $10^4\times$ with a dispersive grating pair, amplify safely, recompress — petawatt lasers exist because $\beta_2$ is reversible. Photonic-crystal and metamaterial design tailor $\omega(k)$ nearly at will (flat bands for slow light, near-zero-index media where phase velocity diverges while $v_g$ stays modest); and pulsar-timing arrays, de-dispersing across the Galaxy with example 1's formula, announced their gravitational-wave background evidence in 2023 — plasma dispersion subtracted at the microsecond level across decades of data.

Origins · 1 minWilliam Rowan Hamilton and how this was discovered

William Rowan Hamilton

1805–1865 · Dublin

Hamilton had already reformulated mechanics and optics into one variational structure — the characteristic function of 1833, from which "Hamiltonian" everything descends — when, in an 1839 paper on the propagation of light in crystals, he drew the distinction this article rests on: in a dispersive medium, the velocity of a wave's phase and the velocity of a group of waves differ, and he gave the group's velocity as $d\omega/dk$ in all but notation. The insight sat quietly (Stokes set it as a Cambridge exam problem in 1876; Rayleigh systematized it in The Theory of Sound and tied it to energy transport) — a familiar Hamilton pattern: he mined veins decades before the ore had uses. The rest of the biography is legend-adjacent: professor of astronomy at 22 while still an undergraduate; the 1843 flash of insight at Broom Bridge where he carved the quaternion relations into the stonework; a lifelong conviction that quaternions were his greatest work, history preferring instead the mechanics that became the skeleton of quantum theory — where his optical-mechanical analogy returned with interest, since the group velocity of a matter wave is the particle's velocity, a fact Schrödinger leaned on directly. The velocity taxonomy of this article is, fittingly, Hamiltonian optics doing what it always did: geometry on a dispersion surface, tangents distinguished from chords.

History

1839

— Hamilton distinguishes phase and group propagation in dispersive media — the tangent-versus-chord insight

1876

— Stokes poses the group-velocity relation as a Smith's Prize examination question

1877

— Rayleigh's Theory of Sound systematizes v_g = dω/dk and links it to energy transport

1914

— Sommerfeld and Brillouin dissect signal arrival in a Lorentz medium: precursors, signal velocity, front at exactly c

1966

— Kao's fiber proposal makes group-velocity dispersion an engineering budget line

1982

— Chu and Wong measure superluminal group velocities for pulses in absorbing media — reshaping, not signaling

1999

— Hau slows light to 17 m/s via electromagnetically induced transparency

2000

— Wang, Kuzmich and Dogariu measure v_g = −c/310 in gain-assisted cesium; the front still arrives at c

The physical picture

Everything lives on one diagram: the dispersion curve $\omega(k)$ . A single monochromatic wave is one point on the curve, and its phase velocity is the slope of the chord from the origin to that point, $\omega/k$ . A packet — any signal with finite extent — occupies a small neighborhood of the curve around its carrier $k_0$ , and its envelope moves at the slope of the tangent, $d\omega/dk|_{k_0}$ . Vacuum's dispersion is the straight line $\omega = ck$ : chord and tangent coincide everywhere, all speeds are $c$ , packets are rigid — the degenerate case that made the distinction invisible until media bent the line. In matter, the curve bows and the two slopes part company; every phenomenon in this article is geometry on that diagram. Why does the envelope move at the tangent slope? Because an envelope is an interference pattern. Superpose two waves of nearby frequencies: they agree (beat maximum) where their phases coincide, and the point of agreement moves at $\Delta\omega/\Delta k$ — the finite-difference tangent (Derivation A does this in three lines, then upgrades to a continuum of components by stationary phase). The carrier crests inside the envelope still move at $\omega_0/k_0$ ; when the two speeds differ, crests slide through the envelope — born at one end, dying at the other, exactly the ocean-swell spectacle ( $v_p = 2v_g$ for deep water, so crests overtake). Nothing physical travels with a crest; a crest is a phase label, and phase labels are cheap. The stakes are set by what does travel: energy and information. Derivation B shows the cycle-averaged Poynting flux divided by the cycle-averaged energy density — with the medium's mechanical energy honestly counted — is precisely $v_g$ in a transparent medium; that theorem is why "group velocity" appears on fiber datasheets rather than in philosophy seminars. And the boundary of its validity is patrolled by Derivation C: inside an absorption line, where $dn/d\omega < 0$ steeply, the tangent slope exceeds $c$ or goes negative. The resolution is not that the formula fails algebraically — measured pulse peaks in such media genuinely exit before a vacuum-racing peak would — but that a reshaping pulse's peak is not a messenger. The messenger is the front, the first moment anything can differ from zero, and the front is pinned at $c$ by the deepest fact in the subject: at $\omega \to \infty$ , every medium is vacuum, because inertia forbids instantaneous response.

Figure

Computed wave packet at two times — the envelope advances at the group velocity while carrier crests slide through it at twice the speed

Figure

The dispersion diagram is the whole subject — phase velocity is the chord from the origin, group velocity the tangent at the carrier, drawn for the cutoff family where the chord runs above c and the tangent below

The velocity taxonomy

For a wave $\cos(kz - \omega t)$ the crests move at

v_p = \frac{\omega}{k} = \frac{c}{n(\omega)}, \qquad (1)

where: $v_p$ is a statement about one infinite, eternal sinusoid — it carries no information because a pure sinusoid has no features to carry (its future is its past). Nothing prevents $v_p > c$ (X-rays in glass, radio above a plasma's $\omega_p$ , waveguide modes — three cases this track meets), and nothing physical accelerates when it happens. Any actual signal occupies a band of frequencies, and its envelope moves at

v_g = \frac{d\omega}{dk}\bigg|_{k_0} = \frac{c}{n + \omega\,\dfrac{dn}{d\omega}} \equiv \frac{c}{n_g}, \qquad (2)

where: the second form follows from differentiating $k = n(\omega)\,\omega/c$ , and $n_g$ — the group index — is the number laser and fiber engineers tabulate (silica at 1550 nm: $n = 1.444$ but $n_g = 1.462$ ; the 1.2% difference is measurable as nanoseconds per kilometer). Normal dispersion ( $dn/d\omega > 0$ , the transparent-valley rule of the absorption article) makes $n_g > n$ : envelopes lag crests. The useful equivalent forms — worth keeping on one line —

v_g = v_p + k\,\frac{dv_p}{dk} = v_p - \lambda\,\frac{dv_p}{d\lambda}, \qquad (3)

where: the last form (Rayleigh's) reads directly off measured $v_p(\lambda)$ curves and exposes the sign logic: if longer wavelengths travel faster ( $dv_p/d\lambda > 0$ , the usual case), the group lags the crests. Deep-water gravity waves are the extreme classroom case: $\omega = \sqrt{gk}$ gives $v_g = \tfrac{1}{2}v_p$ exactly — the crest-overtaking spectacle, and a reminder that this taxonomy belongs to all wave physics, not just light. Beyond the envelope's speed sits its shape: expanding $k(\omega)$ to second order,

k(\omega) \approx k_0 + \frac{\omega - \omega_0}{v_g} + \frac{1}{2}\beta_2\,(\omega - \omega_0)^2, \qquad \beta_2 \equiv \frac{d^2k}{d\omega^2}, \qquad (4)

where: $\beta_2$ — the group-velocity dispersion (GVD) — makes different slices of the packet's spectrum travel at slightly different group velocities, so the packet spreads and chirps (Derivation A, Step 4). Telecom's headline number, $D = 17$ ps/(nm·km) for silica at 1550 nm, is $\beta_2$ in engineering units: a pulse of 0.1 nm spectral width smears by 170 ps over 100 km — the effect that dispersion-managed fiber links exist to cancel, and that pulsar astronomy (worked example 1) runs in reverse as a measuring instrument. The remaining members of the taxonomy exist because $v_g$ has a jurisdiction. Energy velocity $v_E = \langle S\rangle/\langle u\rangle$ : equal to $v_g$ in transparent media (Derivation B) — that equality is $v_g$ 's license to mean something. Signal velocity: the arrival speed of a detectable feature; equal to $v_g$ in transparent media, detector-dependent inside absorption bands. Front velocity: the speed of the boundary between field and no-field — exactly $c$ , unconditionally (Derivation C). The hierarchy collapses to one number in vacuum and fans out in matter; knowing which number answers which question is the skill this article installs.

Derivation AFrom beats to packets — the group velocity, twice

The goal is $v_g = d\omega/dk$ derived at two levels of honesty: the two-wave beat argument (which contains the whole idea) and the full Gaussian packet through a dispersive medium (which certifies it and prices the spreading). The strategy for the packet is the standard Fourier pipeline: decompose, propagate each component with its own $e^{i(k(\omega)z - \omega t)}$ , expand $k(\omega)$ about the carrier, and resum order by order — watching which order of the expansion produces which phenomenon.

Step 1. Two waves, one lesson. Superpose equal amplitudes at $(k_0 \pm \tfrac{\Delta k}{2},\ \omega_0 \pm \tfrac{\Delta\omega}{2})$ :

\psi = 2\cos\!\left(\tfrac{\Delta k}{2}z - \tfrac{\Delta\omega}{2}t\right)\,\cos(k_0 z - \omega_0 t).

The second factor is the carrier, crests at $\omega_0/k_0$ ; the first is a slow envelope moving at $\Delta\omega/\Delta k \to d\omega/dk$ . Why the envelope speed is a derivative and not a ratio: the beat maximum sits where the two components are in phase, and maintaining that agreement as time advances requires matching phase changes, not phases — a difference quotient by construction.

Step 2. The honest packet. Take a Gaussian spectrum $A(\omega) \propto \exp[-(\omega-\omega_0)^2 T^2/2]$ (a pulse of duration $\sim T$ ), and propagate:

\psi(z,t) = \int A(\omega)\,e^{i(k(\omega)z - \omega t)}\,d\omega, \qquad k(\omega) = k_0 + \frac{\omega - \omega_0}{v_g} + \frac{\beta_2}{2}(\omega-\omega_0)^2 + \cdots

Step 3. Resum through first order (set $\beta_2 = 0$ temporarily). The integral becomes

\psi = e^{i(k_0 z - \omega_0 t)}\int A(\omega)\,e^{i(\omega - \omega_0)(z/v_g - t)}\,d\omega = e^{i(k_0z - \omega_0 t)}\;F\!\left(t - z/v_g\right):

a carrier at the phase velocity multiplied by an envelope $F$ that is a rigid function of $t - z/v_g$ — the packet translates at $v_g$ with unchanged shape. Why this is also the stationary-phase statement: for large $z$ the integral is dominated by the frequency where $d/d\omega[k(\omega)z - \omega t] = 0$ , i.e. $z/t = d\omega/dk$ — the observer moving at the group velocity sees the spectral neighborhood of the carrier conspire constructively; everyone else sees cancellation.

Step 4. Second order prices the spreading. Restoring $\beta_2$ , the Gaussian integral is still exact and yields a packet whose duration grows as

T(z) = T\sqrt{1 + \left(\frac{\beta_2 z}{T^2}\right)^2},

with a linear frequency sweep (chirp) across it — the fast and slow spectral slices physically separating. The characteristic dispersion length $L_D = T^2/|\beta_2|$ marks where spreading doubles the duration ( $\times\sqrt{2}$ ): for a 10 ps telecom pulse in silica ( $\beta_2 \approx -21.7\ \text{ps}^2/\text{km}$ ), $L_D \approx 4.6$ km; for a 10 femtosecond laser pulse in glass, millimeters — why ultrafast optics is a war fought against, and with, $\beta_2$ . Why shorter pulses spread faster: duration $T$ means bandwidth $1/T$ , and GVD's damage scales with bandwidth squared — halve the pulse, quadruple the spreading rate. *Verified: beat algebra is the product-to-sum identity; the Gaussian propagation reproduces Jackson §7.9 / standard fiber-optics results (Agrawal's $T_1 = T_0[1+(z/L_D)^2]^{1/2}$ ); silica numbers $\beta_2 = -21.7\ \text{ps}^2$ /km ↔ $D = +17$ ps/(nm·km) at 1550 nm cross-checked via $D = -2\pi c\beta_2/\lambda^2$ ; deep-water check $\omega = \sqrt{gk} \Rightarrow v_g = \tfrac{1}{2}\sqrt{g/k} = v_p/2$ ✓.*

Go deeper · Derivation B · 2 minEnergy velocity equals group velocity in a transparent medium

Derivation BEnergy velocity equals group velocity in a transparent medium

The goal is the theorem that gives $v_g$ its physical authority: in a lossless dispersive dielectric, the cycle-averaged energy flux divided by the cycle-averaged energy density is exactly $d\omega/dk$ . The subtlety — and the whole content — is that $\langle u\rangle$ must count the energy stored in the medium's oscillators, not just the fields; miss it and the ratio comes out wrong (it gives $v_p$ -flavored nonsense that can exceed $c$ in transparent regions where nothing of the sort happens). The strategy: compute flux and density for a plane wave in the lossless Lorentz medium, using the oscillator energetics recalled from the mechanics Resonance article.

Step 1. The cast. Monochromatic plane wave in a medium of real $n(\omega)$ (transparent window: $\gamma \to 0$ , $\omega$ away from $\omega_0$ ): $E = E_0\cos(kz - \omega t)$ , $B = (n/c)E_0\cos(kz - \omega t)$ , $k = n\omega/c$ . Flux: $\langle S\rangle = \langle EB\rangle/\mu_0 = \tfrac{n}{2\mu_0 c}E_0^2$ .

Step 2. Field energy density. $\langle u_{\rm field}\rangle = \tfrac{\varepsilon_0}{4}E_0^2 + \tfrac{1}{4\mu_0}\langle B^2\rangle\cdot 2 = \tfrac{\varepsilon_0}{4}(1 + n^2)E_0^2$ — deliberately not $\tfrac{\varepsilon_0}{2}n^2E_0^2$ : using the static-dielectric energy formula smuggles in exactly the medium term we are about to count honestly, once, ourselves.

Step 3. The medium's energy. Each Lorentz oscillator carries kinetic plus potential energy; with displacement amplitude $x_0 = (q/m)E_0/(\omega_0^2 - \omega^2)$ (Resonance article, $\gamma \to 0$ ),

\langle u_{\rm med}\rangle = N\left\langle \tfrac{m}{2}\dot{x}^2 + \tfrac{m\omega_0^2}{2}x^2 \right\rangle = \frac{Nm}{4}(\omega^2 + \omega_0^2)\,x_0^2 = \frac{\varepsilon_0 E_0^2}{4}\,\frac{\omega_p^2(\omega^2 + \omega_0^2)}{(\omega_0^2 - \omega^2)^2}.

Why this term is large exactly where dispersion is strong: near resonance the electrons swing hard, banking substantial energy — and $n(\omega)$ varies steeply there for the same reason. Energy storage in the medium and frequency-dependence of the index are the same physics; that is the moral of the theorem.

Step 4. Assemble and compare. With the Lorentz $n^2 = 1 + \omega_p^2/(\omega_0^2 - \omega^2)$ , the total $\langle u\rangle = \langle u_{\rm field}\rangle + \langle u_{\rm med}\rangle$ combines (a few lines of algebra with a common denominator) into the compact general form

\langle u\rangle = \frac{\varepsilon_0}{4}\left[\frac{d(\omega\,\varepsilon_r)}{d\omega} + n^2\right]E_0^2 = \frac{\varepsilon_0}{4}\left[n^2 + n\,\omega\frac{d(n^2)}{d\omega}/n... \right]E_0^2 \;=\; \frac{\varepsilon_0}{2}\,n\,n_g\,E_0^2,

where the middle expression is tidied by using $d(\omega\varepsilon_r)/d\omega = n^2 + 2n\omega\,dn/d\omega$ and $n_g = n + \omega\,dn/d\omega$ : the honest energy density is the geometric pairing of phase and group indices. Then

v_E = \frac{\langle S\rangle}{\langle u\rangle} = \frac{\tfrac{n}{2\mu_0c}E_0^2}{\tfrac{\varepsilon_0}{2}n\,n_g\,E_0^2} = \frac{c}{n_g} = v_g.

Step 5. Jurisdiction. The derivation used $\gamma \to 0$ and real $n$ — the transparent window. Inside an absorption band the oscillator energy bookkeeping acquires a dissipative channel, $\langle u\rangle$ is no longer even well-defined as "stored" energy without care, and the theorem lapses — precisely where $v_g$ goes rogue. The two failures are synchronized because they are the same failure: group velocity means energy speed exactly as far as energy is merely stored and returned. Verified: reproduces the Brillouin/Loudon result $v_E = v_g$ for lossless dispersive media (Jackson §7.8-adjacent; Zangwill §18.6); the general energy density $\tfrac{1}{4}[d(\omega\varepsilon)/d\omega]|E_0|^2 + \tfrac{1}{4\mu_0}|B_0|^2$ matches the Landau–Lifshitz form; algebra spot-checked at $\omega \ll \omega_0$ : $n_g \to n +$ small, $v_E \to c/n$ ✓, and near resonance $n_g$ grows large, $v_E$ small — slow light, as observed.

Go deeper · Derivation C · 3 minFaster than c without paradox — anomalous group velocity and the front theorem

Derivation CFaster than c without paradox — anomalous group velocity and the front theorem

The goal is the honest closure of the superluminal question the absorption article deferred: show why $v_g > c$ (and $v_g < 0$ ) occur, state what a packet actually does there, and prove the front velocity is exactly $c$ in every causal medium. The strategy for the front theorem is the high-frequency argument — short, rigorous in outline, and the reason relativity never needed defending.

Step 1. Where the tangent goes rogue. From (2), $v_g > c$ requires $n_g = n + \omega\,dn/d\omega < 1$ : strong negative slope. The absorption article's Lorentz profiles supply it: across a resonance, $n$ dives (anomalous dispersion) with slope $\propto 1/\gamma$ — arbitrarily steep for narrow lines. Near line center, $n_g$ can pass through zero and go negative: $v_g$ infinite, then negative. These are not algebra artifacts; they are the tangent slope of a real, measured curve.

Step 2. What a packet does there. The Derivation-A pipeline still runs — but its expansion converges only while the packet's bandwidth sits inside the anomalous window, and the medium absorbs and reshapes as it propagates. The measured phenomenology (Chu–Wong 1982 in absorbing media; Wang–Kuzmich–Dogariu 2000 in a gain-assisted "transparent" anomalous medium, $v_g = -c/310$ ): the output pulse's peak exits before the peak of a vacuum-racing reference — even "before the input peak enters," in the negative- $v_g$ case. No paradox: the medium, responding causally to the pulse's leading tail, reconstructs a forward-shifted copy of the smooth envelope while attenuating the rest. A smooth Gaussian's early tail already contains, by analyticity, the information of its peak; the medium is an extrapolator, not a time machine. The proof it cannot cheat: put a genuine surprise on the pulse — a sharp front, a switch-on — and every experiment finds the surprise arriving at $c$ or later.

Step 3. The front theorem. Let a signal with a sharp front (identically zero before $t = 0$ ) enter any medium. Its spectrum, by the sharpness of the front, has components at arbitrarily high frequency, and the front's propagation is governed by their limit. Physically: at $\omega \to \infty$ the electrons' inertia makes them unresponsive — the medium has no time to polarize — so $n(\omega) \to 1$ (the absorption article's high-frequency limit, and a rigorous consequence of the Kramers–Kronig representation: $n(\omega) - 1 \sim -\omega_p^2/2\omega^2 \to 0$ ). The first flicker therefore propagates in effective vacuum:

v_{\rm front} = \lim_{\omega\to\infty}\frac{c}{n(\omega)} = c, \qquad \text{in every medium, always}.

Why nothing can outrun it even transiently: causality of the response function makes the propagated field an analytic functional of the input's past; a region where the input was identically zero propagates to identically zero — Sommerfeld's 1914 contour argument makes this exact: the response ahead of $z = ct$ vanishes because the contour closes in the analytic half-plane.

Step 4. The residue: precursors. Behind the front and ahead of the main signal, Sommerfeld and Brillouin found the medium's transient reply — first a feeble high-frequency ripple (the Sommerfeld precursor: the barely-slowed $\omega \to \infty$ components), then a low-frequency swell (the Brillouin precursor: the below-resonance components), and only then the main body at the signal velocity. The 1914 analysis, done by steepest descent through a Lorentz medium, remains the definitive audit of "what actually arrives when" — and its headline is the taxonomy's moral: five velocities, each answering a different question, only the front's pinned by principle. Verified: anomalous-window slope from the absorption article's Eq. (5) profiles; negative- $v_g$ phenomenology matches Wang et al., Nature 406, 277 (2000) ( $v_g = -c/310$ , pulse advance 62 ns) and the standard reshaping interpretation; front theorem per Sommerfeld–Brillouin 1914 (Jackson §7.11); $n(\omega\to\infty) = 1 - \omega_p^2/2\omega^2$ is the universal high-frequency form for any charge-inertia medium.

Limiting cases

Case	Result	Meaning
vacuum	$v_p = v_g = v_E = v_{\rm front} = c$	straight dispersion line: the taxonomy degenerates
transparent normal dispersion	$v_g < v_p$ , $v_E = v_g$	crests outrun packets; energy rides the envelope
deep-water waves	$v_g = v_p/2$ exactly	crests born at the back, dying at the front
plasma / waveguide ( $\omega^2 = \omega_c^2 + c^2k^2$ )	$v_p v_g = c^2$	chord above $c$ , tangent below — next articles' law
steep normal dispersion (EIT)	$n_g \sim 10^7$ : $v_g \sim$ m/s	slow light — meters per second in cold vapors
anomalous band	$v_g > c$ or $< 0$	reshaping, not transport; theorem of Derivation B lapses
sharp front, any medium	exactly $c$	$n(\infty) = 1$ : inertia is the law relativity hired
$\beta_2 \neq 0$ , pulse $T$	spreads at $L_D = T^2/\vert\beta_2\vert$	GVD: fiber penalty, pulsar tool, CPA resource

Common trapNeither velocity is sacred: phase velocity carries no information, and group velocity is only as honest as the medium is transparent

Two symmetric errors patrol this subject. The first — panicking at $v_p > c$ — dissolves immediately: a monochromatic wave's crests are featureless phase labels; nothing rides them (X-ray optics and waveguide phase velocities live above $c$ permanently and blamelessly). The second error is subtler and commoner: promoting $v_g$ to "the real speed" universally. Group velocity earns physical meaning through the energy-transport theorem, whose jurisdiction is transparent media; inside an absorption or gain band the packet reshapes, its peak is not a conserved messenger, and $v_g$ — while still the correct tangent slope and still predictive of where the peak of a smooth, band-limited pulse appears — no longer bounds or measures information flow. Measured $v_g = -c/310$ is a fact; usable superluminal signaling is not, and the discriminator is always the same: put a genuine discontinuity (a switch-on, a random bit) on the wave and it propagates at $c$ , medium be damned, because $n(\infty) = 1$ . When quoting speeds, name your regime: transparent → quote $v_g$ (= energy speed); absorptive/anomalous → quote the front ( $c$ ) and the observed peak shift separately, never as "the signal traveled at...".

Worked examples

1 · The pulsar's built-in spectrometer (confidence builder)

The Crab pulsar's pulses arrive at Earth through the interstellar plasma (dispersion measure DM $= 56.8$ pc cm⁻³ — the column density of free electrons). Using the plasma dispersion $\omega^2 = \omega_p^2 + c^2k^2$ , find the arrival-time delay at 1.4 GHz and at 400 MHz, and the differential delay a pulsar astronomer actually measures.

Step 1. Group velocity in a plasma: $v_g = c\sqrt{1 - \omega_p^2/\omega^2} \approx c(1 - \omega_p^2/2\omega^2)$ for $\omega \gg \omega_p$ (interstellar $f_p \sim$ kHz against GHz observing — deeply valid). Each pulse is slower at lower frequency: the extra delay per unit path is $\omega_p^2/2c\omega^2 \propto n_e/f^2$ .

Step 2. Integrate along the path: the delay relative to infinite frequency is $\Delta t = \dfrac{e^2}{8\pi^2\varepsilon_0 m_e c}\,\dfrac{\rm DM}{f^2}$ , which evaluates to the radio astronomer's constant: $\Delta t = 4.149\ \text{ms} \times \dfrac{\rm DM\,[pc\,cm^{-3}]}{(f/\text{GHz})^2}$ .

Step 3. Crab numbers: at 1.4 GHz, $\Delta t = 4.149\times56.8/1.96 = \mathbf{120\ ms}$ ; at 400 MHz, $4.149\times56.8/0.16 = \mathbf{1.47\ s}$ . The differential delay across that band — the measurable — is $\mathbf{1.35\ s}$ : the low-frequency edge of each pulse arrives more than a second late, for a pulse intrinsically sub-millisecond wide.

Step 4. Read it as an instrument. The $1/f^2$ signature is unmistakable, so the fit returns DM to high precision — a direct integral of interstellar electron density along the line of sight. Every pulsar is thus a plasma probe (DM values map the Galaxy's ionized gas), every fast radio burst carries its host's and the intergalactic medium's electron column, and pulsar-timing arrays must de-disperse to microsecond precision before hunting gravitational waves. WEQ's radar example trusted vacuum's straight dispersion line; this example is the bent line, monetized.

2 · Reading the swell forecast (exam level)

A storm 2000 km offshore generates waves of many wavelengths. Deep-water dispersion: $\omega = \sqrt{gk}$ . (a) Show $v_g = v_p/2$ . (b) When do 200 m swells arrive, and (c) why do the longest swells arrive first — with what practical consequence?

Step 1. The ratio: $\omega = \sqrt{gk} \Rightarrow v_p = \omega/k = \sqrt{g/k}$ , while $v_g = d\omega/dk = \tfrac{1}{2}\sqrt{g/k} = \mathbf{v_p/2}$ . (Watching crests advance through a swell group and vanish at its front is this equation, visible from any headland.)

Step 2. Speeds for $\lambda = 200$ m: $k = 2\pi/200 = 0.0314\ \text{m}^{-1}$ ; $v_p = \sqrt{9.81/0.0314} = 17.7$ m/s; $v_g = \mathbf{8.8\ m/s}$ . Travel time for the energy (which moves at $v_g$ — Derivation B's theorem, water-wave edition): $2\times10^6/8.8 = 2.26\times10^5$ s $= \mathbf{63\ hours}$ . Using $v_p$ would predict 31 hours and miss the arrival by a day and a half — the classic penalty for quoting the wrong velocity.

Step 3. Sorting: $v_g \propto \sqrt{\lambda}$ , so the sea itself is a dispersive delay line — 200 m swells (8.8 m/s) outrun 100 m swells (6.25 m/s) by a day over this path. The beach receives the storm's spectrum sorted, long first, and the arrival schedule runs backwards to the source: surf forecasters (a technique going back to wartime landing planning) read the day-by-day shortening of the arriving swell period to infer the storm's distance and intensity — exactly the pulsar astronomer's $1/f^2$ fit, transposed to $\sqrt{\lambda}$ and run on an ocean.

Going further

References. Griffiths, Introduction to Electrodynamics, 4th ed., §9.4.3 · Jackson, Classical Electrodynamics, 3rd ed., §7.8–7.11 (group velocity, energy transport, signal propagation, precursors) · Zangwill, Modern Electrodynamics, §18.6 · Brillouin, Wave Propagation and Group Velocity (Academic, 1960) · Purcell & Morin, Electricity and Magnetism, 3rd ed., Ch. 10

Provenance. Group-velocity forms $v_g = d\omega/dk = c/(n + \omega\,dn/d\omega)$ cross-checked against Griffiths §9.4.3 and Jackson §7.8; the energy-velocity theorem against the lossless-medium energy density $\tfrac{1}{4}[d(\omega\varepsilon)/d\omega]|E|^2$ form (Landau–Lifshitz ECM §80; Jackson §7.8; Zangwill §18.6); superluminal/negative group velocity phenomenology against Chu & Wong, PRL 48, 738 (1982) and Wang, Kuzmich & Dogariu, Nature 406, 277 (2000); front velocity and precursors per Sommerfeld and Brillouin (1914; Jackson §7.11). A verifier should: (1) re-derive $v_g$ from the beat superposition and from stationary phase; (2) confirm the Gaussian spreading law $T(z) = T[1 + (\beta_2 z/T^2)^2]^{1/2}$ ; (3) check $v_E = v_g$ by assembling flux over honest energy density for the lossless Lorentz medium; (4) recompute the dispersion-measure constant 4.149 ms·GHz²·(pc/cm³)⁻¹ from $e^2/8\pi^2\varepsilon_0 m_e c$ . Addendum for this edition: this article owns the velocity taxonomy the absorption article deferred forward (anomalous-band honesty resolved here); the waveguide instance $v_pv_g = c^2$ is deferred by name to the cutoff-and-dispersion article; the driven-oscillator energetics in Derivation B recalled from the mechanics Resonance article. Numbers recomputed: silica 1550 nm $n = 1.444$ , $n_g = 1.462$ , $D = 17$ ps/(nm·km) ↔ $\beta_2 = -21.7$ ps²/km, 0.1 nm × 100 km → 170 ps smear, $L_D = 4.6$ km for 10 ps; Crab DM 56.8 → delays 120 ms (1.4 GHz), 1.47 s (400 MHz), differential 1.35 s; swell $\lambda = 200$ m: $v_p = 17.7$ , $v_g = 8.8$ m/s, 2000 km in 63 h; Wang et al. advance 62 ns at $v_g = -c/310$ ; packet figure computed with $v_p = 2v_g$ .

Builds on: The Electromagnetic Wave Equation · Absorption and Dispersion · Complex Notation for Electromagnetic Waves · Monochromatic Plane Waves in Vacuum

Leads to: Guided Waves and Transmission Lines · Cutoff Frequencies and Dispersion in Waveguides · Electromagnetic Waves in Linear Dielectric Media

Is the group velocity just an approximation that fails for wide-band pulses, even in transparent media?

Yes — it is the first term of an expansion, and its accuracy is a bandwidth budget. Derivation A truncated $k(\omega)$ at first order to get rigid translation at $v_g$ ; the price of the neglected orders is quantified by the same expansion: second order ( $\beta_2$ ) spreads and chirps the packet over the dispersion length $L_D = T^2/|\beta_2|$ , third order distorts it asymmetrically (important near fiber zero-dispersion points, where $\beta_2$ vanishes and $\beta_3$ inherits the job). So "the packet moves at $v_g$ " is exact only in the narrow-band limit; for real pulses the honest statement is "the packet's center moves at $v_g$ while its shape evolves at rates set by higher $\beta_n$ ." An octave-spanning femtosecond pulse in glass violates the narrow-band premise outright — ultrafast optics therefore works with the full $k(\omega)$ numerically, reserving $v_g$ for narrow-band reasoning. The taxonomy survives: even for wide-band pulses, energy centroid motion in transparent media follows the spectrally-averaged group velocity.

The quantum connection: is the group velocity of a matter wave really the particle's velocity?

Exactly, and it is one of quantum mechanics' foundational consistency checks. A free particle's dispersion is $\omega = \hbar k^2/2m$ (from $E = p^2/2m$ with de Broglie's $E = \hbar\omega$ , $p = \hbar k$ ), giving $v_p = \hbar k/2m = v_{\rm classical}/2$ — the phase velocity is half the particle's speed, an early source of confusion — while $v_g = d\omega/dk = \hbar k/m = p/m$ : the group velocity is precisely the classical velocity. The packet's envelope is where the particle is; the sub-luminal-irrelevant crests inside it are gauge-like phase structure. Schrödinger leaned on this correspondence (which descends directly from Hamilton's optical-mechanical analogy — his mechanics was the geometric-optics limit of a wave theory waiting to be found) when constructing wave mechanics. The relativistic version is tidier still: $\omega^2 = c^2k^2 + (mc^2/\hbar)^2$ gives $v_pv_g = c^2$ — the same hyperbolic dispersion law the waveguide articles are about to meet with $\omega_c$ playing the mass.

If crests carry no information, why does anything measurable travel at the phase velocity at all?

Phase velocity governs everything that depends on phase relationships rather than on transported features: interference fringe positions, the optical path lengths in an interferometer arm, phase-matching in nonlinear crystals (where efficient harmonic generation requires the fundamental's and harmonic's phase fronts to stay locked — an engineering discipline built entirely on $v_p(\omega)$ differences), Doppler and aberration bookkeeping, and the resonance conditions of cavities. None of these is a signal racing anywhere; all are standing comparisons between already-present waves. The division of labor is clean: ask "when does the energy/message arrive?" → group (or front) velocity; ask "where do the fringes sit / does the interaction stay coherent?" → phase velocity. Both appear on the same datasheet — a fiber's $n$ sets connectors' phase behavior and reflection coefficients (Fresnel formulas use $n$ , not $n_g$ ), while its $n_g$ and $D$ set pulse timing — and confusing the columns is a genuine, recurring engineering bug.

What exactly are the Sommerfeld and Brillouin precursors — has anyone seen one?

They are the medium's causal reflexes, arriving between the front and the main signal. Turn on a sinusoid at the edge of a Lorentz medium: the front (traveling at exactly $c$ ) is built from the $\omega \to \infty$ components, which barely interact — behind it rides a faint, rapidly oscillating transient, the Sommerfeld precursor, whose frequency chirps downward as the slower-but-stronger high-frequency components arrive in sequence. Later comes the Brillouin precursor — a low-frequency swell built from the below-resonance components (which see the largest $n$ but the least absorption) — and only then the main signal at its steady-state amplitude. Both were extracted by steepest-descent analysis in 1914, as a direct rebuttal to "group velocity exceeds $c$ , relativity is in trouble." Experimentally: cleanly resolved in microwave waveguide analogs and in water-wave tanks; in optics, single-cycle terahertz and femtosecond experiments in the 1990s–2000s observed precursor-like transients in dielectrics (with some interpretive debate, since ultrashort pulses blur "precursor" and "pulse"). The Brillouin precursor has practical notoriety: its sub-exponential attenuation makes it a candidate for ground- and foliage-penetrating radar waveforms.

How do slow-light experiments reach group indices of ten million without absorbing everything?

By manufacturing a dispersion feature that is steep and transparent — the combination nature's isolated resonances refuse (steep slope normally lives inside the absorption line, per the Kramers–Kronig lock). Electromagnetically induced transparency (EIT) uses a control laser to split an atomic resonance into two (an Autler–Townes/dark-state structure); destructive interference of excitation pathways opens a narrow transparency window between them, and within that window $n(\omega)$ climbs extremely steeply — normal dispersion, since the window is bracketed by two absorption peaks — while $n$ itself stays near 1. Group index $n_g = n + \omega\,dn/d\omega$ is then dominated by the slope term: window widths of tens of kHz at optical carriers ( $10^{14}$ Hz) yield $n_g \sim 10^7$ , hence 17 m/s in Hau's sodium condensate. The same physics inverted (gain doublet) manufactures the anomalous window used for negative- $v_g$ experiments. Both are Derivation B's theorem operating at the edge of its jurisdiction: inside the EIT window the medium is genuinely transparent, energy velocity tracks the tiny $v_g$ , and the pulse's energy is mostly parked in atomic coherence — which is why switching off the control beam "stops" the light: the parked share is, at that moment, everything.

Does the vp times vg equals c-squared relation hold generally, or is it special?

Special — it is the signature of one family of dispersion relations, the hyperbolic $\omega^2 = \omega_c^2 + c^2k^2$ , and worth recognizing on sight. Differentiating: $2\omega\,d\omega = 2c^2k\,dk$ , so $v_pv_g = (\omega/k)(d\omega/dk) = c^2$ — an exact reciprocity, with $v_p > c$ and $v_g < c$ paired multiplicatively. The family's members: plasmas ( $\omega_c = \omega_p$ , worked example 1's medium), hollow waveguides ( $\omega_c$ = the cutoff frequency — the next two articles derive and spend this), and relativistic matter waves ( $\omega_c = mc^2/\hbar$ ) — three physical systems, one "massive photon" mathematics. Ordinary dielectrics do not obey it (their $\omega(k)$ is not hyperbolic; glass has $v_pv_g < c^2$ in the visible), so the relation is a diagnostic: if a medium's $v_pv_g = c^2$ , look for a cutoff/mass gap in its spectrum. The waveguide articles will use the reciprocity as a running consistency check — and its geometric origin (a zigzagging light ray whose axial phase pattern outruns light while its energy zigzags below $c$ ) is the single most illuminating picture in that chapter.