Fourier Modes, Dispersion, and Group Velocity

Guiding question: When does group velocity correctly describe the transport of a localized disturbance?

Group velocity tracks a packet when the packet is narrow in wave number, lies on one smooth dispersion branch, and is observed before higher-order dispersion substantially reshapes its envelope. This approximation can apply from the initial instant; long propagation time is needed only for the separate stationary-phase argument below. Group velocity is not, by definition, the signal-front velocity.

Required background. Ordinary Differential Equations provides initial-value problems, linearization, and asymptotic solution methods; Classical Field Theory provides field actions, Euler–Lagrange equations, and gauge redundancy. Helpful background. Calculus and Linear Algebra provides multivariable differentiation, linear maps, inner products, and basis changes; Partial Differential Equations and Green Functions provides characteristics, boundary data, distributions, and Green-function methods.

Deriving the envelope velocity

For a translation-invariant linear equation, insert

$$\psi=Ae^{i(\mathbf k\cdot\mathbf x-\omega t)}.$$

The resulting algebraic condition $D(\omega,\mathbf k)=0$ is the dispersion relation. On a branch $\omega=\Omega(\mathbf k)$, planes of constant phase move along $\mathbf k$ with phase speed

$$v_{\rm ph}=\frac{\omega}{|\mathbf k|}, \qquad \mathbf v_{\rm ph}=\frac{\omega\mathbf k}{|\mathbf k|^2}.$$

This is the normal velocity of a constant-phase plane. A monochromatic mode is spatially infinite, so following one of its crests is not the same operation as following a localized message.

In one dimension, superpose two nearby modes, $k_\pm=k_0\pm\Delta k/2$ and $\omega_\pm=\Omega(k_\pm)$. Define their average frequency $\bar\omega=(\omega_++\omega_-)/2$ and separation $\Delta\omega=\omega_+-\omega_-$. The exact identity for a sum of cosines gives a rapid carrier multiplied by a slow envelope:

$$\psi =2\cos\left(\frac{\Delta k\,x-\Delta\omega\,t}{2}\right) \cos(k_0x-\bar\omega t).$$

The envelope has constant phase when

$$\frac{x}{t}=\frac{\Delta\omega}{\Delta k} \longrightarrow \Omega'(k_0).$$

Thus

$$\boxed{\mathbf v_{\rm g}=\nabla_{\mathbf k}\Omega.}$$

For a general one-dimensional packet

$$\psi(t,x)=\int_{-\infty}^{\infty}\frac{dk}{2\pi}\, a(k)e^{i[kx-\Omega(k)t]},$$

stationary phase selects modes satisfying $x/t=\Omega'(k)$ when $t$ is large with $x/t$ fixed and the stationary point is nondegenerate. If no wave number on the occupied branch has that group velocity, repeated integration by parts suppresses the smooth far-field signal. This result, including the isolated-branch and nondegenerate-stationary-point assumptions, is derived in Rosales 2023, §§3.1–3.3, pp. 15–23 (official PDF).

Expanding about $k_0$,

$$\Omega(k)=\Omega_0+v_{\rm g}(k-k_0) +\frac12\Omega''_0(k-k_0)^2+\cdots,$$

shows the limitation. The linear term translates the envelope; $\Omega''_0$ broadens and chirps it. To expose the control parameter, write $q=k-k_0$ and factor out the carrier:

$$\psi=e^{i(k_0x-\Omega_0t)} \int\frac{dq}{2\pi}\,a(k_0+q) \exp\!\left[iq(x-v_{\rm g}t)-\frac{i}{2}\Omega''_0q^2t+\cdots\right].$$

For a packet of spectral width $\Delta k$, rigid translation is accurate while

$$\epsilon_{\rm disp}=|\Omega''_0|\,(\Delta k)^2t\ll1.$$

Spreading becomes order one near $t_{\rm disp}\sim1/(|\Omega''_0|(\Delta k)^2)$. A vanishing $\Omega''_0$ does not guarantee no spreading: the first nonzero higher derivative then supplies the relevant scale. Here $\Omega_0=\Omega(k_0)$; in the earlier two-mode identity, $\bar\omega$ equals this only up to terms of order $(\Delta k)^2$.

For a concrete packet, choose the initial complex envelope $A(0,x)=A_0\exp[-x^2/(2\ell^2)]$. Keeping the quadratic dispersion term gives the Gaussian integral

$$A(t,x)=\frac{A_0}{\sqrt{1+i\Omega''_0t/\ell^2}} \exp\!\left[-\frac{(x-v_{\rm g}t)^2}{2(\ell^2+i\Omega''_0t)}\right].$$

Its center moves at $v_{\rm g}$, while the width parameter of its magnitude is

$$\ell(t)=\ell\sqrt{1+\left(\frac{\Omega''_0t}{\ell^2}\right)^2}.$$

This explicitly separates translation from spreading and recovers $t_{\rm disp}\sim\ell^2/|\Omega''_0|$ since $\Delta k\sim1/\ell$. The formula is exact for a quadratic dispersion relation and approximate otherwise. A Gaussian has nonzero tails at every $x$, so it has no sharp signal front; it cannot by itself test the causal-front speed.

Four velocities that should not be conflated

• Phase velocity follows monochromatic crests.
• Group velocity follows a narrowband envelope only on a real, smooth, isolated dispersion branch in the regime just stated.
• Energy velocity is the time-averaged energy flux divided by the time-averaged energy density. It equals group velocity for the lossless isolated single-branch setting developed in Rosales 2023, §6.1, pp. 35–39 (official PDF), but not universally in absorbing or multibranch media.
• Front velocity follows the boundary of exact support. For a local hyperbolic equation it is fixed by the characteristic cone; in a causal dispersive medium it is tied to the high-frequency response rather than to a finite-band peak.

In a lossy medium, either $k$ or $\omega$ is generally complex, so differentiating a real graph and calling the result a transport speed is already an extra approximation. The real, isolated-branch derivation above no longer establishes the motion of a pulse maximum, energy, or information. Causality instead resides in the retarded response, whose frequency-domain analyticity constrains dispersion through Kramers–Kronig relations. See Tong 2015, §§7.5.2–7.5.4, pp. 196–204 (official PDF) for the narrow-packet argument and the causal response of matter.

Checks and failure modes

For a nondispersive vacuum wave, $\Omega=|k|$, so $v_{\rm ph}=1$ and $|v_{\rm g}|=1$; the sign of $v_{\rm g}$ records the propagation direction. A broad pulse, a packet spanning a band crossing, or two coupled branches has no single reliable group velocity. Near $\Omega''=0$ higher dispersion orders may control spreading. Near a turning point, stationary points merge and an Airy-type uniform approximation replaces separate Gaussian stationary-phase contributions; see Rosales 2023, §3.3, pp. 20–23 (official PDF).

Operationally, phase velocity comes from crest timing and group velocity from an envelope fit. A mathematical front is the boundary of exact support, whereas an experimental arrival time is defined by a detector threshold above noise. The measurement protocol is therefore part of any empirical velocity claim.

Transfer task: Klein–Gordon dispersion

For

$$(\partial_t^2-\partial_x^2+m^2)\psi=0,$$

compare phase, group, and front speeds for $m\gt0$. Then estimate the spreading time of a narrow packet centered at $k_0$ with spectral width $\Delta k$.

Solution

On the positive-frequency branch the dispersion relation is

$$\omega(k)=\sqrt{k^2+m^2}.$$

For $k\ne0$,

$$v_{\rm ph}=\frac{\omega}{|k|}\gt1, \qquad |v_{\rm g}|=\frac{|k|}{\omega}\lt1, \qquad v_{\rm ph}|v_{\rm g}|=1.$$

The superluminal phase speed transports no localized feature. As $|k|\to\infty$, $|v_{\rm g}|\to1$, agreeing with the characteristic front speed. In the low-$k$ limit the packet group velocity tends to zero and dispersion is strong.

Differentiating once more gives

$$\omega''(k)=\frac{m^2}{(k^2+m^2)^{3/2}}.$$

Provided $\Delta k$ is narrow enough that $\omega''$ is approximately constant across the occupied band, order-one spreading begins at

$$t_{\rm disp} \sim \frac{(k_0^2+m^2)^{3/2}}{m^2(\Delta k)^2}.$$

Retarded and Advanced Green Functions next makes the causal support choice explicit for sourced waves.

References

• R. R. Rosales, Lecture Topics for MIT 18.376, Massachusetts Institute of Technology, 2023, §§3.1–3.3, pp. 15–23, §4.1, pp. 24–28, and §6.1, pp. 35–39 (official PDF).
• David Tong, Lectures on Electromagnetism, University of Cambridge, 2015, §§7.5.2–7.5.4, pp. 196–204 (official PDF).