Gain-assisted superluminal light propagation
A light pulse crossing a caesium cell with a negative group index leaves before it arrives, undistorted, by classical interference in anomalous dispersion; its front still travels at exactly c.
Kind: physics · Loophole: L5 · Standing: K4 · Bill: B-none · Last reviewed: 2026-09-12
The claim
The paper makes a precise and modest claim, and the faster-than-light reading that the press attached to it is a different claim, so both are stated here. Wang, Kuzmich and Dogariu (Nature 406, 277, 2000) claim that in a region of anomalous dispersion created between two closely spaced Raman gain lines in caesium vapour, a laser pulse propagates with a group velocity that exceeds $c$ and is in fact negative, while its shape is preserved and its intensity is neither strongly absorbed nor strongly amplified. Their measured group-velocity index is $n_g = -310 \pm 5$, so that the peak of a pulse appears at the exit of the cell before it enters, an advance of about 62 ns for a cell 6 cm long. This is the best-controlled demonstration of a negative group velocity, cleaner than the absorptive cases where the transmitted pulse is heavily attenuated and reshaped. The authors say plainly that the effect is "not at odds with causality, being a direct consequence of classical interference between its different frequency components in an anomalous dispersion region" [HIGH] S1.
The FTL reading, made in 2000 by press accounts and by some proponents of signalling schemes, is that a pulse which leaves before it arrives has carried something faster than light and could carry a message. In its strongest form: the pulse is undistorted, so any information encoded in its shape is available at the exit 62 ns early, and a chain of such cells could compound the advance without limit. The dossier grades the effect K4 as an established, peer-reviewed result of classical electrodynamics and quantum optics, and the FTL reading K0 as a misreading that the authors themselves, and the experiments that followed, have closed.
Origin and lineage
Sommerfeld and Brillouin settled the principle in 1914: in a dispersive medium the phase velocity and the group velocity are free to exceed $c$, and the group velocity can become negative near an absorption line, but the front of a signal that is switched on travels at exactly $c$ [HIGH] S2 (Brillouin 1960). Garrett and McCumber predicted in 1970 that a Gaussian pulse can traverse an anomalous-dispersion medium with its peak advanced, undistorted, provided the pulse is long and the medium thin (Phys. Rev. A 1, 305) [HIGH] S1. Chu and Wong observed it in an absorbing medium in 1982 (Phys. Rev. Lett. 48, 738) [HIGH] S1. Wang, Kuzmich and Dogariu's contribution was to replace absorption with a pair of gain lines, so that the region between them has steep anomalous dispersion with nearly zero net gain and loss, and to measure a large negative $n_g$ with pulse shape preserved [HIGH] S1. Kuzmich, Dogariu, Wang, Milonni and Chiao followed in 2001 with an analysis of the quantum noise that accompanies gain, showing why the signal velocity stays below $c$ [MED] S1. Stenner, Gauthier and Neifeld measured the speed of information directly in a fast-light medium in 2003 (Nature 425, 695) [MED] S1. The register cites this paper in LOR-3.
Nothing in fiction descends from it. It belongs to the same family as the tunnelling claims in the Hartman effect and the Nimtz claims: a real superluminal group velocity that is not a signal.
The mechanism
A pulse is a superposition of plane waves of different frequencies. In a medium with refractive index $n(\omega)$ the phase velocity is $v_p = c/n$ and the group velocity, the speed of the envelope for a narrowband pulse, is
Normal dispersion has $dn/d\omega > 0$ and $n_g > 1$. Near a resonance, $dn/d\omega$ is large and negative on one side (anomalous dispersion), and $n_g$ can fall below 1, through zero and negative. A negative $n_g$ means the envelope inside the medium moves backwards: the exit pulse forms and leaves before the entrance pulse has fully arrived, while inside the cell a pulse travels backwards from the exit to meet and cancel the incoming one. For a cell of length $L$ the peak advance relative to vacuum is
which for $n_g = -310$ and $L = 6\ \mathrm{cm}$ gives $\Delta t \approx -62\ \mathrm{ns}$, about 300 times the vacuum transit time of the cell (0.2 ns) [HIGH] S1 (the index is the paper's measured value; the advance follows from it). The authors obtained the anomalous dispersion with two Raman gain lines about 2.7 MHz apart in caesium; between the gain peaks the index falls steeply with frequency while the gain and absorption nearly cancel, so the pulse is neither amplified nor attenuated and, because the pulse bandwidth is small compared with the line spacing, its shape is preserved to good approximation [HIGH] S1.
Why is this not a signal? Three arguments, in increasing strength.
First, the Kramers-Kronig relations tie $n(\omega)$ to the absorption or gain spectrum. Any medium becomes transparent at frequencies far above its resonances, with $n(\omega) \to 1$ as $\omega \to \infty$. A signal that is switched on has a front, a point where the field is non-analytic, and the front's spectrum extends to arbitrarily high frequency, so the front propagates at exactly $c$. Everything the medium does happens behind the front. This is Sommerfeld and Brillouin's theorem [HIGH] S2.
Second, a smooth Gaussian pulse is an analytic function. Its entire future shape is encoded in any small segment of its leading edge. What the medium does with anomalous dispersion is to extrapolate: the leading tail, which is already present at the exit window at the time the peak would have arrived in vacuum, is reshaped by interference into a copy of the whole pulse. No part of the output depends on any part of the input that has not yet arrived at the light-cone speed. The output "peak" is built from input that arrived early. Garrett and McCumber's 1970 analysis is exactly this [HIGH] S1. The pulse is preserved only because it is bandlimited; any attempt to encode new information, a step or a discontinuity, injects high frequencies that fall outside the anomalous-dispersion window and travel at $c$ or slower.
Third, the experiments. Stenner, Gauthier and Neifeld encoded bits as sharp transitions on pulses sent through a fast-light medium with $n_g$ large and negative, and measured the time at which a receiver could distinguish the bit: the information velocity was not superluminal, and the detection was in fact slightly later than in vacuum, even though the pulse peak arrived early [MED] S1. Kuzmich, Dogariu, Wang, Milonni and Chiao showed that in a gain medium the quantum noise added by spontaneous emission grows in exactly the way needed to prevent the early part of the pulse from being read reliably, so that the signal-to-noise defined operationally arrives no earlier than $c$ allows; Aharonov, Reznik and Stern had argued in 1998 on general grounds that quantum noise must enforce this limit on any superluminal group velocity (Phys. Rev. Lett. 81, 2190) [MED] S1.
The register's LOR-3 entry summarises this as classical interference between frequency components in anomalous dispersion, which is the authors' own phrase.
What it costs
Nothing. Caesium vapour, two Raman pumps and a probe laser. The bill is B-none and the standing is K4 because the effect is an established, replicated result of linear optics and there is no exotic requirement. The FTL reading would need the front velocity theorem to fail, which would need the medium to remain dispersive at arbitrarily high frequency, which no medium does. That reading is scored below as a violation of LOR-3 attributed to the reading rather than to the paper, and it gets K0.
Constraint scoring
| Constraint | Verdict | Note |
|---|---|---|
| CAU-1 | SATISFIES | No signal outruns light: the front moves at c and the information velocity is measured at or below c (Stenner, Gauthier and Neifeld). The authors state the result is not at odds with causality |
| CAU-2 | N/A | No spacetime geometry |
| CAU-3 | N/A | No wormhole or bubble |
| CAU-4 | N/A | No preferred frame is needed, because nothing superluminal carries information; the medium's rest frame is a frame like any other |
| CAU-5 | N/A | No entanglement is involved |
| ENE-1 | N/A | An atomic vapour is ordinary matter; the energy conditions price geometry |
| ENE-2 | N/A | No gravitational field; ENE-2 concerns time advance through curved space |
| ENE-3 | N/A | No negative energy |
| ENE-4 | N/A | Warp-class |
| ENE-5 | N/A | Warp-class |
| ENE-6 | N/A | Casimir effect not invoked |
| ENE-7 | N/A | Pulse reshaping in an optical medium on flat spacetime; there is no topology and no geometric shortcut, so the censorship theorem does not bear. |
| CON-1 | N/A | Nothing is piloted |
| CON-2 | N/A | No route is laid; the cell is the whole apparatus, and stacking cells does not compound an advance beyond the pulse length |
| CON-3 | N/A | No tube |
| STA-1 | N/A | No bubble |
| STA-2 | N/A | No throat |
| STA-3 | N/A | No chronology horizon |
| HAZ-1 | N/A | Nothing is swept up |
| HAZ-2 | N/A | No horizon |
| HAZ-3 | N/A | No throat |
| LOR-1 | N/A | Light is massless and nothing is accelerated |
| LOR-2 | N/A | No tachyon; a negative group index is a property of the envelope, not of a particle with spacelike momentum |
| LOR-3 | SATISFIES | This is the entry's own worked example. The paper claims a group velocity above c and a front velocity of c, and says so; the FTL reading would violate LOR-3 and is refuted by the same analysis (K0) |
| LOR-4 | N/A | Neither Scharnhorst nor OPERA |
| WRP-1 | N/A | Not a warp drive |
| WRP-2 | N/A | Not a warp shell |
| WRP-3 | N/A | Not in the warp class |
| MAN-1 | N/A | No extra dimension |
Status of the argument
- 1914: Sommerfeld and Brillouin prove the front velocity is c in any causal medium [HIGH] S2.
- 1970: Garrett and McCumber predict undistorted advanced pulses in anomalous dispersion [HIGH] S1. 1982: Chu and Wong observe them in an absorber [HIGH] S1.
- 2000: Wang, Kuzmich and Dogariu measure $n_g = -310$ with a gain doublet and state that causality is untouched [HIGH] S1. Press coverage reports "light faster than light" and the authors spend the following year correcting it.
- 2001: Kuzmich, Dogariu, Wang, Milonni and Chiao analyse signal velocity and quantum noise in the same system [MED] S1.
- 2003: Stenner, Gauthier and Neifeld measure the information velocity in a fast-light medium and find it is not superluminal [MED] S1.
- No peer-reviewed claim that the gain-assisted advance carries information exists as of 2026-09-12. The effect is textbook physics; the FTL reading survives only in popular accounts.
Sources
- Wang, Kuzmich and Dogariu, "Gain-assisted superluminal light propagation", Nature 406, 277 (2000), doi:10.1038/35018520. S1
- Garrett and McCumber, "Propagation of a Gaussian light pulse through an anomalous dispersion medium", Phys. Rev. A 1, 305 (1970). S1
- Chu and Wong, "Linear pulse propagation in an absorbing medium", Phys. Rev. Lett. 48, 738 (1982). S1
- Kuzmich, Dogariu, Wang, Milonni and Chiao, "Signal velocity, causality, and quantum noise in superluminal light pulse propagation", Phys. Rev. Lett. 86, 3925 (2001). S1
- Stenner, Gauthier and Neifeld, "The speed of information in a 'fast-light' optical medium", Nature 425, 695 (2003), doi:10.1038/nature02016. S1
- Aharonov, Reznik and Stern, "Quantum limitations on superluminal propagation", Phys. Rev. Lett. 81, 2190 (1998). S1
- Brillouin, Wave Propagation and Group Velocity (Academic Press, 1960). S2
- Winful, "Tunneling time, the Hartman effect, and superluminality: a proposed resolution of an old paradox", Phys. Rep. 436, 1 (2006), for the general front-velocity discussion. S2