Light travelling down a glass fibre does not pass through the glass without touching it. Some of it scatters off sound waves moving through the same material, and the scattered light comes back slightly lower in frequency, having handed part of its energy to the sound. The effect is called stimulated Brillouin scattering, and in telecommunications it is usually a nuisance, because it steals power from a signal that was supposed to arrive intact.
Turned around, it is also one of the strongest nonlinear effects available in an optical fibre. A fibre that does it strongly can amplify a signal, filter a microwave tone, build a very narrow laser line, or store a pulse of light by writing it into sound and reading it back. The measure of how strongly a particular fibre does this is its Brillouin gain, quoted in inverse watts per metre, and standard telecom fibre sits low. In work published in Optica on 19 July 2026, Simon Seiderer and seven colleagues report 434 per watt per metre. They got it by freezing the liquid in the fibre’s core. The paper places that figure more than three orders of magnitude above standard silica.
A fibre with a liquid core
The device is simpler than the number suggests. It is a sealed silica capillary five metres long, with a nominal inner diameter of 1.37 micrometres, filled with carbon disulfide. Both ends are fusion spliced, through ultrahigh numerical aperture bridge fibres, to ordinary single mode fibre pigtails. Liquid core fibres of this kind are an established platform for nonlinear optics, and carbon disulfide is one of the standard fillings.
Because the capillary is sealed and completely full, the liquid inside cannot expand or contract freely. Heat part of the fibre and the pressure rises everywhere along it. Cool part of it and the pressure falls, and if the core is narrow enough the liquid column may hold together under tension instead of breaking into vapour. The same group has previously reported absolute negative pressures as low as minus 300 bar on this platform, in an earlier study of the same kind of fibre.
This experiment did not go there. Cavitation would interrupt the light, so the sample was kept at positive internal pressure throughout. Suppressing cavitation is what keeps the fibre in the isochoric state the measurement depends on.
Carbon disulfide melts at 162 kelvin, or about minus 112 degrees Celsius. That is warm enough that liquid nitrogen at 77 kelvin will freeze it, which is why this experiment needs a dewar rather than a cryostat. A 27.5 centimetre stretch of the fibre was lowered into liquid nitrogen. A separate four metre section, kept physically apart from it, could be warmed to as much as 343 kelvin to raise the internal pressure before freezing, which is how the group tuned the result.
One assumption is worth registering before any of the numbers. The pump power inside the fibre is not measured directly. The authors infer it from the power sent into the pigtails by assuming that the sample’s total loss, 5.6 decibels at ambient conditions, splits evenly between the input and output ends. Every figure below with watts in its denominator follows from that.
The jump on freezing
Why should freezing matter so much? The Brillouin gain coefficient of a material scales with the eighth power of the effective refractive index of the optical mode. An eighth power is unforgiving in both directions, and a modest rise in refractive index becomes a large rise in gain.
Solid carbon disulfide is denser than the liquid, and the paper expects its refractive index to sit well above the ambient liquid value of 1.5885 at a wavelength of 1550 nanometres. From time domain measurements the group puts the effective refractive index of the fundamental optical mode in the frozen fibre at 1.94, with an uncertainty of 0.21. Converting that back through the simulation gives a core index of 2.07, again with an uncertainty of 0.21, and that is the value the rest of the mode calculations use. Those simulations, not a measurement, are also where the smallest optical and acoustic overlap area comes from: 1.28 square micrometres for the fundamental pair, a tighter confinement that lifts the gain again.
The measured result is a Brillouin gain of 434 per watt per metre with an uncertainty of 22, against 47 with an uncertainty of 3 for the same fibre in its liquid state. The paper calls that nearly an order of magnitude. The frequency at which the effect peaks moves too, from 2.46 gigahertz in the liquid to 4.81 gigahertz in the solid, and the resonance narrows from 71 to 24 megahertz.
The gain does not land on the same number every time. The authors report reproducible Brillouin responses over hundreds of freezing cycles, and also that some cycles produce lower gain, with the majority above 390 per watt per metre. That is a narrower claim than the single headline number implies: 434 is the best of a spread, and the honest floor is that most freezing cycles clear 390 per watt per metre.
The transition looks instant. The experimental setup resolves events on the scale of seconds, and at that resolution the spectrum simply switches from one state to the other with nothing in between. A gradual change is expected, and Rolle and colleagues have demonstrated one in bulk glycerol and aqueous solutions; this instrument cannot resolve it, which settles nothing either way.
The practical numbers moved the right way in one respect. Cooling widened the mode enough to improve the splice coupling, so total transmission rose by 0.15 decibels rather than falling. Propagation loss in the frozen section measured 0.20 decibels per metre, with an uncertainty of 0.08.
A memory made of sound
With that much gain, a small pump does real work. The group reports 6.94 decibels of on and off amplification using 13.4 milliwatts of pump power, with an uncertainty of 0.7, over the 27.5 centimetre frozen stretch.
They then used it to store light. Two counter propagating pulses, a data pulse and a control pulse separated in frequency by the Brillouin shift, meet inside the frozen section, and the data pulse’s information is transferred into an acoustic wave. A second control pulse converts it back into light. The technique itself is not new; storing light in a fibre this way was demonstrated in 2007. What is new is the energy budget.
A data pulse of 0.030 nanojoules and pulses 1.7 nanoseconds long were used, with a storage time of 3.0 nanoseconds. Readout rose above the noise floor by three standard deviations with a control pulse of only 0.205 nanojoules. Pushed to 2.05 nanojoules, total write efficiency reached up to 63 per cent and readout efficiency up to 23 per cent. At the lowest control pulse energy demonstrated, storing one bit per pulse in amplitude, the paper puts the cost at 205 picojoules per bit.
The authors also convert that into a comparison. Using the pulse area criterion the memory literature uses to score storage efficiency, they calculate that a highly nonlinear fibre held in a cryostat would need 150 times more pump power to reach the same value.
The frozen section pays for this with an impairment. A core index of 2.07 gives a V number of 4.1, so light there is multi mode. The large mismatch in sound speed between carbon disulfide and silica supports several guided acoustic modes too. Between them those produce the extra Brillouin resonances visible as sidebands around 4.52 and 5.06 gigahertz. Broad data and control pulses interacting with those, the authors note, can generate weak, phase shifted readouts that interfere with the primary signal. They suggest it could be mitigated by tuning the storage time so the stray contributions phase-match the main readout. A smaller core, they add separately, could also make the whole fibre single mode.
Two disclosures sit alongside the memory result. Mario Chemnitz and Markus Schmidt, two of the eight authors, hold patents on the fabrication and the locally distributed temperature control of liquid core optical fibres. The paper declares this while stating no conflict of interest. The underlying data are not publicly posted either, and may be obtained from the authors on reasonable request, so none of the numbers above can be recomputed from a repository today.
One thing the fibre has not yet told them
The nearest term uses are unglamorous and specific. Distributed fibre sensing, microwave photonics, narrow linewidth lasers, and the low power optical memory and activation functions that optical computing needs if it is to beat electronics on energy. A gain figure that high, in a fibre you can splice into existing equipment, is worth more to those applications than an exotic waveguide needing free space alignment. The paper also puts the strong coupling regime of quantum optomechanics within reach, which is a projection rather than a result here.
Two of the headroom figures are projections too, and of different kinds. A core diameter of 0.8 micrometres could shrink the mode overlap area by up to a factor of 1.5 and reach roughly 650 per watt per metre, which comes from numerical simulation. A fibre long enough for the loss limited effective length to apply could in principle reach 42 decibels per milliwatt against the 0.52 obtained here, which the authors calculate analytically from the measured propagation loss. Neither has been built.
It is not quite the record holder. Microstructured and tapered chalcogenide fibres reach 550 per watt per metre, a margin the paper calls only slight, and they pay for it with free space coupling and propagation loss of 0.65 decibels per metre. The authors describe that trade off by quoting a chalcogenide group: “it is a challenge to obtain such a small-core fiber with low optical losses.”
The piece of physics sitting unresolved in the middle of it is the frozen core itself. Solid carbon disulfide has been studied in bulk, by X-ray crystallography in the 1960s and by Brillouin and Raman scattering under high pressure since. But the paper points to no prior characterisation of it frozen inside a 1.37 micrometre fibre core, and does not claim to have settled the structure here.
What the fibre gives up is an effective index for its fundamental optical mode, 1.94 with an uncertainty of 0.21. Worked back from that index and the measured frequency shift, it also gives up an effective speed of sound: 1920 metres per second, with an uncertainty of 210. Because the spectrum fits a homogeneous, isotropic profile, the authors narrow the structure to two candidates and cannot choose between them. That choice, they say, needs temperature resolved Brillouin or Raman work they have not done. Is the material in there a glass, or a crowd of very small crystals?