KM3NeT: The Deep-Sea Neutrino Telescope

A Neutrino Nobody Was Ready For

At 01:16 UTC on 13 February 2023 a muon crossed the water above the Mediterranean seabed off Sicily, travelling almost exactly along the horizon, 0.6 degrees above it. It triggered 3,672 photomultiplier tubes and left 28,086 recorded hits across twenty-one strings of sensors hanging in the dark. Reconstructed, that muon carried about 120 PeV of energy, and the neutrino that made it is estimated at roughly 220 PeV: around 25 times more energetic than any neutrino previously observed. The collaboration named the event KM3-230213A, which is simply the date.

The awkward part is what was in the water that night. ARCA, the detector that caught it, was running about one tenth of its planned sensors. The most energetic neutrino in the history of the subject walked into a construction site, and the construction site saw it anyway. KM3NeT published the result in Nature on 12 February 2025, two years after the fact, because defending a 220 PeV reconstruction from a partly built telescope takes that long.

Why Hang a Telescope in the Open Sea

Every high-energy neutrino telescope works the same way, and none of them detect neutrinos directly. A neutrino occasionally hits a nucleus and produces a charged particle, usually a muon, that keeps going at very nearly the speed of light in vacuum. That is faster than light travels in the surrounding medium, so the particle drags a shockwave of light behind it, Cherenkov radiation, cast forward in a cone. Line a cubic kilometre of transparent material with light sensors, record which sensor saw a photon and exactly when, and you can reconstruct where the neutrino came from and roughly its energy. A cubic kilometre is the scale you need because the chance of any individual neutrino interacting at all is absurdly small.

IceCube solved the transparent-material problem by drilling into the Antarctic ice sheet and freezing its sensors permanently in place. KM3NeT did the opposite: it lowers its sensors into open seawater and lets them hang. There is no drilling, the available volume is effectively unlimited, and the Mediterranean puts the instrument in the northern hemisphere, watching a different half of the sky from the South Pole. Both share the logic of every deep underground neutrino detector, which is to stack enough rock, ice or water overhead that almost nothing except a neutrino gets through. The price of water is that nothing stays still.

Seven Hundred Metres of String

KM3NeT is two telescopes with one design. ARCA, for Astroparticle Research with Cosmics in the Abyss, sits about 3,450 metres down some 80 kilometres off Portopalo di Capo Passero in Sicily, and hunts the highest-energy neutrinos. ORCA, Oscillation Research with Cosmics in the Abyss, sits about 2,450 metres down roughly 40 kilometres off Toulon in France, and is tuned instead to neutrinos of a few GeV, to measure oscillations and settle the neutrino mass ordering.

Both are assembled from the same unit. A detection unit is a vertical string about 700 metres tall, held to the seabed by a dead weight and pulled upright by a buoy at the top, carrying 18 Digital Optical Modules. Each module is a glass pressure sphere holding 31 small photomultiplier tubes aimed in different directions. The lowest module sits about 70 metres above the seabed and the rest are spaced roughly 36 metres apart up the rope, and neighbouring ARCA strings stand about 90 metres from each other. Thirty-one eyes per sphere is the quiet cleverness in the design: a single module already knows roughly which way a photon came from.

Depth 3,450 m · 349.5 bar

Track distance from module 100 m · 303 ns

What Seawater Costs You

Ice has one enormous advantage: it holds still. Seawater does not. At the Capo Passero site the currents push the strings sideways, and the top of a detection unit can be displaced by well over a hundred metres when the current runs at 30 centimetres per second. You cannot survey the detector once and be finished; it is a slightly different shape every hour.

So KM3NeT measures itself continuously, with sound. Acoustic emitters anchored on the seabed at known positions send out pings; hydrophones at the base of each string and piezo sensors glued inside the glass spheres hear them; the arrival times are triangulated on shore to fix every module to within about 10 to 20 centimetres. That demands timing the pulses to better than 50 microseconds, and because the speed of sound in seawater depends on temperature, salinity and pressure, a dedicated instrumentation line carries velocimeters and conductivity-temperature-depth probes to keep that speed honest. It is exactly the physics behind sonar and echo location, run backwards to find the listener instead of the target.

Then there is the light the sea makes by itself. Seawater contains potassium-40, a naturally radioactive isotope whose beta decays shed Cherenkov light everywhere, continuously, at a rate of tens of kilohertz in a large photomultiplier. On top of that the deep Mediterranean is full of organisms that manufacture their own light, and a bioluminescent burst can drive hit rates into the megahertz range. Antarctic ice has neither problem. Neither background can be removed, so both have to be modelled, monitored and calibrated around.

What Seawater Gives Back

The compensation is optical. In the clearest Antarctic ice, blue light travels roughly 200 metres before it is absorbed but only about 50 metres before it is scattered, so photons reach the sensors smeared in both time and direction. Deep Mediterranean water is the reverse: measurements at the KM3NeT site give an absorption length near 60 metres and a scattering length near 250 metres. Photons disappear sooner, which is why water detectors need their sensors packed more densely, but those that do arrive have travelled almost exactly along the line they were emitted, and arrive when geometry says they should.

That is worth a great deal, because the Cherenkov angle is not a free parameter. With cos θ𝒸 = 1/n and n = 1.35 for 450 nm light in seawater, the cone opens at θ𝒸 ≈ 42°, and the first photon to reach a module arrives a fixed 3.03 nanoseconds later for every metre the track passed away from it. Clean arrival times therefore convert straight into direction. ARCA is designed for a median angular resolution around 0.1° for muon tracks at a few hundred TeV and below about 2° for the fuzzier shower-like events. For neutrino astronomy that is the difference between naming a patch of sky and naming a source.

The Trouble With One Event

KM3-230213A is the most interesting neutrino ever recorded and nobody can say where it came from. Its direction is known to 1.5° at 68 per cent confidence, and that uncertainty is dominated not by physics but by how well the collaboration knew the absolute orientation of a half-built detector sitting on the seabed. A dedicated survey and a full recalibration should eventually bring it towards the 0.12° statistical uncertainty of the muon track itself. Several candidate sources sit inside the current error region. None of them produced a coincident gamma-ray signal.

Then the harder problem. Treat the event as a fair sample of a steady all-sky flux, and the flux it implies sits in roughly 2.5 to 3 sigma tension with the upper limits that IceCube and the Pierre Auger Observatory have set at those energies. IceCube has watched longer, with a larger instrumented volume, and has not reported a comparable track from that direction. The tension is real and unresolved. The candidate explanations include a single source that flared for less than about two years, a cosmogenic neutrino born when an ultra-high-energy cosmic ray collided with a microwave-background photon, ordinary statistical bad luck, and physics nobody has written down yet. Anyone who tells you which, is guessing.

Sixty-Two Strings and Counting

In September 2026 the collaboration reported a seventeen-day campaign out of Malta that installed a dozen more detection units at the ARCA site, bringing the detector to 62 strings linked to shore by two long electro-optical cables and seven submarine junction boxes. The final ARCA design calls for 230. The telescope that caught a 220 PeV neutrino using one tenth of its sensors now has roughly a quarter of them, and every one of those strings is still swaying in the dark, being re-measured by sound, while the sea glows faintly around it.

Run it yourself: the Cherenkov cone simulation lets you push a particle past the speed of light in seawater and watch the cone snap into existence at cos θc = 1/(βn).