On 16 June 2023, an interaction inside the LUX-ZEPLIN experiment produced two brief flashes. One came promptly as atoms in liquid xenon emitted ultraviolet light. The other appeared after freed electrons drifted upward into xenon gas. From those signals, researchers reconstructed something striking: a xenon nucleus appeared to have recoiled with about 248 kiloelectronvolts of energy.
The event was deep inside the carefully defined central region of the detector. It did not coincide with activity in the surrounding systems designed to expose unwanted particles. Its signals looked unlike the far more common electron recoils produced by radioactive decay.
Then the event remained alone. Across 220 live days, no second interaction appeared with the same combination of high energy, position and nuclear-recoil-like behaviour.
That is why the result is both enticing and limited. Northwestern physicist Eric Dahl called it the most interesting single event he had seen in 20 years of searching. In the same account, he warned that nobody should claim a discovery from one event.
LZ was built to notice almost nothing
LUX-ZEPLIN, usually shortened to LZ, operates at the 4,850-foot level of the Sanford Underground Research Facility in Lead, South Dakota. The site occupies part of the former Homestake gold mine. Nearly a mile of rock overhead filters out most cosmic rays before they can reach the experiment.
The full detector contains about 10 tonnes of ultrapure xenon, with seven active tonnes in its central time projection chamber. For this analysis, researchers defined a smaller fiducial mass of 4.71 tonnes. Keeping the search volume away from surfaces reduces contamination from the walls and electrodes.
A particle depositing energy in the liquid produces an immediate scintillation signal called S1. The same interaction also frees electrons. An electric field draws them upward into a thin layer of xenon gas, where they generate a larger second signal called S2.
The time between the two flashes gives the depth of the interaction, while the light pattern reveals its horizontal position. The ratio between S1 and S2 helps distinguish recoiling electrons from recoiling xenon nuclei. Dark-matter particles known as WIMPs should produce nuclear recoils, but so can neutrons and neutrinos.
LZ therefore surrounds the target with more layers of detection. Additional xenon, a liquid-scintillator veto and a tank containing hundreds of tonnes of purified water help identify particles entering or leaving the central chamber. The aim is not to make backgrounds literally impossible, but to measure and reject them so effectively that an untagged nuclear recoil becomes informative.
The event opened a higher-energy part of old data
The collaboration collected the 220 live days between March 2023 and April 2024. An earlier analysis had searched the dataset for the low-energy recoils expected from the simplest WIMP interactions. The new work asked whether less conventional interactions could deposit much more energy.
The collaboration extended its nuclear-recoil window to almost 270 keV. After accounting for the working mass and time, the exposure was 2.84 tonne-years. That is the equivalent of observing one tonne of target for 2.84 years, although the actual detector watched a larger mass for a shorter period.
The event of interest was recorded at 21:22:39 UTC. The team reconstructed its energy as 248 keV, with statistical and systematic uncertainties of 23 keV each. It sat 26.4 centimetres above the cathode and more than 20 centimetres inward from the chamber wall.
Its position matters because wall events can lose charge and masquerade as something more signal-like. This interaction was comfortably inside the selected volume. The shape of its S2 pulse fit a point-like deposit at the reconstructed depth, and the way S1 light divided between upper and lower sensors agreed with that location.
The event lay 1.5 standard deviations below the middle of the calibrated nuclear-recoil band. It was 6.7 standard deviations below the middle of the electron-recoil band. Those numbers make a nuclear recoil the better fit, but the S1 pulse shape could not independently classify it with certainty.
Researchers kept asking how the detector could be fooled
Rare-event experiments become credible by distrusting their most interesting data. The LZ collaboration examined beta and gamma radiation, radon decay chains, radioactive isotopes in the xenon, solar neutrinos, neutrons from detector materials and accidental pairings of unrelated pulses.
Investigators also tested more peculiar event shapes. An interaction can deposit energy in two places while producing charge from only one, creating misleading S1 and S2 signals. Electric-field distortions can pull reconstructed wall events inward. Delayed signals can accidentally align with another flash. None of the modelled possibilities provided a likely explanation for this event’s full signature.
That conclusion has a precise boundary. It means no background process included in the analysis produced a convincing expected rate and signal pattern. It does not mean every possible ordinary mechanism has been ruled out. An exceptionally rare process may be missing from a model because no one has yet recognised or measured it.
The analysis includes another reason for restraint. Researchers normally hide the final search data or insert artificial “salt” events while fixing their cuts, reducing the chance that knowledge of an interesting point changes the procedure. Here, the artificial events did not adequately cover the newly examined high-energy region.
The collaboration fixed its selections and likelihood models before revealing the remaining salt, but describes the finished high-energy analysis as non-blind. That disclosure does not erase the event. It tells other physicists how much protection against unconscious analyst choices was available.
One point can favour many versions of dark matter
Weakly interacting massive particles are not one particle with one predicted signature. The label covers a broad family of hypothetical candidates. In the simplest models, a WIMP elastically strikes a nucleus and produces a spectrum weighted toward low recoil energies.
A 248 keV recoil sits unusually high for those standard searches. The new analysis tested a wider effective-field-theory framework, including interactions that depend more strongly on momentum transfer or nuclear spin. It also considered inelastic collisions in which the incoming particle changes into a heavier state.
If a WIMP produced this event, the collaboration estimates that its mass would probably be at least 200 times the proton’s mass. Several different combinations of mass and interaction could be made consistent with the same point. A lone recoil therefore cannot identify a particle even if its dark-matter origin were assumed.
Astronomers already have overwhelming evidence that unseen mass influences galaxies and galaxy clusters. ScienceBlog has previously covered how wide-field observations can map dark matter through its gravitational effects. LZ is attempting the more direct and difficult task of detecting an individual interaction by whatever particle supplies that mass.
Why the result is 2.6 sigma rather than 3.4
For the model that happened to fit best, the event reached a local significance of 3.4 sigma. “Local” means the calculation asks how unusual the result would be for that particular model at that particular part of parameter space.
The analysis tested many operators, particle masses and inelastic mass splittings. When enough possibilities are examined, random data have more chances to resemble at least one. This is the look-elsewhere effect, and ignoring it would make the most favourable result appear stronger than the search as a whole.
After simulating that wider search, the collaboration reported a global significance of 2.6 sigma. Institutional summaries translate this into a background-only fluctuation probability of about 0.5 per cent within the analysis framework.
That figure is not a 99.5 per cent probability that dark matter caused the event. A p-value asks how surprising the data are under a specified background model. It does not assign probabilities to every unmodelled background or calculate the chance that a chosen dark-matter theory is true.
Particle physics normally reserves “discovery” for results reaching five sigma, corresponding to a far smaller background-only tail probability. Even then, independent checks and a pattern of events would be needed to establish what had been found. At 2.6 sigma, the LZ result is a reason to investigate, not a reason to rename a particle.
Dahl’s excitement and caution belong together
Dahl’s description carries weight because outliers are routine in this work. Most become less mysterious when researchers examine their positions, veto coincidences or pulse shapes. This one continued to look valid as the collaboration tested increasingly unusual explanations.
His full position is more cautious than the memorable line alone. Detectors have repeatedly surprised researchers with behaviours they did not anticipate. LZ is much better understood than earlier instruments, but no team can prove it has imagined every way a complicated machine might generate an exceptional event.
The phrase “most interesting single event” also preserves the central limitation. A population has a spectrum, rate, spatial distribution and perhaps a change through the year. One point has none of those patterns. It can suggest models, but cannot test their wider predictions.
The experiment still has most of its planned exposure ahead
LZ is working toward 1,000 live days, so the 220-day sample represents only around 22 per cent of that goal. The detector has already collected more data than appear in this analysis, and future work can reopen the high-energy window with improved calibrations and a larger exposure.
If the event belongs to a genuine dark-matter population, additional recoils should eventually appear with a distribution consistent with the same interaction. Significance should grow as the detector accumulates exposure. Other experiments using different targets may also see compatible effects.
If it was a rare background, more data may reveal companions that expose the mechanism. If it remains alone, its statistical importance will generally weaken as clean observation time increases without a repeat. Either outcome teaches researchers something about the detector.
For now, LUX-ZEPLIN has produced exactly the kind of result that makes precision science uncomfortable in a productive way: too well checked to dismiss, too isolated to trust as a discovery. The single event deserves attention because ordinary explanations have struggled. It does not yet deserve an identity.