Key Points
ใปThe LZ collaboration announced on September 1, 2026 at a conference in Tendo, Yamagata Prefecture, Japan, that one event consistent with a nuclear recoil of 248 plus or minus 23 (statistical) plus or minus 23 (systematic) keV survived its analysis. The global significance is 2.6 sigma, with a local significance reaching 3.4 sigma. Particle physics reserves the word discovery for 5 sigma.
ใปThe event turned up because LZ widened its search window. Earlier analyses looked at nuclear recoil energies between 5.4 and 55 keV; this one extended to 270 keV to test inelastic and effective-field-theory models. The single event sits in the newly opened region.
ใปLZ also disclosed something unflattering. According to the collaboration’s preprint posted to arXiv on September 2, 2026, the salting procedure meant to prevent analyzer bias was unsuccessful, and the authors classify the work as a non-blind analysis.
LZ Reports One Event Its Known Backgrounds Cannot Account For
The LUX-ZEPLIN (LZ) dark matter experiment announced on September 1, 2026, at the TeV Particle Astrophysics 2026 conference in Tendo, Yamagata Prefecture, that it had observed a single event consistent with a dark matter particle striking a xenon nucleus. The preprint appeared on arXiv on September 2 and has been submitted to a physics journal.
According to the LZ collaboration’s own experiment documentation, LZ sits about 1,480 meters underground at the Sanford Underground Research Facility, a former gold mine in Lead, South Dakota, and uses liquid xenon as its target.
According to the LZ collaboration’s preprint, an analysis of 220 live days of data taken between March 27, 2023 and April 1, 2024, an exposure of 2.84 tonne-years, left one event at a nuclear recoil energy of 248 plus or minus 23 (statistical) plus or minus 23 (systematic) keV.
The collaboration reports a global significance of 2.6 sigma after correcting for the number of models tested, and a local significance of up to 3.4 sigma for individual models. The team examined neutrons, radioactivity in detector materials and other known backgrounds, and states that it could not identify a plausible explanation for the event.
Rick Gaitskell of Brown University, the LZ spokesperson, said in a Brown University statement dated September 1, 2026 that the collaboration is not claiming to have seen dark matter, but has seen something interesting enough to share. LZ has continued taking data under the same conditions since April 2024.
Related Articles
A Detector That Waits 1.5 Kilometers Underground for a Nucleus to Move
LUX-ZEPLIN (LZ) is a two-phase liquid xenon detector built to catch the rare moment when a dark matter particle strikes an atomic nucleus and knocks it loose. That knock is called a nuclear recoil, and it is the signature the experiment has spent years waiting for.
Dark matter is the name given to a form of matter that neither emits nor absorbs light but does exert gravity. Its existence is inferred from three independent observations: the flat rotation curves of spiral galaxies measured precisely by Vera Rubin and Kent Ford in the 1970s, the bending of light from distant galaxies by intervening mass, and the pattern of temperature fluctuations in the cosmic microwave background. Roughly 85 percent of the matter in the universe is thought to be dark matter. What remains unknown is which particle it is.
How the detector separates a signal from noise
One leading class of candidate is the WIMP, a weakly interacting massive particle. Such a particle would pass through the Earth constantly and interact almost never, but would very occasionally strike a xenon nucleus. LZ uses seven tonnes of active liquid xenon as the target and records two signals: a prompt flash of light where the collision happened, and a second flash produced when the electrons freed by the collision are drawn upward into a gas layer. The delay between the two gives the depth of the interaction.
The difficulty is that neutrons and gamma rays produce similar signals. That is why, according to the LZ collaboration’s experiment documentation and its September 2026 preprint, the detector sits under 1,480 meters of rock, wrapped in a two-tonne liquid xenon skin and an outer detector of gadolinium-loaded scintillator plus 229 tonnes of ultrapure water. When the outer layers catch a particle at the same instant as an inner signal, the inner event is discarded as intruding noise.
The event announced this month carried a recoil energy of 248 keV. That figure is the energy the nucleus received, not the mass of whatever hit it. According to the LZ preprint, a real signal at that energy would typically require a particle heavier than 200 GeV, more than 200 times the mass of a proton, and Nature’s September 2026 report on the result put the plausible range at 200 GeV and possibly around 1,000 GeV.
What 2.6 sigma actually says
Two numbers appear in the paper and they mean different things. LZ did not test one dark matter hypothesis. Per the LZ preprint posted to arXiv on September 2, 2026, the collaboration laid out a large family of models with different masses and interaction types, including the 15 independent operators of non-relativistic effective field theory, and fitted each. The best-fitting model gives the 3.4 sigma local significance the paper reports.
Hitting one target with one shot is not the same as lining up a hundred targets and pointing afterward at whichever one was struck. The second case has to be discounted for the number of chances taken. That discounted figure is the global significance of 2.6 sigma that LZ reports for the event, and it is the number to quote.
Sigma measures how unlikely a result is if nothing unusual is happening. A common way to state it goes wrong: this event is not 99.5 percent likely to be dark matter. Berkeley Lab published the figure as about 0.5 percent in its September 1, 2026 announcement, and the correct statement of it is that if only known backgrounds were present, data this discrepant would arise by chance that often. Particle physics calls 3 sigma evidence and 5 sigma a discovery, the standard CERN applied when it announced the Higgs boson in 2012, and 5 sigma corresponds to roughly one chance in 3.5 million.
The field has reason for that caution. Several signals that once drew attention have since faded.
| Experiment (year reported) | What was reported | Significance | What happened next |
|---|---|---|---|
| DAMA/LIBRA (1998 onward) | An annual modulation in the detection rate | About 12.9 sigma cumulative | Unresolved. ANAIS-112 and COSINE-100, using the same sodium iodide crystals, have failed to reproduce it, and ANAIS-112 reports a 3.5 to 4.0 sigma disagreement |
| CDMS-II (2013) | Three candidate events against an expected background below one | About 3 sigma | Faded. More sensitive later experiments excluded the region and additional data did not raise the significance |
| XENON1T (2020) | An excess of 53 events over expectation at low energy | 3.5 sigma | Explained. XENONnT did not reproduce it in 2022, and trace tritium contamination became the leading explanation |
Sources: the CDMS-II paper abstract on arXiv (2013), a 2025 review preprint on DAMA/LIBRA, and the XENONnT electronic recoil paper on arXiv (2022).
Japan is not in LZ, but Japanese detectors bound the interpretations
According to the LZ experiment’s own collaboration listing, no Japanese institution belongs to LZ, which comprises 39 institutions across the United States, the United Kingdom, Portugal, Switzerland, South Korea and Australia. The result was announced in Japan only because TeVPA 2026, one of the field’s main international meetings, was held in Tendo, Yamagata Prefecture, this year.
Japan did run its own liquid xenon dark matter search. XMASS operated 1,000 meters underground in the Kamioka mine in Gifu Prefecture from 2010 and ended observations in 2019. Researchers from the University of Tokyo, Nagoya University and Kobe University subsequently joined XENONnT at the Gran Sasso laboratory in Italy, one of the few detectors in the world capable of testing the same hypothesis independently.
Kamioka enters the story a second time. KamLAND, a 1,000-tonne liquid scintillator detector running there since 2002 for neutrino physics, has archival data that the fermionic dark matter absorption paper published on arXiv on September 1, 2026 uses to constrain one of the theoretical explanations offered for the LZ event, as described below.
Why One Event Moves a Field That Usually Waits for Thousands
Why does one event matter in a dark matter search?
The paper reports that of the 1,710 events that passed every selection cut in the science dataset, exactly one landed in the nuclear recoil region at high energy. The expected background there, as the LZ preprint of September 2, 2026 reports, is small: atmospheric neutrino coherent scattering contributes about 0.1 events, and the other channels examined fall below one event as well.
That is the difference between one event among a hundred expected and one event where the calculation says none should appear. Sam Eriksen of the University of Bristol, the paper’s lead author, said in the Berkeley Lab announcement of September 1, 2026 that a single outstanding event matters precisely because the detector and its backgrounds are so well understood. Aaron Manalaysay of Lawrence Berkeley National Laboratory, chair of the LZ institutional board, said in the same announcement that he had never before seen an outlier that looked reasonable from every angle.
The backgrounds the team examined include beta decays of radon daughters, radioactive xenon isotopes, neutrons from detector materials, and accidental coincidences in which two unrelated flashes happen to pair up. None provided a plausible account. That is not the same as saying no explanation exists, and with a single event no statistical test can separate an unknown background from a real signal.
Was the field looking in the wrong energy range all along?
According to the same preprint, LZ’s previous analysis covered nuclear recoil energies from 5.4 to 55 keV, because the simplest WIMP models predict that most collisions deposit little energy. Direct detection has concentrated on that low-energy band for decades without a confirmed signal.
According to the LZ collaboration’s preprint posted to arXiv on September 2, 2026, this analysis widened the window to 270 keV in order to test more complicated models. In inelastic scattering, where the dark matter particle converts into a heavier state during the collision, the signal is predicted to shift toward higher energies. The single event sits inside the newly opened band.
That invites the reading that dark matter was not missing, only being looked for in the wrong place. It remains one hypothesis among several. LZ also reports that the analysis set world-leading constraints across every model tested, which means the same dataset closed off far more possibilities than it opened.
LZ wrote in its own paper that the blinding failed
Large experiments guard against analyzer bias by injecting artificial signal-like events into the real data, fixing the selection criteria, and only then revealing and removing the fakes. LZ calls this salting. It is the equivalent of sealing the answer in an envelope and agreeing on the marking scheme before opening it.
The preprint states that the attempt was unsuccessful. The distribution used to generate the fake events was chosen before the detector calibration and did not adequately cover the high-energy signal region. LZ therefore classifies the work as a non-blind analysis, and says that above 55 keV its protection against bias rests on having reused the selection criteria fixed in the previous analysis without modification.
Some secondary coverage has described the event as having survived the removal of the injected fakes, which reads as though the safeguard worked. The paper says the opposite, and this is the point at which following second-hand reports goes wrong.
Publishing a failed safeguard rather than burying it reflects well on the collaboration’s practice. It does not raise the odds that the event is real. And even a successful salting would only have blocked the analysts’ hindsight, not guaranteed the absence of an unknown background. Only more data and independent experiments can address the second question.
Why did theory papers appear before the paper itself?
According to arXiv submission timestamps, two interpretation papers went up on September 1, 2026, the day of the announcement and before the LZ preprint itself appeared. A third followed on September 2. Their authors had begun calculating from conference slides.
The three point in similar directions. Elastic scattering, in which the particle simply bounces off, would produce many events at low energy, and none appear there. What fits is an inelastic process in which the particle converts to a heavier state, an entrance fee that only the most energetic collisions can pay, which pushes the signal upward. Across the three interpretation papers published on arXiv on September 1 and 2, 2026, the proposed masses cluster around 1 TeV, with mass splittings between roughly 300 and 380 keV. The candidates include a nearly pure higgsino, fermionic dark matter absorption, and endothermic scattering at the kinematic edge.
A pile of papers is not evidence that the signal is real. It shows a field testing which existing models can accommodate an anomaly. Fitting one data point is the easy part. Remaining consistent with decades of null results from detectors around the world is the hard part.
The papers demonstrate that difficulty themselves. The fermionic absorption paper states that a reinterpretation of KamLAND’s neutron emission data excludes its benchmark parameter region, and acknowledges a significant tension between the LZ interpretation and existing limits from large liquid scintillator detectors. The higgsino interpretation, by contrast, advertises that it has no free parameters to tune, which makes it straightforward to falsify. Theories offered in a form that can be refuted are the ones the next dataset will settle.
Japanese Reactions to the LZ Dark Matter Signal
What follows is a sample of posts on X, not a measure of Japanese public opinion. The posts are in Japanese and translations are given here in English. Engagement figures were read on September 4, 2026 and change over time.
The notable feature of the Japanese-language response is how little heat it carries. Science media and newspapers put the 2.6 sigma figure and the phrase this is not a discovery inside the same post as the headline, from the first day. The more expert the account, the heavier the caveats. The sarcasm that ran through some English-language threads, along the lines of another fifteen years of waiting, is largely absent here.
Japan’s main business daily carried the story with the reservation built into the same post.
Translation: A possible sign of the unidentified dark matter, reported at an international conference. Research groups in the United States and Britain have caught one particle interaction that known phenomena struggle to explain. The confidence level reaches only 2.6 sigma, and no conclusion has been drawn.
A Japanese science writer who explains research to a general audience summarized the finding and then applied the brake in the same breath.
Translation: A report from the LZ experiment, which hunts for WIMPs, the unknown particles thought to make up dark matter. They may have found a single signal produced by a dark matter particle colliding with an atomic nucleus. It does not yet reach the threshold for a discovery, but it is the most compelling hint so far.
The most detailed Japanese explainer came from an astronomer at a university, who walked readers from galaxy rotation curves through the detector’s discrimination principle to the gap between 2.6 and 5 sigma.
Translation (excerpt): This field is littered with two and three sigma hints that later vanished. And the energy of this event is higher than a textbook WIMP would produce. Both things are worth saying at once: it is genuinely interesting, and it is nowhere near settled.
One science communicator drew the whole result as a four-panel comic with the detectors as characters, working through the rejected candidates in the panels.
Translation (excerpt): LZ went looking for bigger collisions than usual, and found one suspicious signal. A neutron? It would have scattered more than once. A gamma ray that bounced twice? The numbers do not fit. One event is not enough to tell, so now everybody has to look harder.
One caveat about the shape of this sample. A search for the Japanese word for dark matter on X returns mostly game and anime usages, character names and a running joke about inedible cooking, so the accounts engaging with the actual result are almost entirely news outlets, researchers and science communicators rather than the general public.
The Question Is Not This Event but the Next One
A single event drawn from the 220 days of data the LZ collaboration reported on September 1, 2026 has moved a field that normally waits for thousands, because the region it landed in was calculated to be empty. That is the whole of the story, and it is also the limit of it.
LZ has been running under the same conditions since April 2024 and has already banked more data than it analyzed here. If second and third events appear at the same energy, in the spectral shape the theories predict, the candidate is promoted. If the next events scatter across unrelated energies, or nothing appears at all, a background or a statistical fluctuation becomes the likely account. XENONnT in Italy and PandaX in China can examine the same high-energy window, and Jianglai Liu of PandaX told Nature that the result is intriguing.
The history of this search is partly a history of signals that dissolved. The reason the field does not discard a single event is that the procedure for finding out is established and works. The most accurate way to read the 248 keV event is as the moment that procedure started running.
Sekahan on YouTube
We publish video summaries of articles like this one, along with short clips built around Japanese reactions.
Frequently Asked Questions
Did the LZ experiment discover dark matter?
No. The global significance is 2.6 sigma, well short of the 5 sigma that particle physics requires for a discovery and below the 3 sigma usually called evidence. According to the LZ collaboration’s preprint posted to arXiv on September 2, 2026, one event consistent with a 248 keV nuclear recoil survived all selection cuts and no plausible known background was identified for it. LZ spokesperson Rick Gaitskell said the collaboration is not claiming to have seen dark matter.
What does 2.6 sigma mean in the LZ dark matter result?
According to Berkeley Lab’s September 1, 2026 announcement of the result, it means that if only known backgrounds were present, data this discrepant would arise by chance roughly 0.5 percent of the time. It does not mean the event is 99.5 percent likely to be dark matter. The figure is the global significance, which discounts for the large number of models LZ fitted; the local significance for the single best-fitting model reaches 3.4 sigma.
Why is 248 keV unusual for a dark matter signal?
Because previous LZ analyses covered nuclear recoil energies only up to 55 keV, where the simplest WIMP models predict most collisions. According to the LZ preprint, the search window was extended to 270 keV in this analysis to test inelastic and effective-field-theory models, which shift the expected signal to higher energies, and a signal at 248 keV would typically require a particle heavier than 200 GeV.
How did Japanese social media react to the LZ dark matter result?
With unusual restraint. Japanese science media and newspapers published the 2.6 sigma figure and the caveat that this is not a discovery inside the same posts as the headline. Nikkei’s post noted that no conclusion had been drawn; a science writer called it the most compelling hint so far while stressing it falls short of the discovery threshold; and an astronomer pointed out that the field is littered with two and three sigma hints that later vanished and that the event’s energy is higher than a textbook WIMP would produce.
Reference Links
- Search for dark matter particle interactions in an extended nuclear recoil energy window with the LUX-ZEPLIN (LZ) experiment | arXiv
- LZ sees surprising result in search for dark matter | Berkeley Lab News Center
- Is this the first glimpse of dark matter? Data point excites physicists | Nature
- LZ experiment sees surprising result in search for dark matter | Brown University
- The LZ Dark Matter Experiment | LZ Collaboration
- Higgsino Dark Matter Interpretation of the LUX-ZEPLIN 248 keV Nuclear-Recoil Event | arXiv
- Fermionic Dark Matter Absorption and the High-Energy Event in LUX-ZEPLIN | arXiv
- Search for New Physics in Electronic Recoil Data from XENONnT | arXiv
