The fourth neutrino was excluded at 99.99%. The anomalies are still there.

Two experiments built to test the same thirty-year-old idea reported within days of each other in December 2025, and both came back empty. MicroBooNE excluded the sterile-neutrino explanation of the LSND and MiniBooNE anomalies at 95% confidence. KATRIN found no kink in 36 million tritium decay electrons and rejected the field’s strongest positive claim at 99.99%. The eV-scale sterile neutrino is now effectively closed. The keV and GeV versions, which are the ones that do real work in cosmology, remain barely tested.

A particle defined by what it does not do

The Standard Model contains three neutrinos, and they are defined by a behaviour: they couple to the W and Z bosons. A sterile neutrino is a neutral lepton that does not. It carries no weak charge, so a beam of them passes through a detector and leaves nothing behind. MicroBooNE’s Nature paper defines it simply as an extra neutrino state that does not interact directly with matter.

That makes it undetectable head-on. What makes it findable is mixing. If a sterile state has mass, it can share quantum identity with the three active neutrinos, so an electron neutrino travelling a few metres has some probability of turning into something no detector can see. In the minimal “3+1” model the mixing matrix grows to 4×4, and one new mass splitting, Δm²₄₁, sets the oscillation length.

Two consequences follow, and they structure everything below. First, all evidence is indirect: a deficit where none should be, a distortion in a spectrum, a photon from a decay. Second, “sterile” describes an interaction rather than a mass, so this is not one hypothesis but a family of them spanning fifteen orders of magnitude. The eV-scale version was invented to fit anomalies. The keV version is a dark-matter candidate. The GeV version can generate the matter-antimatter asymmetry. The 10⁹ to 10¹⁵ GeV version is what the classic seesaw asks for and no experiment can reach. A null result at one scale says almost nothing about another.

Four anomalies pointing at roughly the same place

Nobody proposed the eV-scale sterile neutrino for its elegance. It was proposed because several independent classes of experiment counted fewer, or more, neutrinos than they should have, at baselines and energies where three-flavour oscillation predicts nothing at all.

• LSND (1990s). An excess of electron antineutrinos in a muon antineutrino beam, at about 30 m and tens of MeV. Never confirmed. KARMEN saw no excess in the same channel.
• MiniBooNE (2007–2021). A surplus of electron-like events in a muon-flavour beam at about 540 m and roughly 1 GeV. The excess stands. Its sterile interpretation does not.
• The reactor antineutrino anomaly (2011). Measured reactor flux below prediction at 10 to 100 m. Reported at 2.5σ in 2011, reduced to roughly 1σ by the 2021 flux models, and revived at 2.2σ by a 2026 reanalysis using the 2023 summation model.
• The gallium anomaly (1990s, revisited by BEST in 2022). An electron neutrino deficit from a ⁵¹Cr source at two baselines, 0.4 m and 0.8 m. BEST measured a rate ratio of 0.80 ± 0.05, and the combined gallium significance exceeds 5σ. The combined best fit is sin²2θ_ee ≈ 0.34, at Δm² ≈ 1.25 eV² with no upper bound on the mass splitting.
• Neutrino-4 (2019–2022). A claimed positive oscillation pattern at reactor baselines of metres, with m₄ ≈ 2.70 ± 0.22 eV. The strongest positive claim in the field.

Individually these are weak: a few standard deviations each, none independently reproduced. What kept them alive was the coincidence. Four unrelated setups appeared to point at roughly the same mass splitting. What has undermined them is that the coincidence has never survived a purpose-built test.

MicroBooNE: two beams into one detector

MicroBooNE is a liquid-argon time projection chamber at Fermilab, sitting about 70 m closer to the beam target than MiniBooNE and taking data from 2015 to 2021. Argon TPCs can separate an electron from a single photon or a π⁰ with a precision MiniBooNE’s mineral oil could not reach, which matters because photons faking electrons is exactly the background that could produce a false appearance signal.

The decisive move was to run two beams into one detector: the Booster Neutrino Beam and NuMI. That breaks the degeneracy between electron neutrino appearance and electron neutrino disappearance, which would otherwise blunt the sensitivity of either beam alone. The collaboration describes it as the first sterile search done with one detector and two simultaneous beams.

The single light sterile interpretation of LSND and MiniBooNE is excluded at 95% CL. A significant portion of the gallium parameter space is ruled out as well. An earlier 2025 analysis in Physical Review Letters, searching for charged-current electron neutrino events without visible pions, found no anomalous excess either.

KATRIN: no beam, no baseline, no kink

KATRIN attacks the same question with none of the same equipment. There is no beam and no baseline: just a tritium source and the world’s most precise electron spectrometer, reading the last 40 eV below the 18.57 keV decay endpoint. Campaigns KNM1 to KNM5, running from 2019 to 2021, delivered 36 million electrons over 259 measurement days at sub-percent accuracy.

If a fourth mass state exists, it is emitted occasionally in place of the light ones, leaving a kink in the spectrum at E₀ minus m₄ plus a global distortion set by the mixing. The signature lives in curvature, which means the analysis depends on spectral shape rather than absolute normalisation. That is precisely the weakness that made the reactor flux anomaly so hard to interpret, and KATRIN does not share it.

No kink was found. The Neutrino-4 best-fit point is rejected at 99.99% CL, with its favoured region fully covered. The combined gallium best fit is excluded at 96.56% CL. Taken together with other disappearance experiments, KATRIN’s reach runs from a fraction of an eV² to several hundred eV², and the collaboration states that mixing angles above a few percent are excluded across that range.

Why the pair matters more than either result

These two experiments fail in unrelated ways, which is the reason to take them together. Reactor searches are sharpest below a few eV² and depend on knowing the flux they started with. KATRIN covers a few to several hundred eV² and needs no flux model at all. A claim only counts when two methods with unrelated systematics agree, and here they agree on nothing being there.

What survives in the eV band is the small-mixing corner: parameter space too weakly coupled to have produced the anomalies in the first place. A sterile neutrino could still be hiding there. It could not have caused LSND.

What did not go away

The anomalies outlived their explanation. The MiniBooNE excess is unexplained. The gallium deficit was measured by BEST at two baselines and did not shrink; if a sterile neutrino is not the cause, then something is wrong in the cross-section, the source calibration, or the ⁷¹Ge counting, which is itself a result worth having.

The reactor anomaly has moved back up. A May 2026 reanalysis using the 2023 French summation flux model, the first with a comprehensive uncertainty budget, revives it at 2.2σ. The same work finds a 3.8σ tension between the gallium data and everything else once solar, KATRIN, reactor spectral-ratio and Daya Bay constraints are folded in, falling to 1.3σ if the gallium uncertainties are enlarged.

The theoretical case never rested on the anomalies

This is the part the null results do not touch. The reason to want a right-handed neutrino was never LSND. It was the fact that neutrinos have mass at all.

The Standard Model as written has no right-handed neutrino and no neutrino mass term. Oscillation experiments proved neutrinos have mass twenty-five years ago, so the model is already known to be incomplete, and the only question is what to add. The cheapest addition is a fermion that is a singlet under every Standard Model gauge group: no colour, no weak isospin, no hypercharge. It needs no new force and no new energy scale, and because it is a singlet it is allowed a Majorana mass of its own, which no other Standard Model fermion is.

That one property drives the rest. In the type-I seesaw the light neutrino mass comes out as roughly m_ν ≈ y²v²/M, the Higgs scale squared over the heavy Majorana mass, with y the Yukawa coupling. Smallness becomes automatic rather than fine-tuned: neutrinos are light because their partners are heavy. But the equation has two unknowns. Order-one Yukawas put M near 10¹⁵ GeV; Yukawas of 10^-7 put M near a GeV. Both reproduce the observed 0.05 eV. Nothing in the theory picks the scale, which is why excluding one decade barely dents the framework.

The neutrino-minimal Standard Model, the νMSM, adds three right-handed neutrinos and nothing else. One at keV mass is the dark matter. Two nearly degenerate GeV states generate the matter-antimatter asymmetry through CP-violating oscillations before the electroweak transition. All three supply the active neutrino masses through the seesaw. The constraint that makes the model interesting is also what makes it hard to test: the mixing must stay small, or the states equilibrate too early and the asymmetry washes out. The model’s own structure predicts that nothing will be easy to see.

The 3.5 keV line, squeezed by a factor of five

A keV sterile neutrino decays to a photon of half its mass, so a 7.1 keV particle emits at 3.55 keV. In 2014 an unidentified line near that energy was reported at 4 to 5σ in stacked XMM-Newton observations of 73 galaxy clusters, with a consistent signal from Perseus. It was contested immediately: highly ionised potassium and sulfur charge exchange both produce lines nearby that low-resolution CCDs cannot separate.

Hitomi reached Perseus with a microcalorimeter in 2016 and failed after weeks in orbit, setting a limit instead of making a measurement. XRISM’s Resolve instrument has now stacked 3.75 Ms across ten clusters from 2.5 to 15 keV in the rest frame and found no unidentified lines. Its 3σ limit on the decay rate of a 7.1 keV dark-matter particle is about 1.0 × 10^-27 s^-1, three to four times below Hitomi’s Perseus limit, and still a factor of five above the 2014 claim.

The honest reading of that null result: the claim has not been refuted, it has been squeezed and left standing in a corner. Resolve’s spectral resolution is sufficient and its field of view is small, so the remaining factor of five is a question of accumulating cluster observations over several more years.

Whether the keV candidate is excluded depends on how it was made, and observation cannot fix that independently. In the Dodelson-Widrow mechanism active neutrinos oscillate into steriles as the plasma cools, leaving a suppressed non-thermal population. In the Shi-Fuller variant a pre-existing lepton asymmetry drives the mixing through a resonance and produces a colder spectrum for the same abundance. Non-resonant production of all of the dark matter is essentially closed by X-ray limits combined with structure-formation bounds. Resonant production, and models where the sterile neutrino is only part of the dark matter, remain open.

Cosmology has no room for a thermalised fourth neutrino

A fourth species that reached thermal equilibrium before neutrino decoupling would add one unit to the radiation density, predicting N_eff ≈ 4.04. Planck with BAO measures 2.99 ± 0.17, against a Standard Model prediction of 3.044. There is no room.

So the eV sterile that would have fitted the anomalies has to be prevented from thermalising, by secret self-interactions, low reheating, or a large lepton asymmetry. Each fix adds parameters to rescue a hypothesis whose original evidence has now evaporated. What cosmology constrains is production history, not existence.

What would change the picture

The field is not winding down. It is changing target, and almost every programme below moves away from the band that has just been closed.

• KATRIN’s full dataset, now in analysis, holds more than 220 million electrons in the region of interest, over six times the published statistics. It will close or confirm the small-mixing corner that survives.
• TRISTAN, a KATRIN detector upgrade from 2026, bypasses the main spectrometer to read the full differential beta spectrum rather than the endpoint. It would be the first laboratory search for keV-scale dark-matter steriles, with projected sensitivity to mixing as low as one part per million.
• The Short-Baseline Neutrino programme at Fermilab, with ICARUS running since 2021 and SBND since 2024, puts near and far liquid-argon detectors on the same beam. The near/far comparison removes flux modelling entirely and tests models MicroBooNE could not: 3+2, sterile decay, non-standard interactions.
• XRISM, accumulating. Further stacked cluster spectra should reach the sensitivity that tests the 2014 line directly.
• Heavy-neutral-lepton searches at beam dumps, in kaon and beauty decays, and at future electron-positron colliders. A displaced vertex in the GeV window would be the seesaw itself, not an anomaly fix.
• Neutrinoless double beta decay at ton scale in germanium and xenon. It establishes whether neutrinos are Majorana, the property the whole framework rests on, independent of any sterile mass.

Bottom line

Thirty years of short-baseline anomalies pointed at an eV-scale fourth neutrino. Two experiments built specifically to test that idea have closed it from opposite methodological directions, and nothing about that verdict is ambiguous. What it does not touch: neutrinos still have mass, the Standard Model still cannot explain it, and a gauge-singlet fermion is still the cheapest fix. The keV window remains the best-motivated dark-matter candidate that also explains something else. Watch TRISTAN’s keV search, XRISM’s next factor of five, and any displaced vertex in the GeV window. The eV chapter is closed. The particle is not.

References

• MicroBooNE Collaboration, Search for light sterile neutrinos with two neutrino beams at MicroBooNE, Nature 648, 64–69 (3 December 2025); preprint arXiv:2512.07159
• KATRIN Collaboration, Sterile-neutrino search based on 259 days of KATRIN data, Nature 648, 70 (3 December 2025); preprint arXiv:2503.18667
• MicroBooNE Collaboration, Search for an anomalous production of charged-current ν_e interactions without visible pions across multiple kinematic observables, Phys. Rev. Lett. 135, 081802 (2025); preprint arXiv:2412.14407
• C. Giunti, Y.-F. Li, R.-P. Zhang, Revival of the Reactor Antineutrino Anomaly, arXiv:2605.10353 (11 May 2026, preprint)
• C. Giunti, Y. F. Li, C. A. Ternes, Z. Xin, Reactor antineutrino anomaly in light of recent flux model refinements, Phys. Lett. B 829, 137054 (2022); arXiv:2110.06820
• V. V. Barinov et al. (BEST), Search for electron neutrino transitions to sterile states in the BEST experiment, Phys. Rev. C 105, 065502 (2022); arXiv:2201.07364
• XRISM Collaboration, Constraints on unidentified X-ray emission lines, including the 3.5 keV line, in the stacked spectrum of ten galaxy clusters, ApJL 994, L28 (2025); preprint arXiv:2510.24560
• Hitomi Collaboration, Hitomi constraints on the 3.5 keV line in the Perseus galaxy cluster, ApJL 837, L15 (2017); arXiv:1607.07420
• Planck Collaboration, Planck 2018 results. VI. Cosmological parameters, A&A 641, A6 (2020); arXiv:1807.06209. Source of N_eff = 2.99 ± 0.17
• B. Dasgupta & J. Kopp, Sterile neutrinos, Phys. Rep. 928, 1 (2021); arXiv:2106.05913
• D. Gorbunov, M. Shaposhnikov et al., Search for GeV-scale sterile neutrinos responsible for active neutrino oscillations and baryon asymmetry of the Universe, arXiv:1301.5516
• M. Drewes et al., A White Paper on keV sterile neutrino dark matter, JCAP 01 (2017) 025; arXiv:1602.04816
• A. Boyarsky, M. Drewes, T. Lasserre, S. Mertens, O. Ruchayskiy, Sterile neutrino dark matter, Prog. Part. Nucl. Phys. 104, 1 (2019); arXiv:1807.07938
• O. Benevides Rodrigues et al., Towards a robust exclusion of the sterile-neutrino explanation of short-baseline anomalies, arXiv:2503.13594 (preprint; source of the LSND significance in Figure 2)