Sterile Neutrinos and the LSND/MiniBooNE Mystery

In 1995, the Liquid Scintillator Neutrino Detector (LSND) at Los Alamos reported an excess of electron antineutrino events at a baseline where the standard three-flavor oscillation framework predicted none. The anomaly was statistically modest — about 3.8 standard deviations — but physically remarkable: it required a mass-squared splitting of approximately $1$ eV², three orders of magnitude larger than the atmospheric splitting measured a few years later by Super-Kamiokande. No combination of the three known neutrino flavors could accommodate it.

The most economical explanation was a fourth neutrino species: a sterile neutrino that does not couple directly to the weak interaction, mixing with active neutrinos only through the PMNS-like structure. The idea itself was not new — sterile neutrinos appear naturally in seesaw models of neutrino mass — but LSND was the first experiment whose data suggested one might exist at a mass accessible to terrestrial experiments.

Thirty years later, the question of whether sterile neutrinos exist remains one of the most contested in particle physics. A parade of follow-up experiments has alternately supported and excluded the LSND interpretation. Parallel anomalies in reactor, gallium, and accelerator experiments have muddied rather than clarified the picture. Each new experimental result tightens certain parameter regions while opening others.

This essay is an attempt to describe the state of the sterile-neutrino question as of 2026 — what we know, what we don’t, and why the field has been unable to settle a question that seemed answerable three decades ago.

The LSND experiment

LSND operated at the Los Alamos Meson Physics Facility from 1993 to 1998. An 800-MeV proton beam struck a water-and-iron target, producing stopped pions that decayed to muons that, in turn, decayed at rest: ${\pi }^{+}\to {\mu }^{+}+{\nu }_{\mu }$ ${\mu }^{+}\to e^{+}+{\bar{\nu }}_{\mu }+{\nu }_e$ The resulting neutrino flux from ${\mu }^{+}$ decay contained ${\bar{\nu }}_{\mu }$ but essentially no ${\bar{\nu }}_e$. The ${\bar{\nu }}_e$ component was suppressed by approximately $10^{-3}$ relative to ${\bar{\nu }}_{\mu }$ — the ratio of electrons to muons in the decay products, since electron capture at rest is kinematically suppressed.

The detector was a 167-ton liquid scintillator tank 30 metres from the beam target. It searched for ${\bar{\nu }}_e$ interactions through the inverse beta decay reaction ${\bar{\nu }}_e+p\to e^{+}+n$ with the delayed-coincidence signature of a positron followed by a 2.2-MeV gamma from neutron capture. The beam’s ${\bar{\nu }}_e$ contamination produced a well-characterized background; any excess above this would be a signal of ${\bar{\nu }}_{\mu }\to {\bar{\nu }}_e$ oscillations over the 30-metre baseline.

LSND observed 88.6 ± 25.2 excess events — a 3.8σ signal. The inferred oscillation parameters, treating the result as a two-flavor oscillation with one new mass splitting, were: ${\sin }^{2}2\theta \approx 0.003,\Delta m^2\approx 1{\text{ eV}}^2$ This mass-squared splitting was incompatible with atmospheric ($\Delta m^2\sim 10^{-3}$ eV²) or solar ($\sim 10^{-5}$ eV²) splittings. A new oscillation frequency appeared to be operating at this short baseline.

The sterile-neutrino interpretation

The simplest way to accommodate the LSND result is to add a fourth neutrino mass eigenstate ${\nu }_4$ with mass-squared splittings of $\sim 1$ eV² from the other three. For this new state to exist without contradicting other experiments, it must not couple to the $Z$ or $W$ bosons — that is, it must be sterile. Its effect on oscillations comes through mixing with the active flavors in a 4x4 PMNS-like matrix.

Constraints on the number of active (non-sterile) neutrino species from LEP ($N_{\nu }=2.984\pm 0.008$) explicitly exclude a fourth active state at the $Z$ boson mass scale. A sterile state would evade this constraint by definition.

If sterile neutrinos exist with $\Delta m^2\sim 1$ eV², they would produce observable effects in several places:

• Short-baseline reactor experiments (a few metres to kilometres): ${\bar{\nu }}_e$ disappearance at the few-percent level
• Gallium-based solar neutrino calibration with radioactive sources: anomalous ${\nu }_e$ disappearance
• Short-baseline accelerator experiments with conventional neutrino beams: anomalous ${\nu }_e/{\bar{\nu }}_e$ appearance in a predominantly ${\nu }_{\mu }/{\bar{\nu }}_{\mu }$ beam
• Cosmological observables through additional relativistic species in the early universe: $N_{\text{eff}}\gtrsim 3$

The LSND anomaly was therefore followed by a wave of experiments designed specifically to test it.

MiniBooNE and the emerging picture

The MiniBooNE experiment at Fermilab ran from 2002 to 2019 with the specific goal of testing the LSND signal. It used an 8-GeV proton beam, a decay-in-flight neutrino source, and a 450-ton mineral-oil Cherenkov detector at 541 metres baseline. The combination of beam and baseline was designed such that MiniBooNE’s $L/E$ was matched to LSND’s within the relevant parameter space.

MiniBooNE’s results were complicated. It observed an excess of electron-like events in both neutrino mode and antineutrino mode, at $4.5\sigma$ combined significance. But the kinematic distribution of the excess was not what a simple two-flavor oscillation would predict: the events were concentrated at the low-energy end of the spectrum, below about 400 MeV. Within the standard LSND-sterile interpretation, MiniBooNE should have seen oscillation-shaped excesses at higher energies.

Alternative explanations were proposed: the excess could reflect mis-identified photons from neutral-current interactions (a “photon-like” signal mimicking the “electron-like” selection in Cherenkov reconstruction). MiniBooNE itself could not distinguish electrons from photons in its detector. If the signal was actually photons, a completely different mechanism would be needed to explain it.

The MicroBooNE verdict

The MicroBooNE experiment, a liquid argon time projection chamber (LArTPC) at Fermilab, began operation in 2015 with the explicit goal of resolving the MiniBooNE photon-versus-electron ambiguity. LArTPCs provide excellent particle identification — electron tracks are distinguishable from photon showers on the basis of track width, ionization density, and conversion length.

Over four papers published between 2021 and 2023, MicroBooNE analyzed its data and found:

• No excess of electron events at the location MiniBooNE would have predicted
• No excess of photon events either
• The data are compatible with Standard Model backgrounds within statistical uncertainty

The MicroBooNE result is widely interpreted as excluding the simplest LSND-like sterile-neutrino interpretations of the MiniBooNE anomaly. Extensions of the sterile-neutrino model to more complex frameworks (3+2 or 3+1+decay scenarios) remain viable but are significantly constrained.

The LSND result itself, however, is in a sense untouchable: the experiment was decommissioned 25 years ago, and the same configuration cannot be rerun. The LSND signal remains an unresolved historical anomaly.

The reactor and gallium anomalies

Two parallel threads of evidence for sterile neutrinos have evolved independently of LSND.

The reactor antineutrino anomaly. In 2011, a reanalysis of reactor ${\bar{\nu }}_e$ flux predictions found that the measured rates at short baselines (10–100 metres) were about 6% below the expected rates. If interpreted as oscillation, this suggested ${\bar{\nu }}_e$ disappearance into a sterile state at $\Delta m^2\sim 1$ eV² — remarkably similar to LSND. A series of dedicated short-baseline reactor experiments (PROSPECT, STEREO, DANSS, NEOS) then ran to test whether the disappearance was oscillatory (L-dependent) or a flat systematic effect.

As of 2023, the combined short-baseline reactor data strongly disfavor oscillation at the originally suggested parameters. The PROSPECT experiment at the HFIR reactor in Oak Ridge published definitive exclusion curves in 2023: the LSND-favored region is excluded at 95% CL. The source of the 6% flux deficit is now widely believed to be an error in the reactor-neutrino flux predictions (specifically the ${}^{235}$U spectrum), not oscillation.

The gallium anomaly. The SAGE and GALLEX experiments calibrated their gallium-based solar neutrino detectors by exposing them to intense ${}^{51}$Cr and ${}^{37}$Ar radioactive sources, then measuring the neutrino capture rate. In both experiments, the measured rate came in roughly 20% below the predicted rate — a discrepancy that has persisted across multiple calibration campaigns.

In 2022, the BEST experiment at Baksan (Russia) repeated the gallium calibration with improved systematics and reported the same 20% deficit at $4\sigma$ significance. Interpreted as oscillation, this would again imply sterile neutrinos at $\Delta m^2\sim 1$ eV², with a particularly large mixing angle.

The gallium anomaly is currently the strongest surviving piece of evidence for sterile neutrinos. It has resisted several attempts at conventional explanation (cross-section corrections, nuclear matrix element re-evaluation). A clean resolution will require either a new independent measurement (perhaps with a different target) or acceptance that the gallium signal is real.

Cosmological constraints

If sterile neutrinos with $m\sim 1$ eV and substantial mixing exist, they should have thermalized in the early universe and contributed to $N_{\text{eff}}$ at BBN and CMB. Planck data constrain $N_{\text{eff}}<3.4$ at 95% CL, which in turn constrains the combination of sterile mass and mixing.

Combined cosmology-plus-terrestrial-constraint analyses tend to disfavor the LSND sterile-neutrino scenario at high confidence: the required mixing would have equilibrated the sterile state before BBN, leading to a clearly observable CMB distortion that is not seen.

Modifications to the cosmological model (secret sterile-neutrino interactions, non-standard thermal histories) can evade these constraints but require additional new physics.

The current consensus (such as it is)

As of 2026, the field is in the following tense state:

• LSND remains unexplained. No direct repeat of its specific configuration exists.
• MiniBooNE’s excess is now widely believed to be an identification issue, not a sterile-neutrino signal.
• Reactor anomaly is attributed to flux-prediction errors.
• Gallium anomaly remains a real open discrepancy.
• Global fits to all short-baseline data show severe tension between appearance and disappearance channels — no single sterile-neutrino parameter choice fits everything.
• Cosmology disfavors thermalized eV-scale sterile neutrinos, but non-thermal and secret-interaction models remain viable.

The sterile-neutrino hypothesis is not dead, but the simplest versions are disfavored. Complex models survive, at the cost of theoretical elegance. A clean experimental resolution would require a new measurement that definitively confirms or refutes one of the surviving anomalies — most likely the gallium signal — without introducing new tensions.

The open experimental program

Several current and proposed experiments could close the question:

Short-baseline neutrino programme (SBN) at Fermilab. Combines three LArTPCs (SBND, MicroBooNE, ICARUS) at different baselines in the same neutrino beam. Their combined analysis, expected to be released in stages through 2026–2028, will constrain sterile neutrinos across the LSND parameter space with orders-of-magnitude better sensitivity than any single experiment.

DUNE near detector. The DUNE near detector, 574 m from the beam source, will also probe short-baseline oscillations with LArTPC precision.

Source experiments with chromium or other isotopes. Repeating the gallium calibration with a different isotope (or different target) would test whether the gallium anomaly survives a change of systematics.

IsoDAR and other reactor-like sources. Dedicated sterile-search beams with known flux and controlled baselines.

By the end of the decade, at least one of these programmes should have produced a result definitive enough to settle the sterile question. Until then, the field lives with a low-grade anomaly that has survived longer than any other open question in neutrino oscillation physics.