MiniBooNE: An Experiment That Refused to Settle Down

By the late 1990s, the LSND result at Los Alamos had presented the field with an awkward problem. A 3.8σ excess of electron antineutrinos at a 30-metre baseline, with parameters incompatible with the three-flavor framework. Either it was a statistical fluke, a systematic error in the experiment, or evidence of a new neutrino — sterile, with a mass-squared splitting around 1 eV². The community needed a clean test.

That test became MiniBooNE — the Mini Booster Neutrino Experiment at Fermilab. A 450-tonne mineral-oil Cherenkov detector at 541 metres baseline from an 8-GeV proton beam, designed specifically to probe the same oscillation parameters as LSND with a different energy and a different baseline. The L/E was the key: matched to LSND’s, so any oscillation phenomenon that produced LSND’s signal should also produce a MiniBooNE signal.

The experiment ran from 2002 to 2019. What it saw became one of the most discussed and contested results in modern neutrino physics. An excess of electron-like events, but at low energies that didn’t fit a simple oscillation pattern. After 17 years of data and analysis, the question was no longer “does the LSND result reproduce” but “what does this excess actually mean”.

The story is, in many ways, the model case for how a clear hypothesis test can produce a result that is itself ambiguous — and how the community handles that.

The original logic

LSND ran at the Los Alamos Meson Physics Facility from 1993 to 1998. Its source was stopped pions decaying at rest, producing a beam dominated by ${\nu }_{\mu }$ and ${\bar{\nu }}_{\mu }$ with energies up to about 50 MeV. Its detector, 30 metres downstream, looked for electron antineutrinos via inverse beta decay. It saw 88 events excess over background, fitted to $\Delta m^2\sim 1$ eV² and ${\sin }^2(2\theta )\sim 0.003$.

For MiniBooNE, the design philosophy was to match LSND’s L/E with completely different beam energy and detector technology. If oscillation was real:

• LSND: $L=30$ m, $E\sim 30$ MeV → $L/E\sim 1$ m/MeV
• MiniBooNE: $L=541$ m, $E\sim 600$ MeV → $L/E\sim 1$ m/MeV

Same L/E, totally different absolute scales. A consistent oscillation interpretation would predict roughly the same fractional appearance probability at both experiments. Different oscillation parameters wouldn’t.

The MiniBooNE detector — 12.2 metres in diameter, filled with 800 tonnes of pure mineral oil (used as both target and scintillator), surrounded by 1,280 inward-facing 8-inch photomultipliers — was designed to identify ${\nu }_{\mu }\to {\nu }_e$ and ${\bar{\nu }}_{\mu }\to {\bar{\nu }}_e$ appearance. The signature: an electron event in the appropriate energy range, against a background of ${\nu }_{\mu }$ charged-current and neutral-current events.

The 2007 first result

Initial neutrino-mode running, 2002 through 2007, was analyzed and published in 2007. The result: 380 ± 19 (stat) ± 35 (syst) electron-like events observed against 358 ± 19 (stat) ± 35 (syst) background expected. Above 475 MeV, the data matched the no-oscillation hypothesis. Below 475 MeV — in a “low-energy excess” region — there was a 3σ excess of about 100 events.

The puzzle was immediate. The standard sterile-neutrino interpretation of LSND would predict an excess at all energies in the MiniBooNE acceptance, peaked roughly in the middle of the spectrum. MiniBooNE saw an excess only at the bottom of the range. Either the oscillation parameters didn’t match LSND’s, or the excess was something other than oscillation, or the LSND result was a statistical fluctuation.

The collaboration’s official position: above 475 MeV, the data is consistent with no oscillation. Below 475 MeV, an unexplained low-energy excess is observed. The excess might be background, but doesn’t appear to fit oscillation either.

The antineutrino runs

From 2008 to 2014, MiniBooNE ran in antineutrino mode. The motivation: LSND saw the excess in antineutrino appearance, so a clean test should run with antineutrinos too. Antineutrino running was technically harder — wrong-sign contamination is larger, and statistics are about a factor of 2 lower for the same beam exposure.

The 2014 antineutrino-mode result: another 78 events of excess at 2.8σ, this time spread more broadly across the energy range. Neutrino-mode + antineutrino-mode combined gave a 4.5σ excess, but with the signature heavily dependent on which energy bins you included.

Various analyses tried to extract a coherent oscillation interpretation. Some sterile-neutrino models could fit the combined data, but only at parameters that contradicted other constraints (atmospheric data, IceCube limits, cosmology). No single, simple sterile-neutrino picture worked.

The photon hypothesis

MiniBooNE’s principal weakness was its inability to distinguish single electrons from single photons. Both produce electromagnetic showers in mineral oil that look essentially identical to a Cherenkov detector. The “electron-like” sample was therefore an “EM-shower-like” sample, possibly contaminated by mis-reconstructed photons.

If the low-energy excess was actually photons from poorly-modeled background processes, the entire interpretation would change. The candidate processes: neutral-current radiative photon emission (where a ${\nu }_{\mu }$ knocks out a single photon), single photon production from $\Delta (1232)$ resonance decay, and similar non-oscillation backgrounds.

By 2018-2019, it was clear that distinguishing electrons from photons in the MiniBooNE data required either an experimental upgrade or a follow-up experiment with better technology. The collaboration could not internally resolve the ambiguity.

MicroBooNE enters

MicroBooNE was a liquid-argon time-projection chamber at Fermilab, deployed at the same baseline as MiniBooNE on the same beam. Its TPC technology gave sub-millimetre tracking and electron/photon discrimination at the per-event level. An electron leaves a track that ionises uniformly along its length; a photon converts to an electron-positron pair after traversing some distance and produces a track-cluster of two particles emerging from the conversion vertex.

Between 2021 and 2024, MicroBooNE published a series of papers analyzing different signal hypotheses for the MiniBooNE excess:

Single-electron scenario — most direct mapping to oscillation. MicroBooNE’s data showed no excess of single-electron events at the location MiniBooNE saw an excess. This essentially rules out the simplest oscillation interpretation of MiniBooNE.

Single-photon scenario — neutral-current radiative photons. MicroBooNE’s data showed no excess of single-photon events either. The hypothesis that the MiniBooNE excess was misidentified photons from this process is also disfavored.

Multi-particle scenarios — final states with multiple electromagnetic showers. MicroBooNE’s discrimination is good for these too, and again no excess.

The verdict: at the location and energy range where MiniBooNE saw an excess, MicroBooNE sees neither electrons nor photons in any plausible configuration. The MiniBooNE excess does not match any of the natural new-physics interpretations.

Where this leaves things

The MiniBooNE result is, as of 2026, in an unusual position. It is statistically real (4.5σ combined excess) and reproducible by reanalysis of the same data. But it cannot be explained by any of the standard hypotheses (sterile-neutrino oscillation, single-photon backgrounds, single-electron new physics). MicroBooNE has effectively excluded the obvious explanations.

What remains:

• Mismodeled backgrounds at low energy in MiniBooNE specifically, perhaps due to detector or particle-identification issues that don’t apply to MicroBooNE. Several specific candidates (neutron-induced backgrounds, secondary interactions in the oil) have been examined but none individually accounts for the full excess.
• Exotic models — for example, sterile neutrinos with additional decay channels, or non-standard interactions that produce excesses preferentially at low energies. These are model-dependent and not particularly compelling.
• Statistical fluctuation — at 4.5σ this is unlikely but not impossible, especially if background uncertainties are larger than estimated.

The community consensus, by 2024, is that the MiniBooNE excess is most likely a misunderstood detector or background effect specific to the mineral-oil + Cherenkov technology, rather than new physics. But “most likely” is not “definitively”. A specific, identified explanation remains elusive.

What the saga teaches

Several methodological lessons.

Detector technology matters more than statistics. MiniBooNE accumulated more events and ran longer than MicroBooNE, but MicroBooNE’s superior particle identification settled the question. Statistical precision can’t substitute for systematic clarity.

Anomalies should be cross-checked with independent technology. The MiniBooNE-MicroBooNE comparison is the model: the same physical question (does the LSND excess reproduce?) addressed by two different detector approaches. The answer is more informative than either alone.

Persistent anomalies don’t have to be new physics. The MiniBooNE excess is a real statistical effect, but the most likely explanation now is detector physics rather than fundamental physics. The same logic applies to other anomalies: the gallium anomaly, the reactor anomaly, ANITA anomalies. Each requires careful systematic investigation before it can be claimed as new physics.

Closing an anomaly takes longer than opening it. LSND was a 5-year experiment. MiniBooNE ran for 17 years. MicroBooNE took 4 more years to close the question. Resolution of the LSND saga has therefore taken close to 30 years.

For sterile-neutrino physics specifically, MiniBooNE’s resolution effectively closes one major anomaly while leaving the gallium anomaly, the LSND result itself (which can’t be re-tested without rebuilding the experiment), and short-baseline reactor measurements as remaining open questions. The Short-Baseline Neutrino programme at Fermilab — combining SBND, MicroBooNE, and ICARUS — will provide the next generation of clean tests through the late 2020s.

The neutrino is, after seventy years, still capable of producing experimental results that take a generation to settle. That MiniBooNE’s excess is now most likely not new physics is itself a kind of progress: it tells us where to look next, and where not to.