PROSPECT, STEREO, and the Short-Baseline Reactor Programme

The reactor antineutrino anomaly emerged in 2011 from a careful reanalysis of historical data. Earlier short-baseline reactor experiments — Bugey, Goesgen, others from the 1980s — had measured antineutrino fluxes that, when reanalysed against updated theoretical predictions, fell systematically short of expectation by 5-6%. Combined with the LSND/MiniBooNE and gallium anomalies, the deficit suggested the possibility of a fourth, sterile, neutrino.

If true, the implications would be profound. A sterile neutrino at $\Delta m^2\sim 1$ eV² with ${\sin }^2(2\theta )\sim 0.1$ would represent the first observed beyond-Standard-Model particle. It would also raise questions about cosmological consistency (a thermalized eV-scale sterile would shift $N_{\text{eff}}$, contradicting CMB measurements), about the mass-generation mechanism (no satisfactory theoretical framework for an eV sterile is widely accepted), and about supernova nucleosynthesis (sterile-active oscillation in the supernova environment would alter the r-process in observable ways).

So the question demanded a clean experimental test. By 2017, three new experiments at small research reactors were taking data: PROSPECT at the High Flux Isotope Reactor in Oak Ridge, STEREO at the ILL reactor in Grenoble, and DANSS at the Kalinin power plant in Russia. By 2020, two more had joined: NEOS at Hanbit and PROSPECT-II in advanced design.

The results, accumulated and combined through 2023, are remarkably consistent. The LSND-favored region of sterile-neutrino parameter space is excluded. The reactor anomaly remains, but it is not due to sterile-neutrino oscillation. The most plausible explanation is errors in the predicted antineutrino flux from ²³⁵U specifically — a cleaner picture than was available a decade ago, and one that has substantially clarified the broader sterile-neutrino landscape.

This post walks through the experiments, what they measured, and what their combined verdict means.

The hypothesis to test

In a sterile-neutrino model with one additional flavor ${\nu }_4$ and a mixing parameter ${\theta }_{14}$, the survival probability of an electron antineutrino is: $P({\bar{\nu }}_e\to {\bar{\nu }}_e)=1-{\sin }^2(2{\theta }_{14}){\sin }^2\left(\frac{\Delta m_{14}^{2}L}{4E}\right)$

For $\Delta m^2\sim 1$ eV² and $E\sim 4$ MeV (typical reactor antineutrino energy), the first oscillation maximum is at $L\approx 1$ metre. So if the LSND-favored sterile exists, the short-baseline survival probability should oscillate as a function of distance — with a wavelength of just a few metres.

Standard reactor experiments at long baselines (KamLAND at 180 km, Daya Bay at 1.6 km) average over many oscillation cycles and see only the oscillation-averaged deficit. A short-baseline experiment, with the source close enough to resolve the oscillation pattern, sees the deficit modulated by distance.

The cleanest test: place a detector close to a research reactor (say, 5-15 metres) and measure the antineutrino spectrum as a function of position. If the deficit oscillates with $L/E$, sterile neutrinos. If the deficit is uniform (only an overall reduction), flux mismodeling.

This is exactly what PROSPECT, STEREO, and the others were designed to do.

The experiments

PROSPECT at the High Flux Isotope Reactor (HFIR), Oak Ridge National Laboratory. HFIR uses 93%-enriched ²³⁵U fuel and produces approximately 85 MW thermal power. PROSPECT’s detector — 4 tonnes of liquid scintillator — was placed in a corridor 7-9 metres from the reactor core, with the detector segmented into 154 cells along a 2-metre baseline span. Each cell separately measured the antineutrino interaction rate, allowing direct comparison of nearby vs. far cells within a single detector.

PROSPECT operated from 2018 to 2020, accumulating roughly 50,000 inverse beta decay events. The sterile-neutrino analysis directly compared the ratio of events in different cells against expectations under both no-oscillation and sterile-oscillation hypotheses.

STEREO at the ILL reactor in Grenoble. ILL uses 90%-enriched ²³⁵U and produces 58 MW. STEREO’s detector consisted of six segmented cells of liquid scintillator at baseline 9-12 metres, with total active volume of 2 tonnes.

STEREO operated from 2017 to 2020, accumulating roughly 100,000 events. Like PROSPECT, the cell-by-cell rate comparison provided direct sensitivity to oscillation patterns.

DANSS at the Kalinin power plant. Kalinin is a 3-GW commercial reactor (much higher power than research reactors) burning standard low-enriched ²³⁵U+²³⁹Pu mix. DANSS’s detector is 1 tonne of plastic scintillator on a movable platform that can scan baseline from 9.7 to 12.2 metres. The same detector at multiple positions provides the cleanest possible cancellation of detector systematics.

DANSS has been running since 2016 and continues today. It accumulates ~5,000 events per day.

NEOS at the Hanbit reactor in South Korea. Single-detector experiment at 24 metres baseline, taking single-position measurements but comparing to other reactor data.

PROSPECT-II at HFIR — an upgraded version of PROSPECT with improved analysis and additional running. Currently in operation.

What they found

By 2023, combined analyses of PROSPECT + STEREO + DANSS + NEOS data had produced clear results.

The sterile neutrino interpretation of the reactor anomaly is excluded:

• LSND-favored region ($\Delta m^2\sim 1$ eV², ${\sin }^2(2\theta )\sim 0.1$) ruled out at >95% C.L.
• Most of the $\Delta m^2=0.1$–10 eV² parameter space is now constrained.
• Some parameter space remains at very small mixing or very large $\Delta m^2$, but it is theoretically uninteresting.

The flux deficit remains:

• Both PROSPECT and STEREO observe an overall ~5% deficit relative to the standard Hubert-Mueller flux prediction.
• The deficit is not modulated by distance — it’s a flat reduction.
• This is consistent with the prediction being too high; not with oscillation.

The 5-MeV bump also remains:

• Both experiments see the same low-energy excess that Daya Bay first identified.
• The bump is independent of baseline, again pointing to a flux-prediction issue rather than oscillation.

Why the prediction is wrong

The Hubert-Mueller and Schreckenbach-Klapdor flux predictions, used since the 1980s, are based on conversion of measured beta spectra of fission products into corresponding antineutrino spectra. They assume specific shapes for “forbidden” decays (where the nuclear matrix element doesn’t simplify) and integrate over all fission products.

Recent improvements in nuclear-data evaluations (particularly for ²³⁵U-specific decays in the 5-7 MeV antineutrino energy range) have revealed that some forbidden-decay shape factors were inaccurate. New predictions, incorporating improved matrix elements:

• Mueller et al. 2024 update — slightly lower flux, partially closing the deficit
• Estienne et al. 2019 update — different shape, also partially closing the bump

These updates haven’t fully eliminated the discrepancy, but they make it clearer that the deficit is most plausibly a prediction error rather than new physics.

A definitive resolution will require a direct measurement of the antineutrino spectrum from individual fission products — a proposed experiment that would need a different infrastructure (TANL-type beam at radioactive ion facility) but is in advanced planning at multiple labs.

Context for the LSND saga

The PROSPECT/STEREO results are part of a broader picture in 2026:

• LSND’s anomaly remains experimentally confirmed but unrepeated.
• MiniBooNE’s anomaly is most likely a misidentified-photon background (per MicroBooNE’s 2024 verdict).
• The reactor anomaly is most likely flux mismodeling, not oscillation.
• The gallium anomaly remains real but isolated (no broad confirmation in other experiments).

Combined, the picture is that there is no clean, broad evidence for a sterile neutrino at the LSND-favored parameters. The remaining anomalies (LSND itself, gallium) are isolated and most likely either statistical fluctuations or misidentified backgrounds in their respective experiments. The “sterile neutrino in the eV-mass region” hypothesis is, by 2026, considered disfavored by most of the community.

This does not exclude sterile neutrinos at other mass scales. keV-scale sterile neutrinos remain candidates for warm dark matter. TeV-scale right-handed Majorana neutrinos remain a target of LHC searches and seesaw scenarios. The LSND-style eV sterile is what has been excluded; other sterile-neutrino scenarios are very much alive.

What’s next

The PROSPECT-II programme continues at HFIR, accumulating exposure to refine flux measurements. DANSS continues at Kalinin. New experiments are being proposed:

Reactor flux measurements at percent precision — directly measuring antineutrino spectra from individual fission isotopes via dedicated radioactive-source experiments. These would settle the flux-prediction question directly.

Very-short-baseline searches at <1 metre to extend coverage to higher $\Delta m^2$. Some scenarios (LSND, gallium) might still hide at larger mass-squared values not yet probed.

Coherent scattering at short baselines — experiments like CONUS, RED-100, NUCLEUS using CEvNS at reactors test the antineutrino flux through a completely different channel. Independent cross-check of any flux-prediction explanation.

By the end of the decade, the reactor antineutrino anomaly should be definitively resolved — most likely through improved flux predictions matching observed data, but with the systematic uncertainty quantified and well-understood.

The methodological lesson

The PROSPECT/STEREO/DANSS programme is, in some respects, the model for how a major particle-physics anomaly should be tested.

A clear experimental hypothesis (sterile neutrino at LSND parameters). Multiple independent experiments at different reactors with different detector technologies. Different baselines for direct oscillation-pattern fits. Sufficient exposure (tens of thousands of events per experiment) to provide high statistical sensitivity. And, importantly, willingness to publish results that exclude the originally motivating hypothesis.

Compare this with the long history of the LSND anomaly itself — single experiment, never directly repeated, ongoing 30 years after the original publication. The reactor programme cleanly resolved the corresponding question for reactor experiments in less than a decade.

The lesson: anomalies should be tested promptly with multi-experiment campaigns at different sites and energies. When an anomaly is “tested away” — as the reactor anomaly’s sterile-neutrino interpretation has been — the field updates its understanding and moves on. The remaining puzzles get clearer attention; resources stop being expended on theoretical models that don’t fit.

For the broader sterile-neutrino question, PROSPECT and STEREO have done their job. The eV-scale sterile is unlikely to be the answer to LSND, MiniBooNE, gallium, or the reactor anomaly. The remaining anomalies have specific local explanations (or remain genuinely puzzling). The field has progressed.

That progression — clean experiments, clean conclusions, willingness to update — is what makes physics work. Even when the answer is “no new physics”, the path to that answer is itself worth the investment.