PROSPECT: Precision Reactor Oscillation Experiment

• PROSPECT is a very-short-baseline reactor antineutrino experiment at HFIR using a 3.68-ton, 154-segment, 6Li-doped scintillator to search for eV-scale sterile neutrino oscillations.
• It employs a reactor-model independent method by comparing prompt-energy spectra across 7–13 m baselines, achieving exclusions of key sterile neutrino parameters with over 95% confidence.
• It delivers a precise pure-U-235 antineutrino spectrum measurement that benchmarks reactor models and aids in resolving spectral-shape anomalies in the 4–7 MeV range.

PROSPECT, the Precision Reactor Oscillation and Spectrum Experiment, is a very-short-baseline reactor antineutrino experiment at the High Flux Isotope Reactor (HFIR) at Oak Ridge National Laboratory. It was designed to perform two coupled measurements with the same segmented detector: a reactor-model-independent search for $\bar{\nu}_e$ disappearance consistent with eV-scale sterile neutrinos, and a precision measurement of the antineutrino spectrum from an almost pure $^{235}\mathrm{U}$ fission source. Because HFIR is a compact, highly enriched uranium (HEU) reactor with more than $99\%$ of fissions from $^{235}\mathrm{U}$, PROSPECT occupies a distinctive position in reactor-neutrino physics: it probes meter-scale oscillation structure while isolating the $^{235}\mathrm{U}$ component of the reactor spectrum (Collaboration et al., 2018).

1. Scientific motivations and programmatic evolution

PROSPECT emerged in response to two anomalies in reactor neutrino physics. The first was the reactor antineutrino anomaly, namely an approximately $5\%$–$6\%$ deficit in observed $\bar{\nu}_e$ rates relative to then-standard flux predictions. The second was the spectral-shape anomaly, especially the excess-like deviation in the $4$–$7\ \mathrm{MeV}$ or $^{235}\mathrm{U}$0–$^{235}\mathrm{U}$1 region reported by $^{235}\mathrm{U}$2 reactor experiments. From the outset, the collaboration framed PROSPECT as a dual-purpose experiment: a short-baseline oscillation search for a sterile state with $^{235}\mathrm{U}$3, and a direct spectrum measurement capable of testing whether reactor-model deficiencies, rather than new oscillation physics, lay behind the anomalies (Ashenfelter et al., 2015).

HFIR was central to that strategy. In the proposal-era papers, the collaboration emphasized that an HEU research reactor permits a nearly pure $^{235}\mathrm{U}$4 measurement, unlike commercial low-enriched uranium reactors where $^{235}\mathrm{U}$5, $^{235}\mathrm{U}$6, $^{235}\mathrm{U}$7, and $^{235}\mathrm{U}$8 contributions evolve with burnup. The original staged design therefore combined near-field oscillation sensitivity with a benchmark $^{235}\mathrm{U}$9 spectrum program, first through a movable $99\%$0-ton detector at $99\%$1–$99\%$2, and later through an additional farther detector at $99\%$3–$99\%$4 or $99\%$5–$99\%$6, depending on the design iteration (Ashenfelter et al., 2015).

A useful distinction is therefore required between the original proposal, the commissioned detector, and the later PROSPECT-II concept.

Configuration	Baseline coverage	Defining features
Original Phase I/II proposal	$99\%$7–$99\%$8 and $99\%$9–$^{235}\mathrm{U}$0	Movable near detector plus planned far detector
Commissioned PROSPECT detector	About $^{235}\mathrm{U}$1–$^{235}\mathrm{U}$2 practical coverage	Single movable 154-segment detector at HFIR
PROSPECT-II proposal	$^{235}\mathrm{U}$3–$^{235}\mathrm{U}$4 through segmentation	Upgraded HFIR detector with PMTs outside scintillator

The commissioned experiment realized part of that original program in a single detector. Later, after the first run and its technical lessons, the collaboration reformulated the continuation as PROSPECT-II, an upgraded detector intended to revisit the remaining short-baseline parameter space and improve pure-$^{235}\mathrm{U}$5 measurements (Andriamirado et al., 2022).

2. Reactor source, detector architecture, and instrumentation

HFIR is an $^{235}\mathrm{U}$6 research reactor with a compact core, a decisive advantage for very-short-baseline oscillation studies because source-size smearing remains small. In the commissioned-detector papers, the core is described as a compact cylindrical source with diameter $^{235}\mathrm{U}$7 and height $^{235}\mathrm{U}$8, and a detailed core model showed that the $^{235}\mathrm{U}$9 fission fraction remained above $^{235}\mathrm{U}$0 throughout operation (Ashenfelter et al., 2018).

The commissioned PROSPECT detector was a movable, optically segmented $^{235}\mathrm{U}$1Li-doped liquid scintillator antineutrino detector with an active scintillator volume of 3760 liters, corresponding to 3.68 metric tons of active target. The active volume measured 1.176 m wide $^{235}\mathrm{U}$2 2.045 m long $^{235}\mathrm{U}$3 1.607 m tall and was divided into 154 segments arranged in a 14 $^{235}\mathrm{U}$4 11 grid. Each segment had a square cross section of 0.145 m $^{235}\mathrm{U}$5 0.145 m and a length of 1.176 m, and each was viewed from both ends by photomultipliers. The nearest installed configuration placed the detector center at $^{235}\mathrm{U}$6 from the core center, with moves planned to 9.06 m and 12.36 m, yielding practical baseline coverage from about 7 m to 13 m (Collaboration et al., 2018).

The scintillator technology was a defining subsystem. The detector used a novel $^{235}\mathrm{U}$7Li-loaded liquid scintillator based on EJ-309, loaded to 0.08\% $^{235}\mathrm{U}$8Li by mass in the final detector. The target composition was measured as C $^{235}\mathrm{U}$9\% and H $5\%$0\%, with density $5\%$1. Optical segmentation was provided by a low-mass reflective lattice using separators with carbon-fiber cores, highly reflective 3M DF2000MA film, and FEP protection. The non-scintillating mass in the target region was 3.4\%, and optical cross-talk between segments was reported as less than 1\% (Collaboration et al., 2018, Collaboration et al., 2019).

The prototype program established that this architecture could meet the experimental requirements in realistic geometry. PROSPECT-50, a 50 L, two-segment full-scale prototype, demonstrated $5\%$2 light collection, $5\%$3 energy resolution at $5\%$4, effective attenuation length $5\%$5, and a neutron capture time of $5\%$6 for a $5\%$7Li concentration of 0.1\% by mass. It also achieved PSD figure of merit $5\%$8 in both prompt-like and delayed-like energy regions, exceeding the stated PROSPECT requirement of $5\%$9 (Ashenfelter et al., 2018).

Calibration was correspondingly elaborate. The detector incorporated an optical calibration system injecting 450 nm laser light into 42 locations, and a radioactive source system based on 35 PTFE tubes in a $6\%$0 grid. The source suite included $6\%$1Cs, $6\%$2Co, $6\%$3Na, $6\%$4Cf, and AmBe, with source-position reproducibility within 2 mm and absolute positioning accuracy of about 2 cm (Collaboration et al., 2018, Collaboration et al., 2019).

3. Detection principle, event reconstruction, and background control

PROSPECT detects reactor antineutrinos through inverse beta decay,

$6\%$5

The interaction threshold is 1.8 MeV. The positron creates the prompt scintillation signal, typically in the 1–8 MeV prompt-energy range in the commissioned-detector description, while the neutron thermalizes and captures after a characteristic delay of about 40 $6\%$6 (Collaboration et al., 2018).

A principal design choice was neutron capture on $6\%$7Li rather than gadolinium. In PROSPECT,

$6\%$8

or, in the later detector-analysis paper’s reconstruction language, a delayed signal near 0.526 MeV. Because the $6\%$9 and triton are heavily ionizing and localized, the delayed signal is compact and usually contained within a single segment, making it especially compatible with fine segmentation and pulse-shape discrimination (Collaboration et al., 2018, Andriamirado et al., 2020).

Event reconstruction relied on double-ended timing and charge information. In operation, PROSPECT achieved 4.5\% RMS at 1 MeV energy resolution and a reconstructed longitudinal position resolution of 0.05 m along a segment. Waveforms were digitized by CAEN V1725 modules at 250 MHz and 14-bit depth. Trigger thresholds corresponded to segment energies of about 100 keV for the trigger and 40 keV for zero-length-encoding readout. Early data showed distinct PSD populations: electron/gamma-like pulses near tail fraction 0.1, neutron captures on $\bar{\nu}_e$0Li near 0.55 MeV$\bar{\nu}_e$1 and tail fraction around 0.25, and fast-neutron recoil protons forming a separate band (Collaboration et al., 2018).

Background control was necessarily central because the detector operated essentially on the surface, under less than 1 meter-water-equivalent overburden, and only a few meters from an operating reactor. The shielding package combined lead, polyethylene, borated polyethylene, water, structural HDPE, and an outer aluminum/fire-safe covering, together with a fixed local lead wall near the reactor pool wall. Outer detector segments functioned as an effective veto or fiducial layer because simulations and commissioning data showed that edge segments experienced IBD-like background rates 10–100 times higher than inner ones (Collaboration et al., 2018).

Operationally, PROSPECT combined shielding with topology, timing, PSD, and reactor-off subtraction. Accidentals were measured directly from delayed-time sidebands, while cosmogenic correlated backgrounds were constrained by reactor-off data. The result was a robust aboveground antineutrino signal: the first physics letter reported $\bar{\nu}_e$2 detected IBDs in 33 live-days of reactor-on data and stated that reactor antineutrinos could be identified at $\bar{\nu}_e$3 statistical significance within two hours of reactor-on data taking (Ashenfelter et al., 2018).

4. Oscillation search and the sterile-neutrino question

The oscillation search was deliberately reactor-model independent. Rather than fitting an absolute flux deficit, PROSPECT compared prompt-energy spectra measured at different baselines within the same detector, exploiting the expected short-baseline survival form

$\bar{\nu}_e$4

This relative strategy addressed a common misconception: PROSPECT was not primarily a total-rate test of reactor models, but a search for baseline- and energy-dependent distortions characteristic of oscillations (Kyzylova, 2019).

The first published oscillation result used 33 reactor-on days and 28 reactor-off days. The analysis used 6 baseline bins and 16 prompt-energy bins, and the event rate followed the expected $\bar{\nu}_e$5 trend with a 40\% decrease from the shortest to the longest baseline. The data did not show the oscillatory structure expected for the canonical reactor-antineutrino-anomaly best-fit point $\bar{\nu}_e$6, $\bar{\nu}_e$7. PROSPECT excluded that point at greater than 95\% confidence level, summarized in the proceedings paper as $\bar{\nu}_e$8, and also disfavored the then-recent Neutrino-4 best-fit point at $\bar{\nu}_e$9 confidence (Kyzylova, 2019).

A larger and more detailed oscillation analysis later used data collected from March 5, 2018 to October 6, 2018, corresponding to 95.6 calendar days reactor-on and 73.1 calendar days reactor-off, or 82.2 live days reactor-on and 65.2 live days reactor-off after veto dead time. In that data set, PROSPECT extracted $4$0 reactor antineutrino events and performed the oscillation fit in 10 baseline bins covering 6.7–9.2 m. The best-fit point,

$4$1

improved the fit by only $4$2 over no oscillation, and Feldman–Cousins toys showed the no-oscillation hypothesis to be fully compatible with the data, with a $4$3-value of 0.57. The same analysis disfavored the reactor-antineutrino-anomaly best-fit point at $4$4 (Andriamirado et al., 2020).

These exclusions were substantial but not exhaustive. Later PROSPECT-II motivation papers explicitly noted that, although PROSPECT together with STEREO, NEOS, and DANSS excluded large portions of the low-$4$5 parameter space previously suggested by the reactor and gallium anomalies, a sizeable region above about $4$6, and especially above $4$7, remained comparatively open. They also stressed that the broader short-baseline landscape had become more complex, not simpler, because BEST strengthened the gallium anomaly while MicroBooNE did not fully eliminate sterile-neutrino interpretations of MiniBooNE/LSND (Andriamirado et al., 2022).

5. The $4$8 antineutrino spectrum measurement

The second pillar of PROSPECT was its $4$9 spectrum program. Because HFIR is an HEU reactor with $7\ \mathrm{MeV}$0 contributing $7\ \mathrm{MeV}$1 of the antineutrino flux, PROSPECT could measure a nearly pure-$7\ \mathrm{MeV}$2 spectrum rather than the isotopically mixed spectrum seen at commercial power reactors (Surukuchi, 2019).

An early dedicated spectrum analysis used 40.3 days of reactor-on data and 37.8 days of reactor-off data. After subtraction of accidental backgrounds and atmospheric-pressure-corrected reactor-off backgrounds, the final IBD sample contained

$7\ \mathrm{MeV}$3

signal events, with $7\ \mathrm{MeV}$4. In a shape-only comparison to the Huber-based $7\ \mathrm{MeV}$5 prediction, the prompt-energy spectrum yielded

$7\ \mathrm{MeV}$6

The analysis found two sliding-window regions of disagreement above $7\ \mathrm{MeV}$7, and an ad hoc comparison to the Daya Bay excess gave a best-fit distortion normalization

$7\ \mathrm{MeV}$8

Within that framework, the hypothesis that $7\ \mathrm{MeV}$9 alone accounts for the Daya Bay local excess was disfavored at $^{235}\mathrm{U}$00 (Surukuchi, 2019).

A later and larger analysis, based on the improved detector model and the longer 2018 data set, reported a forward-folded comparison of the prompt reconstructed-energy spectrum to a Huber-based model with HFIR-specific corrections. In that analysis the $^{235}\mathrm{U}$01 prompt spectrum was in excellent agreement with the model, with

$^{235}\mathrm{U}$02

and a one-sided $^{235}\mathrm{U}$03-value of 0.48. The same paper nevertheless found that a Daya Bay-like Gaussian distortion amplitude

$^{235}\mathrm{U}$04

was mildly preferred over no distortion, with $^{235}\mathrm{U}$05 disfavored at $^{235}\mathrm{U}$06, while the hypothesis that $^{235}\mathrm{U}$07 alone generates the full commercial-reactor spectrum discrepancy, $^{235}\mathrm{U}$08, was disfavored at $^{235}\mathrm{U}$09 (Andriamirado et al., 2020).

Taken together, these analyses established several points. First, a pure-$^{235}\mathrm{U}$10 measurement is indispensable for interpreting the LEU spectral anomaly. Second, PROSPECT did not support a simple statement that the commercial-reactor bump is entirely a $^{235}\mathrm{U}$11 phenomenon. Third, the $^{235}\mathrm{U}$12 spectrum could be studied at precision sufficient to discriminate among forward-folded model expectations and to sharpen later data-driven isotopic comparisons (Surukuchi, 2019, Andriamirado et al., 2020).

6. PROSPECT-II, detector reanalysis, and legacy

PROSPECT’s first detector demonstrated that precision aboveground antineutrino measurements at HFIR were feasible, but its operational lifetime was shortened by PMT-base failures associated with scintillator entering PMT housings. PROSPECT-II was proposed as an evolutionary upgrade rather than a new concept: it retained the segmented $^{235}\mathrm{U}$13Li-doped liquid scintillator architecture, shielding, PSD-based background rejection, and short-baseline HFIR location, while redesigning the optical readout so that PMTs would sit outside the liquid scintillator volume. The proposal called for rebuilding the inner scintillator containment vessel, producing new scintillator, and revamping the calibration deployment scheme, while largely reusing the outer containment vessel, shielding package, and data-acquisition electronics (Andriamirado et al., 2022).

The PROSPECT-II motivation was sharpened by the post-2018 anomaly landscape. The collaboration argued that a pure-$^{235}\mathrm{U}$14, reactor-model-independent disappearance measurement at HFIR remained especially valuable in the $^{235}\mathrm{U}$15–$^{235}\mathrm{U}$16 region, with the strongest need above about $^{235}\mathrm{U}$17 and particularly in the $^{235}\mathrm{U}$18–$^{235}\mathrm{U}$19 interval. The proposal stated that PROSPECT-II could surpass the present global reactor-analysis precision above $^{235}\mathrm{U}$20, improve sensitivity by a factor of 2 to 4 for $^{235}\mathrm{U}$21–$^{235}\mathrm{U}$22 mass splittings, and, in combination with projected KATRIN sensitivity, exclude all of the gallium-suggested parameter space. It also projected a pure-$^{235}\mathrm{U}$23 flux measurement with anticipated precision of $^{235}\mathrm{U}$24, dominated by HFIR reactor-power uncertainty of about $^{235}\mathrm{U}$25, with impactful physics after as little as one calendar year and full sensitivity after 14 reactor cycles (Andriamirado et al., 2022).

The PROSPECT data set also continued to evolve analytically after detector operation ended. A 2025 machine-learning reanalysis addressed the 2018 single-ended-segment problem by using waveform-level sparse convolutional neural networks for position reconstruction and GMMConv graph networks for particle classification. In that work, the best single-ended $^{235}\mathrm{U}$26-reconstruction reached 133 mm MAE, and the optimized single-ended event-reconstruction chain increased effective statistics from 244.9/day to 253.1/day, a 3.3\% improvement over the traditional single-ended treatment (Andriamirado et al., 9 Mar 2025).

PROSPECT’s historical significance therefore lies in more than one result. It established that a segmented $^{235}\mathrm{U}$27Li-loaded scintillator detector could operate near the surface at an HEU research reactor with well-controlled backgrounds; it delivered model-independent short-baseline oscillation constraints that weakened the canonical reactor-antineutrino-anomaly sterile-neutrino interpretation; it produced a benchmark $^{235}\mathrm{U}$28 spectrum against which isotope-specific reactor-model claims can be tested; and it defined the technical and physics basis for PROSPECT-II as a targeted continuation of the short-baseline reactor program (Collaboration et al., 2018, Andriamirado et al., 2022).