ANNIE: Neutrino Neutron Interaction Experiment

• ANNIE is a compact neutrino experiment at Fermilab using a 26-ton water Cherenkov target to measure final-state neutron multiplicity from neutrino-nucleus interactions.
• The experiment integrates gadolinium loading, LAPPDs, and a Muon Range Detector to enhance neutron tagging, improve muon reconstruction, and reduce background in proton-decay searches.
• ANNIE’s dual mission advances both precise neutrino-interaction measurements and detector R&D for future water-based and liquid scintillator neutrino detectors.

Searching arXiv for ANNIE-related papers to ground the article and verify the provided paper set. The Accelerator Neutrino Neutron Interaction Experiment (ANNIE) is a compact neutrino experiment at Fermilab built around a 26-ton water Cherenkov target on the Booster Neutrino Beam (BNB). Its central scientific objective is to measure the multiplicity of final-state neutrons from neutrino–nucleus interactions in water, an observable that is directly relevant to neutrino-interaction modeling, neutrino-energy reconstruction, and background rejection in water-based detectors. In parallel, ANNIE functions as a detector-development platform for gadolinium-loaded water, Large Area Picosecond PhotoDetectors (LAPPDs), and later water-based liquid scintillator (WbLS), with the explicit aim of testing technologies for future neutrino detectors (Drakopoulou et al., 2019, Tiras, 2019).

1. Scientific motivation and conceptual origin

ANNIE was proposed to address a specific deficiency in water-Cherenkov neutrino physics: the lack of direct measurements of neutron yield as a function of interaction kinematics, especially momentum transfer. The original rationale was closely tied to atmospheric-neutrino backgrounds in proton-decay searches. If the true neutron multiplicity distribution is $P(N)$ and the neutron-tagging efficiency is $\epsilon$, the fraction of atmospheric-neutrino background events that survive a zero-neutron selection is written as

$$f = P(0) + P(1)(1-\epsilon) + P(2)(1-\epsilon)^2 + P(3)(1-\epsilon)^3 + \cdots$$

so proton-decay background rejection depends not only on the efficiency of neutron tagging but on the full multiplicity distribution itself (Anghel et al., 2015).

The experiment was also framed from the beginning as a neutrino-interaction measurement. Final-state neutrons encode information about many-body nuclear dynamics, final-state interactions, meson exchange currents, stuck-pion backgrounds, and the distinction between truly quasi-elastic and CCQE-like topologies. This is why ANNIE’s program is repeatedly described as a measurement of neutron yield from $\nu_\mu$ interactions as a function of $Q^2$: neutron counting is not merely a background-rejection tool, but a handle on the hadronic final state and therefore on the systematics of neutrino-energy reconstruction in oscillation analyses (Drakopoulou, 2018, Back et al., 2017).

A persistent misconception is that ANNIE is principally an instrumentation demonstrator. The published descriptions do not support that reduction. They consistently present ANNIE as a dual-mission experiment: a neutron-physics program in water and a test bed for next-generation optical detector technologies. This suggests that the experiment’s design choices—small scale, hybrid instrumentation, staged operation, and emphasis on delayed neutron capture—were driven by the need to make a targeted physics measurement under conditions that also stress new detector concepts (Tiras, 2019).

2. Beam environment and detector architecture

ANNIE is installed in the Booster Neutrino Beam at Fermilab, roughly 100 m downstream of the BNB target. In neutrino mode, the beam is described as about 93% $\nu_\mu$, with a spectrum peaking near 700 MeV; the Phase-I background paper quotes an average beam rate of 5 Hz, a spill composed of 84 bunches over 1.6 $\mu$s, and a nominal intensity of $5\times 10^{12}$ protons on target per spill (Drakopoulou et al., 2019, Back et al., 2019).

The detector is a hybrid system. Its core is an upright cylindrical tank containing water, later operated with 0.1% gadolinium by weight in the target medium in later Phase-II-era descriptions. Upstream sits the Front Muon Veto (FMV), and downstream the Muon Range Detector (MRD). The FMV is used to reject entering charged particles from upstream interactions and cosmic-ray muons; one later detector description gives an entering-muon tagging efficiency of $(95.6 \pm 1.6)\%$. The MRD is an iron–plastic scintillator sandwich calorimeter that ranges out muons and supplies direction and stopping-energy information; one R&D description specifies 11 alternating vertical and horizontal layers with 5 cm iron absorbers and 306 scintillator paddles (Tiras, 2019, Collaboration et al., 2023, Abubakar et al., 30 Oct 2025).

ANNIE’s optical instrumentation evolved across phases and studies. Phase I used a water tank with 58 upward-facing $8''$ Hamamatsu R5912 PMTs at the base and a movable Neutron Capture Volume (NCV) for background measurements (Drakopoulou et al., 2019). Phase-II reconstruction studies compared a PMT-only configuration with 128 $\epsilon$0-inch PMTs against a configuration adding 5 LAPPDs on the downstream wall (Drakopoulou, 2018). A later baseline reconstruction paper described the installed conventional system as 132 PMTs of three types—72 Hamamatsu R7081 $\epsilon$1-inch, 20 ETEL D784UKFLB $\epsilon$2-inch, and 40 Hamamatsu R5912-100 $\epsilon$3-inch—with about 14% photocathode coverage (Abubakar et al., 30 Oct 2025). These differing numbers reflect different design stages, simulation benchmarks, and detector configurations rather than a single immutable layout.

The basic architectural logic remained stable across those iterations. The water target provides the neutrino interaction medium and delayed neutron-capture signal, the PMTs provide broad photon collection, the MRD constrains muon range and direction, the FMV rejects entering backgrounds, and the LAPPDs provide high-precision spatio-temporal information for prompt-event reconstruction. In this sense ANNIE is best understood as a deliberately hybrid water-Cherenkov detector rather than a conventional PMT-only tank (Drakopoulou, 2018, Tiras, 2019).

3. Gadolinium neutron tagging and the Phase-I background program

A defining feature of ANNIE is the use of gadolinium to make neutron capture efficient and observable in water. In pure water, thermal neutrons are predominantly captured on hydrogen, producing a 2.2 MeV gamma on a timescale of roughly 200 $\epsilon$4s, which is difficult to detect in a water Cherenkov detector. With gadolinium loading, the thermal-neutron capture cross section rises to about 49,000–49,700 barns, capture produces an $\epsilon$5 MeV gamma cascade, and the time constant falls to tens of microseconds, often quoted as about 20 $\epsilon$6s in the proposal literature or about 30 $\epsilon$7s at 0.1% Gd by mass in later Phase-II discussions (Anghel et al., 2015, Back et al., 2019).

Phase I was designed to determine whether beam-correlated neutron backgrounds would compromise this delayed-capture strategy. The detector was operated as a pure-water system equipped with the movable NCV, a $\epsilon$8 acrylic vessel containing 25 gallons of 0.25% by weight Gd-loaded liquid scintillator and viewed by two $\epsilon$9 PMTs. The NCV was optically isolated and calibrated with a $$f = P(0) + P(1)(1-\epsilon) + P(2)(1-\epsilon)^2 + P(3)(1-\epsilon)^3 + \cdots$$0Cf source. Its combined neutron-detection efficiency was later quoted as

$$f = P(0) + P(1)(1-\epsilon) + P(2)(1-\epsilon)^2 + P(3)(1-\epsilon)^3 + \cdots$$1

after combining source-based and threshold-based estimates (Drakopoulou et al., 2019, Back et al., 2019).

The Phase-I analyses identified two principal beam-correlated neutron backgrounds. “Sky-shine” neutrons are produced in the beam dump, leak into the atmosphere, scatter multiple times, and enter the detector from above. “Dirt” neutrons are produced by neutrino interactions in upstream rock and soil. ANNIE mapped the beam-correlated neutron-candidate rate by moving the NCV to multiple positions with different water overburdens and lateral positions, thereby measuring the background field throughout the tank (Drakopoulou et al., 2019).

The principal result was that the background was both small and strongly depth-dependent. The strongest signal appeared at V4, near the top center of the tank, with a beam-correlated background-neutron rate of

$$f = P(0) + P(1)(1-\epsilon) + P(2)(1-\epsilon)^2 + P(3)(1-\epsilon)^3 + \cdots$$2

while within the future active volume the highest measured rate was reported at V2 as $$f = P(0) + P(1)(1-\epsilon) + P(2)(1-\epsilon)^2 + P(3)(1-\epsilon)^3 + \cdots$$3 neutrons per $$f = P(0) + P(1)(1-\epsilon) + P(2)(1-\epsilon)^2 + P(3)(1-\epsilon)^3 + \cdots$$4 per spill, and the rate at the center of the tank was stated to be consistent with zero within uncertainties. The top-heavy pattern, including the observation that V4 exceeded even the most upstream position, favored sky-shine rather than dirt neutrons as the dominant source and was interpreted as evidence of a soft neutron spectrum that can be attenuated by modest water shielding (Back et al., 2019, Drakopoulou et al., 2019).

This result settled a key feasibility question. Because ANNIE expects fewer than one neutrino interaction per spill at BNB fluxes, the per-spill neutron background rate can be interpreted as the probability of observing a background neutron after an otherwise independent signal interaction. Phase I therefore established that beam-related backgrounds were “sufficiently low” for the Phase-II neutron-multiplicity measurement and that the active volume could be optically isolated and shielded by a relatively small amount of water without spoiling the program (Drakopoulou et al., 2019).

4. Event reconstruction and kinematic inference in Phase II

The Phase-II reconstruction program was built around two stated goals: the first measurement of the neutron yield from $$f = P(0) + P(1)(1-\epsilon) + P(2)(1-\epsilon)^2 + P(3)(1-\epsilon)^3 + \cdots$$5 interactions as a function of $$f = P(0) + P(1)(1-\epsilon) + P(2)(1-\epsilon)^2 + P(3)(1-\epsilon)^3 + \cdots$$6 and the first deployment of LAPPDs in a physics experiment. The reconstruction framework begins from the standard water-Cherenkov picture of a charged lepton emitting Cherenkov light. One Phase-II reconstruction study parameterized the charged-particle track with six parameters—three for vertex position, one for event time, and two for direction—and used a maximum-likelihood fit based on photon hit times and Cherenkov-cone pattern information. It defined

$$f = P(0) + P(1)(1-\epsilon) + P(2)(1-\epsilon)^2 + P(3)(1-\epsilon)^3 + \cdots$$7

and

$$f = P(0) + P(1)(1-\epsilon) + P(2)(1-\epsilon)^2 + P(3)(1-\epsilon)^3 + \cdots$$8

with “resolution” taken from the 68th percentile of the corresponding cumulative distributions (Drakopoulou, 2018).

That simulation study reported a large LAPPD-driven gain. With 128 $$f = P(0) + P(1)(1-\epsilon) + P(2)(1-\epsilon)^2 + P(3)(1-\epsilon)^3 + \cdots$$9-inch PMTs only, the vertex resolution was about 38 cm; with the same PMTs plus 5 LAPPDs on the downstream wall, the vertex resolution improved to about 12 cm. The angular resolution improved to about $\nu_\mu$0, roughly a factor of two better than the PMT-only case. For quasi-elastic events with stopped muons, a TensorFlow 1.3.0 deep neural network reconstructed the muon track length in water, and a Boosted Decision Tree in Scikit-Learn 0.18.2 combined that estimate with MRD information and other event observables. In the 5 LAPPD + 128 PMT configuration, the paper quoted about 10% muon-energy resolution and about 14% neutrino-energy resolution at the 68th percentile (Drakopoulou, 2018).

A later reconstruction paper established a different reference point: the baseline performance before implementation of novel technologies. There the reconstruction deliberately used only the conventional PMT array and the muon spectrometer, reflecting the fact that in a detector of ANNIE’s size nanosecond-scale timing is not as useful as in a large detector. The paper therefore emphasized pattern recognition rather than a conventional global timing fit. For BNB muon neutrino Charged Current Zero Pion (CC0pi) events, the combined PMT+MRD fit yielded a muon vertex uncertainty of 60 cm, a directional uncertainty of 13.2 degrees, and energy reconstruction uncertainty of about 10% (Abubakar et al., 30 Oct 2025).

The coexistence of these two reconstruction narratives is instructive rather than contradictory. One describes the expected gain when LAPPDs are added to a simulated Phase-II optical configuration; the other defines the experimentally relevant baseline against which LAPPDs and WbLS can be measured. Taken together, they show that ANNIE’s reconstruction strategy is not generic water-Cherenkov practice transplanted to a small tank, but a detector-specific program that couples Cherenkov pattern recognition, MRD track information, and eventually picosecond photodetection to obtain $\nu_\mu$1-resolved neutron measurements (Drakopoulou, 2018, Abubakar et al., 30 Oct 2025).

5. LAPPDs in ANNIE: from detector R&D to beam-neutrino observation

ANNIE’s detector-development program gave a central role to Large Area Picosecond PhotoDetectors. These are $\nu_\mu$2 microchannel-plate imaging photodetectors with single-photon timing resolutions of roughly 50 ps and millimeter-scale spatial sensitivity. ANNIE was described as the first particle-physics application of this technology and the first use of LAPPDs in Gd-loaded water and in a neutrino beam (Tiras, 2019).

The collaboration developed a dedicated characterization program at Iowa State University and Fermilab. The Fermilab setup used a dark box, a 50-ps-pulsed 405-nm diode laser, a 420-nm LED, a 2D motion stage, and PSEC4 readout sampling at 10 Gsample/s. It mapped quantum efficiency on an $\nu_\mu$3 grid with 2.5 mm steps and measured gain, timing, and transit-time spread. Reported performance included pulse rise times of about 850 ps, a full width at half maximum of about 1.1 ns, time resolution below 60 ps, less than 4% after-pulsing, and gains exceeding $\nu_\mu$4 (Tiras, 2019).

The subsequent beam-operation milestone was reported in “First Beam Neutrinos Observed with an LAPPD in the ANNIE Experiment” (Adams et al., 14 Aug 2025). That work deployed a fully integrated underwater module, the Packaged ANNIE LAPPD (PAL), containing the detector, readout electronics, slow controls, environmental monitoring, and waterproof housing. The first installed unit, LAPPD-40, had 28 silk-screened silver anode microstrips, nominal gain $\nu_\mu$5 at 950 V per MCP, and average quantum efficiency $\nu_\mu$6. Signals were digitized by two ACDC boards per LAPPD, each using the PSEC4 ASIC with 10 GS/s sampling, 1 GHz analog bandwidth, and 256-sample ring buffers. The first ANNIE LAPPD was deployed on March 29, 2022 and integrated into the full DAQ by April 2022 (Adams et al., 14 Aug 2025).

The analysis used data from June 2022 to July 2023, corresponding to 107 stable beam days with all subdetectors functioning nominally and the LAPPD fully operational in the DAQ. Event building required matching timestamps from the central trigger, PMTs, MRD, and the LAPPD local clock; the synchronization ambiguity was resolved using the unique time structure of Booster Neutrino Beam spills. After a sequence of cuts—paired LAPPD-to-PMT match, PMT clustering, MRD reconstruction, in-time MRD coincidence, muon-topology selection, and FMV veto—the final selected sample contained 1030 estimated true beam events and was about 98% pure. In an illustrative event, the LAPPD waveform pattern exhibited a $\nu_\mu$7 ns time gradient across the detector surface, consistent with ray-tracing predictions from the MRD-reconstructed muon track (Adams et al., 14 Aug 2025).

This established more than simple sensor operability. It showed that an LAPPD in ANNIE could be packaged, powered, triggered, synchronized, deployed underwater, and used to record neutrino-induced Cherenkov light in a running beamline detector. A plausible implication is that ANNIE converted LAPPDs from a promising photodetector technology into an experimentally constrained reconstruction element whose performance can be assessed against neutrino data rather than laboratory surrogates alone (Adams et al., 14 Aug 2025).

6. WbLS deployment and broader significance for water-based neutrino detection

ANNIE’s R&D role later expanded to water-based liquid scintillator (WbLS), motivated by the possibility of combining Cherenkov directionality with scintillation light yield in a single target medium. In March 2023, the collaboration deployed SANDI—Scintillator for ANNIE Neutrino Detection Improvement—a 366 L acrylic vessel of WbLS inside the main detector. The vessel had 90 cm interior height, 72 cm diameter, and 2.54 cm acrylic walls, and it was operated from March to May 2023. The WbLS was approximately 1% organic material by mass and 99% water, produced at Brookhaven National Laboratory and monitored for optical stability before and after deployment (Collaboration et al., 2023).

The reported result was the detection in ANNIE of both Cherenkov light and scintillation light from the WbLS. Using throughgoing muons, the analysis extracted

$\nu_\mu$8

for pure water and

$\nu_\mu$9

for the WbLS-crossing sample, corresponding to

$Q^2$0

A Michel-electron analysis yielded

$Q^2$1

and therefore

$Q^2$2

The paper was explicit that these are not intrinsic WbLS light-yield measurements; they are detector-level increases in observed photoelectrons within the ANNIE geometry, including vessel absorption, shadowing, and geometry effects (Collaboration et al., 2023).

The broader significance of ANNIE follows directly from this combined physics-and-R&D structure. Its neutron-yield program is relevant to oscillation experiments, proton-decay searches, diffuse supernova neutrino background searches, and core-collapse supernova neutrino studies, all of which benefit from neutron tagging in water (Anghel et al., 2015, Back et al., 2017). Its detector-development program is relevant to future water, Gd-water, and water-based liquid scintillator detectors, and one R&D paper explicitly positions it as complementary to proton-multiplicity studies in liquid-argon detectors on the same beamline (Tiras, 2019).

Another common simplification is that ANNIE’s neutron program concerns only delayed captures. The published record points to a more integrated picture. Final-state neutron counting depends on prompt-event fiducialization, muon kinematics from the MRD, low background from the hall environment, delayed capture visibility through gadolinium loading, and increasingly on precision optical reconstruction from LAPPDs and hybrid media. ANNIE’s importance therefore lies not in any one subsystem, but in the demonstration that these elements can be combined in a small beam detector to turn neutron production from a poorly constrained byproduct into a measurable neutrino observable (Drakopoulou et al., 2019, Adams et al., 14 Aug 2025).