Performance of the Eos detector with water

This paper presents the first results from the Eos detector, demonstrating its performance and calibration capabilities using water as a Cherenkov-only medium to validate reconstruction algorithms and detector models for future hybrid neutrino experiments.

The Gist

Imagine a giant, transparent jellyfish floating in a dark room. Inside this jellyfish is a smaller, delicate glass bowl. The goal of this experiment, called Eos, is to teach this jellyfish how to "see" tiny particles of light that fly through it, so that in the future, it can help scientists understand the secrets of the universe, like how stars burn or why there is more matter than antimatter.

This specific paper is like the "training manual" written after the jellyfish was filled with plain water. The scientists wanted to prove that their high-tech jellyfish works perfectly before they fill it with a special, glowing liquid later on.

Here is a simple breakdown of what they did and what they found:

1. The Setup: A High-Tech Fishbowl

The Eos detector is a 4-tonne glass tank (the inner bowl) sitting inside a larger 30-tonne steel tank (the outer bowl).

• The Water Phase: For this experiment, they filled the inner bowl with plain water. Water is special because when a particle zips through it, it creates a faint, blue flash of light called Cherenkov light (think of the sonic boom a jet makes, but for light).
• The Eyes: Surrounding the glass bowl are 239 giant "eyes" (photomultiplier tubes or PMTs). These eyes are incredibly sensitive; they can detect a single photon (a particle of light). Some of these eyes are big, some are small, and some have special sunglasses (called dichroicons) that help sort different colors of light.

2. The Training: Teaching the Eyes to See

Before they could trust the detector, they had to teach it. They used a "calibration crew" to lower different light sources down the center of the tank, like a diver lowering a flashlight into a pool.

• The Laserball: They lowered a glowing ball that flashed laser light in all directions. This was like a "test pattern" on a TV screen. It helped them measure exactly how fast the light traveled and how long it took for each "eye" to blink. They found that some eyes were slightly slower than others due to long cables, so they adjusted the timing for each one.
• The Thorium Source: They lowered a radioactive source that shoots out gamma rays. When these rays hit the water, they create a predictable amount of light. This helped them figure out how "sensitive" each eye was. Some eyes were a bit dimmer than expected, so they adjusted the software to give them a little boost.
• The Directional Source: They used a special source that shoots particles in a straight line, like a laser pointer. This helped them test if the detector could tell which way a particle was moving.
• The AmBe Source: This source shoots out neutrons and gamma rays. It's like a two-step dance: first a flash, then a second flash a tiny fraction of a second later. The detector successfully caught this "dance," proving it could spot neutrons even in a noisy environment.

3. The Computer Brain: Simulation vs. Reality

The scientists built a perfect digital twin of the detector on their computers. They fed this computer model the same data they got from the real detector.

• The Goal: They wanted to see if the computer's predictions matched the real-world results.
• The Result: It was a match! The computer model predicted exactly how the light would travel, where the particles would hit, and how bright the flashes would be. The differences between the real detector and the computer model were tiny (usually less than a few centimeters in position).

4. The "Reconstruction" Magic

Once the detector saw the light, the scientists had to figure out where the particle came from and which way it was going. They used three different "mathematical detectives" (algorithms) to solve the puzzle:

• The Quad Fitter: A fast, simple method that uses four eyes to guess the location.
• The Likelihood Fitters (SeedNDestroy & Mimir): Smarter detectives that use probability to find the best answer.
• The Deep Learning Detective (HITMAN): A modern AI tool trained on millions of simulated events to guess the answer instantly.

All three detectives did a great job. They could pinpoint the location of the light source and the direction it was traveling with high accuracy.

5. The Big Takeaway

The paper concludes that the Eos detector works exactly as the scientists hoped.

• They proved that their "hybrid" technology (which can see both the faint Cherenkov light and, in the future, bright scintillation light) is ready for the next step.
• They showed that even with a small detector near the surface (where there is a lot of background noise from cosmic rays), they could still find clean signals.
• Most importantly, they built a reliable computer model. Because the model matches the real water-filled detector so well, they can now trust it to predict how the detector will behave when they fill it with the special, glowing liquid scintillator in the future.

In short: The scientists built a high-tech underwater camera, filled it with water, tested it with various light sources, and proved that their computer simulations are perfect. Now, they are ready to fill it with the "real stuff" to start doing serious physics.

Technical Summary

Technical Summary: Performance of the Eos Detector with Water

Problem and Motivation
The detection of neutrinos has historically relied on either water-based Cherenkov detectors (e.g., Super-Kamiokande) or liquid scintillator detectors (e.g., Borexino). While water offers excellent directionality via Cherenkov light, it produces low light yields. Conversely, liquid scintillators provide high light yields but make the identification of Cherenkov photons challenging due to the overwhelming scintillation background. Future hybrid detectors, such as Theia, aim to simultaneously detect both Cherenkov and scintillation photons to achieve low energy thresholds, excellent vertex resolution, and event-by-event directionality. However, validating the reconstruction algorithms and Monte Carlo (MC) models required for these hybrid systems is difficult without a controlled environment. The Eos detector was constructed to demonstrate the performance capabilities of this hybrid technology and to test MC simulation predictions before the deployment of scintillating materials.

Methodology
The Eos detector is a four-tonne optical detector located at the University of California, Berkeley. It consists of a cylindrical acrylic inner vessel (IV) with a 91 cm radius, surrounded by a 30-tonne stainless steel outer vessel (OV). The detector is instrumented with 239 photomultiplier tubes (PMTs) of three types: 204 eight-inch Hamamatsu R14688-100 PMTs, 24 twelve-inch R11780 PMTs, and 11 ten-inch R7081 PMTs. A unique feature is the deployment of 12 dichroicons in front of 12 of the eight-inch PMTs to enable spectral sorting of Cherenkov and scintillation light.

For the "water phase" described in this manuscript, the IV was filled with water, acting as a medium that produces only Cherenkov light. This provided a well-understood baseline for calibration and algorithm development. The study utilized a suite of calibration sources deployed along the central vertical axis:

• Laserball: An isotropic light source with tunable wavelengths (374–515 nm) used for timing, charge, and pulse-shape calibration.
• Thorium Source: A $^{232}$Th source emitting 2.6 MeV $\gamma$-rays to study position reconstruction and PMT efficiency.
• Directional Sources: Collimated $\beta$-emitters ($^{90}$Sr and $^{106}$Ru) with self-triggering capabilities to test direction reconstruction.
• AmBe Source: A $^{241}$Am-Be source producing prompt 4.4 MeV $\gamma$-rays and delayed 2.2 MeV $\gamma$-rays from neutron capture, used to test coincidence tagging and neutron detection.
• Cosmic Muons: Used to tag Michel electrons for high-energy calibration.

Data processing involved extracting PMT hit times and charges using lognormal fits to waveforms. Reconstruction algorithms included a "quad fitter," likelihood-based fitters (SeedNDestroy and Mimir), and a deep-learning-based inference tool (HITMAN). The simulation framework, RAT-PAC-2 (based on Geant4), was calibrated using the laserball and thorium data to match detector response parameters (timing delays, pulse shapes, charge scales, and efficiencies).

Key Contributions

1. First Water-Phase Results: This paper presents the first results from Eos, establishing a baseline performance for a multi-tonne hybrid detector using only water.
2. Comprehensive Calibration: The authors detail a full calibration chain, including PMT timing delays (achieved via 515 nm laserball data), pulse shape modeling, charge scaling, and dark-rate characterization.
3. Model Validation: The study validates the RAT-PAC-2 simulation against data across various source types, positions, and energies. It specifically demonstrates the successful modeling of dichroicon responses and the ability to simulate the complex geometry of the detector.
4. Reconstruction Algorithm Benchmarking: The paper compares four reconstruction methods (Quad, SeedNDestroy, Mimir, HITMAN) for vertex and direction reconstruction, demonstrating that likelihood-based and machine-learning approaches can achieve comparable or superior performance to traditional methods in this asymmetric detector geometry.
5. Neutron Detection in Water: The paper demonstrates the capability to detect neutron capture events (via the AmBe source) in a surface-level, water-filled detector, validating the coincidence tagging logic required for future reactor antineutrino monitoring.

Results

• Calibration: After applying timing delay corrections, the time-residual width for a central laserball run was measured at 450 ps. PMT charge and efficiency calibrations brought the simulation into agreement with data, with efficiency scale factors determined as 0.78 (barrel), 0.90 (top), and 0.88 (bottom).
• Vertex Reconstruction: Using the Thorium source, the reconstruction bias and resolution were evaluated. For central 2.0 MeV electrons (simulated), the vertex resolution ($\sigma$) was found to be approximately 8–9 cm for the likelihood and ML-based fitters, compared to ~11 cm for the quad fitter. The bias between data and simulation was generally within 1–2 cm, meeting the paper's goal of agreement within 3 cm. The $z$-resolution showed a dependence on source position due to detector asymmetry (higher PMT coverage at the bottom), which the simulation accurately predicted.
• Direction Reconstruction: Using directional sources, the angular resolution ($\alpha_{1\sigma}$) was measured. The agreement between data and simulation for $\alpha_{1\sigma}$ was within 0.1 across various source positions and types, meeting the stated performance goal.
• Dichroicon Performance: The simulation of dichroicon response showed good agreement with data for laserball deployments, particularly above the cut-on wavelength (450 nm), though some discrepancies were noted at lower source positions due to incidence angle dependencies.
• AmBe Coincidence: The detector successfully identified the prompt 4.4 MeV and delayed 2.2 MeV $\gamma$-rays with a measured neutron capture time of $\tau = 206.2 \pm 6.4$ $\mu$s, consistent with expectations.

Significance
The paper claims that these results serve as a critical validation of the detector modeling and reconstruction software required for future hybrid detectors. By demonstrating that a detailed MC model can accurately predict detector behavior in a water-only environment, the authors establish confidence in extrapolating these tools to future phases involving liquid scintillators (both water-based and pure). The successful reconstruction of vertex and direction, along with the validation of neutron detection capabilities, supports the feasibility of using this technology for physics goals such as neutrinoless double beta decay, solar neutrino measurements, and nuclear nonproliferation monitoring. The work provides a benchmark for the Theia collaboration and future large-scale hybrid detector designs.