Design and performance of a large-area scintillator-based chamber for the MID subsystem of ALICE 3

This paper presents the design, construction, and beam test results of a large-area scintillator-based chamber for the ALICE 3 muon identifier, demonstrating over 99% muon detection efficiency and effective pion rejection via a machine learning algorithm.

The Gist

Imagine you are trying to find a specific type of needle in a giant, chaotic haystack. But this isn't just any haystack; it's a storm of particles flying at nearly the speed of light, and the "needles" are muons (ghostly particles that pass through almost everything), while the "hay" is made of pions (heavier, messier particles that crash into things).

This paper is about building a super-sensitive "net" to catch those muons for a massive physics experiment called ALICE 3, which will study the "primordial soup" of the universe (quark-gluon plasma) in the future.

Here is the story of how they built and tested this net, explained simply:

1. The Goal: Catching Ghosts in a Storm

The scientists need to identify muons with very specific energy levels (between 1.5 and 5 GeV/c). Why? Because other experiments can only see muons that are "running fast" (above 6 GeV/c). ALICE 3 wants to catch the "slow walkers" too, which is unique.

To do this, they need a detector that can tell the difference between a muon (which zips right through) and a pion (which usually gets stopped or creates a mess).

2. Building the Net: The "Scintillator Sandwich"

The team built a prototype chamber that acts like a giant, high-tech sandwich.

• The Bread: Two layers of aluminum plates.
• The Filling: Inside, there are 48 long, thin plastic bars (scintillators). Think of these like glow-in-the-dark candy bars.
  • Layer 1: 24 bars running horizontally.
  • Layer 2: 24 bars running vertically (perpendicular to the first layer).
  • The Gap: There is a 1 cm air gap between the two layers.
• The Magic Trick: When a particle hits these bars, they flash with light. To catch this light, they threaded special fibers (like optical fibers) through the middle of the bars and attached tiny, super-sensitive cameras (called SiPMs) to the ends.
  • Analogy: Imagine a row of fireflies in a dark room. If you touch a firefly, it flashes. The fiber is a straw that carries that flash to the camera so you know exactly which firefly was touched.

They tested two sizes of these "straws" (fibers) and found that the thicker ones (2mm) collected 40% more light, making the camera's job much easier.

3. The Test Drive: The CERN "Wind Tunnel"

They took this giant sandwich to CERN (the European lab for particle physics) to test it in a real particle beam.

• The Setup: They placed a massive block of iron in front of the detector.
  • Analogy: Imagine a hailstorm. You want to see who is wearing a raincoat (muons) and who is just getting wet (pions). You put a thick wall of lead (the iron absorber) in the path. The "wet" pions get stopped by the wall or bounce off, while the "raincoat-wearing" muons punch right through.
• The Beam: They shot a mix of pions and muons at the detector. To make sure they weren't tricked by electrons (which are like tiny, fast mosquitoes), they used a special "Cherenkov counter" (a pressure filter) to block the mosquitoes out.

4. The Brain: Teaching a Computer to Spot the Difference

Just looking at the data wasn't enough. The pion beams were messy, creating "showers" of particles that could fake a muon signal. So, they used Machine Learning (AI).

• The Training: They fed the computer 50% of the data, telling it: "This is a muon (good), and this is a pion (bad)."
• The Variables: The AI looked at clues like:
  • How much light was detected?
  • Where exactly did the hit happen?
  • How long did the signal last?
• The Result: The AI learned to spot the subtle differences. When a muon hits, it's a clean, direct hit. When a pion hits, it's often a messy cluster of hits.

5. The Results: A High-Performance Net

The results were impressive:

• Catching Muons: When the detector was set to catch 94% of the real muons, it worked perfectly.
• Rejecting Pions: The "fake muon" rate (mistaking a pion for a muon) dropped to just 2.4% when using a 70 cm thick iron wall.
• The Math: They found that as they made the iron wall thicker, the number of fake muons dropped exponentially (like a balloon deflating). The "slope" of this drop was very steep, meaning the detector is very good at filtering out the noise.

Why This Matters

This paper proves that their design works. They built a large, cheap, and effective detector that can be used in the future ALICE 3 experiment. By using simple plastic bars, light fibers, and smart AI, they can finally catch those elusive, slow-moving muons that other experiments miss.

In a nutshell: They built a giant, glowing, two-layered net, put it behind a thick iron wall, and taught a computer to ignore the noise. Now, they are ready to catch the ghosts of the universe.

Technical Summary

1. Problem Statement

The ALICE experiment at the LHC aims to study the strongly-interacting quark–gluon plasma (sQGP). While the current apparatus has been successful, fundamental questions regarding sQGP properties and strong interaction remain open after LHC Runs 3 and 4. To address these, the ALICE 3 upgrade is proposed for Run 5 (2036–2041).

A critical component of ALICE 3 is the Muon Identifier Detector (MID), designed to reconstruct $J/\psi$ mesons at rest via the dimuon decay channel. This requires identifying muons with momenta in the 1.5–5 GeV/c range. Existing LHC experiments struggle to identify muons below 6 GeV/c. The challenge is to design a detector that:

• Achieves high muon identification efficiency ($\sim$94%).
• Minimizes "fake-muon" efficiency (misidentifying pions as muons) to below 3%.
• Operates within a compact space behind a standard magnetic iron absorber (approx. 70 cm thick).
• Provides sufficient time resolution (few ns) to suppress slow particles and position resolution (few cm) limited by scattering.

2. Methodology

A. Detector Design and Construction

The authors designed and constructed a large-area prototype chamber ($1 \times 1 \text{ m}^2$) to validate the MID concept.

• Geometry: The chamber consists of two sensitive layers separated by a 1 cm air gap.
• Scintillator Bars: Each layer contains 24 scintillator bars (dimensions: $100 \times 4 \times 1 \text{ cm}^3$), manufactured by FNAL-NICADD. The bars are arranged orthogonally between layers to create an overlapping cell size of $4 \times 4 \text{ cm}^2$.
• Readout System:
  • Light Collection: Wavelength-shifting (WLS) fibers (Kuraray Y-11) are embedded in a central groove of the bars. Two fiber diameters (1.5 mm and 2.0 mm) were tested; the 2.0 mm fibers yielded a 40% increase in light yield.
  • Photodetectors: Light is read out by Hamamatsu SiPMs (model 14160-3050HS, $3 \times 3 \text{ mm}^2$ active area).
  • Mechanical Assembly: Bars are wrapped in aluminum tape. SiPMs are mounted using 3D-printed snap-fit holders fixed with optical cement mixed with carbon powder to avoid screws.
• Mechanical Structure: The assembly is supported by an aluminum framework with sills and angle supports. Simulations confirmed the structure could withstand a 10 kg load with a maximum deflection of only 0.386 mm.

B. Experimental Setup

The prototype was tested at the CERN T10 beamline using:

• Beams: 3 GeV/c $\pi^-$-enriched and $\mu^+$ beams.
• Absorber: A variable-length iron absorber (46 cm to 86 cm) placed in front of the chamber to simulate the ALICE 3 environment.
• Trigger & Veto:
  • A 5-fold coincidence of scintillator paddles provided the trigger.
  • A high-pressure threshold Cherenkov counter (XCET) was used to veto electrons, reducing electron contamination in the pion beam to $<0.3\%$.
  • A small reference chamber provided a precise time-of-arrival reference ($ToA_{ref}$).

C. Data Analysis Strategy

• Machine Learning: The analysis utilized Boosted Decision Trees (BDT) implemented in TMVA.
• Training Data: 50% of the collected data (pure muon signal and pion background) was used for training.
• Input Variables: The BDT was trained on:
  • Number of hits per event ($N_{hit}$).
  • Charge deposition.
  • Time-of-arrival relative to reference ($\Delta ToA$).
  • Fired channel positions.
• Validation: The remaining 50% of data was used to measure muon efficiency and hadron suppression.

3. Key Contributions

1. Large-Scale Prototype: Successful construction and testing of a full-scale ($1 \text{ m}^2$) two-layer scintillator chamber, moving beyond previous small-scale prototypes.
2. Optimization of Light Yield: Demonstration that increasing WLS fiber diameter from 1.5 mm to 2.0 mm significantly improves light collection (40% gain), a critical factor for SiPM performance.
3. Mechanical Innovation: Development of a screw-free, 3D-printed snap-fit holder system for SiPMs, ensuring mechanical stability and ease of assembly.
4. ML-Driven Analysis: Implementation of a BDT algorithm specifically trained to distinguish muons from pions in a high-background environment, optimizing the trade-off between efficiency and purity.

4. Results

• Muon Efficiency: The detector achieved a muon identification efficiency above 99% for the "OR" condition (detection in either Layer 1 or Layer 2). When constrained to a 94% efficiency (the target for ALICE 3), the system performed robustly.
• Fake-Muon Efficiency (Pion Rejection):
  • For an absorber thickness of 70 cm (the target for ALICE 3), the fake-muon efficiency was 2.4% $\pm$ 0.1%.
  • The fake-muon efficiency as a function of absorber length follows an exponential decay ($e^{-x/\lambda}$) with a slope parameter $\lambda = 18.79$ cm.
  • A constant background term of 3.1% was observed, attributed to residual muon contamination from pion decays in flight.
• Performance Validation: The experimental results align well with GEANT 4 simulations. The data-driven background efficiency using the large prototype was found to be lower than simulations including all secondary particles but higher than simulations considering only primary pions, validating the effectiveness of the template-fit method used to suppress secondary contributions.

5. Significance

This work validates the feasibility of the scintillator-bar technology for the ALICE 3 MID subsystem. The results confirm that:

• A two-layer scintillator design is sufficient to meet the physics requirements for muon reconstruction in the 1.5–5 GeV/c momentum range.
• The combination of FNAL-NICADD scintillators, optimized WLS fibers, SiPM readout, and Machine Learning analysis can achieve the required >94% muon efficiency and <3% fake-muon rate.
• The mechanical design is robust and scalable for the full 180-chamber ALICE 3 system.

These findings provide a strong technical foundation for proceeding with the full construction of the ALICE 3 detector, enabling unique physics capabilities in heavy-ion collisions that are currently inaccessible to other LHC experiments. Future work involves testing longer bars (1.5 m) and scintillators from different vendors.