Electromagnetic Shower Reconstruction and Identification in FASER's Emulsion Detector for LHC Forward Neutrino Measurements

This paper presents a validated framework for reconstructing and identifying electromagnetic showers in the FASERnu emulsion detector using CERN test-beam data, achieving high background rejection and efficient energy reconstruction to support electron neutrino measurements at the LHC.

The Gist

Imagine the Large Hadron Collider (LHC) as a massive, high-speed particle highway. Most of the traffic consists of heavy trucks (protons) smashing into each other, but hidden among the debris are tiny, ghostly cars called neutrinos. These neutrinos are so shy and light that they usually pass right through the walls of the giant detectors without anyone noticing.

The FASER experiment is like a specialized toll booth built 480 meters down the road, specifically designed to catch these ghostly neutrinos as they zoom past. Inside this booth sits a detector called FASERν, which is essentially a giant, high-tech sandwich made of hundreds of thin layers of tungsten (a heavy metal) and photographic film (emulsion).

When a neutrino hits the tungsten, it sometimes creates an electron. This electron doesn't just travel in a straight line; it hits the metal and sparks a chain reaction, creating a "shower" of smaller particles, much like a firework exploding or a snowball rolling down a hill and gathering more snow. This is called an Electromagnetic Shower.

The problem? The detector is also flooded with muons (another type of particle) that look very similar to these electron showers but are just boring, straight lines. The scientists need to find the "fireworks" (electrons) in a blizzard of "straight lines" (muons).

Here is how the paper explains the new method they developed to do this, using simple analogies:

1. The Detective's Strategy: Finding the "Fireworks"

The scientists built a three-step "filter" to separate the electron showers from the muon background.

• Step 1: The Rough Filter (Pre-selection)
  Imagine you are looking for a specific type of bird in a forest. First, you ignore anything that is too small or too big. The scientists do this by looking at how much a particle "wobbles" as it moves. Muons are like straight arrows; they don't wobble much. Electrons, because they are creating a shower, wobble and scatter a lot. They throw out the "straight arrows" immediately.

• Step 2: The Smart Cluster (Reconstruction)
  Once they have a candidate, they need to map out the shape of the shower. They use a computer algorithm called DBSCAN.

  • The Analogy: Imagine a crowded dance floor. Some people are dancing in a tight, energetic circle (the electron shower), while others are just walking around the edges (background noise). The algorithm looks for the "dense circle" of dancers. It doesn't need to know how many people are in the circle beforehand; it just finds the group where the people are packed tightest. This helps them draw the exact path of the electron shower, even if the data is messy.

• Step 3: The Expert Judge (Identification)
  After mapping the shower, they need to be 100% sure it's not a fake. They use a "Smart Judge" (a machine learning tool called a BDT).

  • The Analogy: Think of this like a bouncer at a club who checks a list of 10 different rules. Is the shower too wide? Is it too short? Does it spread out too fast? The "bouncer" looks at the shape of the particle trail. If it matches the "firework" pattern perfectly, it gets in. If it looks like a "straight arrow," it gets kicked out.
  • The Result: This system is incredibly good at its job. It successfully rejects 99.99% of the fake muons at lower energies and 99.94% at higher energies, while still catching about 60% to 70% of the real electron showers.

2. Measuring the Energy: Counting the Snowflakes

Once they have identified a real electron shower, they need to know how much energy it had.

• The Analogy: Imagine you are trying to guess how big a snowball was before it started rolling down a hill. You can't see the original snowball, but you can count how many snowflakes it collected by the time it stopped.
• In this experiment, the "snowflakes" are the tiny tracks left by particles in the film. The scientists simply count the total number of tracks in the shower. The more tracks, the more energy the original electron had.
• They found this method is very accurate. At 100 GeV (a specific energy unit), their guess was off by less than 1%. At 200 GeV, it was also off by less than 1%.

3. The "Fog" Problem (Systematic Uncertainties)

The biggest source of error isn't the math or the computer; it's the film itself.

• The Analogy: Imagine trying to count snowflakes, but sometimes the camera lens is slightly foggy, and you miss a few flakes. If the film is "foggy" (less efficient), you count fewer tracks and think the energy was lower than it really was.
• The paper admits that the biggest uncertainty comes from how well the film captures the tracks. Depending on how "foggy" the film is, their energy measurement could be off by about 10%. This is the main thing they need to be careful about.

Summary

The paper presents a new, highly effective way to find and measure electron neutrinos in the FASER detector. They built a digital "net" that:

1. Filters out straight-line particles.
2. Uses smart clustering to find the "dense" particle showers.
3. Uses a trained AI to verify the shape.
4. Counts the tracks to measure energy.

This method allows them to see the "ghostly" neutrinos clearly, even when they are hiding in a massive crowd of other particles, paving the way for better measurements of how neutrinos interact at the LHC.

Technical Summary

Technical Summary: Electromagnetic Shower Reconstruction and Identification in FASER's Emulsion Detector

Problem Statement
The FASER$\nu$ experiment at the LHC aims to study high-energy neutrino interactions, specifically electron neutrino ($\nu_e$) charged-current (CC) interactions, which produce electromagnetic (EM) showers. A primary challenge in this environment is the intense background of muons generated by the LHC, which outnumber signal electrons by orders of magnitude (approximately 1500 muons/cm² vs. 8 electrons/cm² in test conditions). Furthermore, existing reconstruction methods for the FASER$\nu$ emulsion detector were optimized for energies above 200 GeV and exhibited limitations at lower energies (tens of GeV) where showers produce fewer segments, making them harder to distinguish from background tracks. The paper addresses the need for a robust, energy-independent reconstruction and identification framework capable of operating across the full FASER$\nu$ energy range (from tens of GeV to multi-TeV) while maintaining high signal efficiency and extreme background rejection.

Methodology
The authors present a multi-stage pipeline for EM shower reconstruction and identification, validated using 100 GeV and 200 GeV electron test-beam data from the CERN SPS H4 beamline.

1. Data Processing and Pre-selection:

   • Emulsion films are scanned to reconstruct 3-D particle trajectories (base-tracks).
   • A track reconnection algorithm merges broken tracks.
   • Pre-selection: Tracks are filtered based on segment count ($3 \le n_{seg} \le 30$ for 100 GeV; $3 \le n_{seg} \le 40$ for 200 GeV) and scattering angle ($\theta_s \ge 2$ mrad). This removes fragmented tracks and through-going muons, retaining approximately 29% of tracks originating from the first plate.

2. Shower Reconstruction (Adaptive Clustering):

   • Initial Axis: Determined by fitting the first three segments of primary tracks.
   • Cylindrical Search: A preliminary volume (radius 100 $\mu$m, length 8 cm) is defined around the initial axis.
   • DBSCAN Clustering: An adaptive Density-Based Spatial Clustering of Applications with Noise (DBSCAN) algorithm is applied to segments on each plate. Crucially, the algorithm parameters are not energy-dependent but adapt to local segment density:
     • Neighborhood radius $\epsilon = 2.5 \times \text{median}(d_{NN})$, bounded between 10–80 $\mu$m.
     • Minimum cluster size $N_{min} = \max(2, \lfloor N_{seg}/4 \rfloor)$.
   • Axis Refinement: Centroids of selected clusters undergo sequential validation (rejecting deviations $>60$ $\mu$m) and are fitted using Huber robust regression to determine the final shower axis. This minimizes the influence of spurious centroids.

3. Identification Chain:

   • Loose ID (Cut-based): Removes bulk muon backgrounds mimicking showers using variables like maximum segment count per plate ($N_{max} \ge 4$), shower start plate ($\le 20$), and angular spread ($\sigma_\theta \ge 4$ mrad).
   • Medium ID (BDT-based): A Boosted Decision Tree (XGBoost) classifier uses ten shower shape variables designed to be energy-independent. These include:
     • Longitudinal: Concentration ratio, core fraction, peak fraction, early fraction, longitudinal RMS, and asymmetry.
     • Transverse/Clustering: Widening ratio, mean cluster number, and mean largest-cluster fraction.
     • Angular: Angular spread RMS ($\sigma_\theta$).
   • The classifier is trained on MC electrons and visually identified muon backgrounds from test-beam data.

4. Energy Reconstruction:

   • The total number of reconstructed segments ($N_{total}$) serves as the calorimetric estimator.
   • Calibration curves ($N_{total} = a(\epsilon) \cdot E + b(\epsilon)$) are derived from MC samples with varying single-film efficiencies to account for efficiency-dependent biases.
   • A "close-pair" inefficiency correction is applied for the 200 GeV module to account for overlapping grain patterns in dense shower cores.

Key Results

• Background Rejection: The combined identification chain achieves background rejection rates of 99.99% at 100 GeV and 99.94% at 200 GeV. The BDT classifier achieves an Area Under the Curve (AUC) of 0.997.
• Signal Efficiency: Total reconstruction and identification efficiencies are 58.9% at 100 GeV and 70.8% at 200 GeV (evaluated on MC).
• Energy Resolution and Bias:
  • At 100 GeV: Relative bias of +0.6%, resolution of 25.4%.
  • At 200 GeV: Relative bias of −0.8%, resolution of 22.6%.
• Systematic Uncertainties: The dominant systematic uncertainty arises from variations in single-film detection efficiency.
  • 100 GeV: $+10.9\% / -8.2\%$.
  • 200 GeV: $+10.3\% / -6.9\%$.
  • Background contamination contributes negligibly ($\sim 0.1\%$).

Significance and Claims
The paper claims to provide a validated framework for electron neutrino identification using the FASER$\nu$ detector. The primary significance lies in the development of an energy-independent reconstruction algorithm that does not require tuning for different energy scales, enabling its application across the broad FASER$\nu$ energy range (20 GeV to 300 GeV and beyond). The methodology successfully demonstrates that high-purity electron neutrino samples can be extracted from the intense LHC muon background using emulsion detectors. The authors state that once validated in the high-background LHC environment (with appropriate beam track masking), this method will be deployed to measure electron neutrino charged-current interaction cross-sections and study neutrino flavor composition at TeV energies. The work represents an improvement over previous methods optimized only for higher energies, extending the physics reach of the experiment to lower energy regimes.