TL;DR
This paper introduces a novel particle identification method for liquid argon time projection chambers (LArTPCs) that accounts for angular dependencies in calorimetric measurements. This new approach significantly improves the ability to distinguish between protons and muons, achieving a 93.7% proton selection efficiency with only a 10% muon mis-identification rate, and is applicable to future LArTPC experiments.
What Problem Does This Paper Address?
Neutrino detectors like MicroBooNE use liquid argon time projection chambers (LArTPCs) to reconstruct particle trajectories and identify their types based on how they deposit energy (calorimetry). A key challenge is that the accuracy of these calorimetric measurements, specifically the energy deposited per unit length (dE/dx), is not uniform across the detector. This non-uniformity arises from the geometry of the detector and the way particle tracks interact with the wire planes, leading to a dependence on the track's angle relative to the detector's drift and wire directions. This angular dependence directly impacts the performance of particle identification (PID) algorithms, making it harder to reliably distinguish between different particle species, particularly protons and muons, which have similar masses and energy deposition patterns.
Methods
The core of this work is a new likelihood-based method for particle identification. This method addresses the angular dependency by:
- Modeling the dE/dx Probability Density Function (PDF): Instead of relying solely on theoretical models, the dE/dx PDFs for muons and protons are estimated from detailed MicroBooNE simulations. These PDFs are not just a function of dE/dx and residual range, but crucially also incorporate the local pitch (a measure related to the track's angle and wire spacing). This three-dimensional histogram approach captures the complex interplay of factors affecting the measured dE/dx.
- Combining Information from All Three Wire Planes: LArTPCs typically have three wire planes (collection, and two induction planes). The new method computes a single likelihood for a track by combining the PDFs from each of these planes. This leverages the full calorimetric information available, even if the reconstruction is less accurate on one or more planes due to angular effects.
- Likelihood Ratio Test Statistic (PID Score): The method calculates a likelihood ratio between the proton and muon hypotheses. This ratio is then transformed into a PID score (P) that ranges from -1 to 1, making it easier to interpret and display. A score closer to 1 indicates a higher probability of being a muon, while a score closer to -1 indicates a higher probability of being a proton.
This approach is a significant improvement over previous methods that either used simpler models or relied on single-plane information, which were more susceptible to angular biases.
Datasets
The primary dataset used for developing and evaluating this new PID method is MicroBooNE simulation data. This simulation is crucial for accurately modeling the complex detector effects and angular dependencies. The performance of the method is then validated using MicroBooNE beam data collected during the winter and spring 2016 runs, corresponding to 4.8 x 10^19 protons on target (POT). The paper also compares its performance against a previous state-of-the-art method using a "chi-squared test with respect to the proton hypothesis" which was based on Bethe-Bloch theory.
Key Insights and Ideas
- Calorimetry is Angle-Dependent: The most fundamental insight is that even in a "perfect" detector, calorimetric measurements like dE/dx are inherently dependent on the track's trajectory angle. This is a crucial realization for designing accurate PID systems in LArTPCs.
- Simulation is Key for Complex Effects: Accurately capturing these angular dependencies and detector distortions requires sophisticated simulation. The paper demonstrates the power of using simulated data to build robust PID models.
- Multi-Plane Information is Essential: Combining information from all three wire planes is critical for overcoming the limitations of individual planes, especially for tracks with challenging angles. This holistic approach significantly boosts separation power.
- A Unified PID Score: The transformation of the likelihood ratio into a bounded PID score (P) simplifies interpretation and comparison, making it a more user-friendly tool for physicists.
Key Results
- Improved Proton/Muon Separation: The new three-plane PID method achieves a 93.7% proton selection efficiency at a 10% muon mis-identification rate. This is a substantial improvement over previous methods, such as the collection-plane-only PID (83.4% efficiency) and the chi-squared test (81.6% efficiency) at the same muon mis-identification rate.
- Higher Area Under the ROC Curve (AUC): The overall separation power, quantified by AUC, is 0.977 for the new method, compared to 0.955 for the collection-plane-only method and 0.838 for the chi-squared test.
- Reduced Angular Dependence: The new method shows a significantly reduced dependence on the track's azimuthal angle compared to older methods, making it more robust across different track orientations.
- Successful Application to Exclusive Channels: The paper demonstrates the method's utility by successfully classifying exclusive neutrino interaction final states (e.g., νμCC0π1p, νμCC1π0p), which is a novel result for liquid argon detectors.
Limitations and Future Work
- Dependence on Simulation Accuracy: While the simulation is detailed, its accuracy in modeling all detector effects (like electron-ion recombination) can introduce systematic uncertainties. The paper acknowledges this and tests the robustness of its performance against such uncertainties.
- Pion and Kaon Classification: The current method primarily focuses on distinguishing muons from protons. Pions are treated similarly to muons, and kaons are omitted due to their rarity. Future work could extend the PID to include these particles.
- Further Improvement with Additional Cuts: While the new PID score is powerful, the paper notes that further improvements in classifying exclusive final states could benefit from additional cut variables to further reduce background.
- Applicability to Future Detectors: The authors emphasize that this methodology is directly transferable to other LArTPC experiments like SBND, ICARUS, and DUNE, making it a valuable tool for the broader liquid argon neutrino physics program.
Summary updated