Abstract: The development and operation of Liquid-Argon Time-Projection Chambers for neutrino physics has created a need for new approaches to pattern recognition in order to fully exploit the imaging capabilities offered by this technology. Whereas the human brain can excel at identifying features in the recorded events, it is a significant challenge to develop an automated, algorithmic solution. The Pandora Software Development Kit provides functionality to aid the design and implementation of pattern-recognition algorithms. It promotes the use of a multi-algorithm approach to pattern recognition, in which individual algorithms each address a specific task in a particular topology. Many tens of algorithms then carefully build up a picture of the event and, together, provide a robust automated pattern-recognition solution. This paper describes details of the chain of over one hundred Pandora algorithms and tools used to reconstruct cosmic-ray muon and neutrino events in the MicroBooNE detector. Metrics that assess the current pattern-recognition performance are presented for simulated MicroBooNE events, using a selection of final-state event topologies.
Introduction
Liquid-Argon Time-Projection Chambers (LArTPCs) serve as high-definition imaging detectors for neutrino physics, capturing subatomic interactions with remarkable detail. However, the automated reconstruction of 3D particle interactions from raw 2D data is a formidable challenge. In the MicroBooNE experiment, this task is further complicated by a high background of cosmic-ray muons—approximately 20 for every single neutrino interaction.
The Pandora Software Development Kit (SDK) addresses this "needle in a haystack" problem using a multi-algorithm approach. By breaking down complex pattern recognition into a chain of over 100 specialized, "safe" steps, it provides a robust framework for distinguishing rare neutrino events from the constant rain of cosmic-ray background.
The Multi-Algorithm Philosophy
Rather than relying on a single monolithic "black box" algorithm, Pandora utilizes a "divide and conquer" strategy. Each algorithm in the chain is designed to perform a specific, well-defined task, such as merging clusters that are clearly contiguous or splitting them at sharp kinks. By ensuring each individual step is highly reliable, the system avoids cascading errors where a single incorrect merge could ruin the entire reconstruction. Complex topological decisions are deferred until simpler, unambiguous ones are resolved.
A central concept in this framework is the PFParticle (Particle Flow Particle) hierarchy. As shown in Figure 1, a PFParticle serves as a parent object that links various reconstruction products, including 3D vertices, 2D clusters, and 3D tracks or showers. This hierarchical structure allows the software to represent complex decay chains, such as a muon producing a delta ray or a neutral pion decaying into two photons.
Reconstruction Workflow: PandoraCosmic and PandoraNu
The reconstruction process is divided into two distinct passes, each tailored to a specific goal:
- PandoraCosmic: This initial pass focuses on identifying and tagging obvious cosmic-ray muons. By removing these hits early, the system significantly reduces the complexity of the remaining data.
- PandoraNu: The remaining hits are passed to
PandoraNu, which is optimized for neutrino interaction topologies. It performs event slicing, identifies the neutrino interaction vertex, and reconstructs the resulting daughter particles.
To transition from 2D views to 3D space, Pandora employs rank-three tensors to compare clusters across the three wire planes ($u, v, w$). By looking for consistent overlaps in time and space, the software resolves ambiguities where multiple particles might overlap in a single 2D projection—a common occurrence that can create "ghost hits" if not handled mathematically.
Performance and Results
The framework's performance was evaluated using large-scale simulations tailored to the MicroBooNE detector, utilizing tools like GENIE for neutrino interactions and CORSIKA for cosmic-ray backgrounds. The results demonstrate high efficiency, particularly for track-heavy events:
- Simple Interactions: For quasi-elastic events (one muon, one proton), Pandora achieves a correct reconstruction rate of 86%.
- Muon Reconstruction: Individual muon reconstruction efficiency is consistently high, often exceeding 95%.
- Shower Challenges: Performance is lower for events involving neutral pions ($\pi^0 \to \gamma\gamma$), with a 50% perfect reconstruction rate. This is primarily due to the difficulty of capturing sparse, low-energy photon hits.
Conclusion and Future Directions
The Pandora multi-algorithm approach provides a robust and scalable framework for automated pattern recognition in LArTPCs. While highly effective, challenges remain in separating collinear tracks and reconstructing fragmented, low-energy showers. Future iterations of the framework may benefit from integrating deep learning techniques, such as Convolutional Neural Networks (CNNs), to enhance specific tasks like hit classification within the existing multi-algorithm chain. This hybrid approach could combine the interpretability of traditional algorithms with the feature-extraction power of modern machine learning.
Summary updated