Abstract Wire-Cell is a 3D event reconstruction package for liquid argon time projection chambers (LArTPCs). Through geometry, time, and drifted charge from multiple readout wire planes, 3D space points with associated charge are reconstructed prior to the pattern recognition stage. Pattern recognition techniques, including track trajectory and dQ/dx (ionization charge per unit length) fitting, 3D neutrino vertex fitting, track and shower separation, particle-level clustering, and particle identification are then applied on these 3D space points as well as the original 2D projection measurements. A deep neural network is developed to enhance the reconstruction of the neutrino interaction vertex. Compared to traditional algorithms, the deep neural network boosts the vertex efficiency by a relative 30% for charged-current $\nu_e$ interactions. This pattern recognition achieves 80–90% reconstruction efficiencies for primary leptons, after a 65.8% (72.9%) vertex efficiency for charged-current $\nu_e$ ($\nu_\mu$) interactions. Based on the resulting reconstructed particles and their kinematics, we also achieve 15-20% energy reconstruction resolutions for charged-current neutrino interactions.
The Challenge of Neutrino Reconstruction
Neutrinos are fundamental particles that interact only via the weak force, making them exceptionally difficult to detect. Liquid Argon Time Projection Chambers (LArTPCs), such as the MicroBooNE detector, serve as high-resolution "electronic bubble chambers." When a neutrino interacts with an argon nucleus, it produces charged particles that ionize the liquid argon. An electric field drifts these electrons toward sense wire planes.
The primary reconstruction challenge lies in the data format: the detector provides 2D projections from multiple wire planes. Accurately identifying the neutrino interaction vertex—the point of origin for all secondary particles—is critical. Misidentification at this stage leads to cascading errors in particle classification and energy estimation. Traditional algorithms often struggle with complex topologies, such as low-energy "wiggled" electrons or events where multiple particles emerge from the same point.
The Wire-Cell Reconstruction Pipeline
The Wire-Cell framework addresses these challenges by performing 3D tomographic imaging before pattern recognition. This approach uses geometry and charge-matching constraints to convert 2D wire signals into a 3D point cloud of ionization charge.
The reconstruction follows a two-stage "Hybrid" process:
- First Iteration (Neutrino Vertexing): This stage combines traditional geometric algorithms with a deep learning component. Traditional methods search for topological features like kinks and "Bragg peaks" (high ionization at a particle's stopping point). Simultaneously, a Regressional Segmentation Network—a Sparse U-Net based on SparseConvNet—processes the 3D voxels to predict a confidence map for the vertex location. By using Sparse Convolutional Networks, the framework can process large 3D volumes with high resolution (0.5 cm voxels) without the massive memory overhead of dense 3D CNNs.
- Second Iteration (Particle Flow Tree): Once the vertex is established, the algorithm builds a hierarchical "Particle Flow Tree." This establishes parent-child relationships between particles, such as a primary muon producing a Michel electron, enabling a complete picture of the interaction.
Evaluation and Key Results
The framework was evaluated using high-statistics MicroBooNE Monte-Carlo (MC) simulations, including "Overlay MC" where simulated signals are superimposed on real cosmic-ray background data to ensure realistic noise conditions.
Key findings include:
- Vertex Efficiency: The hybrid approach achieved a vertex reconstruction efficiency (within 1 cm of the true location) of 65.8% for $\nu_e$CC and 72.9% for $\nu_\mu$CC interactions.
- DNN Impact: The integration of the Sparse U-Net provided a 30% relative boost in vertex identification efficiency for $\nu_e$ events compared to traditional algorithms alone. The DNN is excellent at finding the general region of a vertex in messy events, while traditional geometric fitting provides sub-centimeter precision.
- Particle Reconstruction: Primary leptons (muons and electrons) are reconstructed with 80–90% efficiency once the vertex is correctly identified.
- Energy Resolution: The system achieves an energy reconstruction resolution of 15–20% for charged-current interactions.
Conclusion and Future Outlook
The Wire-Cell 3D pattern recognition framework represents a significant advancement in LArTPC data analysis. By leveraging the sparsity of neutrino events through Sparse Convolutional Networks and combining them with robust geometric fitting, the team has achieved high-precision reconstruction suitable for large-scale experiments like DUNE. Future work will focus on improving voxel resolution and distinguishing between particles and anti-particles (e.g., $e^-$ vs $e^+$) to further enhance the physics reach of these detectors.
Summary updated