Electron Neutrino Energy Reconstruction in NOvA Using CNN Particle IDs

Shiqi Yu

Abstract NOvA is a long-baseline neutrino oscillation experiment. It is optimized to measure $\nu_e$ appearance and $\nu_\mu$ disappearance at the Far Detector in the $\nu_\mu$ beam produced by the NuMI facility at Fermilab. NOvA uses a convolutional neural network (CNN) to identify neutrino events in two functionally identical liquid scintillator detectors. A different network, called prong-CNN, has been used to classify reconstructed particles in each event as either lepton or hadron. Within each event, hits are clustered into prongs to reconstruct final-state particles and these prongs form the input to this prong-CNN classifier. Classified particle energies are then used as input to an electron neutrino energy estimator. Improving the resolution and systematic robustness of NOvA's energy estimator will improve the sensitivity of the oscillation parameters measurement. This paper describes the methods to identify particles with prong-CNN and the following approach to estimate $\nu_e$ energy for signal events.

Introduction

The NOvA experiment seeks to observe "$\nu_e$ appearance"—the oscillation of muon neutrinos into electron neutrinos over an 810 km baseline. To extract fundamental physics parameters such as the neutrino mass hierarchy and the CP-violating phase ($\delta_{CP}$), precise knowledge of the incoming neutrino energy is paramount. However, neutrino interactions produce complex final states with multiple particles, making energy estimation a significant challenge. Traditional methods often struggle to distinguish between electromagnetic and hadronic components, leading to systematic uncertainties and lower resolution.

Reconstruction and Particle Identification

The NOvA reconstruction pipeline begins by grouping detector hits into "slices" based on space and time, followed by identifying the 3D interaction vertex. Individual particle trajectories are then clustered into "prongs" using the FuzzyK algorithm.

The core of the new energy estimation method is the prong-CNN, a Siamese neural network designed to classify these prongs. The network takes two 2D views ($xz$ and $yz$ planes) of a single 3D prong as input. By using a Siamese architecture, the model can extract features from both views and combine them to output a particle identification score, effectively performing 3D classification with reduced computational overhead.

The Siamese architecture of the prong-CNN, which processes orthogonal 2D views to classify individual particles.

Prongs are classified as Electromagnetic (EM) if they receive high electron or photon ID scores, while others are categorized as Hadronic.

Energy Estimation

Once particles are classified, the total energy ($E_{reco}$) is estimated by summing the calibrated energies of the EM and hadronic components. To account for non-linear detector responses, a second-order polynomial fit is applied:

$$E_{reco} = 0.9595E_{EM} + 1.0198E_{HAD} + 0.0122E_{EM}^2 + 0.2163E_{HAD}^2$$

This estimator is tuned using Monte Carlo (MC) simulations. To ensure the model is not biased by the specific neutrino beam flux, a weighting technique is applied to the MC data to remove energy dependence before the final fit.

Results and Impact

The prong-CNN effectively separates EM showers from hadronic activity. The correlation between reconstructed components and the true neutrino energy is shown in the mapping below, which serves as the basis for the energy estimation.

Mapping of reconstructed hadronic and electromagnetic energy components against the average true neutrino energy.

This refined reconstruction method significantly enhances the precision of NOvA’s oscillation analysis. By improving energy resolution and reducing dependence on simulation details, the experiment can more effectively constrain $\delta_{CP}$ and determine the neutrino mass ordering.

Conclusion

The application of deep learning at the particle level represents a significant step forward for neutrino physics. By decomposing complex interactions into individual prongs and applying a Siamese CNN for identification, NOvA has developed a robust and accurate energy estimator. Future work will focus on improving the initial clustering of "fuzzy" electron tracks and expanding these CNN-based techniques to other interaction channels.

Summary updated

Abstract

NOvA is a long-baseline neutrino oscillation experiment. It is optimized to measure $ν_e$ appearance and $ν_μ$ disappearance at the Far Detector in the $ν_μ$ beam produced by the NuMI facility at Fermilab. NOvA uses a convolutional neural network (CNN) to identify neutrino events in two functionally identical liquid scintillator detectors. A different network, called prong-CNN, has been used to classify reconstructed particles in each event as either lepton or hadron. Within each event, hits are clustered into prongs to reconstruct final-state particles and these prongs form the input to this prong-CNN classifier. Classified particle energies are then used as input to an electron neutrino energy estimator. Improving the resolution and systematic robustness of NOvA's energy estimator will improve the sensitivity of the oscillation parameters measurement. This paper describes the methods to identify particles with prong-CNN and the following approach to estimate $ν_e$ energy for signal events.
BibTeX
@misc{yu2019electron,
  author = {Shiqi Yu},
  title = {{Electron Neutrino Energy Reconstruction in NOvA Using CNN Particle IDs}},
  year = {2019},
  eprint = {1910.06953},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/1910.06953},
}

Loading related content...

Figure