Reconstruction of Neutrino Interactions using Machine Learning and Calculation of Neutrino Energy from CCQE Interactions
Session Number
2
Advisor(s)
Dr. Zelimir Djurcic, Argonne National Laboratory
Location
A 113
Discipline
Physical Science
Start Date
15-4-2026 11:10 AM
End Date
15-4-2026 11:55 AM
Abstract
Neutrinos are one of the most abundant subatomic particles in the universe, yet they interact with matter very infrequently. They oscillate between three flavors: electron, muon, and tau neutrinos, with these oscillations depending on neutrino energy. Determining neutrino energy is therefore vital to studying neutrino behavior. The focus for this research was to use machine learning to model neutrino interactions, and identify specific interaction types. Code written in Python in a Jupyter Notebook was used to visualize reconstructed neutrino interactions and to calculate variables needed to estimate the neutrino energy. ML was used to determine which particles were produced, and to find Charged-Current Quasi-Elastic interactions. Determining the energy of the produced particles was used to calculate neutrino energy, specifically in CCQE interactions that produced one proton and one muon. Results showed that calculated neutrino energies aligned with predictions, and that reconstructions were generally able to identify particles produced in the interactions correctly. These findings showed the feasibility of the application of machine learning for event reconstruction, and the use of proton kinematics together with the muon direction to constrain neutrino energy, strategies which will be important for future projects like the DUNE, that will have to analyze large data sets.
Reconstruction of Neutrino Interactions using Machine Learning and Calculation of Neutrino Energy from CCQE Interactions
A 113
Neutrinos are one of the most abundant subatomic particles in the universe, yet they interact with matter very infrequently. They oscillate between three flavors: electron, muon, and tau neutrinos, with these oscillations depending on neutrino energy. Determining neutrino energy is therefore vital to studying neutrino behavior. The focus for this research was to use machine learning to model neutrino interactions, and identify specific interaction types. Code written in Python in a Jupyter Notebook was used to visualize reconstructed neutrino interactions and to calculate variables needed to estimate the neutrino energy. ML was used to determine which particles were produced, and to find Charged-Current Quasi-Elastic interactions. Determining the energy of the produced particles was used to calculate neutrino energy, specifically in CCQE interactions that produced one proton and one muon. Results showed that calculated neutrino energies aligned with predictions, and that reconstructions were generally able to identify particles produced in the interactions correctly. These findings showed the feasibility of the application of machine learning for event reconstruction, and the use of proton kinematics together with the muon direction to constrain neutrino energy, strategies which will be important for future projects like the DUNE, that will have to analyze large data sets.