Machine Learning Neutrino-Nucleus Cross Sections II: More Leptons
Abstract
Neutrino-nucleus cross sections are essential inputs for long-baseline neutrino oscillation experiments, yet remain challenging to model with high precision. Recent work demonstrated that fully black-box data-driven cross-section models cannot accurately generalize between near- and far-detector fluxes, but imposing Standard-Model symmetries via structure-function parameterizations provides a well-posed learning problem in a simplified DUNE-inspired muon-neutrino disappearance analysis. Here, we extend this framework to multiple lepton flavors and both neutrino and antineutrino beam modes, incorporating the electron appearance channel and enabling a simultaneous determination of sin2θ23, Δm231, and the CP-violating phase δCP. We find that the learned cross section model recovers the injected oscillation parameters without significant bias, with constraints comparable to those obtained using the exact underlying cross section. Flux uncertainties have more significant effects in the multi-channel case, and incorporating flux nuisance parameters into the learning problem is required for accurate results. Our results demonstrate that this data-driven approach can be extended to the multi-channel measurements needed for long-baseline neutrino oscillation analyses.
Related
- Sterile Neutrino Dark Matter Cries for GeV Heavy Neutral Leptons
- Machine-learning-based neutrino-event generation constrained by experimental measurements
- Anomalous horizontal track-like events at the HAWC observatory: candidate neutrino-induced charged leptons from the Pico de Orizaba volcano
- First Measurement of Differential Cross Sections in Kinematic Imbalance Variables with Electron Neutrino Interactions Using the MicroBooNE Detector
- Testing coherent elastic neutrino-nucleus scattering N dependence
BibTeX
@misc{hackett2026machine,
title = {{Machine Learning Neutrino-Nucleus Cross Sections II: More Leptons}},
author = {Daniel C. Hackett and Joshua Isaacson and Shirley Weishi Li and Karla Tame-Narvaez and Michael L. Wagman},
howpublished = {arXiv},
year = {2026},
month = oct,
doi = {10.48550/arxiv.2610.30338},
eprint = {2610.30338},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2610.30338},
}
RIS
TY - UNPB
TI - Machine Learning Neutrino-Nucleus Cross Sections II: More Leptons
AU - Daniel C. Hackett
AU - Joshua Isaacson
AU - Shirley Weishi Li
AU - Karla Tame-Narvaez
AU - Michael L. Wagman
PY - 2026
DA - 2026/10/21
PB - arXiv
DO - 10.48550/arxiv.2610.30338
UR - https://arxiv.org/abs/2610.30338
AB - Neutrino-nucleus cross sections are essential inputs for long-baseline neutrino oscillation experiments, yet remain challenging to model with high precision. Recent work demonstrated that fully black-box data-driven cross-section models cannot accurately generalize between near- and far-detector fluxes, but imposing Standard-Model symmetries via structure-function parameterizations provides a well-posed learning problem in a simplified DUNE-inspired muon-neutrino disappearance analysis. Here, we extend this framework to multiple lepton flavors and both neutrino and antineutrino beam modes, incorporating the electron appearance channel and enabling a simultaneous determination of sin^2θ_23, Δm^2_31, and the CP-violating phase δ_CP. We find that the learned cross section model recovers the injected oscillation parameters without significant bias, with constraints comparable to those obtained using the exact underlying cross section. Flux uncertainties have more significant effects in the multi-channel case, and incorporating flux nuisance parameters into the learning problem is required for accurate results. Our results demonstrate that this data-driven approach can be extended to the multi-channel measurements needed for long-baseline neutrino oscillation analyses.
ER -