Improving Neutrino Oscillation Measurements through Event Classification

TL;DR

This work tackles the challenge of energy reconstruction systematics in long-baseline neutrino experiments by classifying events according to their underlying interaction type prior to energy reconstruction. Using supervised learning on labeled generator data, the authors demonstrate cross-generator generalization across $QE$, $MEC$, $RES$, and $DIS$ channels and integrate the classifier into a toy DUNE $\nu_\mu$ disappearance analysis. They show that event-class-based analyses can reduce oscillation-parameter uncertainties by $\sim$10–20% under perfect modeling and substantially mitigate biases under mismodeling, highlighting a practical method to lessen reconstruction-driven systematics. The findings suggest broad applicability to other experiments and potential improvements in cross-section tuning and extrapolation between near and far detectors.

Abstract

Precise neutrino energy reconstruction is essential for next-generation long-baseline oscillation experiments, yet current methods remain limited by large uncertainties in neutrino-nucleus interaction modeling. Even so, it is well established that different interaction channels produce systematically varying amounts of missing energy and therefore yield different reconstruction performance--information that standard calorimetric approaches do not exploit. We introduce a strategy that incorporates this structure by classifying events according to their underlying interaction type prior to energy reconstruction. Using supervised machine-learning techniques trained on labeled generator events, we leverage intrinsic kinematic differences among quasi-elastic scattering, meson-exchange current, resonance production, and deep-inelastic scattering processes. A cross-generator testing framework demonstrates that this classification approach is robust to microphysics mismodeling and, when applied to a simulated DUNE $ν_μ$ disappearance analysis, yields improved accuracy and sensitivity. These results highlight a practical path toward reducing reconstruction-driven systematics in future oscillation measurements.

Figures (7)

• Figure 1: Neutrino event spectrum in the DUNE near detector split according to interaction type. Quasi-elastic (QE), resonant (RES), deep inelastic scattering (DIS), and meson exchange currents (MEC) are shown from top to bottom in each panel.
• Figure 2: Fraction of missing energy for DUNE near detector events as a function of true neutrino energy. Different colors refer to the type of interaction, whereas G0.8ENIE and N0.8uWro results are distinguished via solid and dashed lines. Note that the DIS cross section is negligible below 1 GeV.
• Figure 3: Efficiency and contamination for the classification of QE and MEC events, assuming QE events as the signal. Results are shown for G0.8ENIE (blue) and N0.8uWro (orange) events. The dots indicate benchmark points obtained using the cuts defined in the main text.
• Figure 4: Efficiency and contamination for the classification of QE and MEC events in the generalization scheme, i.e., with the data and model corresponding to different generators. As before, QE events are treated as the signal. Results are shown for G0.8ENIE (blue) and N0.8uWro (orange) samples. The dots represent benchmark points obtained using cuts defined in the main text.
• Figure 5: Efficiency and contamination for G0.8ENIE (left) and N0.8uWro (right) in a multi-class classification using the one-vs-rest strategy.