Evaluating Neutrino Energy Reconstruction Biases in DUNE

University of Edinburgh | Senior Honours Project

Overview

This repository contains the code and analysis for my Senior Honours Project investigating neutrino energy reconstruction methods for the Deep Underground Neutrino Experiment (DUNE).

Due to complex nuclear effects (such as Final State Interactions and pion absorption) within the argon nucleus, traditional analytic methods for calculating neutrino energy often suffer from systematic biases. This project evaluates traditional calorimetric and quasi-elastic (QE) kinematic formulas against a Machine Learning approach to provide a more robust, data-driven energy estimation.

Project Scope & My Contributions

Base Framework: The foundational CAFAna analysis scripts (located in /baseline_analysis/) were provided as part of the course materials. I completed these scripts to perform traditional calorimetric and kinematic energy reconstructions.

My Original Contribution (The ML Extension): Recognizing the limitations and biases of the traditional analytic formulas, I independently designed and implemented a Machine Learning solution.

• Developed a Gradient Boosted Decision Tree (BDT) regression model using ROOT's TMVA framework (see /ml_bdt_extension/).
• Engineered relevant features from the reconstructed detector data to train the model to recognize events with missing hadronic energy.
• Validated the model against an independent test set, proving a significant reduction in reconstruction bias compared to traditional methods.

Repository Structure

• /ml_bdt_extension/: Custom Machine Learning pipeline (BDT training, evaluation, and residual analysis).
• /baseline_analysis/: Foundational CAFAna scripts for traditional energy reconstruction and interaction mode breakdowns.
• /plots_and_results/: High-resolution ROOT canvas outputs showing smearing matrices, resolution distributions, and BDT model validation.

Prerequisites & Dependencies

To run the scripts in this repository, you must have the following High-Energy Physics (HEP) software stack installed:

• C++ Compiler: GCC (v8.2.0 or higher recommended)
• ROOT: v6.22 or higher (must be compiled with TMVA support for the ML extension)
• CAFAna: DUNE's Common Analysis File framework (cafanacore v01.17 or similar)
• Boost: v1.73.0 or higher

Note: The original analysis was performed on the University of Edinburgh compute cluster utilizing the Fermilab software environment (CVMFS). If running locally, ensure your ROOT and CAFAna environment variables are sourced correctly.

How to Run

1. Running the Baseline Analytic Reconstructions

The baseline analysis utilizes standard CAFAna cafe commands to parse the simulation data and plot the traditional reconstruction methods.

To execute an interaction mode breakdown or analytic comparison:

# Execute the base exercises
cafe baseline_analysis/Exercise1.C
cafe baseline_analysis/Exercise2.C

# Execute the analytic method comparison
cafe baseline_analysis/Methodcomparison.C