Efficient Neutrino Oscillation Parameter Inference with Gaussian Process

Lingge Li, Nitish Nayak, Jianming Bian, Pierre Baldi

Introduction

Neutrino oscillation studies are fundamental to our understanding of particle physics, yet the statistical methods required to infer oscillation parameters are notoriously computationally expensive. The standard Feldman-Cousins (FC) method, while robust for small datasets and complex physics, requires exhaustive Monte Carlo simulations across a dense parameter grid. This often translates to millions of CPU hours for a single analysis. This paper presents an iterative approach using Gaussian Processes (GP) to significantly reduce this computational burden by focusing resources on the most critical regions of the parameter space.

Methodology: Gaussian Process Iterative Refinement

The proposed method replaces the brute-force grid search with an active learning framework based on Bayesian Optimization. Instead of calculating critical values for every point, a Gaussian Process serves as a surrogate model to estimate the underlying function of critical values, $c(\theta)$.

The GP iterative process: (Top-Left) Surrogate model of critical values; (Top-Right) Smart sampling of grid points; (Bottom-Left) Acquisition function identifying areas of high uncertainty; (Bottom-Right) Resulting confidence contours.

The core of the efficiency lies in the acquisition function, which identifies points where the GP is most uncertain about the boundary of the confidence contour. By iteratively sampling only these high-information points and updating the surrogate model, the algorithm converges on the final contour without needing to explore the entire grid.

Experimental Evaluation

The method was tested using a toy neutrino oscillation experiment focusing on the parameters $\sin^2 \theta_{23}$ and $\delta_{CP}$. A "true" ground-truth contour was established using a standard 400-point FC grid search. The GP-based approach was then evaluated against this baseline across 200 simulated datasets to ensure statistical consistency.

Results and Performance

The GP iterative method achieved a 70% reduction in computational cost compared to the traditional grid search. As shown in the performance analysis, the method reaches high accuracy with a fraction of the total grid points.

Confidence contour overlap versus the percentage of grid points sampled. The method achieves near-perfect overlap with the ground truth using only 30% of the computational budget.

Key findings include:

  • Efficiency: Identical results were produced using only 30% of the original computation budget.
  • Accuracy: The confidence contours generated by the GP showed exceptional overlap with the ground-truth FC contours, validating the method's reliability.
  • Convergence: The acquisition function effectively prioritized sampling near the 68% and 95% confidence boundaries, ignoring regions deep inside or far outside the contours.

Conclusion

By shifting from exhaustive calculation to active learning, researchers can accelerate neutrino parameter inference by an order of magnitude. While currently demonstrated in 2D, future work will focus on scaling this approach to higher-dimensional spaces involving multiple systematic uncertainties. This technique holds broad potential for any high-energy physics application currently limited by the computational intensity of the Feldman-Cousins method.

Summary updated

Abstract

Many experiments have been set-up to measure the parameters governing the neutrino oscillation probabilities accurately, with implications for the fundamental structure of the universe. Very often, this involves inferences from tiny samples of data which have complicated dependencies on multiple oscillation parameters simultaneously. This is typically carried out using the unified approach of Feldman and Cousins which is very computationally expensive, on the order of tens of millions of CPU hours. In this work, we propose an iterative method using Gaussian Process to efficiently find a confidence contour for the oscillation parameters and show that it produces the same results at a fraction of the computation cost.
BibTeX
@inproceedings{li2019efficient,
  author = {Lingge Li and Nitish Nayak and Jianming Bian and Pierre Baldi},
  title = {{Efficient Neutrino Oscillation Parameter Inference with Gaussian Process}},
  booktitle = {AAAI Conference on Artificial Intelligence},
  year = {2019},
  doi = {10.1609/aaai.v33i01.33019967},
}
Figure