Feldman-Cousins' ML Cousin: Sterile Neutrino Global Fits using Simulation-Based Inference

TL;DR

This paper introduces a machine learning-based method that significantly accelerates global fits in particle physics, especially for sterile neutrino analyses, by approximating Bayesian inference more efficiently than traditional Feldman-Cousins procedures.

Contribution

The authors develop a novel simulation-based inference technique that is 200 times faster than Feldman-Cousins, enabling practical global fits in complex particle physics analyses.

Findings

Achieved a 200-fold speedup over Feldman-Cousins method.

Successfully applied the method to sterile neutrino global fits.

Demonstrated potential for broader application in particle physics analyses.

Abstract

For many small-signal particle physics analyses, Wilks' theorem, a simplifying assumption that presumes log-likelihood asymptotic normality, does not hold. The most common alternative approach applied in particle physics is a highly computationally expensive procedure put forward by Feldman and Cousins. When many experiments are combined for a global fit to data, deviations from Wilks' theorem are exacerbated, and Feldman-Cousins becomes computationally intractable. We present a novel, machine learning-based procedure that can approximate a full-fledged Bayesian analysis 200 times faster than the Feldman-Cousins method. We demonstrate the utility of this novel method by performing a joint analysis of electron neutrino/antineutrino disappearance data within a single sterile neutrino oscillation framework. Although we present a prototypical simulation-based inference method for a sterile neutrino global fit, we anticipate that similar procedures will be useful for global fits of all kinds, especially those in which Feldman-Cousins is too computationally expensive to use.