Determination of the Atmospheric Neutrino Fluxes from Atmospheric Neutrino Data

TL;DR

The paper tackles the challenge of determining atmospheric neutrino fluxes directly from experimental data rather than relying solely on theoretical flux calculations. It introduces a data-driven framework that uses artificial neural networks as unbiased interpolants of the flux energy dependence, paired with Monte Carlo replicas to faithfully estimate uncertainties. By forward-folding event rates through a discretized, precomputed flux representation and optimizing network parameters with genetic algorithms, the authors extract an energy-dependent flux that is compatible with Honda/Bartol calculations within uncertainties, and they demonstrate robustness against training choices and oscillation-parameter variations. The approach provides a cross-check of flux models and lays the groundwork for incorporating high-energy data and achieving a full energy–zenith–flavor flux determination in future neutrino experiments.

Abstract

The precise knowledge of the atmospheric neutrino fluxes is a key ingredient in the interpretation of the results from any atmospheric neutrino experiment. In the standard atmospheric neutrino data analysis, these fluxes are theoretical inputs obtained from sophisticated numerical calculations based on the convolution of the primary cosmic ray spectrum with the expected yield of neutrinos per incident cosmic ray. In this work we present an alternative approach to the determination of the atmospheric neutrino fluxes based on the direct extraction from the experimental data on neutrino event rates. The extraction is achieved by means of a combination of artificial neural networks as interpolants and Monte Carlo methods for faithful error estimation