Modeling inclusive electron-nucleus scattering with Bayesian artificial neural networks

TL;DR

This paper presents a Bayesian neural network approach to model inclusive electron-nucleus scattering, providing quantified uncertainties and novel response function extractions, improving predictive accuracy over previous methods.

Contribution

It introduces a Bayesian neural network framework for modeling response functions in electron-nucleus scattering, enabling uncertainty quantification and improved data analysis.

Findings

Accurately models scattering response functions across various nuclei.

Provides quantified uncertainties for the response functions.

Offers new extractions of longitudinal and transverse response functions.

Abstract

We introduce a Bayesian protocol based on artificial neural networks that is suitable for modeling inclusive electron-nucleus scattering on a variety of nuclear targets with quantified uncertainties. Unlike previous applications in the field, which directly parameterize the cross sections, our approach employs artificial neural networks to represent the longitudinal and transverse response functions. In contrast to cross sections, which depend on the incoming energy, scattering angle, and energy transfer, the response functions are determined solely by the energy and momentum transfer to the system, allowing the angular component to be treated analytically. We assess the accuracy and predictive power of our framework against the extensive data in the quasielastic inclusive electron-scattering database. Additionally, we present novel extractions of the longitudinal and transverse response functions and compare them with previous experimental analysis and nuclear ab-initio calculations.