Neural network generated parametrizations of deeply virtual Compton form factors

TL;DR

This paper introduces a neural network-based method to parametrize the Compton form factor H from DVCS data, enabling model-independent fits with realistic uncertainty estimates and predictions for experimental observables.

Contribution

It presents a novel neural network approach for extracting CFFs from DVCS data, improving uncertainty quantification and propagation compared to traditional methods.

Findings

Neural network parametrization accurately fits DVCS data

Provides realistic uncertainty estimates for CFF H

Predicts beam charge-spin asymmetry for COMPASS II

Abstract

We have generated a parametrization of the Compton form factor (CFF) H based on data from deeply virtual Compton scattering (DVCS) using neural networks. This approach offers an essentially model-independent fitting procedure, which provides realistic uncertainties. Furthermore, it facilitates propagation of uncertainties from experimental data to CFFs. We assumed dominance of the CFF H and used HERMES data on DVCS off unpolarized protons. We predict the beam charge-spin asymmetry for a proton at the kinematics of the COMPASS II experiment.