A Modern Extraction of Compton Form Factors from Deeply Virtual Compton Scattering

Over the last 20 years, there has been intense experimental activity dedicated to the measurement of observables to help build a 3D description of the nucleon. Generalized Parton Distributions (GPDs) describe complementary aspects of the structure of hadrons, providing qualitative and quantitative information about the partonic structure and dynamics such as orbital angular momentum. The cleanest process to access GPDs is the Deeply Virtual Compton Scattering (DVCS), where the cross-section is parametrized in terms of Compton Form Factors (CFFs) which are convolutions of GPDs with coefficient functions computed from perturbative QCD. The CFFs are extracted from DVCS experimental data taken at Jefferson Lab, including the most recent Hall A data. This analysis consists of a novel local fitting technique using chi-squared maps to constrain the CFFs. The CFFs, ReH, ReE, and ReHt are determined independently in each kinematic bin for the unpolarized beam-target configuration at twist-2 approximation using the formalism developed by A.V. Belitsky, D. Müller, and A. Kirchner (BKM10). The resulting CFFs are used to train and regularize a Deep Neural Network (DNN) to obtain a global behavior of the CFFs with minimal model dependency. An advanced DNN-driven extraction with prediction capabilities that shows potential for improving the accuracy, precision, and reliability of future CFFs extractions is introduced. These procedures are tested and systematically studied using pseudo-data generated with kinematics mimicking the experimental data.