Speaker
Description
Generalized Parton Distributions (GPDs) encode the three-dimensional partonic structure of the nucleon, but are accessible experimentally only through Compton form factors (CFFs), and only in part of their kinematic domain. Reconstructing the full GPD H(x,ξ) therefore requires a parametrization that links the measured DGLAP region (|x|>ξ) to the unmeasured ERBL region (|x|<ξ). We present a neural-field reconstruction of the underlying double distribution (DD) F(β,α), from which H(x,ξ) is obtained by the Radon transform — a construction that builds polynomiality into the architecture and propagates information from the DGLAP region into the ERBL region by design. Uncertainty is propagated via a replica ensemble with a Goloskokov–Kroll closure test, fitting Compton form factors with experimental ξ coverage, and against a direct H parametrization, without the double distribution as a baseline. We find that the ERBL region of the GPD is recovered through the DD prior and that the recovered α-profile depends on the regularization prescription in a way that is consistent with the existence of shadow double distributions. Ongoing work targets Bayesian uncertainty quantification such as MC-dropout and stochastic weight averaging for null-space sampling, sensitivity analysis via the input–output Jacobian to identify the most informative kinematic regions, which provides opportunities to understand the impact of data provided by future experiments and facilities.