Using analytic models to describe effective PDFs

Parton distribution functions (PDFs) play a pivotal role in hadron collider phenomenology. They are non-perturbative quantities extracted from fits to available data, and their scale dependence is dictated by the DGLAP evolution equations. In this article, we discuss machine learning (ML) strategies to efficiently compute PDFs by bypassing the scale evolution. Analytical approximations to the PDFs as a function of x and Q², including up to next-to-leading order (NLO) QCD effects, are obtained. The methodology is tested by reproducing the HERAPDF2.0 set and implementing the analytical expressions in benchmarking codes. It is found that the computational cost is reduced while the precision of the simulations stays well under control.