A first unbiased global NLO determination of parton distributions and their uncertainties

TL;DR

This paper provides the first unbiased global determination of parton distribution functions at NLO using a comprehensive dataset and advanced neural network techniques, achieving high accuracy and consistency across datasets.

Contribution

It introduces an improved neural network training method and a novel approach to normalization uncertainties in global PDF fitting at NLO.

Findings

High consistency among datasets and with NLO QCD.

Some PDFs relevant for LHC are determined more accurately than previous fits.

No evidence of tension between datasets.

Abstract

We present a determination of the parton distributions of the nucleon from a global set of hard scattering data using the NNPDF methodology: NNPDF2.0. Experimental data include deep-inelastic scattering with the combined HERA-I dataset, fixed target Drell-Yan production, collider weak boson production and inclusive jet production. Next-to-leading order QCD is used throughout without resorting to K-factors. We present and utilize an improved fast algorithm for the solution of evolution equations and the computation of general hadronic processes. We introduce improved techniques for the training of the neural networks which are used as parton parametrization, and we use a novel approach for the proper treatment of normalization uncertainties. We assess quantitatively the impact of individual datasets on PDFs. We find very good consistency of all datasets with each other and with NLO QCD, with no evidence of tension between datasets. Some PDF combinations relevant for LHC observables turn out to be determined rather more accurately than in any other parton fit.