Parton Distribution Functions

The analysis of LHC data is complicated by the fact that it is a proton-proton collider. The underlying hard processes which may produce new physics involve instead parton-parton collisions, where a 'parton' is a quark, antiquark or gluon. These hard processes may be computed using perturbative QCD. However to turn these QCD calculations into physical cross-sections, which can be compared to actual data, we need to know the probability of finding a given parton in the proton. This information is encoded in so-called 'parton distribution functions' (or PDFs for short). Since PDFs cannot be computed in perturbation theory, they must in turn be inferred from experimental data. PDF determinations are thus the key to understanding LHC physics: without them we would have no way of comparing theory with data, and thus no way of reliably exploring possible new physics.

There are a number of international collaborations around the world working on PDF determination. In Edinburgh we belong to one of the largest and most successful of these, the NNPDF Collaboration. NNPDF use neural networks to extract the PDFs, and Monte Carlo methods to determine their statistical distribution. In recent years deep learning has attracted a lot of attention and research and my project will be working on improving the methodology behind determining the PDFs using novel machine learning techniques.