Exploring the substructure of matter with Machine Learning
One of the most important features of NNs is its compliance with the universal approximation theorem...
2 minute read
In 1983 at CERN, a difference between the probability that an electron scatters off an atomic nucleus and that of the same number of free protons and neutrons was observed...
2 minute read
Fragmentation Functions (FFs) encode the long-distance dynamics of the
interactions among quarks and gluons which lead to their hadronisation...
2 minute read
The strong force, described by the theory of quantum chromodynamics (QCD), manifests itself through the constant
exchange of gluons between the quarks. Gluons, like photons, are elementary massless particles that define the nature of
the force they carry. Gluons interact 137 stronger than photons at distances similar to the size of the nucleus (~10
-15
m). The gluon interaction becomes even stronger for larger distances and weaker for smaller ones, a feature called
asymptotic freedom. This property leads to the confinement of quarks and gluons (collectively called partons) within a
nucleon and is also responsible for the fact that a parton cannot be measured or observed as a free particle. In fact
whenever a parton gets knocked out of a nucleon, it instantly starts to fragment into other partons until it hadronises,
forming a new hadron (a bound state of 2 or more quarks).
QCD yields many quantities that are not calculable by means of perturbative
methods and needs to be extracted from measured data. That is due to the observed nature of the strong force that makes
the interactions between quarks and gluons becomes strong (non-perturbative) at low-energies, leading to their
confinement within composite hadrons.
← My research focused on analysing the one-dimensional substructure of the nucleons and nuclei using machine learning
techniques. By one-dimensional, I refer to the determination of the distribution of a nucleon's and nuclei'
longitudinal momentum (along their beam axis) among their quarks and gluons constituents at high-energies. That is
achieved by parameterising non-perturbative quantities called parton distribution functions (PDFs) by means of
artificial neural networks. In addition to PDFs, I studied related non-perturbative dynamics that describe the hadronisation of
partons into hadrons, which are encoded by the fragmentation functions (FFs). Being universal approximators to any
continuous function, the usage of neural networks is well justified to fit quantities like PDFs whose functional form is
not predicted by QCD. While nucleons PDFs are currently determined up to high accuracies due to the
abundant data from the LHC and older collider and fixed target experiments, nuclear PDFs determination is less accurate
and their deviation from the nucleons PDFs is yet to be understood theoretically.