Collider, neutrino and dark relics of Grand Unification with Deep Learning


  • Call:

    PT-CERN Call 2020/2

  • Academic Year:


  • Domain:

    Astroparticle Physics

  • Supervisor:

    António Morais

  • Co-Supervisor:

    Felipe Ferreira de Freitas

  • Institution:

    Universidade de Aveiro

  • Host Institution:

    CIDMA - Centro de Investigação em Matemática e Aplicações da Universidade de Aveiro

  • Abstract:

    The last decade witnessed the confirmation of three outstanding milestones in fundamental physics recognized with the attribution of three Noble Prizes. In 2013, such an award was granted to the Higgs boson discovery dated from 2012 thus completing the Standard Model (SM) of Particle Physics. 2017 was the year of conceding the Nobel Prize to the discovery of Gravitational Waves in 2015, 100 years after Einstein has published his theory of General Relativity. Last but not least, 2015 was also the year where the discovery of neutrino flavour oscillations in 2002 was rewarded with a Nobel Prize. The latter discovery is one of the strongest evidences that the SM is by no means a complete description of particles and interactions. In fact, flavour oscillations between different neutrino species imply a quantum superposition of mass eigenstates which the SM cannot offer. Furthermore, the existence of dark matter (DM), from galaxy rotation curves and anisotropies in the Cosmic Microwave Background is becoming increasingly favoured by a particle physics explanation that is absent in the SM. With this thesis we aim at exploring both the DM and neutrino mass problems focusing on a novel trinification Grand Unified Theory (T-GUT) supplemented with a local family symmetry. Besides the unification of forces, this high-scale theory also features the unification of the Higgs and matter sectors. Therefore, the existence of a large unified representation yields a rich neutrino sector at the low-scale containing 9 Dirac weakly interacting components, of which 3 are sub-eV, and 6 heavy Majorana ones. One of the key studies proposed to develop in this thesis is to understand which regions of the parameter space allow for a Pontecorvo–Maki–Nakagawa–Sakata (PMNS) mixing and which implications it poses to the high-scale theory. Furthermore, the lightest Majorana neutrino can be sterile enough to provide a potential candidate for the DM. The new Dirac-type heavy neutrinos are accompanied by three generations of vector-like leptons offering an interesting opportunity for collider searches and phenomenological studies at the Large Hadron Collider. The low energy limit of the T-GUT framework also offers 2, down-type, singlet vector-like quarks (VLQ) not far from the TeV scale and at the reach of the LHC or future colliders. Furthermore, it follows from the unification picture that the mass spectrum of such vector-like fermions are directly related to the ultra-violet limit of the model and a potential discovery could be regarded as a relic of a more fundamental theory. The techniques we aim to develop in order to address the above questions will involve optimization algorithms based on Evolutionary searches for Deep Learning and Machine Learning tools. As a novel application, we intend to use such techniques as a “control” environment for Monte-Carlo event generator software, such as e.g. MadGraph, Pythia or MicrOmegas, in combination with model building tools such as SARAH and Feynrules. These studies also involve the development of algorithms based on probabilistic programming language and Bayesian Deep Learning. Such techniques will be used as a generative model environment, allowing us o make Monte-Carlo event generators to work in a more efficient ways by looking for model parameter space points consistent with both theory and experimental data. Last but not least, such tools will be built following a model-independent philosophy in order to promptly apply them to different New Physics models.