PhD Studentship: Machine Learning Techniques and High Performance Computing for Higgs Physics

A Higgs boson was discovered in July 2012 at the CERN Large Hadron Collider (LHC) during Run 1, which led to the 2013 Nobel Prize in Physics. In May 2015, the LHC started colliding beams at Run 2. The new stage, at higher energy and luminosity, will enable precision measurements of the properties of the discovered Higgs boson as well afford one with excellent discovery prospects of additional Higgs particles predicted by several new physics models.

A light Higgs boson, the Standard (SM) one or else one belonging to some Beyond the SM (BSM) scenario, decays to b-quark pairs most of the times. However, the current description of strong interactions, Quantum Chromo-Dynamics (QCD), predicts 'confinement': that is, particles carrying a colour charge, such as b-quarks, cannot exist in free form. Rather, they fragment into colourless hadrons before they can be indirectly detected as 'jets'. 

Jets have long been studied and a very good understanding of their dynamics has been achieved. Nonetheless, the advent of the LHC calls for gaining a much deeper insight into their behavior, in view of the fact that the ever larger energy available therein produces jets over new kinematic ranges. 

Specifically, for the case of b-jets originating from light Higgs boson decays, these tend to be highly boosted, therefore merging into 'fat' structures that ought to be recognised and resolved into their constituents in order to access the Higgs boson  properties. Even the exploitation of the fact that b-quarks have a finite lifetime (unlike lighter quarks or gluons), hence that the hadrons they produce eventually decay away from the interaction point (via displaced vertices), thereby rendering b-jets in principle distinguishable from other jets, requires re-assessment in the new kinematic regime. 

The project seeks to clarify the dynamics of such b-jets above and beyond current knowledge, by exploiting advanced computational models of multi-particle interactions relying upon (e.g., Markov chain) Monte Carlo event generation for the fragmentation and hadronisation process combined with very advanced QCD predictions for the hard scattering and fragmentation of multi-b-quark final states, in order to closely mimic the actual conditions existing at the LHC. Novel jet reconstruction algorithms will have to be developed. 

The improved jet algorithms to be used in LHC analysis (within the CMS experiment) rely on constructing suitable discriminators, typically using either a Neural Net (NN) or Boosted Decision Tree (BDT),  e.g., through the Toolkit for Multi-Variate Analysis (TMVA) within CERN ROOT. In fact, consideration to scikit-learn tools will also be given. Indeed, the development of totally novel approaches, thus going well beyond simply applying existing algorithms to new problems, will be within the remit of this project. As a result, a significant step change with respect to ongoing studies of jet dynamics will occur and this will be facilitated by both novel machine learning techniques and very advanced high performance computing environments.

If you wish to discuss any details of the project informally, please contact Prof Stefano Moretti, Southampton High Energy Physics (SHEP) research group, Email: stefano@soton.ac.uk, Tel: +44 (0) 2380596829.

This project is run through participation in the EPSRC Centre for Doctoral Training in Next Generation Computational Modelling (http://ngcm.soton.ac.uk). For details of our 4 Year PhD programme, please see http://www.findaphd.com/search/PhDDetails.aspx?CAID=331&LID=2652 

For a details of available projects click here http://www.ngcm.soton.ac.uk/projects/index.html