Mr Joshuha Thomas-Wilsker

Supervised by

Personal profile

I am a postdoctoral research assistant at Royal Holloway, University of London where my contract will last until 30th September 2016. During the first year of my PhD I worked with the ATLAS flavour tagging group to obtain my ATLAS authorship. I calibrated one of the ATLAS b-tagging algorithms using the “tt dilepton kinematic selection” method which targets tt dilepton events to obtain a very pure, inclusive sample of b-jets in real data. Selecting b-jets inclusively prevents any potential bias being introduced, as can be the case when using b- jets selected according to the decay of the b hadron. I furthered this study by investigating if such a potential bias exists by performing the same cali- bration on two samples of jets: those which contain a muon and those which do not. More on this work can be found in the internal notes listed below.

My current research is on the search for the stan- dard model Higgs boson produced in association with a pair of top-quarks and decaying to a bb-pair; “ttH(bb)”. This search channel is of particular importance as it is used to measure the top Yukawa coupling. Furthermore, the bbdecay mode has the largest branching fraction for the 125 GeV Higgs boson and therefore contributes a large proportion of the statistics in the context of the wider ttH search. The analysis is divided into two channels according to the decay of the W bosons in the ttH(bb) system: the so called “single lepton” channel where one W boson decays hadronically and the other leptonically and the “dilepton” analysis where both W bosons decay leptonically. The single lepton analysis provides most of the statistics due to the large branching fraction of the W bosons to leptons and jets whereas the dilepton analysis provides a much cleaner signal due to the two high-pT leptons. The focus of my work has been on the dilepton analysis. For the 2015 internal note and paper (now published in the European Physical Journal C) I evaluated the systematic uncertainty associated with the choice of parton shower hadronisation model for the ttHsignal sample and helped with the optimisation of the analysis selection.

A large part of my PhD thesis investigated the use of multivariate classifiers and introduc- ing new variables in future ttH(bb) analyses. I compared the performance of boosted decision trees (BDT) with neural networks (NN) when given the same information. I also investigated additional/alternative MVA input variables with better separation power and studied ways of improving the matching between the reconstructed objects and their parton-level counterparts to obtain a more reliable description of the selected events. I have used this experience to good effect in the ttH(bb) 2016 ICHEP analysis for which I trained and optimised a neural network used to discriminate between ttH(bb) dilepton signal and background events.

One of the most interesting parts of the ttH(bb) analysis is the binned likelihood fit used to obtain the final results. In the analysis, a binned likelihood function is used which is dependent on the signal strength parameter (the parameter of interest) and a collection of nuisance parameters (priors that incorporate the effect of systematic uncertainties on signal and background predictions). The nuisance parameters in the fit act to adjust the expected amount of signal and background. Their final fitted values equate to the amount that best fits data. By fitting simultaneously in both control and signal regions, systematic uncertainties can be reduced in the signal regions by taking advantage of the more highly populated control regions. The likelihood is then used to calculate a test statistic which evaluates the probability of obtaining a result as signal like as the one observed, under the background-only hypothesis. I have taken on a leading role in this area performing numerous studies using the dilepton fit. I have also worked actively in the combination team helping to ensure orthogonality between the various ttH channels as well as providing the most up-to-date dilepton analysis models for the ttH combination ICHEP 2016 result.

ID: 3909164