Nutzung multivariater Analysemethoden zur Optimierung von Top-Tagging-Algorithmen

Author: 
Torben Dreyer
Date: 
Mar 2013

Thesis Type:

Many extensions of the Standard Model of particle physics predict heavy particles, which decay into top-quarks. Top-quarks produced by such decays have high transverse momentum and are therefore boosted in the direction of their flight. As a result, all the decay products of the top-quark are collimated and may be reconstructed in one big jet. This poses a big challenge especially for the hadronic decay channel of top-quarks.
Top-tagging algorithms use substructure information of jets in order to separate top-decays from the overwhelming QCD-background in proton-proton collisions at the LHC experiments.
In this thesis multivariate methods and the variables Qjets volatility and N-subjettiness are employed to study the possibility to improve the CMS-tagger and HEPTopTagger. For this study, simulated events from top-quark pair production and light QCD jet production in pp-collisions are used.
It is found, that for similar signal efficiency for the CMS-tagger the background efficiency can be reduced from 2.68% $\pm$ 0.02% for the CMS-tagger to 1.44% $\pm$ 0.01%. In case of the HEPTopTagger it’s possible to reduce the backgroundefficiency from 5.32% $\pm$ 0.01% to 1.86% $\pm$ 0.01%.