Performance of boosted top quark identification in 2012 ATLAS data

This note presents the performance of a variety of techniques used to identify highly-boosted top quarks. The studies presented here use the full 2012 ATLAS dataset taken at a center of mass energy of $\sqrt{s}$ =8 TeV, corresponding to an integrated luminosity of (20.3$\pm$0.6) fb$^{-1}$ of proton-proton collisions produced by the Large Hadron Collider. A sample enriched in $t\bar{t}\rightarrow(Wb)(Wb)\rightarrow(qqb)(\mu\nu b)$ events is used for the study. Following the jet trimming procedure, large-R jet substructure properties are shown to be well described by the simulation. Additionally, the distribution of subjets from the trimming process is used to characterize events in which the hadronic daughter particles of a boosted top quark are fully contained within the distance parameter of a large-R jet. Performance of top quark mass reconstruction using the HEPTopTagger algorithm is also shown and demonstrates an insensitivity to additional proton-proton interactions per event. The distributions of the tagger internal substructure variables and the tagged jet mass are well described by the simulation. Various working points of different tagging approaches are compared in simulation between a signal of large-R jets containing highly boosted top quarks and a background of large-R jets originating from hard light quarks or gluons. In addition to the HEPTopTagger, which gives a high rejection of background jets and a medium top quark jet tagging efficiency of 30-40% depending on the algorithm parameters, several taggers that are based on selecting on substructure variables are studied and shown to yield higher efficiencies at reduced rejection.