Finding top quarks with shower deconstruction

TL;DR

Finding top quarks with shower deconstruction introduces a likelihood-ratio based jet tagging framework that discriminates hadronically decaying top jets from QCD backgrounds. It extends the shower deconstruction approach to the hadronic top decay chain, modeling the decay sequence and soft/collinear radiation through a simplified shower history and comparing performance to public taggers. The method delivers superior discrimination for both moderately boosted and lower-pt tops, with robustness to jet cone size and modest generator dependence, and enables extracting signal-model parameters such as the W mass. These results suggest a practical, high-precision tool for top-quark tagging with potential extensions to event-level analyses and new-physics searches.

Abstract

We develop a new method for tagging jets produced by hadronically decaying top quarks. The method is an application of shower deconstruction, a maximum information approach that was previously applied to identifying jets produced by Higgs bosons that decay to bottom quarks. We tag an observed jet as a top jet based on a cut on a calculated variable that is an approximation to the ratio of the likelihood that a top jet would have the structure of the observed jet to the likelihood that a non-top QCD jet would have this structure. We find that the shower deconstruction based tagger can perform better in discriminating boosted top quark jets from QCD jets than other publicly available tagging algorithms.

Figures (8)

• Figure 1: Shower history for a top quark jet. The hard interaction is indicated by a star. Initial state emissions are indicated by diamonds. Parton decays are indicated by large filled circles and QCD splittings are indicated by small filled circles.
• Figure 2: Shower history for a QCD jet.
• Figure 3: $(1/N)\, dN/ d\log\chi$ for signal events (upper curve) and $(1/N)\, dN/ d\log\chi$ for background events (lower curve) for samples of signal and background events generated by Pythia. We use the cuts described in Sec. \ref{['sec:EventSelection']}.
• Figure 4: Background fake rate $F$ as a function of signal acceptance $A$ for shower deconstruction with the signal and background event samples described in Sec. \ref{['sec:EventSelection']}. The curve for shower deconstruction is compared to $F$ vs $A$ points for the Johns Hopkins top tagger (JH), the top tagger of the CMS group (CMS), the Heidelberg-Eugene-Paris top tagger (HEP), and the use of N-subjettiness as a top tagger (NSUB). We show the results on a linear scale (left) and on a logarithmic scale (right).
• Figure 5: Results using Herwig++ compared to those using Pythia from Fig. \ref{['fig:acceptanceL']}. The solid curve is the $F$ versus $A$ curve from shower deconstruction using events generated with Pythia; the dashed curve uses events generated with Herwig++. The solid circles show $F$ versus $A$ results for the top taggers using events generated with Pythia; the open squares use events generated with Herwig++.