Building a better boosted top tagger

Distinguishing hadronically decaying boosted top quarks from massive QCD jets is an important challenge at the Large Hadron Collider. In this paper we use the power-counting method to study jet substructure observables designed for top tagging, and gain insight into their performance. We introduce a powerful new family of discriminants formed from the energy correlation functions which outperform the widely used $N$-subjettiness. These observables take a highly nontrivial form, demonstrating the importance of a systematic approach to their construction.

Figure 1

Three-prong jet configurations (a) triple splitting, (b) strongly ordered splitting, and (c) soft emission, required to understand the $(e_2^{(\alpha )},e_3^{(\beta )},e_4^{(\gamma )})$ phase space. The structure of dominant energy flow for each configuration is denoted by solid blue lines, as well as soft wide angle radiation (green), and radiation from between the dipoles (orange, light blue).

Figure 2

Phase space defined by the energy correlation functions, with contours of $D_3^{(2,\beta ,\gamma )}$, and showing the effect of a jet mass cut.

Figure 3

Signal and background distributions for $D_3$ for the pythia8 samples. Here, the parameters of $D_3$ are $\alpha =2,\beta =0.8,\gamma =0.6$, and $x=5$, $y=0.35$.

Figure 4

Signal vs background efficiency curves comparing $C_3^{(1)}$, $D_3^{(2,0.8,0.6)}$, and ${\tau }_{3,2}^{(1)}$ from pythia8 (left) and herwig++ (right) samples.