The dependency of boosted tagging algorithms on the event colour structure

TL;DR

This paper investigates how the colour structure of events affects the performance of various top-tagging algorithms, highlighting the importance of optimizing parameters to reduce model dependency in searches.

Contribution

It provides a detailed analysis of the influence of event colour charge on top-tagging efficiency and offers recommendations for parameter optimization in experimental searches.

Findings

Significant differences in top-tagging efficiency based on colour charge.

Quantification of efficiency variations as a function of algorithm parameters.

Recommendation to optimize parameters to reduce model dependency.

Abstract

The impact of event colour structure on the performance of the Johns-Hopkins, CMS, HEPToptagger and N-Subjettiness algorithms is investigated by studying colour singlet and colour octet resonances decaying to top-quark pairs. Large differences in top-tagging efficiency are observed due to the different colour charge of each resonance. These differences are quantified as a function of the algorithm parameters, the jet size parameter and the probability to misidentify light quarks and gluons as top candidates. We suggest that future experimental searches would benefit from optimising the choice of algorithm parameters in order to minimise this source of model dependency.