Hadron-level quark/gluon tagging for ATLAS

The ATLAS experiment at the LHC is a general purpose particle physics detector used to study pp collisions at √s = 13TeV. The two main types of particles produced in the initial hard scatter are quarks and gluons, collectively called “partons”. Since partons cannot exist in an isolated state, they will quickly decay into stable hadrons in the showering and hadronization process. This results collimated sprays of hadrons known as jets. One of the challenges faced by the ATLAS collaboration is to understand the type of particle which initiated a jet, given its signature in the trackers and calorimeters. A jet in ATLAS simulations has multiple levels. A jet is called “parton-level” before it hadronizes and “hadron-level” after it hadronizes. The jet is then simulated as it enters the ATLAS calorimeters, and the energy depositions are grouped into a “reco-level” jet. We are interested in the detector response, or jet energy scale (JES), which is defined as the ratio of reco jet transverse momentum to truth jet transverse momentum pTreco/pTtruth. This quantity measures the fraction of the energy of a jet which is captured in the calorimeters. There have been many previous efforts to understand the JES [1, 2]. It has been observed that jets initiated by different partons have different average responses. This is known as the JES flavour uncertainty. To properly correct for this effect, it is useful to be able to identify each jet with the parton which initiated it. At hadron-level, jets initiated by gluons and by light quarks (u, d, s) can be distinguished only by properties such as jet width and number of constituent hadrons; quark jets are typically narrower and have fewer constituents than gluon jets. Identifying quark and gluon jets based on their properties at hadron-level is the topic for my project. Jets are currently labelled by the highest pT parton in their volume at parton-level. However, this is not ideal for a number of reasons. Partons are not stable, and so their properties are unphysical. There are a number of different Monte Carlo (MC) generators which are used to simulate the parton showering and hadronization, and the way they handle parton-level information can be very different. Relying on unphysical parton-level information will introduce a large dependence on the MC generator used. The goal of this project was to classify jets based on physical hadron-level topology, without using unphysical parton-level information. This classifier should take jet properties at hadron-level as input, and should output a value from −1 to +1 depending on how “quark-like” or “gluon-like” the hadron jet is. The relationship between the classifier and the JES was then studied. This project is described in further detail in a full length report [3], which includes more technical information and results, and additional studies performed for this project.