Wing Sheung Chan

Object reconstruction and identification 49 to automatically take into account the large number of varying factors. During the Run-2 data-taking period, significant improvement in the reconstruction and identification per- formance has been made by introducing machine learning. In particular, the author has significantly contributed to the development of an MVA-based τ had - vis energy calibration algorithm and an electron rejection algorithm, which are both widely used in the collabo- ration currently. Such developments are valuable to many analyses that consider events with τ had - vis in the final states, including, for example, the H → τ τ measurement. 3.2.1. Baseline reconstruction There are two different approaches for reconstructing a τ had - vis . The simpler of the two uses mainly information from the calorimeters and considers all the visible decay products within a circular cone as one single collective object [77, 78] . We will refer to this approach as the baseline reconstruction. As hadrons, τ had - vis can almost always be reconstructed as a hadronic jet. Therefore, the natural starting point of the baseline reconstruction is the jets reconstructed by the method mentioned in Section 3.1.1. All reconstructed hadronic jets are initially considered as possible candidates of τ had - vis . Tracks measured in the inner detector are associated to the τ had - vis candidates if they have matching directions. The direction of a τ had - vis candidate is defined by the barycentre of the clusters in the jet, calibrated with the LC scheme and assumed to have zero rest mass. Tracks that are in the “core region” ∆ R < 0 . 2 around the τ had - vis direction are the associated tracks. Since a τ had - vis consists of one or three charged particles over 99.9% of the time, it can be expected that jets originated from τ had - vis have exactly one or three associated tracks. τ had - vis candidates are classified into “1-prong” or “3-prong” depending on the number of associated tracks. Tracks that are in the “isolation region” 0 . 2 < ∆ R < 0 . 4 are also considered, but only for identification against quark- or gluon-initiated jets, which will be detailed later. The momentum of a τ had - vis candidate is first calculated by summing the four-momenta of calorimeter clusters in the core region. Since the clusters are calibrated with the LC scheme, the τ had - vis momentum obtained at this stage is referred to as the momentum at LC scale ( p LC ). However, the LC scheme is neither optimised for the cone size ∆ R = 0 . 2 , nor the expected hadronic composition of τ had - vis . This is improved by applying a correction derived using simulations. The correction is a function of the transverse momentum and pseudorapidity at LC scale. The corrected momentum defines a tau energy scale (TES) known as the “calo TES”. The above procedures are the baseline reconstruction for τ had - vis . It is a relatively simple yet robust method of τ had - vis reconstruction. It has an advantage of being relatively easy to model due to the fact that it is only based on averaged properties of calorimeter clusters, and hence does not rely heavily on detailed simulations of jet substructures. It also uses computational resources efficiently by capitalising on the regular jet reconstruction.