Wing Sheung Chan

50 Object reconstruction and identification 3.2.2. Substructure reconstruction Even though it is simple and robust, the baseline reconstruction is omitting a lot of valuable information about the τ decay that could be used to improve analyses. To recover that information, a more elaborated reconstruction method is used to complement the baseline reconstruction. This method attempts to reconstruct the individual τ decay products instead of treating all of them as a collective object [79] . We will refer to this approach as the substructure reconstruction. A τ had - vis that is reconstructed with this approach is sometimes called a “pantau” within the ATLAS collaboration. Similar to the baseline reconstruction, the substructure reconstruction is also seeded by jets. However, the momentum of τ had - vis is not simply calculated by adding the energies and directions of all the clusters in the reconstructed jet cone. Instead, it is calculated as the sum of individually reconstructed charged hadrons ( h ± , predominantly pions π ± ) and neutral pions ( π 0 ). The major challenge of this approach is to disentangle the energy deposits of the h ± ’s and π 0 ’s. To achieve that, the hadrons are sequentially reconstructed using an algorithm called Tau Particle Flow. The main idea of Tau Particle Flow is to first make use of the inner detector tracks to reconstruct the h ± ’s, and then use that information to approximate the h ± energy deposits in the EM calorimeter, so that they can be disentangled from the π 0 energy deposits. In Tau Particle Flow, h ± ’s are first identified using the associated tracks. For τ had - vis that has an energy around or below 100 GeV , the h ± ’s can be correctly identified by the tracks about 98% of the time. Then for each h ± , both the direction and magnitude of the momentum can be reconstructed using the track. Since π 0 showers (dominantly 2 γ ) very rarely extend beyond the EMCal, the energy deposits in the HCal that matches the direction of the h ± track can be fully assigned to the h ± . By subtracting the energy deposits in the HCal from the energy calculated using the track, one can then estimate the amount of h ± energy deposits in the EMCal. Then, π 0 candidates are reconstructed using clusters in the EMCal. If the π 0 clusters are overlapping with a reconstructed h ± shower, the estimated energy deposits from the h ± are subtracted from the clusters. This systematically disentangles the energy deposits of the h ± ’s and π 0 and avoids double counting. After that, a BDT classifier is used to identify real π 0 ’s from the π 0 candidates by exploiting the relatively regular shape and size of π 0 showers. At last, after the h ± ’s and π 0 ’s are reconstructed, the full τ had - vis can then be recon- structed by treating the h ± ’s and π 0 ’s as constituents and summing their momenta. The calculated momentum defines a TES known as the “pantau TES”. By incorporating information from the inner detector, the substructure reconstruction is able to determine the direction of τ had - vis more precisely than the baseline reconstruction. For low- p T τ had - vis ( p T . 100 GeV ), the momentum resolution is also superior. Even more importantly, it allows one to determine the actual decay mode of τ had - vis . A BDT classification algorithm has been developed to achieve that by considering the number of reconstructed h ± ’s and π 0 ’s and correcting for potential misreconstructions of π 0 ’s.