Tau reconstruction at CMS with a focus on high p T taus

We present the algorithm and performance of tau reconstruction at the CMS experiment, while highlighting a dedicated reconstruction algorithm that uses calorimeter hits instead of tracks to reconstruct taus with high transverse momentum. Describing the standard Hadrons-Plus-Strips (HPS) algorithm and its dependence on track reconstruction and shower modelling, we present the calorimetric tau (calo-tau) reconstruction that uses minimal track information for high pT taus. The pros and cons of these algorithms are discussed along with their performance and potential uses. It is found that the calo-tau algorithm outperforms the HPS algorithm in the high efficiency region. This study is work in progress, and is an attempt to tune the reconstruction for high pT taus. The calo-tau algorithm is not yet an official tau reconstruction algorithm for CMS. Copyright S. Bhattacharya. This work is licensed under the Creative Commons Attribution 4.0 International License. Published by the SciPost Foundation. Received 22-11-2018 Accepted 17-01-2019 Published 22-02-2019 Check for updates doi:10.21468/SciPostPhysProc.1.050


46
The Hadrons-Plus-Strips (HPS) algorithm is the default algorithm used for τ h reconstruction at the CMS experiment. The basic steps of the algorithm are as follows [1]:

48
• The HPS algorithm is seeded by anti-k T jets with a distance parameter of 0.4 (AK4 49 jets).

50
• The electron and photon constituents in the jet are clustered into "strips" which try 51 to capture the neutral pion decay.

54
• The algorithm forms the following τ h candidates (corresponding to the τ h decay 55 modes shown in (Table 1)). The correlation between generated and reconstructed 56 τ h decay modes is shown in Figure 1 (right).  h ± : A single charged hadron candidate without any strips.

58
h ± π 0 : Combination of one charged hadron and one strip.

60
h ± h ± h ± : Combination of three charged hadrons without any strips. strips, helps to reduce the misidentification probability [1].

79
A BDT (Boosted Decision Tree) is also trained to discriminate τ h jets from QCD jets.

80
Its relative performance w.r.t. the isolation-based discrimination is shown in Figure 2.  The calorimetric tau (calo-tau) algorithm 86 In the previous section we saw that that HPS algorithm relies heavily on track resolution, 87 track momentum measurement, and electron, photon, and charged hadron reconstruction.

88
The calo-tau reconstruction algorithm's robustness lies in its simplicity. The algorithm 89 has been constructed in way such that its dependence on track momentum measurement 90 in minimized, and unlike the HPS-tau algorithm, electron/photon/charged hadron recon-91 struction plays no role here. The main steps are as follows.

92
• Seed the algorithm with a calorimeter jet (calo-jet) reconstructed with the anti-k T 93 algorithm with a distance parameter of 0.4 (AK4 calo-jet).

94
• The existence of a track with the following condition is required to select the jet as a Similar to HPS-taus, we define the following isolation-sum.

108
• ρ is the energy density in the event, and is defined as the median of the calo-jet 109 energies divided by their respective jet-areas.

110
• A eff is an effective area whose value (0.2) is chosen such that the efficiency is inde-111 pendent of pileup.

112
• The product ρA eff is the contribution from pileup to the ECAL energy deposits.

113
The performance of the different working points of the isolation-based discriminant 114 (as a function of p T and pileup conditions) in Z (2 TeV) → τ τ events is shown in Figure   115 4. Both the efficiency and misidentification probability are flat across a wide range of p T .

116
The effect of pileup has also been minimized, as can be seen from flatness of the efficiency 117 and misidentification probability across the number of vertices.  isolation-based discriminant, as expected. However, the interesting feature is that the calo-148 tau algorithm is able to reach higher efficiencies (> 70 %) than the HPS-tau algorithm 149 and performs better in that region .  [3] The CMS Collaboration, Performance of τ -lepton reconstruction and identification in