Performance of the pile up jet identification in CMS for Run 2

The excellent operation of the LHC during the Run 2 data taking (2015-2018) has allowed the machine to reach a peak instantaneous luminosity delivered to ATLAS and CMS as high as $2\times10^{34}~\mathrm{Hz~cm^{-2}}$, leading to a large number of proton-proton inelastic scatterings to occur during the same bunch crossing (``pile up''). For the 2017 and 2018 datasets, this results in an average pile up of 32. A major impact of the pile up is to generate extra jets in the detector, in addition to those associated to the hard scattering under study. Various mitigation techniques exist to identify and reject such pile up jets. The most commonly used algorithm is the ``Charged Hadron Subtraction'' (CHS), which discards all charged constituents associated to a vertex other than the leading vertex before starting the jet clustering procedure. The resulting jet energy is then corrected for the average energy associated to neutral particles from pile up. This procedure ensures an almost 100\% efficiency for hard scatter jets. In order to further reject pile up jets, a pile up identification (PU ID) discriminator is built using a boosted decision tree (BDT), trained with 12-15 input variables characterizing the jet and the event. This note presents the performance of the discriminator that has been trained on simulated samples representing the 2016, 2017 and 2018 detector geometry and data taking conditions. It highlights the improvement brought by the Phase 1 upgrade pixel detector installed in the beginning of 2017 and that allowed to extend the tracking coverage from $\lvert\eta\lvert\approx$ 2.5 to $\lvert\eta\lvert\approx$ 2.7. The note finally illustrates the importance of such a PU ID even for jets inside the tracking coverage by studying the central jet multiplicity in events containing a leptonically decaying Z boson.