Development of the leptonic production tagger in the diphoton decay mode

Abstract

The future detection capacity of Higgs production in tHq mode was studied using a standard preselection already used in previous analyzes within the context of the Compact Muon Solenoid (CMS) collaboration on CERN. In particular, a neural network made up of Long Short-Term Memory (LSTM) layers was used, which is generated using the TensorFlow analysis package. The Toolkit for Multivariate Data Analysis (TMVA) package was also used to train boosted decision trees (BDT). The resulting predictive methods were used to suppress the resonant ttH and nonresonant noise by 35.30% and 22.34 %, respectively. Using the expected reconstructed Higgs mass distribution by the standard model (SM), exclusion upper limits were set on the signal strength () of 10.2 and 21.8 at 68% and 95% confidence (C.L.) at a luminosity of 150.2 fb-1. Additionally, these limits are reduced by 94.27% by increasing the luminosity to 4000 fb-1 consistent with what is expected for the high luminosity stage of the LHC (HL-LHC). In the end, a 90% noise reduction scenario was analyzed and it was obtained that the limit at 2 is reduced by 96.78% and the lower limit reaches an exclusion level of 2 for the case  = 0 at a luminosity of 4000 fb-1.