Introduction

• PLImprovedTight and PLImprovedVeryTight are two new working points based on the PromptLeptonImprovedVeto algorithm.
• PromptLeptonImprovedVeto is the improved version of the prompt lepton tagging algorithm developed in 2019-2020 by Fudong He and Rustem Ospanov.
  • Hanlin Xu has helped with developing the electron version.
  • The previous versions of these algorithms are called PromptLeptonVeto and PromptLeptonIso and described in PromptLeptonTagging.

Presentations

• Introduction of PromptLeptonImprovedVeto algorithm (only showing muon results): ATLAS weekly, December 10, 2019
• Updated with electron results for PromptLeptonImprovedVeto: IFF, March 9, 2020
• Introduction to PromptLeptonImprovedVeto muon working points: MCP October 7, 2020
• Focus on muon working points and performance checks in data: IFF, October 19, 2020
• Focus on electron working points and performance checks in data: Egamma, November 18, 2020

PromptLeptonImprovedVeto

• Several new techniques were developed in order to identify and reject the non-prompt electrons and muons candidates from decays of heavy flavour (b and c) hadrons.
• We designed novel isolation variables that are sensitive to the properties of heavy flavour jets.
• We developed a dedicated secondary vertex reconstruction algorithm to identify the non-prompt leptons produced in the b- and c-hadron decays. Reconstructed secondary vertices are used to estimate flight path of non-prompt leptons.
• We designed and optimised a recurrent neural network (RNN) using inner detector track impact parameters.
  • Electron RNN was trained to separate prompt electrons from non-prompt electrons and from fake electrons produced in material photon conversions.
  • Muon RNN was trained to separate prompt muons from non-prompt muons.
  • Top quark pair Monte-Carlo simulation samples were used for prompt and non-prompt lepton samples. Z boson with photon production was used for electron fakes from photon conversions.
• Boosted Decision Tree algorithm was carefully designed and optimised to combine the isolation parameters, lifetime information, and the RNN output.
  • Three separate BDTs were trained: one for muons, one for barrel region (|η|<1.37) and another endcap region (|η|>=1.37) electrons.
  • The choice of the barrel or endcap BDT version for electrons is done internally by the isolation tool.
  • Top quark pair Monte-Carlo simulation samples were used for prompt and non-prompt lepton samples.
  • BDTs were trained to separate prompt leptons from non-prompt leptons.
• The input variables are summarised in the table below:
  

Working points

• PLImprovedTight and PLImprovedVeryTight selection criteria were optimised as a function of the lepton pT in order to provide better non-prompt lepton rejection than the tightest previously available isolation working point, while providing similar prompt lepton efficiency.
• PLImprovedTight was optimised to provide of about a factor of two better rejection against non-prompt leptons than the PLVTight working point with similar prompt efficiency.
  • The PLVTight working point was derived with the previous version of our algorithm called PromptLeptonVeto.
• PLImprovedVeryTight was optimised to provide even stronger rejection of non-prompt leptons at the price of reduced prompt lepton efficiency.
• The working points were then tuned to provide the continuous efficiency as function of prompt lepton pT.
• The optimisation procedure was presented in the IFF meeting on October 19th, 2020.
  • Each working point was chosen from thousands of candidates in order to give a best overall significance for several benchmark analyses.
• Working point definitions are documented in git: PLIVCutValue.
• Working point efficiency for prompt and non-prompt leptons is documented in git: PLImprovedWPs

Required DxAOD variables

• The required discriminant variable for muons is "PromptLeptonImprovedVeto".
• The required discriminant variable for barrel region electrons is "PromptLeptonImprovedVetoBARR" and for endcap region electrons is "PromptLeptonImprovedVetoECAP".
• The pT bin variable called "PromptLeptonImprovedInput_MVAXBin" is also necessary for applying correct pT-dependent cuts. This is because PromptLeptonImprovedVeto distributions are slightly different in different pT bins.
• These variables are calculated at the derivation level by the LeptonTagger package.
• Shown below are example configuration lines for adding the required variables to release 21.2 derivations:

import LeptonTaggers.LeptonTaggersConfig as LepTagConfig

# Add the PromptLeptonImprovedVeto algorithm to AthSequencer
MUON5Seq = CfgMgr.AthSequencer("MUON5Sequence")
MUON5Seq += LepTagConfig.GetDecorateImprovedPromptLeptonAlgs()

# Get the auxiliary variable list 
ExtraVariables += LepTagConfig.GetExtraImprovedPromptVariablesForDxAOD() # including all input variables and discriminant variable of the PromptLeptonImprovedVeto.
# ExtraVariables += LepTagConfig.GetExtraImprovedPromptVariablesForDxAOD(onlyBDT=True)  # only get the discriminant variable  and pT bin variable of electron and muon.

Working points usage

• PLImprovedTight and PLImprovedVeryTight working points are defined in IsolationSelectionTool with AnalysisBase/AthAnalysis release 21.2.156 or later.
  • Merge requests: add new working points update working point names
• The working points can be configured using the following example python file:

isoTool = CfgMgr.CP__IsolationSelectionTool('IsolationSelectionTool')
isoTool.MuonWP = "PLImprovedTight" # or "PLImprovedVeryTight"
isoTool.ElectronWP = "PLImprovedTight" # or "PLImprovedVeryTight"
ToolSvc += isoTool
isoLowPtPLVTool = CfgMgr.CP__IsolationLowPtPLVTool( 'IsolationLowPtPLVTool')
athTool = CfgMgr.Ath__IsoTool('Ath_IsoTool')
athTool.isoTool = isoTool
athTool.isoLowPtPLVTool = isoLowPtPLVTool
athTool.doPLV = False
athTool.varName = 'isoPLImprovedTight' # or  "isoPLImprovedVeryTight"
ToolSvc += athTool

Working point calibrations

• Calibrations of electron and muon working points are ongoing
  • Initial calibrations for muons are available, and currently being checked and validated: MCP presentation on January 13th, 2021
  • Electron calibrations are also underway, currently planning for results in February or March of 2021