e/gamma trigger efficiencies
DESCRIPTION
E/gamma Trigger Efficiencies. Sam Harper RAL Trigger Performance Meeting 17/05/10. E/g HLT 101: Basic Trigger. L1 Trigger Accept. HLT Starts. core e/gamma triggers are a supercluster with E T > X matched to a L1 Seed L1 is a trigger tower with E T > X - PowerPoint PPT PresentationTRANSCRIPT
E/gamma Trigger Efficiencies
Sam Harper
RAL
Trigger Performance Meeting
17/05/10
2/14
E/g HLT 101: Basic Trigger• core e/gamma triggers are a
supercluster with ET > X matched to a L1 Seed– L1 is a trigger tower with ET > X
• electron triggers additionally require a PM match– note most “electron” triggers are
not electrons at HLT, they are SC + PM
– only EleId + TrackIsol triggers run electron tracking (which is different to RECO…)
• Currently L1 trigger almost completely drives the eff curve for |η| < 2.1
L1 Trigger Accept
Match SuperClus to L1 seed
ET filter (SC ET, w.r.t 0,0,0)
Pixel Matching Step
Hcal Isolation Filter
Spike Filter (E1/E9<X)
end of pho triggers
not currently enabled
HLT Starts
additional ele trig step
note: that L1 η/φ is fairly coarse
3/14
Analysis Details• data: /MinimumBias/Commissioning10-GOODCOLL-v9/RAW-RECO
– runs: 133874, 133877, 133881,133885, 133928• good lumi sec in backups
– spike cleaning: E1/E3x3 <0.9, bx = 1 or 1786
• MC: /MinBias/Spring10-START3X_V25B_v1/GEN-SIM-RAW
– RECOd in CMSSW_3_5_7, START3X_V26A
• HLT: – /cdaq/physics/firstCollisions10/v5.1/HLT_7TeV_HR/V1– using CMSSW_ 3_5_7_p2 , GR10_H_V5
• https://twiki.cern.ch/twiki/bin/viewauth/CMS/HowToInstallONLINErelease
– not the menu/release that the data were taken with but the menu the next data are being taken with
– MC: HLT_GRun in 3_5_7, START3X_V27– FYI: all menus can be found at:
• http://cms-project-confdb-hltdev.web.cern.ch/cms-project-confdb-hltdev/browser/
• online is under ORCOFF, development on HLT Development
4/14
Previous Work
• HLT Eff. to be shown to be 100% for |η| <2.1– therefore L1 completely dominates turn on curve– S. Harper: http://indico.cern.ch/conferenceDisplay.py?confId=93821
• Large amount of work done on understanding L1 Efficiency and turn on curves by LLR and Wisconsin groups in L1 DPG– J. Efrons: http://indico.cern.ch/conferenceDisplay.py?confId=94122 – C. Broutin: http://indico.cern.ch/conferenceDisplay.py?confId=92206
5/14
Efficiency Method
• simply take efficiency as number of SC passing HLT (or filter) / total number of SC– after applying spike rejection and correct bx of course
• cross-checked biasness of sample by selecting only those events which pass HLT_MinBiasBSC or HLT_MinBiasHcal– looks okay…
6/14
L1+HLT Eff Barrel: Pho10
• L1 efficiency drives the curve…• bad MC agreement for all data, good agreement with
H/E<0.05 cut is applied
all SC H/E<0.05
7/14
L1+HLT Eff Endcap: Pho10
• again better MC/data agreement once H/E cut on• different turn on curves for data/MC
– LLR group shown that L1 not efficient for SC where energy is spread over multiple trigger towers
• typical of jets, less typical of electrons
all SC H/E<0.05
8/14
L1+HLT Eff Endcap: Pho10
• Shown by LLR that L1 trigger efficiency drop is due to SC being spread out over multiple trigger towers
• sigmaIEtaIEta < 0.0275 cut improves turn on but clus = 1 is better– H/E does nothing to turn on– likewise EM isolation
9/14
Eff vs η/φ
• all plots have a SC
ET>10 GeV and H/E<0.05 cut applied
• no real surprises here
barrel
endcap
9/14
10/14
HLT Clustering
• my last talk implied this was ~100% – this is not quite true as was starting from L1 Match filter– sometimes the HLT fails to reconstruct a SC, this is almost the only way to fail the
L1 match (the matching is generous…)
• In EB HLT requires seed ET > 1.5 GeV, RECO ET > 1 GeV
• results in small inefficiency for jets
Eff to get HLT SC L1 Match given RECO SC L1 Match
11/14
Ele Trigger Eff
• Efficiency for ecal driven GsfEle electrons to pass Ele10_LW, having passed Pho10– 100% for |η|<2.1
– disagreement in MC
12/14
Ele10LW Eff vs φ
• φ dependence of MC points to overly pessimistic startup conditions– I used START3X_V27 tag, is this correct?
13/14
Ele10LW Eff vs η
• hard to tell with stat. but don’t see a clear ET dependence on PM efficiency
• need more stat to check but likely eff as a function of ET should suffice– note real ele efficiency
will be different…
14/14
Conclusions
• measured L1+HLT efficiency for Photon and Electron triggers– rate is completely driven by L1– shower shape cuts help improve
turn-on• new RCT calibrations are being
tested that will help this
– some differences with MC observed
• currently main sources of inefficiency in data are understood
H/E<0.05
H/E<0.05
|η|<1.442
1.56<|η|<2.5
Pho10 Eff
Pho10 Eff
14/14
15/14
Backups
good lumisec used:• '133874:166-133874:814', '133874:817-
133874:875', '133875:1-133875:20', '133875:22-133875:49', '133877:1-133877:1640', '133877:1643-133877:1997', '133881:1-133881:71', '133881:74-133881:223', '133881:225-133881:562', '133885:1-133885:132', '133885:134-133885:728', '133927:1-133927:57', '133928:1-133928:645',
16/14
E/g HLT 101: Glossary• ET: super cluster ET w.r.t to 0,0,0
• LW PM: large window pixel matching– concerns about beamspot stability in very early running so very loose
pixel matching windows initially– currently whats running for electron triggers
• SW PM: startup window pixel matching– slightly tighter pixel matching requirements which will be introduced
around 1E31 once we have shown that it doesn’t impact electron efficiency
• Electron Tracking:– only done for EleId + Track Isolation Trigger– CTF tracking algo not GSF– less forgiving on missing hits, not as efficient as RECO CTF tracking
• HCAL isolation: – actually more H/E, sum of rec-hits in cone of 0.15 – not currently used
17/14
SW vs LW
• first look at startup pixel match windows in data
18/14
L1+HLT Pho10 Eff: ID Cuts
σiηiη < 0.0275
σiηiη < 0.0275
H/E < 0.05
Isol EM < 3 GeV
σiηiη < 0.0275
H/E < 0.05
Isol EM < 3 GeV
Isol Had < 1 GeV
σiηiη < 0.0275
H/E < 0.05
19/14
Ele Trigger Eff with EleID
• applied HEEP ID to electrons to see if MC efficiencies are changed
20/14
C. Broutin: L1 DPG 22/04/10
21/14
C. Broutin: L1 DPG 22/04/10
22/14
J. Efron: L1 DPG 06/05/10
23/14
HLT ET Turn vs RECO SC ET
• Eff of ET filter w.r.t L1 Match– this should be close to a step function
• for SC with ET > 10 GeV
barrel endcap