update on h zz 2l2n analysisindiacms/indiacms-meetings/august-2012/arun-kumar... · 2 arun kumar,...
TRANSCRIPT
Update on HZZ2l2n Analysis
Arun Kumar, Kirti Ranjan, Ashok Kumar University of Delhi
India-CMS Meeting, 3-4 August,2012 BARC (Mumbai)
04/08/2012 Arun Kumar, University of Delhi 1
Introduction
ICHEP Update
Further improvements in Analysis
Summary and Outlook
Outline
04/08/2012 Arun Kumar, University of Delhi 2
Motivation and Introduction
Motivation:
SM Higgs is already excluded in mass region (127.5 – 600 GeV) but we still need to
understand processes involving scattering of vector bosons.
The most sensitive analysis at high mass region, also in region which has not been
excluded (>600 GeV).
It has 6 times higher branching ratio than HZZ4l channel.
Signatures: Signature is given by two high PT leptons of same flavor with di-lepton mass close to Z
peak and large MET.
Both Glu-Glu and VBF production modes considered
Full reconstruction of Higgs mass not possible.
Backgrounds: Z+Jets (Most Dominant, difficult to model in MC) TTJets (2nd Dominant) ZZ,WZ,WW (Irreducible ZZ) Single T & W+Jets (Minimal Contribution) QCD
04/08/2012 Arun Kumar, University of Delhi 3
Previous Results and Improvements
2011 Analysis:
Results: exclusion in range 310 – 465 GeV using cut & count and 270 – 440 GeV
using mass shape analysis
Published in JHEP 1203 (2012) 040
New Features of Analysis:
Analysis of 2012 dataset
Reanalyzing 2011 datasets
Updated physics objects
VBF selection
Jet binning
Extension of the analysis to region 200-250GeV ( good sensitivity from VBF category)
04/08/2012 Arun Kumar, University of Delhi 4
ICHEP Update Analysis Strategy:
Analysis is optimized separately for Gluon-Gluon and VBF categories.
Selection can be divided into two parts : Preselection + Final Selection
Pre-selection:
Events are selected with 2 same flavored, well identified and isolated leptons with di-
Lepton mass |Mll - MZ | < 15 GeV
Third Lepton Veto: to suppress WZ background.
Min D(jet,PFMET) < 0.5 reduce the instrumental background from jet mis-measurements.
PT (Z) > 55 GeV For background control sample
Anti b-Tagging
Soft Muon Veto
Final Selection:
PFMET with Type1 corrections is used. Mass dependent cut for non-VBF category.
Mass dependent two sided Transverse Mass of Higgs cut is applied for non-VBF categories.
PAS: HIG-12-023 AN: AN-12-138
To suppress TTbar, Single T
We categorize all events based on the number of jets (PT >30 GeV) in event.
04/08/2012 Arun Kumar, University of Delhi 5
VBF Category
An Event is VBF Tagged if it has:
2 non-b-tagged jets with PT > 30 GeV
|Dh| > 4.0 between jets
di-jet mass Mjj > 500 GeV
di-lepton between jets
Central Jet Veto
Including VBF improves sensitivity greatly at low mass
Overall increase in sensitivity 2% – 50 %
Final Selection: PFMET > 70 GeV
04/08/2012 Arun Kumar, University of Delhi 6
Background Estimation
Z+Jets (instrumental background with fake missing energy) is estimated from
data using g+jets events.
Non Resonant backgrounds (Top/WW/W+Jets, have real MET) are estimated
from data using em events.
Irreducible electroweak ZZ background and fully leptonic WZ decays are
estimated from Monte Carlo.
04/08/2012 Arun Kumar, University of Delhi 7
Z+Jets Background Estimation
7TeV
Non-VBF
Non-VBF
8TeV
g+jets events are used to model Z+Jets from data.
g+jets has similar kinmeatic distributions as
Z+Jets
Larger Statistics
Photons are selected using V + g photon selection
Procedure PT distribution of g’s reweighted to match
to Z’s Number of primary vertex distribution in
g’s is reweighted to match to Z’s Each photon is assigned a mass by
sampling Z line shape from data done separately for each event category
Final estimate is taken as half of the prediction from photon events with 100% systematic error on prediction
04/08/2012 Arun Kumar, University of Delhi 8
Non-Resonant Background Estimation
Estimation of non-resonant backgrounds is performed with standard technique, using em events passing analysis selection (Nem
SIG )
Using sidebands (55 < mll < 70 and 110 < mll < 200 GeV) of the Z peak a scale factor a is determined from ratio of ee/mm and em events passing MET > 70 GeV and requiring at least one b-tagged jet.
Predicted number of background events is given then as
This method treats contribution from HWW*2l2n as background and properly accounts on it.
7TeV+8TeV
04/08/2012 Arun Kumar, University of Delhi 9
Results: Combined Categories Exclusion Limit, R
Expected Exclusion: 299 – 505 GeV
Observed Exclusion: 258 – 589 GeV
June 4, 2012 Arun Kumar, University of Delhi 11
Sample Used:
Double Electron dataset 2012A and 2012B
Single Muon dataset 2012A and 2012B
Lumi ~ 2.0 fb-1
Introduction and Data Samples
Trigger Studied
HLT path :
HLT_Ele17_CaloIdT_CaloIsoVL_TrkIdVL_TrkIsoVL_Ele8_CaloIdT_CaloIsoVL_TrkIdVL_TrkIsoVL_v*
HLT_Mu17_Mu8_v* and HLT_IsoMu24
Tag&Probe method used.
HLT path has two legs so efficiency is calculated per leg
June 4, 2012 Arun Kumar, University of Delhi 12
Selection and Definition Tag Selection
Tag must pass the Medium WP of the Electron ID and Tight ID for Muons •(Tables are in back-up) Effective Area-Based PF Isolation < 0.15(0.2) for Electrons (Muons) Tag must pass through the first leg of the HLT path:
HLT_Ele23_CaloIdT_CaloIsoVL_TrkIdVL_TrkIsoVL_HFT30_v* (for Electrons) HLT_IsoMu24_eta2p1_v* (for Muons) Tag PT > 30 GeV |h| < 2.5(2.4), for Electrons(Muons)Transition region excluded DR matching of 0.2 has been used.
Probe Selection
All cuts are same as Tag except the requirement of passing the Trigger and PT requirement. PT > 20 GeV
Tag&Probe pairs are then selected using Z mass window of 15 GeV from the nominal Z mass, 76.2 GeV < M
ee < 106.2 GeV
Definition
Probe Passed
Probe Passed + Probe Falied Trigger Efficiency =
Efficiencies for di-electron Trigger in PT and Eta Bins
HLT_Ele17* HLT_Ele8*
04/08/2012 Arun Kumar, University of Delhi 13
Fig. 43 in AN-12-138 Fig. 43 in AN-12-138
Fig. 43 in AN-12-138 Fig. 43 in AN-12-138
04/08/2012 Arun Kumar, University of Delhi 14
Efficiencies for di-muon Trigger in PT and Eta Bins
HLT_Mu17 HLT_Mu8
IsoMu_24
Fig. 44 in AN-12-138 Fig. 44 in AN-12-138
Fig. 44 in AN-12-138
Arun Kumar, Delhi University 16
TMVA based BDT studies for HZZ2l2n
TMVA Based BDT is chosen with Random Forest Technique. Various BDT configurations
trained and tested.
Training/Testing is done separately for each Higgs Mass point from 200 GeV, to 600 GeV
Analysis is done in inclusive jet bins with VBF.
We allow more phase space in BDT than C&C by either not imposing cut on variables
like MT (Higgs), di-Lepton PT, Min(d (Jet,MET)) OR relaxing the cuts on variables like
M(ll) and PFMET.
Di-Lepton Invariant Mass cut : 60GeV < Mll < 120 GeV
PFMET > 60 GeV
Input variables chosen for training, testing:
di-Lepton invariant Mass
di-Lepton PT
di-Lepton Eta
di-Lepton Phi
PFMET
Min(d (Jet,MET))
Transverse Mass of Higgs
Only 7TeV for the moment
PU Reweighing is applied
04/08/2012 Arun Kumar, University of Delhi 17
Data/MC Comparison Plots for HZZ 2m2n
< 5% differences in data/MC distributions.
04/08/2012 Arun Kumar, Delhi University 18
Input Variables for HZZ2m2n at m(H) = 200 GeV
M(ll) Di-Lepton PT
Di-Lepton Eta
Di-Lepton Phi
PFMET MT Higgs
dPhi (Jet,MET)
N_Jets 1. MT Higgs 2. dPhi(Jet,MET) 3. PFMET 4. Di-Lepton PT 5. N_Jets 6. di-Lepton Mass 7. Di-Lepton Eta 8. Di-Lepton Phi
Ranking of Variables:
9 July 2012 Arun Kumar, Delhi University 19
TMVA BDT Response and ROC Curves for HZZ2m2n
200 GeV
400 GeV
600 GeV
9 July 2012 Arun Kumar, Delhi University 20
Mass (GeV) Cut&Count* BDT Improvement
200 10.38 2.67 ~74%
250 1.84 0.89 ~51%
300 1.10 0.60 ~45%
350 0.72 0.45 ~37%
400 0.86 0.48 ~44%
450 1.05 0.68 ~35%
500 1.52 0.97 ~36%
550 2.27 1.41 ~38%
600 2.85 1.63 ~42%
• Limits are calculated using Higgs combination package after optimizing the mva cut value using LandS package. • Figure of merit for optimizing the MVA output cut is Expected Limit on sSM
Table for Mean Expected Limits at 5fb-1 7TeV Analysis
04/08/2012 Arun Kumar, University of Delhi 21
Summary and Outlook
Presented the ICHEP Update of the HZZ2l2n Analysis
Our contribution was mainly in Trigger Efficiencies, Data/MC Comparison.
We are introducing TMVA BDT analysis for further improvements. First Results
looks promising.
I took 22 Trigger Online Shifts in my last CERN Visit ( 27/03/2012 to 27/06/2012)
and 1 week of ECAL PFG Expert Shift.
Involved in ECAL Reconstruction Software development service tasks.
Perform the TMVA BDT Analysis for both 7TeV and 8TeV. Carry out the analysis
for full 2012 dataset.
Try more signal samples with the same final state.
I will take 2 more weeks of ECAL PFG Expert shifts this year.
Planning for Tracker Alignment Service Task this year.
Poster has been approved for PIC 2012 for HZZ->2l2nu Analysis.
7TeV 8TeV
7TeV+8TeV
04/08/2012 Arun Kumar, University of Delhi 28
Results: Combined Categories Exclusion Limit, R
04/08/2012 Arun Kumar, University of Delhi 29
Trigger Efficiency Numbers
Di-Electron 2012A+2012B
HLT Leg (-2.5,-1.5) (-1.5,0) (0,1.5) (1.5,2.5)
First leg 97.9+-0.23 98.3+-0.03 98.2+-0.03 97.9+-0.23
Second leg 96.5+-0.29 97.6+-0.03 97.5+-0.03 96.1+-0.31
HLT Leg 20-30 GeV 30-40GeV 40-50GeV 50-100GeV
First leg 96.3+-0.08 98.2+-0.03 98.7+-0.03 98.7+-0.05
Second leg 96.1+-0.09 97.4+-0.04 97.8+-0.03 97.8+-0.06
04/08/2012 Arun Kumar, University of Delhi 30
Di-Muon (2012A)
HLT Leg (-2.1,-1.5) (-1.5,0) (0,1.5) (1.5,2.1)
Mu17 93.8+-1.05 95.7+-0.47 95.2+-0.49 93.2+-1.11
Mu8 95.6+-0.90 96.1+-0.45 95.9+-0.46 93.9+-1.06
Di-Muon (2012B)
HLT Leg (-2.4,-1.5) (-1.5,0) (0,1.5) (1.5,2.4)
Mu17 90.7+-0.73 95.2+-0.32 95.3+-0.33 93.1+-0.65
Second leg 93.9+-0.61 95.8+-0.31 95.9+-0.31 95.0+-0.56
Arun Kumar, Delhi University 39
Introduction and Motivation
We aim to investigate the scope of improvement in the analysis with the application
of MVA BDT.
MVA utilizes the whole information of a discriminating variable so better chances of
rejecting background and also it takes care of the correlations between the variables.
Many major analyses like HWW*2l2n and Hgg are getting benefitted so it can
be tried in this analysis.
Fall 11 MC samples and whole 2011 DoubleElectron based HZZ skim datasets are used.
Apply the same ID and Isolation to Leptons as C&C
3rd Lepton Veto and Anti b-Tagging is used.
Di-Lepton Invariant Mass cut relaxed. 60GeV < Mll < 120 GeV
MET > 60 GeV is required to suppress the Drell-Yan background
Main idea is to loosen the cuts used in C&C and open more phase space to do the analysis
Samples and Event Selection:
Only 7TeV for the moment
PU Reweighing is applied
9 July 2012 Arun Kumar, Delhi University 42
Input Variables for HZZ2e2n at m(H) = 200 GeV
M(ll)
Di-Lepton PT
Di-Lepton Eta
Di-Lepton Phi
PFMET MT Higgs
dPhi (Jet,MET) N_Jets
Ranking of Variables:
1. MT Higgs 2. Di-Lepton PT 3. PFMET 4. N_Jets 5. dPhi(Jet,MET) 6. Di-Lepton Eta 7. di-Lepton Mass 8. Di-Lepton Phi
9 July 2012 Arun Kumar, Delhi University 43
TMVA BDT Response and ROC Curves for HZZ2e2n
200 GeV
400 GeV
600 GeV
√s=8 TeV: 25-30% higher s than √s=7 TeV at low mH
All production modes to be exploited
gg VBF VH ttH
Latter 3 have smaller cross sections but better S/B in many cases
Higgs Production at the LHC
5 decay modes exploited
High mass: WW, ZZ
Low mass: bb, ττ, WW, ZZ, gg
Low mass region is very rich but
also very challenging:
main decay modes (bb, ττ) are hard
to identify in the huge background
Very good mass resolution (1%):
H gg and HZZ4l
Higgs Decay Modes The lifetime of Higgs itself is very short (10-22 second). It is only by measuring the characteristics of its decay via various final states (decay channels) into other subatomic particles according to the SM that the Higgs boson will be identified.
Decay products: e, m, t, g, g,qjets, n->missing energy