Seminario di fine III anno Dottorato in Fisica ed Astrofisica Ciclo XXIV Federico De Guio Università degli Studi di Milano Bicocca ed INFN Outline: Vertex selection in Hgg events W search with 1/fb
Vertex selection in Hgg events 2
Introduction Context: Higgs boson search within the SM low mass Higgs (~120 GeV) larger cross section, negligible width clean diphoton final state Mass resolution depends on energy resolution - intercalibration precision - laser correction - etc.. opening angle resolution - depends on vertex Z 0 - no tracks most of the times m H = 2E 1 E 2 (1 cosα) m H m H = 1 2 E1 E 1 E 2 E 2 α tan(α/2) 3
Converted photons Thick tracker in 27% of the events, at least one photon converts in e + e - Conversion tracks provide the z of the vertex But bad energy resolution -> worse mass resolution In the following slides focus on unconverted photons 4
Vertex selection with unconv photons ECAL has no longitudinal segmentation The vertex identification algorithm has to be based on tracking informations and tracking-calorimetry correlation. The standard CMS vertex ID criterion (i.e. SumPt 2 ) is not optimal for H gg. Discriminants Vertex properties correlation vtx-diphoton system 20 variables studied subset of three variables selected - limit correlation MVA approach Ranking method 5
Vertex ID variables Select variables where signal vertex has highest probability of ranking first best three variables: ptbal ptasym - where ptvx = sumpt2 - default criterion in CMS } } kinematic vertex Signal = Hgg event Bkg = minbias event 6
Combining the information The information from different variables is combined using the ranks product algorithm: Vertexes are ranked in each of the three variables The product of the ranks is used as the final discriminant More complicated approaches (such as BDT) have been studied and give equivalent results Efficiency of finding true vertex within 1 cm improves up to 5-% for 2011 PU wrt sumpt2 criterion. At higher PU the gain is even larger. Impact on mass resolution is negligible for pt(gamma-gamma) > 0 GeV/c 7
Vertex ID efficiency from data Idea: use Z events in data to model H production. Method: Select Z ll events and re-run the vertex fitting algorithm removing the leptons tracks. Apply the vertex selection algorithm used for H gg and study the selection efficiency. Good modelling of H gg can be obained ~5% differences at low Higgs pt (due to effect of conversion track). Tried both with electrons and muons: Very similar results. 8
data-mc comparison of vtx variables good modelling of variables in MC 9
vertexid efficiencies data-mc agreement at 5-% level
Conclusions Correct vertex identification is very important in the search for narrow resonances. Choosing the wrong vertex spoils the detector resolution. Presented vertex identification strategy for H gg searches. Exploiting the correlation between the di-photon system and the vertex kinematics improves the vertex ID efficiency (wrt the standard CMS vertex ID algorithm). Overall vertex ID efficiency with PU conditions observed in 2011 data is ~80% for Higgs from gluon-gluon fusion with M H =120 GeV. Studied the vertex ID modelling from data: Z ll events were shown to provide a very good model of H gg Data / MC agreement on the vertex ID efficiency has been observed to be within 5-% at low boson pt and very good at high pt. 11
W search with 1/fb 12
Model and assumptions Altarelli Reference Model (directly comparable to CDF, D0 and Atlas) Neutrino is light and stable Important in the context of the left-right symmetric model Coupling of W to fermions is the same as for W CKM matrix is the same as well No mixing between W and other gauge bosons Excludes mixing between W and either W or Z Decay channels W WW, WZ, and ZZ are suppressed Occurs in many extended gauge models Decay width of W scales with its mass Additional generations of fermions (if exist) are too heavy to be produced 13
Signature W -> e nu signature: single, isolated high-pt electron + large missing transverse energy Main irreducible background: off-shell W and boosted W Search for excess in the transverse mass tail Sensible to detector effect Current best limit from CMS 1.3 TeV: 36/pb, pubblished 2.1 TeV: 1/fb, PAS --> this talk 14
High Level Trigger and eleid Important to keep the W peak in order to compare the data-mc shape/ normalization and test the DD methods for the backgrounds Crossed HLT path used: SingleElectron + Mt lower HLT rate keep all the Ws Crossed path with reasonable selection available for the first inverse femtobarn only develop and test the analysis methods on the first inv femto and use them on the tails > Hz for the crossed HLT path @ 1E33. Not sustainable anymore @ 3.5E33 15
data-mc comparison and selections events 80000 70000 CMS Preliminary 2011 s=7 TeV L=1132.000 pb -1 01_DATA 02_Dibosons 03_TTbar+SingleTop 05_DYll 06_Wmunu 07_Wtaunu 08_gammaJets 09_fakes _Wenu_Z2 _Wenu_Z2_tail 11_Wprime1500 11_Wprime2000 11_Wprime2500 events 160 140 3! CMS Preliminary 2011 s=7 TeV L=1132.000 pb -1 01_DATA 02_Dibosons 03_TTbar+SingleTop 05_DYll 06_Wmunu 07_Wtaunu 08_gammaJets 09_fakes _Wenu_Z2 _Wenu_Z2_tail 11_Wprime1500 11_Wprime2000 11_Wprime2500 60000 120 50000 0 40000 80 30000 60 20000 40 000 20 0 0 0.5 1 1.5 2 2.5 3 "! miss e E T 0 0 1 2 3 4 5 6 7 8 9 ele miss E T /E T data/mc 2 1.8 1.6 1.4 1.2 1 0.8 0.6 Kinematic selections: 0.4 0.2 Δφ (lepton, MET) > 2.5 0.4 < lepton ET / MET < 1.5 0 0 0.5 1 1.5 2 2.5 3 } lepton data/mc 2 1.8 1.6 1.4 1.2 1 0.8 0.6 0.4 0.2 and MET balance: 0 0 1 2 3 4 5 6 7 8 9 both module and direction QCD shape and normalization with fake rate method different strategy wrt 20 dataset SM W background produced in 2 pt bins 16
multijet fake rate estimate define the fake rate as the probability for a SC to pass the eleid selections run over Photon PD To reduce Z->ee and W->enu contamination: Exactly one SC with Pt > 35 GeV MET < GeV apply the fake rate at the same SC sample after the deltaphi and Et/MET selections get both shape and normalization for the fake electrons Use the MC to subtract the W contamination from the estimated fakes sample. 17
Transverse mass events 6 5 4 CMS Preliminary 2011 s=7 TeV L=1132.000 pb -1 01_DATA 02_Dibosons 03_TTbar+SingleTop 05_DYll 06_Wmunu 07_Wtaunu 08_gammaJets 09_fakes _Wenu_Z2 _Wenu_Z2_tail 11_Wprime1500 11_Wprime2000 11_Wprime2500 events 6 5 4 CMS Preliminary 2011 s=7 TeV L=1132.000 pb -1 01_DATA 02_Dibosons 03_TTbar+SingleTop 05_DYll 06_Wmunu 07_Wtaunu 08_gammaJets 09_fakes _Wenu_Z2 _Wenu_Z2_tail 11_Wprime1500 11_Wprime2000 11_Wprime2500 3 3 2 2 1 1-1 -1-2 -2-3 -3 data/mc 0 200 400 600 800 00 1200 1400 2 M T (GeV/c ) 2 1.8 1.6 Good data-mc agreement 1.4 1.2 1 tail monitoring 0.8 0.6 0.4 check evt by evt 0.2 0 0 200 400 600 800 00 1200 1400 data/mc 2 1.8 1.6 1.4 1.2 1 0.8 0.6 0.4 0.2 0 200 400 600 800 00 1200 1400 2 M T (GeV/c ) 0 0 200 400 600 800 00 1200 1400 Laser problems or spikes could manifest themselves with a electron+met signature 18
Efficiencies A tag and probe technique with Z->ee is used to estimate the HLT, reconstruction and ID efficiencies SuperCluster --> gsfelectron --> eleid (WP80) --> HLT (WP80) (1) (2) (3) We expect: (1) few percent inefficiency mainly due to track requirement (2) O(%) inefficiency due to the eleid (WP80) (3) few percent inefficiency wrt the offline selection Full analysis chain efficiency = eff(1) * eff(2) * eff(3) = 96 % used to scale MC 19
Bkg determination and signal estraction Method Get total pp background from M T data spectrum Choose a region with signal contamination << 1% sideband extrapolation Fit sideband, extrapolate to high MT tail to predict # of background events in signal region Method proven to work in MC - Background in MC is sum of all SM contributions (but mainly: off-peak W e ν) 20
Background uncertainties Three ansatz function tested: sideband extrapolation Lower edge varied between 170 and 2 GeV Upper edge varied between 550 and 650 GeV Differences quoted as a systematic on the method 21
Setting the limit Cut and count method: use a sliding search windows to optimize the expected limit flat prior for signal exclusion up to a mass of 2.1 TeV in the electron channel 22
Analysis reload with 3.2/fb same condition as before exclusion up to 2.5 TeV in the combined channel 23
Conclusions Exclusion limit extended up to 2.1 TeV in the electron channel with 1/fb collected in 2011. 24
Backup 25
cross section and BR 26
Systematics 12/04/2011 F. De Guio 27
Bayesian upper limit A Bayesian tool to calculate the expected and observed 95% C.L. upper limits is used 12/04/2011 F. De Guio 28
Channel combination 12/04/2011 F. De Guio 29