Presentation is loading. Please wait.

Presentation is loading. Please wait.

William P. Edson, Teeba Rashid, Dr. Sajjad Alam 1 Dr. Dick Greenwood, Anirvan Sircar 2 Dr. Patrick Skubic, Dr. Muhammad Saleem, Dr. Brad Abbot, Dr. Phil.

Similar presentations


Presentation on theme: "William P. Edson, Teeba Rashid, Dr. Sajjad Alam 1 Dr. Dick Greenwood, Anirvan Sircar 2 Dr. Patrick Skubic, Dr. Muhammad Saleem, Dr. Brad Abbot, Dr. Phil."— Presentation transcript:

1 William P. Edson, Teeba Rashid, Dr. Sajjad Alam 1 Dr. Dick Greenwood, Anirvan Sircar 2 Dr. Patrick Skubic, Dr. Muhammad Saleem, Dr. Brad Abbot, Dr. Phil Gutierrez, Christopher Walker, David Bertsche 3 Dr. Serban Protopopescu 4 1: State University of New York at Albany 2: Louisiana Tech University 3: University of Oklahoma 4: Brookhaven National Laboratory Feb. 7 th 2014 1

2  Physics Channel  Analysis Samples  Object Selection  Event Selection  Tau Selection  QCD Multijet Template method  Variables  Iterative Procedure  TauID Systematic Study  Results  Systematic Uncertainties  Previous Question Responses  Internal Note: http://cds.cern.ch/record/1627649http://cds.cern.ch/record/1627649 2

3 3 Top Pair Branching Fractions for ttbar decays[1] t W+W+ b t W-W- b ντντ τ q q’ g g g

4 ProcessGeneratorSamples Data 20114.6fb -1 (_p822) ttbar semileptonicMC@NLO105200 W( l ν l ) + jetsAlpgen107680-107685 (e ν e ) 107690-107695 ( μν μ ) 107700-107705 ( τν τ ) W(bb) + jetsAlpgen107280-107283 Z( ll ) + jetsAlpgen107650-107655 (ee) 107660-107665 ( μμ ) 107670-107675 ( ττ ) Z( ll )bb + jetsAlpgen109300-109303 (ee) 109305-109308 ( μμ ) 109310-109313 ( ττ ) WWHerwig105985 ZZHerwig105986 WZHerwig105987 Single top (t-channel)AcerMC117360-117362 Single top (s-channel)MC@NLO108343-108345 Single top (Wt)MC@NLO108346 4

5  Trigger:  Tau29_medium_xe35_noMu (B-K)  Tau29T_medium_xe35_noMu_3L1J10(L-M)  Jets:  Anti-K t 0.4 TopoCluster jets  P T > 20 GeV  JVF > 0.75  | η | < 2.5  B-tagging:  BTaggingCalibrationDataInterface 00-01-02  Tau:  Anti-K t 0.4 TopoCluster jets  P T > 20 GeV  | η | < 2.3  tau_EleBDTMedium != 1  tau_muonVeto != 1 5

6 CutCut Description Cut 0Event Cleaning: GRL and larError ≠ 2 Cut 1Trigger (See Slide 5) Cut 2Primary vertex with ntrk > 4 Cut 3LAr FEB Cut 4Event Cleaning: All Good Jets Cut 5N jets ≥ 4 Cut 6Selected tau (See Slide 7) with P T > 40 GeV with Tight Likelihood OR Tight BDTJetScore selection Cut 7Lepton Veto Cut 8MET > 60 GeV Cut 9M T W < 80 GeV Cut 10N bjets ≥ 1 with MV1 > 0.60 Cut 11tau P T < 120 GeV 6 Efficiency (%)Combined1-prong3-prong τ LH or BDT0.1270.0820.046

7 If more than 1 tau candidate following object selection:  Check for number of prongs per candidate  multiple 1-prongs, event is rejected  single 1-prong present, proceed using this tau for analysis  else (only 3-prong) use highest BDTJetscore tau for analysis  other 3-prong taus are kept as jets 7

8  Templates for the MET and dijet mass are constructed for three cases:  1-prong reconstructed taus  3-prong reconstructed taus  Combined 1 and 3-prong 8 BaseLine:  Full Event and Tau Selection Control:  Full Event Selection  Tau Identification passing either:  Likelihood > -40  BDTJetScore > 0.1  AND not passing Tight selection for same tauID

9  Template for QCD taken as the shape difference in the concatenated MET and dijet mass histograms for Data control sample with the control samples from MC:  ttbar semileptonic  W + jets  Z + jets  single top  diboson  9

10 10 MET (GeV) ) Dijet Mass (GeV)

11 11 Plot comparing the distribution of the normalized control and baseline samples with those of the different MC backgrounds following cut 6 (chosen due to increased statistics)

12 Comparison of normalized MET distribution for baseline and control samples following cut 4 Comparison of normalized Dijet mass distribution for baseline and control samples following entire cutflow 12

13 13  Previous Results relied on a fixed scale factor to luminosity for signal MC in the control region. Scaling the MC in this way implied a prior knowledge of what the top cross section value is.  To this end, we have altered the analysis to instead determine the scale factor for signal MC using an iterative approach beginning with an arbitrary cross section from which the scale factor is calculated.  The cross section resulting from the fit using this scale factor is then used to recalculate new scale factors and the process repeated until the cross section value converges.

14  Samples:  Signal Samples: Z → ττ final states selected with one tau decaying via muon while the other hadronically  Control Samples: Primarily W → μν + jet data driven events  Background further reduced by subtracting the number of events with the same charge (SS) for muon and tau candidate from the number of events where the muon and candidate have opposite charge (OS) 14

15  Samples Divided into 2 regions: 1. Those passing TauID OR tight selection and OS-SS subtraction 2. Those failing both tight selections but passing OS- SS subtraction  Further define five variables:  N i W : Number of control events in region i after MC predicted Z ττ contribution removed  N i d : Number of data events in region i  N i S : Number of MC predicted Z ττ events in region i  N i bkd : Number of background events in region i  N pred : Predicted number of signal in region 1 15

16  Tau ID Uncertainty is Estimated based on comparison of N 1 S and N pred.  Equations:  N 2 bkd = N 2 d - N 2 S  N 1 bkd = N 2 bkd * (N 1 W / N 2 W )  This assumes the shape of the background is given by the shape of the control sample  N pred = N 1 d - N 1 bkd 16 Combined (1 & 3-prong) 1-prong3-prong Total ID Uncertainty (%) 4.14.24.8 Exp. to pred. difference (%) -2.2+1.1-10

17 17 Comparison of stacked Signal and Control Samples scaled to N i S and N i bkd respectively to data in defined regions 1 (left) and 2 (right) for the combined 1 and 3-prong case

18 TypeCombined (1 & 3-prong) 1-prong3-prong Signal0.482 ± 0.0340.530 ± 0.0510.416 ± 0.050 QCD multijet0.311 ± 0.0300.216 ± 0.0440.434 ± 0.050 Other MC backgrounds 0.206 ± 0.0030.260 ± 0.0010.147 ± 0.001 18 Fit Result of Data to Signal MC, QCD Multijet, and other MC Backgrounds for concatenated MET & M jj for Combined (1 & 3-prong), chi 2 /ndf: 1.21609 METM jj

19 19 Both plots are for the combined 1 & 3-prong case

20 20 Both plots are for the combined 1 & 3-prong case Additional comparison plots using other variables found in backup slides 42 and 43

21  Signal and QCD output event count results are determined using the output fractions from the fitting analysis  21

22 22 Tau case Signal Single top W + jets Z + jets DibosonQCDSumData Combined (1 & 3-prong) 51244138180.502991012979 1-prong3323194110.42106574534 3-prong181144470.12192438445

23  Combined (1 & 3-prong):   1-prong:   3-prong:   SM NNLO prediction: for top mass of 173.3 GeV [2]  NikHEF Result [3]:  23

24 SampleCross section Including ttbar leptonic fakes151.8 ± 10.3 Current Result149.0 ± 11.0 Difference1.8%

25 25

26 26

27  1-prong 27  3-prong

28 Mass pointX-section for corresponding mass point Nominal signal sample Error (%) M top = 167 GeV121.2112.757.4 M top = 170 GeV115.9112.752.8 M top = 175 GeV99.4112.7512.0 M top = 177 GeV95.5112.7515.3 28 All available samples with different mass points are fast simulated. We compare them with fast simulated signal sample.

29  LAr hole treatment:  In 3.1 you say you are vetoing events with jets / electrons in the LAr hole. This doesn't seem to correspond to the jet / Etmiss and egamma recommendations  >Our analysis now does as prescribed on the twiki pages.  Then in 3.3 you say you reject electrons in the hole (this is inconsistent with 3.1, where you say you reject the event), but egamma recommends not to do this,  >The discussion was misleading. This discussion was about Lar noise bursts and was mixed up with LAr Hole. We fixed the text now.  Muon & electron scale factors: In sections 3.3 and 3.4 you say you are using scale factors for electrons and muons - how are these used in your case where you veto electrons & muons, rather than select them?  >This was a misinterpretation, electron energy is scaled using the prescribed correction factor and the muon p T is smeared again following prescription  Please provide details on which jet calibration configuration is used and which b-tagging calibration is used.  >We used anti-kt4 algorithm with EM+JES calibration for 2011 data as recommended by the top working group. (slide 5)  >For b-tagging calibration we used MV1 tagger with the calibration recommended by the flavor tagging group for 2011 data. (slide 6)  >Version 00-01-02 of the BTaggingCalibrationDataInterface was used. (slide 5) 29

30  Has this idea to use an OR of the likelihood and BDT tau ID been discussed with the tau CP group? How are the systematic uncertainties for this OR determined?  >No, This has not been shown in tau CP group meetings. However, the study is done separately by Serban for the tau ID related systematics and added as Appendix C to the note (see slides 14-17). This study tells us that tau ID related systematic uncertainties are not much different in the case of OR than as those if evaluated separately.  Please provide some details on the trigger scale factors and efficiencies you are using.  >The triggers used in our analysis have expected average efficiencies in the appropriate data-periods around 70% as measured on the signal sample. These triggers selections are initially used by the charged higgs analysis (for mH < mt) in ttbar decay with tau + jets final states: https://cds.cern.ch/record/1419805?ln=en#  Correlation between m(jj) and MET:  Can you provide some plots demonstrating in the MC that m(jj) and MET are uncorrelated?  >See slide 10  The mTW plot … Can you also add the same plot for m(jj) - this is an important check.  >See slide 12  Can you add a plot showing the shape of the different backgrounds for the two variables you are fitting?  >See slide 11  Can you provide the formula used in the fit and which minimization technique that is used?  >We used the TFraction Fitter. The fit is performed using the signal templates taken from the MC, QCD template taken from the data (loose sample) (see slide 10).  For minimization this uses MINUIT2 30

31  Ensemble tests:  To test the impact of fitting two 1D distributions, please could you make ensemble tests based on the 2D MET-m(jj) distribution - you create the 2D distribution (with the same >>binning you have in the fit) and then draw toy experiments using poisson random numbers for each bin. You can then extract the 1D distributions from each 2D toy and pass them >>to the fit procedure. In addition to the linearity test - can you also produce the pull distributions?  > Work still in progress.  The offset and slope of the linearity test look to be inconsistent with 0 and 1, respectively. Is this corrected for? Also, the linear fit in Fig 8 seems to be quite poor - chi2 of 54 for 4 >>degrees of freedom - have you investigated this?  > The non-linearity is very small compared to our errors. Currently working on the 2D MET-Dijet Mass Distribution from last suggestion to hopefully solve this.  Could you clarify where ttbar events other than the signal (e.g. e/mu+jets) are included in the background estimates?  > The fake ttbar events have been added to the list of Other MC backgrounds subtracted away during the QCD template process (slide 9) and included in the Other MC backgrounds fraction (slide 18). It is treated as a constant background and scaled to luminosity. The result is a small correction to the previously stated result (slide 24). 31

32  There should be 17 components for the JES uncertainty  > Work is in progress to produce and analyze the output.  516-527: tau ID / energy scale uncertainties: Please could you provide some details and numbers here and references to where the numbers are derived?  >Since this analysis is closely following the charged higgs analysis with mH < mtop and have similar final state (as referenced in our note). These numbers are borrowed from charged higgs note  Tables 6-8: The b-tag uncertainty seems to be missing from the tables?  > See slides 26, 27  Results: Please can you provide the dependence of the measured cross section as a function of the top mass - you can do this by re-running the analysis on the ttbar samples with a different top mass and then looking at how the measured cross section changes.  > See slide 28 32

33  Aiming for journal paper for the winter conference  Editorial Board desired for publication  Internal Note: http://cds.cern.ch/record/1627649 http://cds.cern.ch/record/1627649 33

34  Dr. Patrick Skubic  Dr. Muhammad Saleem  Dr. Brad Abbot  Dr. Phil Gutierrez  Dr. Dick Greenwood  Christopher Walker  Dr. Serban Protopopescu 34

35 1) Neil Collins, TopCross Section (Current Status and Early LHC Prospects). 10 March 2010. 2) M. Czakon, P. Fiedler, and A. Mitov, The total top quark pair production cross-section at hadron colliders through O( α S 4 ), arXiv:1303.6254 [hep-ph]. 3) ATLAS Collaboration, Measurement of the ttbar production cross section in the tau+jets channel using the ATLAS detector, arXiv:1211.7205v2 [hep-ex].arXiv:1211.7205v2 4) CMS Collaboration, Measurement of the top-antitop production cross section in the tau+jets channel in pp collisions at sqrt(s) = 7 TeV, arXiv:1301.5755v2arXiv:1301.5755v2 5) ATLAS Collaboration, Measurement of the top quark pair production cross section in pp collisions at s√= 7 TeV in μ + τ final states with ATLAS, ATLAS-CONF-2011-119 35

36 36

37  Pseudo data was created using constant fraction for other MC backgrounds and varying signal fractions (0.2 to 0.7 in steps of 0.1)  For each signal fraction, the bin content was randomized for templates for Signal and other MC backgrounds from baseline samples and QCD from the control samples  The three randomized templates coming together formed the pseudo data  The procedure was repeated 10,000 times for each fraction resulting in a histogram for each fraction which could be fit to a Gaussian 37

38 Input Signal Fraction: 0.2Input Signal Fraction: 0.7 38 Remaining result histograms included in Back up slides 31 and 32

39 Input Signal Fraction: 0.3Input Signal Fraction: 0.4 39

40 Input Signal Fraction: 0.5Input Signal Fraction: 0.6 40

41 41  The Gaussian mean values and errors of the Ensemble test runs (y) were plotted versus their corresponding input fraction (x) with the resulting linear fit below: Input Signal Fraction Output Signal Fraction

42 42 Both plots are for the combined 1 & 3-prong case

43 43 Both plots are for the combined 1 & 3-prong case

44 TypeCombined (1 & 3-prong) 1-prong3-prong Signal0.463 ± 0.0380.510 ± 0.0540.419 ± 0.056 QCD multijet0.320 ± 0.0340.227 ± 0.0460.422 ± 0.052 Other MC backgrounds 0.217 ± 0.0040.270 ± 0.0010.160 ± 0.002 44 Likelihood Analysis (Analysis A): TypeCombined (1 & 3-prong) 1-prong3-prong Signal0.528 ± 0.0410.564 ± 0.0550.495 ± 0.064 QCD multijet0.252 ± 0.0350.168 ± 0.0450.350 ± 0.057 Other MC backgrounds 0.218 ± 0.0030.264 ± 0.0020.154 ± 0.001 BDT Analysis (Analysis B):

45 TypeCombined (1 & 3-prong) 1-prong3-prong Analysis A0.1040.0700.040 Analysis B0.0920.0600.033 45

46  Analysis A  Combined (1 & 3-prong):   1-prong:  3-prong:  Analysis B  Combined (1 & 3-prong):   1-prong:  3-prong: 46

47 47 Distribution of M T W from signal and Multijet control sample demonstrating the contribution of fakes in control sample and real τ in signal distribution


Download ppt "William P. Edson, Teeba Rashid, Dr. Sajjad Alam 1 Dr. Dick Greenwood, Anirvan Sircar 2 Dr. Patrick Skubic, Dr. Muhammad Saleem, Dr. Brad Abbot, Dr. Phil."

Similar presentations


Ads by Google