Search for Electroweak Top Quark Production at DØ

Transcription

1 Search for Electroweak Top Quark Production at DØ Columbia University SCIPP Seminar University of California, Santa Cruz December 12, 26

2 The Top Quark Discovered in 1995 at the Tevatron Heaviest particle to date: mt = ± 2.1 GeV / c 2 close to the scale of electroweak symmetry breaking. Beyond the Standard Model effects? Has a large effect in loop corrections that are in powers of mt Predicted lifetime of s, too short to form bound states Retains more information from the fundamental collision 2 m W!GeV" LEP1 and SLD LEP2 and Tevatron (prel.) 68# CL 8.3!" m H!GeV" m t!gev"

3 Top Pair Production So far, the top quark has only been observed in pairs, produced by the strong force. 85% qq annihilation at the Tevatron proton antiproton q q g t Best Independent Measurements of the Mass of the Top Quark (*=Preliminary) DØ Run II Preliminary Fall 26 15% gluon fusion CDF-I dilepton ± 11.4 DØ-I dilepton ± 12.8 CDF-II dilepton* ± 5.5 dilepton/l+jets combined pb 1 dilepton (topological) 37 pb pb pb g t DØ-II dilepton* ± 8.3 CDF-I lepton+jets ± 7.3 DØ-I lepton+jets 18.1 ± 5.3 CDF-II lepton+jets* 17.9 ± 2.5 DØ-II lepton+jets* 17.3 ± 4.5 CDF-I alljets 186. ± 11.5 CDF-II alljets* 174. ± 5.2 CDF-II b decay length* ± 15.8 χ 2 / dof = 1.6/1 Tevatron Run-I/II* ± 2.1 ltrack/emu combined pb 1 τ+jets (b-tagged) NEW 35 pb all-jets (b-tagged) NEW pb 1 l+jets (µ-tagged) NEW pb 1 l+jets (b-tagged) NEW pb 1 l+jets (topological) NEW 91 pb m top = 175 GeV Kidonakis and Vogt, PRD 68, (23) Cacciari et al., JHEP 44, 68 (24) pb +.3 pb pb +.4 pb pb pb g g g g t Top Quark Mass [GeV] σ (pp tt) [pb] g 3

4 Electroweak Top Quark Production NLO cross sections at mt = 175 GeV, Phys.Rev. D7 (24) q s-channel t DØ Limits (95% confidence level) q' q' g W b W + σ tb =.88 ±.7 pb t-channel σ tqb = 1.98 ±.23 pb tw production is negligible at the Tevatron q t b Methodology Integrated Luminosity s-channel t-channel Cut-based 23 pb pb 11.3 pb Neural Network 23 pb pb 5. pb Likelihood 37 pb pb 4.4 pb Methodology CDF Limits Integrated Luminosity s+t-channel Neural Network 695 pb -1 < % CL Likelihood 955 pb -1 < % CL Matrix Element 955 pb -1 = pb 4

5 Why is Electroweak Production Interesting? Electroweak production is directly proportional to Vtb 2 d s b = V ud V us V ub V cd V cs V cb V td V ts V tb Current ual limits matrix to Vtb elements 2 assume can the CKM in principle matrix is unitary. Without that assumption, it is virtually unconstrained:.7 Vtb.9993 (9% CL) d s b Single top production tests that assumption Good place to study the V A charged current interaction The Standard Model predicts the top quark to be produced highly polarized Because the top quark decays before it has time to hadronize, it preserves its polarization S. Eidelman et al, Phys. Lett. B 592, 1 (24) q' g W b q t b 5

6 Why is Electroweak Production Interesting? Sensitive to new physics. s-channel and t-channel have different sensitivities. The s-channel is more sensitive to charged resonances, like top pions or charged Higgs particles. The t-channel is more sensitive to FCNC and other new interactions. Tait, Yuan PRD63, 1418 (21) 6

7 Fermilab and the Tevatron The Tevatron is a pp collider with s = 1.96 TeV Just a few weeks ago reached the milestone: 2 fb -1 have been delivered in Run II. Where the top quark was discovered 1 years ago using less than 1 pb -1 It is currently the only place to study the top quark. 7

8 The Tevatron Performance: over 2 fb -1 Delivered End of Run IIa Current Single Top Analyses End Here 8

9 The DØ Detector N Forward Mini- Drift Tubes Muon Toroid Muon Scintillation Counters S PDTs CC Shielding EC EC Platform Tracking System: : Silicon, Fiber Tracker, Solenoid, Central & Forward Preshowers Fiber Tracker/Preshower VLPC Readout System 9

10 How does Single Top Look? The event signature for single top is: a charged lepton and a neutrino (missing ET) coming from the W decay a bottom quark coming from the top quark decay proton antiproton s-channel q q' W + t W + e + ν b b in s-channel, the other bottom quark (b on the diagram) in t-channel, the light quark (q in the diagram). This one is often quite forward. in t-channel, the second bottom quark from gluon splitting (b on the diagram) is usually not observed. proton antiproton t-channel q' W t b g W + q µ + ν b b 1

11 oson. and W boson. and W boson. shows the transverse momenta Figure and2pseudorapidities shows the transve fo els of the s-channel and t-channel Carlo models single top of the processes, s-channea son. Single Top Parton Distributions lepton and W boson lepton 14 lepton lepton b from t b from t b from t b from t other b other b 12 other b other b lepton b from t 12 other b lepton lepton lepton lepton 14 b b 12 6 b from t b from t b from t b from t other b 6 6 other b 12 other b other b other b other b s-channel Pt, [GeV/c] Pt, [GeV/c] Eta*Q(lepton) Pt, [GeV/c] Pt, [GeV/c] pt [GeV/c] Pt, [GeV/c] lepton b from t t other b other light q b light q Pt, [GeV/c] t-channel 8 6 lepton 1 b 2 from t other b light q pt [GeV/c] Eta*Q(lepton) Q(lepton) 4 η Eta*Q(lepton) Q(lepton) η Eta lepton lepton lepton 2 b from b tt b t other 12 b from t b from t other 12 other light qb 12 other b other 12 light qb light q 1 light q light q Eta*Q(lepton) Pt, [GeV/c] lepton b from t other b Eta η 14 lepton 6 b from t other b 12 light q Jovan Mitrevski Pt, [GeV/c] Pt, [GeV/c] Eta*Q(lepton) Pt, [GeV/c] light q 8 2 η 1

12 Why hasn t Single Top been observed yet? The single top cross section is not that much smaller than that of tt, which was first observed over 1 years ago Top pair production has a much more unique event signature. Single-top is kinematically between top pair and W+jets. The main background is W+jets. W+bb q' g b b W+charm and W+light jets that get tagged B-tagging is key to reducing the large background q W Other Backgrounds q g q e Multijet (i.e. fake lepton) tt (more so in higher jet multiplicity bins) q g q jet 12 jet

13 Analysis Overview Loose event selection (leave more to the NN, DT, ME). Work in separate lepton, jet, and tag bins. Background Modeling W+jets and tt : Alpgen/Pythia with MLM matching The heavy flavor content for the W+jets sample is measured in data and a k-factor of 1.5 is applied to the W+cc and W+bb MC samples. multijet (fake lepton) background: orthogonal data sample. Before b-tagging (where single-top is negligible), the multijet sample is scaled to the fake lepton yield, and the W+jets sample is scaled to the real lepton yield minus the tt yield. We model the single-top signal with the Monte Carlo generator SingleTop, a version of CompHEP modified to reproduce NLO distributions. 13

14 Event Selection Good data quality Good primary vertex lepton+jets triggered data Leptons: tight electron with pt > 15 GeV, η < 1.1, or tight muon with pt > 18 GeV, η < 2.. Veto on second charged lepton Jets: leading pt > 25 GeV, second jet pt > 2 GeV, others pt > 15 GeV. leading η < 2.5, η < 3.4 for subsequent jets. 15 GeV < ET < 2 GeV Triangle cuts: don t take events that have the missing ET aligned or anti-aligned with the lepton or the leading jet, since that s probably due to mismeasurement. 14

15 Yields Before b-tagging Yields Before b-tagging Electron Channel Muon Channel 1 jet 2 jets 3 jets 4 jets 5+ jets 1 jet 2 jets 3 jets 4 jets 5 jets Signals tb tqb tb+tqb Backgrounds t t ll t t l+jets W b b W c c 1, ,45 1, W jj 23,417 5,437 1, ,476 4,723 1, Multijets 1,691 1, Background Sum 27,37 8,22 3, ,816 6,434 2, Data 27,37 8,22 3, ,816 6,432 2,

16 Data/MC Comparisons Before b-tagging (2 jet bin, electron channel) Yield [counts/1gev] 2-1 DØ Run II Preliminary.9 fb e+jets pre tag ==2 jets Yield [counts/1gev] DØ Run II Preliminary.9 fb e+jets pre tag ==2 jets Yield [counts/1gev] 4-1 DØ Run II Preliminary.9 fb e+jets pre tag ==2 jets Lepton p [GeV] T p (jet1) [GeV] T p (jet2) [GeV] T Yield [counts/1gev] DØ Run II Preliminary.9 fb e+jets pre tag ==2 jets Yield [counts/1gev] DØ Run II Preliminary.9 fb e+jets pre tag ==2 jets Yield [counts] 2-1 DØ Run II Preliminary.9 fb e+jets pre tag ==2 jets Missing E [GeV] T M T (W) [GeV] Δ R(jet1,jet2)

17 B-Tagging As mentioned in the introduction, b- tagging is one of the key ways we decrease the large background One of the recent improvements at DØ is the new Neural Network Tagger. The plots on the right show its performance on data in comparison to JLIP, an older tagger. Rank Variable Description 1 SV T SL DLS Decay Length Significance of the Secondary Vertex (SV) 2 CSIP Comb Weighted combination of the tracks IP Significances 3 JLIP Prob Probability that the jet originates from the PV 4 SV T SL χ 2 dof Chi Square per degree of freedom of the SV 5 SV T L N T racks Number of tracks used to reconstruct the SV 6 SV T SL Mass Mass of the SV 7 SV T SL Num Number of SV found in the jet Table 1: NN input variables ranked in order of power. b-jet Efficiency (%) % 37 % % 32% p T > 3 and <! <.8 Tagger NN JLIP Fake Rate (%) TIGHT operating point Efficiency: 5%, Fake Rate:.5% 17

18 Signal : Background 18

19 Yields with One b-tagged Jet Yields with One b-tagged Jet Electron Channel Muon Channel 1 jet 2 jets 3 jets 4 jets 5+ jets 1 jet 2 jets 3 jets 4 jets 5 jets Signals tb tqb tb+tqb Backgrounds t t ll t t l+jets W b b W c c W jj Multijets Background Sum Data

20 Data/MC Comparisons After b-tagging (2 jet bin, electron channel, one tag) Yield [counts/1gev] 1-1 DØ Run II Preliminary.9 fb e+jets ==1 tag ==2 jets 5 Yield [counts/1gev] DØ Run II Preliminary.9 fb e+jets ==1 tag ==2 jets Yield [counts/1gev] 2-1 DØ Run II Preliminary.9 fb e+jets ==1 tag ==2 jets Lepton p [GeV] T p (jet1) [GeV] T p (jet2) [GeV] T Yield [counts/1gev] 1-1 DØ Run II Preliminary.9 fb e+jets ==1 tag ==2 jets 5 Yield [counts/1gev] DØ Run II Preliminary.9 fb e+jets ==1 tag ==2 jets Yield [counts] DØ Run II Preliminary.9 fb e+jets ==1 tag ==2 jets Missing E [GeV] T M T (W) [GeV] Δ R(jet1,jet2)

21 How to proceed Method 1: Optimized cuts-based, but shown not to be competitive Method 2: Neural Networks. Bayesian Neural Networks show lots of promise Method 3: Boosted Decision Trees Start Key HT_AllJets > 96.82? Branch node false Test? true TransverseMass_Jet1Jet2 > 117.3? TransverseMass_Jet1Jet2 > 15.7? BestTopMass > 25.6? Method 4: Matrix Element. This is the focus of this presentation Purity Jet2Pt_NotBest > 47.16?.72 End node: return purity Cos_LeptonQZ_BestTop >.171?

22 Matrix Element Method The matrix element encodes all the kinematic information. dσ = M 2 F dq Where F is the flux factor and dq is the Lorentz invariant phase space factor. The method makes maximal use of the information. The matrix element contains all the properties of the interaction. (practical limitation: LO matrix element.) The method requires no training. It is firmly grounded in the physics. 22

23 The Main Idea of the Matrix Element Method Assume a particular process (e.g. t-channel single top). The probability to observe a particular configuration of jets and leptons given that process is proportional to the differential cross section of that process for such an configuration. Can determine the probability of an observed configuration assuming (given) signal and its probability assuming (given) background. Bayes rule: P (config process) dσ process d(config) P (signal config) = P (config signal) P (signal) P (config signal)p (signal) + P (config back)p (back) 23

24 The Matrix Element Analysis Being more explicit: where P (x S) = 1 σ s ( dσs dx dσ dx i perm d n y ) D(x) = M(y, i) 2 F (y) P (x B) = i P (x S) P (x S) + P (x B) w i P (x B i ) f 1 (y)f 2 (y)tf(y, x, i) P (x B i ) = 1 σ bi σ = ( dσbi dx dx dσ dx ) y: the parton-level information x: the reconstructed information F: the flux factor f1 and f2: the two PDFs TF: the transfer function, relating parton and reconstructed information. 24

25 The Transfer Functions We measure reconstructed values, but the Matrix Element uses parton values. Transfer Functions We assume: can use per-object transfer functions the angles are perfectly measured For jets, we use a double Gaussian form parameterized in energy in η bins. There are separate parameterizations for light, b, and b( μ) jets jets. W object!e parton, E object #$ *!E object *E parton *p 1 # 2 *!E object *E parton *p 4 # 2 1 '2(!p 2 )p 3 p 5 # %!e p 2 2 p )p 3 e 5 # For electrons, we use a single Gaussian parameterized in energy and η. For muons we use a single Gaussian parameterized in (q/pt), η, and if there are silicon tracker hits. 25

26 Transfer Functions (cont.) b jets:.5< " <1 det DØ Run II In Progress E p =15. GeV E p =25. GeV E p =5. GeV E p =9. GeV JES=1. The electron transfer function, Ep = 3 GeV, η =.3 The electron generated xfer func DØ Run II In Progress helxfer Entries 5 Mean 3.32 RMS E (GeV) [GeV] E jet The jet transfer function for b-jets The jet5 p.5.4 T DØ Run II In Progress tchan pos unsmeared smeared dgstar matched.3 Applying the jet transfer function on the bottom quark from the top decay (blue) vs. full GEANT simulation (yellow) [GeV/c] p T

27 B-Tagging Another feature that I view as part of the transfer function is the assignment of jets to partons. We sum all possible assignments. If we do not have extra information, we weigh each permutation equally: dσ =.5dσ(j 1 q 1, j 2 q 2 ) +.5dσ(j 2 q 1, j 1 q 2 ) However, if we do have one b-tagged jet, and we are calculating the t-channel cross-section, where we are using diagrams of the type b u e + νe b d, we can do better: dσ = ɛ b (j tag )[1 ɛ l (j untag )]dσ(j tag b, j untag d) +[1 ɛ b (j untag )]ɛ l (j tag )dσ(j untag b, j tag d) The extension to 3 jets is straightforward. 27

28 Muon Charge The s-channel, which has a b-jet and a b-jet as the two jets, can benefit from b-tagging since the tagging efficiency depends on η and pt. But maybe if a b-quark decays leptonically we can use the muon charge: Muon Charge Assigment Efficiency hmuonchargeeff_ptrel Entries 11 Mean RMS all good: b μ - ν c b μ + νc but also: b xx c μ + ν s b xx c μ - νs.2 DØ Run II In Progress p [GeV/c] Trel Use ptrel, or the pt of the muon relative to the jet. Muons from charm quarks are expected to have a lower ptrel. Do the same as for b-tagging. It will be used in the next round. µ p Trel jet 28

29 The Details We are using MadEvent LO matrix elements, and CTEQ6L1 pdfs. We look at events with 2 and 3 jets. The Matrix Elements we now use are: Signal: s-channel or t-channel single-top Background: Wbb, Wcg, and Wgg (2 jets); Wbbg (3 jets) P (x B) = w W bb P (x W bb) + w W cg P (x W cg) + w W gg P (x W gg) Major upgrade under way: tt matrix elements. Also adding extra 3 jet W+jets and signal matrix elements. 29

30 s-channel Discriminant Plots, 2 Jets DØ Run II Preliminary tb Discriminant.1 DØ Run II Preliminary tb Discriminant tb Wbb tb Discriminant DØ Run II Preliminary tb Discriminant.1 DØ Run II Preliminary tb Discriminant tb Wjj tb Discriminant.8.6 tb tt l+jets.8.6 tb tt ll tb Discriminant tb Discriminant

31 t-channel Discriminant Plots, 2 Jets DØ Run II Preliminary tqb Wbb tq Discriminant DØ Run II Preliminary tq Discriminant tq Discriminant tqb tt l+jets tq Discriminant DØ Run II Preliminary.18 tqb Wjj tq Discriminant DØ Run II Preliminary tq Discriminant tq Discriminant tqb tt ll tq Discriminant

32 Discriminant Results, 2 Jets tb Discriminant : e+µ w/ =2 Jets and Tags Combined DØ Run II KS: Preliminary.978 Data tb tqb Wbb wcc Wjj QCD tt! ll tt! l+jets.5 1 tb Discriminant tq Discriminant : e+µ w/ =2 Jets and Tags Combined KS: 1 s-channel discriminant DØ Run II Preliminary Data tb tqb Wbb wcc Wjj QCD tt! ll tt! l+jets.5 1 tq Discriminant t-channel discriminant tb Discriminant : e+µ w/ =2 Jets and Tags Combined DØ Run II KS: Preliminary tb Discriminant s-channel discriminant tq Discriminant : e+µ w/ =2 Jets and Tags Combined KS: 1 45 DØ Run II Preliminary tq Discriminant t-channel discriminant

33 Discriminant Results, 3 Jets tb Discriminant : e+µ w/ =3 Jets and Tags Combined KS:.912 Data tb tqb Wbb wcc Wjj QCD tt! ll tt! l+jets.5 1 tb Discriminant tq Discriminant : e+µ w/ =3 Jets and Tags Combined KS: 1 Data tb tqb Wbb wcc Wjj QCD tt! ll tt! l+jets tq Discriminant s-channel discriminant t-channel discriminant tb Discriminant : e+µ w/ =3 Jets and Tags Combined KS: tb Discriminant 5 s-channel discriminant tq Discriminant : e+µ w/ =3 Jets and Tags Combined KS: tq Discriminant t-channel discriminant

34 Systematics We have two types of systematics: overall normalization and shapechanging. Currently, we treat the jet energy scale and the tag rate function systematics as shape-changing systematics, everything else is flat. Components for Normalization Single-Tagged Two-Jets Electron Channel Percentage Errors tb tqb t tlj t tll W bb W cc W jj Mis-ID e Luminosity ( 6.1) ( 6.1) Cross section ( 16.) ( 15.) Branching fraction ( 1.) ( 1.) Matrix method Primary vertex Electron ID Jet ID Jet fragmentation Trigger Components for Normalization and Shape Jet energy scale Flavor-dependent TRFs Statistics Combined Acceptance uncertainty Yield uncertainty TABLE 29: Electron channel uncertainties, requiring exactly one tag and two jets. 34

35 Expected Sensitivity no Data, just MC We build a 2-D histogram of s-channel vs t-channel discriminant and run our Bayesian limit setting framework to determine the posterior. We can require the SM ratio, σt/σs = 2.25, or in future updates, not. ] -1 Assuming SM ratio Posterior Density: e+µ w/ 2+3 Jets and >=1 Tag post Prob. Density [pb DØ Run II Preliminary σ s+t δσ σ s+t s+t +1.8 = = Cross Section [pb] t-channel cross section [pb] Not assuming SM ratio DØ DØ Run Run II In II Progress Preliminary 9 pb Flat Systematics Only 3 std. dev. 2 std. dev. 1 std. dev. MPV SM s-channel cross section [pb] -1 35

36 MC Ensemble Test Results Input: pb Cross Section Per Ensemble Entries: 4 Mean:.622 Mode:.1 RMS: Cross Section [pb] Cross Section Per Ensemble Input: 2.86 pb Entries: 4 Mean: 3.18 Mode: 3.5 RMS: Cross Section [pb] Cross Section Per Ensemble Entries: 2 Input: 2 pb Mean: 2.34 Mode: 2.5 RMS: 2.73 DØ Run 5II Preliminary 1 Cross Section [pb] ME analysis Cross Section Per Ensemble Input: 7.89 pb Entries: 2 Mean: 8.28 Mode: 7.5 RMS: Cross Section [pb] Cross Section Per Ensemble Input: 5.92 pb Entries: 199 Mean: 6.58 Mode: 6.5 RMS: Cross Section [pb] Ensemble response s+t cross section [pb] χ 2 /ndof =.64/4 Slope = 1.5 ±.8 Intercept =.21 ± Input s+t cross section [pb] 36

37 Observed Results (Matrix Element) Posterior Density: e+µ w/ 2+3 Jets and >=1 Tag Entries/.2 pb -1 post Prob. Density [pb ] σ s+t = δσ σ s+t s+t = Cross Section [pb] 1 Cross Section For Zero Signal Ensembles DO Run II Preliminary! s+t Obs Entries: p-value:.211 Sigma: 2.87 = 4.6 pb Cross Section [pb] 37 DØ Run II Preliminary.9 fb -1 e / 2jets / 1tag e / 2jets / 2tags e / 3jets / 1tag e / 3jets / 2tags mu / 2jets / 1tag mu / 2jets / 2tags mu / 3jets / 1tag mu / 3jets / 2tags Combined (Matrix elements) Z. Sullivan PRD 7, (24), m t = 175 GeV pb pb pb pb pb pb pb pb pb σ(pp _ tb+tqb) [pb]

38 Observed Results (Preliminary Evidence!) Decision Trees - p-value A 3.4σ excess!! Dugan O Neil (SFU) First Evidence for Single Top Dec. 8, 26 5 / 69 38

39 Observed Results SM compatibility = 21% DØ Run II Preliminary.9 fb -1 Decision trees pb ME Matrix elements pb Bayesian NNs pb Z. Sullivan PRD 7, (24), m t = 175 GeV σ(pp _ tb+tqb) [pb] DT 39

40 Summary / Outlook Preliminary first evidence for single-top production! The matrix element method is being updated with the addition of tt matrix elements and other improvements. An extra ~1 pb -1 of data will probably be added soon. In the longer term, the Tevatron is gathering new data as we speak. By the end or Run IIb, it is on track to have delivered 7-8 fb -1. Run IIb has improvements that will help the physics. The new Silicon Tracker Layer will improve b-tagging. Upgraded trigger will ensure we can keep triggering on the data we need. At the LHC the cross section is two orders of magnitude greater. The excitement is just beginning! 4

41 Backup Slides 41

42 Estimating the Multijet Background Yield The multijet background yield is estimated by solving the following system equations for the real and fake lepton yields: N loose N tight = Nloose fake l = Ntight fake l + Nloose real l + Ntight real l = ε fake l N fake l loose + ε real l N real l loose Nloose is the data yield with a loose lepton ID. Ntight is the data yield with a tight lepton ID. εreal-l is the probability of a real lepton to pass the tight requirements given it has passed the loose. It is measured in MC, with a data/mc scale factor measured in Z l + l - events. Loose vs. Tight loose electron: isolation EM fraction, shower shape, track match. tight electron: loose + likelihood loose muon: medium muon with cosmic veto, Δr(jet, muon) >.5 tight muon: loose + track isolation εfake-l is the probability of a fake lepton to pass the tight requirements given it has passed the loose. It is measured in data with the standard selections but missing ET < 1 GeV. 42

43 Yields Before b-tagging For all MC, apply: Cross section weights (including heavy flavor corrections) Trigger weights: the trigger efficiency for such an event measured in data. object ID weights: correction factors to take into account differences in data and MC object reconstruction efficiency. Create an orthogonal sample by reversing the tight lepton criteria, and normalize it to N tight fake-l. This is the multijet sample. From N tight real-l, subtract the tt yield. Scale the W+jets MC sample to this amount. This is the W sample. 43

44 The Matrix Method The multijet background yield is estimated by solving the following equations: N loose N tight = Nloose fake l = Ntight fake l Measure εfake-l on data using the same selection except missing ET < 1 GeV Measure εreal-l in three steps: + Nloose real l + Ntight real l Measure ε MC real-l using MC truth and the analysis selection criteria = ε fake l N fake l loose + ε real l N real l loose Measure ε Z-data real-l / ε Z-MC real-l using the EM-ID or Muon ID tools ɛ real l = ɛz data real l ɛ Z MC real l ɛ MC real l 44 njets == 2, all trigs tight/loose ratio ! 2 / ndf / 4 Prob " QCD !.52!.1 electron εfake-l Missing E T [GeV] njets == 2, all trigs tight/loose ratio ! 2 / ndf / 4 Prob.5216 " QCD !.351!.4 muon εfake-l Jovan Missing Mitrevski E [GeV] T

45 Yields Yields Before b-tagging Electron Channel Muon Channel 1 jet 2 jets 3 jets 4 jets 5+ jets 1 jet 2 jets 3 jets 4 jets 5 jets Signals tb tqb tb+tqb Backgrounds t t ll t t l+jets W b b W c c 1, ,45 1, W jj 23,417 5,437 1, ,476 4,723 1, Multijets 1,691 1, Background Sum 27,37 8,22 3, ,816 6,434 2, Data 27,37 8,22 3, ,816 6,432 2, Yields with One b-tagged Jet Electron Channel Muon Channel 1 jet 2 jets 3 jets 4 jets 5+ jets 1 jet 2 jets 3 jets 4 jets 5 jets Signals tb tqb tb+tqb Backgrounds t t ll t t l+jets W b b W c c W jj Multijets Background Sum Data TABLE 7: Yields after selection and before b tagging. Yields with Zero b-tagged Jets Electron Channel Muon Channel 1 jet 2 jets 3 jets 4 jets 5+ jets 1 jet 2 jets 3 jets 4 jets 5 jets Signals tb tqb tb+tqb Backgrounds t t ll t t l+jets W b b W c c 1, , W jj 23,242 5,376 1, ,351 4,665 1, Multijets 1,655 1, Background Sum 26,886 7,845 2, ,476 6,124 2, Data 26,925 7,833 2, ,527 6,122 2, Yields with Two b-tagged Jets Electron Channel Muon Channel 1 jet 2 jets 3 jets 4 jets 5+ jets 1 jet 2 jets 3 jets 4 jets 5 jets Signals tb tqb tb+tqb Backgrounds t t ll t t l+jets W b b W c c W jj Multijets Background Sum Data TABLE 8: Yields after selection for events with no b-tagged jets. 45

46 CDF s DPF Results, shown by Bernd Stelzer Compatibility of the New Results Compatibility of the two new results? Performed common pseudo-experiments Fitting EPD and LF discriminants Correlation among fit results: ~53% 6% of the pseudo-experiments had a difference in fit results at least as bad as the difference observed in data The results we observe in the data are compatible at the ~6% level 36 46

47 Measuring Heavy Flavor Fraction in Data Scale Factor α to Match Heavy Flavor Fraction to Data 1 jet 2 jets 3 jets 4 jets Electron Channel tags 1.53 ± ± ± ±.4 1 tag 1.29 ± ± ±.2.69 ±.6 2 tags 1.71 ± ± ± 3.5 Muon Channel tags 1.54 ± ± ± ±.2 1 tag 1.11 ± ± ± ±.5 2 tags 1.4 ± ± ± 2.8 TABLE 21: Scale factor α for the W b b and W c c yields to match the data in each jet bin, for tag, 1 tag, and 2 tag samples. The uncertainties are statistical only. 47

48 Background Sample Optimization Different W+jets backgrounds have different processes, so ideally, for P(x B) we would use a background model that well describes the data. Alternately, we can treat it as optimization: P (x B) = w W bb P (x W bb) + w W cg P (x W cg) + w W gg P (x W gg) We optimized the discriminator by assigning different weights: weight per background wbb wcb wgg Electron 1tag Electron 2tag Muon 1tag Muon 2tag 1 48

49 Cross Checks tb Discriminant tb Discriminant - Ht < 175: e+µ w/ =2 Jets and Tags Combined KS: 1 Data tb tqb Wbb wcc Wjj QCD tt! ll tt! l+jets tq Discriminant - Ht < 175: e+µ w/ =2 Jets and Tags Combined KS: tq Discriminant Data tb tqb Wbb wcc Wjj QCD tt! ll tt! l+jets tb Discriminant - Ht > 3: e+µ w/ =2 Jets and Tags Combined KS:.718 Data tb tqb Wbb wcc Wjj QCD tt! ll tt! l+jets.5 1 tb Discriminant tq Discriminant - Ht > 3: e+µ w/ =2 Jets and Tags Combined KS:.429 Data tb tqb Wbb wcc Wjj QCD tt! ll tt! l+jets.5 1 tq Discriminant