New Results from CDF Single Top Searches Bernd

New Results from CDF Single Top Searches Bernd Stelzer UCLA on behalf of the CDF Collaboration � 2006 Joint Meeting of Pacific Region Particle Physics Communities Honolulu, Hawaii, November 1 st 2006

Outline 1. Top Quark Production at the Tevatron 2. Motivation for Single Top 3. The Experimental Challenge 4. Analyses Techniques at CDF • Likelihood Function Analysis • Neural Network Analysis • Matrix Element Analysis (NEW) 5. Results with 695/pb and 955/pb 6. Conclusions 2

The Tevatron Collider • Tevatron is a proton-antiproton collider with ECM=1. 96 Te. V • Only place where top quarks are produced • ~1/fb for analysis (good silicon), >= 1. 3/fb being processed! Cross Sections at s = 1. 96 Te. V 3

Top Quark Production at Tevatron • We are in the eleventh year since top discovery! • No evidence for single top yet! • The challenge is the large W+jets background! All cross-sections at Mtop=175 Ge. V/c 2 NLO rved Obse ! 1995 NLO = 6. 7 pb ed! Want ? /7 2006 s-channel = 0. 88± 0. 07 pb NLO t-channel = 1. 98± 0. 21 pb B. W. Harris, E. Laenen, L. Phaf, Z. Sullivan, S. Weinzierl hep-ph/0207055 (2002) 4

Motivation 5

Single Top within the Standard Model • Cross section is proportional to |Vtb|2 ü Single top allows direct measurement ü No assumption about unitarity of CKM matrix (using unitarity we know: Vtb = 0. 99) (S. Willenbrock, Rev. Mod. Phys. 72, 1141 -1148) • Source of ~100% polarized top quarks ü Test W-t-b coupling (V-A) (G. Mahlon, hep-ph/9811219) cos (lepton, d-quark) in top frame • Important background to low mass Higgs (m. H<130 Ge. V) WH ü Single top analysis is a benchmark for the WH analysis, WH ~ 1/10 Single Top 6

Single Top Beyond the Standard Model • Single top rate can be altered due to the presence of new Physics - Heavy W’ boson, charged Higgs H+, Kaluza Klein excited WKK (s-channel signature) - Flavor changing neutral currents: t-Z/γ/g-c couplings (t-channel signature) - 4 th generation of quarks • Given speculation that the top quark may play a special role in electroweak t (pb) • s-channel and t-channel have different sensitivity to new physics symmetry breaking, studying the top quark’s electroweak production is important! s (pb) Tait, Yuan PRD 63, 014018(2001) 7

Experimental Challenge 8

CDF II Detector ■ Silicon tracking detectors ■ Central drift h = 1. 0 chambers (COT) ■ Solenoid Coil h = 2. 0 ■ EM calorimeter ■ Hadronic calorimeter h = 2. 8 ■ Muon scintillator counters ■ Muon drift chambers ■ Steel shielding 9

Single-Top Signature at CDF Event Selection: • 1 Lepton, ET >15 Ge. V, | |< 2. 0 • Missing ET (MET) > 25 Ge. V • 2 Jets, ET > 15 Ge. V, | |< 2. 8 • Veto Fake W, Z, Dileptons, Conversions, Cosmics • At least one b-tagged jet, (secondary vertex tag) • Single top mostly in the W+2 jets bin • W+1 jet is dominated by W+jet (S/B~1/72) • W+3 jets is dominated by ttbar (S/B~1/28) (investigating gain in sensitivity) 10

Bottom Quark Tagging Secondary Vertex Tagging • Use long lifetime of B hadrons: c ~450 m + large boost from top decay B hadrons travel Lxy~ 3 mm before decaying with large charged track multiplicity Electron • Signature of bottom quark decay is a displaced secondary vertex CDF W+2 jet Candiate Event: Close-up View of Layer 00 Silicon Detector Jet 2 • Tagging efficiency per jet ~40% 12 mm Jet 1 Run: 205964, Event: 337705 Electron ET= 39. 6 Ge. V, MET = 37. 1 Ge. V Jet 1: ET = 62. 8 Ge. V, Lxy = 2. 9 mm Jet 2: ET = 42. 7 Ge. V, Lxy = 3. 9 mm 11

The Experimental Challenge Number of Events / 955 pb-1 Single Top Background S/B S/ B W(l ) + 2 jets 74 15500 ~1/210 ~ 0. 6 W(l ) + 2 jets + b-tag 38 540 ~1/15 ~ 1. 6 W(l ) + 2 jets + b-tag + discriminant 21 67 ~1/3 ~ 2. 6! • Single top search suffers from large amount of W+jets backgrounds • b-tagging is essential for the analysis to improve signal purity • The use of multivariate analysis techniques to distinguish signal from background a good understanding of background is key! 12

Summary of Backgrounds W+HF jets (Wbb/Wcc/Wc) Top/EWK (WW/WZ/Z→ττ, ttbar) • W+jets normalization from data and heavy flavor (HF) fractions from Alpgen Monte Carlo • MC normalized to theoretical cross-section Non-W (QCD) b tt Di • Fit low MET data and extrapolate into signal region non-W Z/ • Multijet events and jets with semileptonic b-decay Wbb W+HF jets (Wbb/Wcc/Wc) Mistags Wcc Wc • W+jets normalization from data and heavy flavor (HF) fraction from MC Mistags (W+2 jets) • Falsely tagged light quark or gluon jets • Mistag probability parameterization obtained from generic jet data 13

Signal and Background Event Yield CDF Run II Preliminary, L=955 pb-1 Event yield in W+2 jets s-channel 15. 4 ± 2. 2 t-channel 22. 4 ± 3. 6 tt 58. 4 ± 13. 5 Diboson 13. 7 ± 1. 9 Z + jets 11. 9 ± 4. 4 Wbb 170. 9 ± 50. 7 Wcc 63. 5 ± 19. 9 Wc 68. 6 ± 19. 0 Non-W 26. 2 ± 15. 9 Mistags 136. 1 ± 19. 7 Single top 37. 8 ± 5. 9 Total background 549. 3 ± 95. 2 Total prediction 587. 1 ± 96. 6 Observed Single top hidden behind background uncertainty! Makes counting experiment impossible! 644 14

Analyses Techniques 15

$Neural Network Extension to b Tagging • A large fraction of backgrounds are W+charm$

Neural Network Extension to b Tagging • A large fraction of backgrounds are W+charm jets and Mistags! • Distinguish b-quark tags from charm / mistags using a Neural Network trained with secondary vertex information W + 2 jet events with ≥ 1 b-tag – Applied to b-tagged jets with secondary vertex – 25 input variables: Lxy, vertex mass, track multiplicity, impact parameter, semilepton decay information, etc. . . • Good separation! • Network output is used as continuous input variable in all multivariate single top analyses 16

Multivariate Analysis Techniques • Multivariate Likelihood Method – Based on collection of signal and background Monte Carlo distributions – Multiply probability densities for each variable in the signal templates, divide by sum of probability densities in signal and background • Neural Network – Train artificial neural network on distributions of signal and background Monte Carlo events – Maps correlated input distributions to continuous output distribution between -1 (background) and +1 (signal) • Matrix Element Technique – For each candidate calculate an event probability d / for signal and background hypothesis – Build discriminant based on event probabilities • All analysis techniques construct a final discriminating variable which is evaluated for signal/background Monte Carlo and fitted to the data 17

The Likelihood Function Analysis § Multiply probability densities for signal input variables, and divide by sum of probability densities in signal and background • i: variable index, k: sample index (s or t) ji: histogram bin • Four background classes used: Wbb, tt, Wcc/Wc and mistags pik=Normalized bin-content t-channel LF Variables: • total transverse energy: HT • Ml b (neutrino pz from kin. fitter) • Cos (lepton, light jet) in top decay frame • Qlepton* untagged jet aka Qx. Eta • mj 1 j 2 • log(MEtchan) from MADGRAPH • Neural Network b-tagger • LF=0. 01 for double tagged events s-channel LF Variables: • Ml b • log(HT* Ml b ) • ET(jet 1) • log(MEtchan) • HT • Neural Network b-tagger 18

Input Variables to Likelihood Function Analysis 19

Input Variables to Likelihood Function Analysis II 20

Likelihood Function Discriminants • Unfortunately, there is no single ‘golden’ variable to do the single top search t-channel s-channel • Combining information from several ‘input variables’ in likelihood function discriminant is powerful • Both, s-channel and the t-channel likelihood function discriminants show deficit in signal region! 21

Likelihood Function Results Best fit Separate Search: Best fit Combined Search: 95% upper limit on combined single top cross section Current result excludes models Beyond Standard Model 95 s+t channel Expected 2. 9 pb Observed 2. 7 pb Note: Expected limit assumes no single top 22

Neural Network Analysis - Combined Search • Single Neural Network trained with SM combination of s- and t-channel as signal • 14 Variables: top and dijet invariant masses, Qlx q, angles, jet ET 1/2 and j 1+ j 2, W-boson , lepton p. T, kinematic top mass fitter quantities, Neural Network b-tag output etc. . Current result using 695/pb (update with 955/pb expected shortly!) Yield Estimate [695/pb]: Single-Top: 28± 3 events, Total Background: 646± 96 events 23

Neural Network Analysis - Separate Search t-channel W+heavy flavor s-channel ttbar • Two NN’s trained separately for s-channel and t-channel (similar variables) 24

Neural Network Analysis - Results Best fit separate search: Best fit combined search: Channel s+t-channel s-channel Expected 95% C. L. Limit 5. 7 pb 4. 2 pb 3. 7 pb Observed 95% C. L. Limit 3. 4 pb 3. 1 pb 3. 2 pb Note: Expected limit assumes single top at Standard Model rate 25

Matrix Element Method Single Top kinematic quantities: § 2(initial) + 12(final) = 14 degrees of freedom § Assume leptons and angles well measured § 3(l)+4(angle)+3(Pin=Pfin)+1(Ein=Efin) = 11 constraints § 14 – 11 = 3 integrals => Integrate over neutrino pz and jet energy of both jets. Event probability for signal and background hypothesis: Input only lepton and 2 jets 4 -vectors! Leading Order matrix element (Mad. Event) Integration over part of the phase space Φ 4 W(Ejet, Epart) is the probability of measuring a jet energy Ejet when Epart was produced Parton distribution function (CTEQ 5) 26

Transfer Function Full simulation vs parton energy: Double Gaussian parameterization: E pa rto n E jet Double Gaussian parameterization: where: E = (Eparton–Ejet) 27

Event Probability Discriminant (EPD) • We compute probabilities for signal and background hypothesis per event Use full kinematic correlation between signal and background events • Define ratio of probabilities as event probability discriminant (EPD): Note: Neural Network b-tagger is used as b-jet probability: b 28

Cross-Checks in Data `Control Regions • Validate method using data without looking at single top candidates • Compare the Monte Carlo prediction of the shape of the discriminant to various control samples in data • W+2 jets data (require no b-tagged jet) CDF Run II Preliminary Dilepton+2 jets Lepton+4 jets • b-tagged dilepton+2 jets data (99% ttbar) • b-tagged lepton+4 jets data (85% ttbar) 29

Input Variables to Matrix Element Analysis • Input to the Matrix Element Analysis are the measured four-vectors of the Lepton, Jet 1 and Jet 2 in the W+2 jets data (>=1 b-tagged jet) Lepton Jet 1 Jet 2 30

Look at Data • Matrix Element analysis observes excess over background expectation • Likelihood fit result for combined search: • Hypothesis test based on CLs method 31

Hypothesis Test We use the CLs Method developed at LEP L. Read, J. Phys. G 28, 2693 (2002) T. Junk, Nucl. Instrum. Meth. A 434, 435 (1999) http: //www. hep. uiuc. edu/home/trj/cdfstats/mclimit_csm 1/ CDF Run. II Preliminary, L=955 pb-1 Define Likelihood ratio test statistic: b s+b Most sensitive bins Median p-value = 0. 6% (2. 5 ) Observed p-value = 1. 0% (2. 3 ) 32

Single Top Candidate Event Central Electron Candidate Charge: -1, Eta=-0. 72 MET=41. 85, Met. Phi=-0. 83 Jet 1: Et=46. 7 Eta=-0. 61 b-tag=1 Jet 2: Et=16. 6 Eta=-2. 91 b-tag=0 Qx. Eta = 2. 91 (t-channel signature) EPD=0. 95 Jet 1 Run: 211883, Event: 1911511 Lepton Jet 2 33

Qx. Eta for Candidate Events in Signal Region Look for signal features (Qx. Eta) in signal region 4) EPD>0. 95 3) EPD>0. 90 2) EPD>0. 80 1) EPD>0. 60 34

Qx. Eta Distributions in Signal Region 1) 2) 3) 4) 35

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

Conclusions • Search for Single Top is an exciting challenge! • We developed three powerful analysis techniques at CDF • New results with 695/pb and 955/pb are promising! • We are on the verge of being sensitive to a combined s+t channel single top signal with an expected sensitivity of 2 - 2. 5 per analysis! Technique s+t cross-section Expected p-value Observed p-value Likelihood Function (955/pb) 0. 3(+1. 2/-0. 3)pb 2. 3% 51. 3% Neural Network (695/pb) 0. 8(+1. 3/-0. 8)pb coming soon Matrix Element (955/pb) 2. 7(+1. 5/-1. 3)pb 0. 6% 1. 0% • Likelihood Function and Matrix Element results consistent at the ~6% level • Plan to combine all three analyses • Looking forward to analyzing more data! 37

Backup Slides Backup 38

Independent Compatibility Study • Perform common pseudo-experiments • Compute reference (averaged) fit result for each pseudo-experiment • M = (w 1* 1 +w 2* 2 )/(w 1+w 2) • Then compute | 1 -M|/e 1 and see how many times this is worse than what we see in data • Result is 6% of the time! – correlation 0. 532 – err 1 average 0. 490 – err 2 average 0. 527 – Probability 1 5. 91 – Probability 2 6. 08 • Similar result by throwing correlated Gaussian random numbers (I. e. fit results) 39

• Divide the 2 D (ME-LF) discriminant space into 4 Regions LF Simple 2 D ‘Combination’ Region 1) "background-like" contains events with EPD<0. 9, LF<0. 9 Region 2) "background/LFsignal-like" contains events with EPD<0. 9, LF>0. 9 Region 3) "background/MEsignal-like" contains events with EPD>0. 9, LF<0. 9 Region 4) "signal-like" contains events with EPD>0. 9, LF>0. 9 2 = 4. 55/4 P=33. 7 Signal Hypothesis Preferred 2 = 3. 37/4 P=49. 8 40

Correlation of Discriminants Correlation between Likelihood Function and Matrix Element Analysis • • s-channel signal: 37. 3% t-channel signal: 65. 1% ttbar: 43. 6% Wbbbar: 53. 6% Wccbar: 59. 1% Wc: 62. 3% total expected: 50% data: 55. 2% 41

Pseudo-Experiments with Features of Results in Data Matrix Element: 1: Likelihood Fkt: 2: = FIT/ SM Require: a) 0. 9 < 1. 1 b) ( 1 - 2 ) > 2 e 1 42

Information used by the Neural Network B-tagger Use NN-btagger output as b-jet probability b: b = 0. 5 * ( NNout +1 ) 43

Likelihood Function 2 D Templates 44

Event probability discriminants nt iscrimina d l e n n a s-ch criminan l dis t-channe t • Overall good separation of signal from background • For ‘combined search’ define: d combine ant discrimin • Trade less good separation for higher signal rate 45

Transfer Function Tests • Jet energies corrected up to Level 5 • Select Jets which are matched to partons (bquarks here) • • Use double Gaussian parameterization (tails) • Transfer Functions act like ‘single top specific corrections’ Compare Transfer Functions for different slices of Epar 10 < Epart< 60 Ge. V 60 < Epart< 80 Ge. V 80 < Epart< 100 Ge. V 100 < Epart< 120 Ge. V 120 < Epart< 150 Ge. V 150 < Epart< 180 Ge. V d. E = (Epart – Ejet) distributions in slices of Epart 46

Including Systematic Uncertainty (in Likelihood) Likelihood Function (CDF 7106): Expected mean in bin k: βj = σj/σSM parameter single top (j=1) W+bottom (j=2) W+charm (j=3) Mistags (j=4) ttbar (j=5) k = Bin index i = Systematic effect δi = Strength of effect εji± = ± 1σ norm. shifts κjik± = ± 1σ shift in bin k ®Correlation between Shape/Normalization uncertainty included (δi) ®Profile Likelihood with respect to all nuisance parameters 47

Sources of Systematic Uncertainty CDF Run. II Preliminary, L=955 pb-1 Systematic (-1 /+1 ) s-channel t-channel All Single top Shape Variations Jet Energy Scale -1. 4% / 1. 3% -2. 4% / 1. 8% -2. 0% / 1. 6% Initial State Radiation 1. 1% / -2. 0% 2. 6% / 2. 0% / 0. 3% Final State Radiation 1. 3% / 1. 4% 3. 4% / 2. 2% 2. 6% / 1. 9% Parton Dist. Function 1. 0% / -0. 6% 1. 7% / -0. 3% 1. 4% / -0. 4% 1% 2% 1. 6% 6. 1% 7. 8% 7. 4% 6% 6% 6% Neural Net b-tagger N/A N/A Mistag Model N/A N/A Non-W Model N/A N/A Q 2 Scale in Alpgen MC N/A N/A Total Rate Uncertainty 9. 1% 11. 3% 10. 5% N/A Monte Carlo Generator Event Det. Efficiency Luminosity • All rate and shape systematic uncertainties are included as nuisance parameters in the analyses! 48

Kinematic Fitter used in Neural Net and Likelihood Analysis • In the top mass reconstruction we have ambiguities from: • choosing the Pz(ν) solution from W-mass constraint • choosing b quark from top decay (s-channel) • Use a 2 in which Pb, MET, Φv is allowed to float – central values = measured values – uncertainties derived from HEPG comparisons with reconstructed values • Without looking at the b-tag, minimize 2 under four scenarios – 2 choices of which jet is labeled ‘b from top decay’ – 2 neutrino pz solutions 49

CLs and p-values CLs+b= P(Q Qobs|s+b): probability of missing a signal as badly as the data if the signal is really there. CLb = P(Q Qobs|b): Probability of the background looking more signal like than the observed data. If CLs+b < 0. 05, we can reject the s+b hypothesis at the 95% CL. p-value=(1 -CLb) < 1. 35 10 -3 “ 3 ” excess is present. p-value=(1 -CLb) < 2. 87 10 -7 “ 5 ” discovery (1 -CLb) CLs+b 50