410 likes | 422 Views
Preparation for top physics at the LHC. Top Physics at ATLAS day one. 1) Top properties and basic SM physics at s = 14 TeV : Estimate of σ top : interesting even if error is large ( first measurement at s = 14 TeV Start to tune Monte Carlo
E N D
Preparation for top physics at the LHC Pamela Ferrari
Top Physics at ATLAS day one 1)Top properties and basic SM physics at s = 14 TeV : • Estimate of σtop : interesting even if error is large ( first measurement at s = 14 TeV • Start to tune Monte Carlo • Measure top mass feedback on detector performance 2)Understand/calibrate detector and trigger: tt bl bjj • Light jet energy scale selecting a pure sample of W jj in tt events (< 1%) • b-tag efficiency (~ 5%) • Understand missing energy: using W semi-leptonic decays. 3) Prepared to observe spectacular effects due to new physics ( being prepared to discover spectacular bugs..) • Resonances, MSSM higgses, SUSY, FCNC • Measure differential cross sections (ds/dpT,ds/dMtt) sensitive to new physics (provides also an accurate test of SM predictions) Pamela Ferrari
100 pb-1 100pb-1 Trigger studies/ttbar (I) - An example of trigger studies on top events in lepton+jets channel • “Tag and Probe” Efficiency Correction • Select events in Z → ee sample with 2 e and 70 < MZ < 100 GeV • Calculate efficiency in pT and η regions (separate barrel and endcap) • A = Acceptance (from MC) • eT = Trig Efficiency (from Z→ee) • Ndet = Signal + Background • Nback = From AlpGen W+jets Cross Section * Branching Ratio = 124.3 pb (Expect 12/81 * 830 pb = 123 pb for electron + jets channel) Pamela Ferrari
Trigger commissioning: overlaps • Leptons: • Correlation between trigger menus useful for commissioning of lepton triggers: same approach as will be adopted for the 10^31 menu. • How to assess unbiased trigger efficiencies from data? Can we use the correlation matrix to unfold efficiencies? Pamela Ferrari
Trigger commissioning • What can we expect with different integrated luminosities: • @10 pb-1= verify detection of top get single efficiency number with • 5%-10% accuracy • @50-100 pb-1= we get enough events to see turn on curves. • An important check is to evaluate the absolute fraction muons versus electrons in semileptonic ttbar events: Check that can be done without depending on acceptances etc.. • Verify electron trigger in Z ee:the environment and kinematics are different, at present the differences are within few %, but this has to be verified with data. The isolation of electrons could play a different role. • Commissioning of MET and jets trigger: • By using events triggered by lepton trigger and verify that jet trigger and • MET fired at proper thresholds. Pamela Ferrari
b-jets in top events b-jet properties in top events (Wuppertal) • b-jet energy and angular resolution for different jet algorithms • b-tagging performance in top events, for different jet algorithms B-tagging performance: it is important to check it in top environment ( other methods –system 8- use di-jet events) • ttbar events can provide a ‘pure’ sample of b-jets. • Event-tag counting method • Different methods exploiting tt decay kinematics: • Using mass distributions • Using kinematic fit 2 • Using a likelihood based on kinematic variables Pamela Ferrari
Measuring b: Event/tag counting method • Count events with 0,1,2,.. tagged b-jets, likelihood fit for b, c and tt • Semileptonic (L+J) channel: lepton + ET miss + >=4 jets , 0,1,2,3…b-tagged • New: Di-lepton (LL) channel: two leptons + ET miss +>=2 jets, 0,1,2 .. b-tagged LL channel L+J channel H. Bachacou • Uncertainty so far around 5% relative or 0.03 absolute at b=0.6 • Still to consider: b-jet energy scale, ISR/FSR, top mass, generator dependence Pamela Ferrari
Measuring b: Kinematic/topological selections • Use data to extract background breakdown in different jet ET • Topological method(CERN) - use mbl to region enhanced in b-jets Estimate bkg: shape from a control sample where hadronic side mbjj>200 GeV, and leptonic top jet is anti b-tagged. Estimate flavour composition from signal sample where mbl outside mtop 2 • Kinematic method (Wuppertal)- kinematic fit assuming mbjj =m bl= mtop, select combination with best fit 2 Estimate bkgfrom sample requiring one W jet to be b-tagged. Subtract from signal region at low 2, normalising using high 2 • Select using b-tag on hadronic side W jets: ET>40, 20 b-veto (weight < 3) Hadronic side b-jet: ET>40 b-tag (weight>3) Lepton: ET>20 Leptonic side b-jet: ET>20 No tag requirement ETmiss>20 Pamela Ferrari
B-tag efficiency: topological method • Estimated statistical error scaled to sample size of 100 pb-1 • At b=0.6 (weight cut at ~0.8), error is 0.041 or 7% relative for 100 pb-1 • Systematics are small since the method is data driven Scaled to L=100 pb-1 CERN Pamela Ferrari
Cross-section determination • Cross-section measurements • This measurement is relevant from the very beginning even if it is extracted with fairly large errors, since it is the first time the ttbar cross-section is extracted at the LHC ECM. • Estimation of cross-section in semi-leptonic channel to ~20% from the first 100 pb-1 of data. • Methods to extract cross-section: • 1) fitting of distributions, extraction of background from fit on the tail • 2) Counting experiment, normalising W+jets backg with Z+jets • 3) Counting events in 1,2,3,4, jet bins events and subtract W background in each bin. • CDF top discovery method. Pamela Ferrari
Cross-section commissioning analysis: leptons+jets channel with • Define acceptance and efficiency • Acceptance: electrons within |h| < 2.5 • Should split efficiency into: trigger, reconstruction and selection efficiencies. • For now: all e’s and A merged into one efficiency. • Acceptances have to be calculated from MC • Efficiencies can be calculated from MC. • Final efficiencies also calculated from Data • At present no data: split both the top signal and the background sample in 2/3 and 1/3. • The 2/3 sample is our “data” sample. • The 1/3 sample is our “Monte Carlo”: used to extract efficiencies, or to subtract background Pamela Ferrari
Standard semi-leptonic selection non perfect detector simple selection:no b-tag • Selection A: • Trigger L1, L2, EF • PT(l,n) > 20 GeV • 3 jets with PT > 40 GeV • A 4th jet with PT > 20 GeV • h(lep) < 2.5, h(jet) < 2.5 • Top is reconstructed as the 3-jet combination with highest vector sum PT • Selection B: Same as A plus W-mass cut: • Require that of the 3-jets there is a pair whose mass is within 10 GeV of W-mass Hadronic top=3 jets maximising pT top 3 jets pT> 40 GeV 1 jets pT> 20 GeV Isolated lepton pT> 20 GeV ETmiss > 20 GeV Selection B |mjj-mW| < 10 GeV Pamela Ferrari
100 pb-1 100pb-1 X-section by counting experiment Selection B: after MW cut Selection A INFN Udine (Udine Univ&ICTP) CERN Mt=165.5+-2.2 Mt=167.0+-1.8 S/B = 4 Eff=5.5% S/B =5.8 Eff=2.7% • Improve S/B with b-tagging: factor 10 improvement • Being calculated these days: • Wbb backgorund • Zbb background • single top background Pamela Ferrari
Systematics and cross-section measurement Cross-section selection A: 842.9 ± 2.4% (stat) ± 17.4% (syst) Cross-section selection B: 820.4 ± 3.2% (stat) ± 12.0% (syst) Systematics studies still to be finalised Significance vs Lumi on x-section analysis NIKHEF Pamela Ferrari
100 pb-1 100 pb-1 100pb-1 100pb-1 Understand the detector |h|<1 S/B=6.5 • Extract x-section in central and forward region • We are using medium electrons exploring also loose (tight). • Analysis without Etmiss cut INFN Udine (Udine Univ&ICTP) CERN NIKHEF Pamela Ferrari
8 pb-1 100pb-1 Stream test exercise InclEle 8pb-1 Data streams:datasets of raw data categorized by physics content (a group of trigger signatures). Streaming test:there are five stream definitions labelled: Electron, Muon, Tau and missing energy, Photon, Jet and Sum ET. • In streaming test events are written to different datasets according to their trigger signatures. • The stream test data represent the output of 10 half-hour runs in early data-taking. INFN Udine (Udine Univ&ICTP) CERN The method: • Same selection as x-sec analysis • Use reconstructed Z’s to identify number of Z+4j, Z+5j events. • Use to get number of W+4j,W+5j. • Use MC to calculate reco. efficiency of background and signal 873pb ±80 pb (sys) ±68 pb (stat) Pamela Ferrari
8 pb-1 100 pb-1 8 pb-1 100pb-1 100pb-1 100pb-1 W mass contraint in streamtest data • Use W mass constraint to commission the detector and detect miscalibrations! • The W peak is already visible in stream test data with 8pb-1 • Stability of the fit? INFN Udine (Udine Univ&ICTP) Pamela Ferrari
Streamtest data: pre-commissioning analysis strategy Using CDF top discovery method. • Validate reconstruction and trigger performance • Tag and probe with Z events to measure trigger efficiency • Z events to measure electron energy scale and linearity • Electron ID efficiency from Z before/after IsEM cuts • In real data, use ID variables to estimate fake background • Missing ET and transverse mass in W events to set systematic uncertainty on MET scale • Use electron Stream: select inclusive W eν sample. Baseline analysis chosen to provide clean W sample (Cut on W transverse mass) • Exactly one electron: pT > 25 GeV, isEM&0x7FF = 0, • |η| < 2.4 (excluding [1.37,1.52]). • Missing ET > 25 GeV • MT (W) > 45 GeV • Jets: pT > 25 GeV, |η| < 2.5, dR(e, jet) > 0.3. • Unlike CDF can do without b-tagging given good S/B. Pamela Ferrari
Streamtest exercise a la CDF • counting experiment: # W+>4 jet events • Measure # of events in W+0 jet, W+1 jet bins and use this to normalize W x-section • Extrapolate or fit MC to predict LBL • systematic from MC dependence of ratio (50% uncertainty on MC bckg prediction) • Single top contribution non negligible! @15 pb-1 • σ = 941 ± 92 pb counting • σ= 896 ± 181pb fitting Pamela Ferrari
15 pb-1 100pb-1 Missing ET validation with data • After correcting electron energy-scale: Look at transverse mass of W sample (ET>20, MET>25) • Comparison among data and MC to assess systematics LBL W+ 0 Jet Sample W+ 1 Jet Sample Shift: 1.026 ± 0.01 Shift: 1.011 ± 0.03 Pamela Ferrari
Di-leptonic • Cut based analysis: • 2 isolated, opposite charged leptons pT>20 GeV , • at least two Jets, no b-tagging • |η|<2.5 for all visible objects • Veto on M(ll) 85-95GeV • eµ:ETMiss>20 GeV, ee and µµ: ETMiss>35 GeV • In addition also a likelihood analysis exists Bonn+Stockohlm Pamela Ferrari
JES calibration 2bjet+≥2 light j PT jets 40 GeV Rmin(J1;J2) Rmin(W;b) Select pure sample of W’s. • Template method: ensemble test with scale variations (Saclay) • Template histograms of mjj=mW with different E scales a and relative E resolutions b (PYTHIA tt events) • compared with MC@NLO tt events. Fit each template histogram to mjj in the « data », find best c2 • Iterative method (Clermont): • find the ratio ai = MWPDG/Mi where Mi=peak value of dijets forming W the for the Energy interval i. • Reiterate procedure few times, with new calibration (converges after few iterations) Pamela Ferrari
1 fb-1 100pb-1 JES calibration The 2 methods agree, @ 1 fb-1 (syst=stability of the result when changing the combinatorial background, effect of pile-up has still to be checked ) Saclay Stat err~ 0.4% Syst err <1% Mtop After calibr. Before calibr. • stat error scaling with Lumi: MW MW bias from jet selection Pamela Ferrari Smaller errors if we consider only a global JES (not in function of Ejet)
Top mass at LHC Many analyses performed in the past or being developed recently • Cut based • kinematic fit • Template method, • Matrix Element method, • Analyses with soft muon b-tagging, • high pT top, • Analysis in leptonic final states with J/y • correlation between lepton pT and m_top I will concentrate on the methods that are cut based and that have less dependencies from MC shapes. Essentially the recipe to extract top mass at ATLAS in the early days. Pamela Ferrari
1 fb-1 100pb-1 mT in l+jets channel: with 2 b-tag Saclay Common selection: • 1 isol. Lepton pT <20 GeV/c, ETmiss> 20 GeV/c2 • 4 jets (2 b-tagged) pT> 40 GeV/c Saclay Analysis: cut based • Choice of light jet pair and in-situ energy rescaling using a 2 based on the WPDG mass constraint • Systematic uncertainties: • Jet energy scale • 1% on light jet 0.2 GeV/c2 on mtop • 1% on b-jet 0.7GeV/c2 on mto • ongoing ISR/FSR , b-quark fragm. (*w.r.t. l(e,µ)+jets events) Pamela Ferrari
LEPTONIC TOP Lepton+jets leptonic side • Leptonic side reconstruction: needs a good ETmiss understanding • assumption : ETmiss = pT(n) • b-jets association : 4 possibilities (2 pz, 2 b jets, minimize the difference between the two top masses) • To reduce the impact of FSR a kinematic fit on the full final state can also be performed • c2 based on kinematic constraints (El,j & directions vary within resolution) c2 minimisation, event by event • Mtop fitted in slices ofc2 • estrapolation from linear fit: mtop = mtop(2 = 0) Saclay Pamela Ferrari
90 pb-1 100pb-1 Top mass in semileptonic channel Clermont 2nd method (Clermont): • No in-situ rescaling • Harder cuts to remove background • for 90 pb-1:180 events @ 90 pb-1 Mtop = 174.6 ± 1.1(stat)± 3(syst) GeV/c2 • Analysis with no b-tag ongoing • More stringent cuts against bkg @ 100 pb-1 Mtop = 174.6 ± 0.9(stat) GeV/c2 Pamela Ferrari
1 fb-1 100pb-1 Top mass in fully hadronic channel Sensitive to sources of systematics in different way (mainly due to different background) • Event selection: • 6 jets ( pT > 30 GeV and |η|< 3) • High pT lepton veto: isol. e pT > 80 GeV, isol. m pT > 50 GeV • MET veto:cut events w MET > 60 GeV • Cut events > 8 jets (against combinatorics) • eff = 1.1 %, • S/B ≈ 0.79 obtained • Signif. ≈ 32.0 for 1 fb-1 Victoria Mtop=169.13 ± 0.45 (stat) GeV ( generated with 175 GeV) Pamela Ferrari
Single top • The main bkg at LHC is ttbar (~900 pb) in the final states : lνbjjb, lνblνb, τνb jjb (unlike Tevatron where it is W+jets) • W/Z+jets production • W/Z+lightjets: eνjj+X[~1,160pb], eejj+X[116pb] • W+bb+jets: eνbb+X[5.2 pb] Highest xsec and S/B=0.3 After cut based selection S/B of 0.08 to 0.12 in the best case Pamela Ferrari
Use of MVanalyses: single top At the beginning of data taking top group will use cut based analyses, and only later rely on MVA techniques. • One case where multivariate techniques are very important is the single top because of the high tt background: • Cut based analysis works for t-channel (S/B is quite high) • W-t and s-channels the numbers are not encouraging (high backgrounds). Anyway, the cut-based analyses results will be developed for comparison purposes wrt the MVanalyses. • Common single top pre-selection: • triggers (e25i or e60) or (mu20). • 1 high-pT lepton (pT (e/m) 25/20 GeV/c), • MET 20 GeV • a veto on any other isolated lepton > 10 GeV • at least 2 jets of 30 GeV and at most 4 jets • at least one b-tag above 30 GeV At this point ttbar is by far main background. Pamela Ferrari
1 fb-1 100pb-1 Single top: t-channel Cut analysis: low multiplicity jet events, 1 jet in forward pT>50 |h|>2.5, 1 b jet, centrality • S/B ~ 30% • Significance 18.6 @ 1fb-1 • Significance 5.8 @ 100pb-1. Could be interesting for discovery at 100 pb-1 In case no b-tag have to apply cuts on • mt(lept) & total energy of the events • angular correlations between jets, lepton and jets. MVA analysis: Boosted decision tree improves quite a lot the Cut flow analysis results Pamela Ferrari
Single top: Wt-channel Also benefits from a high cross-section. Cut based selection: • high tt back from which signal differs only by one b-tagged jet • pT of the b-tag jet (above 50 GeV) against W+jets • veto of a second b-tagged jet pT25 GeV/c • 50 mWhad 90 GeV/c • Likelihood based selection: slightly looser cuts+ 4 likelihoods against top pair (“l+jets”,“di-lepton”), t-channel, W+jets Doubles S/B S/√B 3.9 s (stat!!) @ 100 pb-1 S/√B ~12 s with 1 fb-1 LPSC Pamela Ferrari
Single top: s-channel Cut based analysis • Exactly 2 b-jets: against Z/W+(b)jets, • Only 2 jet events: against ttbar • Use angular correlations b-jets-lepton, b-j1 - b-j2 and global ET, M Improves with MVA: • requires knowledge of backgrounds: from data (using side distributions). • 4 different Likelihoods against: W+(b)jets, top pair (“di-lepton”, “l+tau, tau+tau” ,“l+jets) LPSC Pamela Ferrari
Conclusions • I could show just a very limited amount of material today • There are many other activities going on • In ATLAS we are mainly concentrating on Cut based analyses in order to avoid to depend from the shapes of the variables • We concentate in understanding • Our detector • The tools we will need for our analysis: JES, b-tagging • The generators • Methods to evaluate the background from the data • The systematics • At the same time we are keeping an eye on what is being done at Tevatron and are trying to learn from their experience: there is much activity going on to exploit techniques adopted by D0 and CDF. Pamela Ferrari
Back-up Pamela Ferrari
B-tag systematics summary for counting method • Still to consider: b-jet energy scale, ISR/FSR, top mass, generator dependence • Uncertainty so far around 5% relative or 0.03 absolute at b=0.6 Pamela Ferrari
Top mass measurement using a soft-µ b-tagging (1 b-jet tagged) • Soft-muon b-tagging: • Main advantage does not rely on the vertex detector • Main disadvantage low branching ratios • BR(B lnX) 11 % and BR(B c lnX) 10 % (lower pT) • Analysis to be performed with at least 1 b-jet et not 2 b-jets • Optimised on tt events: • e = 6.20 ± 0.03 % • light jet rejection = 143 ± 2 • c jet rejection = 69 ± 2 • tau jet rejection = 138 ± 6 • S/B = 1.24 ± 0.04 • 605 signal events @ 100 pb-1 The top mass value is still measured with a reasonable accuracy (statistical error : 0.5 GeV/c2) W + jets background Pamela Ferrari
Matrix element method (CDF) • Calculate probability density for each event (x vector of measured variables) • Use LO Matrix Element and transfer functions to calculate differential cross-section (Transfer function: probability to measure j when parton level p was produced ) • add signal and back: • Multiply the event probabilities to extract the most likely mass Pamela Ferrari
LPNHE Matrix element at ATLAS: di-leptonic channel • This method gives good results even with a few amount of data • Weak point: huge CPU needed for integration • First steps for integration in ATLAS are being done Pamela Ferrari
Top mass measurement in the l+jets channel using lepton’s p_T Method based on the correlation between <p_T> and m_top • Correlation : • Uncertainties: • Statistical : 7 GeV/c2 @ 1 fb-1 • Systematic (ISR,FSR,PDF,..): • to be estimated, • of the order of 4 GeV/c2 l = 13 ± 1 % Stat.error % luminosity (Athens) Pamela Ferrari
Normalisation of W bckg with Z sample? • The method uses the ratio (“MULTI JET PRODUCTION IN W, Z EVENTS AT p~ COLLIDERS”, F.A. BERENDS, R. KLEISS, W.J. STIRLING et al., Phys. Lett B, Vol 224, p237, 1989) • We obtain the ratio for n=1 to 5 for the Z->ee samples samples 8130-8135 before any selection • We use the inclusive W->enu cross-section calculated by SM group 20.81 nb and we multiply it for the acceptance (0.24%) and luminosity Wtot= 80.24 nb *0.24 • We obtain W+1,2,3,4,5jets events before our selection as W+nj=(Wtot * Z+nj/Ztot)* 100pb-1 • Finally we apply the selection efficiency for each of the W+nj from our MC and we sum the individual contributions to get the total and the background from W->mn. Pamela Ferrari