Measurement on the mass difference between top and anti-top quarks Vikash Chavda 1, Un-Ki Yang 1, Jahred Adelman 2 University of Manchester 1, Yale University 2 ATLAS Top UK, August 15,
Mass Difference? Motivation Event selection How to reconstruct mass difference? Template fitting Systematic studies Final result 2
Top Antitop Mass Difference A mass difference between top and antitop quarks: – Implies CPT violation, thus constraint on CPT Violation raised by a new physics process Currently, there have been measurements made on the ttbar mass difference by D0, CDF and CMS m = m t -m tbar = 0.8 ± 1.8(stat.) ± 0.5(syst.) GeV (3.6fb -1 ) : event by event m = -3.3 ± 1.4(stat.) ± 1.0(syst.) GeV (5.6fb -1 ) : event by event m= ± 0.46(stat.) ± 0.27(syst.) GeV (4.9fb -1 ): using diff. samples 3
Event Selection Semi-leptonic channel: follow the top group selection cuts and have full agreement in acceptance challenge Cuts are: – Exactly one el with E T > 25 GeV and |η| 20 GeV and |η| < 2.5 with single lepton triggers – Remove events tagged as e-mu overlap – Require a primary vertex with N tracks > 4 – MET > 35 GeV for the el channel, OR MET > 20 GeV for the mu channel – M t (W) > 25 GeV for the el channel OR MET + M t (W) > 60 GeV for the mu channel – ≥ 4 jets with E T > 25 GeV and |η| 0.75 – Events with loose bad jets with p T > 20 GeV are rejected – ≥ 2 jet tagged with MV1 weight >
Signal MC Generation – ΔMtop ≠ 0 5 Modification made to Pythia to allow generation of samples in the case that m t ≠ m tbar – Modifications were placed in package PythiaExo_i, and 15 Signal samples ranging with mass differences between -15GeV to 15GeV were produced centrally, in release 17
How to get mass difference Modify Kinematic fitter to allow different mass between top and anti-top quark Use up to leading 5 jets to find best combination of 4 jets Select the best comb. for 6 3 rd term: W mass constraint (leptonic & jet energies) Last two terms: fit ΔM: Δm reco = q(lep)*ΔM (=m(lep) - m(had)) (m t +m tbar )/2 is constrained to be the PDG value Jet and leptons energies allowed to be varied within their uncertainties
Event Flow (Electron) 7
Event Flow (Muon) 8
Data-background Comparison 9 Light Jet PtLeptonic Bjet Jet Pt Lepton Pt Missing Et
Reconstructed Top and Anti-top 10 Anti-top MassTop Mass
Mass Difference 11 Fitted Mass Difference (Combined Electron and Muon Channels)
Signal Template Use a modified Pythia for signal samples (central and private production) – -15GeV to +15GeV (15 samples) Fit a double Gaussian to the reconstructed mass difference from each of the MC signal sample (left) Plot dependencies of variables of Gaussian (mean, sigma) as a function of input mass difference: see templates in backup slides A good linear dependence: 2 nd Gaussian (purple) 12
Background Template 13 Single Top W+Jets QCD Various backgrounds added together (Single Top, W+Jets,QCD) Stop from MC, W+jets from the data-driven method QCD from the data-driven method event estimated from JetElectronQCDModel – Non-ttbar background (8.2%)
Likelihood Use an extended maximum likelihood fit with 3 fit parameters – Expected number of signal events (n s ) – Expected number of background events (n b ) – Fitted mass difference ( ) 14 The fit machinery and parameterizations are tested using PE’s The fit with a Gaussian bkgd constraint is also checked out. PE setup: Draw events from Δm reco histograms 2000 PEs Fit using the LH, take the mean on the fitted mass difference from the 2000 PE results
Sensitivity Expected stat. error in 2-tag for 4.7/fb based on PEs with m=0 GeV sample (left) Expected statistical error (2-tag) plotted as a function of the mass difference (right) 15 Exp. statistical error on m: =0.6 to 0.7 GeV Exp. error dist. for m=0 GeV Exp. error for different m
Systematic 16 Systematic uncertainties applied to both the top and antitop equally, is expected to have a small contribution – Jet Energy Scale, Lepton ID, B Jet Energy Scale But systematics which are related to an asymmetry between top and antitop can be large General procedure – Apply uncertainty to key variables in analysis (vary by ± 1σ ) – Re run full analysis – Get the distribution – Run through 2000 PE’s to get ΔM – The shift with respect to a nominal sample is taken as the systematic Δm reco
Systematic Definitions 17 In addition, the effect from non-zero Δm in single-top production: 20 MeV for Δm = 2.5 GeV (check, not syst. Item)
Check on Parameterisation In the mean of the narrow Gaussian, see large fluctuation in the sub 1 GeV range – Due to low statistics signal samples – Uncertainty on the slope: 25 MeV effect on ΔM Question raised whether the fitter has a sensitivity for a precision measurement at the sub-GeV level 18
Test parameterisation sensitivity for a precision measurement at the sub-GeV level Strategy: Check the parameterization at at Δm =0 GeV using high-statistics Pythia ISR/FSR Avg sample Use high statistics (15M) SM ttbar sample and reweight them to our current signal samples Use high statistics new Pythia samples (0, 1, -1 GeV) Precision at the Sub GeV? 19
Sub GeV sensitivity - ISR/FSR Use mixture of the ISR More/Less sample to create an ISR Average sample – Given that the ISR/FSR systematic is small, ideal sample to test ISR More, ISR Less, average sample All fitted values in our standard double gaussian fit to the templates agree with our parameterised function at ΔM=0 The PE result is Δm= 97 MeV ( probably due to the ISR/FSR syst. effect plus fast vs fullsim) 20
Sub GeV Sensitivity: Reweighting A second method we used was to reweight sample to ±1 GeV, ±0.6 GeV, ±0.3 GeV, and to the Signal Pythia 0 GeV sample to obtain high stats signal samples – The mass of the top and antitop were distributed independently during generation – Reweight top and anti-top mass distributions separately at truth level: for 1 GeV case: reweight top mass to 173 GeV, and antitop mass to 172 Gave, keeping the average mass remains GeV 21 sample Pythia 0 GeV sample
PE results The values in the previous tables are fairly consistent with the input mass values within 50 MeV with an offset of 175 MeV Offset mostly due to difference between Pythia and (NLO, Generator, Parton Shower) plus a possible bias from fitter – 3 high statistic samples will help in the understanding of the offset Measurement from fit will be calibrated to sample – 175 MeV will be subtracted from fitted value as final measurement with additional calibration uncertainty of 50 MeV 22
Syst. Errors on M 23 CMS Syst ± ± ± ± ±
m Result 24 Final measurement is presented Mass difference obtained by fitting to 4.7 pb -1 of 2011 data Final result including the subtraction of the 175 MeV offset
Cross checks on the results Results with default fitting Results with a Gaussian bkgd constraint(30%) Electron vs muon channel: consistent 200 MeV difference is expected based on the Pes Electron Muon 25
Conclusions and Plans The top anti-top mass difference analysis presented: The current sensitivity on the ΔM; – Stat: 610 MeV – Syst: 192 MeV – Achieve great sensitivity Paper written – One cross check left before ATLAS circulation 26
Backup Slides 27
Performance Plots Using ttbar sample, show performance of the Kinematic fitter on the mass difference and hadronic top mass See improvement in distribution when using fitted energy of jets and leptons (black curve) Further improvement is seen when applying cut (pink curve) The cut value 10 was tuned to achieve the best sensitivity 28
Mass Difference? Why? CPT violation What events? tTbar in semi-leptonics How to reconstruct mass difference? Kinematic fitter Template fitting What systematics? Final result 29
Data and MC Samples Data High-pt lepton (e/mu) trigger sample in 2011, 4.7pb -1, periods B-M (the standard single lepton sample the Top group used) MC Samples – ttbar events (dominant): SM MC (Δm=0): signal MC (Δm≠0):??? – Backgrounds Single top (Wt, s chan) : – Single top, tchan : AcerMC+Pythia – W/Z+jets: ALPGEN+Herwig – Diboson: – QCD : data driven method 30
Motivation Top quark has largest mass of all the fundamental particles Only fermion with unsurpressed coupling to electroweak symmetry breaking sector 31 CP non conservation and T violation seen in neutral kaon system, and the test of CPT violation have been explored in neutrino sector Top most precisely measured quark, (assuming top mass equal anti-top mass), but the test of the CPT violation has been just started. With High-statistics LHC data, a great opportunity to search for it top quark.. e e - u d s c b
Mass Difference Reconstruct hadronic and leptonic top masses in semi-leptonic decay channel Take the mass difference between top and anti-top quarks – If positive lepton in final state, the mass difference will be top (leptonic) – anti-top (hadronic) Apply template fitting 32
Motivation The CPT Theorem (Combination of Charge, Parity and Time reversal) states: – Any local theory, which is invariant under Lorentz Transformations and defined by a Hermitian Hamiltonian is said to converse CPT CPT Conservation for particles and antiparticles implies: – Equal masses – Equal lifetimes Any mass difference between a particle and its antiparticle is unambiguous evidence of CPT Violation CPT is a fundamental piece of QFT, on which particle physics is based 33
Top Antitop Mass Difference A mass difference between top and antitop quarks – Implies a CPT violation, thus constraint on CPT Violation raised by a new physics process There have been measurements made on the ttbar mass difference by D0, CDF and CMS ΔMtop =M t -M tbar = 0.8 ± 1.8(stat.) ± 0.5(syst.) GeV (3.6fb -1 ) ΔMtop= -3.3 ± 1.4(stat.) ± 1.0(syst.) GeV (5.6fb -1 ) ΔMtop= ± 0.46(stat.) ± 0.27(syst.) GeV (4.9fb -1 ) Analysis done in Semi-leptonic channel with 2 b-tag events ( 4.7 fb -1 in Rel 17) Note: ATL-COM-PHYS
Previous Measurements 35 Template method, semi-leptonic channel, B-tagged, PRL 106, (2011)
PE results PE results are consistent with 175 MeV in the range of ΔM= +1 to -1 GeV.: the offset: Pythia vs diff. (NLO, PS, Gen) plus a bias from the fit Thus, 175 MeV offset will be applied to the measured value, But what syst error? 50 MeV or any dependence on ΔM? – 3 high statistics samples (0,+1,-1 GeV) will help here 36
Data and MC Samples Data Using data recorded in 2011, 4.7pb -1, periods B-M MC (AFII, FS, mc11b, mc11c) Single top (Wt, s chan) -> Single top, tchan -> AcerMC+Pythia Diboson (WW,WZ,ZZ) -> Z+bb -> Alpgen+Herwig Nominal ttbar -> ISR More/Less -> AcerMC+Pythia Parton Shower Model -> Powheg+Pythia(Herwig) Signal (FS, mc11c) 15 samples -> Pythia 37
Jet Energy Scale Uncertainty 38 Evaluated: – Vary jet momentum in analysis by ± 1σ Result GeV ΔM reco distribution with JES ± σ (left), and ΔM reco distribution after the kinematic fit with JES ± σ (right)
Asymmetry between top and anti-top The identification (ID) of the top and anti-top quarks is based on the lepton charge – B/Bbar oscillation has no effect on the top ID – Mis-charge on the lepton? (muon: prob. < , electron:~0.001) Asymmetry in energy response between top and anti-top quarks – B-jet response: in principle, no different at the generator level, but a possible small difference at the detector level ( different k-P/k+P interaction rate in the calorimeter), but the simulation knows it. – W+/W- response: asymmetry in c/cbar, s/sbar, lepton: the effect will be further reduced due to the W mass constraint – Production asymmetry (t/tbar: different eta distributions due to qqbar process): PDF uncertainty is the smallest one. 39
Asymmetry in JES for b/bbar, c/cbar,s/sbar 40 From FS sample b/bbar asymmetry: % c/cbar asymmetry: < % s/sbar asymmetry: no asymm.
b/bbar asymm.: FullSim and FastSim results for the b/bar asymmetry are consistent each other: different interaction rate of K- vs K+ on proton in calorimeter is not observed. But the Powheg+Pythia samples are consistent with no asymmetry. The half difference is taken as the systematic: 80 MeV Possible asymm. (0.1%) in for c/cbar: 13 MeV
Parton Shower Model Systematic PS samples: Powheg interfaced with Pythia and Herwig – m: GeV for Herwig, and GeV for Pythia – A large difference, 0.40 GeV on m Source of the difference – b/bbar-JES asymmetry Powheg with Herwig (AF-II): 0.28% Powheg with Pythia (AF-II): 0.09% The difference 0.21% can only example 0.12 GeV on m, but not able to explain a whole difference 42
Parton Shower Model A new Powheg+Pythia sample generated (using Perugia 2011 tune) Produced on the basis of top mass studies which found a big discrepancy with data due to the Pythia tune. terialId=slides&confId= Using this sample, the PS systematic reduces dramatically from 400 -> 56 MeV! Sample is currently being validated by many analysis to check viability for Top2012 deadlines 43
Sanity Checks No bias (Residuals are consistent with zero) Pull widths are consistent with unity 44
Sub GeV Sensitivity - ISR/FSR ISR/FSR average sample agrees with all of the parameterized value at Δm=0 (plots and PE result) – The PE result is Δm= 97 MeV ( probably due to the ISR/FSR syst. effect plus fast vs fullsim) 45