1 / 27

QCD Multi-jet background in W+jets and Rjets

QCD Multi-jet background in W+jets and Rjets. Alessandro Tricoli (CERN) o n behalf of W+jets and Rjets groups. SM W/Z-Physics Group Meeting. 25 th February 2013. Overview. QCD multi-jet background to Z+jets is small and under control Not discussed today

milt
Download Presentation

QCD Multi-jet background in W+jets and Rjets

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. QCD Multi-jet background in W+jets and Rjets Alessandro Tricoli (CERN) on behalf of W+jets and Rjets groups SM W/Z-Physics Group Meeting 25thFebruary 2013

  2. Overview • QCD multi-jet background to Z+jets is small and under control • Not discussed today • QCD background to W+jets has shown some unexpected features in both electron and muon channels • Far larger than in 2010 data analysis • Muon channel has as much or higher background than the electron channel • Poor fit quality and some model dependence in electron channel • Discrepancy wrt W inclusive analysis results • A lot of effort in W+jets and Rjets groups to understand it • Some questions answered, some to be further investigated A. Tricoli

  3. Relevance of QCD background • QCD background is an important source of uncertainty for both W+jets and Rjets (no cancellation as QCD is very small in Z+jets) • While Z+jets events are very pure, W+jets events are not, does not cancel in Rjets • QCD is dominant background in 0,1,2,3 jet bins, up to ˜15% of data sample • In 2010 QCD multijet uncertainty was 2nd or 3rd of all syst’s for W+ ≥4 jets (11-20%) • How sure are we that we are measuring features of W+jets in all kinematic regions and not its background? With no bias? • Need to have a precise and robust estimation of the background level to reduce uncertainty and produce accurate cross-section measurements • In 2011 JES has been reduced so must QCD uncertainty • JES 4-5% in W+1 jet, mostly cancels in Rjets(≤1% for 1 jet) • Need precision ≤30% (W+1) and ≤10% (R+1) not to be dominated by QCD • 2011 W+jetshas had some issues in both electron and muon channels • Why multi-jet background ˜3-5x larger in 2011 than 2010? • Why multi-jet background level similar in electron and muon channels? A. Tricoli

  4. 2010 vs 2011 - electrons • Large increase in the QCD fraction between 2010 and 2011 • Same template definition in both years (and same lepton thresholds in both years for this comparison) • Different electron identification tight (2010) -> tight++ (2011) • Different primary and supporting triggers in 2010 and 2011 • Is factor 3-5 difference due to pileup? • Same kinematric selection in both years • MET cut is applied to both fractions A. Tricoli

  5. 2010 vs 2011 - muons • Large increase in the QCD fraction between 2010 and 2011 • The same isolation cut is used in both years • Is difference of 3x-5x between 2010 and 2011 due to the broadening of the MET distribution with more pile-up? MET cut is applied to both fractions A. Tricoli

  6. Technique – electrons (I) • Wen+jetsmethod (similar to W inclusive analysis): • QCD template from data: • use OR of three prescaled loose(1) triggers: • EF_e20(22,22vh)_loose, • EF_e20(22,22vh)_looseTrk • EF_e20(22,22vh)_loose1 • select events with loose electrons passing track quality cuts and failing tight • anti-isolation applied • QCD normalization: • fitting MET • QCD template from data + Signal and other BKGs from MC • Fit done in exclusive jet bins up to last bin which is inclusive • Fit done in two different data-periods: D-K, L-M • Templates extracted for D-K and L-M periods in data and MC A. Tricoli

  7. Technique – electrons (II) • QCD Template definition • Electron identification cut reversal is the same as W inclusive analysis • Pass loose and track quality cuts, • Fail tight, ignoring Conversion cut, Cluster-Track Matching cut in Phi and Tight in Eta • Anti-isolation (to reduce signal contamination) is different • W+jets uses reversal of isolation cut used in signal selection: • not isEiso98Etcone20 || not isEiso97Ptcone40 • W inclusive uses EtCone30/Et > 0.2 constunsigned inttemplate_mask_pass = (0               | 0x1 << egammaPID::TrackPixel_Electron               | 0x1 << egammaPID::TrackSi_Electron               | 0x1 << egammaPID::TrackMatchEta_Electron); constunsignedinttemplate_mask_fail = (egammaPID::ElectronTight               & ~(egammaPID::ElectronLoose)               & ~(0x1 << egammaPID::ConversionMatch_Electron)               & ~(0x1 << egammaPID::TrackMatchPhi_Electron)               & ~(0x1 << egammaPID::TrackMatchEtaTight_Electron)); ((el_isEM->at(qcd_i) & egammaPID::ElectronLoose) == 0)  ((el_isEM->at(qcd_i) & template_mask_pass) == 0) && ((el_isEM->at(qcd_i) & template_mask_fail) != 0)) A. Tricoli

  8. Technique – Comparison with W inc. - electrons • W+jets fit results different from W inclusive: • larger background and worse fit chi2 • A lot of effort put in comparing our results • Cutflow, distributions, MC’s samples, binning, fit range code checks etc. • Coarser binning choice in W inclusive (2 GeVvs 5 GeV) improves chi2 • Reduced fit range as in W inclusive improves fit quality • Missing supporting trigger reduced QCD template statistics in W+jets analysis • Differences in QCD template definition and W MCdataset (small effect) W+jets group Winclusive group • With same binning, MC DS, fit range and triggers we find consistent background fractions for ≥ 0 jets, but chi2/dof is still different (31 vs 3) • differences may be due to different calibration? • W incl. cuts out a small region 1.6 < |η| < 1.7 A. Tricoli

  9. Robert King Addendum • Source of disagreement on χ2 between between W inclusive and W+jets • TFractionFitterdoes not look in the bin error, instead looks at bin content assumes error is √(N) • Should scale N and error such that the fractional error stays the same while the absolute error changes to be equal to √ (N) • This is what Inclusive group does => by applying this method we get lower χ2 • Remaining sources of (small) disagrement (<4%) - do we need to pursue this further? Should we make these changes in W+jets,Rjets analyses? • rescaling W pT • cutting out 1.7>|eta|>1.6 • different calibration package • different background MCs W incl. W+jets A. Tricoli

  10. Technique – Fit stability - electrons • Fit Stability and Optimisation: • QCD fit results have been checked with different • Anti-isolation definitions: etcone20,30,40, ptcone40 / ET >0.20 • W contamination in control sample reduced by 3x or more with anti-isolation • After anti-isolation ≤1% W contamination with any anti-isolation choice • (AlpgenWenu+jetsthrough QCD selection) • Etcone has lower W contamination • QCD fraction with and without anti-isolation changes by 1.5% at most • Both QCD fraction, EW SF and Chi2 are little affected by different isolations • Etcone has EW SF and chi2 closer to 1 • Identification cut reversal • Different combinations of isEM bits to fail or pass • Small effects on QCD background fractions A. Tricoli

  11. Background composition - electrons • Since backgrounds are of the same level in e and m channels, has background composition in e-channel has changed in 2011? Fakes -> H.F.? • Check assumption of multi-jet background dominated by fakes in electron channel • Separate electron fakes from H.F. real electrons in W(ev)+jets • Use 3 components to fit to the data: • EW+Top and 2 QCD templates • (cut reversal+anti-isolation) with addition of • impact param significance >10 • Impact param significance <10 •  Fit quality improves • Fake electrons are dominating contribution • (still to assess contributions from conversions) A. Tricoli

  12. Alternative fitting variables - electrons • Tried Df between electron and vector sum of jets • Dftemplate gives very different results from MET template leading to much worse agreement with data • Df fit gives suspiciously larger EW SF (>7%) • Method abandoned • Exploring now fit of Isolation, extracting W template • from Z sample • Work in progress, promising preliminary results W+1 Period L-M A. Tricoli

  13. Technique - muons • Wmn+jets method: • QCD template from data: with reverted impact parameter cut • d0 significance > 3 • differently from W inclusive analysis that uses anti-isolation, due to bias in jet kinematics • QCD normalization by fitting MET on data with QCD template + Signal and other BKGs (from MC) • Fit done in exclusive jet bins up to • last bin which is inclusive • No splitting in data-periods • Consistent results by Monica and • Andrew Reversed Iso Reversed D0 • Using isolation leads to a bias in the leading jet pT distribution • Shown here for QCD b-bbar MC A. Tricoli

  14. Technique – Comparison with W inc. - muons • W+jets fit results different from W inclusive: • larger background and worse chi2 • Trying to compare our results • Inclusive group uses range of 0-40 GeV to avoid high-MET region that suffers from large shape differences • Tried Reverse Isolation consistently with W inclusive analysis • 1.5% difference in fraction, no big difference in Chi2 • Using Sherpa: 1.5% difference in fraction, no big difference in Chi2 • Chi2 still very different between W+jets and W inclusive analyses • No MET cut applied here A. Tricoli

  15. Fitting Tools - muons • Errors are about ~1.5x smaller with RooFit than with TFRactionFitter. Central values are comparable • No MET cut applied here • Similar effect seen in Top background fit • is TFractionFitter uncertainty realistic? Aren’t they overestimated? • Problem of fit convergence in Top background fit with TFractionFitter while none with RooFit A. Tricoli

  16. Fitting Isolation - muons • Use Isolation as alternative fitting variable to cross check results with MET Binning was too coarse, so perhaps the fit is not the best. Black = Wjets Yellow = QCD Blue = Fit result • Chi2 is not better than with MET • Fit results vary quite a bit from MET fits • up to 2.7% • Further investigation needed MET cut is applied for both A. Tricoli

  17. Pileup – electrons (I) • Big jump in pileup in Period L and M wrt previous periods • <m>~5, 10 in Periods D-K and L-M • Primary and Supporting (background) electron triggers changed during the year 2011 and not always coherently in terms of selection and rates • data period statistical weight is different in signal and background samples: • D-K weigh more than L-M in background sample than in signal sample • Pileup effect in whole 2011 data sets not reflected in QCD background sample • MET distribution is pileup sensitive • Period dependent fits in electron channel: D-K, L-M (fit quality improves) A. Tricoli

  18. Pileup – electrons (II) • Splitting Data and MC sets in D-K and L-M templates • Fit quality improves (chi2/dof and EW SF closer to 1) • QCD fraction increase by 1%-2% overall • Low QCD fraction in D-K, large in L-M => Pileup Effect !? • Results irreproducible on MC due to small statistics • Used in both W+jets and Rjets as default EW Scale Factors Chi2/dof • EW S.F. are at times still large 1%-12% • indication EW process mis-modeling? QCD Fraction • Chi2/dof not yet 1, especially with 0,1,2 jets • QCD Template mis-modeling? A. Tricoli

  19. MC study– electron channel • MC does not show same data-period dependence as in data, probably due to poor statistics • However well reproduces pileup • Cannot use MC to understand data-period dependence A. Tricoli

  20. Pileup – muons (I) • No need of period dependence fit in Muon Channel since same trigger used for signal and background samples • pileup effects (e.g. broadening of MET distribution) in signal sample is well reproduced by background sample • Oppositely to electron channel, QCD fraction seems to go down in L-M, by a factor ˜3 wrt D-K • Is this because isolation becomes tighter with more pileup? • Still to be explained larger QCD fraction in 2011 than in 2010 Clear period-dependence of isolation distribution (D-K,L-M) QCD control sample Signal sample A. Tricoli

  21. Pileup – muons (II) • Factor up to 3.4x in difference between L-M and D-K • Results to be cross checked No MET cut applied here A. Tricoli

  22. Uncertainties • Electron channel • Preliminary uncertainties, including: • W/Z model dependence (Sherpa vsAlpgen) • Change of fit range • Change of anti-isolation - dominating source • Statistical uncertainty Uncertainties do not cover difference between data-periods • Muon channel • Preliminary uncertainties, including: • Template shape (anti-isolation and changes of cuts) • Change of fit range • Alternative fit variable [Df(l,sumed-jets)] • background dominant systematics on Rjets:˜3% on 0,1,2,3 jets A. Tricoli

  23. On-going Discussion • Ongoing discussion between W+jets and Rjets groups on • should we worry/understand disagreement between 2010 and 2011? • should we worry/understand levels and data-period dependence of background in electron and muon channels? • should we worry/understand remaining disagreement wrt W inclusive analysis? • No evidence of mistakes in 2011 analysis that could lead to such a difference wrt 2010 • Detector levels are in reasonable agreement • “things” have changed between 2010 and 2011 (what?) • Data-period dependence not confirmed by MC but not refuted either • Or maybe 2010 analysis was wrong? • Large chi2/dof and other unknowns will/can eventually translate in larger uncertainties on background estimation A. Tricoli

  24. Conclusions and Outlook • Great progress in the last few week in understanding QCD background in W+jets events – fast pace! • Better understanding now of pileup contribution in the electron channel • Improved agreement with W inclusive analysis results in electron channel • Systematic studies of stability of results undertaken • e.g. Model dependence, change of control sample definition, fitting tools • A few systematics checks remain to finish in electron channel • Trigger bias, optimal control sample definition (e.g. isEM tight++/medium++) • Cross check results with alternative fitting variable: isolation • Reassess W model dependence in fit results • Still a few open questions in muon channel • Source of increase in background fraction in 2011 wrt 2010 • Remaining discrepancies with W inclusive analysis • Debate within W+jets and Rjets group on priorities A. Tricoli

  25. Backup A. Tricoli

  26. W+jets and R+jets Selection (II) object selection • R+jets uses same selections as in W+jets and Z+jets Preselection: -trigger - primary vertex (at least one good vertex with >=3 track ) -MET cleaning and LAr Hole Veto Lepton selection: Z selection: 2 OS leptons, 66 GeV< mll<116 GeV W selection: MET > 25 GeV, mT(W) > 40 GeV Jets selection: Electron Trigger W: e20_medium(D-I), e22_medium (K), e22vh_medium1 (L-M) Z : 2e12_medium(D-I), 2e12T_medium (K), 2e12Tvh_medium (L-M) MuonTrigger: mu18_MG (D-I), mu18_MG_medium (J,M) Electrons: Author 1or 3, Quality flag (OQ&1446 !=0) pT>25 GeV (W),20 GeV (Z) |n|<2.47 (excluding 1.37-1.52) Muons: STACO Combined with cleaning cuts (MCP recommendations), pT>25 GeV (W),20 GeV(Z) |n|<2.4 Electrons: 2 OS medium electrons Muons: 2 OS muons, at least the 1st with d0sign (<3), z0(<10mm), isolation requ. (ptcone20 / pt < 0.1) MET: MET_RefFinal, JES and lepton scale prop with MissingETUtility Muons: 1 muon with d0sign,z0, isolation requ. Electrons: 1 tight++ electron Antikt4TopoEM at EM*JES scale pT > 30 GeV, |y| < 4.4 |JVF| > 0.75 ( for jets with |η| < 2.4) ΔR(jet-lepton) > 0.5 A. Tricoli

  27. Fit Stability Different anti-isolations A. Tricoli

More Related