1 / 19

Single Top search using Likelihood discriminants and JLIP b-tagging

Single Top search using Likelihood discriminants and JLIP b-tagging. Benoit Clément, Denis Gelé – IReS Strasbourg. Overview Analysis procedure compared to the submitted analysis. Selection and Yield calculation. Taggability and b-tagging. Likelihood discriminant method. Limits.

naif
Download Presentation

Single Top search using Likelihood discriminants and JLIP b-tagging

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. Single Top search using Likelihood discriminants and JLIP b-tagging Benoit Clément, Denis Gelé – IReS Strasbourg • Overview • Analysis procedure compared to the submitted analysis. • Selection and Yield calculation. • Taggability and b-tagging. • Likelihood discriminant method. • Limits. σ = 0.88 pb σ = 1.98 pb

  2. Specificites of the analysiscompared to the publication Full pass 2 dataset, processed with Ipanema ~360 pb-1.(Publi : ~230pb-1) JES 5.3 and corresponding b-tagging and TRFs. (Publi : Stradivarius, JES 5.1) JLIP b-tagging. (tight and tight/loose combinations) (Publi : SVT tight) Likelihood Discriminant method to separate signal (s or t channel) from main backgrounds. (W+jets, ttbar) Publi : Neural Network / Decision trees.

  3. Triggers and luminosities Identical to submitted single top analysis Luminosities, good run, bad lbn : top_dq_data (fall2004-pass2-04) Trigger efficiencies in MC : top_trigger (v01-04-07)

  4. Selection cuts • 1 isolated lepton. • 2  # Good jets  4 • Leading jet Pt > 25 GeV , || < 2.5 • Other jets Pt > 15 GeV and || <3.4 • mj_met > 15 GeV, jes_met > 15 GeV • Triangle cuts (next slide). Identical to submittedsingle top analysis Electron channel Muon channel • medium • nseg = 3 • cut on timing cuts (cosmics) • DCA significance < 3 • |z – zvtx| < 1cm • 2 of the track < 4 • R(mu,jets) > 0.5 • Etrack < 0.06*Pt , Ehalo < 0.08*pt • Pt > 15 GeV • EM ID = 10 || 11 • EM fraction > 0.9 • isolation < 0.15 • Hmatrix7 < 50 • |ze – zvtx| < 1 cm • track match spacial chi2 !=-1 • Likelihood > 0.75 • Pt > 15 GeV • |eta| < 1.1 (CC) Reject events with additionnal isolated leptons.

  5. Triangle cuts To remove events where MET comes from badly reconstructed jets or lepton. Qcd Signal (tb) Qcd Signal (tb) Δφ(l,Met) Δφ(l,Met) Identical to submittedsingle top analysis Δφ(j1,Met) Δφ(j1,Met) Muon channel Cuts inΔφ(xx,Met) – Met plane. Δφ(j2,Met) Electron channel

  6. Yields estimate Identical to submittedsingle top analysis Data : just apply full selection and tagging. QCD : built an orthogonal sample enriched in fake isolated leptons by reversing isolation cuts. The number of selected events before b-tagging is normalized to the data (Matrix Method), then tagged. Wjj and Wbb : selected Wjj+Wbb are normalized to the data (MM). Relative normalization is done using Wjj and Wbb NLO cross-sections. Tagging weights are applied. (TRFs) ttbar, VV, signal : Events are passed through selection. Trigger weights, tagging weights and Data/MC id correction are applied. Normalization to luminosity and theoritical x-sections.

  7. Matrix Method Estimate fake / real lepton content of the data sample. We used numbers listed on ttbar production webpage Exept for muon fake rate -> introduce eta dependence. average relative error : 4.8%

  8. Taggability New taggability rate functions. Introduce PVz dependence. Estimated on very loose preselected sample. (1 loose lepton, 2 jets, Met, DQ) EMqcd TRF MUqcd TRF

  9. Taggable jet physics η Taggable jet pT Jet multipliciy in events with at least 2 taggable jets Jet multipliciy in events with at least 1 taggable jet Taggability RF : closure plots (MU) On preselected analysis sample (MM loose)

  10. Asymetric b-tagging Use 2 working point instead of one : Tight (~0.3% mistag) and Loose. (~1% mistag) Split dataset in simple and double tag in 2 ways : = 1 Tight (STag) and >= 2 Tight (DTag) =1 Tight, 0 other Loose (STagVeto), at least 1 Tight and 1 Loose (DTagAsym) Asymetric double tag should reduce b-tagging systematics (Loose TRF) , increase double tag statistics and optmize t/s channels separation.

  11. Asymetric tagging and B-tagging systematics

  12. Likelihood discriminant Psignal(x) L= __________________ (Psignal(x) + Pbackground(x) ) Where x is a set of discriminant variables. We use Psignal(x) = ΠPsignal(xi) . This is correct if variables are uncorrelated. Psignal(xi) : Spline of final distributions. If correlations exist (as it is the case), the likelihood will not be as dicriminant as it would be with uncorrelated variable, but the method remains valid. Signal is s-channel or t-channel. Background is Wjj+Wbb+Qcd (simple tag), Wbb only (double tag) or ttbar.

  13. Likelihood input variables tt / signal Likelihood s and t channel : Sphericity Min Dr between jets Centrality Pt of 1st, 2nd and 3rd jet Invariant mass of all jets W tranverse mass Met+PT Lepton t channel only : Q.η W+jet / signal Likelihood s and t channel : Top mass (tagged jet) cos(j2,l) in top rest frame Pt of 1st and 2nd jet Invariant mass of all jets W tranverse mass Met+PT Lepton t channel only : Q.η

  14. Likelihood bias To increase the statistics, we used the same events to build the likelihood pdfs and to compute envent yields. This may introduce a bias due to sensitivity to statistical fluctuation. s-channel efficiency s-channel efficiency Green : full stat Red : Even/Odd Blue : Odd/Even ttbar efficiency Wbb efficiency To validate the method : reproduce performance curves using 1 events over 2 for pdf, the other for likelihood computation. -> 2 perf curves (even/odd & odd/even)

  15. Likelihood input variables EM+MU, STag+DTag All jets invariant mass Centrality min dR between 2 jets Leading jet pT

  16. - tt s-channel - tt t-channel Likelihood Plots STagVeto+DTagAsym , EM+MU Wjj s-channel Wjj t-channel

  17. Limits Limits computed with Bayesian fit on 2d LH distribution (LHWjj:LHtt) -> top_statistics package. JES, JER, JId, TRF, Trigger... errors are introduced as histograms (“shape” changing systematics) Other systematics are produced with top_counts. Identical to published single top analysis Asymetric tagging improve expected limits -> final result

  18. tqb STagVeto+ DTagAsym tb STagVeto+ DTagAsym Posterior densities Posterior densities : expected expected expected observed observed tqb STagVeto+ DTagAsym tb STagVeto+ DTagAsym

  19. Summary Analysis very similar to the published one : selection, yields and systematics estimates, limit computation. Improvements : Ipanema dataset (360 pb-1), JES 5.3, JLIP b-tagging. New tools : Asymetric b-tagging : reduces systematics Likelihood discriminant : robust and well known method, if not as sensitive as NN. ---> Improved limits <--- t-channel : 4.44 pb (4.27) - s-channel : 4.97 pb (3.29) Analysis note under EB review : D0Note 4825

More Related