Effects of Jet Substructure Selection in tt WbWb µνbqqb Ziggy Zacks Wesleyan University July 31, 2012 1 / 29
Optimization Optimize data selection from the semi-leptonic boosted tops analysis. Specifically with the use of jet substructure selection of Large-R jets in the muon channel. Substructure is a good means of studying highly boosted events, but is sensitive to pile up 2 / 29
Boosted Tops With high momentum, tops quarks (and other heavy particles) can decay with overlapping jets in their final state Large-R jets jets in our analysis refer to when all of the overlapping R=0.4 jets are contained inside one large jet with R=1.0 3 / 29
Jet Reconstruction ATLAS primarily uses three jet reconstruction algorithms: the anti-k t algorithm, the Cambridge-Aachen (C/A) algorithm, and the k t algorithm. These are the most widely used infrared and collinear-safe jet algorithms available for proton-proton collider physics The k t algorithm clusters the smallest p T constituents creating randomly shaped jets The anti-k t algorithm clusters the largest p T jet first providing large, robust, and cone-shaped jets The C/A algorithm reconstructs entirely off angular separation rather than p T. ATLAS has adopted the anti-k t algorithm as the standard in all of its physics analyses 4 / 29
Jet Trimming Jet mass is an important measurement due to the fact that all of the top decay products are inside large-r jet Mass measurement becomes difficult with pileup Trimming helps to remove softer jet constituents Procedure: Start with Anti-k t R=1.0 Jet k t algorithm to create subjets of R sub Remove any subjets with pt i /pjet T < f cut Anti-k t R=1.0 Jets, trimmed with f cut =0.05,R sub =0.3 Optimized by jet performance subgroup 5 / 29
d12 M T 2 d23 M W 2 Large-R Jet Substructure k t splitting scale: k t-distance of the final step in combining two subjets into the final large-r jet. It is defined by the equation: dij =min(p Ti, p Tj ) x R ij QCD expected to have a steeply falling spectrum 6 / 29
Large-R Jet Substructure N-subjettiness: decides whether jet is described as having N or fewer subjets. τ N = (1/d 0 )( k p Tk x min(δ R 1k,δ R 2k,...,δ R Nk )) with d 0 k p Tk x R Figure: R Nk (min) 0 for boosted W (or top) Figure: R Nk (min) > 0 for QCD jets 7 / 29
Data and Monte Carlo Data sample: DataJetSkim Full 2011 dataset s = 7 TeV Ldt = 4.7 fb 1 Monte Carlo: Signal: MC@NLO tt sample PYTHIA Z tt Background: PYTHIA for QCD ALPGEN for W µν+jets Sample W+jets mc11 7TeV.10769*.AlpgenJimmyWmunuNp* pt20.merge.ntup TOPBOOST.e825 s1299 s1300 r3043 r2993 p841 QCD mc11 7TeV.10501*.J4,5,6,7,8 pythia jetjet.merge.ntup TOPBOOST.e815 s1310 s1300 r3043 r2993 p841 Z mc11 7TeV.105594.Pythia Zprime tt1600.merge.ntup TOPBOOST.e997 s1372 s1370 r3043 r2993 p841 tt mc11 7TeV.105200.T1 McAtNlo Jimmy.merge.NTUP TOPBOOST.e835 s1272 s1274 r3043 r2993 p841 8 / 29
Object Selection Muons: Tight author==12 p T >25 GeV η < 2.5 ID track requirements MiniIsolation MI/p T < 0.05 Anti-k t R=0.4 Jet: p T > 30 GeV JVF > 0.75 Anti-k t R=1.0 Jet: η < 2.0 p T > 350 GeV 120 < mass < 250 GeV 9 / 29
Figure: For top with p T > 350 GeV, the separation between W and b is mostly less than R=1.0 10 / 29
Event Selection Event Selection No trigger requirement First vertex has 5 tracks and type=1 or 3 Exactly one muon No LooseBad Anti-k t 4 jets (jet cleaning) E t > 20 GeV E t +M W T > 60 GeV At least 1 Anti-k t R=10 (Large-R) Jet b-tag requirement (MV1>0.607) 11 / 29
Why these cuts? trimmed, Anti-k t R=1.0 jets, no b-tag, 600< p T <800 GeV 12 / 29
Why these cuts? trimmed, Anti-k t R=1.0 jets, no b-tag, 600< p T <800 GeV Decided on no τ 21 cut Cuts on splitting scale clearly would eliminate high QCD regions, but the N-subjettiness cut seems less clear 13 / 29
Why these cuts? trimmed, Anti-k t R=1.0 jets, no b-tag 14 / 29
Why these cuts? trimmed, Anti-k t R=1.0 jets, no b-tag τ 21 spectrum extremely uniform, decided on no cut Splitting scale cuts would eliminate high W+jets background regions but the N-subjettiness cut seems less conclusive 15 / 29
Proposed Selection d12 > 40 GeV d23 > 20 GeV τ 32 < 0.8 16 / 29
Mass With Current Selection (btag0) Not including dibosons or single top as background, which could account for Data/MC discrepancy Error bars are statistical uncertainty only, no systematics Mass distribution with current semi-leptonic selection, no substructure cuts applied 17 / 29
d12 > 40 GeV Variable distributions inside mass cut window on left Cut removes minimal signal or background from mass cut window Efficiencies in mass cut window (120-250 GeV) Signal Efficiency : 95.76% Background Efficiency : 95.43% 18 / 29
d23 > 20 GeV Pretty strong cut. Removes 63% of background from mass cut window, while losing 38% of the signal Efficiencies in mass cut window (120-250 GeV) Signal Efficiency : 62.78% Background Efficiency : 37.46% 19 / 29
τ 32 < 0.8 Cut reduces background by 32% and signal by 18% in the mass cut window Efficiencies in mass cut window (120-250 GeV) Signal Efficiency : 81.85% Background Efficiency : 68.50% 20 / 29
All Proposed Cuts Purer sample Efficiencies in mass cut window (120-250 GeV) Signal Efficiency : 57.88% Background Efficiency : 30.38% 21 / 29
MV1 Distribution Highest MV1 value of all anti-k t R=0.4 jets in the event Mass after btag cut (.607): Signal Efficiency: 61.85% Background Efficiency: 11.58% 22 / 29
Mass With Current Selection (btag1) Mass distribution with current semi-leptonic selection, no substructure cuts applied 23 / 29
d12 > 40 GeV Variable distributions inside mass cut window on left Cut removes minimal signal or background from mass cut window Efficiencies in mass cut window (120-250 GeV, 1 btag) Signal Efficiency : 97.43% Background Efficiency : 97.96% 24 / 29
d23 > 20 GeV Pretty strong cut. Removes 64% of background from mass cut window, while only losing 31% of the signal Efficiencies in mass cut window (120-250 GeV, 1 btag) Signal Efficiency : 69.16% Background Efficiency : 35.58% 25 / 29
τ 32 < 0.8 Cut reduces background by 28% and signal by 13% in the mass cut window Efficiencies in mass cut window (120-250 GeV, 1 btag) Signal Efficiency : 87.15% Background Efficiency : 72.45% 26 / 29
All Proposed Cuts with btag1 Even purer sample Ratio of background eliminated to signal lost 2 Efficiencies in mass window with respect to btag cut (120-250 GeV, btag1) Signal Efficiency : 66.88% Background Efficiency : 33.54% 27 / 29
Conclusions With one trimmed, Anti-k t R=1.0 jet: The d 12 > 40 GeV cut seems to have very little use except that it cuts out large region leading up to mass window The d 23 > 20 GeV cut appears to have good discriminating power The τ 32 < 0.8 cut had some positive effect Making substructure cuts is not as powerful as b-tagging alone If both b-tagging and new selection were applied 41% of signal would remain, with only 3% of W+jets background Ultimately, b-tagging is extremely strong discriminant and making substructure cuts after the fact is up for debate Selection Signal Efficiency Background Efficiency New Selection NO btag 58% 30% ONLY btag 62% 12% New Selection WITH btag 41% 3% 28 / 29
Acknowledgements Emily for EVERYTHING! Andrew for help whenever I needed it Professor Parsons for the opportunity 29 / 29