Evidence for Evidence for Single Top Quark Production Single Top - PowerPoint PPT Presentation

Evidence for Evidence for Single Top Quark Production Single Top Quark Production Reinhard Schwienhorst Reinhard Schwienhorst MSU high energy seminar, 1/9/2007

Outline • Motivation • Preparation • Optimized Event Analysis • Sensitivity • Cross section measurement • |V tb | • Conclusions/Outlook Emphasis on what's new and different compared to previous analyses 2 Reinhard Schwienhorst, Michigan State University

Motivation What is single top and why does it matter?

Top Quark properties: • Mass, strong coupling gluon t q  q  t King of the Fermions 4 Reinhard Schwienhorst, Michigan State University

Top Quark properties: • Mass, strong coupling • Charge, Spin • Electroweak interactions – Charged current (W boson) • Total width • CKM matrix – Neutral current (Z boson) New physics? • Coupling to Higgs? • Modified weak coupling? • SUSY? Coupling to new particles? 5 Reinhard Schwienhorst, Michigan State University

Top Quark properties: Study in single top • Mass, strong coupling quark production • Charge, Spin • Electroweak interactions – Charged current (W boson) • Total width top quark • CKM matrix – Neutral current (Z boson) Really don't know: • Coupling to Higgs? • Modified weak coupling? W boson • SUSY? Coupling to new particles? 6 Reinhard Schwienhorst, Michigan State University

Top quark electroweak charged current interaction 7 Reinhard Schwienhorst, Michigan State University

SM single top quark production Associated t-channel s-channel production u q t g d W t W  b b t q' W b σ tot = 3 pb TeV: LHC: σ tot = 326 pb 8 Reinhard Schwienhorst, Michigan State University

New physics Associated t-channel s-channel production q g q q t W' t Z, γ , g t c  b W q' b Flavor New heavy boson Modified Changing Wtb coupling Neutral Current 9 Reinhard Schwienhorst, Michigan State University

s-channel Tevatron signature b t “ tb ” q W ν l  q'  b 10 Reinhard Schwienhorst, Michigan State University

s-channel Tevatron signature b t “ tb ” q W ν l  q'  b g Cao, RS, Yuan PRD71, 054023 (2005) 11 Reinhard Schwienhorst, Michigan State University

Tevatron signature t-channel q' q b W “ tqb ” t b ν g l  b Cao, RS, Benitez, Brock, Yuan, PRD72, 094027 (2005) 12 Reinhard Schwienhorst, Michigan State University

Results, Run I and Run II tqb 95% upper limit < 58 pb < 22 pb < 25 pb < 5.0 pb < 5.0 pb < 5.0 pb < 4.4 pb 2005 “Search for Single Top Quark Production using likelihood discriminants,” DØ Note 4825 (2005). Plus 7 PhDs so far 13 Reinhard Schwienhorst, Michigan State University

Tevatron single top goals Production cross sections: s-channel t-channel NLO calculation: 0.88 pb (±8%) 1.98 pb (±11%) current 95% CL limits, DØ: < 5.0 pb < 4.4 pb CDF: < 3.1 pb < 3.2 pb • Discover single top quark production! • Measure production cross sections → CKM quark mixing matrix element V tb • Look for physics beyond the standard model – Coupled to the heavy top quark • Study top quark spin correlations • Understand as background to many searches 14 Reinhard Schwienhorst, Michigan State University

Preparation How do we collect a sample of candidates? 15 Reinhard Schwienhorst, Michigan State University

Experimental setup: Batavia, Illinois Fermilab Tevatron in Run II CDF Proton-antiproton collider CM energy 1.96TeV → Energy frontier Instantaneous luminosity >250E30cm -2 s -1 − ∼4 interactions per crossing, 1.7M crossing per second → Luminosity frontier DØ 16 Reinhard Schwienhorst, Michigan State University

Apparatus: Run II DØ Detector Silicon Tracker Fiber Tracker Calorimeter Muon System

910 pb -1 analysis December 2006 370 pb -1 analysis July 2005 230 pb -1 analysis March 2005, Phys. Lett. B

Improvements • More than twice as much data • Improved W+jets background modeling • Fully reprocessed dataset – New calibrations, lower thresholds, ... • Neural Network b-quark tagging • Split analysis into 12 separate channels ➔ By lepton, jet multiplicity, tag multiplicity • Combined s+t search – Assuming SM ratio of cross sections 19 Reinhard Schwienhorst, Michigan State University

Single top quark event selection (b) quark jet ● Leading jet: E T > 25 GeV ● Second jet: E T > 20 GeV ● 0-2 additional jets (ET > 15 GeV) W  t b-quark jet High-momentum lepton (E T > 15 GeV) Missing (unbalanced) energy (> 15 GeV) 20 Reinhard Schwienhorst, Michigan State University

Event Sample Composition W+jets Top quark pairs ( ~ 1 ev pb ) Single top 21 Reinhard Schwienhorst, Michigan State University

Data-background comparison 22 Reinhard Schwienhorst, Michigan State University

Optimized Event Analysis How do we find the needle in the haystack? 23 Reinhard Schwienhorst, Michigan State University

Multivariate Methods Method: Input: Output: multivariate analysis discriminating variables signal probability Event energy Quark jet angle P(signal) Reconstructed top mass Reconstructed top spin ..... Bayesian Neural Networks Matrix Elements Boosted Decision Trees 24 Reinhard Schwienhorst, Michigan State University

Discriminating Variables • Event kinematics Object kinematics – H (total energy) Jet p T for different jets – H T (transverse energy) – M (invariant mass) – M T (transverse mass) – Summing over various objects in the event • Angular variables – Jet-jet separation – Jet pseudorapidity (t-channel) – Top quark spin – Sphericity, aplanarity 49 variables total 25 Reinhard Schwienhorst, Michigan State University

Decision Trees • Send each event down the tree • Each node corresponds to a cut H T >212 – Divide sample in two: Pass ↔ Fail Fail Pass • A leaf corresponds to a node M t <352 p t <31.6 without branches F P F P – Defines purity = N S /(N S +N B ) from MC sample purity • Training: optimize Gini improvement – Gini = 2 N S N B /(N S + N B ) • Output: purity for each event 0 1 • Boosting: average over many trees (~100) – Iterative tree building: train each new tree focusing more and more on misclassified events 26 Reinhard Schwienhorst, Michigan State University

Bayesian Neural Networks • NN with three layers – 24 in put nodes (variables) – 40 hidden nodes – Each node has a weight • Bayesian Idea: explore all possible weights • Average over 100 individual neural networks – Each network gets a weight based on training performance 27 Reinhard Schwienhorst, Michigan State University

Matrix Elements • Calculate signal discriminant directly for each event • Signal/Background probabilities are calculated from the differential cross section • Calculate differential cross section for each event based on Feynman diagram and event kinematics • Integrate over ME and measured momenta 28 Reinhard Schwienhorst, Michigan State University

Measurement Procedure Multivariate Output Bayesian posterior For each, measure peak position σ • Separate optimization for each process – s-channel, t-channel • Different processes, sensitivity to new physics – s+t combined • Assuming ratio of SM XS – Maximize sensitivity to SM single top 29 Reinhard Schwienhorst, Michigan State University

Measurement Procedure Multivariate Output Bayesian posterior For each, measure peak position σ • Separate optimization for each process – s-channel, t-channel • Different processes, sensitivity to new physics – s+t combined Focus on this for the • Assuming ratio of SM XS remainder of the talk! – Maximize sensitivity to SM single top 30 Reinhard Schwienhorst, Michigan State University

Ensemble Tests – Draw ~1,000,000 “pseudo-data” sets of events from the signal+background MC • Bootstrap with replacement • Several different signal XS values – Repeat full statistical analysis and measure σ for each 31 Reinhard Schwienhorst, Michigan State University

Sensitivity, p-value • P-value: fraction of 0-signal ensembles measuring σ above observed value • Expected p-value: fraction of 0-signal ensembles measuring σ above SM value • Expected p-values: 32 Reinhard Schwienhorst, Michigan State University

Test model on data • W+jets sample Top quark pairs = 4 jets, high event energy (H T = 2 jets, low event energy (H T (l,j) < 175 GeV) (l,j) > 300 GeV) 33 Reinhard Schwienhorst, Michigan State University

Results Let Data speak! 34 Reinhard Schwienhorst, Michigan State University

Bayesian Neural Network 35 Reinhard Schwienhorst, Michigan State University

Matrix Element 36 Reinhard Schwienhorst, Michigan State University

Boosted Decision Trees 37 Reinhard Schwienhorst, Michigan State University

Reconstructed top quark mass Low DT High DT region region Signal DT region 38 Reinhard Schwienhorst, Michigan State University

Summary Submitted to PRL 39 Reinhard Schwienhorst, Michigan State University