Presentation on theme: "January 6, 20111 B Tagging in Jets at Hadron Colliders Matthew Jones Purdue University Heavy Quark Production in Heavy Ion Collisions Purdue University,"— Presentation transcript:
January 6, 20111 B Tagging in Jets at Hadron Colliders Matthew Jones Purdue University Heavy Quark Production in Heavy Ion Collisions Purdue University, West Lafayette IN January 4-6, 2011
January 6, 20112 Disclaimer Results are primarily from the CDF experiment Most apply to high energy jets as in Z, H, t decay Performance limited by detector capabilities... New LHC experiments work exceptionally well! ANY MATERIALS ARE PROVIDED ON AN "AS IS" BASIS. MATTHEW JONES SPECIFICALLY DISCLAIMS ALL EXPRESS, STATUTORY, OR IMPLIED WARRANTIES RELATING TO THESE MATERIALS, INCLUDING BUT NOT LIMITED TO THOSE CONCERNING MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, SUCH AS APPLICATION IN A HEAVY ION EXPERIMENT, OR NON-INFRINGEMENT OF ANY THIRD-PARTY RIGHTS REGARDING THE MATERIALS. VOID WHERE PROHIBITED BY LAW.
January 6, 20113 Typical Applications Cleanest environment at LEP, –measurement of –Electroweak couplings via A b FB –search for At the Tevatron, –b production cross sections –reconstruction of top decays, t Wb –Higgs searches, –Single top production Typical jet energies are E T > 50 GeV These analyses generally require well-understood b-tag efficiencies (acceptance) and fake rates (backgrounds).
January 6, 20114 Properties of B Hadrons B hadrons have unique properties because: The b quark mass is large, –high multiplicity of decay products –decay products can have a hard momentum spectrum The CKM matrix element |V cb | is small, –Relatively long lifetime,
January 6, 20115 B Production in collisions Imprecise knowledge of initial state. Decay products of massive initial state.
January 6, 20116 Modeling B Production Accurate modeling of B production and decay is an essential tool –Allows development and tuning of b tagging algorithms –First-order acceptance/efficiency estimates While important, these models are never perfect –Uncertain production mechanisms and p T spectrum –Unknown contributions to inclusive B decays –Detector effects not accurately modeled Corrections to efficiency must be measured using data Fake rate difficult to model accurately and must also be measured.
January 6, 20117 Monte Carlo Event Generators Hard process: –Event generation (eg, Pythia, Herwig) Structure functions, matrix element, parton shower Explicit calculations of are less applicable –Fragmentation function, –Include some description of the underlying event B decay generators: –Model much of what we have learned about B decays from ARGUS, CLEO, BaBar, Belle,... –CLEO developed QQ, later interfaced with event generators at CDF in Run I –Now superseded by EvtGen...
January 6, 20118 EvtGen B Decay Generator Provides a convenient framework for modeling a wide variety of decays Efficient decay chain simulation via helicity amplitude formalism Decay table: –Only about half of the decay width is accounted for in exclusive final states. –Naive spectator model invoked for B 0 s and Λ b decays. –The remaining inclusive decays are simulated using a variant of the Lund string fragmentation model. Generally good agreement with observed B decay properties...
January 6, 20119 EvtGen B Decay Generator Inclusive processes Lepton momentum in semi-leptonic decays
January 6, 201110 Properties of B Decays Leptons (e or μ), Displaced tracks: –significantly non-zero impact parameter –reconstructed secondary vertices –approximately Lorentz invariant
January 6, 201111 The CDF Detector Muon systems: CMP CMX CMU SVX-II COT TOF Hadronic EM Calorimeter
January 6, 201112 CDF II Calorimeter Pb-Scintillator EM section Fe-Scintillator Hadronic section PMT’s Light guides Δη x Δφ = 0.1 x 0.25
January 6, 201113 CDF II Tracker Small cell drift chamber in a 1.4 Tesla field 4 axial, 4 stereo superlayers, 12 sense wires per layer Fast input to level 1 track trigger –Finds tracks with p T >1.5 GeV/c –Extrapolates to EM calorimeter and muon chambers Highest quality tracks found the tracker, extrapolated into the silicon detector.
January 6, 201114 CDF II Silicon Detector 5 layers of double- sided sensors: –3 φ-z (90°) –2 φ-SAS (±1.2°) 1 single-sided inner layer attached to beam pipe Not a pixel detector: most accurate reconstruction is in the r-φ plane. 90 cm 10.6 cm Luminous region: σ z ~ 30 cm
15 Jet Reconstruction Iterative cone algorithm, typically using Corrections applied for –Non-uniformity in η –Event pileup: 350 MeV in cone per additional vertex –Non-linear tower response –Unrecognizable as jets until E T > 20 GeV Associate tracks that lie within ΔR < 0.4 ~ η φ
January 6, 201116 Secondary Vertex Tagging Algorithm SecVtx algorithm: Phys. Rev. D71, 052003 (2005).Phys. Rev. D71, 052003 (2005) –Applied to tracks with p T >0.5 GeV/c in a cone around a jet within ΔR < 0.4. –Find all tracks with impact parameter significance, –Fit a vertex to all pairs of tracks Associate other tracks if Refit all tracks to common vertex Remove tracks with large –Require significant displacement, L xy > 0: dominated by b-jets L xy < 0: mis-tagged jets
January 6, 201117 –Second pass: If no vertex found in pass 1, search lower-quality two-track vertex using higher p T tracks with |d 0 /σ d0 |>3. –Different operating points: Tight (as described) Ultra-tight (without lower-quality second pass) Loose (relaxed impact parameter significance cuts, additional attempts to seed pass 1 vertices) Many parameters that can be tuned or adjusted to manipulate efficiency/purity. Secondary Vertex Tagging Algorithm
January 6, 201118 Heavy flavor jets: –vertices with positive 2d decay length Light flavor jets: –equal numbers of positive and negative 2d decay length vertices –not quite... correct for: K 0 S and Λ decays nuclear interactions
January 6, 201119 Measuring B Tag Efficiencies In principle, this is trivial in Monte Carlo In practice: –Apply same B tag algorithm to data and Monte Carlo –Correct the efficiency in Monte Carlo using scale factors: Efficiency measured in data: –Select a sample of jets with enhanced b fraction –Measure the efficiency and heavy flavor fraction simultaneously
January 6, 201120 Measuring Efficiency with Leptons Efficiency measured in data: –Tag one jet with a high p T muon –Tag opposite side jet with positive SecVtx tag –Require M vtx > 1.5 GeV/c 2 to suppress light flavor and charm –Fit for heavy flavor fraction using lepton p T,rel Example: E T extrapolation
January 6, 201121 Measuring Efficiency with Electrons 8 GeV electron trigger sample enriched in semi- leptonic B decays Apply tag to away-side jet Naive efficiency for positive tagged electron jet: Subtract expected light-flavor mis-tags:
January 6, 201122 Measuring Efficiency with Electrons Heavy flavor fraction of away side jet: Light jet fraction Probability of tagging light away-side jet Measured using yield of e+D 0 From Monte Carlo estimated using a sample enriched in photon conversions (mostly light flavor)
January 6, 201123 SecVtx Tag Efficiency Efficiency calculated for b-jets in a Monte Carlo sample of top decays –Scale factor applied to give efficiency in data.
January 6, 201124 Fake Rates Probability that a light jet is tagged as a B jet At Tevatron energies, a generic jet sample is mostly light flavor –Measure negative tag rate as a function of: E T jet, ΣE T jet, |η|, N Z vtx, Z pv –Provides an estimator for positive mis-tag rate Typically of order 1%... But this needs to be corrected for: –N LF + > N LF - due to K 0 S and Λ decays: α –Generic jet sample contains heavy flavor: β
January 6, 201125 Tag rate asymmetry: α correction Definition: Application: Fit bottom, charm, light flavor fractions using templates constructed from –Signed vertex mass, M vtx –Pseudo-proper time, Typical result: α ~ 1.2 – 1.5 (function of E T )
January 6, 201126 Tag rate asymmetry: β correction Definition: Application: Fit the flavor fractions in the pre-tagged samples: β ~ 1.1
January 6, 201127 Fake Rates Higher fake rates when track occupancy is high (high E T jets) and near edge of tracking acceptance.
January 6, 201128 Jet Probability Algorithm Jet axis used to construct signed impact parameter for high quality, high p T tracks in a jet: Impact parameter significance: Likely to be positive for tracks from displaced secondary vertices Symmetric about zero for tracks from the primary vertex. Phys. Rev. D74, 072006 (2006)Phys. Rev. D74, 072006 (2006).
January 6, 201129 Jet Probability Algorithm Signed impact parameter significance... Negative side fitted with resolution function R(S). Track probability: –uniformly distributed between 0 and 1 for prompt tracks. Jet probability: Uniformly distributed between 0 and 1 for light flavor jets.
January 6, 201130 Jet Probability Algorithm The efficiency/purity can be continuously adjusted by selecting P J < P J cut. Typical operating points: P J < 1% or P J < 5%. Monte Carlo CDF 50 GeV Jet sample Electron sample
January 6, 201131 Scale Factor and Mistag Asymmetry Heavy flavor fraction measured using electron/conversion technique
January 6, 201132 Jet Probability Efficiency/Fake Rate Better efficiency at high E T than SECVTX.
January 6, 201133 Neural Networks The SecVtx algorithm presumably does not use up all available information Further discrimination possible using advanced multivariate methods (eg, ANN’s) Train network using signal, background from mistags 25 variables had at least 3.5σ discriminating power used for input –# tracks with d 0 significance > 3, –signed d 0 significance of tracks, –vertex mass, –and many others...
January 6, 201135 Neural Networks Network output is insensitive to the origin of the B jet Corrections calculated for sample dependence for mis- tagged jets as a function of E T, N track, ΣE T NN output used as input to another network to discriminate between signal and background in single top production.
January 6, 201136 Summary Characteristic features of heavy flavor decays exploited in various ways to tag B jets An extremely valuable element: –Ability to estimate light-flavor contamination using negative decay length/impact parameter jets –Even this is not ideal, but can be improved by α,β corrections Neural networks can be useful –provided they don’t sacrifice the ability to measure efficiency and fake rates in data Hopefully, some of these ideas can be translated to an environment with lower E T and higher track multiplicity –high quality 3d tracking using pixel detectors may be key