Flavour T agging Methods used in the ATLAS B-Physics Programme - PowerPoint PPT Presentation

Flavour T agging Methods used in the ATLAS B-Physics Programme Patrick Jussel Institut für Astro- und T eilchenphysik Universität Innsbruck FAKT – Aflenz, 23.9.2008

Overview • The ATLAS Detector • ATLAS B-Physics Programme _ • Flavour Tagging Methods: b or b? ➢ Jet Charge Tagging ➢ Soft Muon Tagging • Austrian Federated Tier 2 • Summary and Outlook Patrick Jussel, FAKT, 23.9.2008

The ATLAS Detector Muon System Inner detector Toroidal magnetic field Solenoidal magnetic field Standalone muon reconstruction Pixel detector: 50 x 400 μm 2 Semi Conductor Tracker: 17 μm Transition Radiation Tracker: 170 μm Precision measurement for tracks and vertices Calorimeters LAr calorimeters @ -183°C - EM calorimeter endcaps & centre - Hadronic calorimeter endcaps Trigger Tile Calorimeter Fast identification of interesting events, e.g. muons, jets, missing energy etc. - Hadronic calorimeter barrel Patrick Jussel, FAKT, 23.9.2008

The ATLAS Detector 2008-09-10 10:19 Patrick Jussel, FAKT, 23.9.2008

ATLAS B-Physics Programme σ(bb) ≈ 500 μb 5·10 5 bb pairs/s @ L=10 33 cm -2 s -1 (low lumi) B @ ATLAS - features / drawbacks: + precise vertexing and tracking σ bb + high resolution calorimetry + good muon identification + flexible trigger scheme - no kaon/pion separation Physics Programme: Study heavy flavour mesons and baryons e.g. heavy quarkonium production Measure rare and very rare decays 0 → μ + μ - ) sensitive to new physics (e.g. B s _ CP Violation: 0 - B s 0 meson system parameters of the B s _ → here we need flavour tagging: b or b? Patrick Jussel, FAKT, 23.9.2008

ATLAS B-Physics Programme 0 mixing parameter Δm s : B s _ 0 and B s 0 Oscillations between the flavour eigenstates B s are proportional to mass difference of the two mass eigenstates: − t  1 ± cos  m s t  P ±  t ~ e 0 ,  0  D s     − B s Channels: 0 ,  0  D s     a 1 − B s Method: measure the flavour of the B s meson at decay time and the proper decay time (Reconstruction), tag the flavour at production (Flavour Tagging), maximum likelihood fit method to extract Δm s A  t = N mix  t − N nomix  t  N mix  t  N nomix  t  ∝ cos  m s t  Patrick Jussel, FAKT, 23.9.2008

ATLAS B-Physics Programme Measurement of the weak mixing phase Φ s = -2ηλ 2 :   −  0  J / 0 B s Method: angular distribution of the final state particles in is depending on Φ s , 8 parameters extracted in a maximum likelihood fit SM Φ s not accessible by ATLAS, but deviations SM = -0.036 +/- 0.003 CKM Fitter: Φ s Patrick Jussel, FAKT, 23.9.2008

Flavour Tagging Methods • Opposite Side Tagging: • Same Side Tagging: - Reconstruction - Jet charge - Jet charge - Kaon charge - Lepton charge Oscillations of the opposite side B hadron are possible! Patrick Jussel, FAKT, 23.9.2008

Flavour Tagging Methods Jet Charge Tagging • Principle: due to hadronisation a meson with a b quark is accompanied more likely by negatively charged mesons.  q i p i Q jet = ∑ • Jet charge: ∣ p i  ∣  R  R =   2  2 … with • Sum over all tracks in cone except the decay products of the signal channel. • p L found to be the best choice to use κ in this case as p i Patrick Jussel, FAKT, 23.9.2008

Flavour Tagging Methods Jet Charge Tagging When using a flavour tagger on events with well known flavour, count the events with a good tag ( N g ), with a wrong tag ( N w ) and the events without a tag ( N no ), then we receive... N w  N g = Tagging Effciency: N w  N g  N no N w = Wrong tag fraction: 0  J / N w  N g B s D = N w − N g Wrong tag fraction 0.37 Dilution: N w  N g Dilution 0.25 2  D Tagging Power: Tagging Power 0.04 Expected Performance of the ATLAS Experiment, Detector, Trigger and Physics; ATLAS Collaboration; CERN-OPEN- 2008-020, Geneva, 2008, to appear. Patrick Jussel, FAKT, 23.9.2008

Flavour Tagging Methods Soft Muon Tagging • Principle: opposite side b hadron decaying semi- leptonically 0 flavour • Search for highest p T muon in event, muon charge → signal B s 0 → D s 0 → D s - (Φ,π)π + and: B s - (Φ,π)a 1 • Used in the channel: B s + • Need to understand the sources of wrong tags: – b → c → μ decay chain – additional bb and cc pairs – muons from from J/ψ , τ , ρ , … 0 , B s – oscillations of neutral opposite side b hadrons ( B d 0 ) Patrick Jussel, FAKT, 23.9.2008

Flavour Tagging Methods 0 ,  − B s 0 ,    −  0  D s 0  D s   −  a 1 − B s Soft Muon Tagging Luminosity 10 fb -1 10 fb -1 Comparision of the two signal Rec. Events 6657 3368 channels and histograms Efficiency 0.988 0.985 showing the wrong tag Wrong tag Fraction 0.223 0.233 fractions depending on the tagging muons transverse Dilution 0.554 0.25 momentum. Tagging Power 0.30 0.28 BsDsPi BsDsa1 Expected Performance of the ATLAS Experiment, Detector, Trigger and Physics; ATLAS Collaboration; CERN- OPEN-2008-020, Geneva, 2008, to appear. Patrick Jussel, FAKT, 23.9.2008

Flavour Tagging Methods 0 → D s - (Φ,π)π + ) • different distributions, here wrong tag fraction (9500 B s – distributions depending on e.g. p T ( μ ), p T rel ( μ ) and p T (B S 0 ) – distributions are in different WTF ranges, e.g. 14%-27% for p T rel ( μ ) • PROBLEM: we would like the wrong tag fraction for each event... • QUESTION: Which wrong tag fraction should go into the fit? What is the “real” event by event wrong tag fraction? How can all this information be combined? • ANSATZ: combine variables with TMVA Patrick Jussel, FAKT, 23.9.2008

Flavour Tagging Methods • TMVA = Toolkit for multivariate data analysis in ROOT • Easy to use framework to train, test and use a classifier with a common interface for each technique • Techniques available: Fisher Discriminant, Multidimensional Likelihood Estimator, various Artificial Neural Networks, … • Here: combine p T ( μ ), p T rel ( μ ) and p T (B S 0 ) to one tmva-classifier by: – train classifier with a small fraction of events – evaluate the classifier – test resulting classifier with remaining events • Knowing the truth from Monte Carlo, calculate the wrong tag fraction depending on the tmva-classifier … ( more on http://tmva.sourceforge.net/ ) Patrick Jussel, FAKT, 23.9.2008

Flavour Tagging Methods Fisher Discriminant x 2 x 2 H 1 H 1 Wrong tag Good tag • Geometric classification: search for axis in hyperspace pushing the means of the Axis projections as far as possible from each other H 0 H 0 x 1 x 1 Patrick Jussel, FAKT, 23.9.2008

Flavour Tagging Methods Neural Network • Combine variables in a Clermont-Ferrand Neural Network Patrick Jussel, FAKT, 23.9.2008

Flavour Tagging Methods next steps... • Combine Jet Charge Tagger and Soft Muon Tagger to one combined Flavour Tagger to calculate one event by event wrong tag fraction • Add additional tagging methods: – soft electron tagger – same side K*(892) tagger → – same side lambda tagger • Prepare software for calibration with real data MC study Patrick Jussel, FAKT, 23.9.2008

Austrian Federated Tier 2 Some words on Grid Computing... What is the Austrian Federated Tier 2? • LHC experiments produces a high data rate (more then 1 GB/s) • Worldwide LHC Computing Grid (WLCG) • embedded: Austrian Federated Tier 2 Patrick Jussel, FAKT, 23.9.2008

Austrian Federated Tier 2 • Grid Pilot Project in Innsbruck (2002) • Austrian Grid Phase I (2004-2006) • Austrian Grid Phase II (2007-2010) Status of the Austrian Federated Tier 2 until end of 2008: Vienna Innsbruck CPU > 1000 cores > 200 cores Storage > 100 TB ~ 50 TB Cost estimate: 1.053k€, funded by Infrastructure and manpower to be provided by participating Institutions Patrick Jussel, FAKT, 23.9.2008

Summary • Jet Charge Tagger and Soft Muon Taggers for the 0 channels measurements of B s • Sources of wrong tags are understood (in theory...) • Multivariate Analysis used for an event by event wrong tag fraction • Waiting for data, need calibration Patrick Jussel, FAKT, 23.9.2008

FIN Patrick Jussel, FAKT, 23.9.2008

ATLAS B-Physics Programme Likelihood function of Φ s measurement: Patrick Jussel, FAKT, 23.9.2008