1 / 19

B IDentification

B IDentification. Frank Filthaut University of Nijmegen. Goals. Basic goal: efficient b-tagging in both high-p T (Higgs, top, SUSY) and low-p T (B) physics Benchmarks set in Run 2 Workshops Higgs / Supersymmetry (’98) for high p T

infinity
Télécharger la présentation

B IDentification

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. B IDentification Frank Filthaut University of Nijmegen D0-Germany meeting

  2. Goals • Basic goal: efficient b-tagging in both high-pT (Higgs, top, SUSY) and low-pT (B) physics • Benchmarks set in Run 2 Workshops • Higgs / Supersymmetry (’98) for high pT • Using secondary vertex tag and assuming “nominal” Run 2 detector performance, estimated close to 60% efficiency for mistag rate below 1% • B physics (’00) for low pT • More difficult to give a single number (trigger, analysis details) • Charge of the DØ b-id group: • Provide the physics groups with the algorithms and the tools to study their results, both off-line and (where relevant) at trigger level • Cooperate with physics groups in optimisation D0-Germany meeting

  3. Tags • Conceptually, all possibilities for tags exhausted (we think!): • Soft lepton (±, e±) tags • : Paul Balm (L3), Onne Peters • e: Abid Patwa, André Turcot (L3), Georg Steinbrück, Florian Beaudette, Jean-François Grivaz • Secondary vertex tag • Axel Naumann (L2), Arnaud Duperrin, Mossadek Talby, F. Villeneuve-Séguier (L3), Ariel Schwartzman, Marcel Vreeswijk • Impact parameter tag • Jon Hays, Ian Blackler (L3), Bram Wijngaarden, Frank Filthaut, Sasha Khanov, Flera Rizatdinova • Multivariate combinations of the above • Pavel Demine, Strasbourg (likelihood), Andy Haas (NN), Sherry Towers (guru) • Requires discriminating information from individual tags (rather than yes/no) • flavour tag • No manpower yet (may come from within B physics group) • Thought this was a pure B physics issue, but it turns out other groups also need this (e.g. t  tbar distinction) • In contrast, in Run I DØ used only its muon tags (J/ for B physics, inclusive semileptonic decays in general)! D0-Germany meeting

  4. Muon tag • L3: starting from previous L3 jet and muon “tools” • Associate muon with jet within some cone • Calculate pTrel of muon w.r.t. jet axis to distinguish between muons from b quarks and from ,K decays (and c quarks) • Using pT from muon chambers or central tracker? Resolution vs probability of wrong track – muon association • Effort not yet started (work on input L3 muons) • Offline: • Same variable, plus: P / Ejet , DCA (significance), z (significance) • +jet reco efficiency only ~ 50% for B physics (ttbar) D0-Germany meeting

  5. Electron tag • L3: effort mainly geared towards recognising J/ • B physics as well as low-energy calibration tool • Elements in common with generic L3 electron tag: electron recognition tools (track-CAL, track-PS, CAL-PS match) • Studies so far (April Vert Review): • MC-track match (R < 0.07) • Track-CAL match: • : 63 mrad (20 mrad core) • : 0.03 core but large tails (PV position!)  match in z! (changed) • CPS-CAL match: • : 29 mrad • : same PV tails • Track-CPS: • : 6.1  4.5 mrad • z: 10 mm  z  vs z  (2) D0-Germany meeting

  6. Electron tag • L3 cont’d: • Total e± tagging efficiency ~ 26% • Good for (part of) B physics studies • What about: • High pT? • Semileptonic decays? • FPS? • Offline: • Improved soft electron (E > 2 GeV/c) recognition using track extrapolation in CAL, reducing #cells taken into account • Still need PS match to reduce fake rate! • Variables: pTrel, pe/Ejet , Ejet, soft electron EEM/ptrack D0-Germany meeting

  7. Electron tag • Example for high pT: ttbar sample pTrel • Example for low pT: J/ KS sample pTrel D0-Germany meeting

  8. Electron tag • Performance for high-pT Z bb sample: • Efficiency includes b  e branching ratio • Background taken from same sample • Efficiency as fct of pT D0-Germany meeting

  9. Secondary vertex tag • L3: fast algorithm based on Hough transform • tracks in 2D space (r,  plane)  hits in 2D parametric space (d, 0) • In current implementation, start from tracks that have been found previously using a similar algorithm • but should be possible to use “official” L3 track reconstruction • Look for clustering in 0 coordinate, then “optimise” distance d • Problem: many PV tracks included in SV thus reconstructed (try to distinguish using 2 fit to either PV or SV, and cut on dt) • Intrinsic to method: binning not very fine • SV: require |d| > 1 mm, at least 3 tracks • All highly optimised for high-pT samples; 35% SV prob vs 10% PV prob D0-Germany meeting

  10. Secondary vertex tag • Offline: can do vertex finding in 3D: Kalman filter • Start by clustering tracks (simple cone, R = 0.5) • Build up SV starting from track pairs, reject tracks associated to PV and MB interactions; track pT and opening angle cuts • When SV found: associate with jet within R < 0.3 • Tag: Lxy/xy > 3 • Constrained fits  also track parameters improved • Works rather well for high-pT events (also optimised for ttbar!) D0-Germany meeting

  11. Secondary vertex tag • How well do things work for B physics? • Tracking efficiency in jets as fct of pT  down to 40% from tracking alone • Boost much smaller (<c> ~ 6 mm)  PV track rejection: 24% • After all cuts: efficiency ~ 15% Separate B physics selection required! Quality OK: resolution ~ 50 m (r,), 80 m (z) D0-Germany meeting

  12. Impact parameter tag • Offline: • Take collection of tracks • Select best PV based on z coordinates • Calculate each track’s impact parameter w.r.t. PV • Can be 2D (r- plane) or 3D • So far, studies have concentrated on 2D • Either cut on #tracks above given (physics-)signed i.p. significance, or multiply tracks’ PV probabilities to yield a discriminant (both possibilities implemented) • Need to reject tracks from , K (preferably explicitly) Z  bb ttbar(b) ttbar(l) Z light 2D impact parameter significance D0-Germany meeting

  13. Impact parameter tag • Copying CDF cuts: • 3 tracks with d/d > 2, or • 2 tracks with d/d > 3 • Starting effort on 3D tags • “Real” 3D: distance between track and PV, physics signed • Pseudo 3D: combining separate (r,) and (s, z) information (when useful) • Performance potentially more sensitive to luminosity • L3 effort has just started • Trying to re-use existing off- line code D0-Germany meeting

  14. Multivariate tags • Likelihood tag • Basic use: combination of independent 1D distributions • Higher dimensionality of the problem taken into account by doing this as a function of jet , pT • Also looking into 2D distributions • Variables used so far: pTrel,, Lxy/xy, mSV, charged energy fraction If a value is found NB issue of how to deal with “missing” data otherwise f(x|H) is distribution of variable x for hypothesis H PH is probability to find a value for hypothesis H D0-Germany meeting

  15. Multivariate tags • Results (for Z  bb vs Z  light quarks) • NN tag: using the same input, but (in principle) allows to consider full dimensionality of the problem. Started recently • Perhaps harder to understand  keep also likelihood method • NB: also individual tags can use neural nets (some do already) • NB: • 0.1 < efficiency < 0.4 • rejection > 0.992 D0-Germany meeting

  16. Common issues • Tracking efficiency in jets • Low even for MC • Luminosity dependence • Tracking efficiency • Vertex finding and selection • Jet direction (for pTrel) and energy (some criteria relative to Ejet) • Jet algorithm dependence • Cone vs. kT, algorithm parameters (so far we’ve used R=0.7 cones??) • Also: use of tracks during jet reco (instead of association afterwards) Cone jets kT jets D0-Germany meeting

  17. Common issues • Jet algorithm dependence • E resolution • MC parentage • At moderate pTjet ( ~ 50 GeV/c), large fraction of b jets originates from gluon splitting rather than lowest order production of b quarks • Makes definition of efficiency ambiguous • Lack of large (recent) MC samples of wide range of processes D0-Germany meeting

  18. Schedule • Presently, largest effort into understanding / improving performance on MC • Our inputs are also continuously changing • Takes time to find out and recover from • About to study effect of trigger • Was difficult so far, as there was no common n-tuple with both trigger and offline information • Should start trying to understand the quality of the data • Muon, dimuon, and muon+jet trigger exists now • Difficult, as b-ID is at the end of the food chain • Calorimetry, tracking, muons all need to work • Software: n-tuple, thumbnail support • Try to study / implement as much as possible of the triggers • Mainly muons • After shutdown (December), phase in other triggers • As soon as possible (allowing time for commissioning) • For our physics coordinator: first physics results by Moriond? • Is really pushing it D0-Germany meeting

  19. Conclusions • A fairly solid start has been made with b tagging • But much remains to be done • Our group is clearly manpower-limited • Algorithm development in the DØ environment is not very efficient • Especially if you’re “overseas” • DØ tends to “institutionalise” responsibilities • But one person’s effort cannot be spread too thin • Most of the people in the group are also working on other – and often more urgent – projects. • More than enough room to accommodate new collaborators D0-Germany meeting

More Related