TrackFixup (overview) David Nathan Brown Motivation Evaluation of existing tracks Conceptual design of TrkFixup Functional description Algorithm overview Loop branch filter (Gerry Lynch) Phyics evaluation tools (David Lopes Pegna) Schedule Discussion
TrkFixup Motivation Tracks are the core of BaBar analysis Precisely reconstruct P, trajectory of charged particles Implicitly define neutral particles (= clusters with no track) Track efficiency is combinatoric in most analyses Small improvements can have important net results Improved tracking efficiency = ‘free luminosity’ Especially relevant in competing with Belle The Mini contains detailed detector-level information Allows running ‘reconstruction’ algorithms in analysis The existing track selections predate the mini The next full reprocessing will be in …. There is room for improvement in BaBar tracking David Nathan Brown 2 Tools Workshop October 1, 2005
MC True Charged Particles (18.3.1b, B generic + semi-leptonic) 14% track-finding inefficiency Momentum (GeV) Cos() 2-3% (absolute) improvement possible Recoverable tracks are not ‘junk’ David Nathan Brown 3 Tools Workshop October 1, 2005
The Truth about Reconstructed Tracks =‘Default’ TrkRecoTrk =|d0|<1.5cm && |z0| < 10cm 90% 10% 8% 20% 2% 89% 8% 2% 6% 1% Gev Gev = GTVL && 0.1<Pt<10.0Gev && NDch>=20 78% 6% 1% 3% 0% ‘good reco’ 2 (reco-true params)<100 ‘bad reco’ 2 (reco-true params)>100 ‘duplicates’ ghosts, loopers,… ‘Phys. Bkg.’ decay-in-flight, … ‘Det. Bkg.’ mat. Int., albedo, … ‘Machine Bkg.’ Background frame Gev David Nathan Brown 4 Tools Workshop October 1, 2005
TrkFixup Goals Increase the track finding efficiency Recover tracks with >= 4(6) Svt hits Improve track resolution (good track efficiency) Cleanup pat. rec. errors Accommodate non-Gaussian effects hit resolution, hard scattering, … Add information to the track fits dE/dx constraint on momentum Recover lost (inner) Svt hits Filter ‘Junk’ tracks explicitly by process Duplicate tracks (loop branches, ghosts, …) Physics and detector background (kinks, had. Int., …) Machine background (svt only random combinations) Define ‘physics lists’ of TrkRecoTrk objects Standardized pre-selection for analysis David Nathan Brown 5 Tools Workshop October 1, 2005
… TrkRecoTrk Lists duplicate candidate: Promote to Kalman fit ‘Bad’ track fits or ‘Work’ List ‘Inner’ List TrkHitFix TrkLoopFix … ‘Merged’ List ‘Default’ List ‘Outer’ List Good track fits, No duplicates: Set aside ‘Junk’ List ‘Good’ List David Nathan Brown 6 Tools Workshop October 1, 2005
How it works TrkFixup works on TrkRecoTrk objects Full reconstruction interface, not Beta! TrkFixup runs in cache mode questionable tracks are ‘promoted’ to a full Kalman fit Roughly double the time of normal cache mode event read Refit mode is 10x cache mode TrkFixupSequence is inserted into BetaMiniSequence Before track-cluster match BtaCandidates are rebuilt from tracks Two possible configurations for BtaCandidate lists Produce ‘Fixup’ lists in parallel with ‘normal’ lists Tracks are copied (to avoid changing ‘normal’ tracks) New lists are ‘ChargedTracksFixup’, etc. Overwrite ‘normal’ lists (direct) David Nathan Brown 7 Tools Workshop October 1, 2005
TrkFixup Algorithm Status Infrastructure TrkRecoTrk vertexing W. Hulsbergen V0 reconstruction T. Tanabe Algorithm modules under development Looper removal G. Lynch In TrkFixup Ghost track removal D. Brown In TrkFixup dE/dx momentum constraint M. Battaglia In Development Hit filtering K. Tackmann In TrkFixup Svt hit recovery E. Prencipe In Development Short-track recovery Open Albedo removal Open Kink finding Open Material interaction Open Bremstrahlung recovery Open David Nathan Brown 8 Tools Workshop October 1, 2005
TrkRecoTrk Vertex Reconstruction Primary Vertex reconstruction Uses same algorithm as Bta vertexing Comparable resolution Vx ~ 70 m, pull width ~ 1.3 W. Hulsbergen T.Tanabe (LBL/UCB) V0 reconstruction Uses TrkPocaVtx W. Hulsbergen Finds Ks, Efficiency >90% Purity ~ 90% Run3 BSemiExclusive David Nathan Brown 9 Tools Workshop October 1, 2005
TrkHitFix - Method Worst hit: > 8 Disable worst hit and refit track Preselection: Track probability < 10-5 Kerstin Tackmann, LBL/UCB
Kerstin Tackmann, LBL/UCB TrkHitFix - Results Worst hit in DCH log10track probability Worst hit in SVT Fraction of tracks with pull < 5 Kerstin Tackmann, LBL/UCB
TrkHitFix - Summary and Plans TrkHitFix module improves resolution and track fit probability Largest effect in intermediate momentum region Possible improvements: Try to separate SVT hits in tail and bump of pull distribution Allow disabling several hits Worst hit Kerstin Tackmann, LBL/UCB
dE/dx constraint M. Battaglia UCB/LBL dE/dx curvature combined Idea: constrain momentum using (Svt) dE/dx by inverting B.B. curve Provides competitive P measurement as curvature at low momentum Makes P/P ~ flat vs P at <5% Reduces outlyers Need improved calibration of dE/dx to go below 90Mev M. Battaglia UCB/LBL Hist = before constraint Point=after constraint dE/dx curvature combined David Nathan Brown 13 Tools Workshop October 1, 2005
Ghost Tracks Ghost = 2nd track reconstructed from the same physical (charged) particle, excluding loopers view z view David Nathan Brown 14 Tools Workshop October 1, 2005
Ghost ‘Reconstruction’ Primary David Nathan Brown 15 Tools Workshop October 1, 2005
Ghost Selection Composition Put ‘Ghost’ on ‘Outer’ list Junk Ghost Primary Ghost <50% efficient >90% pure cm cm David Nathan Brown 16 Tools Workshop October 1, 2005
TrkGhostFix to-do Try to improve efficiency Loosen cuts? New selection? Find better algorithm to define ‘Primary’ track when neither have Svt hits Can momentum be improved by merging? Preliminary result: no David Nathan Brown 17 Tools Workshop October 1, 2005