Tracking Validation outline Giulia Casarosa ★ Current Performances INFN - Sezione di Pisa ‣ Overall Efficiency & Purity ‣ Track Quality Eugenio Paoloni ~ unipi ‣ Pattern Recognition ★ How Jakob Fix Looks Like Face2Face Tracking Meeting Prague, January 19 th 2015
Current Performances efficiency track pattern + + & purity quality recognition ➡ used release r14908 (except otherwise stated), TrackingPerformanceEvaluation Module Belle CDC pattern recognition (Trasan) ‣ Merger of VXD TrackCand + CDC TrackCand uses MC truth information ‣ no extrapolation from CDC to VXD and vice-versa ‣ 2 Giulia Casarosa F2F Prague - 19 th January 2015
efficiency & purity Integrated Efficiency & Purity tracking ( * ) VXD only CDC only VXD+CDC (r14200) (r14200) purity (%) 95.72±0.08 77.1±0.1 - not updated not updated see next slide ε = efficiency (%) 68.4±0.2 75.3±0.1 82.0±0.1 efficiency ’ (%) ε ’ = 77.8±0.1 91.1±0.1 91.8±0.1 ( * ) numbers and plots shown are based on 8k Y(4S) events purity = probability to find an MCParticle associated to a Track, given a Track # MCParticles with at least one associated Track ε = physical efficiency # MCParticles # MCTrackCands with at least one associated Track geometrical acceptance and ε ’ = detector efficiency are factored out # MCTrackCands 3 Giulia Casarosa F2F Prague - 19 th January 2015
Reconstruction run-time errors related to tracking ➡ r14908, during this study, after a few events, basf2 exited with this error: [ERROR] Uncaught exception encountered: Out-of-range access in StoreArray::operator[], for array 'TrackCands' (durability: event) { module: MCTrackMatcher } Traceback (most recent call last): File "/gpfs/home/belle2/casarosa/releases/rValidation2/workdir/../tracking/validation/ a02_trackingEfficiency_runTracking.py", line 85, in <module> process(path) File "/gpfs/home/belle2/casarosa/releases/rValidation2/lib/Linux_x86_64/debug/basf2_def.py", line 155, in process fw.process(path) IndexError: Out-of-range access in StoreArray::operator[], for array 'TrackCands' (durability: event) ➡ build-2014-10-18, Skimming MC-4.5 failed jobs are almost all due to: [ERROR] Uncaught exception encountered: Out-of-range access in StoreArray::operator[], for array 'TrackFitResults' (durability: event) { module: ParticleLoader } Traceback (most recent call last): File "/home/belle2/casarosa/releases/build-2014-10-18/workdir/scripts/skim.py", line 85, in <module> process(analysis_main) File "/sw/belle2/releases/build-2014-10-18/lib/Linux_x86_64/debug/basf2_def.py", line 153, in process fw.process(path) IndexError: Out-of-range access in StoreArray::operator[], for array 'TrackFitResults' (durability: event) 4 Giulia Casarosa F2F Prague - 19 th January 2015
efficiency & purity Efficiency VS Transverse Momentum ε ’ = 97% @ p t = 500 MeV/c ε ’ = 77% @ p t = 100 MeV/c 1- ε ’ = 3% @ p t = 1 GeV/c VXD+CDC VXD+CDC legend: • ε , physical efficiency • ε ’, geometrical acceptance and detector efficiency factored out 5 Giulia Casarosa F2F Prague - 19 th January 2015
efficiency & purity Efficiency VS Polar and Azimuthal Angles flat as expected clear effect of the geometrical acceptance lower ε in the forward and backward directions VXD+CDC VXD+CDC legend: • ε , physical efficiency • ε ’, geometrical acceptance and detector efficiency factored out 6 Giulia Casarosa F2F Prague - 19 th January 2015
track quality Transverse Impact Parameter VXD+CDC ⟨ σ d0 ⟩ = 131 μ m VXD+CDC ➡ almost gaussian pulls distribution 7 Giulia Casarosa F2F Prague - 19 th January 2015
track quality σ d0 VS transverse momentum VXD+CDC ➡ 3 families of tracks: ‣ CDC only tracks ‣ tracks with no PXD clusters attached ‣ tracks with PXD clusters attached 8 Giulia Casarosa F2F Prague - 19 th January 2015
r14200 Comparison with B A B AR Tracking VXDonly B A B AR σ z0 r14200 ➡ Belle2 VXD Tracking when PXD B A B AR σ d0 clusters are used in the track B2 σ d0 Tracks with PXD cluster fit performs better than B A B AR B2 σ d0 Tracks with NO PXD cluster ➡ Belle2 Tracking when no PXD clusters are used in the track fit performs worse than B A B AR 9 Giulia Casarosa F2F Prague - 19 th January 2015
track quality Relating Track Parameters Errors (1) ➡ Can relate the errors on d0 and ϕ and estimate the distance between the origin and the nearest hit (i.e. the hit that has more weight in the d 0 estimation ) VXD+CDC track PXD layer beampipe σ ϕ p t σ d0 σ d0 ≃ r t · σ ϕ ➡ valid when first hit is near the beam pipe MC-truth-information free plot! ➡ neglecting detector hit resolution 10 Giulia Casarosa F2F Prague - 19 th January 2015
track quality Relating Track Parameters Errors (2) ➡ Can also user the errors on z0 and cot θ to estimate the distance between the origin and the nearest hit (i.e. the hit that has more weight in the z 0 estimation ) σ z0 ≃ r t · σ cot θ VXD+CDC VXD+CDC MC-truth-information free plot! 11 Giulia Casarosa F2F Prague - 19 th January 2015
track quality Relating Track Parameters Errors (3) ➡ Can also look at the longitudinal view and plot the distance between the origin and the nearest hit to the origin. ProjectionY( ) of the left plot VXD+CDC VXD+CDC MC-truth-information free plots! beampipe 12 Giulia Casarosa F2F Prague - 19 th January 2015
track quality Effect of Multiple Scattering σ d0 β p sin 3/2 θ · (13.6MeV/c) –1 VS p t σ d0 β p sin 3/2 θ · (13.6MeV/c) –1 VS p t VXD+CDC ProfileX( ) of the left plot VXD+CDC ➡ taking into account the error introduced by the multiple scattering of the first layer crossed by the track: p t · σ d0 β p sin 3/2 θ = cost 13 Giulia Casarosa F2F Prague - 19 th January 2015
pattern recognition Patter Recognition Efficiency & Purity not updated Pattern VXD only CDC only VXD+CDC Recognition purity (%) 94.88±0.08 75.5±0.1 - not updated ε = efficiency (%) 79.3±0.2 91.3±0.1 94.3±0.1 ➡ above numbers are relative to release r14200 VXD only PR efficiency increased to 85% with the patch, same purity • VXD+CDC PR efficiency increased to 96.1% with the patch • purity = probability to find an MCTrackCand associated to a TrackCand, given a TrackCand # MCTrackCand with at least one associated TrackCand ε = # MCTrackCand 14 Giulia Casarosa F2F Prague - 19 th January 2015
pattern recognition PR Efficiency on VXD Layers PR efficiency VS VXD Layer Number layer1 layer2 layer3 layer4 layer5 layer6 VXD+CDC • Efficiency is maximum on layer3 (80%) • Efficiency decreases from inner to the outer layers (70% on layer1, 60% on layer6) 15 Giulia Casarosa F2F Prague - 19 th January 2015
pattern recognition VXD Clusters used in the Track Fit hits used in the Track Fit: transverse view hits used in the Track Fit: longitudinal view L3 hot spots L2 L2 L1 L1 VXD+CDC VXD+CDC lower efficiency in PXD slanted ladders? 16 Giulia Casarosa F2F Prague - 19 th January 2015
How Jakob Fix Looks Like ➡ compare VXD only tracking in r14200 (before the patch) and r14200+patch patch = svn update of: tracking/modules/VXDTF/{include,src}/VXDTFModule.{h,cc} • tracking/modules/VXDTFHelperTools/{include,src}/FilterCalculatorModule.{h,cc} • tracking/spacePointCreation/{include,src}/spacepoint.{h,cc} • tracking/vectorTools • 17 Giulia Casarosa F2F Prague - 19 th January 2015
efficiency & purity Efficiency VS Polar and Azimuthal Angles known modulation known drop for θ = π /2, of 2 π /7 is still present still there VXDonly VXDonly ➡ improvement in efficiency but undesired feature are still present 18 Giulia Casarosa F2F Prague - 19 th January 2015
efficiency & purity Efficiency VS p t and VXD Layers PR efficiency VS VXD Layer Number layer3 layer4 layer5 layer1 layer2 layer6 VXDonly VXDonly ➡ improvement in efficiency VS p t ➡ huge increase of the efficiency on layers 1 and 2, but an undesired 10% decrease on outer layers shows up 19 Giulia Casarosa F2F Prague - 19 th January 2015
track quality Transverse Impact Parameter Resolution r14200 patch VXDonly VXDonly Tracks with PXD cluster Tracks with PXD cluster Tracks with NO PXD cluster Tracks with NO PXD cluster 20 Giulia Casarosa F2F Prague - 19 th January 2015
track quality σ d0 with and without PXD clusters VXDonly VXDonly 21 Giulia Casarosa F2F Prague - 19 th January 2015
track quality Relating Track Parameters Errors projection on r t of σ d0 / σ ϕ (v. slide 12) VXDonly ➡ striking difference between the two curves 22 Giulia Casarosa F2F Prague - 19 th January 2015
Conclusions σ d0 β p sin 3/2 θ VS p t ➡ Good improvements in the tracking Profiled on p t performances after Jakob Fix! (slide 13) ➡ Some plots that may be implemented in our new validation: PR efficiency VS σ d0 / σ ϕ z-r t view VXD Layer Number (slide 15) projected on r t σ d0 VS p t w/wo PXD hits (slide 12) Profiled on p t (slide 21) Thank You! 23 Giulia Casarosa F2F Prague - 19 th January 2015
backup slides r14200 validation plots here: https://agenda.infn.it/getFile.py/access?resId=0&materialId=slides&contribId=7&sessionId=3&subContId=1&confId=8680 24 Giulia Casarosa F2F Prague - 19 th January 2015
Recommend
More recommend
