CMS Tracking POG Performance Plots 2014

Muon Efficiency With Dedicated Iterations

Plot Description
eff_eta_dr030e030_corr.png In 2012 data it was noticed a loss of muon reconstruction efficiency in the tracker. In order to recover it, 2 additional iterations have been designed: 1. an Outside-in iteration, seeded from the muon system, designed to recover the missing muon-track in the tracker and 2. an Inside-Out iteration designed to re-reconstruct muon-tagged tracks with looser requirements to improve the hit-collection efficiency. The plot on the left shows mainly the effect of the Outside-in iteration. The efficiency is computed or all tracks passing a loose selection (pt > 2 GeV || (abs(eta) > 1 && p > 2 GeV) to ensure they reach the muon system, using a Tag&Probe method on a small fraction of the data from the SingleMu dataset collected in the Run2012C period. The red bands represent the efficiency computed using the standard tracking of 2012, while the black points represent the efficiency used by adding the 2 new iterations describe above. The recovery is clearly visible across the whole tracker-eta coverage.
eff_vtx_dr030e030_corr.png In 2012 data it was noticed a loss of muon reconstruction efficiency in the tracker. In order to recover it, 2 additional iterations have been designed: 1. an Outside-in iteration, seeded from the muon system, designed to recover the missing muon-track in the tracker and 2. an Inside-Out iteration designed to re-reconstruct muon-tagged tracks with looser requirements to improve the hit-collection efficiency. The plot on the left shows mainly the effect of the Outside-in iteration. The efficiency is computed for all tracks passing a loose selection (pt > 2 GeV || (abs(eta) > 1 && p > 2 GeV) to ensure they reach the muon system, using a Tag&Probe method on a small fraction of the data from the SingleMu dataset collected in the Run2012C period. The red bands represent the efficiency computed using the standard tracking of 2012, while the black points represent the efficiency used by adding the 2 new iterations describe above. The new iterations are clearly much less sensitive to the underlying PU conditions.
TkLay7_from_TMA_eta.png In 2012 data it was noticed a loss of muon reconstruction efficiency in the tracker. In order to recover it, 2 additional iterations have been designed: 1. an Outside-in iteration, seeded from the muon system, designed to recover the missing muon-track in the tracker and 2. an Inside-Out iteration designed to re-reconstruct muon-tagged tracks with looser requirements to improve the hit-collection efficiency. The plot on the left shows mainly the effect of the Inside-out iteration. The efficiency is computed for all tracks passing a tighter selection (pt > 20 GeV and having hits in at least 7 layers in the tracker) to ensure they reach the muon system, using a Tag&Probe method on a small fraction of the data from the SingleMu dataset collected in the Run2012C period. The red bands represent the efficiency computed using the standard tracking of 2012, while the black points represent the efficiency used by adding the 2 new iterations describe above. The effect of recovering hits for tracks associated to muons is clearly visible.

Efficiency Plots Per Iteration

Plot Description
effic_vs_vertpos.png TTbar sample with average pileup of 20 (no out-of-time PU) reconstructed with release CMSSW620 (2012-like tracking). Denominator: simulated tracks with pT>0.8 GeV, |eta|<2.5, production vertex with z<150 cm; Numerator: denominator + associated to a reconstructed track; Association: reconstructed track has >75% of hits from the simulated track. The plot shows that iter0, iter1 and iter2 reconstruct prompt tracks, iter3 and iter4 reconstruct non-prompt tracks produced within the pixel region, iter5 and iter6 reconstruct displaced tracks outside the pixel region.
efficPt.png TTbar sample with average pileup of 20 (no out-of-time PU) reconstructed with release CMSSW620 (2012-like tracking). Denominator: simulated tracks with |eta|<2.5, production vertex with R<20 cm and z<30 cm; Numerator: denominator + associated to a reconstructed track; Association: reconstructed track has >75% of hits from the simulated track. The plot highlights the contribution from Iter1 at pT<0.5 Iter3.

Timing Plots Per Iteration

Plot Description
timingNew_iter_comp_pu_20_40_60.png TTbar sample with average pileup of 20, 40 or 60 (no out-of-time PU) reconstructed with release CMSSW620 (2012-like tracking). Shown is time per iterative tracking step, normalized to the time of iter0 for <PU>=20. Note that iter0, iter1 and iter3 (seeded by pixel triplets) scale linearly (or almost linearly) with PU, while iter2 (pixel pairs), iter4 (mixed triplets), iter5 and iter6 (strip pairs) quickly degrade with increasing pileup.
timingNew_track_comp_pu_20_40_60.png TTbar sample with average pileup of 20, 40 or 60 (no out-of-time PU) reconstructed with release CMSSW620 (2012-like tracking). Shown is time per track in each iterative tracking step, normalized to the time of iter0 for <PU>=20; only tracks passing the highPurity quality flag are counted. Note that iter0, iter1 and iter3 (seeded by pixel triplets) are reasonably PU-independent, while iter2 (pixel pairs), iter4 (mixed triplets), iter5 and iter6 (strip pairs) quickly degrade with increasing pileup.
timingNew_step_comp_pu_20_40_60.png TTbar sample with average pileup of 20, 40 or 60 (no out-of-time PU) reconstructed with release CMSSW620 (2012-like tracking). Shown is time per tracking stage, normalized to the time of building at <PU>=20. Fitting scales linearly with PU (i.e. number of reconstructed tracks is linear). Due to increasing combinatorics, seeding and building scale non linearly. Particularly important is the degradation of seeding performance as seeding is the input to building and a large contribution to the time growth in building is due to the increase in fake seeds. Timing of selection is negligible.

Masked Hits Per Iteration

Plot Description
barrel_hits_per_iter.png TTbar sample with average pileup of 20 (no out-of-time PU). Reconstructed with release CMSSW620 (2012-like tracking). After each iteration, the hits associated to high quality tracks are masked so that the combinatorics is reduced for next iterations. Shown is the fraction of unmasked hits per tracking stage in barrel detectors. The largest reduction comes after iter0, which reconstructs the majority of tracks. Total reduction after all steps is ~45% for pixel barrel layer1, ~30% for TIB layer 1 and ~20% for TOB layer 1. Single sided hits only are considered for TIB and TOB.
forward_hits_per_iter.png TTbar sample with average pileup of 20 (no out-of-time PU). Reconstructed with release CMSSW620 (2012-like tracking). After each iteration, the hits associated to high quality tracks are masked so that the combinatorics is reduced for next iterations. Shown is the fraction of unmasked hits per tracking stage in forward detectors. The largest reduction comes after iter0, which reconstructs the majority of tracks. Total reduction after all steps is ~40% for pixel disk 1, ~25% for TID disk 1 and ~20% for TEC disk 1. Single sided hits only are considered for TID and TEC.
mtch_tib1_per_iter_pu.png TTbar sample with average pileup of 20, 40 or 60 (no out-of-time PU). Reconstructed with release CMSSW620 (2012-like tracking). After each iteration, the hits associated to high quality tracks are masked so that the combinatorics is reduced for next iterations. Shown is the number of unmasked TIB1 hits per tracking stage, normalized to the number of hits with <PU>=20. Shown in red are single-sided strip hits (mono), shown in black are double-sided strip hits (matched). Note that the number of matched hits increase more with PU due to ghost hits when multiple single sided hits are present on one module. On the other hand, this has the effect that a larger fraction of hits is masked.
ghost_pu.png TTbar sample with average pileup of 20, 40 or 60 (no out-of-time PU). Reconstructed with release CMSSW620 (2012-like tracking). Ghost matched hits are produced when more than one track crosses a glued strip sensor. In this plot Nghost is defined as Nmatched-Nmono (correct in the reasonable approximation of high hit efficiency and low noise). Shown here is the ratio of ghost matched hits and mono hits in double-sided barrel layers. In TIB1 (where the occupancy is higher) for PU≥40, the number of ghost hits exceeds the number of true (mono) hits.

-- GiuseppeCerati - 26 May 2014

Topic attachments
I Attachment History Action Size Date Who Comment
PNGpng TkLay7_from_TMA_eta.png r3 r2 r1 manage 27.5 K 2014-05-28 - 09:03 MarcoRovere  
PNGpng barrel_hits_per_iter.png r2 r1 manage 14.6 K 2014-05-26 - 14:53 GiuseppeCerati  
PNGpng eff_eta_dr030e030_corr.png r2 r1 manage 25.9 K 2014-05-28 - 09:03 MarcoRovere  
PNGpng eff_vtx_dr030e030_corr.png r2 r1 manage 27.2 K 2014-05-28 - 09:04 MarcoRovere  
PNGpng efficPt.png r3 r2 r1 manage 13.8 K 2014-05-30 - 17:34 GiuseppeCerati  
PNGpng effic_vs_vertpos.png r3 r2 r1 manage 16.1 K 2014-05-30 - 17:33 GiuseppeCerati  
PNGpng forward_hits_per_iter.png r2 r1 manage 14.4 K 2014-05-26 - 14:52 GiuseppeCerati  
PNGpng ghost_pu.png r3 r2 r1 manage 11.7 K 2014-05-30 - 17:36 GiuseppeCerati  
PNGpng mtch_tib1_per_iter_pu.png r3 r2 r1 manage 17.3 K 2014-05-30 - 17:35 GiuseppeCerati  
PNGpng timingNew_iter_comp_pu_20_40_60.png r2 r1 manage 12.2 K 2014-05-26 - 14:53 GiuseppeCerati  
PNGpng timingNew_step_comp_pu_20_40_60.png r2 r1 manage 12.5 K 2014-05-26 - 14:54 GiuseppeCerati  
PNGpng timingNew_track_comp_pu_20_40_60.png r2 r1 manage 13.7 K 2014-05-26 - 14:54 GiuseppeCerati  
Edit | Attach | Watch | Print version | History: r5 < r4 < r3 < r2 < r1 | Backlinks | Raw View | WYSIWYG | More topic actions
Topic revision: r5 - 2014-06-02 - GiuseppeCerati
 
    • Cern Search Icon Cern Search
    • TWiki Search Icon TWiki Search
    • Google Search Icon Google Search

    CMSPublic All webs login

This site is powered by the TWiki collaboration platform Powered by PerlCopyright & 2008-2023 by the contributing authors. All material on this collaboration platform is the property of the contributing authors.
or Ideas, requests, problems regarding TWiki? use Discourse or Send feedback