recent improvements in track reconstructionbmangano/cms_talks/byboris/tracking... · g.petrucciani)...
TRANSCRIPT
B.ManganoOpening Session- CMS Week - February 2008
Boris Mangano
(University of California, San Diego)
on behalf of the tracking group (Tracker Project)
Recent Improvements in TrackReconstruction
B.ManganoOpening Session- CMS Week - February 2008
Outline
• one slide description of track reconstruction in CMS
• improvements in tracking
- description
- effects on tracking performance
- effects on “physics”
• evolution of timing performance over past CMSSW releases
B.ManganoOpening Session- CMS Week - February 2008
Tracking modules
Trajectory seeding
Trajectory fitting
Trajectory building
Initial estimate of trajectory parametersfrom a small subset of trackermeasurements (hits on seeding layers).
An iterative process which collects all themeasurements associated to the samecharged particle.
Estimate of final track parameters from the fitof the full set of measurements associated tothe same charged particle.
Track collectionfiltering
new
Removal of “ghost” tracks (fakes) + qualityfilter.
B.ManganoOpening Session- CMS Week - February 2008
glob
al e
ffici
ency
|η|
Porting of code from ORCA
-•- single muon, pt = 1 GeV
-•- single muon, pt = 10 GeV
-•- single muon, pt = 100 GeV
ORCA performance
CMSSW_1_3_X
performance
ORCA tracking performance ~ re-obtained with CMSSW_1_3_X release(beginning of 2007)
Actually some new features were already introduced: e.g. mixed seedinggave better tracking efficiency in the high eta region.
B.ManganoOpening Session- CMS Week - February 2008
New Feature: Tracking with “overlaps”
Original trajectory builder was able to collect only onemeasurement per layer per track.
Nevertheless the tracker has overlapping sensors:
sometimes there is more than one measurementproduced by the same charged particle on the sametracker layer
B.ManganoOpening Session- CMS Week - February 2008
standard building
building “with overlap”
hit c
olle
ctio
n ef
ficie
ncy
|η|
New Feature: Tracking with “overlaps”
New trajectory builder is able tocollect hits on all the overlapping
sensors of the CMS tracker
hit collection efficiency:
from ~88% to 98%
new trajectory builder
old trajectory builder
fully integrated and used by defaultin CMSSW_1_7_X
Thanks to:
- W.Adam (Wien)
- P.Lenzi (INFN Florence)
B.ManganoOpening Session- CMS Week - February 2008
New Feature: Handling of Magnetic FieldIn-Homogeneities During Track Fit
-•- single muon, pt = 1 GeV
-•- single muon, pt = 10 GeV
-•- single muon, pt = 100 GeV
|η|
approximate handling ofmagnetic fieldinhomogeneities produces abias in the reconstructedvalue of transversemomentum
Effect is apparent in high etaregion of the tracker
average (pt reco - pt sim)/pt error
B.ManganoOpening Session- CMS Week - February 2008
-•- single muon, pt = 1 GeV-•- single muon, pt = 10 GeV-•- single muon, pt = 100 GeV
|η|
Thanks to:
- T.Todorov (now in ATLAS)
- M.Mulders (CERN)fully integrated and used by defaultin CMSSW_1_8_X
Track parameterspropagator based onRunge-Kutta methodallows to properly take intoaccount the magnetic fieldinhomogeneities duringthe final fit
New Feature: Handling of Magnetic FieldIn-Homogeneities During Track Fit
average (pt reco - pt sim)/pt error
B.ManganoOpening Session- CMS Week - February 2008
J/ψ reconstructed mass
mean = 3.096 GeV
plot by M.Cepeda(CIEMAT)
New Feature: Handling of Magnetic FieldIn-Homogeneities During Track Fit
Bias in the reconstructed J/ψ mass iscompletely removed thanks to theRunge-Kutta propagator
B.ManganoOpening Session- CMS Week - February 2008
CMSSW_1_7_X
CMSSW_1_7_0+ 7 fixes/improvements
|η|
aver
age χ2 / n
dfAdditional New Features: Impact on
Quality of Reconstructed Track
7 independent new features oftrack reconstruction haveexceptionally improved the χ2 ofreconstructed tracks
Individual contributions are shown inthe following slides
QCD 80-120sample
B.ManganoOpening Session- CMS Week - February 2008
|η|
aver
age χ2 /n
df
default tracking inCMSSW_1_7_X
+ updated materialbudget parameterizationfor track fitting Thanks to:
- A.Bocci (SNS Pisa)
- W.Adam (Wien)
Additional New Features: Impact onQuality of Reconstructed Track
QCD 80-120sample
B.ManganoOpening Session- CMS Week - February 2008
+ better track anglehypothesis duringclusters re-fitting
|η|
aver
age χ2 / n
df
Thanks to:
-G.Cerati (INFN Milano)
-C.Genta (INFN Florence)
default tracking inCMSSW_1_7_X
Additional New Features: Impact onQuality of Reconstructed Track
QCD 80-120sample
B.ManganoOpening Session- CMS Week - February 2008
+ splitting of matchedhits before final fit of thetrack
|η|
aver
age χ2 /n
df
Thanks to:
-G.Cerati (INFN Milano)
-C.Genta (INFN Florence)
default tracking inCMSSW_1_7_X
Additional New Features: Impact onQuality of Reconstructed Track
QCD 80-120sample
B.ManganoOpening Session- CMS Week - February 2008
+ Runge-Kuttapropagator during finalfit of the track
|η|
aver
age χ2 /n
df
default tracking inCMSSW_1_7_X
Additional New Features: Impact onQuality of Reconstructed Track
QCD 80-120sample
Thanks to:
- T.Todorov (now in ATLAS)
- M.Mulders (CERN)
B.ManganoOpening Session- CMS Week - February 2008
+ new pixel parametersestimator based ontemplates
|η|
aver
age χ2 / n
df
Thanks to:
-M.Swartz, D.Fehling,G.Giurgiu, P.Maksimovic(JHU)
- V.Chiochia (UniversityZurich)
default tracking inCMSSW_1_7_X
Additional New Features: Impact onQuality of Reconstructed Track
QCD 80-120sample
review of the severalbenefits for trackingdue to new pixel CPEare shown inCMS NOTE 2007-033
B.ManganoOpening Session- CMS Week - February 2008
+ tool for rejection of
outlier hits
|η|
aver
age χ2 / n
df
Thanks to:
G.Cerati (INFN Milano)
default tracking inCMSSW_1_7_X
Additional New Features: Impact onQuality of Reconstructed Track
QCD 80-120sample
Impact of outlier rejection ontracking efficiency, fake rateand estimation ofparameters at vertex will bedescribed in a note, currentlyin preparation
B.ManganoOpening Session- CMS Week - February 2008
+ finer description of
material budget for final fit
of tracks
|η|
aver
age χ2 /n
df
Thanks to:
A.Bocci (SNS Pisa)
default tracking inCMSSW_1_7_X
Additional New Features: Impact onQuality of Reconstructed Track
QCD 80-120sample
B.ManganoOpening Session- CMS Week - February 2008
old standard tracking + newtrack collection filter
old standard tracking
QCD 80-120sample
old standard tracking+ new track collectionfilter
old standard tracking
CMSSW_16X
new features: track collection filterbased on vertex compatibility
Thanks to:-P.Azzurri (SNS Pisa)-M.Pioppi, P.Janot (CERN)-F.Ambroglini (INFN Perugia)
|η|
|η|Fake rate is almost completely removedwithout spoiling the track reconstructionefficiency
selection is based on d0 and dz(and significances) respect toreconstructed primary vertex
efficiency and fake forpt > 0.9 GeV
B.ManganoOpening Session- CMS Week - February 2008
New Pixel CPE and New Track Filter:effects on b-tagging
QCD 50-120sample
b-tagging efficiency
mis
-tag
Tests on b-tagging byM.Swartz, D.Fehling,G.Giurgiu, P.Maksimovic(JHU) and T.Boccali (INFNPisa)
track counting, 2nd track (w/o new CPE)
track counting, 2nd track (new pixel CPE)
track counting, 3rd track (w/o pixel CPE)
track counting, 3rd track (new pixel CPE)
Impact on btagging performance are impressive forboth track counting configurations.
The btagging algorithms have not been changed
B.ManganoOpening Session- CMS Week - February 2008
new features: track collection filterbased on vertex compatibility
studies and plot byM.Lebourgeios (UCSD)
“cut-based” loose electrons:isolation performance
Lowering the tracking fake rate, the electron isolation performanceimproves (without any changes in the isolation algorithm itself!)
B.ManganoOpening Session- CMS Week - February 2008
old standard tracking
new trackingconfiguration
QCD 80-120sample
new features: seeder based on tripletsand vertex-constrained hit pairs
pt spectrum of reconstructedtracks can extended towardlower values:
New seed generator is based ontriplets of hits + hit pairs constrainedby the primary vertex.
Both high efficiency and high seedcollection purity are achieved
up to +65% reconstructed trackincreasing only slightly thealgorithm timing
Thanks to:
-M.Konecki (Warsaw)
-G.Boudoul, T.Legrand (IPNL Lyon)
-G.Petrucciani (SNS Pisa)
normalizeddistributions
B.ManganoOpening Session- CMS Week - February 2008
Evolution of Timing PerformanceAlmost factor 3 reduction in timing:
SMatrix migration and C++ codeoptimization (V.Innocente,G.Petrucciani)
Almost same timing, despite +65%reconstructed tracks (thanks to newseed generator):
minimum Pt from 0.9 GeV to 0.3GeV
aver
age
time/
even
t (se
cs)
all tests done with:
- BJets_Pt_50-120sample w/o PU
- 2.0 GHz E5335 intelprocessor
- gcc345 compiler
B.ManganoOpening Session- CMS Week - February 2008
Evolution of Timing Performance
mode
old code (before June2007)
global unpacking
regional unpacking
unpacking on-demand
861
86
7
4
- sample: single tau
- time is per event.Averaged on eventswhich pass L2.5
Thanks to:
- M.Wingham, N.Cripps,R.Bainbridge (Imperial college)
- D.Giordano (INFN Bari)
- J-R. Vlimant (UCSB)
More than a factor 100X inreducing the timing of trackerclusters unpacking.
Clear impact on HLT performance
SiStrip clustersunpacking time (ms)
B.ManganoOpening Session- CMS Week - February 2008
Conclusion
• Last 12 months have been rich of activities related to track reconstruction
• The current status of the tracking software is well beyond what was availableafter the porting from ORCA:
- almost no fake tracks (from 20% to few percent)
- low pt tracking (from 0.9 GeV to 0.3 GeV)
- faster reconstruction (factor 2 in speed, despite the more sophisticated
algorithms)
• Many additional ideas are under investigation and development.
• Current studies are more “real tracker conditions” oriented.
B.ManganoOpening Session- CMS Week - February 2008
Conclusion
Higgs -> GammaGamma
sample
performance for 10pb-1 conditions areexpected to be under control:
-resolutions of parameters at vertex arenecessarily degraded
- efficiency and fake rate are almostexactly the same
tracking for ideal conditions
tracking for alignmentconditions after 10pb-1