Tracking for 2015
Target
Verify the performance of LHCb Tracking algorithms with the new beam conditions in 2015 (

=13TeV, bunch spacing=25,50 ns) and be prepared for various data-taking scenarios.
Useful samples
There are many useful samples in bookkeeping that can be used to perform your studies:
Data Samples
- 2011, 25ns: /LHCb/Collision11_25/Beam3500GeV-VeloClosed-MagDown/Real Data/Reco11a/90000000 ( Full stream )/RAW : these events are taken without any trigger. No data is actually on disk. If needed, we can request to copy them temporarily to disk.
- 2012, 25ns: /LHCb/Collision12_25/Beam4000GeV-VeloClosed-MagDown/Real Data/90000000 ( Full stream )/RAW : useful for tracking performance studies
- 2012, 25ns: /LHCb/Collision12_25/Beam4000GeV-VeloClosed-MagDown/Real Data/Reco14/Stripping20/90000000 ( Full stream )/(various streams) : useful for comparing physics signal yields and efficiencies
- 2012, High Luminosity: /LHCb/Collision12hl/Beam4000GeV-VeloClosed-MagDown/Real Data/90000000 ( Full stream )/RAW : useful for tracking performance studies
- 2012, High Luminosity: /LHCb/Collision12hl/Beam4000GeV-VeloClosed-MagDown/Real Data/Reco13g/Stripping20/90000000 ( Full stream )/(various streams) : useful for comparing physics signal yields and efficiencies [beware: Reco13g, not Reco14]
N.B.: The RAW data samples are on tape, they can be duplicated on disk upon request. For tracking studies one can refer to the DST streams in Stripping20.
MC Samples
N.B.: There is an issue with the DST samples. The "Link/Raw/(Detector)/Clusters" containers are not filled, therefore clusters truth-matching is not possible. This samples cannot therefore be used for Tracking performance measurements using the
PatChecker algorithm. XDIGI samples need to be used instead.
- 13TeV, RunII, MinBias: /MC/Dev/Beam6500GeV-RunII-MagDown-Nu1.9-25ns-Pythia8/Sim08f/30000000 ( minbias )/XDIGI : for minimum bias studies
- 13TeV, RunII, Physics: /MC/Dev/Beam6500GeV-RunII-MagDown-Nu1.9-25ns-Pythia8/Sim08f/( evtid )/XDIGI : a few physics channels for physics studies
All of this samples are generated for magnet Up and Down and for different running conditions:
- 25ns: Nu=1.6,1.9
- 50ns: Nu=1.6,2.7
- 8TeV, RunI, MinBias: /MC/2012/Beam4000GeV-2012-MagDown-Nu2.5-Pythia8/Sim08e/Trig0x409f0045 /30000000 ( minbias )/XDIGI : for minimum bias studies
- 8TeV, RunI, Physics: /MC/2012/Beam4000GeV-2012-MagDown-Nu2.5-Pythia8/Sim08e/Trig0x409f0045 /( evtid )/XDIGI : a few physics channels for physics studies
All of this samples are generated for magnet Up and Down.
Condition Databases
The suggested condition databases for running on these samples are (11/2/2014):
- CONDDB: cond-20130710
- DDDB: dddb-20130929-1
- SIMCOND: sim-20131023-vc-md100, sim-20131023-vc-mu100
Ongoing activities and responsible people
This is a list of specific topics to see who is currently working on what, broader topics and responsible people are listed below.
Topic |
Description |
Contact |
Link |
Combinatorics in T-stations |
Check if hits can be (better) preselected |
??? |
??? |
Fitting framework |
Provide fitting framework for all chi2-fits, vectorisable |
Manuel Schiller,... |
??? |
Vectorising pat reco |
Vectorise the hotspots of the pattern recognition |
Gerhard Raven, ... |
??? |
Speeding up Kalman filter |
Can we speed up the Kalman filter, using vectorised libraries/code |
??? |
??? |
Adaptive windows |
Can we use adaptive windows in the patreco (depending on position, event multiplicity, ...) |
??? |
??? |
Floats |
Rewrite the patreco using float instead of doubles |
??? |
??? |
Online -> Offline tracks |
Can/should offline tracks be constructed from online tracks? |
??? |
??? |
Update / Check params in pat reco |
Update all parametrisations in the patreco algorithms to latest MC |
??? |
??? |
Detailed Wish List
Topic |
Algorithm |
Proponent |
Comment |
Timescale |
Started |
Finished |
Responsible |
Pattern Recognition Fit |
General |
Manuel Schiller |
|
|
|
|
|
Trade ghost rate for efficiency |
Velo |
David Hutchcroft |
almost at limit |
|
|
|
|
Lite cluster decoding |
Velo |
Kurt Rinnert |
|
|
|
|
|
double vs. float |
Velo |
Kurt Rinnert |
|
|
|
|
|
2nd metal layer effect |
Velo |
|
|
|
|
|
|
Clustering thresholds |
Velo |
|
|
|
|
|
|
Improve clustering |
Upstream |
Barbara Storaci |
|
|
|
|
|
Velo+VeloTT+Forward efficiency loss recovered by matching? |
Upstream |
Barbara Storaci |
|
|
|
|
|
Code review |
Upstream |
Barbara Storaci |
|
|
|
|
|
Optimised search regions |
Upstream |
Barbara Storaci |
|
|
|
|
|
TT addition after Velo+VeloTT+Forward still needed? |
Upstream |
Barbara Storaci |
|
|
|
|
|
Require confirming stereo hit when adding x hit? |
Forward |
Michel De Cian |
|
|
|
|
|
"Precluster" layer |
Forward |
Michel De Cian |
|
|
|
|
|
Better way to sort subranges |
Forward |
Michel De Cian |
|
|
|
|
|
Region and occupancy dependent search windows |
Forward |
Michel De Cian |
|
|
|
|
|
TT hit adding smarter and faster |
Forward |
Michel De Cian |
waiting until TrackMatching in 'obsolete' to simplify interface |
Middle of November |
yes |
yes |
Michel |
Pooled allocation for hits |
Forward |
Michel De Cian |
|
|
|
|
|
floats vs. double |
Forward |
Michel De Cian |
|
|
|
|
|
AOS vs SOA hits |
Forward |
Michel De Cian |
|
|
|
|
|
High threshold bit for silicon detectors |
Forward |
Michel De Cian |
|
|
|
|
|
Ambiguity resolution via pitch residuals? Simfit in x,y? |
Forward |
Michel De Cian |
|
|
|
|
|
Hit cleaning in busiest region of OT |
Seeding |
Manuel Schiller |
|
|
|
|
|
Kill unused OT hits early in the reco |
Seeding |
Manuel Schiller |
|
|
|
|
|
Kill hits from previous/next bunch crossing |
Seeding |
Manuel Schiller |
|
|
|
|
|
Use wire propagation time? |
Seeding |
Manuel Schiller |
|
|
|
|
|
Port upgrade studies? |
Matching |
Michel De Cian |
Tuning and some improvements done. Small improvements still possible, if errors correctly propagated from Velo and Seed |
End of December |
yes |
yes |
Michel |
Backport upgrade version? |
Downstream |
Michel De Cian |
basic version backported, gain smaller than expected, investigate... |
End of year (?) |
yes |
|
Michel |
Improvement of equation of motion of charged particle in magnetic field |
Kalman Fit |
Wouter Hulsbergen, Gerhard Raven |
|
|
|
|
|
Calculate more quickly energy loss and multiple scattering corrections |
Kalman Fit |
Wouter Hulsbergen, Gerhard Raven |
|
|
|
|
|
Node/measurement creation, improve the allocation |
Kalman Fit |
Wouter Hulsbergen, Gerhard Raven |
|
|
|
|
|
TrackProjectors: TrajPoca really needed for straight line hits? |
Kalman Fit |
Wouter Hulsbergen, Gerhard Raven |
|
|
|
|
|
Matrix operations |
Kalman Fit |
Wouter Hulsbergen, Gerhard Raven |
|
|
|
|
|
OT hits in high occupancy regions |
Kalman Fit |
Wouter Hulsbergen, Gerhard Raven |
|
|
|
|
|
OT decoding improvements possible? |
Kalman Fit |
Wouter Hulsbergen, Gerhard Raven |
|
|
|
|
|
Check tracks for hit overlap very fast? |
Kalman Fit |
Wouter Hulsbergen, Gerhard Raven |
|
|
|
|
|
Advanced algorithms monitoring (efficiency, timing, ghosts) |
Monitoring |
Michel De Cian, Manuel Schiller |
|
|
yes |
|
Svende |
High-level Quantities Monitoring |
Monitoring |
|
|
|
|
|
|
Tune ghost probability |
ghost probability |
Michel De Cian |
|
end of 2014 |
|
|
Paul |
Overlap online-offline |
online-offline |
Vava Gligorov |
|
end of January 2015 |
|
|
Sascha |
Implement improved multiple-scattering correction |
Kalman Fit |
Miriam Hess |
|
|
|
|
Miriam |
Tracking Algorithms Performance
Task Force: Manuel Schiller, Michel De Cian, Sebastian Neubert.
The following tables show the output of the
PrChecker for 1000 events of B->K*mumu for different running scenarios:
Additionally it was checked how the behaviour would be for the highest running scenarios when cutting on the SPD multiplicity:
- 25ns, Nu 2.6, SPD < 450: PrCheckerReco15DevKstmumuMagDown25nsNu26SPD450.txt
- 25ns, Nu 2.6, SPD < 600: PrCheckerReco15DevKstmumuMagDown25nsNu26SPD600.txt
- 25ns, Nu 2.6, SPD < 900: PrCheckerReco15DevKstmumuMagDown25nsNu26SPD900.txt
- 50ns, Nu 4.8, SPD < 450: PrCheckerReco15DevKstmumuMagDown50nsNu48SPD450.txt
- 50ns, Nu 4.8, SPD < 600: PrCheckerReco15DevKstmumuMagDown50nsNu48SPD600.txt
- 50ns, Nu 4.8, SPD < 900: PrCheckerReco15DevKstmumuMagDown50nsNu48SPD900.txt
The upstream numbers in all tables are wrong due to a wrong setting.
The following
PrChecker tables (5k events) are with Reco15 for 2015 data taking and obtained from Sim08f samples:
For reference the same with the Sim08e samples:
Velo Tracking
Velo Pattern Recognition contacts:
David Hutchcroft and
Kurt Rinnert
Standalone T reconstruction
T seeding contact:
Manuel Schiller
Long Track Reconstruction
There's a presentation by Paul
https://indico.cern.ch/conferenceDisplay.py?confId=280286
It can be reproduced with the instruction in Phys/TrackingScripts/TrackEffMCStudies/README. The Brunel options produce two TTrees, one comparing the
TrackBestTrackCreator to the
TrackEventCloneKiller, one for comparing
PatForward with
PatMatch.
Forward Tracking
Forward contacts:
Michel De Cian
Track Matching
Downstream Tracking
Downstream contacts:
Adam Davis,
Mike Sokoloff and
Mika Vesterinen
Upstream Tracking
VeloTT contacts:
Barbara Storaci and
Marco Tresch
Clone Killer
Open question: Why are forward tracks preferred over match tracks after the Kalman filter? Better chi2? More hits on track?
Possible Improvements
There are a few studies performed on the tracking for the upgrade that may be ported to the current tracking:
Running the tracking on top of HLT found tracks:
Tracking Efficiency
contact:
Michael Kolpin
Detectors Performance
Velo
TT
IT
contact:
Mark Tobin
OT
contact:
Francesco Dettori and
Niels Tuning
Further checks are ongoing on the drift time.
Outer Tracker performance with 25ns has been studied using the data recorded in 2012. These studies are outlined in the
Performance of the LHCb Outer Tracker
.
Muon
Validation
A tool is needed for validating the reconstruction software. Differences among two reconstruction software versions should be analysed by means of tracking specific quantities and physical quantities. Contact:
Svende Braun
(Open) Issues
- The definition of "from B" and "from C" changed in summer (see Rev 159594). The PatChecker / TrackChecker have been updated on 23.1.2014 to only flag daughter particles from mesons and baryons as "from B" or "from C" (which is the correct definition). Without this update there might be slightly lower efficiencies for these categories.
- There may be differences between the simulated and actual detector response with 25ns. Investigation with detector and simulation experts is ongoing.
- The "Link/Raw/(Detector)/Clusters" containers are not filled, therefore clusters truth-matching is not possible on the DST Monte Carlo since Sim08.
Obsolete code
The following classes have not been used since a long time in the tracking and are not forseen to be used in the future. They might become subject to removal (after an announcement and contacting the authors).
All of this code has been put in obsolete (November 15 2014).
Presentations with 2015 tracking contributions
- Maurizio's kick-off presentation Kick-off for 2015 improvements
- Ideas by Manuel Schiller: Possible improvements of Pattern Recognition
- Daniel's (explicit) vectorsation for the upgrade: Explicit vectorisation for PrPixel
- Manuel's brainstorm summary and first developments Output of the brainstorming for Reco15
- Svende's presentation of the new PrChecker SuperTrackEffChecker
- Espen's improvement for VeloTT: VeloTT improvements
- Tracking & Alignment meeting, November 20: T&A, Nov 20
- Tracking & Alignment meeting, November 27, 2014: T&A, Nov 27, 2014
- Tracking & Alignment meeting, January 8, 2015: T&A, Jan 8, 2015
- Tracking & Alignment meeting, January 13, 2015: T&A, Jan 13, 2015
- Tracking & Alignment meeting, January 27, 2015: T&A, Jan 27, 2015
- Tracking & Alignment meeting, February 5, 2015: T&A, Feb 5, 2015
- Tracking & Alignment meeting, February 10, 2015: T&A, Feb 10, 2015
- Tracking & Alignment meeting, February 19, 2015: T&A, Feb 19, 2015
--
MichelDeCian - 30 Jan 2015