tracking software status tracking software status norman a. graf for the tracking group ( prime...
TRANSCRIPT
DD
Tracking Software StatusTracking Software Status
Norman A. GrafNorman A. Graf
for the tracking groupfor the tracking group( prime architect: David Adams )
Software and Data Analysis Workshop
PragueSeptember 23, 1999
DD IntroductionGoal of global tracking is to find and fit
the tracks in a D0 Event using event data from one or more of the D0 subdetectors.
The input data Event is a collection of clusters from each subdetector.
The output is a collection of global tracks where each track contains a list of clusters and one or more kinematic fits based on these clusters.
Group is also responsible for overseeing algorithms for the Level 3 Trigger.
DD L3 CFT TrackingLink-and-Tree algorithm has been
implemented which successfully finds tracks in the CFT.
Can search in selected region, or all CFT.
Adjacent fibers (up to 3) are clustered.Search for tracks as circular arcs passing
near the origin using axial fiber information.
Matching stereo fiber hits are then found.A fast circle fit is performed in the x-y
plane, along with a straight-line fit in s-z.
DD L3 CFT TrackingTracking efficiencies and
resolutions have been studied; clearly functions of event complexity.
Efficiency versus Purity is clearly an issue, but efforts underway to quantify.
Timing obviously a functon of minimum pT cut, but also dependent on samples used; currently under investigation.
DD L3 CFT Tracking tt +2MBClusters...
Links...
Trees...
Tracks...
DoneDone...Done
DD L3 TrackingThe first version of the tracking
framework has been completed. l3ftrack_base Base classes. l3ftrack_smt SMT extensions l3ftrack_mc MC extensions
SMT tracking tool exists. Reconstructs tracks within a region or
in the entire detector. Uses algorithm similar to CFT link-and-
tree. Reasonably fast and efficient.
MC Tracking tool exists.
DD L3 Tracking PlansImprove existing CFT algorithm.
Correct high-pT inefficiencies in r-. Make r-z code more efficient.
Incorporate CFT algorithm into the framework.
Add extension capability from CFT to SMT. Provide multiple algorithms with varying
speed and pT efficiencies.Use L3 unpacking code as it becomes
available from subdetectors.Test, debug and optimize.
DD TRF++TRF++ is an extensible object-
oriented framework for finding and fitting tracks in particle physics detectors. Written in C++. Provides extensive base class libraries. Modular, with acyclic package
dependencies. Track-finding strategy based on a road-
following algorithm. Track-fitting based on a Kalman-filter
algorithm
DD Global Tracking System
The global tracking system (GTR) defines the following event data: GTrackChunk : holds reconstructed
tracks. McTrackChunk : holds Monte Carlo tracks.
and the following packages: GtrFind: finds tracks GtrMcFill: generates MC tracks. GtrClusterSim: generates MC subdetector
clusters. GtrTuple: matches found and MC tracks
and generates analysis ntuple.
DD Data FlowFollowing figure shows the flow of
data from the Central Fiber Tracker through the global tracking system.
Data is indicated within horizontal bars.
Reconstructors which operate on or create Chunks are contained in packages indicated in ellipses.
The GTR system is shown in red, the CFT system is shown as an example implementation and is in green.
DD
DD Offline Tracking Status
DD
DO UPGRADE TRACKING SYSTEM
Central Tracking Regions
The Central Tracking Volume is divided into three regions of interest:
DD Central RegionTracking in Central Region has been
available for quite some time. Path requires all 16 CFT layers to be hit.
Internally simulated tracks are found with 100% efficiency.
Effects of MS are correctly handled in the thin-scatterer approximation.
Tracking efficiency for GEANT simulated tracks had been less than 100%, even for high momentum single muons. Problem was in CFT digitization.
DD Central Region Tracking Tests
Start with sample of high momentum tracks in full CFT fiducial region. 50GeV pT
z=dca=0, -1<tan(λ)<1, 0<φ<2 π Reconstruction efficiency
1979/1991 events (99.40%) Good track fit χ2 . Good track match χ2 .
Work starting to develop additional clustering algorithms. Current algorithm is simple Nearest-
Neighbor.
DD Track Quality Metrics
The quality of the reconstructed tracks is represented by the following quantities:
Track χ2 : For the CFT, requiring 16 hits results in
a track χ2 with 11 d.o.f. (5 constraints).
Probability to exceed χ2: Produces a distribution flat between 0
and 1 if the values really are χ2 distributed.
DD Track Fit χ2
DD Track Quality Metrics
Match χ2 : χ2, i.e. ( fit-MC )2 /2
fit , which should be distributed with 5 d.o.f.
Parameter Significances: Normalized residuals, ( fit-MC )/ ,
giving gaussian distributions with mean of 0 and width of 1.
DD Track Match χ2
DD Central Region Tracking Tests
Start with sample of high momentum tracks in full CFT fiducial region. z=dca=0, -1<tan(λ)<1, 0<φ<2 π
50GeV pT : find 1980/1991 events (99.4%)
3GeV pT : find 1976/2000 events (98.8%)
1GeV pT : find 1874/2000 events (93.7%)
DD Central Region Tracking Tests
Multimuon sample in CFT fiducial volume. 0.5 GeV< |pT | < 10 GeV dca=0, z gaussian σ=28cm. π/4 < Θ < 3π/4 0<φ<2 π 4856/4959 found (98%)
Tracking done only in CFT. Adding SMT to tracking in multi-track
events causes slight inefficiency and poorly understood track and match χ2.
DD Central Region Tracking Tests
Analyze Z μμ sample with underlying event. Use Isajet events with underlying event and
require both muons to pass through the CFT fiducial volume.
52/100 events pass cuts 104 muons with pt>20GeV
103/104 muons found.Generating larger samples of Z μμ
with 0, 1 and 2 additional minimum bias events to study efficiencies and resolutions as a function of hit density.
DD Forward TrackingTracks pass through SMT Barrels
and some portion of F and H disks.Start tracks with hits in H or F disks.
DD Forward TracksRequire tracks to have at least 4 hits.
Most hits are 2D, therefore 3 hits constrain the track parameters.
Only one miss allowed in track.Constrain track to come from beam
axis and apply minimum pT cut to improve performance.
Prune track list at each layer to remove tracks with hits in common.
Studies conducted with GEANT samples of 10GeV muons.
DD Overlap TrackingWork started to develop appropriate paths.Tactic is to use the equivalent of the
current CFT Path and remove successive outer layers.
Paths orthogonal to existing path by requiring z of stereo hits to lie appropriately close to edge.
Object-reader capabilities of GTR system allow paths to be defined in external file. No coding required!
First tracks have been found in single muon GEANT files.
DD Muon TrackingThe trf interface for the muon
detector and muon hits has been coded.
Modifications to the gtr system made.WAMUS internally generated events :
~100% efficiency for 10 hit planes. ~55% overall acceptance*efficiency
Plan to: Analyze GEANT data. Integrate FAMUS. Use field map (TIM package).
DD Muon Tracking Results
DD Material InteractionEffects of Multiple Scattering and energy
loss are handled via Interactors.Thin Surface Multiple Scattering on
cylindrical and planar (xy and z) surfaces is released.
TRF interacting detectors currently account for MS on measurement surfaces and some passive elements ( beampipe, SMT support, solenoid)
Energy loss code written, being incorporated into the interacting detectors.
All parameters under RCP control.
DD Interacting Propagators
Current Propagators simply transport tracks from one surface to another. Interactions (MS & dE/dx) are handled by the surfaces.
Work is underway to develop Propagators which allow tracks to be arbitrarily transported, and have the track modified by any surfaces it may have crossed in the interval.
D0Propagator exists as well as CFTPropagator implementation.
Work proceeding for other subdetectors.
DD StatusGlobal tracking software system is composed
of a number of packages which define and implement the interface between the individual detector components (e.g. CFT) and the actual track finding and fitting software (TRF++).
The system defines and manages the interface to global tracks, which are composed of a list of constituent clusters and a list of kinematic fits.
MC tracks are also defined and utilities exist to facilitate the association to and comparison between simulated and found tracks.
DD Short Term GoalsAdd and integrate central and forward.Add and integrate muon system.Improve performance for complicated
events.Optimize fitting to account for material.Account for non-uniform magnetic
fields.Produce tracks with optimal fits
everywhere.Add Monte Carlo Fitter.Develop and release L3 filters.
DD Tracking on the Web
User’s Guide How to generate, find and analyze.
Software Links to GTR, TRF and subdetectors.
Projects What is (and isn’t) being done.
Results Canonical Plots.
Project Status Milestones and schedules.
DD ConclusionsThe tracking software continues to
improve both in quality and performance. Doing more and doing it better.
More people are (slowly) becoming involved.
Just starting to seamlessly incorporate subdetectors.
Have not yet demonstrated capability to reconstruct “real” events in reasonable time with requisite efficiency.
DD ConclusionsWelcome contributions from non-
coders: Systematically investigate
efficiencies, resolutions and timing. Generate “Physics” samples and
analyze standard ntuples. Contribute to path algorithms.
Many of the building blocks are in place, but much more work still needs to be done to have a fully working system.
Optimize, optimize, optimize.