Christian Grefe – July 15, 2010 Motivation (cont.)

Iterative Tracking in LCSim
July 20, 2010
Christian Grefe
CERN, Bonn University
Motivation
• Currently tracking in LCSim is steered by a list of strategies
• Each strategy defines the roles of each layer
• 3 seed layers + 1 confirm layers + several extension layers (usually all other)
• χ2 cut for the helix fit
• A strategy list should cover the minimum number of layer combinations
needed to find all tracks (above certain pt and minimum nHits req)
• This is automatically done using a training sample
• Weighting layers is possible to favor certain strategies (i.e. inside-out)
• All Strategies are equal
• All are tried and a list of candidates is formed
• If candidates share hits the “best” is chosen (based on χ2 of fit)
Christian Grefe – July 15, 2010
Page 2
Motivation (cont.)
• Iteration allows another level of steering
• Each strategy list is used to form tracks and all used hits are removed
• Following strategy lists suffer from less confusion
• Possible scenario to increase performance
• Start with well defined tracks (many hits, moderate χ2-cut)
• Find low pt tracks (less hits, tighter χ2-cut)
Christian Grefe – July 15, 2010
Page 3
Status
• Implemented IterativeSeedTracker as derived class from SeedTracker
• Small changes to HitManager (does not break SeedTracker)
• No other changes needed
• Compatible with existing SeedTrackerDiagnostics drivers
• First tests done
Christian Grefe – July 15, 2010
Page 4
XML Steering Example
• Usage almost identical to SeedTracker
<lcsim xmlns:lcsim="http://www.lcsim.org/schemas/lcsim/1.0"
xmlns:xs="http://www.w3.org/2001/XMLSchema-instance"
xs:noNamespaceSchemaLocation="http://www.lcsim.org/schemas/lcsim/1.0/lcsim.xsd">
<inputFiles>
<file>/afs/cern.ch/eng/clic/work/events/clic01_sid/uds91_09_.slcio</file>
</inputFiles>
<control>
<numberOfEvents>-1</numberOfEvents>
<verbose>false</verbose>
<printDriverStatistics>true</printDriverStatistics>
<printDriversDetailed>true</printDriversDetailed>
</control>
<execute>
<driver name="hitMaker" />
digitization driver
<driver name="iterativeSeedTracker" />
</execute>
<driver name="hitMaker"
type="org.lcsim.recon.tracking.seedtracker.steeringwrappers.SiD02HybridHitMakerNoCCD" >
</driver>
<driver name="iterativeSeedTracker"
type="org.lcsim.recon.tracking.seedtracker.IterativeSeedTracker" >
<inputCollection type="String">HelicalTrackHits</inputCollection>
<trkCollectionName type=“String”>Tracks</trkCollectionName>
<strategyFile type="String">StrategyList1.xml</strategyFile>
<strategyFile type="String">StrategyList2.xml</strategyFile>
<timingPlots type="boolean">true</timingPlots>
</driver>
</drivers>
Use as many as you want,
</lcsim>
order is important!
Christian Grefe – July 15, 2010
Page 5
First Tests
• 1000 uds @ 3 TeV, first use 7Hit strategies and 6Hit strategies later
• The 6 hit strategy picks up some extra tracks
Christian Grefe – July 15, 2010
Page 6
First Tests (cont.)
• No notable impact on track reconstruction time
Christian Grefe – July 15, 2010
Page 7
Outlook
• Extend StrategyBuilder to allow for “iterative training”
• Give several prototype strategies with individual definition of findability (nHits,
pT, d0, z0, χ2-cut)
• Train strategy lists successively
• Remove tracks that are findable by this strategy list before training the next
• (This is only iterative training on different phase spaces, can not simulate gain
by removing confusion because it is on MC level)
• Extensive testing!
• Impact on tracking efficiency / purity
Christian Grefe – July 15, 2010
Page 8