80 likes | 250 Views
Status of Event Overlay in SiD. February 8, 2011. Christian Grefe CERN, Bonn University. Time to Overlay Events. Investigated huge difference in performance between Marlin and lcsim implementation Culprit: different treatment of SimCalorimeterHits
E N D
Status of Event Overlay in SiD February 8, 2011 Christian Grefe CERN, Bonn University
Time to Overlay Events • Investigated huge difference in performance between Marlin and lcsim implementation • Culprit: different treatment of SimCalorimeterHits • SimCalorimeterHits are in fact a list of several contributions • each one has an energy and a time • getEnergy() returns the sum of the energy of all contributions in the hit • getTime() returns the time of the earliest contribution • Marlin checks getTime() to decide if a hit falls into a readout time window -> overestimation of energy • lcsim checks time of every contribution and only keeps the relevant contributions -> correct energy but very inefficient • Solution: added boolean switch in the lcsim version to select full treatment of SimCalorimeterHits (default: off)
Performance Test • Signal event: Z -> qq (uds), 500 GeV • Background: gg -> hadrons (3.2 events/BX) • Time between BX: 0.5ns • Signal event placed at BX 10 • Integration time for Si detectors: 10ns • Integration time for calorimeters: varied • Java memory limit: 1024MB • Tested overlaying on 100 events in a single lcsim job without problems • Very long readout windows might lead to memory problems!
Performance Test • Excellent performance now • Using the correct treatment of SimCalorimeterHits is at least 100 times slower! • Even when overlaying no background and just cutting on the content of the SimCalorimeterHits in the signal event
How good is the simplification? • Shown is the fraction of energy in a hit deposited only after a certain amount of time (with respect to the first deposition) • Checked only gg->hadron events, physics events might behave differently • Note: every digitization code uses the same simplification EcalEndcap
How good is the simplification? • No visible difference in Hcal and Ecal hits • The effect is smaller than 1% of the deposited energy, even when assuming a very short time window of 1ns HcalEndcap
Next Steps • (Re)check track reconstruction performance with background • Write adapted version of Mark’s driver to prepare collections for Pandora (should be trivial: no TPC with dangling tracks) • Check full chain including Pandora
Update on Tracking • Rich has committed some code, but problems are still obvious • Jeremy has also started working on this issue (mostly cleaning up the treatment of geometry in the tracking code)