GlueX Level-3 Trigger Meeting, Jul 8, 2016

From GlueXWiki
Jump to: navigation, search

Meeting Info.

Meeting Time And Location

13:00 EST (JLab time)

CC F326-327


You can connect using BlueJeans Video conferencing (ID: 198 332 417). (Click "Expand" to the right for details -->):

  1. Make sure you have created a BlueJeans account via your JLab CUE account using this link:

  2. Meeting ID: 198 332 417
    • (you may need to type this in, depending how you connect)

  3. If connecting via Web Browser: click this link (no passcode is needed):

  4. If connecting via iOS or Android App:
    • Use your JLab e-mail address to log in and then enter the meeting ID given above to join the meeting

  5. If connecting via Phone: Dial one of the following numbers and then enter the meeting ID above and hit "#" or "##"

  6. If connecting via Polycom unit:
    • Dial or
    • Enter meeting ID above

Background Information

(Click expand for some links to background information -->)


  1. Announcements
  2. Previous Meeting (Jun 24, 2016)
  3. Benchmarking status (Cristiano)
  4. Simulation Status (Adesh)
  5. Resurrecting BDT for simulation (Justin and Adesh)
  6. Classifying Spring 2016 data (David)
  7. AOT


Attendees: David L. (chair), Cris F., Simon T., Adesh S., Curtis M., Mike W., Justin S.


  • Meeting time is moved to 13:00 every other Friday to alternate with DIRC meetings

Benchmarking Status

  • Cristiano continues to work on benchmarking reconstruction algorithms using janadot
  • Looking at real data run 10913 and simulated data from both Adesh's low-threshold data set and the standard "sim1" 3GeV thresh data set
  • Previously saw large discrepancy between real and simulated data but is now seeing better agreement due to:
    • Newer software (unsure if this was important factor)
    • Better job skimming only physics events
  • MC data has more tracks/showers per event than real data
  • Currently the simulated data does not have a L1 trigger emulation applied
    • Alex S. is working on a L1 trigger emulator using configuration info. recorded in RCDB
  • Justin suggested turning off all but pion mass hypothesis for tracking. That will make it faster and we are unlikely to need multiple mass hypotheses to make L3 decision
  • Mike suggested we could improve tracking speed for L3 decisions if we dropped tracks that look to be low momentum early on (this was done in LHCb to significant benefit)
  • Cris will look at comparing DTrackCandidate results to DTrackTimeBased since the candidate momentum would be the natural place to make this cut

Simulation Status

  • Simulated data set is now available
  • Noticed MIP peak in FCAL has more energy for real data that for MC
  • Justin noted that there may be a trigger bias with the real data
    • This can be investigated by looking at events where the trigger condition is met by a different part of the detector

Resurrecting BDT for simulation

  • Justin found old code from few years ago when initial L3 BDT studies were done and applied it to data simulated with Spring 2016 conditions
  • Performance was similar to what was seen 2 years ago
  • L3 trigger efficiency at 20kHz is ~93.5%
  • For 99% L3 trigger efficiency, we'll reject about 50-60% of background with the present scheme

Classifying Spring 2016 data

  • David showed some results from trying to categorize events as "good" or "bad" from 2 runs from the Spring 2016 data set
    • run 11667=1345A and run 11511=1200A
    • Approximately 1/4 of the physics events satisfied the "good" condition of >4GeV reconstructable energy in the event
  • Following up on a suggestion from Eugene, some attempt was also made to identify what fraction of L1 events were due to a small number of clusters near the beamline in the FCAL
    • These would be characteristic of EM events which are of less interest
    • An estimated 10% of L1 events met this condition
    • Excluding first 2 rings of FCAL blocks from trigger seems to have effectively suppressed many of the EM events
  • The "good" classification was used to train a BDT using lower-level event information as variables
    • Retaining ~99% of events would lead to rejecting ~35% of background
    • Higher level reconstruction variables will be added to see how this might be improved