HDGeant4 Meeting, January 15, 2019

From GlueXWiki
Jump to: navigation, search

HDGeant4 Meeting
Tuesday, January 15, 2019
3:00 pm
JLab: CEBAF Center, A110
BlueJeans: 968592007

Agenda

  1. Review of minutes from last time
  2. nightly cannot run MC with old run numbers (either geant 3 or geant 4)
  3. Issues on GitHub (all)
  4. Pull Requests on GitHub (all)
  5. Review of photon pull distributions in 1B bggen sample (Jon)

Minutes

Present:

  • CMU: Naomi Jarvis
  • JLab: Thomas Britton, Sean Dobbs, Mark Ito (chair), Simon Taylor, Beni Zihlmann
  • UConn: Richard Jones

There is a recording of this meeting on the BlueJeans site. Use your JLab credentials.

Review of minutes from last time

We went over the minutes from December 18 without significant comment.

Nightly cannot run MC with old run numbers (either geant 3 or geant 4)

Thomas reported that the simulation chain breaks presently (current master branch) at the reconstruction stage (i. e., hd_root). This turns out to be due to a hard check on the presence of the DIRC geometry in halld_recon; lack of the DIRC triggers an exit. This needs to be relaxed. Sean volunteered to make the change.

Issues on GitHub

We reviewed the issues.

More FDC issues? #66

Richard had several comments on this issue.

  • Richard observed a problem, first reported by Beni, where pristine tracks, headed into the meat of the FDC, leave no hits at all. Richard traced it to a Geant4 (G4) bug. Our representation of the FDC has a internal volume in the form of a cone to allow for cables whose volume increases going upstream. The cone has a very small apex angle. This triggers a condition in the G4 code that finds intersections of lines with cones, one that causes an exit if the discriminant of the quadratic formula used in the calculation is too small (less than 10−6. We are left with no hits at all. The problem has been reported to the G4 Collaboration. It has been corrected in our local of the G4 code.
  • The indexing bug in the FDC was discussed at the last Software Meeting.
  • In the course of his HDGeant4 study, Richard documented an anomaly in the track finding, again, first reported by Beni (same meeting as above) where many hits in the FDC are not assigned to the track at the wire-based stage of tracking, but are recovered at the time-based stage. These are all 4 GeV/c tracks going at 5 degrees polar angle. For hdgeant, he sees 10% of the tracks have only less than 10 pseudo hits associated with the track during the wire-based fit out of the 24 possible hits. For hdgeant4, the fraction is about 25%. In both cases the recovery in the time-based fit is observed. This problem has not been resolved, but neither is it a glaring difference between the behavior of hdgeant and hdgeant4, nor does it appear to affect track reconstruction efficiency because of hit recovery at the time-based stage.
  • Richard also mentioned that when he goes to Geant4 version 10.04, he sees a significant increase in the start-up time of hdgeant4, from 20 seconds to 2 minutes. He traced it back to a more aggressive attempt to optimize the internal representation of the geometry to speed up event generation. It is not clear that we see much of this speed up. Richard introduced a new card for control.in, GEOMOPT, that, when set to 0, will bypass optimization and make start-up faster. This is useful when developing code.
  • Richard also reported that the upgrade to 10.04 requires changes to the use of the G4THitsMap class due a change in the class interface. A large fraction of our source code files had to get the the change, as well as instrumentation to allow operation with older versions of Geant4. Yuck.

Photon energy scale #67

Richard has done studies, presented at the last Calorimeter Meeting, that address this issue. He sees no significant difference between the single-shower response of hdgeant vs. that of hdgeant4. He observed some systematic shifts as a function of polar angle, but those are common to both programs.

Sean will revisit his original study and close the issue if appropriate.

many overlaps in the GlueX geometry #60

This issue was discussed at the last Software Meeting. Richard has eliminated overlaps and will close the issue.

Switching to HDGeant4

We seem to be close to being able to recommend HDGeant4 for regular use. For the DIRC data, there will be no choice, of course. Sean recommended going back to our original list of simulations to compare between HDG3 and HDG4 and repeating them with the current code base. Mark suggested that if the studies show agreement, we document those results in a GlueX Doc which can be used to back-up a recommendation for collaboration-wide use of HDGeant4 by default. Thomas will look into firing up his simulation runs again.