Difference between revisions of "HDGeant4 Meeting, December 1, 2020"

From GlueXWiki
Jump to: navigation, search
m (Minutes)
(Minutes: complete the minutes)
Line 34: Line 34:
  
 
   https://halldweb.jlab.org/dist/gluex_centos-8.2.2004_sng3.6_gin2.15.sif
 
   https://halldweb.jlab.org/dist/gluex_centos-8.2.2004_sng3.6_gin2.15.sif
 +
 +
After the announcement we had a discussion on how to transition to GCC compilers beyond 4.8.5, the default for CentOS 7. Moving to more advanced versions of ROOT and Geant4 depend on making this transition.
 +
 +
* Mark has been thinking that by going to CentOS 8, either via container or actual of the new OS, we get to GCC 8.3.1 a big leap forward with an unambiguous prescription for going forward.
 +
* Richard and Sean advocated keeping at CentOS 7 since, after all, it represents the bulk of our installed base, and using DevToolSets from Software Collections to get an upgraded GCC.
 +
* Richard mentioned the use case where a CentOS 7 container is used with CVMFS to get the software. Here the OS and built software are provided; users in this category don't care how the sausage is made.
 +
* Sean mentioned the use case where only ROOT files are analyzed using gluex_root_analysis. All of the build issues are moot in this case as well.
 +
* Beni pointed out that by going to a new version of the OS, we get upgrades on many packages all at once, not just GCC, and eventually all of these upgrades will have to be faced. If we follow the new OS approach, confronting these problems will give us flexibility and save pain down the road.
 +
* Sean advocated an approach where we specify a minimum compiler version as a starting point and work on guidance for every to get there.
 +
* Mark will continue to think about a recommendation for best practices. He found the discussion very useful in this regard.
 +
 +
=== Review of minutes from the last meeting ===
 +
 +
We went over [[HDGeant4 Meeting, September 22, 2020#Minutes|the minutes from September 22]]. Sean mentioned that the details on how to widen the Δt cuts to make efficiency calculations more reliable was taken up at [[GlueX_PID_Meeting,_December_1,_2020|the PID meeting this morning]].
 +
 +
=== Issues on GitHub ===
 +
 +
We spent most of the remaining time discussing [https://github.com/JeffersonLab/HDGeant4/issues/111 Issue #111], '''Difference in Acceptance between G3 and G4''' Colin lead us through the issue. There was a lot of leading to do since he first submitted the issue in May of 2019 and it has remained active, with many updates over time. [The secretary refers the reader to the link above rather than attempt a comprehensive summary.]
 +
 +
There followed an active discussion of ideas related to the issue.
 +
 +
* With respect to the low acceptance in G4 when using the "no heavy light" (NHL)branch, Richard suspects that simply turning off energy deposition for particles heavier than a proton might be too naive an approach. That and the fact that if the energy reported out by the Monte Carlo is not the energy deposited by particles in sensitive volumes, then there needs to be a calibration step as is done with real data. He outlined ideas for adding detail to the energy read-out from the Monte Carlo where the details of how much energy loss shows up as visible light in the BCAL scintillation fibers. He noted how these details are different depending on particle species: photons are different from pions, and pions perhaps different from protons.
 +
* Sean emphasized that the main problem with getting efficiency calculations to agree with each other is tails on the timing distributions for charged hadrons in the BCAL, rather than the overall energy response, although the two issues could easily have a common cause.
 +
* Alex reminded us that in his study of pion timing from rho events, he tried various time windows, going all the way to ±5$nbsp;ns, and even there he saw differences in efficiency between the different hadronic interaction models.
 +
* Colin had two suggestions for further studies.
 +
*# Rather than looking as the relatively complicated topology of &gamma;p&rarr;&eta;&pi;<sup>&minus;</sup>&Delta;<sup>++</sup>, we drop back to a single particle gun study inspired by differences seen in the full reaction.
 +
*# On and event-by-event basis, study where the cuts differ when using different hadronic interaction models for the same event.
 +
* Richard also had suggestions on what to try next:
 +
*# Look at the timing distributions for charged hadrons in the BCAL on the NHL branch. If the high side tail that we have been discussing is due to late neutrons and/or decaying nuclear fragments as he suspects, then it should absent when using this branch.
 +
*# Look at the effect of the NHL branch on photon energy and if it is present, measure the energy deficit and use that measurement to apply a correction to the reconstructed photon energy. Once that is done, look at the effect on overall event efficiency, again using the NHL branch, on Colin's reaction.
 +
* Sean reminded us of [https://halldweb.jlab.org/doc-private/DocDB/ShowDocument?docid=4053 a talk that Mark Dalton gave at a GlueX workshop] in May 2019. Mark showed a variety of effects in data and Monte Carlo. Sean showed the slide on BCAL timing resolution as a function of energy and BCAL shower time offsets as a function of both energy and z-coordinate. There are many features that we still do not understand in detail. Sean does not want us to lose track of these issues.
 +
* Zisis reminded us that Andrei Semenov developed a detailed Geant4 simulation of the BCAL with the geometry of the fibers and lead sheets expressed explicitly. This may come in handy as we focus in on details.
 +
* The group thanked Zisis for getting the group fired up to re-address these issues and also welcomed Tegan to the effort.

Revision as of 18:44, 4 December 2020

HDGeant4 Meeting
Tuesday, December 1, 2020
3:00 pm EST
BlueJeans: 968 592 007

Agenda

  1. Announcements
  2. Review of minutes from the last meeting (all)
  3. Issues on GitHub
  4. Pull Requests on GitHub
  5. Action Item Review

Minutes

Present: Alex Austregesilo, Tegan Beattie, Sean Dobbs, Colin Gleason, Mark Ito (chair), Igal Jaegle, Naomi Jarvis, Richard Jones, Zisis Papandreou, Simon Taylor, Beni Zihlmann

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

Announcements

Mark announced that he has a complete build of CentOS 8 available at JLab. It uses the default version set: version_4.32.0.xml.

  • The directories are in the standard place on the group disk:
/group/halld/Software/builds/Linux_CentOS8-x86_64-gcc8.3.1-cntr
  • The build is now getting shipped to Oasis and is available via CVMFS.
  • The corresponding singularity container is in the "dist" directory:
/group/halld/www/halldweb/html/dist/gluex_centos-8.2.2004_sng3.6_gin2.15.sif

and is available for download from

 https://halldweb.jlab.org/dist/gluex_centos-8.2.2004_sng3.6_gin2.15.sif

After the announcement we had a discussion on how to transition to GCC compilers beyond 4.8.5, the default for CentOS 7. Moving to more advanced versions of ROOT and Geant4 depend on making this transition.

  • Mark has been thinking that by going to CentOS 8, either via container or actual of the new OS, we get to GCC 8.3.1 a big leap forward with an unambiguous prescription for going forward.
  • Richard and Sean advocated keeping at CentOS 7 since, after all, it represents the bulk of our installed base, and using DevToolSets from Software Collections to get an upgraded GCC.
  • Richard mentioned the use case where a CentOS 7 container is used with CVMFS to get the software. Here the OS and built software are provided; users in this category don't care how the sausage is made.
  • Sean mentioned the use case where only ROOT files are analyzed using gluex_root_analysis. All of the build issues are moot in this case as well.
  • Beni pointed out that by going to a new version of the OS, we get upgrades on many packages all at once, not just GCC, and eventually all of these upgrades will have to be faced. If we follow the new OS approach, confronting these problems will give us flexibility and save pain down the road.
  • Sean advocated an approach where we specify a minimum compiler version as a starting point and work on guidance for every to get there.
  • Mark will continue to think about a recommendation for best practices. He found the discussion very useful in this regard.

Review of minutes from the last meeting

We went over the minutes from September 22. Sean mentioned that the details on how to widen the Δt cuts to make efficiency calculations more reliable was taken up at the PID meeting this morning.

Issues on GitHub

We spent most of the remaining time discussing Issue #111, Difference in Acceptance between G3 and G4 Colin lead us through the issue. There was a lot of leading to do since he first submitted the issue in May of 2019 and it has remained active, with many updates over time. [The secretary refers the reader to the link above rather than attempt a comprehensive summary.]

There followed an active discussion of ideas related to the issue.

  • With respect to the low acceptance in G4 when using the "no heavy light" (NHL)branch, Richard suspects that simply turning off energy deposition for particles heavier than a proton might be too naive an approach. That and the fact that if the energy reported out by the Monte Carlo is not the energy deposited by particles in sensitive volumes, then there needs to be a calibration step as is done with real data. He outlined ideas for adding detail to the energy read-out from the Monte Carlo where the details of how much energy loss shows up as visible light in the BCAL scintillation fibers. He noted how these details are different depending on particle species: photons are different from pions, and pions perhaps different from protons.
  • Sean emphasized that the main problem with getting efficiency calculations to agree with each other is tails on the timing distributions for charged hadrons in the BCAL, rather than the overall energy response, although the two issues could easily have a common cause.
  • Alex reminded us that in his study of pion timing from rho events, he tried various time windows, going all the way to &pm;5$nbsp;ns, and even there he saw differences in efficiency between the different hadronic interaction models.
  • Colin had two suggestions for further studies.
    1. Rather than looking as the relatively complicated topology of γp→ηπΔ++, we drop back to a single particle gun study inspired by differences seen in the full reaction.
    2. On and event-by-event basis, study where the cuts differ when using different hadronic interaction models for the same event.
  • Richard also had suggestions on what to try next:
    1. Look at the timing distributions for charged hadrons in the BCAL on the NHL branch. If the high side tail that we have been discussing is due to late neutrons and/or decaying nuclear fragments as he suspects, then it should absent when using this branch.
    2. Look at the effect of the NHL branch on photon energy and if it is present, measure the energy deficit and use that measurement to apply a correction to the reconstructed photon energy. Once that is done, look at the effect on overall event efficiency, again using the NHL branch, on Colin's reaction.
  • Sean reminded us of a talk that Mark Dalton gave at a GlueX workshop in May 2019. Mark showed a variety of effects in data and Monte Carlo. Sean showed the slide on BCAL timing resolution as a function of energy and BCAL shower time offsets as a function of both energy and z-coordinate. There are many features that we still do not understand in detail. Sean does not want us to lose track of these issues.
  • Zisis reminded us that Andrei Semenov developed a detailed Geant4 simulation of the BCAL with the geometry of the fibers and lead sheets expressed explicitly. This may come in handy as we focus in on details.
  • The group thanked Zisis for getting the group fired up to re-address these issues and also welcomed Tegan to the effort.