Difference between revisions of "HDGeant4 Meeting, November 5, 2019"

From GlueXWiki
Jump to: navigation, search
m
(Added the minutes.)
Line 11: Line 11:
 
# [https://github.com/JeffersonLab/HDGeant4/pulls Pull Requests on GitHub]
 
# [https://github.com/JeffersonLab/HDGeant4/pulls Pull Requests on GitHub]
 
# Action Item Review
 
# Action Item Review
 +
 +
== Minutes ==
 +
 +
Present:
 +
* ''' CMU: ''' Naomi Jarvis
 +
* ''' FSU: ''' Sean Dobbs
 +
* ''' JLab: ''' Alex Austregesilo, Mark Ito (chair), Igal Jaegle, Simon Taylor, Beni Zihlmann
 +
* ''' UConn: ''' Richard Jones
 +
 +
There is a [https://bluejeans.com/s/EGhEW/ recording of this meeting] on the BlueJeans site. Use your JLab credentials to get access.
 +
 +
=== Announcements ===
 +
 +
* Naomi alerted us to a recent Analysis Note from Nilanga Wickramaarachchi where he sees a discrepancy between HDG3 and HDG4. Alex commented that it amounts to a factor of two difference in acceptance for the t-channel and a strange dip in acceptance in the u-channel. He will mention this at tomorrow's Production and Analysis Meeting. He also noted that Peter Pauli has seen similar acceptance differences.
 +
* Beni announced that he has succeeded building the entire GlueX Software stack on Ubuntu 19.04 (Disco). This distribution ships with GCC 8.3. He noted a huge spike in memory use when compiling DRootGeom.cc from the HDGEOMETRY library. % GB of swap space was needed. Richard thought that this might have something to do with compiler optimization on a large, script-generated source code file.
 +
 +
=== Review of minutes from the last meeting ===
 +
 +
We went over [[HDGeant4 Meeting, October 22, 2019#Minutes|the minutes from October 22]] without significant comment.
 +
 +
=== Visualization with OpenGL ===
 +
 +
Igal has been having trouble running hdgeant4 using OpenGL to do event visualization. He gets an error that points to OpenGL libraries and the program crashes. He has not seen this problem with Geant4 version 9, only version 10.
 +
 +
Richard told us he uses OpenGL often and is committed to maintaining this piece of functionality. He will consult with Igal offline.
 +
 +
Richard also mentioned that he uses a package called [https://virtualgl.org/About/Introduction VirtualGL] to access local graphics acceleration hardware while using a VNC to run Geant4 on a remote node.
 +
 +
=== Python Interface Problem ===
 +
 +
Igal reports an issue using the Python interface to hdgeant4. There appears to be a conflict between different versions of the boost libraries. Richard cautioned us that hdgeant4 will not work through the Python interface if multi-threaded operation is turned on in the build. You need a single-threaded build for it to work due to issues with the SciPy library. Igal will try that.
 +
 +
=== Neutron Simulation ===
 +
 +
We reviewed [https://github.com/JeffersonLab/HDGeant4/issues/125 Issue #125]. In light of this issue, Igal suggested that we use the "high precision" option in the physics list if we want to do neutron propagation with Geant4. We need to look into this.
 +
 +
=== Segfault from G4VoxelNavigation::ComputeStep ===
 +
 +
We reviewed [https://github.com/JeffersonLab/HDGeant4/issues/118 Issue #118]. Naomi will return to this when she succeeds in getting a free moment or two.
 +
 +
=== Overall BCAL energy shift? ===
 +
 +
We reviewed [https://github.com/JeffersonLab/HDGeant4/issues/113 Issue #113]. There is an energy shift in the BCAL response when comparing G3 to G4. Richard will be taking another look at this one.
 +
 +
Igal suggested looking at the difference without smearing.
 +
 +
Igal also asked about whether the same low-energy cut-offs. Richard reminded us that the scheme used in G4 is to stop tracking when particle range gets small rather than when particle energy gets small, as is done in G3. This makes matching of thresholds between the two systems hard if not impossible.
 +
 +
=== Difference in Acceptance between G3 and G4 ===
 +
 +
We reviewed [https://github.com/JeffersonLab/HDGeant4/issues/111 Issue #111]. Colin Gleason reported this issue in his analysis. We will go back to him and ask if this is still a problem for him. If so, it may need to be repeated with modern versions of software.
 +
 +
=== Feature request - progress indicator ===
 +
 +
This is [https://github.com/JeffersonLab/HDGeant4/issues/107 Issue #107]. Richard is on it.
 +
 +
=== Reading in # events and source ===
 +
 +
Richard estimated that [https://github.com/JeffersonLab/HDGeant4/issues/105 Issue #105] would take about an hour to address.
 +
 +
=== Terminal cleanup? ===
 +
 +
[https://github.com/JeffersonLab/HDGeant4/issues/102 Issue #102] is on Richard's list.
 +
 +
=== Calorimeter timing mismatch between g3 and g4 ===
 +
 +
We then discussed [https://github.com/JeffersonLab/HDGeant4/issues/93 Issue #93], having to do with charged particle timing measurements in out two calorimeters. Alex wondered if the problem seen here is the same as those reported by Colin and Nilanga, i.e, large differences in overall acceptance for a particular channel as a whole between G3 and G4.
 +
 +
Mark suggested that calculating efficiency from Monte Carlo could be very difficult if accuracy depends on getting the tails of distributions correct. We might be better off studying efficiency in data alone. Richard thought that that approach might be subject to large systematic effects.
 +
 +
Igal suggested that in this case as well, comparisons of the unsmeared reconstruction might be informative.
 +
 +
Alex will try to look at single particle responses in data and see how they compare to the two Monte Carlos.

Revision as of 20:15, 13 November 2019

HDGeant4 Meeting
Tuesday, November 5, 2019
2:00 pm EST
JLab: CEBAF Center, A110
BlueJeans: 968 592 007

Agenda

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

Minutes

Present:

  • CMU: Naomi Jarvis
  • FSU: Sean Dobbs
  • JLab: Alex Austregesilo, Mark Ito (chair), Igal Jaegle, Simon Taylor, Beni Zihlmann
  • UConn: Richard Jones

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

Announcements

  • Naomi alerted us to a recent Analysis Note from Nilanga Wickramaarachchi where he sees a discrepancy between HDG3 and HDG4. Alex commented that it amounts to a factor of two difference in acceptance for the t-channel and a strange dip in acceptance in the u-channel. He will mention this at tomorrow's Production and Analysis Meeting. He also noted that Peter Pauli has seen similar acceptance differences.
  • Beni announced that he has succeeded building the entire GlueX Software stack on Ubuntu 19.04 (Disco). This distribution ships with GCC 8.3. He noted a huge spike in memory use when compiling DRootGeom.cc from the HDGEOMETRY library. % GB of swap space was needed. Richard thought that this might have something to do with compiler optimization on a large, script-generated source code file.

Review of minutes from the last meeting

We went over the minutes from October 22 without significant comment.

Visualization with OpenGL

Igal has been having trouble running hdgeant4 using OpenGL to do event visualization. He gets an error that points to OpenGL libraries and the program crashes. He has not seen this problem with Geant4 version 9, only version 10.

Richard told us he uses OpenGL often and is committed to maintaining this piece of functionality. He will consult with Igal offline.

Richard also mentioned that he uses a package called VirtualGL to access local graphics acceleration hardware while using a VNC to run Geant4 on a remote node.

Python Interface Problem

Igal reports an issue using the Python interface to hdgeant4. There appears to be a conflict between different versions of the boost libraries. Richard cautioned us that hdgeant4 will not work through the Python interface if multi-threaded operation is turned on in the build. You need a single-threaded build for it to work due to issues with the SciPy library. Igal will try that.

Neutron Simulation

We reviewed Issue #125. In light of this issue, Igal suggested that we use the "high precision" option in the physics list if we want to do neutron propagation with Geant4. We need to look into this.

Segfault from G4VoxelNavigation::ComputeStep

We reviewed Issue #118. Naomi will return to this when she succeeds in getting a free moment or two.

Overall BCAL energy shift?

We reviewed Issue #113. There is an energy shift in the BCAL response when comparing G3 to G4. Richard will be taking another look at this one.

Igal suggested looking at the difference without smearing.

Igal also asked about whether the same low-energy cut-offs. Richard reminded us that the scheme used in G4 is to stop tracking when particle range gets small rather than when particle energy gets small, as is done in G3. This makes matching of thresholds between the two systems hard if not impossible.

Difference in Acceptance between G3 and G4

We reviewed Issue #111. Colin Gleason reported this issue in his analysis. We will go back to him and ask if this is still a problem for him. If so, it may need to be repeated with modern versions of software.

Feature request - progress indicator

This is Issue #107. Richard is on it.

Reading in # events and source

Richard estimated that Issue #105 would take about an hour to address.

Terminal cleanup?

Issue #102 is on Richard's list.

Calorimeter timing mismatch between g3 and g4

We then discussed Issue #93, having to do with charged particle timing measurements in out two calorimeters. Alex wondered if the problem seen here is the same as those reported by Colin and Nilanga, i.e, large differences in overall acceptance for a particular channel as a whole between G3 and G4.

Mark suggested that calculating efficiency from Monte Carlo could be very difficult if accuracy depends on getting the tails of distributions correct. We might be better off studying efficiency in data alone. Richard thought that that approach might be subject to large systematic effects.

Igal suggested that in this case as well, comparisons of the unsmeared reconstruction might be informative.

Alex will try to look at single particle responses in data and see how they compare to the two Monte Carlos.