HDGeant4 Meeting, March 26, 2019
Tuesday, March 26, 2019
JLab: CEBAF Center, A110
- Review of minutes from the last meeting
- Do-over on comparison studies
- Further comparative studies of calorimeter response
- Issues on GitHub (all)
- Pull Requests on GitHub (all)
- JLab: Alex Austregesilo, Thomas Britton, Sean Dobbs, Mark Ito (chair),Richard Jones, Simon Taylor
There is a recording of this meeting on the BlueJeans site. Use your JLab credentials to access it.
Review of minutes from the last meeting
We went over the minutes without significant comment.
Do-over on comparison studies
Simon has run some single-particle gun simulations for both HDG3 and HDG4 on the OSG. He has not had a chance to look at them in detail yet. His initial impression is that differences are seen in calorimetry.
SciComp System Perfomance
We got onto a digression on recent experience using the farm and disk resources at JLab.
- Richard reports short file life-times on the cache disk after retrieval from tape, not associated with farm jobs, as short as 20 hours. We speculated that this was due to farm job usage of the cache displacing files that would normally live for a week.
- Random trigger files that are supposed to be "permanently" pinned for year have disappeared. This causes problems for Monte Carlo jobs on the OSG which need these files.
- We commented that although we have a lot of space on the work disk, there are usage storms on work that not only make file retrieval extremely slow, but bring the ifarm nodes to a standstill. When work was on Lustre this was not an issue, but handling small files was a problem.
Further comparative studies of calorimeter response
Richard showed some slides describing extensions of his work on low-level comparisons of HDG3 and HDG4 for calorimetry.
- He looked at the residual non-linearity, albeit up to 8 GeV, for reconstructed shower energy in the FCAL. Agreement between HDG3 and HDG4 is very good now, at all energies.
- He has examined cluster time, relative to "RF". Both HDG3 and 4 show the same linear systematic dependence in relative time as a function of energy.
- For BCAL energy response, he showed the same residual non-linearity study as shown earlier for the FCAL. Here the qualitative agreement is good, with perhaps an 8 MeV upward shift of HDG4 relative to HDG3. The effect of energy leakage out the back of BCAL dominates the changes in the trends as polar angle moves from far forward to 90 degrees.
- For BCAL timing, he noted a difference in the algorithm used to get cluster timing between HDG3 and HDG4
- HDG3 uses energy-weighted average shower times
- HDG4 uses the earliest hit per sector for times
- We noted that mcsmear smears the BCAL timing by 55 ps, not a lot. Richard sees this in his study.
- The difference in the BCAL algorithms does not make much apparent difference. He suggests we use energy weighted for both to be consistent.
- Long ago someone decided to deprecate the bcalSiPMUpHit and bcalSiPMDownHit elements in HDDM in favor of a single bcalTruthHit element that can serve information from both upsteam and downstream ends of the BCAL. In retrospect, this was not a good choice. Richard proposes to reverse that decision and rehabilitate the original bcalSiPMUpHit and bcalSiPMDownHit elements in HDDM. There is no reason at the same time to remove the bcalTruthHit, which can remain alongside the UpHit and DownHit tags, to give additional truth info in case that is needed for acceptance or bcal response studies. But the mcsmear code should act primarily on the UpHit and DownHit info.
- There is still an issue that Alex sees with the ρ's albeit with charged tracks and not electromagnetic showers. Hopefully Simon's single-track studies will address it.