Minutes of 29 Sep 2009 meeting of Geant4 Technical Forum
Editor: G.Cosmo, 29 Sep 2009
- Starting at 16:50.
Update on Open Requirements
+ #1503, low-energy physics-lists: lists available in release 9.3-beta. Migration
of physics models achieved, including Livermore and Penelope models.
Transition energy chosen to be 1 GeV.
+ #1202, RPG model under development, missing resources; priorities to be discussed
with people who are interested.
+ #2501, choices of physics lists by region: specific use-case for EM processes
(msc) demonstrated with existing toolkit; CMS has been able to run
simulation utilizing this.
+ #2502, fluctuation of energy deposition in gas: fix included in 9.3-beta.
Proposed to be closed once included in final release.
+ #2503, steering field tracking by region: implementation expected by end-October.
+ #0402, documentation on physics-lists: prototype pages in place with partial
Geant4 Relase 9.2.p02 - and 9.3-beta
+ Not clear if all developments planned for 2009 will go in release 9.3; certainly
some will not make it.
Period of Support for Geant4 release 9.2
+ From last meeting, support should be provide until November 2010 for release 9.2.
The Steering Board agreed on this, although only restricting to important fixes.
Input from Experiments, Users
+ CMS: using 9.1.p01 for mass production, several hundred million events produced.
Needs to keep stable on that version; no significant problems found so far; will
keep an eye on the current development. Will test in parallel release 9.3;
testing startup performance improvement provided aside on top of 9.1.p01;
looking to 9.2.p02 in the context of the Calorimetry Task Force.
Concerned about the FPE reported in Bertini, G4CollisionOutput.
Needed visualization of Boolean solids; specialized physics-lists per region
(workaround provided, functionality anyhow useful to have); interest in using
Glauber-Gribov cross-section above 90-100 GeV; looking forward for the efforts
going on to resolve discontinuity in energy report in QGSP_BERT.
Abort ratio of jobs due to Geant4 is very low; 1 event out of hundred thousand;
not possible to quantify the impact of the FPE observed. Some evidence of cases
of non-reproducibility, which can become problematic for the debugging of rare
+ LHCb: millions events produced, abort rate of 0.5%. 9.2.p02 to be used in future.
+ ATLAS: using 9.2.p02, just started production; plan to continue to end 2010.
SLC4/gcc-3.4 32-bits, plan to move to SLC5/gcc-4.3. Focus on performance
optimization; happy for the latest fixes introduced in 9.2.p02.
Concerning stability, two open issues: rare crash in hadronics from a corrupt
nucleus; stuck tracks in field, observed now, but probably old story (?)
attempted solution by increasing the threshold for abandoning tracks; some
guidance on how to set geometric tolerances would be useful; a rate from 0.1%
to 1% of stuck tracks which are a concern on ATLAS GRID production.
Other issues concern memory allocated by transportation (>2GB per event).
New class for caching field in Geant4 introduced in ATLAS version, suggest to
include in official Geant4.
Few problems in the physics; constructing EM processes in a vector helped in
saving 3.5 ins per job in initialization time; also improvements in Bertini
memory had helped. A case of Valgrind run-time error detected.
Thanking for help an response from the Geant4 team; would be nice to have more
public patches to avoid divergences; back-porting fixes from 9.3 to 9.2 would be
helpful. 400 million events in 2009; launching several 100 millions events with
9.2 (around 2 million per day).
GC: would be useful to know the exact place where the Valgrind errors occurs.
+ ALICE: activity on Geant4 is ramping up to get Gent4 production ready; the
experience gained so far is rather positive; comparing with Fluka and Geant3;
some issues on TPC may come up, as observed in the past.
An abstract will be submitted at the next Geant4 Users Workshop in Catania.
Investigation on Energy Response
+ See presentation by Alberto.
+ Observations: LEP seems to provide more energy into pions versus the other models;
in general worst than the other models; Bertini and Binary cascade appear to
produce too much energy in protons and neutrons above around 5 GeV; Fritiof, too
many pi-0 below 6 GeV; QGS could be extended down to 8-9 GeV; energy non-
conservation to be fixed in FTF/BIC.
New experimental physics lists to try as better model mixing: FTFP_BERT_TRV and
QGSP_FTFP_BERT; available in 9.3-beta.
Work to do: more validations with thin-target data in 1-20 GeV; continue to
improve/extend models; try to link model-level features to calorimeter
observables. This requires a major effort of the whole Geant4 hadronic group.
+ Next meeting towards the end of November.
Closing meeting at 18:25.
Tuesday, September 29, 2009