Minutes of 2 Jul 2008 meeting of Geant4 Technical Forum Editor: G.Cosmo, 2 Jul 2008 Revisions: A.De Roeck, 18 Jul 2008 - Starting at 16:40. Geant4 9.2-Beta release ======================= + Feedback welcome but use with care (certainly not in production) Report from Performance Improvement Team ======================================== + Problems of connection at 16:55, echoes etc... John Apostolakis trying to repair... lost some attendees ... + Resuming at 17:05. + Slide 4: DB system for recording performance run data; usable only by experts for the moment. + Slide 5: Improvements in the Bertini Cascade brought to 3-4% CPU boost in typical CMS run. Considering inclusion in next Geant4 release. + Slide 6: some irreproducibility issues observed in the GRID were probably due to changes in the GRID environment; discrepancies seen went away after the changes applied to the GRID and the effect is not understood. + Slide 7: Sequence of randoms varies from AMD to Intel. Not clear why. + Slide 8: Targeting CHIPS for re-factoring and design improvement M.Paterno and J.Kowalkoski could do in 3 solid weeks ... Update on Open Issues ===================== + 0601: More details in error messages: ongoing improvements in diagnostics. To be closed + 0605: Adding touchable to secondaries. Implementation complete. To be closed + 1201: Improvements QGSP angular distributions. Under investigation but not applicable. To be closed. + 1202: Alternative models for intermediate energies. Ongoing, FTFP extended, new RPG model, ... + 0304: GDML. Complete. Close + 0106: Setup statistical test suites. Added calorimeter regression tests. + 0303: Performance of Geant4. G3 vs G4... Performance is a constant ongoing effort. To be closed. + 0309: Documentation on technical aspects of physics processes. Concrete issues addressed, propose guidelines, if agreed, close. + 0401: Extension of Ion hadronics. New native QMD model, interface to DPMJet under development. + 0402: Documentation of physics-lists. Ongoing. Registration of energy model ranges provided in 9.2-Beta, improve of coverage in progress. + 0506: Optical photon transport in repeated volumes. To be verified. + 0703: Still open. + 0704: No input. + 1301: Not resourced yet. Workaround identified. + 1302: Message processes in physics list. Partially done + 0502: Proposed to close. No longer big issue in CMS, big effort put in fixing geometry issues. + 1102: Identify a process efficiently. SubType of EM processes implemented. + 1401: Recoil ions. Requested feedback. + 1402: Scintillation. Updated G4Scintillation to use Birk's correction. + 1501: Categorize energy deposit: Fixes to NIEL released in 9.1.p01 + 1503: Physics lists for EM low-energy: option-3 in 9.2-Beta for more precision EM, already available PL for fluorescence. + Underground physics, see slides for status. + No progress on PL documentation on the web yet. + D.Wright reports that the neutron_hp model is going to be replaced with a new one based on Livermore model, to be included in 9.2. Experiment/User Input ===================== - ATLAS: Using SLC4/gcc-3.4.6 in 32 and 64 bits mode, CLHEP-1.9.3.1, Geant4 8.3.p02 plus two patches (BREPS and G4Tubs). Question if going on testing 9.2-Beta or not or sticking on 9.1. 1% aborted events for stuck tracks fixed with G4Tubs fix. Particles beginning out-of-world generated from generators, a check should be added in Geant4 to avoid this. Some exceptions in hadronic processes. Warnings from G4PropagationInField and crash in field. Performance: charged-geantinos take more than muons to propagate, eventually explained by higher accuracy applied to geantinos Using QGSP-BERT and using neutron cuts; implemented stepper-dispatcher to select the optimal stepper per particle-type. Looking forward for using enumeration of processes. D.Wright reports that the exception in the hadronics is now suppressed in 9.2-Beta. - CMS: Calorimeter task force created beginning 2008 to boost analysis of test-beams; useful interaction with Geant4 Collaboration; using Geant4 9.1.p02, choice for the massive production, chosen QGSP_BERT_EMV. Required fix is the high-Pt particles on high-energy hadronic model. Performance tests: 40% loss with BERT, 50-80% bigger event size, 5-10% more memory usage... most expected. Pending issues: Bremmstrhalung process for hadron; correction of quasi-elastic in Bertini; visualization of Boolean solids; hybrid stepper using field values and field gradients. MIP Fraction in ECAL to be improved (quasi-elastic in Bertini will help?); energy deposit in PbWO4 crystals at intermediate energies and in HCAL above 100 GeV to be improved/fixed. J.Apostolakis reports that reproducibility issues in intermediate event startups is still an open issue under investigation. It is certainly a long term goal, given the complexity. - LHCb: Nothing new to report. Now moving to 9.1.p02 to be used in production; to be checked if issues reported in February have been solved. - ALICE: No report, no representatives. Offline request for Arb8 shape. - Exotics in LHC: Quirks fermions... paired particles moving with anomalous curvature... hard to simulate. Markus Luty (UC-Davis) interested to contribute if Geant4 can deal with this. See slides for more details. J.Apostolakis: Geant4 can deal with this but not necessarily in its standard operation. See also next talk, that may help ... - Collective Computations: Beam computations, such as space charge. Merged all managers in a single manager. Implemented way to clone all internal state data of classes for every track (!). Realised a simpler way, using parallel tracking and this would not require any change to Geant4 code. Under development. M.Asai suggests using stacking on suspended tracks; overhead is the re-initialisation of the track when re-activated, but this can be overcome by copying the state and restoring. Validation of the statistical multi-fragmentation model of Geant4 ================================================================= + Simulation of nucleus-nucleus collisions. Some hadronic exceptions experienced while transporting Ne20/Ni58 in water. Tried using statistical multi-fragmentation model written in Fortran, problems of precision, float/double, raised. Fixes introduced to several classes to make data agree and fix crashes. Asking for having multifragmentation better controlled from user application (not possible to switch it on currently from G4ExcitationHandler, since present as private instance). J.Apostolakis comments that V.Lara did implement the model from scratch starting from theoretical model, it has not been a re-engineering. Update on spectra of interactions at 8-9 GeV ============================================ + Following report from HARP-CDP exercising Geant4 9.1 and 9.1.p02. 9.1.p02 partially improves the observed discrepancies. Absence of Fermi-motion in quasi-elastic, problem of scattering in Bertini Cascade and in LHEP are at the origin of the problem. 5 models can handle 8-9 GeV proton and pion beams, LEP, QGS, FTF, Binary and Bertini; QGS and FTF provide correct distributions now in 9.1.p02, where quasi-elastic part was fixed. + Would be useful to have HARP data published for improving simulation of quasi-elastic scattering. + Currently investigating the Bertini Cascade; LEP is more complicated since optimised for calorimeter simulation and not optimal for thin-target experiments, LHEP is kept unchanged as reference results for LHC calorimetry. FTF model is optimal, particularly FTF_BIC; the model is further tuned in 9.2-Beta. Physics lists for moderate energies are FTF_BIC and QBBC as available in 9.2. Beyond 9.2, LEP is planned to be replaced by the new RPG model; Binary and Bertini will be further tuned for the energy range 3-5 up to 10 GeV. Closing meeting at 18:50.