1. Computing Performance Recommendations #13, #14

2. Recommendation #13 (1/3) We recommend providing a simple mechanism for users to turn off “irrelevant” processes for a given region. Several capabilities already exist which can address significant parts of the needs: There is a mechanism to stop EM processes from generating optical photons for “irrelevant” regions. (Relevant processes: transition radiation, Cerenkov and scintillation.)  For other EM processes, "cuts per region" addresses fully this purpose, by not creating EM particles below a threshold range (and thus material- dependent energy) in certain regions. This is the most important capability, as in most applications electrons and gammas consume the majority of CPU time (typically over 80%).  For all particle types, the user can kill tracks when they reach the user- defined limits, e.g. flight time, total track length, etc.  e.g. neutrons, the second most important particle type by average number of steps. continuing…

3. Recommendation #13 (2/3) In general, the user can activate/inactivate individual physics processes for the next and following steps.  In User’s SteppingAction, the user has access to the process vector through the particle definition of the current track.  Information regarding next volume (including material, region, etc.) is fully available.  We opt not to document this possibility explicitly.  The user has to be responsible of the correctness of the results.  For example, if a process is inactivated for a step, the track survived that step should have a smaller track weight. Forcing or prohibiting the occurrence of an interaction, e.g. decay or hadronic interaction, in a certain volume/region is identified as quite useful for boosting simulation. This is a kind of event biasing (variance reduction) option and thus to be discussed with the next recommendation. continuing…

4. Recommendation #13 (3/3) For the related goal of reducing the computing cost in particular regions: Switching the accuracy parameters and methods for trajectory integration (in an external EM field) for each region is being developed in collaboration with Atlas, as a user code. Its effectiveness is being studied. It could be included to Geant4 release (as an example) for use by other large-scale HEP experiments. For hadronic processes, we interpret this requirement to give the user an easy mechanism to switch hadronic models on the fly. A mechanism discussed to address this is the "physics lists per region" or "process managers per region".  This would be a new feature requiring significant changes in the Geant4 kernel (assessed to be of the same order as the major changes for the "cuts per region" development).  We have opted to postpone this development, until we finish other urgent things, including issues for LHC start-up. A decision on making a prototype implementation during 2010 is not taken yet.  It potentially couples with the planned internal overall Geant4 design assessment,

5. Recommendation #14 (1/2) We recommend providing guidance on the use of VRT (variance reduction techniques) in Geant4 so users can better take advantage of them. References to publications and data regarding the validation of the VRT methods in Geant4 should be made available in the performance optimization guide. There is much room to improve in this area. Recent loss of manpower at both SLAC and CERN halted the improvement activities (both in documenting and in enriching options). Currently there is active development of an implementation of Reverse Monte Carlo for electrons and gammas, funded by ESA. This includes draft documentation. Assisted in the implementation of Woodcock tracking by GATE developers. We are discussing making this development available inside the Geant4 toolkit. continuing…

6. Recommendation #14 (2/2) We are seeking for resources to resume other improvement activities in this area. Plans (pending the finding of funding, recruiting of person/s):  With 0.75 FTE-year document the key existing VRT capabilities in Geant4 with realistic examples and results which validate and demonstrate the efficiency improvement.  After another 1.5 FTE-year, we can unify the usage of several options. (Note: currently the usages of VRTs in Geant4 are quite different for each options due to their legacies, for example, the usage of leading particle biasing is not common for EM and hadronic particles).  Further resources are demanded for enriching options.