Minutes Technical Forum 11th March 2021

Minutes of 11th March 2021 Meeting of Geant4 Technical Forum

Editor: Marilena Bandieramonte, 11th March 2021

- Starting at 16:00 (3 hours)

Introduction
============

+ see slides for detail

  • Focus of this TF is the 2021 work plan towards this year major release, Geant4 version 11.0 (10 December 2021)
  • Beta release will be 25th June 2021
  • Highlights of this major release are:
    • New tasking model
    • Improvements and developments regarding EM and Hadronic Physics
    • VecGeom becoming the default library for implemented shapes
    • Adoption of C++17 as minimum standard
    • Improvement in the data library handling
  • Highlights for the R&D part:
    • From CPU to GPU: many ongoing projects
    • Fast simulation 
    • Performance tests on A64FX
       

Patch release 10.7.p01 and 2021 work plan - non-physics part
============================================================

+ see slides for details

Clarification: 

  • Analysis fix in pg 6, is related to MT mode
  • There will be many changes in the next Geant4 release. Proper guidelines for users will be published to help them with the adoption of this major release. Also the beta release of June will be a kind of preview that could be used to do the integration/migration exercise.
  • There are some preliminary benchmarks about the new Task based model, that show a better usage of threads. Large scale tests are still to be done. The way the sub-event parallelism is envisaged: very large events may be split in sub-events that can be assigned to different tasks and then executed in parallel. The major challenge here is to ensure the reproducibility (distribution of random number seeds between tasks), it will be addressed this year. It will be very beneficial for heavy ions simulations. 


Patch release 10.7-p01 and 2021 work plan - EM physics part
===========================================================

+ see slides for details

Major highlights from the work plan: 

  • Evaluation and extensions of the General Gamma Process approach (to be implemented also for Electron and Positron)
  • Evaluation of the G4HepEM project and integration in G4
  • Extensions to dark matter particle interactions
  • Integration of Opticks package on GPUs with new extended examples

Clarification: question about the Quantum Entanglement (slide 9). To do it right one should not only generate against probability distributions but calculating amplitudes, and in G4 this can be done cause the history of 2 interactions is kept and the amplitudes in take into consideration. It sounds like a challenge since the theoretical model is complex and not sure it would work. It seems from the linked publication that this is feasible. 

Geant4 R&D: G4HepEM CPU/GPU library
===========================================================

+ see slides for details

Major highlights:

  • G4HepEM is a R&D project initiated within the G4 Em physics working group
  • A very compact library specialised for EM showers generation and tailored for HEP applications
  • Given its design and implementation, it provides support to R&D projects targeting GPUs 
  • Run-time and initialization time functionalities are well separated
  • The data-layout is designed to be cache efficient
  • The same code can be reused as it is to perform (the physics related part of) the EM shower simulation on the GPU
  • G4HepEM reproduces Geant4 results with a precision within 1 per mille 
  • Already integrated in R&D GPU projects like Adept


Clarifications: 

  • G4HepEM can be already tried out and tested, even if some physics processes are still missing and the code is not optimised yet. To integrate it one need to assign this special process in the physics list for e-/e+ and gammas. The development is still ongoing so several optimizations that are in the native Geant4 EM simulation, are not there (i.e. usage of standard mathematical functions). The physics list to compare with is the HEP default one EM option 0.
  • There are preliminary performance comparisons (even if the library is not yet optimised) showing that G4HepEM is not slower in comparison to the standard Geant4 simulation
  • About the features that allows integration with GPU are already being exploited: the gamma part was already integrated. And in 2 weeks time some verification results will be presented. 


Patch release 10.7-p01 and 2021 work plan - Hadronic physics part
=================================================================

+ see slides for details

Major highlights:

  • Extend validation of charm production for FTF and QGS
  • Improvement of antiproton and light anti-ion annihilations in FTF
  • Improvements of de-excitation models: FermiBreakUp, Evaporation, GEM; validation and tuning to data
  • Improvement of Geant4 for nuclear-fusion applications. Production of Lithium nuclear data libraries, verification and validation.
  • Implement an option that forces ParticleHP to respect event-by-event conservations (energy-momentum, baryonic number, etc.) 
  • Extend ParticleHP model to higher energies 
  • Implement a very detailed physics for organic neutron detectors up to 100 - 200 MeV
  • New reference physics list using LEND. Update GIDI/LEND interface.
  • Add new physics (HighNESS project) + technical improvements
  • Development and validation of neutrino / lepton – nuclear physics
  • Extension of nuclear cross sections for light hyper-nuclei and anti-hyper-nuclei projectiles


Geant4 R&D: activities for 2021
=================================================================

+ see slides for details

Major highlights:

  • improvements to current Geant4 code base
  • fast simulation techniques (“classic” and ML detector-agnostic model)
  • exploration of new hardware (GPUs)
    • AdePT – Accelerated demonstrator of electromagnetic Particle Transport (basic prototype already working, with one particle and shower development done in the GPU)
    • Celeritas 
    • Optiks integration

Clarifications: For now the AdePT prototype works as a demonstrator prototype, but this is not making efficient use of GPUs capabilities. The occupancy comes from the shower development. In order to leverage GPUs some kind of grouping mechanism will have to be implemented, in order to process events concurrently. 

Open requirements
=================

+ see slides for details

There are no New Requirements from Experiments since last TF, this is an update of the current open issues

Efficient discrete-event based particle tracking simulation for high energy physics
=========================================================================

+ see slides for details

Major highlights:

  • Novel QSS integration methods developed for particle transport showing clear performance gains (largely application-dependent)
  • Entering a new benchmark phase (consolidating tests on CMS and starting with ATLAS as next reference model)
  • The group is entering Geant4 collaboration and has developed a plan for integrating the new QSS methods in native Geant4.


Clarifications:

  • The QSS system is fully implemented in C++, no other dependency
  • Geant4 developers are in close contact and the QSS team, and the method provides a promising alternative to other methods that have interpolation capabilities. It should be checked with the new version of the Dormand Prince that has this interpolation capability. This will be very interesting in the ATLAS case, because of the many boundary crossings. 
  • The initial synthetic geometry was realised with simplistic lattices of tubes 
  • Strategy 2 is faster than strategy 1, cause it removes the overhead. At the same time strategy one is needed cause some of the development done in the QSS solver as a standalone solver keeps evolving and many groups are working on developing new models that are not necessarily implementable in G4.