Choose timezone
Your profile timezone:
67th Geant4 Technical Forum
Description
We are pleased to announce the 67th edition of the "Geant4 Technical Forum", the forum for Geant4 users and developers, that will be held on Thursday, April 2nd, 2026, from 15:00 to 18:00 Central European Time (CET).
It will be held in hybrid mode, in person at CERN and over Zoom
The main focus of this edition of the Technical Forum for the developer part is the discussion of this year Geant4 work-plan, that will shape the content of the next Geant4 release. In addition, there will be a section dedicated to the users with the opportunity to give inputs and express requests and requirements to the Geant4 collaboration.
We encourage as usual, all our users to contribute to the session with their inputs that may concern basic functionalities, technical issues, performances, user support or other Geant4-user related aspects. Please use this opportunity to discuss priority or time sensitivity regarding existing open issues and to raise new issues or new aspects of existing ones.
Please send your proposals, issues for discussion, feedbacks and new requirements to the organisers by Tuesday, March 26th, so that the agenda can be finalised accordingly. Thank you.
Geant4 Technical Forum
Zoom Meeting ID
96372102095
Host
Alberto Ribon
Alternative hosts
Marc Verderi, Marilena Bandieramonte
Passcode
54459386
Useful links
Join via phone
Zoom URL
Introduction (Alberto)
- No questions
Geant4 2026 work-plan for the non-physics part & patch release 11.4.1 (Gabriele):
- Alexei: Concerning ML, G4 uses external framers for ML, do you think that there will be at some point a sense of using some ML framework (PyTorch, others) directly inside Geant4?
- Witek: Some of these frameworks are already used within our applications - also ROOT provides some infrastructure for ML. We use them as external libraries, but we will not integrate them in G4
- Gabriele: in G4 for fast simulation you use the hooks and then the rest of the machinery is done by the framework you choose. So it is up to the user to configure properly their application.
- Alexei will follow up offline to solve issue in cmake to be able to plug and play different ML dependencies. That would be helpful for the users.
- Francesco: in the long-term plan in general we could think about possible implementation of ML tools. In the short term we should state clearly in the README of examples using ML what are the steps that the user should make to compile properly
- Mihaly: Are we talking about G4 dependency from external tools or user applications dependency?
- If it is the user application then, the application (so the user) should take care of the dependencies.
Geant4 2026 work-plan for the EM physics part & patch release 11.4.1 (Vladimir):
- Alberto: Slide 8: This goes back to what discussed before. This implementation of a ML model, will be introduced in the source code, not as user application. If it is at source level, this might be a problem, and we need to discuss how to handle it.
- Witek: it could be introduced as an optional dependency as GDML
- Marilena: what kind of dependency will need to be introduced to support this model?
- Some ML framework like PyTorch. This is long-term plan, so we still didn’t decide on the technology. It is possible to convert formats, training with Keras, converted to ONNX, then file, then uploaded to Geant4 with ONNX.
- Alexei: concerning channeling data in slide 8, it is a matter of performing test of synthetic data with experimental data. So it will grow with time.
Geant4 2026 work-plan for the hadronic physics part & patch release 11.4.1 (Pico):
- Alexei: Will it be possible to generate lambda hyperons in G4?
- Alberto: Yes, it is already possible, the new ones are the charm ones
Work plan for the R&D activities (Witek):
- Alexei: on slide 3, nice to see the speedup. What about the energy consumption?
- Witek: I don’t have the plots here. We are at the edge of winning something if we compare running only on CPU and keeping the GPU idle and running with both CPU and GPU. If there’s no GPU at the moment there’s no argument to buy. If there’s a GPU and it is off, there’s no argument for now to switch it on.
- Alexei: concerning slide 7, why transformers? Do they provide gains?
- Alberto: My understanding is that transformers have a better adaptability.
- Marilena: What we are currently seeing in ATLAS is that this model is able to handle high number of input features and this is potentially an advantage over other models when we need to model showers on complex calorimeters.
- Marilena: question on slide 6. VecGeom will adopt these standalone libraries from Celeritas. Does it mean that also AdePT will use them and what is the advantage?
- Severin: this will help with portability. In addition, the way VecGeom is initializing is very slow, and this will go away. Other performance in navigation will not be affected by this change. This is a work in progress from both teams.
Open and New Requirements update (Krzysztof )
- No questions
Foreseen model for hadrontherapy purpose based on measured data (Christian):
- Alberto: this is very interesting. One suggestion: if you have experimental data you will use them, otherwise you plan to use AI. In the second case, if the user cannot install or doesn’t want to install AI infrastructure, a possibility would be to use either ENDL or binary cascade. We are looking forward to have this in Geant4. Next collaboration meeting will be in Bordeaux, you could participate.
- Christian agrees. Introducing the database by then would be nice
Precision inconsistency and transition sampling artifacts in G4LevelReader (Mohamed):
- Vladimir: my impression is that the migration to double could increase the memory but not critically. I suggest to open a bug report and attach the implementation to that with the migration from float to double. This should be enough.
- Mohamed: most of the variables are double now. I didn’t realize why these 5 variables are float
- Vladimir: all database are uploaded for LHC use case, and we made this decision with making them float, but we can definitely make them double
- Mohamed: most of the variables are double now. I didn’t realize why these 5 variables are float
- Alberto: suggestion for the future: in G4 11.2 we have introduced NUDEX that currently is used for neutron capture. The idea is to extend it to any kind of nuclear de-exitation and it is supposed to provide high precision. To use it today you would need to hack the code, but this is a possibility. In future we will generalize its usage and you might be able to use it directly
- Mohamed: Yes, I am aware of it, will try to use it in the future when available
- Pico: very clear and nicely presented talk. Please upload slides on the agenda.
There are minutes attached to this event.
Show them.