SciPost Submission Page
Fast, accurate, and precise detector simulation with vision transformers
by Luigi Favaro, Andrea Giammanco, Claudius Krause
Submission summary
| Authors (as registered SciPost users): | Luigi Favaro · Claudius Krause |
| Submission information | |
|---|---|
| Preprint Link: | https://arxiv.org/abs/2509.25169v2 (pdf) |
| Code repository: | https://github.com/luigifvr/vit4hep |
| Data repository: | https://calochallenge.github.io/homepage/ |
| Date submitted: | Nov. 20, 2025, 9:33 a.m. |
| Submitted by: | Luigi Favaro |
| Submitted to: | SciPost Physics Proceedings |
| Proceedings issue: | The 2nd European AI for Fundamental Physics Conference (EuCAIFCon2025) |
| Ontological classification | |
|---|---|
| Academic field: | Physics |
| Specialties: |
|
| Approach: | Computational |
Abstract
The speed and fidelity of detector simulations in particle physics pose compelling questions about LHC analysis and future colliders. The sparse high-dimensional data, combined with the required precision, provide a challenging task for modern generative networks. We present a comparison between solutions with different trade-offs, including accurate Conditional Flow Matching and faster coupling-based Normalising Flows. Vision Transformers allows us to emulate the energy deposition from detailed Geant4 simulations. We evaluate the networks using high-level observables, neural network classifiers, and sampling timings, showing minimum deviations from Geant4 while achieving faster generation. We use the CaloChallenge benchmark datasets for reproducibility and further development.
Author comments upon resubmission
We addressed the comment raised in the report in this revised version.
A CPU time estimate for the Geant4 generation is only available for datasets-2 and -3 of the CaloChallenge.
We include this number in Table 1, and we refer to it in the text. The Geant4 generation heavily depends on the incident energy since higher energies imply a larger number of particles to track. Given the incident energy distribution used to generate the two datasets, we report an average generation time per shower.
We provide only an order-of-magnitude estimate because the exact number will depend on multiple factors, such as hardware specifics and computational overheads.
Kind regards,
L. Favaro, A. Giammanco, C. Krause