Speaker
Description
Monte Carlo (MC) simulation is essential for ATLAS physics analyses, connecting theoretical predictions to detector-level observables. While Geant4 provides highly accurate detector modelling, it is computationally expensive. To enable large MC samples, especially for the High-Luminosity LHC, ATLAS has developed a broad fast simulation program that replaces selected steps with faster approximations.
A key tool is AtlFast3, which combines classical parametrisations with generative machine learning models for fast simulation of electromagnetic and hadronic calorimeter showers. Recent work includes an improved voxelisation scheme for training and exploration of advanced generative approaches such as diffusion models, transformers, and continuous normalizing flows. A new transformer-based generative model using flow-matching techniques was also developed to better model muon punch-through and capture correlations in secondary particle production.
Fast simulation is also being extended to the inner detector. Fast Track Simulation (FATRAS) uses simplified geometries and parametrized electromagnetic interactions, while retaining Geant4 hadronic models, with ongoing R&D focused on improved accuracy and integration with ACTS. Track overlay accelerates pile-up processing by reusing reconstructed pile-up tracks, with updates ensuring consistent tracking and ML-driven runtime steering. These components are integrated in the FastChain framework, unifying simulation and reconstruction into an efficient end-to-end workflow.
| I read the instructions above | Yes |
|---|