Speaker
Description
The High-Luminosity LHC (HL-LHC) will operate with pile-up levels of up to μ≃ 200, significantly increasing the trigger and reconstruction challenge for the ATLAS experiment due to high detector occupancies and background activity. To meet the stringent computing budget and physics-performance requirements, ATLAS is migrating tracking to the experiment-agnostic A Common Tracking Software (ACTS) framework and exploring machine-learning techniques to reduce the combinatorial load prior to pattern recognition. This contribution presents a prototype standalone muon reconstruction chain for HL-LHC conditions, implemented in the ATLAS reconstruction software stack and integrated into the Event Filter using ACTS-based components. The chain includes space-point formation, Hough-transform pattern recognition in η and ϕ, straight-line segment reconstruction with dedicated treatment of MDT left--right ambiguities, and segment-seeded track building with a global χ2 fit in the inhomogeneous magnetic field. In simulation and in the Run-3 trigger environment, the chain achieves high efficiency and excellent angular and momentum resolution, while delivering a substantial CPU performance improvement over the legacy pipeline. To further reduce background-driven compute cost, a Graph Neural Network filter operating on space-point graphs is introduced, rejecting over 96 % of background-only regions while retaining more than 99% of muon candidates.
| I read the instructions above | Yes |
|---|