Speakers
Description
This contribution presents the expected tracking performance of the ATLAS ITk based on the latest detector layouts and software developments. Excellent efficiency, resolution, and control of mis-reconstructed tracks are demonstrated in high-occupancy environments, validating the ITk design for HL-LHC conditions.
A central element of the Run-4 tracking strategy is the migration to ACTS, an experiment-independent tracking toolkit. The ongoing integration involves a redesign of reconstruction algorithms and the ATLAS Event Data Model, enabling a scalable, thread-safe, and maintainable software architecture. The current status of ACTS-based ITk tracking and its performance and computational characteristics are discussed.
In addition, machine-learning approaches based on Graph Neural Networks (GNNs) are explored as a complementary solution for track reconstruction at extreme pile-up. Recent advances in the GNN4ITk pipeline are presented, including improvements in physics performance and significant gains in computational efficiency through model optimization, GPU acceleration, and integration into the ATLAS software framework.
| I read the instructions above | Yes |
|---|