Geometric deep learning for high-energy particle tracking – DeepTrack

The high-luminosity phase of the LHC (HL-LHC) starting in 2026, with the first data taking planned in Run 4 (2029-2032), will play a crucial role to provide some indications on the shortcomings of the Standard Model of particle physics. A precise, robust and fast particle track reconstruction is essential to any kind of study performed at the LHC. However the track reconstruction from the hits measured by the ATLAS Inner Tracker detector (ITk) will be a real challenge due to the very high combinatorics to be faced in the HL-LHC conditions. It is of tremendous importance to develop more efficient tracking algorithms to fully leverage the data that will be recorded at HL-LHC and ensure impactful physics results.
The aim of this proposal is to study and demonstrate that a Graph Neural Network based charged particle track reconstruction brings significant improvement in terms of physics performances, computing time and robustness compared to the current Combinatorial Kalman Filter approach, improving the physics reach of the ATLAS experiment at HL-LHC.