Speaker
Description
In the ATLAS experiment, colliding proton-proton bunches produce multiple primary vertices per bunch crossing. Typically, the primary vertex with the highest sum of squared transverse momentum of associated tracks is designated as the hard-scatter (HS) vertex, serving as the reference point for all physics objects in the event. However, this method proves suboptimal for scenarios with low track activity in the primary vertex, such as Higgs boson decay into photon pairs or low transverse momentum leptonic signatures. To address this, a novel approach employing graph neural networks has been developed by ATLAS. This new algorithm aims to enhance the efficiency of identifying the correct HS vertex in events by leveraging information from the complete set of reconstructed physics objects in the event. In this presentation, we will discuss the implementation of this algorithm, highlight its performance across various physics processes, and compare its effectiveness to existing algorithms used in ATLAS for HS vertex determination.
Significance
The contribution presents an entirely new algorithm, and can impact the physics reach of the ATLAS collaboration in a large fraction of physics cases.
| Experiment context, if any | ATLAS |
|---|
