Machine Learning for Long-Lived Particle Jet-Tagging at Future Colliders

This study evaluates the performance of ParticleNet in jet-tagging tasks for searches targeting long-lived dark scalars produced in exotic Higgs boson decays. This process is simulated as part of the development of the Future Circular Collider in its e+e- stage. These dark scalars can exhibit lifetimes spanning several orders of magnitude, which makes jet-tagging particularly challenging. In this work, we employ a pre-trained machine learning model, ParticleNet, to perform jet-tagging and study the trend between jet-tagging performance and lifetime. Our results show a strong inverse correlation between the average b-tagging score predicted by ParticleNet and the lifetime of the dark scalars. For short lifetimes, the achieved tagging performance is comparable to that of current jet-taggers, with ParticleNet outperforming existing approaches for some shorter decay lengths but degrades as lifetimes increase. We conclude that the observed degradation in performance is likely due to the fact that ParticleNet was trained on less-displaced decaying jets rather than jets originating from long-lived particle decays.