INFN@Young: 3rd event

Wednesday 15 Mar 2023, 17:00 → 19:00 Europe/Rome

Aula Conversi (Dipartimento di Fisica - Ed. G.Marconi)

1) Black Holes as particle detectors: superradiant instabilities

Speaker: Enrico Cannizzaro

Abstract: It is well known that spinning black holes (BHs) are unstable under ultralight massive bosonic perturbations. Due to this so-called superradiant instability, a macroscopic bosonic condensate can form around a spinning BH, leading to striking observable signatures such as gaps in the BH spin-mass Regge plane or gravitational wave emissions from the condensate. This phenomenon represents a powerful tool to probe ultralight degrees of freedom beyond the Standard Model, such as axions and dark photons. Until recently, studies of the superra-diant instability assumed that the superradiant field was free from interactions, aside from a minimal coupling to gravity. However, as number densities can reach extreme values in the process, the effect of interactions can be crucial, even for very weakly interacting fields. In this talk, I will discuss these phenomena and show how interactions with the Standard Model strongly modify the fate of these systems.

2) Prospects for Particle Flow Reconstruction with a GEANT4 Simulation of a Configurable Detector and Graph Neural Networks

Speaker: Lorenzo Santi

Abstract: This talk will present two papers that showcase cutting-edge techniques for particle analysis in High Energy Physics (HEP). The first paper introduces a novel detector simulation, called COCOA (Configurable Calorimeter Simulation for AI Applications), which is designed for machine learning applications. The simulation is created using Pythia8 and GEANT4 and includes built-in topological clustering and jet clustering at truth and reco level, which can be configured using a file. The second paper demonstrates the application of Graph Neural Networks (GNNs) to global particle flow using the dataset produced by the COCOA simulation. Three separate set-to-set neural network architectures are used to reconstruct particles in events with a single jet in a fully-simulated calorimeter. The performance is evaluated based on particle reconstruction quality, properties regression, and jet-level metrics. The results show that this high-dimensional end-to-end approach outperforms basic parametric approaches, particularly when using a novel architecture based on learning hypergraph structure, HGPflow.

Together, these papers provide valuable insights into the development of advanced computational techniques for improving particle detection and analysis in HEP experiments.