Université de Strasbourg
https://univ-grenoble-alpes-fr.zoom.us/j/98838875999?pwd=eDVaQmp3bmdjMC8rbjJUa2d2c2xBUT09
Globular clusters (GCs) are ubiquitous stellar systems found in nearly all galaxies; however, their origin in the primordial universe remains unknown. Despite their apparent simplicity, GCs present several modelling challenges: due to their old ages, the repeated gravitational interactions between ~1 million stars strongly impact their evolution and need to be calculated directly. Therefore, an efficient multi-scale approach to resolve this so-called “million-body problem” is needed. In this talk, I will present my current efforts in modelling realistic GCs, with a particular focus on a forward-modelling approach to interpret current and future observations. I will first introduce a new suite of >20 realistic star-by-star N-body simulations, run with NBODY6+++GPU, incorporating physics from the scale of stars (e.g. stellar evolution) to the one of external host galaxy (external tidal field), and using a realistic number of stars (N = 250k–1.5M). These simulations allow us to carry a first comparison with present-day GC properties and help establish a link with their formation properties in the high-z universe. I will then demonstrate how deep-learning techniques —specifically convolutional neural networks trained on synthetic images— can maximize the scientific return of these computationally expensive simulations for a direct comparison with observations. In particular, I will highlight recent developments in the pi-DOC deep-learning algorithm, designed to measure the dynamical and morphological properties of GCs in both the Milky Way and Andromeda, from space telescope observations. These promising results suggest that detailed dynamical studies of GCs could soon be extended beyond the Local Group, providing new valuable insights into the formation and evolution of these ancient stellar systems.