Internal dynamics of galaxy clusters from cosmological hydrodynamical simulations

Galaxy clusters are the most massive systems in the Universe. They are usually located at the nodes of the cosmic web from which they continuously accrete matter. In this work, by combining cosmological simulations and local Universe observations, we examined several properties of the different collisionless tracers of the internal dynamics of galaxy clusters - namely Dark Matter (DM), stars, and galaxies -- to gain insights into the main processes operating in structure formation and evolution. We base our analysis on the DIANOGA zoom-in simulation set which is composed of 29 Lagrangian regions at different levels of resolution and under varying physical conditions (full hydrodynamical and/or N-body simulations). Recent measurements (Biviano et al. 2013,2016; Capasso et al. 2019) of the pseudo-entropy (σ^2⁄ρ^(2/3) , where σ velocity dispersion and ρ density of the collisionless tracer) allowed us to study its role in the evolution of clusters as dynamical attractor (e.g., Taylor et al. 2001, Dehnen et al. 2005). Its fingerprint is the universal radial profile described by a simple power-law. We find good agreement in both normalisation and slope between observations and simulations. A significant tension is present with the galaxy member population, we discuss in detail the probable reasons behind this finding. A large body of spectroscopic measurements (Loubser et al. 2018; Sohn et al. 2020, 2021) were able to provide a large statistical sample to study the dynamics of the Brightest Cluster Galaxy (BCG). We compare scaling relations between the BCG and cluster velocity dispersions and corresponding masses: we find in general a good agreement with observational results for the former and significant tension in the latter. We analyse the key features of the velocity dispersion profiles, as traced by stars, DM, and galaxies (Sartoris et al. 2020) and they are in excellent agreement with simulations. We also quantify the assumed impact of the IntraCluster Light (ICL) in these measurements. Furthermore, given the existing dynamical distinction between BCG and ICL, we developed a Machine Learning (ML) method based on a supervised Random Forest to classify stars in simulated galaxy clusters in these two classes. We employ matched stellar catalogues (built from a modified version of Subfind, Dolag et al. 2010) to train and test the classifier. The input features are cluster mass, normalised particle clustercentric distance, and rest-frame velocity. The model is found to correctly identify most of the stars, while the larger errors are exhibited at the BCG outskirts, where the differences between the physical properties of the two components are less obvious. We find that our classifier provides consistent results in simulations for clusters at z<1, using different numerical resolutions and implementations of the feedback. The last part of the project has focused on creating a ML framework to bridge the observational analysis with predictions from simulations. Measuring the ICL in observations is a difficult task which is often solved by fitting functional profiles to the BCG+ICL light profile, but often providing significantly different results. We developed a method based on convolutional neural networks to identify the ICL distribution in mock images of galaxy clusters, according to the dynamical classification we routinely perform in simulations. We construct several sets of mock images based on different observables (i.e., magnitudes, line-of-sight velocity, and velocity dispersion) that can be employed as input by the network to predict the ICL distribution in such images. This project has highlighted the dependence of the ICL build-up on the numerical resolution of the simulations, a problem which requires further investigations.