ERGO-ML: A continuous organization of the X-ray galaxy cluster population in TNG-Cluster with contrastive learning

Abstract

The physical properties of the intracluster medium (ICM) reflect signatures of the underlying gravitational potential, mergers and strong interactions with other halos and satellite galaxies, as well as galactic feedback from supernovae and supermassive black holes (SMBHs). Traditionally, clusters have been characterized in terms of summary statistics, such as halo mass, X-ray luminosity, cool-core state, luminosity of AGN, and number of merging components. In this paper of the Extracting Reality from Galaxy Observables with Machine Learning series (ERGO-ML), we instead consider the full information content available in maps of X-ray emission from the ICM. We employ Nearest Neighbour Contrastive Learning (NNCLR) to identify and populate a low-dimensional representation space of such images. Using idealized X-ray maps of the 352 clusters of the TNG-Cluster cosmological magnetohydrodynamical simulation suite, we take three orthogonal projections of each cluster at eight snapshots within the redshift range $0\leq z<1$, resulting in a dataset of $\sim$8,000 images. Our findings reveal that this representation space forms a continuous distribution from relaxed to merging objects, and from centrally-peaked to flat emission profiles. The representation also exhibits clear trends with redshift, with halo, gas, stellar, and SMBH mass, with time since a last major merger, and with indicators of dynamical state. We show that an 8-dimensional representation can be used to predict a variety of cluster properties, find analogs, and identify correlations between physical properties, thereby suggesting causal relationships. Our analysis demonstrates that contrastive learning is a powerful tool for characterizing galaxy clusters from their images alone, allowing us to derive constraints on their physical properties and formation histories using cosmological hydrodynamical galaxy simulations.