Speaker
Description
The intracluster medium (ICM) holds signatures of the dynamical history of
the galaxy cluster, including the dark matter density profile, mergers with
other clusters, and energetic activity (from supernovae and supermassive
black holes) in its member galaxies. For all but the most relaxed galaxy
clusters observed at high spatial resolution by instruments such as the
Chandra and XMM-Newton X-ray telescopes, it is extremely challenging
to infer such properties as the mass and baryon fraction from the ICM
emission. Reproducing these features is a key test of the realism of a
given cosmological simulation. I use Nearest Neighbour Contrastive
Learning (NNCLR) to reduce images of the X-ray emission of the clusters
in TNG-Cluster to a compact representation space. We find that the
self-supervised representation space forms a continuous distribution from
cool to non-cool core clusters, as well as from relaxed to merging objects.
It also shows trends in redshift, halo mass, stellar mass, time since last
major merger, and offset between the peaks of mass and X-ray emission. The
self-supervised sorting of the images clusters known populations of galaxy
clusters, providing simulated analogues to famous observed objects like the
Bullet cluster and Perseus.
