Oral Presentation

Bridging Simulations and Observations - A deep learning framework applied to Planck clusters

Author(s): Subha Majumdar (TIFR)

Presenter: Subha Majumdar (Tata Institute of Fundamental reserach)

We introduce a deep learning graph convolutional network (GCN) method, ideally tailored for graph-structured data,
for estimating galaxy cluster masses from radial profiles of the intra-cluster medium (ICM) inferred from X-ray observations, as used for clusters detected in the Planck SZ survey.
The GCN is trained and tested using state-of-the-art hydrodynamical simulations of galaxy clusters.
The mass estimates using our method exhibit no systematic bias compared to the true cluster masses in the simulations.
Additionally, we achieve a dispersion in recovered mass versus true mass of around 9%, which is about a factor of three better than
those typically obtained from a standard hydrostatic equilibrium approach. Our algorithm is also robust with respect to the quality of
the data and the morphology of the clusters. Going beyond simulations, we apply our technique to XMM-Newton observed galaxy cluster samples and
compare the GCN-derived mass estimates with those obtained with the standard Y-M500 scaling relations, and naturally recover a bias factor (instead of putting it by hand). We also find
marginal evidence of a mass-dependent bias in SZ-derived masses, with high masses exhibiting a greater
degree of bias.