Abstract

Contributed Talk - Splinter GalaxyClusters

Wednesday, 09 September 2026, 17:30   (MW-1050)

Simulation-Based Inference (SBI) for eROSITA Galaxy Cluster Cosmology Using Set-Based Neural Networks

Silas Zelmer, Esra Bulbul, Kai Lehman, Sven Krippendorf, Emmanuel Artis, Sebastian Grandis, and the eROSITA collaboration
MPE, MPE & LMU, USM & ORIGINS & CCA, DAMTP and Cavendish Laboratory, MPE, Universität Innsbruck

We present a simulation-based inference (SBI) framework for cosmological parameter estimation from eROSITA galaxy cluster catalogs. Our approach combines a realistic forward-modeling pipeline with a set-based neural network and normalizing flows to leverage the full cluster-level information content. Using mock catalogs matching the size and selection properties of the eRASS1 sample, we demonstrate accurate and well-calibrated recovery of cosmological parameters in the presence of survey systematics. Despite the moderate sample size (~3,300 clusters), our method achieves constraints on Ωₘ and σ₈ comparable to traditional analyses based on significantly larger datasets. This highlights the potential of SBI methods for next-generation large-scale structure studies and the analysis of high-dimensional survey data.