Set-based Implicit Likelihood Inference of Galaxy Cluster Mass
This work addresses the challenge of improving mass estimates for galaxy clusters in astrophysics, which is incremental as it builds on existing dynamical methods with machine learning enhancements.
The paper tackles the problem of inferring galaxy cluster masses from projected galaxy dynamics by introducing a set-based machine learning framework that combines Deep Sets and conditional normalizing flows. The result is a significant reduction in scatter and well-calibrated uncertainties across the full mass range compared to traditional methods, as demonstrated on the Uchuu-UniverseMachine simulation.
We present a set-based machine learning framework that infers posterior distributions of galaxy cluster masses from projected galaxy dynamics. Our model combines Deep Sets and conditional normalizing flows to incorporate both positional and velocity information of member galaxies to predict residual corrections to the $M$-$σ$ relation for improved interpretability. Trained on the Uchuu-UniverseMachine simulation, our approach significantly reduces scatter and provides well-calibrated uncertainties across the full mass range compared to traditional dynamical estimates.